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->sctx->cur_inode_next_write_offset)
1277 found->found_refs++;
1278 if (ino < found->ino) {
1280 found->offset = offset;
1281 } else if (found->ino == ino) {
1283 * same extent found more then once in the same file.
1285 if (found->offset > offset + bctx->extent_len)
1286 found->offset = offset;
1293 * Given an inode, offset and extent item, it finds a good clone for a clone
1294 * instruction. Returns -ENOENT when none could be found. The function makes
1295 * sure that the returned clone is usable at the point where sending is at the
1296 * moment. This means, that no clones are accepted which lie behind the current
1299 * path must point to the extent item when called.
1301 static int find_extent_clone(struct send_ctx *sctx,
1302 struct btrfs_path *path,
1303 u64 ino, u64 data_offset,
1305 struct clone_root **found)
1307 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1313 u64 extent_item_pos;
1315 struct btrfs_file_extent_item *fi;
1316 struct extent_buffer *eb = path->nodes[0];
1317 struct backref_ctx *backref_ctx = NULL;
1318 struct clone_root *cur_clone_root;
1319 struct btrfs_key found_key;
1320 struct btrfs_path *tmp_path;
1321 struct btrfs_extent_item *ei;
1325 tmp_path = alloc_path_for_send();
1329 /* We only use this path under the commit sem */
1330 tmp_path->need_commit_sem = 0;
1332 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1338 if (data_offset >= ino_size) {
1340 * There may be extents that lie behind the file's size.
1341 * I at least had this in combination with snapshotting while
1342 * writing large files.
1348 fi = btrfs_item_ptr(eb, path->slots[0],
1349 struct btrfs_file_extent_item);
1350 extent_type = btrfs_file_extent_type(eb, fi);
1351 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1355 compressed = btrfs_file_extent_compression(eb, fi);
1357 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1358 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1359 if (disk_byte == 0) {
1363 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1365 down_read(&fs_info->commit_root_sem);
1366 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1367 &found_key, &flags);
1368 up_read(&fs_info->commit_root_sem);
1372 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1377 ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0],
1378 struct btrfs_extent_item);
1380 * Backreference walking (iterate_extent_inodes() below) is currently
1381 * too expensive when an extent has a large number of references, both
1382 * in time spent and used memory. So for now just fallback to write
1383 * operations instead of clone operations when an extent has more than
1384 * a certain amount of references.
1386 if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) {
1390 btrfs_release_path(tmp_path);
1393 * Setup the clone roots.
1395 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1396 cur_clone_root = sctx->clone_roots + i;
1397 cur_clone_root->ino = (u64)-1;
1398 cur_clone_root->offset = 0;
1399 cur_clone_root->found_refs = 0;
1402 backref_ctx->sctx = sctx;
1403 backref_ctx->found = 0;
1404 backref_ctx->cur_objectid = ino;
1405 backref_ctx->cur_offset = data_offset;
1406 backref_ctx->found_itself = 0;
1407 backref_ctx->extent_len = num_bytes;
1409 * For non-compressed extents iterate_extent_inodes() gives us extent
1410 * offsets that already take into account the data offset, but not for
1411 * compressed extents, since the offset is logical and not relative to
1412 * the physical extent locations. We must take this into account to
1413 * avoid sending clone offsets that go beyond the source file's size,
1414 * which would result in the clone ioctl failing with -EINVAL on the
1417 if (compressed == BTRFS_COMPRESS_NONE)
1418 backref_ctx->data_offset = 0;
1420 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1423 * The last extent of a file may be too large due to page alignment.
1424 * We need to adjust extent_len in this case so that the checks in
1425 * __iterate_backrefs work.
1427 if (data_offset + num_bytes >= ino_size)
1428 backref_ctx->extent_len = ino_size - data_offset;
1431 * Now collect all backrefs.
1433 if (compressed == BTRFS_COMPRESS_NONE)
1434 extent_item_pos = logical - found_key.objectid;
1436 extent_item_pos = 0;
1437 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1438 extent_item_pos, 1, __iterate_backrefs,
1439 backref_ctx, false);
1444 if (!backref_ctx->found_itself) {
1445 /* found a bug in backref code? */
1448 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1449 ino, data_offset, disk_byte, found_key.objectid);
1453 btrfs_debug(fs_info,
1454 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1455 data_offset, ino, num_bytes, logical);
1457 if (!backref_ctx->found)
1458 btrfs_debug(fs_info, "no clones found");
1460 cur_clone_root = NULL;
1461 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1462 if (sctx->clone_roots[i].found_refs) {
1463 if (!cur_clone_root)
1464 cur_clone_root = sctx->clone_roots + i;
1465 else if (sctx->clone_roots[i].root == sctx->send_root)
1466 /* prefer clones from send_root over others */
1467 cur_clone_root = sctx->clone_roots + i;
1472 if (cur_clone_root) {
1473 *found = cur_clone_root;
1480 btrfs_free_path(tmp_path);
1485 static int read_symlink(struct btrfs_root *root,
1487 struct fs_path *dest)
1490 struct btrfs_path *path;
1491 struct btrfs_key key;
1492 struct btrfs_file_extent_item *ei;
1498 path = alloc_path_for_send();
1503 key.type = BTRFS_EXTENT_DATA_KEY;
1505 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1510 * An empty symlink inode. Can happen in rare error paths when
1511 * creating a symlink (transaction committed before the inode
1512 * eviction handler removed the symlink inode items and a crash
1513 * happened in between or the subvol was snapshoted in between).
1514 * Print an informative message to dmesg/syslog so that the user
1515 * can delete the symlink.
1517 btrfs_err(root->fs_info,
1518 "Found empty symlink inode %llu at root %llu",
1519 ino, root->root_key.objectid);
1524 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1525 struct btrfs_file_extent_item);
1526 type = btrfs_file_extent_type(path->nodes[0], ei);
1527 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1528 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1529 BUG_ON(compression);
1531 off = btrfs_file_extent_inline_start(ei);
1532 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1534 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1537 btrfs_free_path(path);
1542 * Helper function to generate a file name that is unique in the root of
1543 * send_root and parent_root. This is used to generate names for orphan inodes.
1545 static int gen_unique_name(struct send_ctx *sctx,
1547 struct fs_path *dest)
1550 struct btrfs_path *path;
1551 struct btrfs_dir_item *di;
1556 path = alloc_path_for_send();
1561 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1563 ASSERT(len < sizeof(tmp));
1565 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1566 path, BTRFS_FIRST_FREE_OBJECTID,
1567 tmp, strlen(tmp), 0);
1568 btrfs_release_path(path);
1574 /* not unique, try again */
1579 if (!sctx->parent_root) {
1585 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1586 path, BTRFS_FIRST_FREE_OBJECTID,
1587 tmp, strlen(tmp), 0);
1588 btrfs_release_path(path);
1594 /* not unique, try again */
1602 ret = fs_path_add(dest, tmp, strlen(tmp));
1605 btrfs_free_path(path);
1610 inode_state_no_change,
1611 inode_state_will_create,
1612 inode_state_did_create,
1613 inode_state_will_delete,
1614 inode_state_did_delete,
1617 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1625 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1627 if (ret < 0 && ret != -ENOENT)
1631 if (!sctx->parent_root) {
1632 right_ret = -ENOENT;
1634 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1635 NULL, NULL, NULL, NULL);
1636 if (ret < 0 && ret != -ENOENT)
1641 if (!left_ret && !right_ret) {
1642 if (left_gen == gen && right_gen == gen) {
1643 ret = inode_state_no_change;
1644 } else if (left_gen == gen) {
1645 if (ino < sctx->send_progress)
1646 ret = inode_state_did_create;
1648 ret = inode_state_will_create;
1649 } else if (right_gen == gen) {
1650 if (ino < sctx->send_progress)
1651 ret = inode_state_did_delete;
1653 ret = inode_state_will_delete;
1657 } else if (!left_ret) {
1658 if (left_gen == gen) {
1659 if (ino < sctx->send_progress)
1660 ret = inode_state_did_create;
1662 ret = inode_state_will_create;
1666 } else if (!right_ret) {
1667 if (right_gen == gen) {
1668 if (ino < sctx->send_progress)
1669 ret = inode_state_did_delete;
1671 ret = inode_state_will_delete;
1683 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1687 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1690 ret = get_cur_inode_state(sctx, ino, gen);
1694 if (ret == inode_state_no_change ||
1695 ret == inode_state_did_create ||
1696 ret == inode_state_will_delete)
1706 * Helper function to lookup a dir item in a dir.
1708 static int lookup_dir_item_inode(struct btrfs_root *root,
1709 u64 dir, const char *name, int name_len,
1714 struct btrfs_dir_item *di;
1715 struct btrfs_key key;
1716 struct btrfs_path *path;
1718 path = alloc_path_for_send();
1722 di = btrfs_lookup_dir_item(NULL, root, path,
1723 dir, name, name_len, 0);
1724 if (IS_ERR_OR_NULL(di)) {
1725 ret = di ? PTR_ERR(di) : -ENOENT;
1728 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1729 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1733 *found_inode = key.objectid;
1734 *found_type = btrfs_dir_type(path->nodes[0], di);
1737 btrfs_free_path(path);
1742 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1743 * generation of the parent dir and the name of the dir entry.
1745 static int get_first_ref(struct btrfs_root *root, u64 ino,
1746 u64 *dir, u64 *dir_gen, struct fs_path *name)
1749 struct btrfs_key key;
1750 struct btrfs_key found_key;
1751 struct btrfs_path *path;
1755 path = alloc_path_for_send();
1760 key.type = BTRFS_INODE_REF_KEY;
1763 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1767 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1769 if (ret || found_key.objectid != ino ||
1770 (found_key.type != BTRFS_INODE_REF_KEY &&
1771 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1776 if (found_key.type == BTRFS_INODE_REF_KEY) {
1777 struct btrfs_inode_ref *iref;
1778 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1779 struct btrfs_inode_ref);
1780 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1781 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1782 (unsigned long)(iref + 1),
1784 parent_dir = found_key.offset;
1786 struct btrfs_inode_extref *extref;
1787 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1788 struct btrfs_inode_extref);
1789 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1790 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1791 (unsigned long)&extref->name, len);
1792 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1796 btrfs_release_path(path);
1799 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1808 btrfs_free_path(path);
1812 static int is_first_ref(struct btrfs_root *root,
1814 const char *name, int name_len)
1817 struct fs_path *tmp_name;
1820 tmp_name = fs_path_alloc();
1824 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1828 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1833 ret = !memcmp(tmp_name->start, name, name_len);
1836 fs_path_free(tmp_name);
1841 * Used by process_recorded_refs to determine if a new ref would overwrite an
1842 * already existing ref. In case it detects an overwrite, it returns the
1843 * inode/gen in who_ino/who_gen.
1844 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1845 * to make sure later references to the overwritten inode are possible.
1846 * Orphanizing is however only required for the first ref of an inode.
1847 * process_recorded_refs does an additional is_first_ref check to see if
1848 * orphanizing is really required.
1850 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1851 const char *name, int name_len,
1852 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1856 u64 other_inode = 0;
1859 if (!sctx->parent_root)
1862 ret = is_inode_existent(sctx, dir, dir_gen);
1867 * If we have a parent root we need to verify that the parent dir was
1868 * not deleted and then re-created, if it was then we have no overwrite
1869 * and we can just unlink this entry.
1871 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1872 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1874 if (ret < 0 && ret != -ENOENT)
1884 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1885 &other_inode, &other_type);
1886 if (ret < 0 && ret != -ENOENT)
1894 * Check if the overwritten ref was already processed. If yes, the ref
1895 * was already unlinked/moved, so we can safely assume that we will not
1896 * overwrite anything at this point in time.
1898 if (other_inode > sctx->send_progress ||
1899 is_waiting_for_move(sctx, other_inode)) {
1900 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1901 who_gen, who_mode, NULL, NULL, NULL);
1906 *who_ino = other_inode;
1916 * Checks if the ref was overwritten by an already processed inode. This is
1917 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1918 * thus the orphan name needs be used.
1919 * process_recorded_refs also uses it to avoid unlinking of refs that were
1922 static int did_overwrite_ref(struct send_ctx *sctx,
1923 u64 dir, u64 dir_gen,
1924 u64 ino, u64 ino_gen,
1925 const char *name, int name_len)
1932 if (!sctx->parent_root)
1935 ret = is_inode_existent(sctx, dir, dir_gen);
1939 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1940 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1942 if (ret < 0 && ret != -ENOENT)
1952 /* check if the ref was overwritten by another ref */
1953 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1954 &ow_inode, &other_type);
1955 if (ret < 0 && ret != -ENOENT)
1958 /* was never and will never be overwritten */
1963 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1968 if (ow_inode == ino && gen == ino_gen) {
1974 * We know that it is or will be overwritten. Check this now.
1975 * The current inode being processed might have been the one that caused
1976 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1977 * the current inode being processed.
1979 if ((ow_inode < sctx->send_progress) ||
1980 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1981 gen == sctx->cur_inode_gen))
1991 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1992 * that got overwritten. This is used by process_recorded_refs to determine
1993 * if it has to use the path as returned by get_cur_path or the orphan name.
1995 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1998 struct fs_path *name = NULL;
2002 if (!sctx->parent_root)
2005 name = fs_path_alloc();
2009 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
2013 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2014 name->start, fs_path_len(name));
2022 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2023 * so we need to do some special handling in case we have clashes. This function
2024 * takes care of this with the help of name_cache_entry::radix_list.
2025 * In case of error, nce is kfreed.
2027 static int name_cache_insert(struct send_ctx *sctx,
2028 struct name_cache_entry *nce)
2031 struct list_head *nce_head;
2033 nce_head = radix_tree_lookup(&sctx->name_cache,
2034 (unsigned long)nce->ino);
2036 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2041 INIT_LIST_HEAD(nce_head);
2043 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2050 list_add_tail(&nce->radix_list, nce_head);
2051 list_add_tail(&nce->list, &sctx->name_cache_list);
2052 sctx->name_cache_size++;
2057 static void name_cache_delete(struct send_ctx *sctx,
2058 struct name_cache_entry *nce)
2060 struct list_head *nce_head;
2062 nce_head = radix_tree_lookup(&sctx->name_cache,
2063 (unsigned long)nce->ino);
2065 btrfs_err(sctx->send_root->fs_info,
2066 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2067 nce->ino, sctx->name_cache_size);
2070 list_del(&nce->radix_list);
2071 list_del(&nce->list);
2072 sctx->name_cache_size--;
2075 * We may not get to the final release of nce_head if the lookup fails
2077 if (nce_head && list_empty(nce_head)) {
2078 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2083 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2086 struct list_head *nce_head;
2087 struct name_cache_entry *cur;
2089 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2093 list_for_each_entry(cur, nce_head, radix_list) {
2094 if (cur->ino == ino && cur->gen == gen)
2101 * Removes the entry from the list and adds it back to the end. This marks the
2102 * entry as recently used so that name_cache_clean_unused does not remove it.
2104 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2106 list_del(&nce->list);
2107 list_add_tail(&nce->list, &sctx->name_cache_list);
2111 * Remove some entries from the beginning of name_cache_list.
2113 static void name_cache_clean_unused(struct send_ctx *sctx)
2115 struct name_cache_entry *nce;
2117 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2120 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2121 nce = list_entry(sctx->name_cache_list.next,
2122 struct name_cache_entry, list);
2123 name_cache_delete(sctx, nce);
2128 static void name_cache_free(struct send_ctx *sctx)
2130 struct name_cache_entry *nce;
2132 while (!list_empty(&sctx->name_cache_list)) {
2133 nce = list_entry(sctx->name_cache_list.next,
2134 struct name_cache_entry, list);
2135 name_cache_delete(sctx, nce);
2141 * Used by get_cur_path for each ref up to the root.
2142 * Returns 0 if it succeeded.
2143 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2144 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2145 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2146 * Returns <0 in case of error.
2148 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2152 struct fs_path *dest)
2156 struct name_cache_entry *nce = NULL;
2159 * First check if we already did a call to this function with the same
2160 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2161 * return the cached result.
2163 nce = name_cache_search(sctx, ino, gen);
2165 if (ino < sctx->send_progress && nce->need_later_update) {
2166 name_cache_delete(sctx, nce);
2170 name_cache_used(sctx, nce);
2171 *parent_ino = nce->parent_ino;
2172 *parent_gen = nce->parent_gen;
2173 ret = fs_path_add(dest, nce->name, nce->name_len);
2182 * If the inode is not existent yet, add the orphan name and return 1.
2183 * This should only happen for the parent dir that we determine in
2186 ret = is_inode_existent(sctx, ino, gen);
2191 ret = gen_unique_name(sctx, ino, gen, dest);
2199 * Depending on whether the inode was already processed or not, use
2200 * send_root or parent_root for ref lookup.
2202 if (ino < sctx->send_progress)
2203 ret = get_first_ref(sctx->send_root, ino,
2204 parent_ino, parent_gen, dest);
2206 ret = get_first_ref(sctx->parent_root, ino,
2207 parent_ino, parent_gen, dest);
2212 * Check if the ref was overwritten by an inode's ref that was processed
2213 * earlier. If yes, treat as orphan and return 1.
2215 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2216 dest->start, dest->end - dest->start);
2220 fs_path_reset(dest);
2221 ret = gen_unique_name(sctx, ino, gen, dest);
2229 * Store the result of the lookup in the name cache.
2231 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2239 nce->parent_ino = *parent_ino;
2240 nce->parent_gen = *parent_gen;
2241 nce->name_len = fs_path_len(dest);
2243 strcpy(nce->name, dest->start);
2245 if (ino < sctx->send_progress)
2246 nce->need_later_update = 0;
2248 nce->need_later_update = 1;
2250 nce_ret = name_cache_insert(sctx, nce);
2253 name_cache_clean_unused(sctx);
2260 * Magic happens here. This function returns the first ref to an inode as it
2261 * would look like while receiving the stream at this point in time.
2262 * We walk the path up to the root. For every inode in between, we check if it
2263 * was already processed/sent. If yes, we continue with the parent as found
2264 * in send_root. If not, we continue with the parent as found in parent_root.
2265 * If we encounter an inode that was deleted at this point in time, we use the
2266 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2267 * that were not created yet and overwritten inodes/refs.
2269 * When do we have orphan inodes:
2270 * 1. When an inode is freshly created and thus no valid refs are available yet
2271 * 2. When a directory lost all it's refs (deleted) but still has dir items
2272 * inside which were not processed yet (pending for move/delete). If anyone
2273 * tried to get the path to the dir items, it would get a path inside that
2275 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2276 * of an unprocessed inode. If in that case the first ref would be
2277 * overwritten, the overwritten inode gets "orphanized". Later when we
2278 * process this overwritten inode, it is restored at a new place by moving
2281 * sctx->send_progress tells this function at which point in time receiving
2284 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2285 struct fs_path *dest)
2288 struct fs_path *name = NULL;
2289 u64 parent_inode = 0;
2293 name = fs_path_alloc();
2300 fs_path_reset(dest);
2302 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2303 struct waiting_dir_move *wdm;
2305 fs_path_reset(name);
2307 if (is_waiting_for_rm(sctx, ino)) {
2308 ret = gen_unique_name(sctx, ino, gen, name);
2311 ret = fs_path_add_path(dest, name);
2315 wdm = get_waiting_dir_move(sctx, ino);
2316 if (wdm && wdm->orphanized) {
2317 ret = gen_unique_name(sctx, ino, gen, name);
2320 ret = get_first_ref(sctx->parent_root, ino,
2321 &parent_inode, &parent_gen, name);
2323 ret = __get_cur_name_and_parent(sctx, ino, gen,
2333 ret = fs_path_add_path(dest, name);
2344 fs_path_unreverse(dest);
2349 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2351 static int send_subvol_begin(struct send_ctx *sctx)
2354 struct btrfs_root *send_root = sctx->send_root;
2355 struct btrfs_root *parent_root = sctx->parent_root;
2356 struct btrfs_path *path;
2357 struct btrfs_key key;
2358 struct btrfs_root_ref *ref;
2359 struct extent_buffer *leaf;
2363 path = btrfs_alloc_path();
2367 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2369 btrfs_free_path(path);
2373 key.objectid = send_root->root_key.objectid;
2374 key.type = BTRFS_ROOT_BACKREF_KEY;
2377 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2386 leaf = path->nodes[0];
2387 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2388 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2389 key.objectid != send_root->root_key.objectid) {
2393 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2394 namelen = btrfs_root_ref_name_len(leaf, ref);
2395 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2396 btrfs_release_path(path);
2399 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2403 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2408 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2410 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2411 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2412 sctx->send_root->root_item.received_uuid);
2414 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2415 sctx->send_root->root_item.uuid);
2417 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2418 le64_to_cpu(sctx->send_root->root_item.ctransid));
2420 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2421 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2422 parent_root->root_item.received_uuid);
2424 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2425 parent_root->root_item.uuid);
2426 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2427 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2430 ret = send_cmd(sctx);
2434 btrfs_free_path(path);
2439 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2441 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2445 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2447 p = fs_path_alloc();
2451 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2455 ret = get_cur_path(sctx, ino, gen, p);
2458 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2459 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2461 ret = send_cmd(sctx);
2469 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2471 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2475 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2477 p = fs_path_alloc();
2481 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2485 ret = get_cur_path(sctx, ino, gen, p);
2488 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2489 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2491 ret = send_cmd(sctx);
2499 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2501 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2505 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2508 p = fs_path_alloc();
2512 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2516 ret = get_cur_path(sctx, ino, gen, p);
2519 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2520 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2521 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2523 ret = send_cmd(sctx);
2531 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2533 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2535 struct fs_path *p = NULL;
2536 struct btrfs_inode_item *ii;
2537 struct btrfs_path *path = NULL;
2538 struct extent_buffer *eb;
2539 struct btrfs_key key;
2542 btrfs_debug(fs_info, "send_utimes %llu", ino);
2544 p = fs_path_alloc();
2548 path = alloc_path_for_send();
2555 key.type = BTRFS_INODE_ITEM_KEY;
2557 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2563 eb = path->nodes[0];
2564 slot = path->slots[0];
2565 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2567 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2571 ret = get_cur_path(sctx, ino, gen, p);
2574 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2575 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2576 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2577 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2578 /* TODO Add otime support when the otime patches get into upstream */
2580 ret = send_cmd(sctx);
2585 btrfs_free_path(path);
2590 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2591 * a valid path yet because we did not process the refs yet. So, the inode
2592 * is created as orphan.
2594 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2596 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2604 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2606 p = fs_path_alloc();
2610 if (ino != sctx->cur_ino) {
2611 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2616 gen = sctx->cur_inode_gen;
2617 mode = sctx->cur_inode_mode;
2618 rdev = sctx->cur_inode_rdev;
2621 if (S_ISREG(mode)) {
2622 cmd = BTRFS_SEND_C_MKFILE;
2623 } else if (S_ISDIR(mode)) {
2624 cmd = BTRFS_SEND_C_MKDIR;
2625 } else if (S_ISLNK(mode)) {
2626 cmd = BTRFS_SEND_C_SYMLINK;
2627 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2628 cmd = BTRFS_SEND_C_MKNOD;
2629 } else if (S_ISFIFO(mode)) {
2630 cmd = BTRFS_SEND_C_MKFIFO;
2631 } else if (S_ISSOCK(mode)) {
2632 cmd = BTRFS_SEND_C_MKSOCK;
2634 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2635 (int)(mode & S_IFMT));
2640 ret = begin_cmd(sctx, cmd);
2644 ret = gen_unique_name(sctx, ino, gen, p);
2648 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2649 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2651 if (S_ISLNK(mode)) {
2653 ret = read_symlink(sctx->send_root, ino, p);
2656 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2657 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2658 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2659 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2660 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2663 ret = send_cmd(sctx);
2675 * We need some special handling for inodes that get processed before the parent
2676 * directory got created. See process_recorded_refs for details.
2677 * This function does the check if we already created the dir out of order.
2679 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2682 struct btrfs_path *path = NULL;
2683 struct btrfs_key key;
2684 struct btrfs_key found_key;
2685 struct btrfs_key di_key;
2686 struct extent_buffer *eb;
2687 struct btrfs_dir_item *di;
2690 path = alloc_path_for_send();
2697 key.type = BTRFS_DIR_INDEX_KEY;
2699 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2704 eb = path->nodes[0];
2705 slot = path->slots[0];
2706 if (slot >= btrfs_header_nritems(eb)) {
2707 ret = btrfs_next_leaf(sctx->send_root, path);
2710 } else if (ret > 0) {
2717 btrfs_item_key_to_cpu(eb, &found_key, slot);
2718 if (found_key.objectid != key.objectid ||
2719 found_key.type != key.type) {
2724 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2725 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2727 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2728 di_key.objectid < sctx->send_progress) {
2737 btrfs_free_path(path);
2742 * Only creates the inode if it is:
2743 * 1. Not a directory
2744 * 2. Or a directory which was not created already due to out of order
2745 * directories. See did_create_dir and process_recorded_refs for details.
2747 static int send_create_inode_if_needed(struct send_ctx *sctx)
2751 if (S_ISDIR(sctx->cur_inode_mode)) {
2752 ret = did_create_dir(sctx, sctx->cur_ino);
2761 ret = send_create_inode(sctx, sctx->cur_ino);
2769 struct recorded_ref {
2770 struct list_head list;
2772 struct fs_path *full_path;
2778 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2780 ref->full_path = path;
2781 ref->name = (char *)kbasename(ref->full_path->start);
2782 ref->name_len = ref->full_path->end - ref->name;
2786 * We need to process new refs before deleted refs, but compare_tree gives us
2787 * everything mixed. So we first record all refs and later process them.
2788 * This function is a helper to record one ref.
2790 static int __record_ref(struct list_head *head, u64 dir,
2791 u64 dir_gen, struct fs_path *path)
2793 struct recorded_ref *ref;
2795 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2800 ref->dir_gen = dir_gen;
2801 set_ref_path(ref, path);
2802 list_add_tail(&ref->list, head);
2806 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2808 struct recorded_ref *new;
2810 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2814 new->dir = ref->dir;
2815 new->dir_gen = ref->dir_gen;
2816 new->full_path = NULL;
2817 INIT_LIST_HEAD(&new->list);
2818 list_add_tail(&new->list, list);
2822 static void __free_recorded_refs(struct list_head *head)
2824 struct recorded_ref *cur;
2826 while (!list_empty(head)) {
2827 cur = list_entry(head->next, struct recorded_ref, list);
2828 fs_path_free(cur->full_path);
2829 list_del(&cur->list);
2834 static void free_recorded_refs(struct send_ctx *sctx)
2836 __free_recorded_refs(&sctx->new_refs);
2837 __free_recorded_refs(&sctx->deleted_refs);
2841 * Renames/moves a file/dir to its orphan name. Used when the first
2842 * ref of an unprocessed inode gets overwritten and for all non empty
2845 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2846 struct fs_path *path)
2849 struct fs_path *orphan;
2851 orphan = fs_path_alloc();
2855 ret = gen_unique_name(sctx, ino, gen, orphan);
2859 ret = send_rename(sctx, path, orphan);
2862 fs_path_free(orphan);
2866 static struct orphan_dir_info *
2867 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2869 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2870 struct rb_node *parent = NULL;
2871 struct orphan_dir_info *entry, *odi;
2875 entry = rb_entry(parent, struct orphan_dir_info, node);
2876 if (dir_ino < entry->ino) {
2878 } else if (dir_ino > entry->ino) {
2879 p = &(*p)->rb_right;
2885 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2887 return ERR_PTR(-ENOMEM);
2890 odi->last_dir_index_offset = 0;
2892 rb_link_node(&odi->node, parent, p);
2893 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2897 static struct orphan_dir_info *
2898 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2900 struct rb_node *n = sctx->orphan_dirs.rb_node;
2901 struct orphan_dir_info *entry;
2904 entry = rb_entry(n, struct orphan_dir_info, node);
2905 if (dir_ino < entry->ino)
2907 else if (dir_ino > entry->ino)
2915 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2917 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2922 static void free_orphan_dir_info(struct send_ctx *sctx,
2923 struct orphan_dir_info *odi)
2927 rb_erase(&odi->node, &sctx->orphan_dirs);
2932 * Returns 1 if a directory can be removed at this point in time.
2933 * We check this by iterating all dir items and checking if the inode behind
2934 * the dir item was already processed.
2936 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2940 struct btrfs_root *root = sctx->parent_root;
2941 struct btrfs_path *path;
2942 struct btrfs_key key;
2943 struct btrfs_key found_key;
2944 struct btrfs_key loc;
2945 struct btrfs_dir_item *di;
2946 struct orphan_dir_info *odi = NULL;
2949 * Don't try to rmdir the top/root subvolume dir.
2951 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2954 path = alloc_path_for_send();
2959 key.type = BTRFS_DIR_INDEX_KEY;
2962 odi = get_orphan_dir_info(sctx, dir);
2964 key.offset = odi->last_dir_index_offset;
2966 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2971 struct waiting_dir_move *dm;
2973 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2974 ret = btrfs_next_leaf(root, path);
2981 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2983 if (found_key.objectid != key.objectid ||
2984 found_key.type != key.type)
2987 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2988 struct btrfs_dir_item);
2989 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2991 dm = get_waiting_dir_move(sctx, loc.objectid);
2993 odi = add_orphan_dir_info(sctx, dir);
2999 odi->last_dir_index_offset = found_key.offset;
3000 dm->rmdir_ino = dir;
3005 if (loc.objectid > send_progress) {
3006 odi = add_orphan_dir_info(sctx, dir);
3012 odi->last_dir_index_offset = found_key.offset;
3019 free_orphan_dir_info(sctx, odi);
3024 btrfs_free_path(path);
3028 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3030 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3032 return entry != NULL;
3035 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3037 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3038 struct rb_node *parent = NULL;
3039 struct waiting_dir_move *entry, *dm;
3041 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3046 dm->orphanized = orphanized;
3050 entry = rb_entry(parent, struct waiting_dir_move, node);
3051 if (ino < entry->ino) {
3053 } else if (ino > entry->ino) {
3054 p = &(*p)->rb_right;
3061 rb_link_node(&dm->node, parent, p);
3062 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3066 static struct waiting_dir_move *
3067 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3069 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3070 struct waiting_dir_move *entry;
3073 entry = rb_entry(n, struct waiting_dir_move, node);
3074 if (ino < entry->ino)
3076 else if (ino > entry->ino)
3084 static void free_waiting_dir_move(struct send_ctx *sctx,
3085 struct waiting_dir_move *dm)
3089 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3093 static int add_pending_dir_move(struct send_ctx *sctx,
3097 struct list_head *new_refs,
3098 struct list_head *deleted_refs,
3099 const bool is_orphan)
3101 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3102 struct rb_node *parent = NULL;
3103 struct pending_dir_move *entry = NULL, *pm;
3104 struct recorded_ref *cur;
3108 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3111 pm->parent_ino = parent_ino;
3114 INIT_LIST_HEAD(&pm->list);
3115 INIT_LIST_HEAD(&pm->update_refs);
3116 RB_CLEAR_NODE(&pm->node);
3120 entry = rb_entry(parent, struct pending_dir_move, node);
3121 if (parent_ino < entry->parent_ino) {
3123 } else if (parent_ino > entry->parent_ino) {
3124 p = &(*p)->rb_right;
3131 list_for_each_entry(cur, deleted_refs, list) {
3132 ret = dup_ref(cur, &pm->update_refs);
3136 list_for_each_entry(cur, new_refs, list) {
3137 ret = dup_ref(cur, &pm->update_refs);
3142 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3147 list_add_tail(&pm->list, &entry->list);
3149 rb_link_node(&pm->node, parent, p);
3150 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3155 __free_recorded_refs(&pm->update_refs);
3161 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3164 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3165 struct pending_dir_move *entry;
3168 entry = rb_entry(n, struct pending_dir_move, node);
3169 if (parent_ino < entry->parent_ino)
3171 else if (parent_ino > entry->parent_ino)
3179 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3180 u64 ino, u64 gen, u64 *ancestor_ino)
3183 u64 parent_inode = 0;
3185 u64 start_ino = ino;
3188 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3189 fs_path_reset(name);
3191 if (is_waiting_for_rm(sctx, ino))
3193 if (is_waiting_for_move(sctx, ino)) {
3194 if (*ancestor_ino == 0)
3195 *ancestor_ino = ino;
3196 ret = get_first_ref(sctx->parent_root, ino,
3197 &parent_inode, &parent_gen, name);
3199 ret = __get_cur_name_and_parent(sctx, ino, gen,
3209 if (parent_inode == start_ino) {
3211 if (*ancestor_ino == 0)
3212 *ancestor_ino = ino;
3221 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3223 struct fs_path *from_path = NULL;
3224 struct fs_path *to_path = NULL;
3225 struct fs_path *name = NULL;
3226 u64 orig_progress = sctx->send_progress;
3227 struct recorded_ref *cur;
3228 u64 parent_ino, parent_gen;
3229 struct waiting_dir_move *dm = NULL;
3235 name = fs_path_alloc();
3236 from_path = fs_path_alloc();
3237 if (!name || !from_path) {
3242 dm = get_waiting_dir_move(sctx, pm->ino);
3244 rmdir_ino = dm->rmdir_ino;
3245 is_orphan = dm->orphanized;
3246 free_waiting_dir_move(sctx, dm);
3249 ret = gen_unique_name(sctx, pm->ino,
3250 pm->gen, from_path);
3252 ret = get_first_ref(sctx->parent_root, pm->ino,
3253 &parent_ino, &parent_gen, name);
3256 ret = get_cur_path(sctx, parent_ino, parent_gen,
3260 ret = fs_path_add_path(from_path, name);
3265 sctx->send_progress = sctx->cur_ino + 1;
3266 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3270 LIST_HEAD(deleted_refs);
3271 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3272 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3273 &pm->update_refs, &deleted_refs,
3278 dm = get_waiting_dir_move(sctx, pm->ino);
3280 dm->rmdir_ino = rmdir_ino;
3284 fs_path_reset(name);
3287 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3291 ret = send_rename(sctx, from_path, to_path);
3296 struct orphan_dir_info *odi;
3299 odi = get_orphan_dir_info(sctx, rmdir_ino);
3301 /* already deleted */
3306 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3312 name = fs_path_alloc();
3317 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3320 ret = send_rmdir(sctx, name);
3326 ret = send_utimes(sctx, pm->ino, pm->gen);
3331 * After rename/move, need to update the utimes of both new parent(s)
3332 * and old parent(s).
3334 list_for_each_entry(cur, &pm->update_refs, list) {
3336 * The parent inode might have been deleted in the send snapshot
3338 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3339 NULL, NULL, NULL, NULL, NULL);
3340 if (ret == -ENOENT) {
3347 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3354 fs_path_free(from_path);
3355 fs_path_free(to_path);
3356 sctx->send_progress = orig_progress;
3361 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3363 if (!list_empty(&m->list))
3365 if (!RB_EMPTY_NODE(&m->node))
3366 rb_erase(&m->node, &sctx->pending_dir_moves);
3367 __free_recorded_refs(&m->update_refs);
3371 static void tail_append_pending_moves(struct send_ctx *sctx,
3372 struct pending_dir_move *moves,
3373 struct list_head *stack)
3375 if (list_empty(&moves->list)) {
3376 list_add_tail(&moves->list, stack);
3379 list_splice_init(&moves->list, &list);
3380 list_add_tail(&moves->list, stack);
3381 list_splice_tail(&list, stack);
3383 if (!RB_EMPTY_NODE(&moves->node)) {
3384 rb_erase(&moves->node, &sctx->pending_dir_moves);
3385 RB_CLEAR_NODE(&moves->node);
3389 static int apply_children_dir_moves(struct send_ctx *sctx)
3391 struct pending_dir_move *pm;
3392 struct list_head stack;
3393 u64 parent_ino = sctx->cur_ino;
3396 pm = get_pending_dir_moves(sctx, parent_ino);
3400 INIT_LIST_HEAD(&stack);
3401 tail_append_pending_moves(sctx, pm, &stack);
3403 while (!list_empty(&stack)) {
3404 pm = list_first_entry(&stack, struct pending_dir_move, list);
3405 parent_ino = pm->ino;
3406 ret = apply_dir_move(sctx, pm);
3407 free_pending_move(sctx, pm);
3410 pm = get_pending_dir_moves(sctx, parent_ino);
3412 tail_append_pending_moves(sctx, pm, &stack);
3417 while (!list_empty(&stack)) {
3418 pm = list_first_entry(&stack, struct pending_dir_move, list);
3419 free_pending_move(sctx, pm);
3425 * We might need to delay a directory rename even when no ancestor directory
3426 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3427 * renamed. This happens when we rename a directory to the old name (the name
3428 * in the parent root) of some other unrelated directory that got its rename
3429 * delayed due to some ancestor with higher number that got renamed.
3435 * |---- a/ (ino 257)
3436 * | |---- file (ino 260)
3438 * |---- b/ (ino 258)
3439 * |---- c/ (ino 259)
3443 * |---- a/ (ino 258)
3444 * |---- x/ (ino 259)
3445 * |---- y/ (ino 257)
3446 * |----- file (ino 260)
3448 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3449 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3450 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3453 * 1 - rename 259 from 'c' to 'x'
3454 * 2 - rename 257 from 'a' to 'x/y'
3455 * 3 - rename 258 from 'b' to 'a'
3457 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3458 * be done right away and < 0 on error.
3460 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3461 struct recorded_ref *parent_ref,
3462 const bool is_orphan)
3464 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3465 struct btrfs_path *path;
3466 struct btrfs_key key;
3467 struct btrfs_key di_key;
3468 struct btrfs_dir_item *di;
3472 struct waiting_dir_move *wdm;
3474 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3477 path = alloc_path_for_send();
3481 key.objectid = parent_ref->dir;
3482 key.type = BTRFS_DIR_ITEM_KEY;
3483 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3485 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3488 } else if (ret > 0) {
3493 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3494 parent_ref->name_len);
3500 * di_key.objectid has the number of the inode that has a dentry in the
3501 * parent directory with the same name that sctx->cur_ino is being
3502 * renamed to. We need to check if that inode is in the send root as
3503 * well and if it is currently marked as an inode with a pending rename,
3504 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3505 * that it happens after that other inode is renamed.
3507 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3508 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3513 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3514 &left_gen, NULL, NULL, NULL, NULL);
3517 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3518 &right_gen, NULL, NULL, NULL, NULL);
3525 /* Different inode, no need to delay the rename of sctx->cur_ino */
3526 if (right_gen != left_gen) {
3531 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3532 if (wdm && !wdm->orphanized) {
3533 ret = add_pending_dir_move(sctx,
3535 sctx->cur_inode_gen,
3538 &sctx->deleted_refs,
3544 btrfs_free_path(path);
3549 * Check if inode ino2, or any of its ancestors, is inode ino1.
3550 * Return 1 if true, 0 if false and < 0 on error.
3552 static int check_ino_in_path(struct btrfs_root *root,
3557 struct fs_path *fs_path)
3562 return ino1_gen == ino2_gen;
3564 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3569 fs_path_reset(fs_path);
3570 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3574 return parent_gen == ino1_gen;
3581 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3582 * possible path (in case ino2 is not a directory and has multiple hard links).
3583 * Return 1 if true, 0 if false and < 0 on error.
3585 static int is_ancestor(struct btrfs_root *root,
3589 struct fs_path *fs_path)
3591 bool free_fs_path = false;
3593 struct btrfs_path *path = NULL;
3594 struct btrfs_key key;
3597 fs_path = fs_path_alloc();
3600 free_fs_path = true;
3603 path = alloc_path_for_send();
3609 key.objectid = ino2;
3610 key.type = BTRFS_INODE_REF_KEY;
3613 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3618 struct extent_buffer *leaf = path->nodes[0];
3619 int slot = path->slots[0];
3623 if (slot >= btrfs_header_nritems(leaf)) {
3624 ret = btrfs_next_leaf(root, path);
3632 btrfs_item_key_to_cpu(leaf, &key, slot);
3633 if (key.objectid != ino2)
3635 if (key.type != BTRFS_INODE_REF_KEY &&
3636 key.type != BTRFS_INODE_EXTREF_KEY)
3639 item_size = btrfs_item_size_nr(leaf, slot);
3640 while (cur_offset < item_size) {
3644 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3646 struct btrfs_inode_extref *extref;
3648 ptr = btrfs_item_ptr_offset(leaf, slot);
3649 extref = (struct btrfs_inode_extref *)
3651 parent = btrfs_inode_extref_parent(leaf,
3653 cur_offset += sizeof(*extref);
3654 cur_offset += btrfs_inode_extref_name_len(leaf,
3657 parent = key.offset;
3658 cur_offset = item_size;
3661 ret = get_inode_info(root, parent, NULL, &parent_gen,
3662 NULL, NULL, NULL, NULL);
3665 ret = check_ino_in_path(root, ino1, ino1_gen,
3666 parent, parent_gen, fs_path);
3674 btrfs_free_path(path);
3676 fs_path_free(fs_path);
3680 static int wait_for_parent_move(struct send_ctx *sctx,
3681 struct recorded_ref *parent_ref,
3682 const bool is_orphan)
3685 u64 ino = parent_ref->dir;
3686 u64 ino_gen = parent_ref->dir_gen;
3687 u64 parent_ino_before, parent_ino_after;
3688 struct fs_path *path_before = NULL;
3689 struct fs_path *path_after = NULL;
3692 path_after = fs_path_alloc();
3693 path_before = fs_path_alloc();
3694 if (!path_after || !path_before) {
3700 * Our current directory inode may not yet be renamed/moved because some
3701 * ancestor (immediate or not) has to be renamed/moved first. So find if
3702 * such ancestor exists and make sure our own rename/move happens after
3703 * that ancestor is processed to avoid path build infinite loops (done
3704 * at get_cur_path()).
3706 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3707 u64 parent_ino_after_gen;
3709 if (is_waiting_for_move(sctx, ino)) {
3711 * If the current inode is an ancestor of ino in the
3712 * parent root, we need to delay the rename of the
3713 * current inode, otherwise don't delayed the rename
3714 * because we can end up with a circular dependency
3715 * of renames, resulting in some directories never
3716 * getting the respective rename operations issued in
3717 * the send stream or getting into infinite path build
3720 ret = is_ancestor(sctx->parent_root,
3721 sctx->cur_ino, sctx->cur_inode_gen,
3727 fs_path_reset(path_before);
3728 fs_path_reset(path_after);
3730 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3731 &parent_ino_after_gen, path_after);
3734 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3736 if (ret < 0 && ret != -ENOENT) {
3738 } else if (ret == -ENOENT) {
3743 len1 = fs_path_len(path_before);
3744 len2 = fs_path_len(path_after);
3745 if (ino > sctx->cur_ino &&
3746 (parent_ino_before != parent_ino_after || len1 != len2 ||
3747 memcmp(path_before->start, path_after->start, len1))) {
3750 ret = get_inode_info(sctx->parent_root, ino, NULL,
3751 &parent_ino_gen, NULL, NULL, NULL,
3755 if (ino_gen == parent_ino_gen) {
3760 ino = parent_ino_after;
3761 ino_gen = parent_ino_after_gen;
3765 fs_path_free(path_before);
3766 fs_path_free(path_after);
3769 ret = add_pending_dir_move(sctx,
3771 sctx->cur_inode_gen,
3774 &sctx->deleted_refs,
3783 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3786 struct fs_path *new_path;
3789 * Our reference's name member points to its full_path member string, so
3790 * we use here a new path.
3792 new_path = fs_path_alloc();
3796 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3798 fs_path_free(new_path);
3801 ret = fs_path_add(new_path, ref->name, ref->name_len);
3803 fs_path_free(new_path);
3807 fs_path_free(ref->full_path);
3808 set_ref_path(ref, new_path);
3814 * This does all the move/link/unlink/rmdir magic.
3816 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3818 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3820 struct recorded_ref *cur;
3821 struct recorded_ref *cur2;
3822 struct list_head check_dirs;
3823 struct fs_path *valid_path = NULL;
3827 int did_overwrite = 0;
3829 u64 last_dir_ino_rm = 0;
3830 bool can_rename = true;
3831 bool orphanized_dir = false;
3832 bool orphanized_ancestor = false;
3834 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3837 * This should never happen as the root dir always has the same ref
3838 * which is always '..'
3840 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3841 INIT_LIST_HEAD(&check_dirs);
3843 valid_path = fs_path_alloc();
3850 * First, check if the first ref of the current inode was overwritten
3851 * before. If yes, we know that the current inode was already orphanized
3852 * and thus use the orphan name. If not, we can use get_cur_path to
3853 * get the path of the first ref as it would like while receiving at
3854 * this point in time.
3855 * New inodes are always orphan at the beginning, so force to use the
3856 * orphan name in this case.
3857 * The first ref is stored in valid_path and will be updated if it
3858 * gets moved around.
3860 if (!sctx->cur_inode_new) {
3861 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3862 sctx->cur_inode_gen);
3868 if (sctx->cur_inode_new || did_overwrite) {
3869 ret = gen_unique_name(sctx, sctx->cur_ino,
3870 sctx->cur_inode_gen, valid_path);
3875 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3881 list_for_each_entry(cur, &sctx->new_refs, list) {
3883 * We may have refs where the parent directory does not exist
3884 * yet. This happens if the parent directories inum is higher
3885 * than the current inum. To handle this case, we create the
3886 * parent directory out of order. But we need to check if this
3887 * did already happen before due to other refs in the same dir.
3889 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3892 if (ret == inode_state_will_create) {
3895 * First check if any of the current inodes refs did
3896 * already create the dir.
3898 list_for_each_entry(cur2, &sctx->new_refs, list) {
3901 if (cur2->dir == cur->dir) {
3908 * If that did not happen, check if a previous inode
3909 * did already create the dir.
3912 ret = did_create_dir(sctx, cur->dir);
3916 ret = send_create_inode(sctx, cur->dir);
3923 * Check if this new ref would overwrite the first ref of
3924 * another unprocessed inode. If yes, orphanize the
3925 * overwritten inode. If we find an overwritten ref that is
3926 * not the first ref, simply unlink it.
3928 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3929 cur->name, cur->name_len,
3930 &ow_inode, &ow_gen, &ow_mode);
3934 ret = is_first_ref(sctx->parent_root,
3935 ow_inode, cur->dir, cur->name,
3940 struct name_cache_entry *nce;
3941 struct waiting_dir_move *wdm;
3943 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3947 if (S_ISDIR(ow_mode))
3948 orphanized_dir = true;
3951 * If ow_inode has its rename operation delayed
3952 * make sure that its orphanized name is used in
3953 * the source path when performing its rename
3956 if (is_waiting_for_move(sctx, ow_inode)) {
3957 wdm = get_waiting_dir_move(sctx,
3960 wdm->orphanized = true;
3964 * Make sure we clear our orphanized inode's
3965 * name from the name cache. This is because the
3966 * inode ow_inode might be an ancestor of some
3967 * other inode that will be orphanized as well
3968 * later and has an inode number greater than
3969 * sctx->send_progress. We need to prevent
3970 * future name lookups from using the old name
3971 * and get instead the orphan name.
3973 nce = name_cache_search(sctx, ow_inode, ow_gen);
3975 name_cache_delete(sctx, nce);
3980 * ow_inode might currently be an ancestor of
3981 * cur_ino, therefore compute valid_path (the
3982 * current path of cur_ino) again because it
3983 * might contain the pre-orphanization name of
3984 * ow_inode, which is no longer valid.
3986 ret = is_ancestor(sctx->parent_root,
3988 sctx->cur_ino, NULL);
3990 orphanized_ancestor = true;
3991 fs_path_reset(valid_path);
3992 ret = get_cur_path(sctx, sctx->cur_ino,
3993 sctx->cur_inode_gen,
3999 ret = send_unlink(sctx, cur->full_path);
4005 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4006 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4015 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4017 ret = wait_for_parent_move(sctx, cur, is_orphan);
4027 * link/move the ref to the new place. If we have an orphan
4028 * inode, move it and update valid_path. If not, link or move
4029 * it depending on the inode mode.
4031 if (is_orphan && can_rename) {
4032 ret = send_rename(sctx, valid_path, cur->full_path);
4036 ret = fs_path_copy(valid_path, cur->full_path);
4039 } else if (can_rename) {
4040 if (S_ISDIR(sctx->cur_inode_mode)) {
4042 * Dirs can't be linked, so move it. For moved
4043 * dirs, we always have one new and one deleted
4044 * ref. The deleted ref is ignored later.
4046 ret = send_rename(sctx, valid_path,
4049 ret = fs_path_copy(valid_path,
4055 * We might have previously orphanized an inode
4056 * which is an ancestor of our current inode,
4057 * so our reference's full path, which was
4058 * computed before any such orphanizations, must
4061 if (orphanized_dir) {
4062 ret = update_ref_path(sctx, cur);
4066 ret = send_link(sctx, cur->full_path,
4072 ret = dup_ref(cur, &check_dirs);
4077 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4079 * Check if we can already rmdir the directory. If not,
4080 * orphanize it. For every dir item inside that gets deleted
4081 * later, we do this check again and rmdir it then if possible.
4082 * See the use of check_dirs for more details.
4084 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4089 ret = send_rmdir(sctx, valid_path);
4092 } else if (!is_orphan) {
4093 ret = orphanize_inode(sctx, sctx->cur_ino,
4094 sctx->cur_inode_gen, valid_path);
4100 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4101 ret = dup_ref(cur, &check_dirs);
4105 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4106 !list_empty(&sctx->deleted_refs)) {
4108 * We have a moved dir. Add the old parent to check_dirs
4110 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4112 ret = dup_ref(cur, &check_dirs);
4115 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4117 * We have a non dir inode. Go through all deleted refs and
4118 * unlink them if they were not already overwritten by other
4121 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4122 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4123 sctx->cur_ino, sctx->cur_inode_gen,
4124 cur->name, cur->name_len);
4129 * If we orphanized any ancestor before, we need
4130 * to recompute the full path for deleted names,
4131 * since any such path was computed before we
4132 * processed any references and orphanized any
4135 if (orphanized_ancestor) {
4136 ret = update_ref_path(sctx, cur);
4140 ret = send_unlink(sctx, cur->full_path);
4144 ret = dup_ref(cur, &check_dirs);
4149 * If the inode is still orphan, unlink the orphan. This may
4150 * happen when a previous inode did overwrite the first ref
4151 * of this inode and no new refs were added for the current
4152 * inode. Unlinking does not mean that the inode is deleted in
4153 * all cases. There may still be links to this inode in other
4157 ret = send_unlink(sctx, valid_path);
4164 * We did collect all parent dirs where cur_inode was once located. We
4165 * now go through all these dirs and check if they are pending for
4166 * deletion and if it's finally possible to perform the rmdir now.
4167 * We also update the inode stats of the parent dirs here.
4169 list_for_each_entry(cur, &check_dirs, list) {
4171 * In case we had refs into dirs that were not processed yet,
4172 * we don't need to do the utime and rmdir logic for these dirs.
4173 * The dir will be processed later.
4175 if (cur->dir > sctx->cur_ino)
4178 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4182 if (ret == inode_state_did_create ||
4183 ret == inode_state_no_change) {
4184 /* TODO delayed utimes */
4185 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4188 } else if (ret == inode_state_did_delete &&
4189 cur->dir != last_dir_ino_rm) {
4190 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4195 ret = get_cur_path(sctx, cur->dir,
4196 cur->dir_gen, valid_path);
4199 ret = send_rmdir(sctx, valid_path);
4202 last_dir_ino_rm = cur->dir;
4210 __free_recorded_refs(&check_dirs);
4211 free_recorded_refs(sctx);
4212 fs_path_free(valid_path);
4216 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4217 void *ctx, struct list_head *refs)
4220 struct send_ctx *sctx = ctx;
4224 p = fs_path_alloc();
4228 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4233 ret = get_cur_path(sctx, dir, gen, p);
4236 ret = fs_path_add_path(p, name);
4240 ret = __record_ref(refs, dir, gen, p);
4248 static int __record_new_ref(int num, u64 dir, int index,
4249 struct fs_path *name,
4252 struct send_ctx *sctx = ctx;
4253 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4257 static int __record_deleted_ref(int num, u64 dir, int index,
4258 struct fs_path *name,
4261 struct send_ctx *sctx = ctx;
4262 return record_ref(sctx->parent_root, dir, name, ctx,
4263 &sctx->deleted_refs);
4266 static int record_new_ref(struct send_ctx *sctx)
4270 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4271 sctx->cmp_key, 0, __record_new_ref, sctx);
4280 static int record_deleted_ref(struct send_ctx *sctx)
4284 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4285 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4294 struct find_ref_ctx {
4297 struct btrfs_root *root;
4298 struct fs_path *name;
4302 static int __find_iref(int num, u64 dir, int index,
4303 struct fs_path *name,
4306 struct find_ref_ctx *ctx = ctx_;
4310 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4311 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4313 * To avoid doing extra lookups we'll only do this if everything
4316 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4320 if (dir_gen != ctx->dir_gen)
4322 ctx->found_idx = num;
4328 static int find_iref(struct btrfs_root *root,
4329 struct btrfs_path *path,
4330 struct btrfs_key *key,
4331 u64 dir, u64 dir_gen, struct fs_path *name)
4334 struct find_ref_ctx ctx;
4338 ctx.dir_gen = dir_gen;
4342 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4346 if (ctx.found_idx == -1)
4349 return ctx.found_idx;
4352 static int __record_changed_new_ref(int num, u64 dir, int index,
4353 struct fs_path *name,
4358 struct send_ctx *sctx = ctx;
4360 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4365 ret = find_iref(sctx->parent_root, sctx->right_path,
4366 sctx->cmp_key, dir, dir_gen, name);
4368 ret = __record_new_ref(num, dir, index, name, sctx);
4375 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4376 struct fs_path *name,
4381 struct send_ctx *sctx = ctx;
4383 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4388 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4389 dir, dir_gen, name);
4391 ret = __record_deleted_ref(num, dir, index, name, sctx);
4398 static int record_changed_ref(struct send_ctx *sctx)
4402 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4403 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4406 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4407 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4417 * Record and process all refs at once. Needed when an inode changes the
4418 * generation number, which means that it was deleted and recreated.
4420 static int process_all_refs(struct send_ctx *sctx,
4421 enum btrfs_compare_tree_result cmd)
4424 struct btrfs_root *root;
4425 struct btrfs_path *path;
4426 struct btrfs_key key;
4427 struct btrfs_key found_key;
4428 struct extent_buffer *eb;
4430 iterate_inode_ref_t cb;
4431 int pending_move = 0;
4433 path = alloc_path_for_send();
4437 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4438 root = sctx->send_root;
4439 cb = __record_new_ref;
4440 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4441 root = sctx->parent_root;
4442 cb = __record_deleted_ref;
4444 btrfs_err(sctx->send_root->fs_info,
4445 "Wrong command %d in process_all_refs", cmd);
4450 key.objectid = sctx->cmp_key->objectid;
4451 key.type = BTRFS_INODE_REF_KEY;
4453 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4458 eb = path->nodes[0];
4459 slot = path->slots[0];
4460 if (slot >= btrfs_header_nritems(eb)) {
4461 ret = btrfs_next_leaf(root, path);
4469 btrfs_item_key_to_cpu(eb, &found_key, slot);
4471 if (found_key.objectid != key.objectid ||
4472 (found_key.type != BTRFS_INODE_REF_KEY &&
4473 found_key.type != BTRFS_INODE_EXTREF_KEY))
4476 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4482 btrfs_release_path(path);
4485 * We don't actually care about pending_move as we are simply
4486 * re-creating this inode and will be rename'ing it into place once we
4487 * rename the parent directory.
4489 ret = process_recorded_refs(sctx, &pending_move);
4491 btrfs_free_path(path);
4495 static int send_set_xattr(struct send_ctx *sctx,
4496 struct fs_path *path,
4497 const char *name, int name_len,
4498 const char *data, int data_len)
4502 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4506 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4507 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4508 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4510 ret = send_cmd(sctx);
4517 static int send_remove_xattr(struct send_ctx *sctx,
4518 struct fs_path *path,
4519 const char *name, int name_len)
4523 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4527 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4528 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4530 ret = send_cmd(sctx);
4537 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4538 const char *name, int name_len,
4539 const char *data, int data_len,
4543 struct send_ctx *sctx = ctx;
4545 struct posix_acl_xattr_header dummy_acl;
4547 p = fs_path_alloc();
4552 * This hack is needed because empty acls are stored as zero byte
4553 * data in xattrs. Problem with that is, that receiving these zero byte
4554 * acls will fail later. To fix this, we send a dummy acl list that
4555 * only contains the version number and no entries.
4557 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4558 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4559 if (data_len == 0) {
4560 dummy_acl.a_version =
4561 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4562 data = (char *)&dummy_acl;
4563 data_len = sizeof(dummy_acl);
4567 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4571 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4578 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4579 const char *name, int name_len,
4580 const char *data, int data_len,
4584 struct send_ctx *sctx = ctx;
4587 p = fs_path_alloc();
4591 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4595 ret = send_remove_xattr(sctx, p, name, name_len);
4602 static int process_new_xattr(struct send_ctx *sctx)
4606 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4607 __process_new_xattr, sctx);
4612 static int process_deleted_xattr(struct send_ctx *sctx)
4614 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4615 __process_deleted_xattr, sctx);
4618 struct find_xattr_ctx {
4626 static int __find_xattr(int num, struct btrfs_key *di_key,
4627 const char *name, int name_len,
4628 const char *data, int data_len,
4629 u8 type, void *vctx)
4631 struct find_xattr_ctx *ctx = vctx;
4633 if (name_len == ctx->name_len &&
4634 strncmp(name, ctx->name, name_len) == 0) {
4635 ctx->found_idx = num;
4636 ctx->found_data_len = data_len;
4637 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4638 if (!ctx->found_data)
4645 static int find_xattr(struct btrfs_root *root,
4646 struct btrfs_path *path,
4647 struct btrfs_key *key,
4648 const char *name, int name_len,
4649 char **data, int *data_len)
4652 struct find_xattr_ctx ctx;
4655 ctx.name_len = name_len;
4657 ctx.found_data = NULL;
4658 ctx.found_data_len = 0;
4660 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4664 if (ctx.found_idx == -1)
4667 *data = ctx.found_data;
4668 *data_len = ctx.found_data_len;
4670 kfree(ctx.found_data);
4672 return ctx.found_idx;
4676 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4677 const char *name, int name_len,
4678 const char *data, int data_len,
4682 struct send_ctx *sctx = ctx;
4683 char *found_data = NULL;
4684 int found_data_len = 0;
4686 ret = find_xattr(sctx->parent_root, sctx->right_path,
4687 sctx->cmp_key, name, name_len, &found_data,
4689 if (ret == -ENOENT) {
4690 ret = __process_new_xattr(num, di_key, name, name_len, data,
4691 data_len, type, ctx);
4692 } else if (ret >= 0) {
4693 if (data_len != found_data_len ||
4694 memcmp(data, found_data, data_len)) {
4695 ret = __process_new_xattr(num, di_key, name, name_len,
4696 data, data_len, type, ctx);
4706 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4707 const char *name, int name_len,
4708 const char *data, int data_len,
4712 struct send_ctx *sctx = ctx;
4714 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4715 name, name_len, NULL, NULL);
4717 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4718 data_len, type, ctx);
4725 static int process_changed_xattr(struct send_ctx *sctx)
4729 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4730 __process_changed_new_xattr, sctx);
4733 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4734 __process_changed_deleted_xattr, sctx);
4740 static int process_all_new_xattrs(struct send_ctx *sctx)
4743 struct btrfs_root *root;
4744 struct btrfs_path *path;
4745 struct btrfs_key key;
4746 struct btrfs_key found_key;
4747 struct extent_buffer *eb;
4750 path = alloc_path_for_send();
4754 root = sctx->send_root;
4756 key.objectid = sctx->cmp_key->objectid;
4757 key.type = BTRFS_XATTR_ITEM_KEY;
4759 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4764 eb = path->nodes[0];
4765 slot = path->slots[0];
4766 if (slot >= btrfs_header_nritems(eb)) {
4767 ret = btrfs_next_leaf(root, path);
4770 } else if (ret > 0) {
4777 btrfs_item_key_to_cpu(eb, &found_key, slot);
4778 if (found_key.objectid != key.objectid ||
4779 found_key.type != key.type) {
4784 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4792 btrfs_free_path(path);
4796 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4798 struct btrfs_root *root = sctx->send_root;
4799 struct btrfs_fs_info *fs_info = root->fs_info;
4800 struct inode *inode;
4803 struct btrfs_key key;
4804 pgoff_t index = offset >> PAGE_SHIFT;
4806 unsigned pg_offset = offset_in_page(offset);
4809 key.objectid = sctx->cur_ino;
4810 key.type = BTRFS_INODE_ITEM_KEY;
4813 inode = btrfs_iget(fs_info->sb, &key, root);
4815 return PTR_ERR(inode);
4817 if (offset + len > i_size_read(inode)) {
4818 if (offset > i_size_read(inode))
4821 len = offset - i_size_read(inode);
4826 last_index = (offset + len - 1) >> PAGE_SHIFT;
4828 /* initial readahead */
4829 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4830 file_ra_state_init(&sctx->ra, inode->i_mapping);
4832 while (index <= last_index) {
4833 unsigned cur_len = min_t(unsigned, len,
4834 PAGE_SIZE - pg_offset);
4836 page = find_lock_page(inode->i_mapping, index);
4838 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4839 NULL, index, last_index + 1 - index);
4841 page = find_or_create_page(inode->i_mapping, index,
4849 if (PageReadahead(page)) {
4850 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4851 NULL, page, index, last_index + 1 - index);
4854 if (!PageUptodate(page)) {
4855 btrfs_readpage(NULL, page);
4857 if (!PageUptodate(page)) {
4866 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4881 * Read some bytes from the current inode/file and send a write command to
4884 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4886 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4889 ssize_t num_read = 0;
4891 p = fs_path_alloc();
4895 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4897 num_read = fill_read_buf(sctx, offset, len);
4898 if (num_read <= 0) {
4904 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4908 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4912 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4913 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4914 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4916 ret = send_cmd(sctx);
4927 * Send a clone command to user space.
4929 static int send_clone(struct send_ctx *sctx,
4930 u64 offset, u32 len,
4931 struct clone_root *clone_root)
4937 btrfs_debug(sctx->send_root->fs_info,
4938 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4939 offset, len, clone_root->root->root_key.objectid,
4940 clone_root->ino, clone_root->offset);
4942 p = fs_path_alloc();
4946 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4950 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4954 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4955 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4956 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4958 if (clone_root->root == sctx->send_root) {
4959 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4960 &gen, NULL, NULL, NULL, NULL);
4963 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4965 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4971 * If the parent we're using has a received_uuid set then use that as
4972 * our clone source as that is what we will look for when doing a
4975 * This covers the case that we create a snapshot off of a received
4976 * subvolume and then use that as the parent and try to receive on a
4979 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4980 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4981 clone_root->root->root_item.received_uuid);
4983 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4984 clone_root->root->root_item.uuid);
4985 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4986 le64_to_cpu(clone_root->root->root_item.ctransid));
4987 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4988 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4989 clone_root->offset);
4991 ret = send_cmd(sctx);
5000 * Send an update extent command to user space.
5002 static int send_update_extent(struct send_ctx *sctx,
5003 u64 offset, u32 len)
5008 p = fs_path_alloc();
5012 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
5016 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5020 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5021 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5022 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5024 ret = send_cmd(sctx);
5032 static int send_hole(struct send_ctx *sctx, u64 end)
5034 struct fs_path *p = NULL;
5035 u64 offset = sctx->cur_inode_last_extent;
5040 * A hole that starts at EOF or beyond it. Since we do not yet support
5041 * fallocate (for extent preallocation and hole punching), sending a
5042 * write of zeroes starting at EOF or beyond would later require issuing
5043 * a truncate operation which would undo the write and achieve nothing.
5045 if (offset >= sctx->cur_inode_size)
5049 * Don't go beyond the inode's i_size due to prealloc extents that start
5052 end = min_t(u64, end, sctx->cur_inode_size);
5054 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5055 return send_update_extent(sctx, offset, end - offset);
5057 p = fs_path_alloc();
5060 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5062 goto tlv_put_failure;
5063 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5064 while (offset < end) {
5065 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5067 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5070 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5071 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5072 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5073 ret = send_cmd(sctx);
5078 sctx->cur_inode_next_write_offset = offset;
5084 static int send_extent_data(struct send_ctx *sctx,
5090 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5091 return send_update_extent(sctx, offset, len);
5093 while (sent < len) {
5094 u64 size = len - sent;
5097 if (size > BTRFS_SEND_READ_SIZE)
5098 size = BTRFS_SEND_READ_SIZE;
5099 ret = send_write(sctx, offset + sent, size);
5109 static int clone_range(struct send_ctx *sctx,
5110 struct clone_root *clone_root,
5111 const u64 disk_byte,
5116 struct btrfs_path *path;
5117 struct btrfs_key key;
5119 u64 clone_src_i_size = 0;
5122 * Prevent cloning from a zero offset with a length matching the sector
5123 * size because in some scenarios this will make the receiver fail.
5125 * For example, if in the source filesystem the extent at offset 0
5126 * has a length of sectorsize and it was written using direct IO, then
5127 * it can never be an inline extent (even if compression is enabled).
5128 * Then this extent can be cloned in the original filesystem to a non
5129 * zero file offset, but it may not be possible to clone in the
5130 * destination filesystem because it can be inlined due to compression
5131 * on the destination filesystem (as the receiver's write operations are
5132 * always done using buffered IO). The same happens when the original
5133 * filesystem does not have compression enabled but the destination
5136 if (clone_root->offset == 0 &&
5137 len == sctx->send_root->fs_info->sectorsize)
5138 return send_extent_data(sctx, offset, len);
5140 path = alloc_path_for_send();
5145 * There are inodes that have extents that lie behind its i_size. Don't
5146 * accept clones from these extents.
5148 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5149 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5150 btrfs_release_path(path);
5155 * We can't send a clone operation for the entire range if we find
5156 * extent items in the respective range in the source file that
5157 * refer to different extents or if we find holes.
5158 * So check for that and do a mix of clone and regular write/copy
5159 * operations if needed.
5163 * mkfs.btrfs -f /dev/sda
5164 * mount /dev/sda /mnt
5165 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5166 * cp --reflink=always /mnt/foo /mnt/bar
5167 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5168 * btrfs subvolume snapshot -r /mnt /mnt/snap
5170 * If when we send the snapshot and we are processing file bar (which
5171 * has a higher inode number than foo) we blindly send a clone operation
5172 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5173 * a file bar that matches the content of file foo - iow, doesn't match
5174 * the content from bar in the original filesystem.
5176 key.objectid = clone_root->ino;
5177 key.type = BTRFS_EXTENT_DATA_KEY;
5178 key.offset = clone_root->offset;
5179 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5182 if (ret > 0 && path->slots[0] > 0) {
5183 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5184 if (key.objectid == clone_root->ino &&
5185 key.type == BTRFS_EXTENT_DATA_KEY)
5190 struct extent_buffer *leaf = path->nodes[0];
5191 int slot = path->slots[0];
5192 struct btrfs_file_extent_item *ei;
5196 u64 clone_data_offset;
5198 if (slot >= btrfs_header_nritems(leaf)) {
5199 ret = btrfs_next_leaf(clone_root->root, path);
5207 btrfs_item_key_to_cpu(leaf, &key, slot);
5210 * We might have an implicit trailing hole (NO_HOLES feature
5211 * enabled). We deal with it after leaving this loop.
5213 if (key.objectid != clone_root->ino ||
5214 key.type != BTRFS_EXTENT_DATA_KEY)
5217 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5218 type = btrfs_file_extent_type(leaf, ei);
5219 if (type == BTRFS_FILE_EXTENT_INLINE) {
5220 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5221 ext_len = PAGE_ALIGN(ext_len);
5223 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5226 if (key.offset + ext_len <= clone_root->offset)
5229 if (key.offset > clone_root->offset) {
5230 /* Implicit hole, NO_HOLES feature enabled. */
5231 u64 hole_len = key.offset - clone_root->offset;
5235 ret = send_extent_data(sctx, offset, hole_len);
5243 clone_root->offset += hole_len;
5244 data_offset += hole_len;
5247 if (key.offset >= clone_root->offset + len)
5250 if (key.offset >= clone_src_i_size)
5253 if (key.offset + ext_len > clone_src_i_size)
5254 ext_len = clone_src_i_size - key.offset;
5256 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5257 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5258 clone_root->offset = key.offset;
5259 if (clone_data_offset < data_offset &&
5260 clone_data_offset + ext_len > data_offset) {
5263 extent_offset = data_offset - clone_data_offset;
5264 ext_len -= extent_offset;
5265 clone_data_offset += extent_offset;
5266 clone_root->offset += extent_offset;
5270 clone_len = min_t(u64, ext_len, len);
5272 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5273 clone_data_offset == data_offset) {
5274 const u64 src_end = clone_root->offset + clone_len;
5275 const u64 sectorsize = SZ_64K;
5278 * We can't clone the last block, when its size is not
5279 * sector size aligned, into the middle of a file. If we
5280 * do so, the receiver will get a failure (-EINVAL) when
5281 * trying to clone or will silently corrupt the data in
5282 * the destination file if it's on a kernel without the
5283 * fix introduced by commit ac765f83f1397646
5284 * ("Btrfs: fix data corruption due to cloning of eof
5287 * So issue a clone of the aligned down range plus a
5288 * regular write for the eof block, if we hit that case.
5290 * Also, we use the maximum possible sector size, 64K,
5291 * because we don't know what's the sector size of the
5292 * filesystem that receives the stream, so we have to
5293 * assume the largest possible sector size.
5295 if (src_end == clone_src_i_size &&
5296 !IS_ALIGNED(src_end, sectorsize) &&
5297 offset + clone_len < sctx->cur_inode_size) {
5300 slen = ALIGN_DOWN(src_end - clone_root->offset,
5303 ret = send_clone(sctx, offset, slen,
5308 ret = send_extent_data(sctx, offset + slen,
5311 ret = send_clone(sctx, offset, clone_len,
5315 ret = send_extent_data(sctx, offset, clone_len);
5324 offset += clone_len;
5325 clone_root->offset += clone_len;
5326 data_offset += clone_len;
5332 ret = send_extent_data(sctx, offset, len);
5336 btrfs_free_path(path);
5340 static int send_write_or_clone(struct send_ctx *sctx,
5341 struct btrfs_path *path,
5342 struct btrfs_key *key,
5343 struct clone_root *clone_root)
5346 struct btrfs_file_extent_item *ei;
5347 u64 offset = key->offset;
5350 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5352 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5353 struct btrfs_file_extent_item);
5354 type = btrfs_file_extent_type(path->nodes[0], ei);
5355 if (type == BTRFS_FILE_EXTENT_INLINE) {
5356 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5358 * it is possible the inline item won't cover the whole page,
5359 * but there may be items after this page. Make
5360 * sure to send the whole thing
5362 len = PAGE_ALIGN(len);
5364 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5367 if (offset >= sctx->cur_inode_size) {
5371 if (offset + len > sctx->cur_inode_size)
5372 len = sctx->cur_inode_size - offset;
5378 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5382 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5383 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5384 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5387 ret = send_extent_data(sctx, offset, len);
5389 sctx->cur_inode_next_write_offset = offset + len;
5394 static int is_extent_unchanged(struct send_ctx *sctx,
5395 struct btrfs_path *left_path,
5396 struct btrfs_key *ekey)
5399 struct btrfs_key key;
5400 struct btrfs_path *path = NULL;
5401 struct extent_buffer *eb;
5403 struct btrfs_key found_key;
5404 struct btrfs_file_extent_item *ei;
5409 u64 left_offset_fixed;
5417 path = alloc_path_for_send();
5421 eb = left_path->nodes[0];
5422 slot = left_path->slots[0];
5423 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5424 left_type = btrfs_file_extent_type(eb, ei);
5426 if (left_type != BTRFS_FILE_EXTENT_REG) {
5430 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5431 left_len = btrfs_file_extent_num_bytes(eb, ei);
5432 left_offset = btrfs_file_extent_offset(eb, ei);
5433 left_gen = btrfs_file_extent_generation(eb, ei);
5436 * Following comments will refer to these graphics. L is the left
5437 * extents which we are checking at the moment. 1-8 are the right
5438 * extents that we iterate.
5441 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5444 * |--1--|-2b-|...(same as above)
5446 * Alternative situation. Happens on files where extents got split.
5448 * |-----------7-----------|-6-|
5450 * Alternative situation. Happens on files which got larger.
5453 * Nothing follows after 8.
5456 key.objectid = ekey->objectid;
5457 key.type = BTRFS_EXTENT_DATA_KEY;
5458 key.offset = ekey->offset;
5459 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5468 * Handle special case where the right side has no extents at all.
5470 eb = path->nodes[0];
5471 slot = path->slots[0];
5472 btrfs_item_key_to_cpu(eb, &found_key, slot);
5473 if (found_key.objectid != key.objectid ||
5474 found_key.type != key.type) {
5475 /* If we're a hole then just pretend nothing changed */
5476 ret = (left_disknr) ? 0 : 1;
5481 * We're now on 2a, 2b or 7.
5484 while (key.offset < ekey->offset + left_len) {
5485 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5486 right_type = btrfs_file_extent_type(eb, ei);
5487 if (right_type != BTRFS_FILE_EXTENT_REG &&
5488 right_type != BTRFS_FILE_EXTENT_INLINE) {
5493 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5494 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5495 right_len = PAGE_ALIGN(right_len);
5497 right_len = btrfs_file_extent_num_bytes(eb, ei);
5501 * Are we at extent 8? If yes, we know the extent is changed.
5502 * This may only happen on the first iteration.
5504 if (found_key.offset + right_len <= ekey->offset) {
5505 /* If we're a hole just pretend nothing changed */
5506 ret = (left_disknr) ? 0 : 1;
5511 * We just wanted to see if when we have an inline extent, what
5512 * follows it is a regular extent (wanted to check the above
5513 * condition for inline extents too). This should normally not
5514 * happen but it's possible for example when we have an inline
5515 * compressed extent representing data with a size matching
5516 * the page size (currently the same as sector size).
5518 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5523 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5524 right_offset = btrfs_file_extent_offset(eb, ei);
5525 right_gen = btrfs_file_extent_generation(eb, ei);
5527 left_offset_fixed = left_offset;
5528 if (key.offset < ekey->offset) {
5529 /* Fix the right offset for 2a and 7. */
5530 right_offset += ekey->offset - key.offset;
5532 /* Fix the left offset for all behind 2a and 2b */
5533 left_offset_fixed += key.offset - ekey->offset;
5537 * Check if we have the same extent.
5539 if (left_disknr != right_disknr ||
5540 left_offset_fixed != right_offset ||
5541 left_gen != right_gen) {
5547 * Go to the next extent.
5549 ret = btrfs_next_item(sctx->parent_root, path);
5553 eb = path->nodes[0];
5554 slot = path->slots[0];
5555 btrfs_item_key_to_cpu(eb, &found_key, slot);
5557 if (ret || found_key.objectid != key.objectid ||
5558 found_key.type != key.type) {
5559 key.offset += right_len;
5562 if (found_key.offset != key.offset + right_len) {
5570 * We're now behind the left extent (treat as unchanged) or at the end
5571 * of the right side (treat as changed).
5573 if (key.offset >= ekey->offset + left_len)
5580 btrfs_free_path(path);
5584 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5586 struct btrfs_path *path;
5587 struct btrfs_root *root = sctx->send_root;
5588 struct btrfs_file_extent_item *fi;
5589 struct btrfs_key key;
5594 path = alloc_path_for_send();
5598 sctx->cur_inode_last_extent = 0;
5600 key.objectid = sctx->cur_ino;
5601 key.type = BTRFS_EXTENT_DATA_KEY;
5602 key.offset = offset;
5603 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5607 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5608 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5611 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5612 struct btrfs_file_extent_item);
5613 type = btrfs_file_extent_type(path->nodes[0], fi);
5614 if (type == BTRFS_FILE_EXTENT_INLINE) {
5615 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5616 extent_end = ALIGN(key.offset + size,
5617 sctx->send_root->fs_info->sectorsize);
5619 extent_end = key.offset +
5620 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5622 sctx->cur_inode_last_extent = extent_end;
5624 btrfs_free_path(path);
5628 static int range_is_hole_in_parent(struct send_ctx *sctx,
5632 struct btrfs_path *path;
5633 struct btrfs_key key;
5634 struct btrfs_root *root = sctx->parent_root;
5635 u64 search_start = start;
5638 path = alloc_path_for_send();
5642 key.objectid = sctx->cur_ino;
5643 key.type = BTRFS_EXTENT_DATA_KEY;
5644 key.offset = search_start;
5645 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5648 if (ret > 0 && path->slots[0] > 0)
5651 while (search_start < end) {
5652 struct extent_buffer *leaf = path->nodes[0];
5653 int slot = path->slots[0];
5654 struct btrfs_file_extent_item *fi;
5657 if (slot >= btrfs_header_nritems(leaf)) {
5658 ret = btrfs_next_leaf(root, path);
5666 btrfs_item_key_to_cpu(leaf, &key, slot);
5667 if (key.objectid < sctx->cur_ino ||
5668 key.type < BTRFS_EXTENT_DATA_KEY)
5670 if (key.objectid > sctx->cur_ino ||
5671 key.type > BTRFS_EXTENT_DATA_KEY ||
5675 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5676 if (btrfs_file_extent_type(leaf, fi) ==
5677 BTRFS_FILE_EXTENT_INLINE) {
5678 u64 size = btrfs_file_extent_ram_bytes(leaf, fi);
5680 extent_end = ALIGN(key.offset + size,
5681 root->fs_info->sectorsize);
5683 extent_end = key.offset +
5684 btrfs_file_extent_num_bytes(leaf, fi);
5686 if (extent_end <= start)
5688 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5689 search_start = extent_end;
5699 btrfs_free_path(path);
5703 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5704 struct btrfs_key *key)
5706 struct btrfs_file_extent_item *fi;
5711 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5714 if (sctx->cur_inode_last_extent == (u64)-1) {
5715 ret = get_last_extent(sctx, key->offset - 1);
5720 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5721 struct btrfs_file_extent_item);
5722 type = btrfs_file_extent_type(path->nodes[0], fi);
5723 if (type == BTRFS_FILE_EXTENT_INLINE) {
5724 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5725 extent_end = ALIGN(key->offset + size,
5726 sctx->send_root->fs_info->sectorsize);
5728 extent_end = key->offset +
5729 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5732 if (path->slots[0] == 0 &&
5733 sctx->cur_inode_last_extent < key->offset) {
5735 * We might have skipped entire leafs that contained only
5736 * file extent items for our current inode. These leafs have
5737 * a generation number smaller (older) than the one in the
5738 * current leaf and the leaf our last extent came from, and
5739 * are located between these 2 leafs.
5741 ret = get_last_extent(sctx, key->offset - 1);
5746 if (sctx->cur_inode_last_extent < key->offset) {
5747 ret = range_is_hole_in_parent(sctx,
5748 sctx->cur_inode_last_extent,
5753 ret = send_hole(sctx, key->offset);
5757 sctx->cur_inode_last_extent = extent_end;
5761 static int process_extent(struct send_ctx *sctx,
5762 struct btrfs_path *path,
5763 struct btrfs_key *key)
5765 struct clone_root *found_clone = NULL;
5768 if (S_ISLNK(sctx->cur_inode_mode))
5771 if (sctx->parent_root && !sctx->cur_inode_new) {
5772 ret = is_extent_unchanged(sctx, path, key);
5780 struct btrfs_file_extent_item *ei;
5783 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5784 struct btrfs_file_extent_item);
5785 type = btrfs_file_extent_type(path->nodes[0], ei);
5786 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5787 type == BTRFS_FILE_EXTENT_REG) {
5789 * The send spec does not have a prealloc command yet,
5790 * so just leave a hole for prealloc'ed extents until
5791 * we have enough commands queued up to justify rev'ing
5794 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5799 /* Have a hole, just skip it. */
5800 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5807 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5808 sctx->cur_inode_size, &found_clone);
5809 if (ret != -ENOENT && ret < 0)
5812 ret = send_write_or_clone(sctx, path, key, found_clone);
5816 ret = maybe_send_hole(sctx, path, key);
5821 static int process_all_extents(struct send_ctx *sctx)
5824 struct btrfs_root *root;
5825 struct btrfs_path *path;
5826 struct btrfs_key key;
5827 struct btrfs_key found_key;
5828 struct extent_buffer *eb;
5831 root = sctx->send_root;
5832 path = alloc_path_for_send();
5836 key.objectid = sctx->cmp_key->objectid;
5837 key.type = BTRFS_EXTENT_DATA_KEY;
5839 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5844 eb = path->nodes[0];
5845 slot = path->slots[0];
5847 if (slot >= btrfs_header_nritems(eb)) {
5848 ret = btrfs_next_leaf(root, path);
5851 } else if (ret > 0) {
5858 btrfs_item_key_to_cpu(eb, &found_key, slot);
5860 if (found_key.objectid != key.objectid ||
5861 found_key.type != key.type) {
5866 ret = process_extent(sctx, path, &found_key);
5874 btrfs_free_path(path);
5878 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5880 int *refs_processed)
5884 if (sctx->cur_ino == 0)
5886 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5887 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5889 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5892 ret = process_recorded_refs(sctx, pending_move);
5896 *refs_processed = 1;
5901 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5912 int need_truncate = 1;
5913 int pending_move = 0;
5914 int refs_processed = 0;
5916 if (sctx->ignore_cur_inode)
5919 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5925 * We have processed the refs and thus need to advance send_progress.
5926 * Now, calls to get_cur_xxx will take the updated refs of the current
5927 * inode into account.
5929 * On the other hand, if our current inode is a directory and couldn't
5930 * be moved/renamed because its parent was renamed/moved too and it has
5931 * a higher inode number, we can only move/rename our current inode
5932 * after we moved/renamed its parent. Therefore in this case operate on
5933 * the old path (pre move/rename) of our current inode, and the
5934 * move/rename will be performed later.
5936 if (refs_processed && !pending_move)
5937 sctx->send_progress = sctx->cur_ino + 1;
5939 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5941 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5944 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5945 &left_mode, &left_uid, &left_gid, NULL);
5949 if (!sctx->parent_root || sctx->cur_inode_new) {
5951 if (!S_ISLNK(sctx->cur_inode_mode))
5953 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5958 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5959 &old_size, NULL, &right_mode, &right_uid,
5964 if (left_uid != right_uid || left_gid != right_gid)
5966 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5968 if ((old_size == sctx->cur_inode_size) ||
5969 (sctx->cur_inode_size > old_size &&
5970 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5974 if (S_ISREG(sctx->cur_inode_mode)) {
5975 if (need_send_hole(sctx)) {
5976 if (sctx->cur_inode_last_extent == (u64)-1 ||
5977 sctx->cur_inode_last_extent <
5978 sctx->cur_inode_size) {
5979 ret = get_last_extent(sctx, (u64)-1);
5983 if (sctx->cur_inode_last_extent <
5984 sctx->cur_inode_size) {
5985 ret = send_hole(sctx, sctx->cur_inode_size);
5990 if (need_truncate) {
5991 ret = send_truncate(sctx, sctx->cur_ino,
5992 sctx->cur_inode_gen,
5993 sctx->cur_inode_size);
6000 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6001 left_uid, left_gid);
6006 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6013 * If other directory inodes depended on our current directory
6014 * inode's move/rename, now do their move/rename operations.
6016 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
6017 ret = apply_children_dir_moves(sctx);
6021 * Need to send that every time, no matter if it actually
6022 * changed between the two trees as we have done changes to
6023 * the inode before. If our inode is a directory and it's
6024 * waiting to be moved/renamed, we will send its utimes when
6025 * it's moved/renamed, therefore we don't need to do it here.
6027 sctx->send_progress = sctx->cur_ino + 1;
6028 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6037 struct parent_paths_ctx {
6038 struct list_head *refs;
6039 struct send_ctx *sctx;
6042 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
6045 struct parent_paths_ctx *ppctx = ctx;
6047 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6052 * Issue unlink operations for all paths of the current inode found in the
6055 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6057 LIST_HEAD(deleted_refs);
6058 struct btrfs_path *path;
6059 struct btrfs_key key;
6060 struct parent_paths_ctx ctx;
6063 path = alloc_path_for_send();
6067 key.objectid = sctx->cur_ino;
6068 key.type = BTRFS_INODE_REF_KEY;
6070 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6074 ctx.refs = &deleted_refs;
6078 struct extent_buffer *eb = path->nodes[0];
6079 int slot = path->slots[0];
6081 if (slot >= btrfs_header_nritems(eb)) {
6082 ret = btrfs_next_leaf(sctx->parent_root, path);
6090 btrfs_item_key_to_cpu(eb, &key, slot);
6091 if (key.objectid != sctx->cur_ino)
6093 if (key.type != BTRFS_INODE_REF_KEY &&
6094 key.type != BTRFS_INODE_EXTREF_KEY)
6097 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6098 record_parent_ref, &ctx);
6105 while (!list_empty(&deleted_refs)) {
6106 struct recorded_ref *ref;
6108 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6109 ret = send_unlink(sctx, ref->full_path);
6112 fs_path_free(ref->full_path);
6113 list_del(&ref->list);
6118 btrfs_free_path(path);
6120 __free_recorded_refs(&deleted_refs);
6124 static int changed_inode(struct send_ctx *sctx,
6125 enum btrfs_compare_tree_result result)
6128 struct btrfs_key *key = sctx->cmp_key;
6129 struct btrfs_inode_item *left_ii = NULL;
6130 struct btrfs_inode_item *right_ii = NULL;
6134 sctx->cur_ino = key->objectid;
6135 sctx->cur_inode_new_gen = 0;
6136 sctx->cur_inode_last_extent = (u64)-1;
6137 sctx->cur_inode_next_write_offset = 0;
6138 sctx->ignore_cur_inode = false;
6141 * Set send_progress to current inode. This will tell all get_cur_xxx
6142 * functions that the current inode's refs are not updated yet. Later,
6143 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6145 sctx->send_progress = sctx->cur_ino;
6147 if (result == BTRFS_COMPARE_TREE_NEW ||
6148 result == BTRFS_COMPARE_TREE_CHANGED) {
6149 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6150 sctx->left_path->slots[0],
6151 struct btrfs_inode_item);
6152 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6155 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6156 sctx->right_path->slots[0],
6157 struct btrfs_inode_item);
6158 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6161 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6162 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6163 sctx->right_path->slots[0],
6164 struct btrfs_inode_item);
6166 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6170 * The cur_ino = root dir case is special here. We can't treat
6171 * the inode as deleted+reused because it would generate a
6172 * stream that tries to delete/mkdir the root dir.
6174 if (left_gen != right_gen &&
6175 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6176 sctx->cur_inode_new_gen = 1;
6180 * Normally we do not find inodes with a link count of zero (orphans)
6181 * because the most common case is to create a snapshot and use it
6182 * for a send operation. However other less common use cases involve
6183 * using a subvolume and send it after turning it to RO mode just
6184 * after deleting all hard links of a file while holding an open
6185 * file descriptor against it or turning a RO snapshot into RW mode,
6186 * keep an open file descriptor against a file, delete it and then
6187 * turn the snapshot back to RO mode before using it for a send
6188 * operation. So if we find such cases, ignore the inode and all its
6189 * items completely if it's a new inode, or if it's a changed inode
6190 * make sure all its previous paths (from the parent snapshot) are all
6191 * unlinked and all other the inode items are ignored.
6193 if (result == BTRFS_COMPARE_TREE_NEW ||
6194 result == BTRFS_COMPARE_TREE_CHANGED) {
6197 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6199 sctx->ignore_cur_inode = true;
6200 if (result == BTRFS_COMPARE_TREE_CHANGED)
6201 ret = btrfs_unlink_all_paths(sctx);
6206 if (result == BTRFS_COMPARE_TREE_NEW) {
6207 sctx->cur_inode_gen = left_gen;
6208 sctx->cur_inode_new = 1;
6209 sctx->cur_inode_deleted = 0;
6210 sctx->cur_inode_size = btrfs_inode_size(
6211 sctx->left_path->nodes[0], left_ii);
6212 sctx->cur_inode_mode = btrfs_inode_mode(
6213 sctx->left_path->nodes[0], left_ii);
6214 sctx->cur_inode_rdev = btrfs_inode_rdev(
6215 sctx->left_path->nodes[0], left_ii);
6216 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6217 ret = send_create_inode_if_needed(sctx);
6218 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6219 sctx->cur_inode_gen = right_gen;
6220 sctx->cur_inode_new = 0;
6221 sctx->cur_inode_deleted = 1;
6222 sctx->cur_inode_size = btrfs_inode_size(
6223 sctx->right_path->nodes[0], right_ii);
6224 sctx->cur_inode_mode = btrfs_inode_mode(
6225 sctx->right_path->nodes[0], right_ii);
6226 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6228 * We need to do some special handling in case the inode was
6229 * reported as changed with a changed generation number. This
6230 * means that the original inode was deleted and new inode
6231 * reused the same inum. So we have to treat the old inode as
6232 * deleted and the new one as new.
6234 if (sctx->cur_inode_new_gen) {
6236 * First, process the inode as if it was deleted.
6238 sctx->cur_inode_gen = right_gen;
6239 sctx->cur_inode_new = 0;
6240 sctx->cur_inode_deleted = 1;
6241 sctx->cur_inode_size = btrfs_inode_size(
6242 sctx->right_path->nodes[0], right_ii);
6243 sctx->cur_inode_mode = btrfs_inode_mode(
6244 sctx->right_path->nodes[0], right_ii);
6245 ret = process_all_refs(sctx,
6246 BTRFS_COMPARE_TREE_DELETED);
6251 * Now process the inode as if it was new.
6253 sctx->cur_inode_gen = left_gen;
6254 sctx->cur_inode_new = 1;
6255 sctx->cur_inode_deleted = 0;
6256 sctx->cur_inode_size = btrfs_inode_size(
6257 sctx->left_path->nodes[0], left_ii);
6258 sctx->cur_inode_mode = btrfs_inode_mode(
6259 sctx->left_path->nodes[0], left_ii);
6260 sctx->cur_inode_rdev = btrfs_inode_rdev(
6261 sctx->left_path->nodes[0], left_ii);
6262 ret = send_create_inode_if_needed(sctx);
6266 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6270 * Advance send_progress now as we did not get into
6271 * process_recorded_refs_if_needed in the new_gen case.
6273 sctx->send_progress = sctx->cur_ino + 1;
6276 * Now process all extents and xattrs of the inode as if
6277 * they were all new.
6279 ret = process_all_extents(sctx);
6282 ret = process_all_new_xattrs(sctx);
6286 sctx->cur_inode_gen = left_gen;
6287 sctx->cur_inode_new = 0;
6288 sctx->cur_inode_new_gen = 0;
6289 sctx->cur_inode_deleted = 0;
6290 sctx->cur_inode_size = btrfs_inode_size(
6291 sctx->left_path->nodes[0], left_ii);
6292 sctx->cur_inode_mode = btrfs_inode_mode(
6293 sctx->left_path->nodes[0], left_ii);
6302 * We have to process new refs before deleted refs, but compare_trees gives us
6303 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6304 * first and later process them in process_recorded_refs.
6305 * For the cur_inode_new_gen case, we skip recording completely because
6306 * changed_inode did already initiate processing of refs. The reason for this is
6307 * that in this case, compare_tree actually compares the refs of 2 different
6308 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6309 * refs of the right tree as deleted and all refs of the left tree as new.
6311 static int changed_ref(struct send_ctx *sctx,
6312 enum btrfs_compare_tree_result result)
6316 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6317 inconsistent_snapshot_error(sctx, result, "reference");
6321 if (!sctx->cur_inode_new_gen &&
6322 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6323 if (result == BTRFS_COMPARE_TREE_NEW)
6324 ret = record_new_ref(sctx);
6325 else if (result == BTRFS_COMPARE_TREE_DELETED)
6326 ret = record_deleted_ref(sctx);
6327 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6328 ret = record_changed_ref(sctx);
6335 * Process new/deleted/changed xattrs. We skip processing in the
6336 * cur_inode_new_gen case because changed_inode did already initiate processing
6337 * of xattrs. The reason is the same as in changed_ref
6339 static int changed_xattr(struct send_ctx *sctx,
6340 enum btrfs_compare_tree_result result)
6344 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6345 inconsistent_snapshot_error(sctx, result, "xattr");
6349 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6350 if (result == BTRFS_COMPARE_TREE_NEW)
6351 ret = process_new_xattr(sctx);
6352 else if (result == BTRFS_COMPARE_TREE_DELETED)
6353 ret = process_deleted_xattr(sctx);
6354 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6355 ret = process_changed_xattr(sctx);
6362 * Process new/deleted/changed extents. We skip processing in the
6363 * cur_inode_new_gen case because changed_inode did already initiate processing
6364 * of extents. The reason is the same as in changed_ref
6366 static int changed_extent(struct send_ctx *sctx,
6367 enum btrfs_compare_tree_result result)
6372 * We have found an extent item that changed without the inode item
6373 * having changed. This can happen either after relocation (where the
6374 * disk_bytenr of an extent item is replaced at
6375 * relocation.c:replace_file_extents()) or after deduplication into a
6376 * file in both the parent and send snapshots (where an extent item can
6377 * get modified or replaced with a new one). Note that deduplication
6378 * updates the inode item, but it only changes the iversion (sequence
6379 * field in the inode item) of the inode, so if a file is deduplicated
6380 * the same amount of times in both the parent and send snapshots, its
6381 * iversion becames the same in both snapshots, whence the inode item is
6382 * the same on both snapshots.
6384 if (sctx->cur_ino != sctx->cmp_key->objectid)
6387 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6388 if (result != BTRFS_COMPARE_TREE_DELETED)
6389 ret = process_extent(sctx, sctx->left_path,
6396 static int dir_changed(struct send_ctx *sctx, u64 dir)
6398 u64 orig_gen, new_gen;
6401 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6406 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6411 return (orig_gen != new_gen) ? 1 : 0;
6414 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6415 struct btrfs_key *key)
6417 struct btrfs_inode_extref *extref;
6418 struct extent_buffer *leaf;
6419 u64 dirid = 0, last_dirid = 0;
6426 /* Easy case, just check this one dirid */
6427 if (key->type == BTRFS_INODE_REF_KEY) {
6428 dirid = key->offset;
6430 ret = dir_changed(sctx, dirid);
6434 leaf = path->nodes[0];
6435 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6436 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6437 while (cur_offset < item_size) {
6438 extref = (struct btrfs_inode_extref *)(ptr +
6440 dirid = btrfs_inode_extref_parent(leaf, extref);
6441 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6442 cur_offset += ref_name_len + sizeof(*extref);
6443 if (dirid == last_dirid)
6445 ret = dir_changed(sctx, dirid);
6455 * Updates compare related fields in sctx and simply forwards to the actual
6456 * changed_xxx functions.
6458 static int changed_cb(struct btrfs_path *left_path,
6459 struct btrfs_path *right_path,
6460 struct btrfs_key *key,
6461 enum btrfs_compare_tree_result result,
6465 struct send_ctx *sctx = ctx;
6467 if (result == BTRFS_COMPARE_TREE_SAME) {
6468 if (key->type == BTRFS_INODE_REF_KEY ||
6469 key->type == BTRFS_INODE_EXTREF_KEY) {
6470 ret = compare_refs(sctx, left_path, key);
6475 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6476 return maybe_send_hole(sctx, left_path, key);
6480 result = BTRFS_COMPARE_TREE_CHANGED;
6484 sctx->left_path = left_path;
6485 sctx->right_path = right_path;
6486 sctx->cmp_key = key;
6488 ret = finish_inode_if_needed(sctx, 0);
6492 /* Ignore non-FS objects */
6493 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6494 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6497 if (key->type == BTRFS_INODE_ITEM_KEY) {
6498 ret = changed_inode(sctx, result);
6499 } else if (!sctx->ignore_cur_inode) {
6500 if (key->type == BTRFS_INODE_REF_KEY ||
6501 key->type == BTRFS_INODE_EXTREF_KEY)
6502 ret = changed_ref(sctx, result);
6503 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6504 ret = changed_xattr(sctx, result);
6505 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6506 ret = changed_extent(sctx, result);
6513 static int full_send_tree(struct send_ctx *sctx)
6516 struct btrfs_root *send_root = sctx->send_root;
6517 struct btrfs_key key;
6518 struct btrfs_path *path;
6519 struct extent_buffer *eb;
6522 path = alloc_path_for_send();
6526 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6527 key.type = BTRFS_INODE_ITEM_KEY;
6530 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6537 eb = path->nodes[0];
6538 slot = path->slots[0];
6539 btrfs_item_key_to_cpu(eb, &key, slot);
6541 ret = changed_cb(path, NULL, &key,
6542 BTRFS_COMPARE_TREE_NEW, sctx);
6546 ret = btrfs_next_item(send_root, path);
6556 ret = finish_inode_if_needed(sctx, 1);
6559 btrfs_free_path(path);
6563 static int tree_move_down(struct btrfs_path *path, int *level)
6565 struct extent_buffer *eb;
6567 BUG_ON(*level == 0);
6568 eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]);
6572 path->nodes[*level - 1] = eb;
6573 path->slots[*level - 1] = 0;
6578 static int tree_move_next_or_upnext(struct btrfs_path *path,
6579 int *level, int root_level)
6583 nritems = btrfs_header_nritems(path->nodes[*level]);
6585 path->slots[*level]++;
6587 while (path->slots[*level] >= nritems) {
6588 if (*level == root_level)
6592 path->slots[*level] = 0;
6593 free_extent_buffer(path->nodes[*level]);
6594 path->nodes[*level] = NULL;
6596 path->slots[*level]++;
6598 nritems = btrfs_header_nritems(path->nodes[*level]);
6605 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
6608 static int tree_advance(struct btrfs_path *path,
6609 int *level, int root_level,
6611 struct btrfs_key *key)
6615 if (*level == 0 || !allow_down) {
6616 ret = tree_move_next_or_upnext(path, level, root_level);
6618 ret = tree_move_down(path, level);
6622 btrfs_item_key_to_cpu(path->nodes[*level], key,
6623 path->slots[*level]);
6625 btrfs_node_key_to_cpu(path->nodes[*level], key,
6626 path->slots[*level]);
6631 static int tree_compare_item(struct btrfs_path *left_path,
6632 struct btrfs_path *right_path,
6637 unsigned long off1, off2;
6639 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
6640 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
6644 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
6645 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
6646 right_path->slots[0]);
6648 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
6650 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
6657 * This function compares two trees and calls the provided callback for
6658 * every changed/new/deleted item it finds.
6659 * If shared tree blocks are encountered, whole subtrees are skipped, making
6660 * the compare pretty fast on snapshotted subvolumes.
6662 * This currently works on commit roots only. As commit roots are read only,
6663 * we don't do any locking. The commit roots are protected with transactions.
6664 * Transactions are ended and rejoined when a commit is tried in between.
6666 * This function checks for modifications done to the trees while comparing.
6667 * If it detects a change, it aborts immediately.
6669 static int btrfs_compare_trees(struct btrfs_root *left_root,
6670 struct btrfs_root *right_root,
6671 btrfs_changed_cb_t changed_cb, void *ctx)
6673 struct btrfs_fs_info *fs_info = left_root->fs_info;
6676 struct btrfs_path *left_path = NULL;
6677 struct btrfs_path *right_path = NULL;
6678 struct btrfs_key left_key;
6679 struct btrfs_key right_key;
6680 char *tmp_buf = NULL;
6681 int left_root_level;
6682 int right_root_level;
6685 int left_end_reached;
6686 int right_end_reached;
6694 left_path = btrfs_alloc_path();
6699 right_path = btrfs_alloc_path();
6705 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
6711 left_path->search_commit_root = 1;
6712 left_path->skip_locking = 1;
6713 right_path->search_commit_root = 1;
6714 right_path->skip_locking = 1;
6717 * Strategy: Go to the first items of both trees. Then do
6719 * If both trees are at level 0
6720 * Compare keys of current items
6721 * If left < right treat left item as new, advance left tree
6723 * If left > right treat right item as deleted, advance right tree
6725 * If left == right do deep compare of items, treat as changed if
6726 * needed, advance both trees and repeat
6727 * If both trees are at the same level but not at level 0
6728 * Compare keys of current nodes/leafs
6729 * If left < right advance left tree and repeat
6730 * If left > right advance right tree and repeat
6731 * If left == right compare blockptrs of the next nodes/leafs
6732 * If they match advance both trees but stay at the same level
6734 * If they don't match advance both trees while allowing to go
6736 * If tree levels are different
6737 * Advance the tree that needs it and repeat
6739 * Advancing a tree means:
6740 * If we are at level 0, try to go to the next slot. If that's not
6741 * possible, go one level up and repeat. Stop when we found a level
6742 * where we could go to the next slot. We may at this point be on a
6745 * If we are not at level 0 and not on shared tree blocks, go one
6748 * If we are not at level 0 and on shared tree blocks, go one slot to
6749 * the right if possible or go up and right.
6752 down_read(&fs_info->commit_root_sem);
6753 left_level = btrfs_header_level(left_root->commit_root);
6754 left_root_level = left_level;
6755 left_path->nodes[left_level] =
6756 btrfs_clone_extent_buffer(left_root->commit_root);
6757 if (!left_path->nodes[left_level]) {
6758 up_read(&fs_info->commit_root_sem);
6763 right_level = btrfs_header_level(right_root->commit_root);
6764 right_root_level = right_level;
6765 right_path->nodes[right_level] =
6766 btrfs_clone_extent_buffer(right_root->commit_root);
6767 if (!right_path->nodes[right_level]) {
6768 up_read(&fs_info->commit_root_sem);
6772 up_read(&fs_info->commit_root_sem);
6774 if (left_level == 0)
6775 btrfs_item_key_to_cpu(left_path->nodes[left_level],
6776 &left_key, left_path->slots[left_level]);
6778 btrfs_node_key_to_cpu(left_path->nodes[left_level],
6779 &left_key, left_path->slots[left_level]);
6780 if (right_level == 0)
6781 btrfs_item_key_to_cpu(right_path->nodes[right_level],
6782 &right_key, right_path->slots[right_level]);
6784 btrfs_node_key_to_cpu(right_path->nodes[right_level],
6785 &right_key, right_path->slots[right_level]);
6787 left_end_reached = right_end_reached = 0;
6788 advance_left = advance_right = 0;
6792 if (advance_left && !left_end_reached) {
6793 ret = tree_advance(left_path, &left_level,
6795 advance_left != ADVANCE_ONLY_NEXT,
6798 left_end_reached = ADVANCE;
6803 if (advance_right && !right_end_reached) {
6804 ret = tree_advance(right_path, &right_level,
6806 advance_right != ADVANCE_ONLY_NEXT,
6809 right_end_reached = ADVANCE;
6815 if (left_end_reached && right_end_reached) {
6818 } else if (left_end_reached) {
6819 if (right_level == 0) {
6820 ret = changed_cb(left_path, right_path,
6822 BTRFS_COMPARE_TREE_DELETED,
6827 advance_right = ADVANCE;
6829 } else if (right_end_reached) {
6830 if (left_level == 0) {
6831 ret = changed_cb(left_path, right_path,
6833 BTRFS_COMPARE_TREE_NEW,
6838 advance_left = ADVANCE;
6842 if (left_level == 0 && right_level == 0) {
6843 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6845 ret = changed_cb(left_path, right_path,
6847 BTRFS_COMPARE_TREE_NEW,
6851 advance_left = ADVANCE;
6852 } else if (cmp > 0) {
6853 ret = changed_cb(left_path, right_path,
6855 BTRFS_COMPARE_TREE_DELETED,
6859 advance_right = ADVANCE;
6861 enum btrfs_compare_tree_result result;
6863 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
6864 ret = tree_compare_item(left_path, right_path,
6867 result = BTRFS_COMPARE_TREE_CHANGED;
6869 result = BTRFS_COMPARE_TREE_SAME;
6870 ret = changed_cb(left_path, right_path,
6871 &left_key, result, ctx);
6874 advance_left = ADVANCE;
6875 advance_right = ADVANCE;
6877 } else if (left_level == right_level) {
6878 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6880 advance_left = ADVANCE;
6881 } else if (cmp > 0) {
6882 advance_right = ADVANCE;
6884 left_blockptr = btrfs_node_blockptr(
6885 left_path->nodes[left_level],
6886 left_path->slots[left_level]);
6887 right_blockptr = btrfs_node_blockptr(
6888 right_path->nodes[right_level],
6889 right_path->slots[right_level]);
6890 left_gen = btrfs_node_ptr_generation(
6891 left_path->nodes[left_level],
6892 left_path->slots[left_level]);
6893 right_gen = btrfs_node_ptr_generation(
6894 right_path->nodes[right_level],
6895 right_path->slots[right_level]);
6896 if (left_blockptr == right_blockptr &&
6897 left_gen == right_gen) {
6899 * As we're on a shared block, don't
6900 * allow to go deeper.
6902 advance_left = ADVANCE_ONLY_NEXT;
6903 advance_right = ADVANCE_ONLY_NEXT;
6905 advance_left = ADVANCE;
6906 advance_right = ADVANCE;
6909 } else if (left_level < right_level) {
6910 advance_right = ADVANCE;
6912 advance_left = ADVANCE;
6917 btrfs_free_path(left_path);
6918 btrfs_free_path(right_path);
6923 static int send_subvol(struct send_ctx *sctx)
6927 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6928 ret = send_header(sctx);
6933 ret = send_subvol_begin(sctx);
6937 if (sctx->parent_root) {
6938 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6942 ret = finish_inode_if_needed(sctx, 1);
6946 ret = full_send_tree(sctx);
6952 free_recorded_refs(sctx);
6957 * If orphan cleanup did remove any orphans from a root, it means the tree
6958 * was modified and therefore the commit root is not the same as the current
6959 * root anymore. This is a problem, because send uses the commit root and
6960 * therefore can see inode items that don't exist in the current root anymore,
6961 * and for example make calls to btrfs_iget, which will do tree lookups based
6962 * on the current root and not on the commit root. Those lookups will fail,
6963 * returning a -ESTALE error, and making send fail with that error. So make
6964 * sure a send does not see any orphans we have just removed, and that it will
6965 * see the same inodes regardless of whether a transaction commit happened
6966 * before it started (meaning that the commit root will be the same as the
6967 * current root) or not.
6969 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6972 struct btrfs_trans_handle *trans = NULL;
6975 if (sctx->parent_root &&
6976 sctx->parent_root->node != sctx->parent_root->commit_root)
6979 for (i = 0; i < sctx->clone_roots_cnt; i++)
6980 if (sctx->clone_roots[i].root->node !=
6981 sctx->clone_roots[i].root->commit_root)
6985 return btrfs_end_transaction(trans);
6990 /* Use any root, all fs roots will get their commit roots updated. */
6992 trans = btrfs_join_transaction(sctx->send_root);
6994 return PTR_ERR(trans);
6998 return btrfs_commit_transaction(trans);
7002 * Make sure any existing dellaloc is flushed for any root used by a send
7003 * operation so that we do not miss any data and we do not race with writeback
7004 * finishing and changing a tree while send is using the tree. This could
7005 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
7006 * a send operation then uses the subvolume.
7007 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
7009 static int flush_delalloc_roots(struct send_ctx *sctx)
7011 struct btrfs_root *root = sctx->parent_root;
7016 ret = btrfs_start_delalloc_snapshot(root);
7019 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7022 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7023 root = sctx->clone_roots[i].root;
7024 ret = btrfs_start_delalloc_snapshot(root);
7027 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7033 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
7035 spin_lock(&root->root_item_lock);
7036 root->send_in_progress--;
7038 * Not much left to do, we don't know why it's unbalanced and
7039 * can't blindly reset it to 0.
7041 if (root->send_in_progress < 0)
7042 btrfs_err(root->fs_info,
7043 "send_in_progress unbalanced %d root %llu",
7044 root->send_in_progress, root->root_key.objectid);
7045 spin_unlock(&root->root_item_lock);
7048 static void dedupe_in_progress_warn(const struct btrfs_root *root)
7050 btrfs_warn_rl(root->fs_info,
7051 "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
7052 root->root_key.objectid, root->dedupe_in_progress);
7055 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
7058 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
7059 struct btrfs_fs_info *fs_info = send_root->fs_info;
7060 struct btrfs_root *clone_root;
7061 struct btrfs_key key;
7062 struct send_ctx *sctx = NULL;
7064 u64 *clone_sources_tmp = NULL;
7065 int clone_sources_to_rollback = 0;
7066 unsigned alloc_size;
7067 int sort_clone_roots = 0;
7070 if (!capable(CAP_SYS_ADMIN))
7074 * The subvolume must remain read-only during send, protect against
7075 * making it RW. This also protects against deletion.
7077 spin_lock(&send_root->root_item_lock);
7078 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
7079 dedupe_in_progress_warn(send_root);
7080 spin_unlock(&send_root->root_item_lock);
7083 send_root->send_in_progress++;
7084 spin_unlock(&send_root->root_item_lock);
7087 * Userspace tools do the checks and warn the user if it's
7090 if (!btrfs_root_readonly(send_root)) {
7096 * Check that we don't overflow at later allocations, we request
7097 * clone_sources_count + 1 items, and compare to unsigned long inside
7100 if (arg->clone_sources_count >
7101 ULONG_MAX / sizeof(struct clone_root) - 1) {
7106 if (!access_ok(arg->clone_sources,
7107 sizeof(*arg->clone_sources) *
7108 arg->clone_sources_count)) {
7113 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
7118 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
7124 INIT_LIST_HEAD(&sctx->new_refs);
7125 INIT_LIST_HEAD(&sctx->deleted_refs);
7126 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
7127 INIT_LIST_HEAD(&sctx->name_cache_list);
7129 sctx->flags = arg->flags;
7131 sctx->send_filp = fget(arg->send_fd);
7132 if (!sctx->send_filp) {
7137 sctx->send_root = send_root;
7139 * Unlikely but possible, if the subvolume is marked for deletion but
7140 * is slow to remove the directory entry, send can still be started
7142 if (btrfs_root_dead(sctx->send_root)) {
7147 sctx->clone_roots_cnt = arg->clone_sources_count;
7149 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
7150 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
7151 if (!sctx->send_buf) {
7156 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
7157 if (!sctx->read_buf) {
7162 sctx->pending_dir_moves = RB_ROOT;
7163 sctx->waiting_dir_moves = RB_ROOT;
7164 sctx->orphan_dirs = RB_ROOT;
7166 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
7168 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
7169 if (!sctx->clone_roots) {
7174 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
7176 if (arg->clone_sources_count) {
7177 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
7178 if (!clone_sources_tmp) {
7183 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
7190 for (i = 0; i < arg->clone_sources_count; i++) {
7191 key.objectid = clone_sources_tmp[i];
7192 key.type = BTRFS_ROOT_ITEM_KEY;
7193 key.offset = (u64)-1;
7195 index = srcu_read_lock(&fs_info->subvol_srcu);
7197 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
7198 if (IS_ERR(clone_root)) {
7199 srcu_read_unlock(&fs_info->subvol_srcu, index);
7200 ret = PTR_ERR(clone_root);
7203 spin_lock(&clone_root->root_item_lock);
7204 if (!btrfs_root_readonly(clone_root) ||
7205 btrfs_root_dead(clone_root)) {
7206 spin_unlock(&clone_root->root_item_lock);
7207 srcu_read_unlock(&fs_info->subvol_srcu, index);
7211 if (clone_root->dedupe_in_progress) {
7212 dedupe_in_progress_warn(clone_root);
7213 spin_unlock(&clone_root->root_item_lock);
7214 srcu_read_unlock(&fs_info->subvol_srcu, index);
7218 clone_root->send_in_progress++;
7219 spin_unlock(&clone_root->root_item_lock);
7220 srcu_read_unlock(&fs_info->subvol_srcu, index);
7222 sctx->clone_roots[i].root = clone_root;
7223 clone_sources_to_rollback = i + 1;
7225 kvfree(clone_sources_tmp);
7226 clone_sources_tmp = NULL;
7229 if (arg->parent_root) {
7230 key.objectid = arg->parent_root;
7231 key.type = BTRFS_ROOT_ITEM_KEY;
7232 key.offset = (u64)-1;
7234 index = srcu_read_lock(&fs_info->subvol_srcu);
7236 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
7237 if (IS_ERR(sctx->parent_root)) {
7238 srcu_read_unlock(&fs_info->subvol_srcu, index);
7239 ret = PTR_ERR(sctx->parent_root);
7243 spin_lock(&sctx->parent_root->root_item_lock);
7244 sctx->parent_root->send_in_progress++;
7245 if (!btrfs_root_readonly(sctx->parent_root) ||
7246 btrfs_root_dead(sctx->parent_root)) {
7247 spin_unlock(&sctx->parent_root->root_item_lock);
7248 srcu_read_unlock(&fs_info->subvol_srcu, index);
7252 if (sctx->parent_root->dedupe_in_progress) {
7253 dedupe_in_progress_warn(sctx->parent_root);
7254 spin_unlock(&sctx->parent_root->root_item_lock);
7255 srcu_read_unlock(&fs_info->subvol_srcu, index);
7259 spin_unlock(&sctx->parent_root->root_item_lock);
7261 srcu_read_unlock(&fs_info->subvol_srcu, index);
7265 * Clones from send_root are allowed, but only if the clone source
7266 * is behind the current send position. This is checked while searching
7267 * for possible clone sources.
7269 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
7271 /* We do a bsearch later */
7272 sort(sctx->clone_roots, sctx->clone_roots_cnt,
7273 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
7275 sort_clone_roots = 1;
7277 ret = flush_delalloc_roots(sctx);
7281 ret = ensure_commit_roots_uptodate(sctx);
7285 mutex_lock(&fs_info->balance_mutex);
7286 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
7287 mutex_unlock(&fs_info->balance_mutex);
7288 btrfs_warn_rl(fs_info,
7289 "cannot run send because a balance operation is in progress");
7293 fs_info->send_in_progress++;
7294 mutex_unlock(&fs_info->balance_mutex);
7296 current->journal_info = BTRFS_SEND_TRANS_STUB;
7297 ret = send_subvol(sctx);
7298 current->journal_info = NULL;
7299 mutex_lock(&fs_info->balance_mutex);
7300 fs_info->send_in_progress--;
7301 mutex_unlock(&fs_info->balance_mutex);
7305 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
7306 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
7309 ret = send_cmd(sctx);
7315 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
7316 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
7318 struct pending_dir_move *pm;
7320 n = rb_first(&sctx->pending_dir_moves);
7321 pm = rb_entry(n, struct pending_dir_move, node);
7322 while (!list_empty(&pm->list)) {
7323 struct pending_dir_move *pm2;
7325 pm2 = list_first_entry(&pm->list,
7326 struct pending_dir_move, list);
7327 free_pending_move(sctx, pm2);
7329 free_pending_move(sctx, pm);
7332 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
7333 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
7335 struct waiting_dir_move *dm;
7337 n = rb_first(&sctx->waiting_dir_moves);
7338 dm = rb_entry(n, struct waiting_dir_move, node);
7339 rb_erase(&dm->node, &sctx->waiting_dir_moves);
7343 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
7344 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
7346 struct orphan_dir_info *odi;
7348 n = rb_first(&sctx->orphan_dirs);
7349 odi = rb_entry(n, struct orphan_dir_info, node);
7350 free_orphan_dir_info(sctx, odi);
7353 if (sort_clone_roots) {
7354 for (i = 0; i < sctx->clone_roots_cnt; i++)
7355 btrfs_root_dec_send_in_progress(
7356 sctx->clone_roots[i].root);
7358 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
7359 btrfs_root_dec_send_in_progress(
7360 sctx->clone_roots[i].root);
7362 btrfs_root_dec_send_in_progress(send_root);
7364 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
7365 btrfs_root_dec_send_in_progress(sctx->parent_root);
7367 kvfree(clone_sources_tmp);
7370 if (sctx->send_filp)
7371 fput(sctx->send_filp);
7373 kvfree(sctx->clone_roots);
7374 kvfree(sctx->send_buf);
7375 kvfree(sctx->read_buf);
7377 name_cache_free(sctx);