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"
29 * Maximum number of references an extent can have in order for us to attempt to
30 * issue clone operations instead of write operations. This currently exists to
31 * avoid hitting limitations of the backreference walking code (taking a lot of
32 * time and using too much memory for extents with large number of references).
34 #define SEND_MAX_EXTENT_REFS 64
37 * A fs_path is a helper to dynamically build path names with unknown size.
38 * It reallocates the internal buffer on demand.
39 * It allows fast adding of path elements on the right side (normal path) and
40 * fast adding to the left side (reversed path). A reversed path can also be
41 * unreversed if needed.
50 unsigned short buf_len:15;
51 unsigned short reversed:1;
55 * Average path length does not exceed 200 bytes, we'll have
56 * better packing in the slab and higher chance to satisfy
57 * a allocation later during send.
62 #define FS_PATH_INLINE_SIZE \
63 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
66 /* reused for each extent */
68 struct btrfs_root *root;
75 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
76 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
79 struct file *send_filp;
85 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
86 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
88 struct btrfs_root *send_root;
89 struct btrfs_root *parent_root;
90 struct clone_root *clone_roots;
93 /* current state of the compare_tree call */
94 struct btrfs_path *left_path;
95 struct btrfs_path *right_path;
96 struct btrfs_key *cmp_key;
99 * infos of the currently processed inode. In case of deleted inodes,
100 * these are the values from the deleted inode.
105 int cur_inode_new_gen;
106 int cur_inode_deleted;
110 u64 cur_inode_last_extent;
111 u64 cur_inode_next_write_offset;
112 bool ignore_cur_inode;
116 struct list_head new_refs;
117 struct list_head deleted_refs;
119 struct radix_tree_root name_cache;
120 struct list_head name_cache_list;
123 struct file_ra_state ra;
128 * We process inodes by their increasing order, so if before an
129 * incremental send we reverse the parent/child relationship of
130 * directories such that a directory with a lower inode number was
131 * the parent of a directory with a higher inode number, and the one
132 * becoming the new parent got renamed too, we can't rename/move the
133 * directory with lower inode number when we finish processing it - we
134 * must process the directory with higher inode number first, then
135 * rename/move it and then rename/move the directory with lower inode
136 * number. Example follows.
138 * Tree state when the first send was performed:
150 * Tree state when the second (incremental) send is performed:
159 * The sequence of steps that lead to the second state was:
161 * mv /a/b/c/d /a/b/c2/d2
162 * mv /a/b/c /a/b/c2/d2/cc
164 * "c" has lower inode number, but we can't move it (2nd mv operation)
165 * before we move "d", which has higher inode number.
167 * So we just memorize which move/rename operations must be performed
168 * later when their respective parent is processed and moved/renamed.
171 /* Indexed by parent directory inode number. */
172 struct rb_root pending_dir_moves;
175 * Reverse index, indexed by the inode number of a directory that
176 * is waiting for the move/rename of its immediate parent before its
177 * own move/rename can be performed.
179 struct rb_root waiting_dir_moves;
182 * A directory that is going to be rm'ed might have a child directory
183 * which is in the pending directory moves index above. In this case,
184 * the directory can only be removed after the move/rename of its child
185 * is performed. Example:
205 * Sequence of steps that lead to the send snapshot:
206 * rm -f /a/b/c/foo.txt
208 * mv /a/b/c/x /a/b/YY
211 * When the child is processed, its move/rename is delayed until its
212 * parent is processed (as explained above), but all other operations
213 * like update utimes, chown, chgrp, etc, are performed and the paths
214 * that it uses for those operations must use the orphanized name of
215 * its parent (the directory we're going to rm later), so we need to
216 * memorize that name.
218 * Indexed by the inode number of the directory to be deleted.
220 struct rb_root orphan_dirs;
223 struct pending_dir_move {
225 struct list_head list;
229 struct list_head update_refs;
232 struct waiting_dir_move {
236 * There might be some directory that could not be removed because it
237 * was waiting for this directory inode to be moved first. Therefore
238 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
244 struct orphan_dir_info {
248 u64 last_dir_index_offset;
251 struct name_cache_entry {
252 struct list_head list;
254 * radix_tree has only 32bit entries but we need to handle 64bit inums.
255 * We use the lower 32bit of the 64bit inum to store it in the tree. If
256 * more then one inum would fall into the same entry, we use radix_list
257 * to store the additional entries. radix_list is also used to store
258 * entries where two entries have the same inum but different
261 struct list_head radix_list;
267 int need_later_update;
273 #define ADVANCE_ONLY_NEXT -1
275 enum btrfs_compare_tree_result {
276 BTRFS_COMPARE_TREE_NEW,
277 BTRFS_COMPARE_TREE_DELETED,
278 BTRFS_COMPARE_TREE_CHANGED,
279 BTRFS_COMPARE_TREE_SAME,
281 typedef int (*btrfs_changed_cb_t)(struct btrfs_path *left_path,
282 struct btrfs_path *right_path,
283 struct btrfs_key *key,
284 enum btrfs_compare_tree_result result,
288 static void inconsistent_snapshot_error(struct send_ctx *sctx,
289 enum btrfs_compare_tree_result result,
292 const char *result_string;
295 case BTRFS_COMPARE_TREE_NEW:
296 result_string = "new";
298 case BTRFS_COMPARE_TREE_DELETED:
299 result_string = "deleted";
301 case BTRFS_COMPARE_TREE_CHANGED:
302 result_string = "updated";
304 case BTRFS_COMPARE_TREE_SAME:
306 result_string = "unchanged";
310 result_string = "unexpected";
313 btrfs_err(sctx->send_root->fs_info,
314 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
315 result_string, what, sctx->cmp_key->objectid,
316 sctx->send_root->root_key.objectid,
318 sctx->parent_root->root_key.objectid : 0));
321 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
323 static struct waiting_dir_move *
324 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
326 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
328 static int need_send_hole(struct send_ctx *sctx)
330 return (sctx->parent_root && !sctx->cur_inode_new &&
331 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
332 S_ISREG(sctx->cur_inode_mode));
335 static void fs_path_reset(struct fs_path *p)
338 p->start = p->buf + p->buf_len - 1;
348 static struct fs_path *fs_path_alloc(void)
352 p = kmalloc(sizeof(*p), GFP_KERNEL);
356 p->buf = p->inline_buf;
357 p->buf_len = FS_PATH_INLINE_SIZE;
362 static struct fs_path *fs_path_alloc_reversed(void)
374 static void fs_path_free(struct fs_path *p)
378 if (p->buf != p->inline_buf)
383 static int fs_path_len(struct fs_path *p)
385 return p->end - p->start;
388 static int fs_path_ensure_buf(struct fs_path *p, int len)
396 if (p->buf_len >= len)
399 if (len > PATH_MAX) {
404 path_len = p->end - p->start;
405 old_buf_len = p->buf_len;
408 * First time the inline_buf does not suffice
410 if (p->buf == p->inline_buf) {
411 tmp_buf = kmalloc(len, GFP_KERNEL);
413 memcpy(tmp_buf, p->buf, old_buf_len);
415 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
421 * The real size of the buffer is bigger, this will let the fast path
422 * happen most of the time
424 p->buf_len = ksize(p->buf);
427 tmp_buf = p->buf + old_buf_len - path_len - 1;
428 p->end = p->buf + p->buf_len - 1;
429 p->start = p->end - path_len;
430 memmove(p->start, tmp_buf, path_len + 1);
433 p->end = p->start + path_len;
438 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
444 new_len = p->end - p->start + name_len;
445 if (p->start != p->end)
447 ret = fs_path_ensure_buf(p, new_len);
452 if (p->start != p->end)
454 p->start -= name_len;
455 *prepared = p->start;
457 if (p->start != p->end)
468 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
473 ret = fs_path_prepare_for_add(p, name_len, &prepared);
476 memcpy(prepared, name, name_len);
482 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
487 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
490 memcpy(prepared, p2->start, p2->end - p2->start);
496 static int fs_path_add_from_extent_buffer(struct fs_path *p,
497 struct extent_buffer *eb,
498 unsigned long off, int len)
503 ret = fs_path_prepare_for_add(p, len, &prepared);
507 read_extent_buffer(eb, prepared, off, len);
513 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
517 p->reversed = from->reversed;
520 ret = fs_path_add_path(p, from);
526 static void fs_path_unreverse(struct fs_path *p)
535 len = p->end - p->start;
537 p->end = p->start + len;
538 memmove(p->start, tmp, len + 1);
542 static struct btrfs_path *alloc_path_for_send(void)
544 struct btrfs_path *path;
546 path = btrfs_alloc_path();
549 path->search_commit_root = 1;
550 path->skip_locking = 1;
551 path->need_commit_sem = 1;
555 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
561 ret = kernel_write(filp, buf + pos, len - pos, off);
562 /* TODO handle that correctly */
563 /*if (ret == -ERESTARTSYS) {
577 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
579 struct btrfs_tlv_header *hdr;
580 int total_len = sizeof(*hdr) + len;
581 int left = sctx->send_max_size - sctx->send_size;
583 if (unlikely(left < total_len))
586 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
587 hdr->tlv_type = cpu_to_le16(attr);
588 hdr->tlv_len = cpu_to_le16(len);
589 memcpy(hdr + 1, data, len);
590 sctx->send_size += total_len;
595 #define TLV_PUT_DEFINE_INT(bits) \
596 static int tlv_put_u##bits(struct send_ctx *sctx, \
597 u##bits attr, u##bits value) \
599 __le##bits __tmp = cpu_to_le##bits(value); \
600 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
603 TLV_PUT_DEFINE_INT(64)
605 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
606 const char *str, int len)
610 return tlv_put(sctx, attr, str, len);
613 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
616 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
619 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
620 struct extent_buffer *eb,
621 struct btrfs_timespec *ts)
623 struct btrfs_timespec bts;
624 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
625 return tlv_put(sctx, attr, &bts, sizeof(bts));
629 #define TLV_PUT(sctx, attrtype, data, attrlen) \
631 ret = tlv_put(sctx, attrtype, data, attrlen); \
633 goto tlv_put_failure; \
636 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
638 ret = tlv_put_u##bits(sctx, attrtype, value); \
640 goto tlv_put_failure; \
643 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
644 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
645 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
646 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
647 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
649 ret = tlv_put_string(sctx, attrtype, str, len); \
651 goto tlv_put_failure; \
653 #define TLV_PUT_PATH(sctx, attrtype, p) \
655 ret = tlv_put_string(sctx, attrtype, p->start, \
656 p->end - p->start); \
658 goto tlv_put_failure; \
660 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
662 ret = tlv_put_uuid(sctx, attrtype, uuid); \
664 goto tlv_put_failure; \
666 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
668 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
670 goto tlv_put_failure; \
673 static int send_header(struct send_ctx *sctx)
675 struct btrfs_stream_header hdr;
677 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
678 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
680 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
685 * For each command/item we want to send to userspace, we call this function.
687 static int begin_cmd(struct send_ctx *sctx, int cmd)
689 struct btrfs_cmd_header *hdr;
691 if (WARN_ON(!sctx->send_buf))
694 BUG_ON(sctx->send_size);
696 sctx->send_size += sizeof(*hdr);
697 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
698 hdr->cmd = cpu_to_le16(cmd);
703 static int send_cmd(struct send_ctx *sctx)
706 struct btrfs_cmd_header *hdr;
709 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
710 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
713 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
714 hdr->crc = cpu_to_le32(crc);
716 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
719 sctx->total_send_size += sctx->send_size;
720 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
727 * Sends a move instruction to user space
729 static int send_rename(struct send_ctx *sctx,
730 struct fs_path *from, struct fs_path *to)
732 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
735 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
737 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
742 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
744 ret = send_cmd(sctx);
752 * Sends a link instruction to user space
754 static int send_link(struct send_ctx *sctx,
755 struct fs_path *path, struct fs_path *lnk)
757 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
760 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
762 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
766 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
767 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
769 ret = send_cmd(sctx);
777 * Sends an unlink instruction to user space
779 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
781 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
784 btrfs_debug(fs_info, "send_unlink %s", path->start);
786 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
790 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
792 ret = send_cmd(sctx);
800 * Sends a rmdir instruction to user space
802 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
804 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
807 btrfs_debug(fs_info, "send_rmdir %s", path->start);
809 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
813 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
815 ret = send_cmd(sctx);
823 * Helper function to retrieve some fields from an inode item.
825 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
826 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
830 struct btrfs_inode_item *ii;
831 struct btrfs_key key;
834 key.type = BTRFS_INODE_ITEM_KEY;
836 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
843 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
844 struct btrfs_inode_item);
846 *size = btrfs_inode_size(path->nodes[0], ii);
848 *gen = btrfs_inode_generation(path->nodes[0], ii);
850 *mode = btrfs_inode_mode(path->nodes[0], ii);
852 *uid = btrfs_inode_uid(path->nodes[0], ii);
854 *gid = btrfs_inode_gid(path->nodes[0], ii);
856 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
861 static int get_inode_info(struct btrfs_root *root,
862 u64 ino, u64 *size, u64 *gen,
863 u64 *mode, u64 *uid, u64 *gid,
866 struct btrfs_path *path;
869 path = alloc_path_for_send();
872 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
874 btrfs_free_path(path);
878 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
883 * Helper function to iterate the entries in ONE btrfs_inode_ref or
884 * btrfs_inode_extref.
885 * The iterate callback may return a non zero value to stop iteration. This can
886 * be a negative value for error codes or 1 to simply stop it.
888 * path must point to the INODE_REF or INODE_EXTREF when called.
890 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
891 struct btrfs_key *found_key, int resolve,
892 iterate_inode_ref_t iterate, void *ctx)
894 struct extent_buffer *eb = path->nodes[0];
895 struct btrfs_item *item;
896 struct btrfs_inode_ref *iref;
897 struct btrfs_inode_extref *extref;
898 struct btrfs_path *tmp_path;
902 int slot = path->slots[0];
909 unsigned long name_off;
910 unsigned long elem_size;
913 p = fs_path_alloc_reversed();
917 tmp_path = alloc_path_for_send();
924 if (found_key->type == BTRFS_INODE_REF_KEY) {
925 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
926 struct btrfs_inode_ref);
927 item = btrfs_item_nr(slot);
928 total = btrfs_item_size(eb, item);
929 elem_size = sizeof(*iref);
931 ptr = btrfs_item_ptr_offset(eb, slot);
932 total = btrfs_item_size_nr(eb, slot);
933 elem_size = sizeof(*extref);
936 while (cur < total) {
939 if (found_key->type == BTRFS_INODE_REF_KEY) {
940 iref = (struct btrfs_inode_ref *)(ptr + cur);
941 name_len = btrfs_inode_ref_name_len(eb, iref);
942 name_off = (unsigned long)(iref + 1);
943 index = btrfs_inode_ref_index(eb, iref);
944 dir = found_key->offset;
946 extref = (struct btrfs_inode_extref *)(ptr + cur);
947 name_len = btrfs_inode_extref_name_len(eb, extref);
948 name_off = (unsigned long)&extref->name;
949 index = btrfs_inode_extref_index(eb, extref);
950 dir = btrfs_inode_extref_parent(eb, extref);
954 start = btrfs_ref_to_path(root, tmp_path, name_len,
958 ret = PTR_ERR(start);
961 if (start < p->buf) {
962 /* overflow , try again with larger buffer */
963 ret = fs_path_ensure_buf(p,
964 p->buf_len + p->buf - start);
967 start = btrfs_ref_to_path(root, tmp_path,
972 ret = PTR_ERR(start);
975 BUG_ON(start < p->buf);
979 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
985 cur += elem_size + name_len;
986 ret = iterate(num, dir, index, p, ctx);
993 btrfs_free_path(tmp_path);
998 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
999 const char *name, int name_len,
1000 const char *data, int data_len,
1001 u8 type, void *ctx);
1004 * Helper function to iterate the entries in ONE btrfs_dir_item.
1005 * The iterate callback may return a non zero value to stop iteration. This can
1006 * be a negative value for error codes or 1 to simply stop it.
1008 * path must point to the dir item when called.
1010 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
1011 iterate_dir_item_t iterate, void *ctx)
1014 struct extent_buffer *eb;
1015 struct btrfs_item *item;
1016 struct btrfs_dir_item *di;
1017 struct btrfs_key di_key;
1030 * Start with a small buffer (1 page). If later we end up needing more
1031 * space, which can happen for xattrs on a fs with a leaf size greater
1032 * then the page size, attempt to increase the buffer. Typically xattr
1036 buf = kmalloc(buf_len, GFP_KERNEL);
1042 eb = path->nodes[0];
1043 slot = path->slots[0];
1044 item = btrfs_item_nr(slot);
1045 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1048 total = btrfs_item_size(eb, item);
1051 while (cur < total) {
1052 name_len = btrfs_dir_name_len(eb, di);
1053 data_len = btrfs_dir_data_len(eb, di);
1054 type = btrfs_dir_type(eb, di);
1055 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1057 if (type == BTRFS_FT_XATTR) {
1058 if (name_len > XATTR_NAME_MAX) {
1059 ret = -ENAMETOOLONG;
1062 if (name_len + data_len >
1063 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1071 if (name_len + data_len > PATH_MAX) {
1072 ret = -ENAMETOOLONG;
1077 if (name_len + data_len > buf_len) {
1078 buf_len = name_len + data_len;
1079 if (is_vmalloc_addr(buf)) {
1083 char *tmp = krealloc(buf, buf_len,
1084 GFP_KERNEL | __GFP_NOWARN);
1091 buf = kvmalloc(buf_len, GFP_KERNEL);
1099 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1100 name_len + data_len);
1102 len = sizeof(*di) + name_len + data_len;
1103 di = (struct btrfs_dir_item *)((char *)di + len);
1106 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1107 data_len, type, ctx);
1123 static int __copy_first_ref(int num, u64 dir, int index,
1124 struct fs_path *p, void *ctx)
1127 struct fs_path *pt = ctx;
1129 ret = fs_path_copy(pt, p);
1133 /* we want the first only */
1138 * Retrieve the first path of an inode. If an inode has more then one
1139 * ref/hardlink, this is ignored.
1141 static int get_inode_path(struct btrfs_root *root,
1142 u64 ino, struct fs_path *path)
1145 struct btrfs_key key, found_key;
1146 struct btrfs_path *p;
1148 p = alloc_path_for_send();
1152 fs_path_reset(path);
1155 key.type = BTRFS_INODE_REF_KEY;
1158 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1165 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1166 if (found_key.objectid != ino ||
1167 (found_key.type != BTRFS_INODE_REF_KEY &&
1168 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1173 ret = iterate_inode_ref(root, p, &found_key, 1,
1174 __copy_first_ref, path);
1184 struct backref_ctx {
1185 struct send_ctx *sctx;
1187 /* number of total found references */
1191 * used for clones found in send_root. clones found behind cur_objectid
1192 * and cur_offset are not considered as allowed clones.
1197 /* may be truncated in case it's the last extent in a file */
1200 /* data offset in the file extent item */
1203 /* Just to check for bugs in backref resolving */
1207 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1209 u64 root = (u64)(uintptr_t)key;
1210 struct clone_root *cr = (struct clone_root *)elt;
1212 if (root < cr->root->root_key.objectid)
1214 if (root > cr->root->root_key.objectid)
1219 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1221 struct clone_root *cr1 = (struct clone_root *)e1;
1222 struct clone_root *cr2 = (struct clone_root *)e2;
1224 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1226 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1232 * Called for every backref that is found for the current extent.
1233 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1235 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1237 struct backref_ctx *bctx = ctx_;
1238 struct clone_root *found;
1240 /* First check if the root is in the list of accepted clone sources */
1241 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1242 bctx->sctx->clone_roots_cnt,
1243 sizeof(struct clone_root),
1244 __clone_root_cmp_bsearch);
1248 if (found->root == bctx->sctx->send_root &&
1249 ino == bctx->cur_objectid &&
1250 offset == bctx->cur_offset) {
1251 bctx->found_itself = 1;
1255 * Make sure we don't consider clones from send_root that are
1256 * behind the current inode/offset.
1258 if (found->root == bctx->sctx->send_root) {
1260 * If the source inode was not yet processed we can't issue a
1261 * clone operation, as the source extent does not exist yet at
1262 * the destination of the stream.
1264 if (ino > bctx->cur_objectid)
1267 * We clone from the inode currently being sent as long as the
1268 * source extent is already processed, otherwise we could try
1269 * to clone from an extent that does not exist yet at the
1270 * destination of the stream.
1272 if (ino == bctx->cur_objectid &&
1273 offset + bctx->extent_len >
1274 bctx->sctx->cur_inode_next_write_offset)
1279 found->found_refs++;
1280 if (ino < found->ino) {
1282 found->offset = offset;
1283 } else if (found->ino == ino) {
1285 * same extent found more then once in the same file.
1287 if (found->offset > offset + bctx->extent_len)
1288 found->offset = offset;
1295 * Given an inode, offset and extent item, it finds a good clone for a clone
1296 * instruction. Returns -ENOENT when none could be found. The function makes
1297 * sure that the returned clone is usable at the point where sending is at the
1298 * moment. This means, that no clones are accepted which lie behind the current
1301 * path must point to the extent item when called.
1303 static int find_extent_clone(struct send_ctx *sctx,
1304 struct btrfs_path *path,
1305 u64 ino, u64 data_offset,
1307 struct clone_root **found)
1309 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1315 u64 extent_item_pos;
1317 struct btrfs_file_extent_item *fi;
1318 struct extent_buffer *eb = path->nodes[0];
1319 struct backref_ctx *backref_ctx = NULL;
1320 struct clone_root *cur_clone_root;
1321 struct btrfs_key found_key;
1322 struct btrfs_path *tmp_path;
1323 struct btrfs_extent_item *ei;
1327 tmp_path = alloc_path_for_send();
1331 /* We only use this path under the commit sem */
1332 tmp_path->need_commit_sem = 0;
1334 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1340 if (data_offset >= ino_size) {
1342 * There may be extents that lie behind the file's size.
1343 * I at least had this in combination with snapshotting while
1344 * writing large files.
1350 fi = btrfs_item_ptr(eb, path->slots[0],
1351 struct btrfs_file_extent_item);
1352 extent_type = btrfs_file_extent_type(eb, fi);
1353 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1357 compressed = btrfs_file_extent_compression(eb, fi);
1359 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1360 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1361 if (disk_byte == 0) {
1365 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1367 down_read(&fs_info->commit_root_sem);
1368 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1369 &found_key, &flags);
1370 up_read(&fs_info->commit_root_sem);
1374 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1379 ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0],
1380 struct btrfs_extent_item);
1382 * Backreference walking (iterate_extent_inodes() below) is currently
1383 * too expensive when an extent has a large number of references, both
1384 * in time spent and used memory. So for now just fallback to write
1385 * operations instead of clone operations when an extent has more than
1386 * a certain amount of references.
1388 if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) {
1392 btrfs_release_path(tmp_path);
1395 * Setup the clone roots.
1397 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1398 cur_clone_root = sctx->clone_roots + i;
1399 cur_clone_root->ino = (u64)-1;
1400 cur_clone_root->offset = 0;
1401 cur_clone_root->found_refs = 0;
1404 backref_ctx->sctx = sctx;
1405 backref_ctx->found = 0;
1406 backref_ctx->cur_objectid = ino;
1407 backref_ctx->cur_offset = data_offset;
1408 backref_ctx->found_itself = 0;
1409 backref_ctx->extent_len = num_bytes;
1411 * For non-compressed extents iterate_extent_inodes() gives us extent
1412 * offsets that already take into account the data offset, but not for
1413 * compressed extents, since the offset is logical and not relative to
1414 * the physical extent locations. We must take this into account to
1415 * avoid sending clone offsets that go beyond the source file's size,
1416 * which would result in the clone ioctl failing with -EINVAL on the
1419 if (compressed == BTRFS_COMPRESS_NONE)
1420 backref_ctx->data_offset = 0;
1422 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1425 * The last extent of a file may be too large due to page alignment.
1426 * We need to adjust extent_len in this case so that the checks in
1427 * __iterate_backrefs work.
1429 if (data_offset + num_bytes >= ino_size)
1430 backref_ctx->extent_len = ino_size - data_offset;
1433 * Now collect all backrefs.
1435 if (compressed == BTRFS_COMPRESS_NONE)
1436 extent_item_pos = logical - found_key.objectid;
1438 extent_item_pos = 0;
1439 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1440 extent_item_pos, 1, __iterate_backrefs,
1441 backref_ctx, false);
1446 if (!backref_ctx->found_itself) {
1447 /* found a bug in backref code? */
1450 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1451 ino, data_offset, disk_byte, found_key.objectid);
1455 btrfs_debug(fs_info,
1456 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1457 data_offset, ino, num_bytes, logical);
1459 if (!backref_ctx->found)
1460 btrfs_debug(fs_info, "no clones found");
1462 cur_clone_root = NULL;
1463 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1464 if (sctx->clone_roots[i].found_refs) {
1465 if (!cur_clone_root)
1466 cur_clone_root = sctx->clone_roots + i;
1467 else if (sctx->clone_roots[i].root == sctx->send_root)
1468 /* prefer clones from send_root over others */
1469 cur_clone_root = sctx->clone_roots + i;
1474 if (cur_clone_root) {
1475 *found = cur_clone_root;
1482 btrfs_free_path(tmp_path);
1487 static int read_symlink(struct btrfs_root *root,
1489 struct fs_path *dest)
1492 struct btrfs_path *path;
1493 struct btrfs_key key;
1494 struct btrfs_file_extent_item *ei;
1500 path = alloc_path_for_send();
1505 key.type = BTRFS_EXTENT_DATA_KEY;
1507 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1512 * An empty symlink inode. Can happen in rare error paths when
1513 * creating a symlink (transaction committed before the inode
1514 * eviction handler removed the symlink inode items and a crash
1515 * happened in between or the subvol was snapshoted in between).
1516 * Print an informative message to dmesg/syslog so that the user
1517 * can delete the symlink.
1519 btrfs_err(root->fs_info,
1520 "Found empty symlink inode %llu at root %llu",
1521 ino, root->root_key.objectid);
1526 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1527 struct btrfs_file_extent_item);
1528 type = btrfs_file_extent_type(path->nodes[0], ei);
1529 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1530 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1531 BUG_ON(compression);
1533 off = btrfs_file_extent_inline_start(ei);
1534 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1536 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1539 btrfs_free_path(path);
1544 * Helper function to generate a file name that is unique in the root of
1545 * send_root and parent_root. This is used to generate names for orphan inodes.
1547 static int gen_unique_name(struct send_ctx *sctx,
1549 struct fs_path *dest)
1552 struct btrfs_path *path;
1553 struct btrfs_dir_item *di;
1558 path = alloc_path_for_send();
1563 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1565 ASSERT(len < sizeof(tmp));
1567 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1568 path, BTRFS_FIRST_FREE_OBJECTID,
1569 tmp, strlen(tmp), 0);
1570 btrfs_release_path(path);
1576 /* not unique, try again */
1581 if (!sctx->parent_root) {
1587 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1588 path, BTRFS_FIRST_FREE_OBJECTID,
1589 tmp, strlen(tmp), 0);
1590 btrfs_release_path(path);
1596 /* not unique, try again */
1604 ret = fs_path_add(dest, tmp, strlen(tmp));
1607 btrfs_free_path(path);
1612 inode_state_no_change,
1613 inode_state_will_create,
1614 inode_state_did_create,
1615 inode_state_will_delete,
1616 inode_state_did_delete,
1619 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1627 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1629 if (ret < 0 && ret != -ENOENT)
1633 if (!sctx->parent_root) {
1634 right_ret = -ENOENT;
1636 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1637 NULL, NULL, NULL, NULL);
1638 if (ret < 0 && ret != -ENOENT)
1643 if (!left_ret && !right_ret) {
1644 if (left_gen == gen && right_gen == gen) {
1645 ret = inode_state_no_change;
1646 } else if (left_gen == gen) {
1647 if (ino < sctx->send_progress)
1648 ret = inode_state_did_create;
1650 ret = inode_state_will_create;
1651 } else if (right_gen == gen) {
1652 if (ino < sctx->send_progress)
1653 ret = inode_state_did_delete;
1655 ret = inode_state_will_delete;
1659 } else if (!left_ret) {
1660 if (left_gen == gen) {
1661 if (ino < sctx->send_progress)
1662 ret = inode_state_did_create;
1664 ret = inode_state_will_create;
1668 } else if (!right_ret) {
1669 if (right_gen == gen) {
1670 if (ino < sctx->send_progress)
1671 ret = inode_state_did_delete;
1673 ret = inode_state_will_delete;
1685 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1689 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1692 ret = get_cur_inode_state(sctx, ino, gen);
1696 if (ret == inode_state_no_change ||
1697 ret == inode_state_did_create ||
1698 ret == inode_state_will_delete)
1708 * Helper function to lookup a dir item in a dir.
1710 static int lookup_dir_item_inode(struct btrfs_root *root,
1711 u64 dir, const char *name, int name_len,
1716 struct btrfs_dir_item *di;
1717 struct btrfs_key key;
1718 struct btrfs_path *path;
1720 path = alloc_path_for_send();
1724 di = btrfs_lookup_dir_item(NULL, root, path,
1725 dir, name, name_len, 0);
1726 if (IS_ERR_OR_NULL(di)) {
1727 ret = di ? PTR_ERR(di) : -ENOENT;
1730 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1731 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1735 *found_inode = key.objectid;
1736 *found_type = btrfs_dir_type(path->nodes[0], di);
1739 btrfs_free_path(path);
1744 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1745 * generation of the parent dir and the name of the dir entry.
1747 static int get_first_ref(struct btrfs_root *root, u64 ino,
1748 u64 *dir, u64 *dir_gen, struct fs_path *name)
1751 struct btrfs_key key;
1752 struct btrfs_key found_key;
1753 struct btrfs_path *path;
1757 path = alloc_path_for_send();
1762 key.type = BTRFS_INODE_REF_KEY;
1765 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1769 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1771 if (ret || found_key.objectid != ino ||
1772 (found_key.type != BTRFS_INODE_REF_KEY &&
1773 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1778 if (found_key.type == BTRFS_INODE_REF_KEY) {
1779 struct btrfs_inode_ref *iref;
1780 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1781 struct btrfs_inode_ref);
1782 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1783 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1784 (unsigned long)(iref + 1),
1786 parent_dir = found_key.offset;
1788 struct btrfs_inode_extref *extref;
1789 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1790 struct btrfs_inode_extref);
1791 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1792 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1793 (unsigned long)&extref->name, len);
1794 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1798 btrfs_release_path(path);
1801 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1810 btrfs_free_path(path);
1814 static int is_first_ref(struct btrfs_root *root,
1816 const char *name, int name_len)
1819 struct fs_path *tmp_name;
1822 tmp_name = fs_path_alloc();
1826 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1830 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1835 ret = !memcmp(tmp_name->start, name, name_len);
1838 fs_path_free(tmp_name);
1843 * Used by process_recorded_refs to determine if a new ref would overwrite an
1844 * already existing ref. In case it detects an overwrite, it returns the
1845 * inode/gen in who_ino/who_gen.
1846 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1847 * to make sure later references to the overwritten inode are possible.
1848 * Orphanizing is however only required for the first ref of an inode.
1849 * process_recorded_refs does an additional is_first_ref check to see if
1850 * orphanizing is really required.
1852 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1853 const char *name, int name_len,
1854 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1858 u64 other_inode = 0;
1861 if (!sctx->parent_root)
1864 ret = is_inode_existent(sctx, dir, dir_gen);
1869 * If we have a parent root we need to verify that the parent dir was
1870 * not deleted and then re-created, if it was then we have no overwrite
1871 * and we can just unlink this entry.
1873 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1874 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1876 if (ret < 0 && ret != -ENOENT)
1886 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1887 &other_inode, &other_type);
1888 if (ret < 0 && ret != -ENOENT)
1896 * Check if the overwritten ref was already processed. If yes, the ref
1897 * was already unlinked/moved, so we can safely assume that we will not
1898 * overwrite anything at this point in time.
1900 if (other_inode > sctx->send_progress ||
1901 is_waiting_for_move(sctx, other_inode)) {
1902 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1903 who_gen, who_mode, NULL, NULL, NULL);
1908 *who_ino = other_inode;
1918 * Checks if the ref was overwritten by an already processed inode. This is
1919 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1920 * thus the orphan name needs be used.
1921 * process_recorded_refs also uses it to avoid unlinking of refs that were
1924 static int did_overwrite_ref(struct send_ctx *sctx,
1925 u64 dir, u64 dir_gen,
1926 u64 ino, u64 ino_gen,
1927 const char *name, int name_len)
1934 if (!sctx->parent_root)
1937 ret = is_inode_existent(sctx, dir, dir_gen);
1941 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1942 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1944 if (ret < 0 && ret != -ENOENT)
1954 /* check if the ref was overwritten by another ref */
1955 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1956 &ow_inode, &other_type);
1957 if (ret < 0 && ret != -ENOENT)
1960 /* was never and will never be overwritten */
1965 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1970 if (ow_inode == ino && gen == ino_gen) {
1976 * We know that it is or will be overwritten. Check this now.
1977 * The current inode being processed might have been the one that caused
1978 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1979 * the current inode being processed.
1981 if ((ow_inode < sctx->send_progress) ||
1982 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1983 gen == sctx->cur_inode_gen))
1993 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1994 * that got overwritten. This is used by process_recorded_refs to determine
1995 * if it has to use the path as returned by get_cur_path or the orphan name.
1997 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
2000 struct fs_path *name = NULL;
2004 if (!sctx->parent_root)
2007 name = fs_path_alloc();
2011 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
2015 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2016 name->start, fs_path_len(name));
2024 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2025 * so we need to do some special handling in case we have clashes. This function
2026 * takes care of this with the help of name_cache_entry::radix_list.
2027 * In case of error, nce is kfreed.
2029 static int name_cache_insert(struct send_ctx *sctx,
2030 struct name_cache_entry *nce)
2033 struct list_head *nce_head;
2035 nce_head = radix_tree_lookup(&sctx->name_cache,
2036 (unsigned long)nce->ino);
2038 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2043 INIT_LIST_HEAD(nce_head);
2045 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2052 list_add_tail(&nce->radix_list, nce_head);
2053 list_add_tail(&nce->list, &sctx->name_cache_list);
2054 sctx->name_cache_size++;
2059 static void name_cache_delete(struct send_ctx *sctx,
2060 struct name_cache_entry *nce)
2062 struct list_head *nce_head;
2064 nce_head = radix_tree_lookup(&sctx->name_cache,
2065 (unsigned long)nce->ino);
2067 btrfs_err(sctx->send_root->fs_info,
2068 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2069 nce->ino, sctx->name_cache_size);
2072 list_del(&nce->radix_list);
2073 list_del(&nce->list);
2074 sctx->name_cache_size--;
2077 * We may not get to the final release of nce_head if the lookup fails
2079 if (nce_head && list_empty(nce_head)) {
2080 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2085 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2088 struct list_head *nce_head;
2089 struct name_cache_entry *cur;
2091 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2095 list_for_each_entry(cur, nce_head, radix_list) {
2096 if (cur->ino == ino && cur->gen == gen)
2103 * Removes the entry from the list and adds it back to the end. This marks the
2104 * entry as recently used so that name_cache_clean_unused does not remove it.
2106 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2108 list_del(&nce->list);
2109 list_add_tail(&nce->list, &sctx->name_cache_list);
2113 * Remove some entries from the beginning of name_cache_list.
2115 static void name_cache_clean_unused(struct send_ctx *sctx)
2117 struct name_cache_entry *nce;
2119 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2122 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2123 nce = list_entry(sctx->name_cache_list.next,
2124 struct name_cache_entry, list);
2125 name_cache_delete(sctx, nce);
2130 static void name_cache_free(struct send_ctx *sctx)
2132 struct name_cache_entry *nce;
2134 while (!list_empty(&sctx->name_cache_list)) {
2135 nce = list_entry(sctx->name_cache_list.next,
2136 struct name_cache_entry, list);
2137 name_cache_delete(sctx, nce);
2143 * Used by get_cur_path for each ref up to the root.
2144 * Returns 0 if it succeeded.
2145 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2146 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2147 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2148 * Returns <0 in case of error.
2150 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2154 struct fs_path *dest)
2158 struct name_cache_entry *nce = NULL;
2161 * First check if we already did a call to this function with the same
2162 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2163 * return the cached result.
2165 nce = name_cache_search(sctx, ino, gen);
2167 if (ino < sctx->send_progress && nce->need_later_update) {
2168 name_cache_delete(sctx, nce);
2172 name_cache_used(sctx, nce);
2173 *parent_ino = nce->parent_ino;
2174 *parent_gen = nce->parent_gen;
2175 ret = fs_path_add(dest, nce->name, nce->name_len);
2184 * If the inode is not existent yet, add the orphan name and return 1.
2185 * This should only happen for the parent dir that we determine in
2188 ret = is_inode_existent(sctx, ino, gen);
2193 ret = gen_unique_name(sctx, ino, gen, dest);
2201 * Depending on whether the inode was already processed or not, use
2202 * send_root or parent_root for ref lookup.
2204 if (ino < sctx->send_progress)
2205 ret = get_first_ref(sctx->send_root, ino,
2206 parent_ino, parent_gen, dest);
2208 ret = get_first_ref(sctx->parent_root, ino,
2209 parent_ino, parent_gen, dest);
2214 * Check if the ref was overwritten by an inode's ref that was processed
2215 * earlier. If yes, treat as orphan and return 1.
2217 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2218 dest->start, dest->end - dest->start);
2222 fs_path_reset(dest);
2223 ret = gen_unique_name(sctx, ino, gen, dest);
2231 * Store the result of the lookup in the name cache.
2233 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2241 nce->parent_ino = *parent_ino;
2242 nce->parent_gen = *parent_gen;
2243 nce->name_len = fs_path_len(dest);
2245 strcpy(nce->name, dest->start);
2247 if (ino < sctx->send_progress)
2248 nce->need_later_update = 0;
2250 nce->need_later_update = 1;
2252 nce_ret = name_cache_insert(sctx, nce);
2255 name_cache_clean_unused(sctx);
2262 * Magic happens here. This function returns the first ref to an inode as it
2263 * would look like while receiving the stream at this point in time.
2264 * We walk the path up to the root. For every inode in between, we check if it
2265 * was already processed/sent. If yes, we continue with the parent as found
2266 * in send_root. If not, we continue with the parent as found in parent_root.
2267 * If we encounter an inode that was deleted at this point in time, we use the
2268 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2269 * that were not created yet and overwritten inodes/refs.
2271 * When do we have orphan inodes:
2272 * 1. When an inode is freshly created and thus no valid refs are available yet
2273 * 2. When a directory lost all it's refs (deleted) but still has dir items
2274 * inside which were not processed yet (pending for move/delete). If anyone
2275 * tried to get the path to the dir items, it would get a path inside that
2277 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2278 * of an unprocessed inode. If in that case the first ref would be
2279 * overwritten, the overwritten inode gets "orphanized". Later when we
2280 * process this overwritten inode, it is restored at a new place by moving
2283 * sctx->send_progress tells this function at which point in time receiving
2286 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2287 struct fs_path *dest)
2290 struct fs_path *name = NULL;
2291 u64 parent_inode = 0;
2295 name = fs_path_alloc();
2302 fs_path_reset(dest);
2304 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2305 struct waiting_dir_move *wdm;
2307 fs_path_reset(name);
2309 if (is_waiting_for_rm(sctx, ino)) {
2310 ret = gen_unique_name(sctx, ino, gen, name);
2313 ret = fs_path_add_path(dest, name);
2317 wdm = get_waiting_dir_move(sctx, ino);
2318 if (wdm && wdm->orphanized) {
2319 ret = gen_unique_name(sctx, ino, gen, name);
2322 ret = get_first_ref(sctx->parent_root, ino,
2323 &parent_inode, &parent_gen, name);
2325 ret = __get_cur_name_and_parent(sctx, ino, gen,
2335 ret = fs_path_add_path(dest, name);
2346 fs_path_unreverse(dest);
2351 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2353 static int send_subvol_begin(struct send_ctx *sctx)
2356 struct btrfs_root *send_root = sctx->send_root;
2357 struct btrfs_root *parent_root = sctx->parent_root;
2358 struct btrfs_path *path;
2359 struct btrfs_key key;
2360 struct btrfs_root_ref *ref;
2361 struct extent_buffer *leaf;
2365 path = btrfs_alloc_path();
2369 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2371 btrfs_free_path(path);
2375 key.objectid = send_root->root_key.objectid;
2376 key.type = BTRFS_ROOT_BACKREF_KEY;
2379 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2388 leaf = path->nodes[0];
2389 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2390 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2391 key.objectid != send_root->root_key.objectid) {
2395 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2396 namelen = btrfs_root_ref_name_len(leaf, ref);
2397 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2398 btrfs_release_path(path);
2401 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2405 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2410 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2412 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2413 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2414 sctx->send_root->root_item.received_uuid);
2416 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2417 sctx->send_root->root_item.uuid);
2419 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2420 le64_to_cpu(sctx->send_root->root_item.ctransid));
2422 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2423 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2424 parent_root->root_item.received_uuid);
2426 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2427 parent_root->root_item.uuid);
2428 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2429 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2432 ret = send_cmd(sctx);
2436 btrfs_free_path(path);
2441 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2443 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2447 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2449 p = fs_path_alloc();
2453 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2457 ret = get_cur_path(sctx, ino, gen, p);
2460 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2461 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2463 ret = send_cmd(sctx);
2471 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2473 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2477 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2479 p = fs_path_alloc();
2483 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2487 ret = get_cur_path(sctx, ino, gen, p);
2490 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2491 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2493 ret = send_cmd(sctx);
2501 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2503 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2507 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2510 p = fs_path_alloc();
2514 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2518 ret = get_cur_path(sctx, ino, gen, p);
2521 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2522 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2523 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2525 ret = send_cmd(sctx);
2533 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2535 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2537 struct fs_path *p = NULL;
2538 struct btrfs_inode_item *ii;
2539 struct btrfs_path *path = NULL;
2540 struct extent_buffer *eb;
2541 struct btrfs_key key;
2544 btrfs_debug(fs_info, "send_utimes %llu", ino);
2546 p = fs_path_alloc();
2550 path = alloc_path_for_send();
2557 key.type = BTRFS_INODE_ITEM_KEY;
2559 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2565 eb = path->nodes[0];
2566 slot = path->slots[0];
2567 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2569 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2573 ret = get_cur_path(sctx, ino, gen, p);
2576 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2577 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2578 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2579 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2580 /* TODO Add otime support when the otime patches get into upstream */
2582 ret = send_cmd(sctx);
2587 btrfs_free_path(path);
2592 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2593 * a valid path yet because we did not process the refs yet. So, the inode
2594 * is created as orphan.
2596 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2598 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2606 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2608 p = fs_path_alloc();
2612 if (ino != sctx->cur_ino) {
2613 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2618 gen = sctx->cur_inode_gen;
2619 mode = sctx->cur_inode_mode;
2620 rdev = sctx->cur_inode_rdev;
2623 if (S_ISREG(mode)) {
2624 cmd = BTRFS_SEND_C_MKFILE;
2625 } else if (S_ISDIR(mode)) {
2626 cmd = BTRFS_SEND_C_MKDIR;
2627 } else if (S_ISLNK(mode)) {
2628 cmd = BTRFS_SEND_C_SYMLINK;
2629 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2630 cmd = BTRFS_SEND_C_MKNOD;
2631 } else if (S_ISFIFO(mode)) {
2632 cmd = BTRFS_SEND_C_MKFIFO;
2633 } else if (S_ISSOCK(mode)) {
2634 cmd = BTRFS_SEND_C_MKSOCK;
2636 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2637 (int)(mode & S_IFMT));
2642 ret = begin_cmd(sctx, cmd);
2646 ret = gen_unique_name(sctx, ino, gen, p);
2650 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2651 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2653 if (S_ISLNK(mode)) {
2655 ret = read_symlink(sctx->send_root, ino, p);
2658 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2659 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2660 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2661 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2662 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2665 ret = send_cmd(sctx);
2677 * We need some special handling for inodes that get processed before the parent
2678 * directory got created. See process_recorded_refs for details.
2679 * This function does the check if we already created the dir out of order.
2681 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2684 struct btrfs_path *path = NULL;
2685 struct btrfs_key key;
2686 struct btrfs_key found_key;
2687 struct btrfs_key di_key;
2688 struct extent_buffer *eb;
2689 struct btrfs_dir_item *di;
2692 path = alloc_path_for_send();
2699 key.type = BTRFS_DIR_INDEX_KEY;
2701 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2706 eb = path->nodes[0];
2707 slot = path->slots[0];
2708 if (slot >= btrfs_header_nritems(eb)) {
2709 ret = btrfs_next_leaf(sctx->send_root, path);
2712 } else if (ret > 0) {
2719 btrfs_item_key_to_cpu(eb, &found_key, slot);
2720 if (found_key.objectid != key.objectid ||
2721 found_key.type != key.type) {
2726 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2727 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2729 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2730 di_key.objectid < sctx->send_progress) {
2739 btrfs_free_path(path);
2744 * Only creates the inode if it is:
2745 * 1. Not a directory
2746 * 2. Or a directory which was not created already due to out of order
2747 * directories. See did_create_dir and process_recorded_refs for details.
2749 static int send_create_inode_if_needed(struct send_ctx *sctx)
2753 if (S_ISDIR(sctx->cur_inode_mode)) {
2754 ret = did_create_dir(sctx, sctx->cur_ino);
2763 ret = send_create_inode(sctx, sctx->cur_ino);
2771 struct recorded_ref {
2772 struct list_head list;
2774 struct fs_path *full_path;
2780 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2782 ref->full_path = path;
2783 ref->name = (char *)kbasename(ref->full_path->start);
2784 ref->name_len = ref->full_path->end - ref->name;
2788 * We need to process new refs before deleted refs, but compare_tree gives us
2789 * everything mixed. So we first record all refs and later process them.
2790 * This function is a helper to record one ref.
2792 static int __record_ref(struct list_head *head, u64 dir,
2793 u64 dir_gen, struct fs_path *path)
2795 struct recorded_ref *ref;
2797 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2802 ref->dir_gen = dir_gen;
2803 set_ref_path(ref, path);
2804 list_add_tail(&ref->list, head);
2808 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2810 struct recorded_ref *new;
2812 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2816 new->dir = ref->dir;
2817 new->dir_gen = ref->dir_gen;
2818 new->full_path = NULL;
2819 INIT_LIST_HEAD(&new->list);
2820 list_add_tail(&new->list, list);
2824 static void __free_recorded_refs(struct list_head *head)
2826 struct recorded_ref *cur;
2828 while (!list_empty(head)) {
2829 cur = list_entry(head->next, struct recorded_ref, list);
2830 fs_path_free(cur->full_path);
2831 list_del(&cur->list);
2836 static void free_recorded_refs(struct send_ctx *sctx)
2838 __free_recorded_refs(&sctx->new_refs);
2839 __free_recorded_refs(&sctx->deleted_refs);
2843 * Renames/moves a file/dir to its orphan name. Used when the first
2844 * ref of an unprocessed inode gets overwritten and for all non empty
2847 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2848 struct fs_path *path)
2851 struct fs_path *orphan;
2853 orphan = fs_path_alloc();
2857 ret = gen_unique_name(sctx, ino, gen, orphan);
2861 ret = send_rename(sctx, path, orphan);
2864 fs_path_free(orphan);
2868 static struct orphan_dir_info *
2869 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2871 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2872 struct rb_node *parent = NULL;
2873 struct orphan_dir_info *entry, *odi;
2877 entry = rb_entry(parent, struct orphan_dir_info, node);
2878 if (dir_ino < entry->ino) {
2880 } else if (dir_ino > entry->ino) {
2881 p = &(*p)->rb_right;
2887 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2889 return ERR_PTR(-ENOMEM);
2892 odi->last_dir_index_offset = 0;
2894 rb_link_node(&odi->node, parent, p);
2895 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2899 static struct orphan_dir_info *
2900 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2902 struct rb_node *n = sctx->orphan_dirs.rb_node;
2903 struct orphan_dir_info *entry;
2906 entry = rb_entry(n, struct orphan_dir_info, node);
2907 if (dir_ino < entry->ino)
2909 else if (dir_ino > entry->ino)
2917 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2919 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2924 static void free_orphan_dir_info(struct send_ctx *sctx,
2925 struct orphan_dir_info *odi)
2929 rb_erase(&odi->node, &sctx->orphan_dirs);
2934 * Returns 1 if a directory can be removed at this point in time.
2935 * We check this by iterating all dir items and checking if the inode behind
2936 * the dir item was already processed.
2938 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2942 struct btrfs_root *root = sctx->parent_root;
2943 struct btrfs_path *path;
2944 struct btrfs_key key;
2945 struct btrfs_key found_key;
2946 struct btrfs_key loc;
2947 struct btrfs_dir_item *di;
2948 struct orphan_dir_info *odi = NULL;
2951 * Don't try to rmdir the top/root subvolume dir.
2953 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2956 path = alloc_path_for_send();
2961 key.type = BTRFS_DIR_INDEX_KEY;
2964 odi = get_orphan_dir_info(sctx, dir);
2966 key.offset = odi->last_dir_index_offset;
2968 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2973 struct waiting_dir_move *dm;
2975 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2976 ret = btrfs_next_leaf(root, path);
2983 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2985 if (found_key.objectid != key.objectid ||
2986 found_key.type != key.type)
2989 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2990 struct btrfs_dir_item);
2991 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2993 dm = get_waiting_dir_move(sctx, loc.objectid);
2995 odi = add_orphan_dir_info(sctx, dir);
3001 odi->last_dir_index_offset = found_key.offset;
3002 dm->rmdir_ino = dir;
3007 if (loc.objectid > send_progress) {
3008 odi = add_orphan_dir_info(sctx, dir);
3014 odi->last_dir_index_offset = found_key.offset;
3021 free_orphan_dir_info(sctx, odi);
3026 btrfs_free_path(path);
3030 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3032 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3034 return entry != NULL;
3037 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3039 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3040 struct rb_node *parent = NULL;
3041 struct waiting_dir_move *entry, *dm;
3043 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3048 dm->orphanized = orphanized;
3052 entry = rb_entry(parent, struct waiting_dir_move, node);
3053 if (ino < entry->ino) {
3055 } else if (ino > entry->ino) {
3056 p = &(*p)->rb_right;
3063 rb_link_node(&dm->node, parent, p);
3064 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3068 static struct waiting_dir_move *
3069 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3071 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3072 struct waiting_dir_move *entry;
3075 entry = rb_entry(n, struct waiting_dir_move, node);
3076 if (ino < entry->ino)
3078 else if (ino > entry->ino)
3086 static void free_waiting_dir_move(struct send_ctx *sctx,
3087 struct waiting_dir_move *dm)
3091 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3095 static int add_pending_dir_move(struct send_ctx *sctx,
3099 struct list_head *new_refs,
3100 struct list_head *deleted_refs,
3101 const bool is_orphan)
3103 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3104 struct rb_node *parent = NULL;
3105 struct pending_dir_move *entry = NULL, *pm;
3106 struct recorded_ref *cur;
3110 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3113 pm->parent_ino = parent_ino;
3116 INIT_LIST_HEAD(&pm->list);
3117 INIT_LIST_HEAD(&pm->update_refs);
3118 RB_CLEAR_NODE(&pm->node);
3122 entry = rb_entry(parent, struct pending_dir_move, node);
3123 if (parent_ino < entry->parent_ino) {
3125 } else if (parent_ino > entry->parent_ino) {
3126 p = &(*p)->rb_right;
3133 list_for_each_entry(cur, deleted_refs, list) {
3134 ret = dup_ref(cur, &pm->update_refs);
3138 list_for_each_entry(cur, new_refs, list) {
3139 ret = dup_ref(cur, &pm->update_refs);
3144 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3149 list_add_tail(&pm->list, &entry->list);
3151 rb_link_node(&pm->node, parent, p);
3152 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3157 __free_recorded_refs(&pm->update_refs);
3163 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3166 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3167 struct pending_dir_move *entry;
3170 entry = rb_entry(n, struct pending_dir_move, node);
3171 if (parent_ino < entry->parent_ino)
3173 else if (parent_ino > entry->parent_ino)
3181 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3182 u64 ino, u64 gen, u64 *ancestor_ino)
3185 u64 parent_inode = 0;
3187 u64 start_ino = ino;
3190 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3191 fs_path_reset(name);
3193 if (is_waiting_for_rm(sctx, ino))
3195 if (is_waiting_for_move(sctx, ino)) {
3196 if (*ancestor_ino == 0)
3197 *ancestor_ino = ino;
3198 ret = get_first_ref(sctx->parent_root, ino,
3199 &parent_inode, &parent_gen, name);
3201 ret = __get_cur_name_and_parent(sctx, ino, gen,
3211 if (parent_inode == start_ino) {
3213 if (*ancestor_ino == 0)
3214 *ancestor_ino = ino;
3223 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3225 struct fs_path *from_path = NULL;
3226 struct fs_path *to_path = NULL;
3227 struct fs_path *name = NULL;
3228 u64 orig_progress = sctx->send_progress;
3229 struct recorded_ref *cur;
3230 u64 parent_ino, parent_gen;
3231 struct waiting_dir_move *dm = NULL;
3237 name = fs_path_alloc();
3238 from_path = fs_path_alloc();
3239 if (!name || !from_path) {
3244 dm = get_waiting_dir_move(sctx, pm->ino);
3246 rmdir_ino = dm->rmdir_ino;
3247 is_orphan = dm->orphanized;
3248 free_waiting_dir_move(sctx, dm);
3251 ret = gen_unique_name(sctx, pm->ino,
3252 pm->gen, from_path);
3254 ret = get_first_ref(sctx->parent_root, pm->ino,
3255 &parent_ino, &parent_gen, name);
3258 ret = get_cur_path(sctx, parent_ino, parent_gen,
3262 ret = fs_path_add_path(from_path, name);
3267 sctx->send_progress = sctx->cur_ino + 1;
3268 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3272 LIST_HEAD(deleted_refs);
3273 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3274 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3275 &pm->update_refs, &deleted_refs,
3280 dm = get_waiting_dir_move(sctx, pm->ino);
3282 dm->rmdir_ino = rmdir_ino;
3286 fs_path_reset(name);
3289 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3293 ret = send_rename(sctx, from_path, to_path);
3298 struct orphan_dir_info *odi;
3301 odi = get_orphan_dir_info(sctx, rmdir_ino);
3303 /* already deleted */
3308 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3314 name = fs_path_alloc();
3319 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3322 ret = send_rmdir(sctx, name);
3328 ret = send_utimes(sctx, pm->ino, pm->gen);
3333 * After rename/move, need to update the utimes of both new parent(s)
3334 * and old parent(s).
3336 list_for_each_entry(cur, &pm->update_refs, list) {
3338 * The parent inode might have been deleted in the send snapshot
3340 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3341 NULL, NULL, NULL, NULL, NULL);
3342 if (ret == -ENOENT) {
3349 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3356 fs_path_free(from_path);
3357 fs_path_free(to_path);
3358 sctx->send_progress = orig_progress;
3363 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3365 if (!list_empty(&m->list))
3367 if (!RB_EMPTY_NODE(&m->node))
3368 rb_erase(&m->node, &sctx->pending_dir_moves);
3369 __free_recorded_refs(&m->update_refs);
3373 static void tail_append_pending_moves(struct send_ctx *sctx,
3374 struct pending_dir_move *moves,
3375 struct list_head *stack)
3377 if (list_empty(&moves->list)) {
3378 list_add_tail(&moves->list, stack);
3381 list_splice_init(&moves->list, &list);
3382 list_add_tail(&moves->list, stack);
3383 list_splice_tail(&list, stack);
3385 if (!RB_EMPTY_NODE(&moves->node)) {
3386 rb_erase(&moves->node, &sctx->pending_dir_moves);
3387 RB_CLEAR_NODE(&moves->node);
3391 static int apply_children_dir_moves(struct send_ctx *sctx)
3393 struct pending_dir_move *pm;
3394 struct list_head stack;
3395 u64 parent_ino = sctx->cur_ino;
3398 pm = get_pending_dir_moves(sctx, parent_ino);
3402 INIT_LIST_HEAD(&stack);
3403 tail_append_pending_moves(sctx, pm, &stack);
3405 while (!list_empty(&stack)) {
3406 pm = list_first_entry(&stack, struct pending_dir_move, list);
3407 parent_ino = pm->ino;
3408 ret = apply_dir_move(sctx, pm);
3409 free_pending_move(sctx, pm);
3412 pm = get_pending_dir_moves(sctx, parent_ino);
3414 tail_append_pending_moves(sctx, pm, &stack);
3419 while (!list_empty(&stack)) {
3420 pm = list_first_entry(&stack, struct pending_dir_move, list);
3421 free_pending_move(sctx, pm);
3427 * We might need to delay a directory rename even when no ancestor directory
3428 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3429 * renamed. This happens when we rename a directory to the old name (the name
3430 * in the parent root) of some other unrelated directory that got its rename
3431 * delayed due to some ancestor with higher number that got renamed.
3437 * |---- a/ (ino 257)
3438 * | |---- file (ino 260)
3440 * |---- b/ (ino 258)
3441 * |---- c/ (ino 259)
3445 * |---- a/ (ino 258)
3446 * |---- x/ (ino 259)
3447 * |---- y/ (ino 257)
3448 * |----- file (ino 260)
3450 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3451 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3452 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3455 * 1 - rename 259 from 'c' to 'x'
3456 * 2 - rename 257 from 'a' to 'x/y'
3457 * 3 - rename 258 from 'b' to 'a'
3459 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3460 * be done right away and < 0 on error.
3462 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3463 struct recorded_ref *parent_ref,
3464 const bool is_orphan)
3466 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3467 struct btrfs_path *path;
3468 struct btrfs_key key;
3469 struct btrfs_key di_key;
3470 struct btrfs_dir_item *di;
3474 struct waiting_dir_move *wdm;
3476 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3479 path = alloc_path_for_send();
3483 key.objectid = parent_ref->dir;
3484 key.type = BTRFS_DIR_ITEM_KEY;
3485 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3487 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3490 } else if (ret > 0) {
3495 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3496 parent_ref->name_len);
3502 * di_key.objectid has the number of the inode that has a dentry in the
3503 * parent directory with the same name that sctx->cur_ino is being
3504 * renamed to. We need to check if that inode is in the send root as
3505 * well and if it is currently marked as an inode with a pending rename,
3506 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3507 * that it happens after that other inode is renamed.
3509 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3510 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3515 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3516 &left_gen, NULL, NULL, NULL, NULL);
3519 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3520 &right_gen, NULL, NULL, NULL, NULL);
3527 /* Different inode, no need to delay the rename of sctx->cur_ino */
3528 if (right_gen != left_gen) {
3533 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3534 if (wdm && !wdm->orphanized) {
3535 ret = add_pending_dir_move(sctx,
3537 sctx->cur_inode_gen,
3540 &sctx->deleted_refs,
3546 btrfs_free_path(path);
3551 * Check if inode ino2, or any of its ancestors, is inode ino1.
3552 * Return 1 if true, 0 if false and < 0 on error.
3554 static int check_ino_in_path(struct btrfs_root *root,
3559 struct fs_path *fs_path)
3564 return ino1_gen == ino2_gen;
3566 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3571 fs_path_reset(fs_path);
3572 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3576 return parent_gen == ino1_gen;
3583 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3584 * possible path (in case ino2 is not a directory and has multiple hard links).
3585 * Return 1 if true, 0 if false and < 0 on error.
3587 static int is_ancestor(struct btrfs_root *root,
3591 struct fs_path *fs_path)
3593 bool free_fs_path = false;
3595 struct btrfs_path *path = NULL;
3596 struct btrfs_key key;
3599 fs_path = fs_path_alloc();
3602 free_fs_path = true;
3605 path = alloc_path_for_send();
3611 key.objectid = ino2;
3612 key.type = BTRFS_INODE_REF_KEY;
3615 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3620 struct extent_buffer *leaf = path->nodes[0];
3621 int slot = path->slots[0];
3625 if (slot >= btrfs_header_nritems(leaf)) {
3626 ret = btrfs_next_leaf(root, path);
3634 btrfs_item_key_to_cpu(leaf, &key, slot);
3635 if (key.objectid != ino2)
3637 if (key.type != BTRFS_INODE_REF_KEY &&
3638 key.type != BTRFS_INODE_EXTREF_KEY)
3641 item_size = btrfs_item_size_nr(leaf, slot);
3642 while (cur_offset < item_size) {
3646 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3648 struct btrfs_inode_extref *extref;
3650 ptr = btrfs_item_ptr_offset(leaf, slot);
3651 extref = (struct btrfs_inode_extref *)
3653 parent = btrfs_inode_extref_parent(leaf,
3655 cur_offset += sizeof(*extref);
3656 cur_offset += btrfs_inode_extref_name_len(leaf,
3659 parent = key.offset;
3660 cur_offset = item_size;
3663 ret = get_inode_info(root, parent, NULL, &parent_gen,
3664 NULL, NULL, NULL, NULL);
3667 ret = check_ino_in_path(root, ino1, ino1_gen,
3668 parent, parent_gen, fs_path);
3676 btrfs_free_path(path);
3678 fs_path_free(fs_path);
3682 static int wait_for_parent_move(struct send_ctx *sctx,
3683 struct recorded_ref *parent_ref,
3684 const bool is_orphan)
3687 u64 ino = parent_ref->dir;
3688 u64 ino_gen = parent_ref->dir_gen;
3689 u64 parent_ino_before, parent_ino_after;
3690 struct fs_path *path_before = NULL;
3691 struct fs_path *path_after = NULL;
3694 path_after = fs_path_alloc();
3695 path_before = fs_path_alloc();
3696 if (!path_after || !path_before) {
3702 * Our current directory inode may not yet be renamed/moved because some
3703 * ancestor (immediate or not) has to be renamed/moved first. So find if
3704 * such ancestor exists and make sure our own rename/move happens after
3705 * that ancestor is processed to avoid path build infinite loops (done
3706 * at get_cur_path()).
3708 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3709 u64 parent_ino_after_gen;
3711 if (is_waiting_for_move(sctx, ino)) {
3713 * If the current inode is an ancestor of ino in the
3714 * parent root, we need to delay the rename of the
3715 * current inode, otherwise don't delayed the rename
3716 * because we can end up with a circular dependency
3717 * of renames, resulting in some directories never
3718 * getting the respective rename operations issued in
3719 * the send stream or getting into infinite path build
3722 ret = is_ancestor(sctx->parent_root,
3723 sctx->cur_ino, sctx->cur_inode_gen,
3729 fs_path_reset(path_before);
3730 fs_path_reset(path_after);
3732 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3733 &parent_ino_after_gen, path_after);
3736 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3738 if (ret < 0 && ret != -ENOENT) {
3740 } else if (ret == -ENOENT) {
3745 len1 = fs_path_len(path_before);
3746 len2 = fs_path_len(path_after);
3747 if (ino > sctx->cur_ino &&
3748 (parent_ino_before != parent_ino_after || len1 != len2 ||
3749 memcmp(path_before->start, path_after->start, len1))) {
3752 ret = get_inode_info(sctx->parent_root, ino, NULL,
3753 &parent_ino_gen, NULL, NULL, NULL,
3757 if (ino_gen == parent_ino_gen) {
3762 ino = parent_ino_after;
3763 ino_gen = parent_ino_after_gen;
3767 fs_path_free(path_before);
3768 fs_path_free(path_after);
3771 ret = add_pending_dir_move(sctx,
3773 sctx->cur_inode_gen,
3776 &sctx->deleted_refs,
3785 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3788 struct fs_path *new_path;
3791 * Our reference's name member points to its full_path member string, so
3792 * we use here a new path.
3794 new_path = fs_path_alloc();
3798 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3800 fs_path_free(new_path);
3803 ret = fs_path_add(new_path, ref->name, ref->name_len);
3805 fs_path_free(new_path);
3809 fs_path_free(ref->full_path);
3810 set_ref_path(ref, new_path);
3816 * This does all the move/link/unlink/rmdir magic.
3818 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3820 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3822 struct recorded_ref *cur;
3823 struct recorded_ref *cur2;
3824 struct list_head check_dirs;
3825 struct fs_path *valid_path = NULL;
3829 int did_overwrite = 0;
3831 u64 last_dir_ino_rm = 0;
3832 bool can_rename = true;
3833 bool orphanized_dir = false;
3834 bool orphanized_ancestor = false;
3836 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3839 * This should never happen as the root dir always has the same ref
3840 * which is always '..'
3842 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3843 INIT_LIST_HEAD(&check_dirs);
3845 valid_path = fs_path_alloc();
3852 * First, check if the first ref of the current inode was overwritten
3853 * before. If yes, we know that the current inode was already orphanized
3854 * and thus use the orphan name. If not, we can use get_cur_path to
3855 * get the path of the first ref as it would like while receiving at
3856 * this point in time.
3857 * New inodes are always orphan at the beginning, so force to use the
3858 * orphan name in this case.
3859 * The first ref is stored in valid_path and will be updated if it
3860 * gets moved around.
3862 if (!sctx->cur_inode_new) {
3863 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3864 sctx->cur_inode_gen);
3870 if (sctx->cur_inode_new || did_overwrite) {
3871 ret = gen_unique_name(sctx, sctx->cur_ino,
3872 sctx->cur_inode_gen, valid_path);
3877 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3883 list_for_each_entry(cur, &sctx->new_refs, list) {
3885 * We may have refs where the parent directory does not exist
3886 * yet. This happens if the parent directories inum is higher
3887 * than the current inum. To handle this case, we create the
3888 * parent directory out of order. But we need to check if this
3889 * did already happen before due to other refs in the same dir.
3891 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3894 if (ret == inode_state_will_create) {
3897 * First check if any of the current inodes refs did
3898 * already create the dir.
3900 list_for_each_entry(cur2, &sctx->new_refs, list) {
3903 if (cur2->dir == cur->dir) {
3910 * If that did not happen, check if a previous inode
3911 * did already create the dir.
3914 ret = did_create_dir(sctx, cur->dir);
3918 ret = send_create_inode(sctx, cur->dir);
3925 * Check if this new ref would overwrite the first ref of
3926 * another unprocessed inode. If yes, orphanize the
3927 * overwritten inode. If we find an overwritten ref that is
3928 * not the first ref, simply unlink it.
3930 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3931 cur->name, cur->name_len,
3932 &ow_inode, &ow_gen, &ow_mode);
3936 ret = is_first_ref(sctx->parent_root,
3937 ow_inode, cur->dir, cur->name,
3942 struct name_cache_entry *nce;
3943 struct waiting_dir_move *wdm;
3945 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3949 if (S_ISDIR(ow_mode))
3950 orphanized_dir = true;
3953 * If ow_inode has its rename operation delayed
3954 * make sure that its orphanized name is used in
3955 * the source path when performing its rename
3958 if (is_waiting_for_move(sctx, ow_inode)) {
3959 wdm = get_waiting_dir_move(sctx,
3962 wdm->orphanized = true;
3966 * Make sure we clear our orphanized inode's
3967 * name from the name cache. This is because the
3968 * inode ow_inode might be an ancestor of some
3969 * other inode that will be orphanized as well
3970 * later and has an inode number greater than
3971 * sctx->send_progress. We need to prevent
3972 * future name lookups from using the old name
3973 * and get instead the orphan name.
3975 nce = name_cache_search(sctx, ow_inode, ow_gen);
3977 name_cache_delete(sctx, nce);
3982 * ow_inode might currently be an ancestor of
3983 * cur_ino, therefore compute valid_path (the
3984 * current path of cur_ino) again because it
3985 * might contain the pre-orphanization name of
3986 * ow_inode, which is no longer valid.
3988 ret = is_ancestor(sctx->parent_root,
3990 sctx->cur_ino, NULL);
3992 orphanized_ancestor = true;
3993 fs_path_reset(valid_path);
3994 ret = get_cur_path(sctx, sctx->cur_ino,
3995 sctx->cur_inode_gen,
4001 ret = send_unlink(sctx, cur->full_path);
4007 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4008 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4017 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4019 ret = wait_for_parent_move(sctx, cur, is_orphan);
4029 * link/move the ref to the new place. If we have an orphan
4030 * inode, move it and update valid_path. If not, link or move
4031 * it depending on the inode mode.
4033 if (is_orphan && can_rename) {
4034 ret = send_rename(sctx, valid_path, cur->full_path);
4038 ret = fs_path_copy(valid_path, cur->full_path);
4041 } else if (can_rename) {
4042 if (S_ISDIR(sctx->cur_inode_mode)) {
4044 * Dirs can't be linked, so move it. For moved
4045 * dirs, we always have one new and one deleted
4046 * ref. The deleted ref is ignored later.
4048 ret = send_rename(sctx, valid_path,
4051 ret = fs_path_copy(valid_path,
4057 * We might have previously orphanized an inode
4058 * which is an ancestor of our current inode,
4059 * so our reference's full path, which was
4060 * computed before any such orphanizations, must
4063 if (orphanized_dir) {
4064 ret = update_ref_path(sctx, cur);
4068 ret = send_link(sctx, cur->full_path,
4074 ret = dup_ref(cur, &check_dirs);
4079 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4081 * Check if we can already rmdir the directory. If not,
4082 * orphanize it. For every dir item inside that gets deleted
4083 * later, we do this check again and rmdir it then if possible.
4084 * See the use of check_dirs for more details.
4086 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4091 ret = send_rmdir(sctx, valid_path);
4094 } else if (!is_orphan) {
4095 ret = orphanize_inode(sctx, sctx->cur_ino,
4096 sctx->cur_inode_gen, valid_path);
4102 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4103 ret = dup_ref(cur, &check_dirs);
4107 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4108 !list_empty(&sctx->deleted_refs)) {
4110 * We have a moved dir. Add the old parent to check_dirs
4112 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4114 ret = dup_ref(cur, &check_dirs);
4117 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4119 * We have a non dir inode. Go through all deleted refs and
4120 * unlink them if they were not already overwritten by other
4123 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4124 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4125 sctx->cur_ino, sctx->cur_inode_gen,
4126 cur->name, cur->name_len);
4131 * If we orphanized any ancestor before, we need
4132 * to recompute the full path for deleted names,
4133 * since any such path was computed before we
4134 * processed any references and orphanized any
4137 if (orphanized_ancestor) {
4138 ret = update_ref_path(sctx, cur);
4142 ret = send_unlink(sctx, cur->full_path);
4146 ret = dup_ref(cur, &check_dirs);
4151 * If the inode is still orphan, unlink the orphan. This may
4152 * happen when a previous inode did overwrite the first ref
4153 * of this inode and no new refs were added for the current
4154 * inode. Unlinking does not mean that the inode is deleted in
4155 * all cases. There may still be links to this inode in other
4159 ret = send_unlink(sctx, valid_path);
4166 * We did collect all parent dirs where cur_inode was once located. We
4167 * now go through all these dirs and check if they are pending for
4168 * deletion and if it's finally possible to perform the rmdir now.
4169 * We also update the inode stats of the parent dirs here.
4171 list_for_each_entry(cur, &check_dirs, list) {
4173 * In case we had refs into dirs that were not processed yet,
4174 * we don't need to do the utime and rmdir logic for these dirs.
4175 * The dir will be processed later.
4177 if (cur->dir > sctx->cur_ino)
4180 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4184 if (ret == inode_state_did_create ||
4185 ret == inode_state_no_change) {
4186 /* TODO delayed utimes */
4187 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4190 } else if (ret == inode_state_did_delete &&
4191 cur->dir != last_dir_ino_rm) {
4192 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4197 ret = get_cur_path(sctx, cur->dir,
4198 cur->dir_gen, valid_path);
4201 ret = send_rmdir(sctx, valid_path);
4204 last_dir_ino_rm = cur->dir;
4212 __free_recorded_refs(&check_dirs);
4213 free_recorded_refs(sctx);
4214 fs_path_free(valid_path);
4218 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4219 void *ctx, struct list_head *refs)
4222 struct send_ctx *sctx = ctx;
4226 p = fs_path_alloc();
4230 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4235 ret = get_cur_path(sctx, dir, gen, p);
4238 ret = fs_path_add_path(p, name);
4242 ret = __record_ref(refs, dir, gen, p);
4250 static int __record_new_ref(int num, u64 dir, int index,
4251 struct fs_path *name,
4254 struct send_ctx *sctx = ctx;
4255 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4259 static int __record_deleted_ref(int num, u64 dir, int index,
4260 struct fs_path *name,
4263 struct send_ctx *sctx = ctx;
4264 return record_ref(sctx->parent_root, dir, name, ctx,
4265 &sctx->deleted_refs);
4268 static int record_new_ref(struct send_ctx *sctx)
4272 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4273 sctx->cmp_key, 0, __record_new_ref, sctx);
4282 static int record_deleted_ref(struct send_ctx *sctx)
4286 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4287 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4296 struct find_ref_ctx {
4299 struct btrfs_root *root;
4300 struct fs_path *name;
4304 static int __find_iref(int num, u64 dir, int index,
4305 struct fs_path *name,
4308 struct find_ref_ctx *ctx = ctx_;
4312 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4313 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4315 * To avoid doing extra lookups we'll only do this if everything
4318 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4322 if (dir_gen != ctx->dir_gen)
4324 ctx->found_idx = num;
4330 static int find_iref(struct btrfs_root *root,
4331 struct btrfs_path *path,
4332 struct btrfs_key *key,
4333 u64 dir, u64 dir_gen, struct fs_path *name)
4336 struct find_ref_ctx ctx;
4340 ctx.dir_gen = dir_gen;
4344 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4348 if (ctx.found_idx == -1)
4351 return ctx.found_idx;
4354 static int __record_changed_new_ref(int num, u64 dir, int index,
4355 struct fs_path *name,
4360 struct send_ctx *sctx = ctx;
4362 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4367 ret = find_iref(sctx->parent_root, sctx->right_path,
4368 sctx->cmp_key, dir, dir_gen, name);
4370 ret = __record_new_ref(num, dir, index, name, sctx);
4377 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4378 struct fs_path *name,
4383 struct send_ctx *sctx = ctx;
4385 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4390 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4391 dir, dir_gen, name);
4393 ret = __record_deleted_ref(num, dir, index, name, sctx);
4400 static int record_changed_ref(struct send_ctx *sctx)
4404 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4405 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4408 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4409 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4419 * Record and process all refs at once. Needed when an inode changes the
4420 * generation number, which means that it was deleted and recreated.
4422 static int process_all_refs(struct send_ctx *sctx,
4423 enum btrfs_compare_tree_result cmd)
4426 struct btrfs_root *root;
4427 struct btrfs_path *path;
4428 struct btrfs_key key;
4429 struct btrfs_key found_key;
4430 struct extent_buffer *eb;
4432 iterate_inode_ref_t cb;
4433 int pending_move = 0;
4435 path = alloc_path_for_send();
4439 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4440 root = sctx->send_root;
4441 cb = __record_new_ref;
4442 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4443 root = sctx->parent_root;
4444 cb = __record_deleted_ref;
4446 btrfs_err(sctx->send_root->fs_info,
4447 "Wrong command %d in process_all_refs", cmd);
4452 key.objectid = sctx->cmp_key->objectid;
4453 key.type = BTRFS_INODE_REF_KEY;
4455 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4460 eb = path->nodes[0];
4461 slot = path->slots[0];
4462 if (slot >= btrfs_header_nritems(eb)) {
4463 ret = btrfs_next_leaf(root, path);
4471 btrfs_item_key_to_cpu(eb, &found_key, slot);
4473 if (found_key.objectid != key.objectid ||
4474 (found_key.type != BTRFS_INODE_REF_KEY &&
4475 found_key.type != BTRFS_INODE_EXTREF_KEY))
4478 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4484 btrfs_release_path(path);
4487 * We don't actually care about pending_move as we are simply
4488 * re-creating this inode and will be rename'ing it into place once we
4489 * rename the parent directory.
4491 ret = process_recorded_refs(sctx, &pending_move);
4493 btrfs_free_path(path);
4497 static int send_set_xattr(struct send_ctx *sctx,
4498 struct fs_path *path,
4499 const char *name, int name_len,
4500 const char *data, int data_len)
4504 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4508 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4509 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4510 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4512 ret = send_cmd(sctx);
4519 static int send_remove_xattr(struct send_ctx *sctx,
4520 struct fs_path *path,
4521 const char *name, int name_len)
4525 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4529 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4530 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4532 ret = send_cmd(sctx);
4539 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4540 const char *name, int name_len,
4541 const char *data, int data_len,
4545 struct send_ctx *sctx = ctx;
4547 struct posix_acl_xattr_header dummy_acl;
4549 /* Capabilities are emitted by finish_inode_if_needed */
4550 if (!strncmp(name, XATTR_NAME_CAPS, name_len))
4553 p = fs_path_alloc();
4558 * This hack is needed because empty acls are stored as zero byte
4559 * data in xattrs. Problem with that is, that receiving these zero byte
4560 * acls will fail later. To fix this, we send a dummy acl list that
4561 * only contains the version number and no entries.
4563 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4564 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4565 if (data_len == 0) {
4566 dummy_acl.a_version =
4567 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4568 data = (char *)&dummy_acl;
4569 data_len = sizeof(dummy_acl);
4573 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4577 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4584 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4585 const char *name, int name_len,
4586 const char *data, int data_len,
4590 struct send_ctx *sctx = ctx;
4593 p = fs_path_alloc();
4597 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4601 ret = send_remove_xattr(sctx, p, name, name_len);
4608 static int process_new_xattr(struct send_ctx *sctx)
4612 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4613 __process_new_xattr, sctx);
4618 static int process_deleted_xattr(struct send_ctx *sctx)
4620 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4621 __process_deleted_xattr, sctx);
4624 struct find_xattr_ctx {
4632 static int __find_xattr(int num, struct btrfs_key *di_key,
4633 const char *name, int name_len,
4634 const char *data, int data_len,
4635 u8 type, void *vctx)
4637 struct find_xattr_ctx *ctx = vctx;
4639 if (name_len == ctx->name_len &&
4640 strncmp(name, ctx->name, name_len) == 0) {
4641 ctx->found_idx = num;
4642 ctx->found_data_len = data_len;
4643 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4644 if (!ctx->found_data)
4651 static int find_xattr(struct btrfs_root *root,
4652 struct btrfs_path *path,
4653 struct btrfs_key *key,
4654 const char *name, int name_len,
4655 char **data, int *data_len)
4658 struct find_xattr_ctx ctx;
4661 ctx.name_len = name_len;
4663 ctx.found_data = NULL;
4664 ctx.found_data_len = 0;
4666 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4670 if (ctx.found_idx == -1)
4673 *data = ctx.found_data;
4674 *data_len = ctx.found_data_len;
4676 kfree(ctx.found_data);
4678 return ctx.found_idx;
4682 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4683 const char *name, int name_len,
4684 const char *data, int data_len,
4688 struct send_ctx *sctx = ctx;
4689 char *found_data = NULL;
4690 int found_data_len = 0;
4692 ret = find_xattr(sctx->parent_root, sctx->right_path,
4693 sctx->cmp_key, name, name_len, &found_data,
4695 if (ret == -ENOENT) {
4696 ret = __process_new_xattr(num, di_key, name, name_len, data,
4697 data_len, type, ctx);
4698 } else if (ret >= 0) {
4699 if (data_len != found_data_len ||
4700 memcmp(data, found_data, data_len)) {
4701 ret = __process_new_xattr(num, di_key, name, name_len,
4702 data, data_len, type, ctx);
4712 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4713 const char *name, int name_len,
4714 const char *data, int data_len,
4718 struct send_ctx *sctx = ctx;
4720 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4721 name, name_len, NULL, NULL);
4723 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4724 data_len, type, ctx);
4731 static int process_changed_xattr(struct send_ctx *sctx)
4735 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4736 __process_changed_new_xattr, sctx);
4739 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4740 __process_changed_deleted_xattr, sctx);
4746 static int process_all_new_xattrs(struct send_ctx *sctx)
4749 struct btrfs_root *root;
4750 struct btrfs_path *path;
4751 struct btrfs_key key;
4752 struct btrfs_key found_key;
4753 struct extent_buffer *eb;
4756 path = alloc_path_for_send();
4760 root = sctx->send_root;
4762 key.objectid = sctx->cmp_key->objectid;
4763 key.type = BTRFS_XATTR_ITEM_KEY;
4765 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4770 eb = path->nodes[0];
4771 slot = path->slots[0];
4772 if (slot >= btrfs_header_nritems(eb)) {
4773 ret = btrfs_next_leaf(root, path);
4776 } else if (ret > 0) {
4783 btrfs_item_key_to_cpu(eb, &found_key, slot);
4784 if (found_key.objectid != key.objectid ||
4785 found_key.type != key.type) {
4790 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4798 btrfs_free_path(path);
4802 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4804 struct btrfs_root *root = sctx->send_root;
4805 struct btrfs_fs_info *fs_info = root->fs_info;
4806 struct inode *inode;
4809 struct btrfs_key key;
4810 pgoff_t index = offset >> PAGE_SHIFT;
4812 unsigned pg_offset = offset_in_page(offset);
4815 key.objectid = sctx->cur_ino;
4816 key.type = BTRFS_INODE_ITEM_KEY;
4819 inode = btrfs_iget(fs_info->sb, &key, root);
4821 return PTR_ERR(inode);
4823 if (offset + len > i_size_read(inode)) {
4824 if (offset > i_size_read(inode))
4827 len = offset - i_size_read(inode);
4832 last_index = (offset + len - 1) >> PAGE_SHIFT;
4834 /* initial readahead */
4835 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4836 file_ra_state_init(&sctx->ra, inode->i_mapping);
4838 while (index <= last_index) {
4839 unsigned cur_len = min_t(unsigned, len,
4840 PAGE_SIZE - pg_offset);
4842 page = find_lock_page(inode->i_mapping, index);
4844 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4845 NULL, index, last_index + 1 - index);
4847 page = find_or_create_page(inode->i_mapping, index,
4855 if (PageReadahead(page)) {
4856 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4857 NULL, page, index, last_index + 1 - index);
4860 if (!PageUptodate(page)) {
4861 btrfs_readpage(NULL, page);
4863 if (!PageUptodate(page)) {
4872 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4887 * Read some bytes from the current inode/file and send a write command to
4890 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4892 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4895 ssize_t num_read = 0;
4897 p = fs_path_alloc();
4901 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4903 num_read = fill_read_buf(sctx, offset, len);
4904 if (num_read <= 0) {
4910 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4914 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4918 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4919 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4920 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4922 ret = send_cmd(sctx);
4933 * Send a clone command to user space.
4935 static int send_clone(struct send_ctx *sctx,
4936 u64 offset, u32 len,
4937 struct clone_root *clone_root)
4943 btrfs_debug(sctx->send_root->fs_info,
4944 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4945 offset, len, clone_root->root->root_key.objectid,
4946 clone_root->ino, clone_root->offset);
4948 p = fs_path_alloc();
4952 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4956 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4960 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4961 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4962 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4964 if (clone_root->root == sctx->send_root) {
4965 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4966 &gen, NULL, NULL, NULL, NULL);
4969 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4971 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4977 * If the parent we're using has a received_uuid set then use that as
4978 * our clone source as that is what we will look for when doing a
4981 * This covers the case that we create a snapshot off of a received
4982 * subvolume and then use that as the parent and try to receive on a
4985 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4986 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4987 clone_root->root->root_item.received_uuid);
4989 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4990 clone_root->root->root_item.uuid);
4991 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4992 le64_to_cpu(clone_root->root->root_item.ctransid));
4993 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4994 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4995 clone_root->offset);
4997 ret = send_cmd(sctx);
5006 * Send an update extent command to user space.
5008 static int send_update_extent(struct send_ctx *sctx,
5009 u64 offset, u32 len)
5014 p = fs_path_alloc();
5018 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
5022 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5026 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5027 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5028 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5030 ret = send_cmd(sctx);
5038 static int send_hole(struct send_ctx *sctx, u64 end)
5040 struct fs_path *p = NULL;
5041 u64 offset = sctx->cur_inode_last_extent;
5046 * A hole that starts at EOF or beyond it. Since we do not yet support
5047 * fallocate (for extent preallocation and hole punching), sending a
5048 * write of zeroes starting at EOF or beyond would later require issuing
5049 * a truncate operation which would undo the write and achieve nothing.
5051 if (offset >= sctx->cur_inode_size)
5055 * Don't go beyond the inode's i_size due to prealloc extents that start
5058 end = min_t(u64, end, sctx->cur_inode_size);
5060 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5061 return send_update_extent(sctx, offset, end - offset);
5063 p = fs_path_alloc();
5066 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5068 goto tlv_put_failure;
5069 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5070 while (offset < end) {
5071 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5073 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5076 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5077 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5078 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5079 ret = send_cmd(sctx);
5084 sctx->cur_inode_next_write_offset = offset;
5090 static int send_extent_data(struct send_ctx *sctx,
5096 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5097 return send_update_extent(sctx, offset, len);
5099 while (sent < len) {
5100 u64 size = len - sent;
5103 if (size > BTRFS_SEND_READ_SIZE)
5104 size = BTRFS_SEND_READ_SIZE;
5105 ret = send_write(sctx, offset + sent, size);
5116 * Search for a capability xattr related to sctx->cur_ino. If the capability is
5117 * found, call send_set_xattr function to emit it.
5119 * Return 0 if there isn't a capability, or when the capability was emitted
5120 * successfully, or < 0 if an error occurred.
5122 static int send_capabilities(struct send_ctx *sctx)
5124 struct fs_path *fspath = NULL;
5125 struct btrfs_path *path;
5126 struct btrfs_dir_item *di;
5127 struct extent_buffer *leaf;
5128 unsigned long data_ptr;
5133 path = alloc_path_for_send();
5137 di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino,
5138 XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0);
5140 /* There is no xattr for this inode */
5142 } else if (IS_ERR(di)) {
5147 leaf = path->nodes[0];
5148 buf_len = btrfs_dir_data_len(leaf, di);
5150 fspath = fs_path_alloc();
5151 buf = kmalloc(buf_len, GFP_KERNEL);
5152 if (!fspath || !buf) {
5157 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
5161 data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di);
5162 read_extent_buffer(leaf, buf, data_ptr, buf_len);
5164 ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS,
5165 strlen(XATTR_NAME_CAPS), buf, buf_len);
5168 fs_path_free(fspath);
5169 btrfs_free_path(path);
5173 static int clone_range(struct send_ctx *sctx,
5174 struct clone_root *clone_root,
5175 const u64 disk_byte,
5180 struct btrfs_path *path;
5181 struct btrfs_key key;
5183 u64 clone_src_i_size = 0;
5186 * Prevent cloning from a zero offset with a length matching the sector
5187 * size because in some scenarios this will make the receiver fail.
5189 * For example, if in the source filesystem the extent at offset 0
5190 * has a length of sectorsize and it was written using direct IO, then
5191 * it can never be an inline extent (even if compression is enabled).
5192 * Then this extent can be cloned in the original filesystem to a non
5193 * zero file offset, but it may not be possible to clone in the
5194 * destination filesystem because it can be inlined due to compression
5195 * on the destination filesystem (as the receiver's write operations are
5196 * always done using buffered IO). The same happens when the original
5197 * filesystem does not have compression enabled but the destination
5200 if (clone_root->offset == 0 &&
5201 len == sctx->send_root->fs_info->sectorsize)
5202 return send_extent_data(sctx, offset, len);
5204 path = alloc_path_for_send();
5209 * There are inodes that have extents that lie behind its i_size. Don't
5210 * accept clones from these extents.
5212 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5213 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5214 btrfs_release_path(path);
5219 * We can't send a clone operation for the entire range if we find
5220 * extent items in the respective range in the source file that
5221 * refer to different extents or if we find holes.
5222 * So check for that and do a mix of clone and regular write/copy
5223 * operations if needed.
5227 * mkfs.btrfs -f /dev/sda
5228 * mount /dev/sda /mnt
5229 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5230 * cp --reflink=always /mnt/foo /mnt/bar
5231 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5232 * btrfs subvolume snapshot -r /mnt /mnt/snap
5234 * If when we send the snapshot and we are processing file bar (which
5235 * has a higher inode number than foo) we blindly send a clone operation
5236 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5237 * a file bar that matches the content of file foo - iow, doesn't match
5238 * the content from bar in the original filesystem.
5240 key.objectid = clone_root->ino;
5241 key.type = BTRFS_EXTENT_DATA_KEY;
5242 key.offset = clone_root->offset;
5243 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5246 if (ret > 0 && path->slots[0] > 0) {
5247 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5248 if (key.objectid == clone_root->ino &&
5249 key.type == BTRFS_EXTENT_DATA_KEY)
5254 struct extent_buffer *leaf = path->nodes[0];
5255 int slot = path->slots[0];
5256 struct btrfs_file_extent_item *ei;
5260 u64 clone_data_offset;
5262 if (slot >= btrfs_header_nritems(leaf)) {
5263 ret = btrfs_next_leaf(clone_root->root, path);
5271 btrfs_item_key_to_cpu(leaf, &key, slot);
5274 * We might have an implicit trailing hole (NO_HOLES feature
5275 * enabled). We deal with it after leaving this loop.
5277 if (key.objectid != clone_root->ino ||
5278 key.type != BTRFS_EXTENT_DATA_KEY)
5281 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5282 type = btrfs_file_extent_type(leaf, ei);
5283 if (type == BTRFS_FILE_EXTENT_INLINE) {
5284 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5285 ext_len = PAGE_ALIGN(ext_len);
5287 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5290 if (key.offset + ext_len <= clone_root->offset)
5293 if (key.offset > clone_root->offset) {
5294 /* Implicit hole, NO_HOLES feature enabled. */
5295 u64 hole_len = key.offset - clone_root->offset;
5299 ret = send_extent_data(sctx, offset, hole_len);
5307 clone_root->offset += hole_len;
5308 data_offset += hole_len;
5311 if (key.offset >= clone_root->offset + len)
5314 if (key.offset >= clone_src_i_size)
5317 if (key.offset + ext_len > clone_src_i_size)
5318 ext_len = clone_src_i_size - key.offset;
5320 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5321 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5322 clone_root->offset = key.offset;
5323 if (clone_data_offset < data_offset &&
5324 clone_data_offset + ext_len > data_offset) {
5327 extent_offset = data_offset - clone_data_offset;
5328 ext_len -= extent_offset;
5329 clone_data_offset += extent_offset;
5330 clone_root->offset += extent_offset;
5334 clone_len = min_t(u64, ext_len, len);
5336 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5337 clone_data_offset == data_offset) {
5338 const u64 src_end = clone_root->offset + clone_len;
5339 const u64 sectorsize = SZ_64K;
5342 * We can't clone the last block, when its size is not
5343 * sector size aligned, into the middle of a file. If we
5344 * do so, the receiver will get a failure (-EINVAL) when
5345 * trying to clone or will silently corrupt the data in
5346 * the destination file if it's on a kernel without the
5347 * fix introduced by commit ac765f83f1397646
5348 * ("Btrfs: fix data corruption due to cloning of eof
5351 * So issue a clone of the aligned down range plus a
5352 * regular write for the eof block, if we hit that case.
5354 * Also, we use the maximum possible sector size, 64K,
5355 * because we don't know what's the sector size of the
5356 * filesystem that receives the stream, so we have to
5357 * assume the largest possible sector size.
5359 if (src_end == clone_src_i_size &&
5360 !IS_ALIGNED(src_end, sectorsize) &&
5361 offset + clone_len < sctx->cur_inode_size) {
5364 slen = ALIGN_DOWN(src_end - clone_root->offset,
5367 ret = send_clone(sctx, offset, slen,
5372 ret = send_extent_data(sctx, offset + slen,
5375 ret = send_clone(sctx, offset, clone_len,
5379 ret = send_extent_data(sctx, offset, clone_len);
5388 offset += clone_len;
5389 clone_root->offset += clone_len;
5390 data_offset += clone_len;
5396 ret = send_extent_data(sctx, offset, len);
5400 btrfs_free_path(path);
5404 static int send_write_or_clone(struct send_ctx *sctx,
5405 struct btrfs_path *path,
5406 struct btrfs_key *key,
5407 struct clone_root *clone_root)
5410 struct btrfs_file_extent_item *ei;
5411 u64 offset = key->offset;
5414 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5416 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5417 struct btrfs_file_extent_item);
5418 type = btrfs_file_extent_type(path->nodes[0], ei);
5419 if (type == BTRFS_FILE_EXTENT_INLINE) {
5420 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5422 * it is possible the inline item won't cover the whole page,
5423 * but there may be items after this page. Make
5424 * sure to send the whole thing
5426 len = PAGE_ALIGN(len);
5428 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5431 if (offset >= sctx->cur_inode_size) {
5435 if (offset + len > sctx->cur_inode_size)
5436 len = sctx->cur_inode_size - offset;
5442 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5446 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5447 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5448 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5451 ret = send_extent_data(sctx, offset, len);
5453 sctx->cur_inode_next_write_offset = offset + len;
5458 static int is_extent_unchanged(struct send_ctx *sctx,
5459 struct btrfs_path *left_path,
5460 struct btrfs_key *ekey)
5463 struct btrfs_key key;
5464 struct btrfs_path *path = NULL;
5465 struct extent_buffer *eb;
5467 struct btrfs_key found_key;
5468 struct btrfs_file_extent_item *ei;
5473 u64 left_offset_fixed;
5481 path = alloc_path_for_send();
5485 eb = left_path->nodes[0];
5486 slot = left_path->slots[0];
5487 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5488 left_type = btrfs_file_extent_type(eb, ei);
5490 if (left_type != BTRFS_FILE_EXTENT_REG) {
5494 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5495 left_len = btrfs_file_extent_num_bytes(eb, ei);
5496 left_offset = btrfs_file_extent_offset(eb, ei);
5497 left_gen = btrfs_file_extent_generation(eb, ei);
5500 * Following comments will refer to these graphics. L is the left
5501 * extents which we are checking at the moment. 1-8 are the right
5502 * extents that we iterate.
5505 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5508 * |--1--|-2b-|...(same as above)
5510 * Alternative situation. Happens on files where extents got split.
5512 * |-----------7-----------|-6-|
5514 * Alternative situation. Happens on files which got larger.
5517 * Nothing follows after 8.
5520 key.objectid = ekey->objectid;
5521 key.type = BTRFS_EXTENT_DATA_KEY;
5522 key.offset = ekey->offset;
5523 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5532 * Handle special case where the right side has no extents at all.
5534 eb = path->nodes[0];
5535 slot = path->slots[0];
5536 btrfs_item_key_to_cpu(eb, &found_key, slot);
5537 if (found_key.objectid != key.objectid ||
5538 found_key.type != key.type) {
5539 /* If we're a hole then just pretend nothing changed */
5540 ret = (left_disknr) ? 0 : 1;
5545 * We're now on 2a, 2b or 7.
5548 while (key.offset < ekey->offset + left_len) {
5549 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5550 right_type = btrfs_file_extent_type(eb, ei);
5551 if (right_type != BTRFS_FILE_EXTENT_REG &&
5552 right_type != BTRFS_FILE_EXTENT_INLINE) {
5557 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5558 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5559 right_len = PAGE_ALIGN(right_len);
5561 right_len = btrfs_file_extent_num_bytes(eb, ei);
5565 * Are we at extent 8? If yes, we know the extent is changed.
5566 * This may only happen on the first iteration.
5568 if (found_key.offset + right_len <= ekey->offset) {
5569 /* If we're a hole just pretend nothing changed */
5570 ret = (left_disknr) ? 0 : 1;
5575 * We just wanted to see if when we have an inline extent, what
5576 * follows it is a regular extent (wanted to check the above
5577 * condition for inline extents too). This should normally not
5578 * happen but it's possible for example when we have an inline
5579 * compressed extent representing data with a size matching
5580 * the page size (currently the same as sector size).
5582 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5587 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5588 right_offset = btrfs_file_extent_offset(eb, ei);
5589 right_gen = btrfs_file_extent_generation(eb, ei);
5591 left_offset_fixed = left_offset;
5592 if (key.offset < ekey->offset) {
5593 /* Fix the right offset for 2a and 7. */
5594 right_offset += ekey->offset - key.offset;
5596 /* Fix the left offset for all behind 2a and 2b */
5597 left_offset_fixed += key.offset - ekey->offset;
5601 * Check if we have the same extent.
5603 if (left_disknr != right_disknr ||
5604 left_offset_fixed != right_offset ||
5605 left_gen != right_gen) {
5611 * Go to the next extent.
5613 ret = btrfs_next_item(sctx->parent_root, path);
5617 eb = path->nodes[0];
5618 slot = path->slots[0];
5619 btrfs_item_key_to_cpu(eb, &found_key, slot);
5621 if (ret || found_key.objectid != key.objectid ||
5622 found_key.type != key.type) {
5623 key.offset += right_len;
5626 if (found_key.offset != key.offset + right_len) {
5634 * We're now behind the left extent (treat as unchanged) or at the end
5635 * of the right side (treat as changed).
5637 if (key.offset >= ekey->offset + left_len)
5644 btrfs_free_path(path);
5648 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5650 struct btrfs_path *path;
5651 struct btrfs_root *root = sctx->send_root;
5652 struct btrfs_key key;
5655 path = alloc_path_for_send();
5659 sctx->cur_inode_last_extent = 0;
5661 key.objectid = sctx->cur_ino;
5662 key.type = BTRFS_EXTENT_DATA_KEY;
5663 key.offset = offset;
5664 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5668 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5669 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5672 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5674 btrfs_free_path(path);
5678 static int range_is_hole_in_parent(struct send_ctx *sctx,
5682 struct btrfs_path *path;
5683 struct btrfs_key key;
5684 struct btrfs_root *root = sctx->parent_root;
5685 u64 search_start = start;
5688 path = alloc_path_for_send();
5692 key.objectid = sctx->cur_ino;
5693 key.type = BTRFS_EXTENT_DATA_KEY;
5694 key.offset = search_start;
5695 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5698 if (ret > 0 && path->slots[0] > 0)
5701 while (search_start < end) {
5702 struct extent_buffer *leaf = path->nodes[0];
5703 int slot = path->slots[0];
5704 struct btrfs_file_extent_item *fi;
5707 if (slot >= btrfs_header_nritems(leaf)) {
5708 ret = btrfs_next_leaf(root, path);
5716 btrfs_item_key_to_cpu(leaf, &key, slot);
5717 if (key.objectid < sctx->cur_ino ||
5718 key.type < BTRFS_EXTENT_DATA_KEY)
5720 if (key.objectid > sctx->cur_ino ||
5721 key.type > BTRFS_EXTENT_DATA_KEY ||
5725 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5726 extent_end = btrfs_file_extent_end(path);
5727 if (extent_end <= start)
5729 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5730 search_start = extent_end;
5740 btrfs_free_path(path);
5744 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5745 struct btrfs_key *key)
5749 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5752 if (sctx->cur_inode_last_extent == (u64)-1) {
5753 ret = get_last_extent(sctx, key->offset - 1);
5758 if (path->slots[0] == 0 &&
5759 sctx->cur_inode_last_extent < key->offset) {
5761 * We might have skipped entire leafs that contained only
5762 * file extent items for our current inode. These leafs have
5763 * a generation number smaller (older) than the one in the
5764 * current leaf and the leaf our last extent came from, and
5765 * are located between these 2 leafs.
5767 ret = get_last_extent(sctx, key->offset - 1);
5772 if (sctx->cur_inode_last_extent < key->offset) {
5773 ret = range_is_hole_in_parent(sctx,
5774 sctx->cur_inode_last_extent,
5779 ret = send_hole(sctx, key->offset);
5783 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5787 static int process_extent(struct send_ctx *sctx,
5788 struct btrfs_path *path,
5789 struct btrfs_key *key)
5791 struct clone_root *found_clone = NULL;
5794 if (S_ISLNK(sctx->cur_inode_mode))
5797 if (sctx->parent_root && !sctx->cur_inode_new) {
5798 ret = is_extent_unchanged(sctx, path, key);
5806 struct btrfs_file_extent_item *ei;
5809 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5810 struct btrfs_file_extent_item);
5811 type = btrfs_file_extent_type(path->nodes[0], ei);
5812 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5813 type == BTRFS_FILE_EXTENT_REG) {
5815 * The send spec does not have a prealloc command yet,
5816 * so just leave a hole for prealloc'ed extents until
5817 * we have enough commands queued up to justify rev'ing
5820 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5825 /* Have a hole, just skip it. */
5826 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5833 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5834 sctx->cur_inode_size, &found_clone);
5835 if (ret != -ENOENT && ret < 0)
5838 ret = send_write_or_clone(sctx, path, key, found_clone);
5842 ret = maybe_send_hole(sctx, path, key);
5847 static int process_all_extents(struct send_ctx *sctx)
5850 struct btrfs_root *root;
5851 struct btrfs_path *path;
5852 struct btrfs_key key;
5853 struct btrfs_key found_key;
5854 struct extent_buffer *eb;
5857 root = sctx->send_root;
5858 path = alloc_path_for_send();
5862 key.objectid = sctx->cmp_key->objectid;
5863 key.type = BTRFS_EXTENT_DATA_KEY;
5865 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5870 eb = path->nodes[0];
5871 slot = path->slots[0];
5873 if (slot >= btrfs_header_nritems(eb)) {
5874 ret = btrfs_next_leaf(root, path);
5877 } else if (ret > 0) {
5884 btrfs_item_key_to_cpu(eb, &found_key, slot);
5886 if (found_key.objectid != key.objectid ||
5887 found_key.type != key.type) {
5892 ret = process_extent(sctx, path, &found_key);
5900 btrfs_free_path(path);
5904 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5906 int *refs_processed)
5910 if (sctx->cur_ino == 0)
5912 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5913 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5915 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5918 ret = process_recorded_refs(sctx, pending_move);
5922 *refs_processed = 1;
5927 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5938 int need_truncate = 1;
5939 int pending_move = 0;
5940 int refs_processed = 0;
5942 if (sctx->ignore_cur_inode)
5945 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5951 * We have processed the refs and thus need to advance send_progress.
5952 * Now, calls to get_cur_xxx will take the updated refs of the current
5953 * inode into account.
5955 * On the other hand, if our current inode is a directory and couldn't
5956 * be moved/renamed because its parent was renamed/moved too and it has
5957 * a higher inode number, we can only move/rename our current inode
5958 * after we moved/renamed its parent. Therefore in this case operate on
5959 * the old path (pre move/rename) of our current inode, and the
5960 * move/rename will be performed later.
5962 if (refs_processed && !pending_move)
5963 sctx->send_progress = sctx->cur_ino + 1;
5965 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5967 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5970 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5971 &left_mode, &left_uid, &left_gid, NULL);
5975 if (!sctx->parent_root || sctx->cur_inode_new) {
5977 if (!S_ISLNK(sctx->cur_inode_mode))
5979 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5984 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5985 &old_size, NULL, &right_mode, &right_uid,
5990 if (left_uid != right_uid || left_gid != right_gid)
5992 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5994 if ((old_size == sctx->cur_inode_size) ||
5995 (sctx->cur_inode_size > old_size &&
5996 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
6000 if (S_ISREG(sctx->cur_inode_mode)) {
6001 if (need_send_hole(sctx)) {
6002 if (sctx->cur_inode_last_extent == (u64)-1 ||
6003 sctx->cur_inode_last_extent <
6004 sctx->cur_inode_size) {
6005 ret = get_last_extent(sctx, (u64)-1);
6009 if (sctx->cur_inode_last_extent <
6010 sctx->cur_inode_size) {
6011 ret = send_hole(sctx, sctx->cur_inode_size);
6016 if (need_truncate) {
6017 ret = send_truncate(sctx, sctx->cur_ino,
6018 sctx->cur_inode_gen,
6019 sctx->cur_inode_size);
6026 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6027 left_uid, left_gid);
6032 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6038 ret = send_capabilities(sctx);
6043 * If other directory inodes depended on our current directory
6044 * inode's move/rename, now do their move/rename operations.
6046 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
6047 ret = apply_children_dir_moves(sctx);
6051 * Need to send that every time, no matter if it actually
6052 * changed between the two trees as we have done changes to
6053 * the inode before. If our inode is a directory and it's
6054 * waiting to be moved/renamed, we will send its utimes when
6055 * it's moved/renamed, therefore we don't need to do it here.
6057 sctx->send_progress = sctx->cur_ino + 1;
6058 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6067 struct parent_paths_ctx {
6068 struct list_head *refs;
6069 struct send_ctx *sctx;
6072 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
6075 struct parent_paths_ctx *ppctx = ctx;
6077 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6082 * Issue unlink operations for all paths of the current inode found in the
6085 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6087 LIST_HEAD(deleted_refs);
6088 struct btrfs_path *path;
6089 struct btrfs_key key;
6090 struct parent_paths_ctx ctx;
6093 path = alloc_path_for_send();
6097 key.objectid = sctx->cur_ino;
6098 key.type = BTRFS_INODE_REF_KEY;
6100 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6104 ctx.refs = &deleted_refs;
6108 struct extent_buffer *eb = path->nodes[0];
6109 int slot = path->slots[0];
6111 if (slot >= btrfs_header_nritems(eb)) {
6112 ret = btrfs_next_leaf(sctx->parent_root, path);
6120 btrfs_item_key_to_cpu(eb, &key, slot);
6121 if (key.objectid != sctx->cur_ino)
6123 if (key.type != BTRFS_INODE_REF_KEY &&
6124 key.type != BTRFS_INODE_EXTREF_KEY)
6127 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6128 record_parent_ref, &ctx);
6135 while (!list_empty(&deleted_refs)) {
6136 struct recorded_ref *ref;
6138 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6139 ret = send_unlink(sctx, ref->full_path);
6142 fs_path_free(ref->full_path);
6143 list_del(&ref->list);
6148 btrfs_free_path(path);
6150 __free_recorded_refs(&deleted_refs);
6154 static int changed_inode(struct send_ctx *sctx,
6155 enum btrfs_compare_tree_result result)
6158 struct btrfs_key *key = sctx->cmp_key;
6159 struct btrfs_inode_item *left_ii = NULL;
6160 struct btrfs_inode_item *right_ii = NULL;
6164 sctx->cur_ino = key->objectid;
6165 sctx->cur_inode_new_gen = 0;
6166 sctx->cur_inode_last_extent = (u64)-1;
6167 sctx->cur_inode_next_write_offset = 0;
6168 sctx->ignore_cur_inode = false;
6171 * Set send_progress to current inode. This will tell all get_cur_xxx
6172 * functions that the current inode's refs are not updated yet. Later,
6173 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6175 sctx->send_progress = sctx->cur_ino;
6177 if (result == BTRFS_COMPARE_TREE_NEW ||
6178 result == BTRFS_COMPARE_TREE_CHANGED) {
6179 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6180 sctx->left_path->slots[0],
6181 struct btrfs_inode_item);
6182 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6185 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6186 sctx->right_path->slots[0],
6187 struct btrfs_inode_item);
6188 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6191 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6192 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6193 sctx->right_path->slots[0],
6194 struct btrfs_inode_item);
6196 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6200 * The cur_ino = root dir case is special here. We can't treat
6201 * the inode as deleted+reused because it would generate a
6202 * stream that tries to delete/mkdir the root dir.
6204 if (left_gen != right_gen &&
6205 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6206 sctx->cur_inode_new_gen = 1;
6210 * Normally we do not find inodes with a link count of zero (orphans)
6211 * because the most common case is to create a snapshot and use it
6212 * for a send operation. However other less common use cases involve
6213 * using a subvolume and send it after turning it to RO mode just
6214 * after deleting all hard links of a file while holding an open
6215 * file descriptor against it or turning a RO snapshot into RW mode,
6216 * keep an open file descriptor against a file, delete it and then
6217 * turn the snapshot back to RO mode before using it for a send
6218 * operation. So if we find such cases, ignore the inode and all its
6219 * items completely if it's a new inode, or if it's a changed inode
6220 * make sure all its previous paths (from the parent snapshot) are all
6221 * unlinked and all other the inode items are ignored.
6223 if (result == BTRFS_COMPARE_TREE_NEW ||
6224 result == BTRFS_COMPARE_TREE_CHANGED) {
6227 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6229 sctx->ignore_cur_inode = true;
6230 if (result == BTRFS_COMPARE_TREE_CHANGED)
6231 ret = btrfs_unlink_all_paths(sctx);
6236 if (result == BTRFS_COMPARE_TREE_NEW) {
6237 sctx->cur_inode_gen = left_gen;
6238 sctx->cur_inode_new = 1;
6239 sctx->cur_inode_deleted = 0;
6240 sctx->cur_inode_size = btrfs_inode_size(
6241 sctx->left_path->nodes[0], left_ii);
6242 sctx->cur_inode_mode = btrfs_inode_mode(
6243 sctx->left_path->nodes[0], left_ii);
6244 sctx->cur_inode_rdev = btrfs_inode_rdev(
6245 sctx->left_path->nodes[0], left_ii);
6246 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6247 ret = send_create_inode_if_needed(sctx);
6248 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6249 sctx->cur_inode_gen = right_gen;
6250 sctx->cur_inode_new = 0;
6251 sctx->cur_inode_deleted = 1;
6252 sctx->cur_inode_size = btrfs_inode_size(
6253 sctx->right_path->nodes[0], right_ii);
6254 sctx->cur_inode_mode = btrfs_inode_mode(
6255 sctx->right_path->nodes[0], right_ii);
6256 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6258 * We need to do some special handling in case the inode was
6259 * reported as changed with a changed generation number. This
6260 * means that the original inode was deleted and new inode
6261 * reused the same inum. So we have to treat the old inode as
6262 * deleted and the new one as new.
6264 if (sctx->cur_inode_new_gen) {
6266 * First, process the inode as if it was deleted.
6268 sctx->cur_inode_gen = right_gen;
6269 sctx->cur_inode_new = 0;
6270 sctx->cur_inode_deleted = 1;
6271 sctx->cur_inode_size = btrfs_inode_size(
6272 sctx->right_path->nodes[0], right_ii);
6273 sctx->cur_inode_mode = btrfs_inode_mode(
6274 sctx->right_path->nodes[0], right_ii);
6275 ret = process_all_refs(sctx,
6276 BTRFS_COMPARE_TREE_DELETED);
6281 * Now process the inode as if it was new.
6283 sctx->cur_inode_gen = left_gen;
6284 sctx->cur_inode_new = 1;
6285 sctx->cur_inode_deleted = 0;
6286 sctx->cur_inode_size = btrfs_inode_size(
6287 sctx->left_path->nodes[0], left_ii);
6288 sctx->cur_inode_mode = btrfs_inode_mode(
6289 sctx->left_path->nodes[0], left_ii);
6290 sctx->cur_inode_rdev = btrfs_inode_rdev(
6291 sctx->left_path->nodes[0], left_ii);
6292 ret = send_create_inode_if_needed(sctx);
6296 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6300 * Advance send_progress now as we did not get into
6301 * process_recorded_refs_if_needed in the new_gen case.
6303 sctx->send_progress = sctx->cur_ino + 1;
6306 * Now process all extents and xattrs of the inode as if
6307 * they were all new.
6309 ret = process_all_extents(sctx);
6312 ret = process_all_new_xattrs(sctx);
6316 sctx->cur_inode_gen = left_gen;
6317 sctx->cur_inode_new = 0;
6318 sctx->cur_inode_new_gen = 0;
6319 sctx->cur_inode_deleted = 0;
6320 sctx->cur_inode_size = btrfs_inode_size(
6321 sctx->left_path->nodes[0], left_ii);
6322 sctx->cur_inode_mode = btrfs_inode_mode(
6323 sctx->left_path->nodes[0], left_ii);
6332 * We have to process new refs before deleted refs, but compare_trees gives us
6333 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6334 * first and later process them in process_recorded_refs.
6335 * For the cur_inode_new_gen case, we skip recording completely because
6336 * changed_inode did already initiate processing of refs. The reason for this is
6337 * that in this case, compare_tree actually compares the refs of 2 different
6338 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6339 * refs of the right tree as deleted and all refs of the left tree as new.
6341 static int changed_ref(struct send_ctx *sctx,
6342 enum btrfs_compare_tree_result result)
6346 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6347 inconsistent_snapshot_error(sctx, result, "reference");
6351 if (!sctx->cur_inode_new_gen &&
6352 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6353 if (result == BTRFS_COMPARE_TREE_NEW)
6354 ret = record_new_ref(sctx);
6355 else if (result == BTRFS_COMPARE_TREE_DELETED)
6356 ret = record_deleted_ref(sctx);
6357 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6358 ret = record_changed_ref(sctx);
6365 * Process new/deleted/changed xattrs. We skip processing in the
6366 * cur_inode_new_gen case because changed_inode did already initiate processing
6367 * of xattrs. The reason is the same as in changed_ref
6369 static int changed_xattr(struct send_ctx *sctx,
6370 enum btrfs_compare_tree_result result)
6374 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6375 inconsistent_snapshot_error(sctx, result, "xattr");
6379 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6380 if (result == BTRFS_COMPARE_TREE_NEW)
6381 ret = process_new_xattr(sctx);
6382 else if (result == BTRFS_COMPARE_TREE_DELETED)
6383 ret = process_deleted_xattr(sctx);
6384 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6385 ret = process_changed_xattr(sctx);
6392 * Process new/deleted/changed extents. We skip processing in the
6393 * cur_inode_new_gen case because changed_inode did already initiate processing
6394 * of extents. The reason is the same as in changed_ref
6396 static int changed_extent(struct send_ctx *sctx,
6397 enum btrfs_compare_tree_result result)
6402 * We have found an extent item that changed without the inode item
6403 * having changed. This can happen either after relocation (where the
6404 * disk_bytenr of an extent item is replaced at
6405 * relocation.c:replace_file_extents()) or after deduplication into a
6406 * file in both the parent and send snapshots (where an extent item can
6407 * get modified or replaced with a new one). Note that deduplication
6408 * updates the inode item, but it only changes the iversion (sequence
6409 * field in the inode item) of the inode, so if a file is deduplicated
6410 * the same amount of times in both the parent and send snapshots, its
6411 * iversion becames the same in both snapshots, whence the inode item is
6412 * the same on both snapshots.
6414 if (sctx->cur_ino != sctx->cmp_key->objectid)
6417 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6418 if (result != BTRFS_COMPARE_TREE_DELETED)
6419 ret = process_extent(sctx, sctx->left_path,
6426 static int dir_changed(struct send_ctx *sctx, u64 dir)
6428 u64 orig_gen, new_gen;
6431 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6436 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6441 return (orig_gen != new_gen) ? 1 : 0;
6444 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6445 struct btrfs_key *key)
6447 struct btrfs_inode_extref *extref;
6448 struct extent_buffer *leaf;
6449 u64 dirid = 0, last_dirid = 0;
6456 /* Easy case, just check this one dirid */
6457 if (key->type == BTRFS_INODE_REF_KEY) {
6458 dirid = key->offset;
6460 ret = dir_changed(sctx, dirid);
6464 leaf = path->nodes[0];
6465 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6466 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6467 while (cur_offset < item_size) {
6468 extref = (struct btrfs_inode_extref *)(ptr +
6470 dirid = btrfs_inode_extref_parent(leaf, extref);
6471 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6472 cur_offset += ref_name_len + sizeof(*extref);
6473 if (dirid == last_dirid)
6475 ret = dir_changed(sctx, dirid);
6485 * Updates compare related fields in sctx and simply forwards to the actual
6486 * changed_xxx functions.
6488 static int changed_cb(struct btrfs_path *left_path,
6489 struct btrfs_path *right_path,
6490 struct btrfs_key *key,
6491 enum btrfs_compare_tree_result result,
6495 struct send_ctx *sctx = ctx;
6497 if (result == BTRFS_COMPARE_TREE_SAME) {
6498 if (key->type == BTRFS_INODE_REF_KEY ||
6499 key->type == BTRFS_INODE_EXTREF_KEY) {
6500 ret = compare_refs(sctx, left_path, key);
6505 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6506 return maybe_send_hole(sctx, left_path, key);
6510 result = BTRFS_COMPARE_TREE_CHANGED;
6514 sctx->left_path = left_path;
6515 sctx->right_path = right_path;
6516 sctx->cmp_key = key;
6518 ret = finish_inode_if_needed(sctx, 0);
6522 /* Ignore non-FS objects */
6523 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6524 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6527 if (key->type == BTRFS_INODE_ITEM_KEY) {
6528 ret = changed_inode(sctx, result);
6529 } else if (!sctx->ignore_cur_inode) {
6530 if (key->type == BTRFS_INODE_REF_KEY ||
6531 key->type == BTRFS_INODE_EXTREF_KEY)
6532 ret = changed_ref(sctx, result);
6533 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6534 ret = changed_xattr(sctx, result);
6535 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6536 ret = changed_extent(sctx, result);
6543 static int full_send_tree(struct send_ctx *sctx)
6546 struct btrfs_root *send_root = sctx->send_root;
6547 struct btrfs_key key;
6548 struct btrfs_path *path;
6549 struct extent_buffer *eb;
6552 path = alloc_path_for_send();
6556 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6557 key.type = BTRFS_INODE_ITEM_KEY;
6560 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6567 eb = path->nodes[0];
6568 slot = path->slots[0];
6569 btrfs_item_key_to_cpu(eb, &key, slot);
6571 ret = changed_cb(path, NULL, &key,
6572 BTRFS_COMPARE_TREE_NEW, sctx);
6576 ret = btrfs_next_item(send_root, path);
6586 ret = finish_inode_if_needed(sctx, 1);
6589 btrfs_free_path(path);
6593 static int tree_move_down(struct btrfs_path *path, int *level)
6595 struct extent_buffer *eb;
6597 BUG_ON(*level == 0);
6598 eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]);
6602 path->nodes[*level - 1] = eb;
6603 path->slots[*level - 1] = 0;
6608 static int tree_move_next_or_upnext(struct btrfs_path *path,
6609 int *level, int root_level)
6613 nritems = btrfs_header_nritems(path->nodes[*level]);
6615 path->slots[*level]++;
6617 while (path->slots[*level] >= nritems) {
6618 if (*level == root_level)
6622 path->slots[*level] = 0;
6623 free_extent_buffer(path->nodes[*level]);
6624 path->nodes[*level] = NULL;
6626 path->slots[*level]++;
6628 nritems = btrfs_header_nritems(path->nodes[*level]);
6635 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
6638 static int tree_advance(struct btrfs_path *path,
6639 int *level, int root_level,
6641 struct btrfs_key *key)
6645 if (*level == 0 || !allow_down) {
6646 ret = tree_move_next_or_upnext(path, level, root_level);
6648 ret = tree_move_down(path, level);
6652 btrfs_item_key_to_cpu(path->nodes[*level], key,
6653 path->slots[*level]);
6655 btrfs_node_key_to_cpu(path->nodes[*level], key,
6656 path->slots[*level]);
6661 static int tree_compare_item(struct btrfs_path *left_path,
6662 struct btrfs_path *right_path,
6667 unsigned long off1, off2;
6669 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
6670 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
6674 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
6675 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
6676 right_path->slots[0]);
6678 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
6680 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
6687 * This function compares two trees and calls the provided callback for
6688 * every changed/new/deleted item it finds.
6689 * If shared tree blocks are encountered, whole subtrees are skipped, making
6690 * the compare pretty fast on snapshotted subvolumes.
6692 * This currently works on commit roots only. As commit roots are read only,
6693 * we don't do any locking. The commit roots are protected with transactions.
6694 * Transactions are ended and rejoined when a commit is tried in between.
6696 * This function checks for modifications done to the trees while comparing.
6697 * If it detects a change, it aborts immediately.
6699 static int btrfs_compare_trees(struct btrfs_root *left_root,
6700 struct btrfs_root *right_root,
6701 btrfs_changed_cb_t changed_cb, void *ctx)
6703 struct btrfs_fs_info *fs_info = left_root->fs_info;
6706 struct btrfs_path *left_path = NULL;
6707 struct btrfs_path *right_path = NULL;
6708 struct btrfs_key left_key;
6709 struct btrfs_key right_key;
6710 char *tmp_buf = NULL;
6711 int left_root_level;
6712 int right_root_level;
6715 int left_end_reached;
6716 int right_end_reached;
6724 left_path = btrfs_alloc_path();
6729 right_path = btrfs_alloc_path();
6735 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
6741 left_path->search_commit_root = 1;
6742 left_path->skip_locking = 1;
6743 right_path->search_commit_root = 1;
6744 right_path->skip_locking = 1;
6747 * Strategy: Go to the first items of both trees. Then do
6749 * If both trees are at level 0
6750 * Compare keys of current items
6751 * If left < right treat left item as new, advance left tree
6753 * If left > right treat right item as deleted, advance right tree
6755 * If left == right do deep compare of items, treat as changed if
6756 * needed, advance both trees and repeat
6757 * If both trees are at the same level but not at level 0
6758 * Compare keys of current nodes/leafs
6759 * If left < right advance left tree and repeat
6760 * If left > right advance right tree and repeat
6761 * If left == right compare blockptrs of the next nodes/leafs
6762 * If they match advance both trees but stay at the same level
6764 * If they don't match advance both trees while allowing to go
6766 * If tree levels are different
6767 * Advance the tree that needs it and repeat
6769 * Advancing a tree means:
6770 * If we are at level 0, try to go to the next slot. If that's not
6771 * possible, go one level up and repeat. Stop when we found a level
6772 * where we could go to the next slot. We may at this point be on a
6775 * If we are not at level 0 and not on shared tree blocks, go one
6778 * If we are not at level 0 and on shared tree blocks, go one slot to
6779 * the right if possible or go up and right.
6782 down_read(&fs_info->commit_root_sem);
6783 left_level = btrfs_header_level(left_root->commit_root);
6784 left_root_level = left_level;
6785 left_path->nodes[left_level] =
6786 btrfs_clone_extent_buffer(left_root->commit_root);
6787 if (!left_path->nodes[left_level]) {
6788 up_read(&fs_info->commit_root_sem);
6793 right_level = btrfs_header_level(right_root->commit_root);
6794 right_root_level = right_level;
6795 right_path->nodes[right_level] =
6796 btrfs_clone_extent_buffer(right_root->commit_root);
6797 if (!right_path->nodes[right_level]) {
6798 up_read(&fs_info->commit_root_sem);
6802 up_read(&fs_info->commit_root_sem);
6804 if (left_level == 0)
6805 btrfs_item_key_to_cpu(left_path->nodes[left_level],
6806 &left_key, left_path->slots[left_level]);
6808 btrfs_node_key_to_cpu(left_path->nodes[left_level],
6809 &left_key, left_path->slots[left_level]);
6810 if (right_level == 0)
6811 btrfs_item_key_to_cpu(right_path->nodes[right_level],
6812 &right_key, right_path->slots[right_level]);
6814 btrfs_node_key_to_cpu(right_path->nodes[right_level],
6815 &right_key, right_path->slots[right_level]);
6817 left_end_reached = right_end_reached = 0;
6818 advance_left = advance_right = 0;
6822 if (advance_left && !left_end_reached) {
6823 ret = tree_advance(left_path, &left_level,
6825 advance_left != ADVANCE_ONLY_NEXT,
6828 left_end_reached = ADVANCE;
6833 if (advance_right && !right_end_reached) {
6834 ret = tree_advance(right_path, &right_level,
6836 advance_right != ADVANCE_ONLY_NEXT,
6839 right_end_reached = ADVANCE;
6845 if (left_end_reached && right_end_reached) {
6848 } else if (left_end_reached) {
6849 if (right_level == 0) {
6850 ret = changed_cb(left_path, right_path,
6852 BTRFS_COMPARE_TREE_DELETED,
6857 advance_right = ADVANCE;
6859 } else if (right_end_reached) {
6860 if (left_level == 0) {
6861 ret = changed_cb(left_path, right_path,
6863 BTRFS_COMPARE_TREE_NEW,
6868 advance_left = ADVANCE;
6872 if (left_level == 0 && right_level == 0) {
6873 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6875 ret = changed_cb(left_path, right_path,
6877 BTRFS_COMPARE_TREE_NEW,
6881 advance_left = ADVANCE;
6882 } else if (cmp > 0) {
6883 ret = changed_cb(left_path, right_path,
6885 BTRFS_COMPARE_TREE_DELETED,
6889 advance_right = ADVANCE;
6891 enum btrfs_compare_tree_result result;
6893 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
6894 ret = tree_compare_item(left_path, right_path,
6897 result = BTRFS_COMPARE_TREE_CHANGED;
6899 result = BTRFS_COMPARE_TREE_SAME;
6900 ret = changed_cb(left_path, right_path,
6901 &left_key, result, ctx);
6904 advance_left = ADVANCE;
6905 advance_right = ADVANCE;
6907 } else if (left_level == right_level) {
6908 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6910 advance_left = ADVANCE;
6911 } else if (cmp > 0) {
6912 advance_right = ADVANCE;
6914 left_blockptr = btrfs_node_blockptr(
6915 left_path->nodes[left_level],
6916 left_path->slots[left_level]);
6917 right_blockptr = btrfs_node_blockptr(
6918 right_path->nodes[right_level],
6919 right_path->slots[right_level]);
6920 left_gen = btrfs_node_ptr_generation(
6921 left_path->nodes[left_level],
6922 left_path->slots[left_level]);
6923 right_gen = btrfs_node_ptr_generation(
6924 right_path->nodes[right_level],
6925 right_path->slots[right_level]);
6926 if (left_blockptr == right_blockptr &&
6927 left_gen == right_gen) {
6929 * As we're on a shared block, don't
6930 * allow to go deeper.
6932 advance_left = ADVANCE_ONLY_NEXT;
6933 advance_right = ADVANCE_ONLY_NEXT;
6935 advance_left = ADVANCE;
6936 advance_right = ADVANCE;
6939 } else if (left_level < right_level) {
6940 advance_right = ADVANCE;
6942 advance_left = ADVANCE;
6947 btrfs_free_path(left_path);
6948 btrfs_free_path(right_path);
6953 static int send_subvol(struct send_ctx *sctx)
6957 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6958 ret = send_header(sctx);
6963 ret = send_subvol_begin(sctx);
6967 if (sctx->parent_root) {
6968 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6972 ret = finish_inode_if_needed(sctx, 1);
6976 ret = full_send_tree(sctx);
6982 free_recorded_refs(sctx);
6987 * If orphan cleanup did remove any orphans from a root, it means the tree
6988 * was modified and therefore the commit root is not the same as the current
6989 * root anymore. This is a problem, because send uses the commit root and
6990 * therefore can see inode items that don't exist in the current root anymore,
6991 * and for example make calls to btrfs_iget, which will do tree lookups based
6992 * on the current root and not on the commit root. Those lookups will fail,
6993 * returning a -ESTALE error, and making send fail with that error. So make
6994 * sure a send does not see any orphans we have just removed, and that it will
6995 * see the same inodes regardless of whether a transaction commit happened
6996 * before it started (meaning that the commit root will be the same as the
6997 * current root) or not.
6999 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
7002 struct btrfs_trans_handle *trans = NULL;
7005 if (sctx->parent_root &&
7006 sctx->parent_root->node != sctx->parent_root->commit_root)
7009 for (i = 0; i < sctx->clone_roots_cnt; i++)
7010 if (sctx->clone_roots[i].root->node !=
7011 sctx->clone_roots[i].root->commit_root)
7015 return btrfs_end_transaction(trans);
7020 /* Use any root, all fs roots will get their commit roots updated. */
7022 trans = btrfs_join_transaction(sctx->send_root);
7024 return PTR_ERR(trans);
7028 return btrfs_commit_transaction(trans);
7032 * Make sure any existing dellaloc is flushed for any root used by a send
7033 * operation so that we do not miss any data and we do not race with writeback
7034 * finishing and changing a tree while send is using the tree. This could
7035 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
7036 * a send operation then uses the subvolume.
7037 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
7039 static int flush_delalloc_roots(struct send_ctx *sctx)
7041 struct btrfs_root *root = sctx->parent_root;
7046 ret = btrfs_start_delalloc_snapshot(root);
7049 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7052 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7053 root = sctx->clone_roots[i].root;
7054 ret = btrfs_start_delalloc_snapshot(root);
7057 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7063 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
7065 spin_lock(&root->root_item_lock);
7066 root->send_in_progress--;
7068 * Not much left to do, we don't know why it's unbalanced and
7069 * can't blindly reset it to 0.
7071 if (root->send_in_progress < 0)
7072 btrfs_err(root->fs_info,
7073 "send_in_progress unbalanced %d root %llu",
7074 root->send_in_progress, root->root_key.objectid);
7075 spin_unlock(&root->root_item_lock);
7078 static void dedupe_in_progress_warn(const struct btrfs_root *root)
7080 btrfs_warn_rl(root->fs_info,
7081 "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
7082 root->root_key.objectid, root->dedupe_in_progress);
7085 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
7088 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
7089 struct btrfs_fs_info *fs_info = send_root->fs_info;
7090 struct btrfs_root *clone_root;
7091 struct btrfs_key key;
7092 struct send_ctx *sctx = NULL;
7094 u64 *clone_sources_tmp = NULL;
7095 int clone_sources_to_rollback = 0;
7096 unsigned alloc_size;
7097 int sort_clone_roots = 0;
7099 if (!capable(CAP_SYS_ADMIN))
7103 * The subvolume must remain read-only during send, protect against
7104 * making it RW. This also protects against deletion.
7106 spin_lock(&send_root->root_item_lock);
7107 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
7108 dedupe_in_progress_warn(send_root);
7109 spin_unlock(&send_root->root_item_lock);
7112 send_root->send_in_progress++;
7113 spin_unlock(&send_root->root_item_lock);
7116 * Userspace tools do the checks and warn the user if it's
7119 if (!btrfs_root_readonly(send_root)) {
7125 * Check that we don't overflow at later allocations, we request
7126 * clone_sources_count + 1 items, and compare to unsigned long inside
7129 if (arg->clone_sources_count >
7130 ULONG_MAX / sizeof(struct clone_root) - 1) {
7135 if (!access_ok(arg->clone_sources,
7136 sizeof(*arg->clone_sources) *
7137 arg->clone_sources_count)) {
7142 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
7147 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
7153 INIT_LIST_HEAD(&sctx->new_refs);
7154 INIT_LIST_HEAD(&sctx->deleted_refs);
7155 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
7156 INIT_LIST_HEAD(&sctx->name_cache_list);
7158 sctx->flags = arg->flags;
7160 sctx->send_filp = fget(arg->send_fd);
7161 if (!sctx->send_filp) {
7166 sctx->send_root = send_root;
7168 * Unlikely but possible, if the subvolume is marked for deletion but
7169 * is slow to remove the directory entry, send can still be started
7171 if (btrfs_root_dead(sctx->send_root)) {
7176 sctx->clone_roots_cnt = arg->clone_sources_count;
7178 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
7179 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
7180 if (!sctx->send_buf) {
7185 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
7186 if (!sctx->read_buf) {
7191 sctx->pending_dir_moves = RB_ROOT;
7192 sctx->waiting_dir_moves = RB_ROOT;
7193 sctx->orphan_dirs = RB_ROOT;
7195 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
7197 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
7198 if (!sctx->clone_roots) {
7203 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
7205 if (arg->clone_sources_count) {
7206 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
7207 if (!clone_sources_tmp) {
7212 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
7219 for (i = 0; i < arg->clone_sources_count; i++) {
7220 key.objectid = clone_sources_tmp[i];
7221 key.type = BTRFS_ROOT_ITEM_KEY;
7222 key.offset = (u64)-1;
7224 clone_root = btrfs_get_fs_root(fs_info, &key, true);
7225 if (IS_ERR(clone_root)) {
7226 ret = PTR_ERR(clone_root);
7229 spin_lock(&clone_root->root_item_lock);
7230 if (!btrfs_root_readonly(clone_root) ||
7231 btrfs_root_dead(clone_root)) {
7232 spin_unlock(&clone_root->root_item_lock);
7233 btrfs_put_root(clone_root);
7237 if (clone_root->dedupe_in_progress) {
7238 dedupe_in_progress_warn(clone_root);
7239 spin_unlock(&clone_root->root_item_lock);
7240 btrfs_put_root(clone_root);
7244 clone_root->send_in_progress++;
7245 spin_unlock(&clone_root->root_item_lock);
7247 sctx->clone_roots[i].root = clone_root;
7248 clone_sources_to_rollback = i + 1;
7250 kvfree(clone_sources_tmp);
7251 clone_sources_tmp = NULL;
7254 if (arg->parent_root) {
7255 key.objectid = arg->parent_root;
7256 key.type = BTRFS_ROOT_ITEM_KEY;
7257 key.offset = (u64)-1;
7259 sctx->parent_root = btrfs_get_fs_root(fs_info, &key, true);
7260 if (IS_ERR(sctx->parent_root)) {
7261 ret = PTR_ERR(sctx->parent_root);
7265 spin_lock(&sctx->parent_root->root_item_lock);
7266 sctx->parent_root->send_in_progress++;
7267 if (!btrfs_root_readonly(sctx->parent_root) ||
7268 btrfs_root_dead(sctx->parent_root)) {
7269 spin_unlock(&sctx->parent_root->root_item_lock);
7273 if (sctx->parent_root->dedupe_in_progress) {
7274 dedupe_in_progress_warn(sctx->parent_root);
7275 spin_unlock(&sctx->parent_root->root_item_lock);
7279 spin_unlock(&sctx->parent_root->root_item_lock);
7283 * Clones from send_root are allowed, but only if the clone source
7284 * is behind the current send position. This is checked while searching
7285 * for possible clone sources.
7287 sctx->clone_roots[sctx->clone_roots_cnt++].root =
7288 btrfs_grab_root(sctx->send_root);
7290 /* We do a bsearch later */
7291 sort(sctx->clone_roots, sctx->clone_roots_cnt,
7292 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
7294 sort_clone_roots = 1;
7296 ret = flush_delalloc_roots(sctx);
7300 ret = ensure_commit_roots_uptodate(sctx);
7304 mutex_lock(&fs_info->balance_mutex);
7305 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
7306 mutex_unlock(&fs_info->balance_mutex);
7307 btrfs_warn_rl(fs_info,
7308 "cannot run send because a balance operation is in progress");
7312 fs_info->send_in_progress++;
7313 mutex_unlock(&fs_info->balance_mutex);
7315 current->journal_info = BTRFS_SEND_TRANS_STUB;
7316 ret = send_subvol(sctx);
7317 current->journal_info = NULL;
7318 mutex_lock(&fs_info->balance_mutex);
7319 fs_info->send_in_progress--;
7320 mutex_unlock(&fs_info->balance_mutex);
7324 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
7325 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
7328 ret = send_cmd(sctx);
7334 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
7335 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
7337 struct pending_dir_move *pm;
7339 n = rb_first(&sctx->pending_dir_moves);
7340 pm = rb_entry(n, struct pending_dir_move, node);
7341 while (!list_empty(&pm->list)) {
7342 struct pending_dir_move *pm2;
7344 pm2 = list_first_entry(&pm->list,
7345 struct pending_dir_move, list);
7346 free_pending_move(sctx, pm2);
7348 free_pending_move(sctx, pm);
7351 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
7352 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
7354 struct waiting_dir_move *dm;
7356 n = rb_first(&sctx->waiting_dir_moves);
7357 dm = rb_entry(n, struct waiting_dir_move, node);
7358 rb_erase(&dm->node, &sctx->waiting_dir_moves);
7362 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
7363 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
7365 struct orphan_dir_info *odi;
7367 n = rb_first(&sctx->orphan_dirs);
7368 odi = rb_entry(n, struct orphan_dir_info, node);
7369 free_orphan_dir_info(sctx, odi);
7372 if (sort_clone_roots) {
7373 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7374 btrfs_root_dec_send_in_progress(
7375 sctx->clone_roots[i].root);
7376 btrfs_put_root(sctx->clone_roots[i].root);
7379 for (i = 0; sctx && i < clone_sources_to_rollback; i++) {
7380 btrfs_root_dec_send_in_progress(
7381 sctx->clone_roots[i].root);
7382 btrfs_put_root(sctx->clone_roots[i].root);
7385 btrfs_root_dec_send_in_progress(send_root);
7387 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) {
7388 btrfs_root_dec_send_in_progress(sctx->parent_root);
7389 btrfs_put_root(sctx->parent_root);
7392 kvfree(clone_sources_tmp);
7395 if (sctx->send_filp)
7396 fput(sctx->send_filp);
7398 kvfree(sctx->clone_roots);
7399 kvfree(sctx->send_buf);
7400 kvfree(sctx->read_buf);
7402 name_cache_free(sctx);