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,
283 static void inconsistent_snapshot_error(struct send_ctx *sctx,
284 enum btrfs_compare_tree_result result,
287 const char *result_string;
290 case BTRFS_COMPARE_TREE_NEW:
291 result_string = "new";
293 case BTRFS_COMPARE_TREE_DELETED:
294 result_string = "deleted";
296 case BTRFS_COMPARE_TREE_CHANGED:
297 result_string = "updated";
299 case BTRFS_COMPARE_TREE_SAME:
301 result_string = "unchanged";
305 result_string = "unexpected";
308 btrfs_err(sctx->send_root->fs_info,
309 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
310 result_string, what, sctx->cmp_key->objectid,
311 sctx->send_root->root_key.objectid,
313 sctx->parent_root->root_key.objectid : 0));
316 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
318 static struct waiting_dir_move *
319 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
321 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
323 static int need_send_hole(struct send_ctx *sctx)
325 return (sctx->parent_root && !sctx->cur_inode_new &&
326 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
327 S_ISREG(sctx->cur_inode_mode));
330 static void fs_path_reset(struct fs_path *p)
333 p->start = p->buf + p->buf_len - 1;
343 static struct fs_path *fs_path_alloc(void)
347 p = kmalloc(sizeof(*p), GFP_KERNEL);
351 p->buf = p->inline_buf;
352 p->buf_len = FS_PATH_INLINE_SIZE;
357 static struct fs_path *fs_path_alloc_reversed(void)
369 static void fs_path_free(struct fs_path *p)
373 if (p->buf != p->inline_buf)
378 static int fs_path_len(struct fs_path *p)
380 return p->end - p->start;
383 static int fs_path_ensure_buf(struct fs_path *p, int len)
391 if (p->buf_len >= len)
394 if (len > PATH_MAX) {
399 path_len = p->end - p->start;
400 old_buf_len = p->buf_len;
403 * First time the inline_buf does not suffice
405 if (p->buf == p->inline_buf) {
406 tmp_buf = kmalloc(len, GFP_KERNEL);
408 memcpy(tmp_buf, p->buf, old_buf_len);
410 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
416 * The real size of the buffer is bigger, this will let the fast path
417 * happen most of the time
419 p->buf_len = ksize(p->buf);
422 tmp_buf = p->buf + old_buf_len - path_len - 1;
423 p->end = p->buf + p->buf_len - 1;
424 p->start = p->end - path_len;
425 memmove(p->start, tmp_buf, path_len + 1);
428 p->end = p->start + path_len;
433 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
439 new_len = p->end - p->start + name_len;
440 if (p->start != p->end)
442 ret = fs_path_ensure_buf(p, new_len);
447 if (p->start != p->end)
449 p->start -= name_len;
450 *prepared = p->start;
452 if (p->start != p->end)
463 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
468 ret = fs_path_prepare_for_add(p, name_len, &prepared);
471 memcpy(prepared, name, name_len);
477 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
482 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
485 memcpy(prepared, p2->start, p2->end - p2->start);
491 static int fs_path_add_from_extent_buffer(struct fs_path *p,
492 struct extent_buffer *eb,
493 unsigned long off, int len)
498 ret = fs_path_prepare_for_add(p, len, &prepared);
502 read_extent_buffer(eb, prepared, off, len);
508 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
512 p->reversed = from->reversed;
515 ret = fs_path_add_path(p, from);
521 static void fs_path_unreverse(struct fs_path *p)
530 len = p->end - p->start;
532 p->end = p->start + len;
533 memmove(p->start, tmp, len + 1);
537 static struct btrfs_path *alloc_path_for_send(void)
539 struct btrfs_path *path;
541 path = btrfs_alloc_path();
544 path->search_commit_root = 1;
545 path->skip_locking = 1;
546 path->need_commit_sem = 1;
550 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
556 ret = kernel_write(filp, buf + pos, len - pos, off);
557 /* TODO handle that correctly */
558 /*if (ret == -ERESTARTSYS) {
572 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
574 struct btrfs_tlv_header *hdr;
575 int total_len = sizeof(*hdr) + len;
576 int left = sctx->send_max_size - sctx->send_size;
578 if (unlikely(left < total_len))
581 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
582 hdr->tlv_type = cpu_to_le16(attr);
583 hdr->tlv_len = cpu_to_le16(len);
584 memcpy(hdr + 1, data, len);
585 sctx->send_size += total_len;
590 #define TLV_PUT_DEFINE_INT(bits) \
591 static int tlv_put_u##bits(struct send_ctx *sctx, \
592 u##bits attr, u##bits value) \
594 __le##bits __tmp = cpu_to_le##bits(value); \
595 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
598 TLV_PUT_DEFINE_INT(64)
600 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
601 const char *str, int len)
605 return tlv_put(sctx, attr, str, len);
608 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
611 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
614 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
615 struct extent_buffer *eb,
616 struct btrfs_timespec *ts)
618 struct btrfs_timespec bts;
619 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
620 return tlv_put(sctx, attr, &bts, sizeof(bts));
624 #define TLV_PUT(sctx, attrtype, data, attrlen) \
626 ret = tlv_put(sctx, attrtype, data, attrlen); \
628 goto tlv_put_failure; \
631 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
633 ret = tlv_put_u##bits(sctx, attrtype, value); \
635 goto tlv_put_failure; \
638 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
639 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
640 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
641 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
642 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
644 ret = tlv_put_string(sctx, attrtype, str, len); \
646 goto tlv_put_failure; \
648 #define TLV_PUT_PATH(sctx, attrtype, p) \
650 ret = tlv_put_string(sctx, attrtype, p->start, \
651 p->end - p->start); \
653 goto tlv_put_failure; \
655 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
657 ret = tlv_put_uuid(sctx, attrtype, uuid); \
659 goto tlv_put_failure; \
661 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
663 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
665 goto tlv_put_failure; \
668 static int send_header(struct send_ctx *sctx)
670 struct btrfs_stream_header hdr;
672 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
673 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
675 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
680 * For each command/item we want to send to userspace, we call this function.
682 static int begin_cmd(struct send_ctx *sctx, int cmd)
684 struct btrfs_cmd_header *hdr;
686 if (WARN_ON(!sctx->send_buf))
689 BUG_ON(sctx->send_size);
691 sctx->send_size += sizeof(*hdr);
692 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
693 hdr->cmd = cpu_to_le16(cmd);
698 static int send_cmd(struct send_ctx *sctx)
701 struct btrfs_cmd_header *hdr;
704 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
705 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
708 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
709 hdr->crc = cpu_to_le32(crc);
711 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
714 sctx->total_send_size += sctx->send_size;
715 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
722 * Sends a move instruction to user space
724 static int send_rename(struct send_ctx *sctx,
725 struct fs_path *from, struct fs_path *to)
727 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
730 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
732 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
736 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
737 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
739 ret = send_cmd(sctx);
747 * Sends a link instruction to user space
749 static int send_link(struct send_ctx *sctx,
750 struct fs_path *path, struct fs_path *lnk)
752 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
755 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
757 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
761 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
762 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
764 ret = send_cmd(sctx);
772 * Sends an unlink instruction to user space
774 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
776 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
779 btrfs_debug(fs_info, "send_unlink %s", path->start);
781 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
785 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
787 ret = send_cmd(sctx);
795 * Sends a rmdir instruction to user space
797 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
799 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
802 btrfs_debug(fs_info, "send_rmdir %s", path->start);
804 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
808 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
810 ret = send_cmd(sctx);
818 * Helper function to retrieve some fields from an inode item.
820 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
821 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
825 struct btrfs_inode_item *ii;
826 struct btrfs_key key;
829 key.type = BTRFS_INODE_ITEM_KEY;
831 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
838 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
839 struct btrfs_inode_item);
841 *size = btrfs_inode_size(path->nodes[0], ii);
843 *gen = btrfs_inode_generation(path->nodes[0], ii);
845 *mode = btrfs_inode_mode(path->nodes[0], ii);
847 *uid = btrfs_inode_uid(path->nodes[0], ii);
849 *gid = btrfs_inode_gid(path->nodes[0], ii);
851 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
856 static int get_inode_info(struct btrfs_root *root,
857 u64 ino, u64 *size, u64 *gen,
858 u64 *mode, u64 *uid, u64 *gid,
861 struct btrfs_path *path;
864 path = alloc_path_for_send();
867 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
869 btrfs_free_path(path);
873 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
878 * Helper function to iterate the entries in ONE btrfs_inode_ref or
879 * btrfs_inode_extref.
880 * The iterate callback may return a non zero value to stop iteration. This can
881 * be a negative value for error codes or 1 to simply stop it.
883 * path must point to the INODE_REF or INODE_EXTREF when called.
885 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
886 struct btrfs_key *found_key, int resolve,
887 iterate_inode_ref_t iterate, void *ctx)
889 struct extent_buffer *eb = path->nodes[0];
890 struct btrfs_item *item;
891 struct btrfs_inode_ref *iref;
892 struct btrfs_inode_extref *extref;
893 struct btrfs_path *tmp_path;
897 int slot = path->slots[0];
904 unsigned long name_off;
905 unsigned long elem_size;
908 p = fs_path_alloc_reversed();
912 tmp_path = alloc_path_for_send();
919 if (found_key->type == BTRFS_INODE_REF_KEY) {
920 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
921 struct btrfs_inode_ref);
922 item = btrfs_item_nr(slot);
923 total = btrfs_item_size(eb, item);
924 elem_size = sizeof(*iref);
926 ptr = btrfs_item_ptr_offset(eb, slot);
927 total = btrfs_item_size_nr(eb, slot);
928 elem_size = sizeof(*extref);
931 while (cur < total) {
934 if (found_key->type == BTRFS_INODE_REF_KEY) {
935 iref = (struct btrfs_inode_ref *)(ptr + cur);
936 name_len = btrfs_inode_ref_name_len(eb, iref);
937 name_off = (unsigned long)(iref + 1);
938 index = btrfs_inode_ref_index(eb, iref);
939 dir = found_key->offset;
941 extref = (struct btrfs_inode_extref *)(ptr + cur);
942 name_len = btrfs_inode_extref_name_len(eb, extref);
943 name_off = (unsigned long)&extref->name;
944 index = btrfs_inode_extref_index(eb, extref);
945 dir = btrfs_inode_extref_parent(eb, extref);
949 start = btrfs_ref_to_path(root, tmp_path, name_len,
953 ret = PTR_ERR(start);
956 if (start < p->buf) {
957 /* overflow , try again with larger buffer */
958 ret = fs_path_ensure_buf(p,
959 p->buf_len + p->buf - start);
962 start = btrfs_ref_to_path(root, tmp_path,
967 ret = PTR_ERR(start);
970 BUG_ON(start < p->buf);
974 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
980 cur += elem_size + name_len;
981 ret = iterate(num, dir, index, p, ctx);
988 btrfs_free_path(tmp_path);
993 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
994 const char *name, int name_len,
995 const char *data, int data_len,
999 * Helper function to iterate the entries in ONE btrfs_dir_item.
1000 * The iterate callback may return a non zero value to stop iteration. This can
1001 * be a negative value for error codes or 1 to simply stop it.
1003 * path must point to the dir item when called.
1005 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
1006 iterate_dir_item_t iterate, void *ctx)
1009 struct extent_buffer *eb;
1010 struct btrfs_item *item;
1011 struct btrfs_dir_item *di;
1012 struct btrfs_key di_key;
1025 * Start with a small buffer (1 page). If later we end up needing more
1026 * space, which can happen for xattrs on a fs with a leaf size greater
1027 * then the page size, attempt to increase the buffer. Typically xattr
1031 buf = kmalloc(buf_len, GFP_KERNEL);
1037 eb = path->nodes[0];
1038 slot = path->slots[0];
1039 item = btrfs_item_nr(slot);
1040 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1043 total = btrfs_item_size(eb, item);
1046 while (cur < total) {
1047 name_len = btrfs_dir_name_len(eb, di);
1048 data_len = btrfs_dir_data_len(eb, di);
1049 type = btrfs_dir_type(eb, di);
1050 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1052 if (type == BTRFS_FT_XATTR) {
1053 if (name_len > XATTR_NAME_MAX) {
1054 ret = -ENAMETOOLONG;
1057 if (name_len + data_len >
1058 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1066 if (name_len + data_len > PATH_MAX) {
1067 ret = -ENAMETOOLONG;
1072 if (name_len + data_len > buf_len) {
1073 buf_len = name_len + data_len;
1074 if (is_vmalloc_addr(buf)) {
1078 char *tmp = krealloc(buf, buf_len,
1079 GFP_KERNEL | __GFP_NOWARN);
1086 buf = kvmalloc(buf_len, GFP_KERNEL);
1094 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1095 name_len + data_len);
1097 len = sizeof(*di) + name_len + data_len;
1098 di = (struct btrfs_dir_item *)((char *)di + len);
1101 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1102 data_len, type, ctx);
1118 static int __copy_first_ref(int num, u64 dir, int index,
1119 struct fs_path *p, void *ctx)
1122 struct fs_path *pt = ctx;
1124 ret = fs_path_copy(pt, p);
1128 /* we want the first only */
1133 * Retrieve the first path of an inode. If an inode has more then one
1134 * ref/hardlink, this is ignored.
1136 static int get_inode_path(struct btrfs_root *root,
1137 u64 ino, struct fs_path *path)
1140 struct btrfs_key key, found_key;
1141 struct btrfs_path *p;
1143 p = alloc_path_for_send();
1147 fs_path_reset(path);
1150 key.type = BTRFS_INODE_REF_KEY;
1153 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1160 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1161 if (found_key.objectid != ino ||
1162 (found_key.type != BTRFS_INODE_REF_KEY &&
1163 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1168 ret = iterate_inode_ref(root, p, &found_key, 1,
1169 __copy_first_ref, path);
1179 struct backref_ctx {
1180 struct send_ctx *sctx;
1182 /* number of total found references */
1186 * used for clones found in send_root. clones found behind cur_objectid
1187 * and cur_offset are not considered as allowed clones.
1192 /* may be truncated in case it's the last extent in a file */
1195 /* data offset in the file extent item */
1198 /* Just to check for bugs in backref resolving */
1202 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1204 u64 root = (u64)(uintptr_t)key;
1205 struct clone_root *cr = (struct clone_root *)elt;
1207 if (root < cr->root->root_key.objectid)
1209 if (root > cr->root->root_key.objectid)
1214 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1216 struct clone_root *cr1 = (struct clone_root *)e1;
1217 struct clone_root *cr2 = (struct clone_root *)e2;
1219 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1221 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1227 * Called for every backref that is found for the current extent.
1228 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1230 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1232 struct backref_ctx *bctx = ctx_;
1233 struct clone_root *found;
1235 /* First check if the root is in the list of accepted clone sources */
1236 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1237 bctx->sctx->clone_roots_cnt,
1238 sizeof(struct clone_root),
1239 __clone_root_cmp_bsearch);
1243 if (found->root == bctx->sctx->send_root &&
1244 ino == bctx->cur_objectid &&
1245 offset == bctx->cur_offset) {
1246 bctx->found_itself = 1;
1250 * Make sure we don't consider clones from send_root that are
1251 * behind the current inode/offset.
1253 if (found->root == bctx->sctx->send_root) {
1255 * If the source inode was not yet processed we can't issue a
1256 * clone operation, as the source extent does not exist yet at
1257 * the destination of the stream.
1259 if (ino > bctx->cur_objectid)
1262 * We clone from the inode currently being sent as long as the
1263 * source extent is already processed, otherwise we could try
1264 * to clone from an extent that does not exist yet at the
1265 * destination of the stream.
1267 if (ino == bctx->cur_objectid &&
1268 offset + bctx->extent_len >
1269 bctx->sctx->cur_inode_next_write_offset)
1274 found->found_refs++;
1275 if (ino < found->ino) {
1277 found->offset = offset;
1278 } else if (found->ino == ino) {
1280 * same extent found more then once in the same file.
1282 if (found->offset > offset + bctx->extent_len)
1283 found->offset = offset;
1290 * Given an inode, offset and extent item, it finds a good clone for a clone
1291 * instruction. Returns -ENOENT when none could be found. The function makes
1292 * sure that the returned clone is usable at the point where sending is at the
1293 * moment. This means, that no clones are accepted which lie behind the current
1296 * path must point to the extent item when called.
1298 static int find_extent_clone(struct send_ctx *sctx,
1299 struct btrfs_path *path,
1300 u64 ino, u64 data_offset,
1302 struct clone_root **found)
1304 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1310 u64 extent_item_pos;
1312 struct btrfs_file_extent_item *fi;
1313 struct extent_buffer *eb = path->nodes[0];
1314 struct backref_ctx *backref_ctx = NULL;
1315 struct clone_root *cur_clone_root;
1316 struct btrfs_key found_key;
1317 struct btrfs_path *tmp_path;
1318 struct btrfs_extent_item *ei;
1322 tmp_path = alloc_path_for_send();
1326 /* We only use this path under the commit sem */
1327 tmp_path->need_commit_sem = 0;
1329 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1335 if (data_offset >= ino_size) {
1337 * There may be extents that lie behind the file's size.
1338 * I at least had this in combination with snapshotting while
1339 * writing large files.
1345 fi = btrfs_item_ptr(eb, path->slots[0],
1346 struct btrfs_file_extent_item);
1347 extent_type = btrfs_file_extent_type(eb, fi);
1348 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1352 compressed = btrfs_file_extent_compression(eb, fi);
1354 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1355 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1356 if (disk_byte == 0) {
1360 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1362 down_read(&fs_info->commit_root_sem);
1363 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1364 &found_key, &flags);
1365 up_read(&fs_info->commit_root_sem);
1369 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1374 ei = btrfs_item_ptr(tmp_path->nodes[0], tmp_path->slots[0],
1375 struct btrfs_extent_item);
1377 * Backreference walking (iterate_extent_inodes() below) is currently
1378 * too expensive when an extent has a large number of references, both
1379 * in time spent and used memory. So for now just fallback to write
1380 * operations instead of clone operations when an extent has more than
1381 * a certain amount of references.
1383 if (btrfs_extent_refs(tmp_path->nodes[0], ei) > SEND_MAX_EXTENT_REFS) {
1387 btrfs_release_path(tmp_path);
1390 * Setup the clone roots.
1392 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1393 cur_clone_root = sctx->clone_roots + i;
1394 cur_clone_root->ino = (u64)-1;
1395 cur_clone_root->offset = 0;
1396 cur_clone_root->found_refs = 0;
1399 backref_ctx->sctx = sctx;
1400 backref_ctx->found = 0;
1401 backref_ctx->cur_objectid = ino;
1402 backref_ctx->cur_offset = data_offset;
1403 backref_ctx->found_itself = 0;
1404 backref_ctx->extent_len = num_bytes;
1406 * For non-compressed extents iterate_extent_inodes() gives us extent
1407 * offsets that already take into account the data offset, but not for
1408 * compressed extents, since the offset is logical and not relative to
1409 * the physical extent locations. We must take this into account to
1410 * avoid sending clone offsets that go beyond the source file's size,
1411 * which would result in the clone ioctl failing with -EINVAL on the
1414 if (compressed == BTRFS_COMPRESS_NONE)
1415 backref_ctx->data_offset = 0;
1417 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1420 * The last extent of a file may be too large due to page alignment.
1421 * We need to adjust extent_len in this case so that the checks in
1422 * __iterate_backrefs work.
1424 if (data_offset + num_bytes >= ino_size)
1425 backref_ctx->extent_len = ino_size - data_offset;
1428 * Now collect all backrefs.
1430 if (compressed == BTRFS_COMPRESS_NONE)
1431 extent_item_pos = logical - found_key.objectid;
1433 extent_item_pos = 0;
1434 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1435 extent_item_pos, 1, __iterate_backrefs,
1436 backref_ctx, false);
1441 if (!backref_ctx->found_itself) {
1442 /* found a bug in backref code? */
1445 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1446 ino, data_offset, disk_byte, found_key.objectid);
1450 btrfs_debug(fs_info,
1451 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1452 data_offset, ino, num_bytes, logical);
1454 if (!backref_ctx->found)
1455 btrfs_debug(fs_info, "no clones found");
1457 cur_clone_root = NULL;
1458 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1459 if (sctx->clone_roots[i].found_refs) {
1460 if (!cur_clone_root)
1461 cur_clone_root = sctx->clone_roots + i;
1462 else if (sctx->clone_roots[i].root == sctx->send_root)
1463 /* prefer clones from send_root over others */
1464 cur_clone_root = sctx->clone_roots + i;
1469 if (cur_clone_root) {
1470 *found = cur_clone_root;
1477 btrfs_free_path(tmp_path);
1482 static int read_symlink(struct btrfs_root *root,
1484 struct fs_path *dest)
1487 struct btrfs_path *path;
1488 struct btrfs_key key;
1489 struct btrfs_file_extent_item *ei;
1495 path = alloc_path_for_send();
1500 key.type = BTRFS_EXTENT_DATA_KEY;
1502 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1507 * An empty symlink inode. Can happen in rare error paths when
1508 * creating a symlink (transaction committed before the inode
1509 * eviction handler removed the symlink inode items and a crash
1510 * happened in between or the subvol was snapshoted in between).
1511 * Print an informative message to dmesg/syslog so that the user
1512 * can delete the symlink.
1514 btrfs_err(root->fs_info,
1515 "Found empty symlink inode %llu at root %llu",
1516 ino, root->root_key.objectid);
1521 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1522 struct btrfs_file_extent_item);
1523 type = btrfs_file_extent_type(path->nodes[0], ei);
1524 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1525 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1526 BUG_ON(compression);
1528 off = btrfs_file_extent_inline_start(ei);
1529 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1531 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1534 btrfs_free_path(path);
1539 * Helper function to generate a file name that is unique in the root of
1540 * send_root and parent_root. This is used to generate names for orphan inodes.
1542 static int gen_unique_name(struct send_ctx *sctx,
1544 struct fs_path *dest)
1547 struct btrfs_path *path;
1548 struct btrfs_dir_item *di;
1553 path = alloc_path_for_send();
1558 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1560 ASSERT(len < sizeof(tmp));
1562 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1563 path, BTRFS_FIRST_FREE_OBJECTID,
1564 tmp, strlen(tmp), 0);
1565 btrfs_release_path(path);
1571 /* not unique, try again */
1576 if (!sctx->parent_root) {
1582 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1583 path, BTRFS_FIRST_FREE_OBJECTID,
1584 tmp, strlen(tmp), 0);
1585 btrfs_release_path(path);
1591 /* not unique, try again */
1599 ret = fs_path_add(dest, tmp, strlen(tmp));
1602 btrfs_free_path(path);
1607 inode_state_no_change,
1608 inode_state_will_create,
1609 inode_state_did_create,
1610 inode_state_will_delete,
1611 inode_state_did_delete,
1614 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1622 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1624 if (ret < 0 && ret != -ENOENT)
1628 if (!sctx->parent_root) {
1629 right_ret = -ENOENT;
1631 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1632 NULL, NULL, NULL, NULL);
1633 if (ret < 0 && ret != -ENOENT)
1638 if (!left_ret && !right_ret) {
1639 if (left_gen == gen && right_gen == gen) {
1640 ret = inode_state_no_change;
1641 } else if (left_gen == gen) {
1642 if (ino < sctx->send_progress)
1643 ret = inode_state_did_create;
1645 ret = inode_state_will_create;
1646 } else if (right_gen == gen) {
1647 if (ino < sctx->send_progress)
1648 ret = inode_state_did_delete;
1650 ret = inode_state_will_delete;
1654 } else if (!left_ret) {
1655 if (left_gen == gen) {
1656 if (ino < sctx->send_progress)
1657 ret = inode_state_did_create;
1659 ret = inode_state_will_create;
1663 } else if (!right_ret) {
1664 if (right_gen == gen) {
1665 if (ino < sctx->send_progress)
1666 ret = inode_state_did_delete;
1668 ret = inode_state_will_delete;
1680 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1684 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1687 ret = get_cur_inode_state(sctx, ino, gen);
1691 if (ret == inode_state_no_change ||
1692 ret == inode_state_did_create ||
1693 ret == inode_state_will_delete)
1703 * Helper function to lookup a dir item in a dir.
1705 static int lookup_dir_item_inode(struct btrfs_root *root,
1706 u64 dir, const char *name, int name_len,
1711 struct btrfs_dir_item *di;
1712 struct btrfs_key key;
1713 struct btrfs_path *path;
1715 path = alloc_path_for_send();
1719 di = btrfs_lookup_dir_item(NULL, root, path,
1720 dir, name, name_len, 0);
1721 if (IS_ERR_OR_NULL(di)) {
1722 ret = di ? PTR_ERR(di) : -ENOENT;
1725 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1726 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1730 *found_inode = key.objectid;
1731 *found_type = btrfs_dir_type(path->nodes[0], di);
1734 btrfs_free_path(path);
1739 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1740 * generation of the parent dir and the name of the dir entry.
1742 static int get_first_ref(struct btrfs_root *root, u64 ino,
1743 u64 *dir, u64 *dir_gen, struct fs_path *name)
1746 struct btrfs_key key;
1747 struct btrfs_key found_key;
1748 struct btrfs_path *path;
1752 path = alloc_path_for_send();
1757 key.type = BTRFS_INODE_REF_KEY;
1760 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1764 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1766 if (ret || found_key.objectid != ino ||
1767 (found_key.type != BTRFS_INODE_REF_KEY &&
1768 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1773 if (found_key.type == BTRFS_INODE_REF_KEY) {
1774 struct btrfs_inode_ref *iref;
1775 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1776 struct btrfs_inode_ref);
1777 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1778 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1779 (unsigned long)(iref + 1),
1781 parent_dir = found_key.offset;
1783 struct btrfs_inode_extref *extref;
1784 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1785 struct btrfs_inode_extref);
1786 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1787 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1788 (unsigned long)&extref->name, len);
1789 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1793 btrfs_release_path(path);
1796 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1805 btrfs_free_path(path);
1809 static int is_first_ref(struct btrfs_root *root,
1811 const char *name, int name_len)
1814 struct fs_path *tmp_name;
1817 tmp_name = fs_path_alloc();
1821 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1825 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1830 ret = !memcmp(tmp_name->start, name, name_len);
1833 fs_path_free(tmp_name);
1838 * Used by process_recorded_refs to determine if a new ref would overwrite an
1839 * already existing ref. In case it detects an overwrite, it returns the
1840 * inode/gen in who_ino/who_gen.
1841 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1842 * to make sure later references to the overwritten inode are possible.
1843 * Orphanizing is however only required for the first ref of an inode.
1844 * process_recorded_refs does an additional is_first_ref check to see if
1845 * orphanizing is really required.
1847 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1848 const char *name, int name_len,
1849 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1853 u64 other_inode = 0;
1856 if (!sctx->parent_root)
1859 ret = is_inode_existent(sctx, dir, dir_gen);
1864 * If we have a parent root we need to verify that the parent dir was
1865 * not deleted and then re-created, if it was then we have no overwrite
1866 * and we can just unlink this entry.
1868 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1869 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1871 if (ret < 0 && ret != -ENOENT)
1881 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1882 &other_inode, &other_type);
1883 if (ret < 0 && ret != -ENOENT)
1891 * Check if the overwritten ref was already processed. If yes, the ref
1892 * was already unlinked/moved, so we can safely assume that we will not
1893 * overwrite anything at this point in time.
1895 if (other_inode > sctx->send_progress ||
1896 is_waiting_for_move(sctx, other_inode)) {
1897 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1898 who_gen, who_mode, NULL, NULL, NULL);
1903 *who_ino = other_inode;
1913 * Checks if the ref was overwritten by an already processed inode. This is
1914 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1915 * thus the orphan name needs be used.
1916 * process_recorded_refs also uses it to avoid unlinking of refs that were
1919 static int did_overwrite_ref(struct send_ctx *sctx,
1920 u64 dir, u64 dir_gen,
1921 u64 ino, u64 ino_gen,
1922 const char *name, int name_len)
1929 if (!sctx->parent_root)
1932 ret = is_inode_existent(sctx, dir, dir_gen);
1936 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1937 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1939 if (ret < 0 && ret != -ENOENT)
1949 /* check if the ref was overwritten by another ref */
1950 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1951 &ow_inode, &other_type);
1952 if (ret < 0 && ret != -ENOENT)
1955 /* was never and will never be overwritten */
1960 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1965 if (ow_inode == ino && gen == ino_gen) {
1971 * We know that it is or will be overwritten. Check this now.
1972 * The current inode being processed might have been the one that caused
1973 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1974 * the current inode being processed.
1976 if ((ow_inode < sctx->send_progress) ||
1977 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1978 gen == sctx->cur_inode_gen))
1988 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1989 * that got overwritten. This is used by process_recorded_refs to determine
1990 * if it has to use the path as returned by get_cur_path or the orphan name.
1992 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1995 struct fs_path *name = NULL;
1999 if (!sctx->parent_root)
2002 name = fs_path_alloc();
2006 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
2010 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
2011 name->start, fs_path_len(name));
2019 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2020 * so we need to do some special handling in case we have clashes. This function
2021 * takes care of this with the help of name_cache_entry::radix_list.
2022 * In case of error, nce is kfreed.
2024 static int name_cache_insert(struct send_ctx *sctx,
2025 struct name_cache_entry *nce)
2028 struct list_head *nce_head;
2030 nce_head = radix_tree_lookup(&sctx->name_cache,
2031 (unsigned long)nce->ino);
2033 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2038 INIT_LIST_HEAD(nce_head);
2040 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2047 list_add_tail(&nce->radix_list, nce_head);
2048 list_add_tail(&nce->list, &sctx->name_cache_list);
2049 sctx->name_cache_size++;
2054 static void name_cache_delete(struct send_ctx *sctx,
2055 struct name_cache_entry *nce)
2057 struct list_head *nce_head;
2059 nce_head = radix_tree_lookup(&sctx->name_cache,
2060 (unsigned long)nce->ino);
2062 btrfs_err(sctx->send_root->fs_info,
2063 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2064 nce->ino, sctx->name_cache_size);
2067 list_del(&nce->radix_list);
2068 list_del(&nce->list);
2069 sctx->name_cache_size--;
2072 * We may not get to the final release of nce_head if the lookup fails
2074 if (nce_head && list_empty(nce_head)) {
2075 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2080 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2083 struct list_head *nce_head;
2084 struct name_cache_entry *cur;
2086 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2090 list_for_each_entry(cur, nce_head, radix_list) {
2091 if (cur->ino == ino && cur->gen == gen)
2098 * Removes the entry from the list and adds it back to the end. This marks the
2099 * entry as recently used so that name_cache_clean_unused does not remove it.
2101 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2103 list_del(&nce->list);
2104 list_add_tail(&nce->list, &sctx->name_cache_list);
2108 * Remove some entries from the beginning of name_cache_list.
2110 static void name_cache_clean_unused(struct send_ctx *sctx)
2112 struct name_cache_entry *nce;
2114 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2117 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2118 nce = list_entry(sctx->name_cache_list.next,
2119 struct name_cache_entry, list);
2120 name_cache_delete(sctx, nce);
2125 static void name_cache_free(struct send_ctx *sctx)
2127 struct name_cache_entry *nce;
2129 while (!list_empty(&sctx->name_cache_list)) {
2130 nce = list_entry(sctx->name_cache_list.next,
2131 struct name_cache_entry, list);
2132 name_cache_delete(sctx, nce);
2138 * Used by get_cur_path for each ref up to the root.
2139 * Returns 0 if it succeeded.
2140 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2141 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2142 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2143 * Returns <0 in case of error.
2145 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2149 struct fs_path *dest)
2153 struct name_cache_entry *nce = NULL;
2156 * First check if we already did a call to this function with the same
2157 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2158 * return the cached result.
2160 nce = name_cache_search(sctx, ino, gen);
2162 if (ino < sctx->send_progress && nce->need_later_update) {
2163 name_cache_delete(sctx, nce);
2167 name_cache_used(sctx, nce);
2168 *parent_ino = nce->parent_ino;
2169 *parent_gen = nce->parent_gen;
2170 ret = fs_path_add(dest, nce->name, nce->name_len);
2179 * If the inode is not existent yet, add the orphan name and return 1.
2180 * This should only happen for the parent dir that we determine in
2183 ret = is_inode_existent(sctx, ino, gen);
2188 ret = gen_unique_name(sctx, ino, gen, dest);
2196 * Depending on whether the inode was already processed or not, use
2197 * send_root or parent_root for ref lookup.
2199 if (ino < sctx->send_progress)
2200 ret = get_first_ref(sctx->send_root, ino,
2201 parent_ino, parent_gen, dest);
2203 ret = get_first_ref(sctx->parent_root, ino,
2204 parent_ino, parent_gen, dest);
2209 * Check if the ref was overwritten by an inode's ref that was processed
2210 * earlier. If yes, treat as orphan and return 1.
2212 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2213 dest->start, dest->end - dest->start);
2217 fs_path_reset(dest);
2218 ret = gen_unique_name(sctx, ino, gen, dest);
2226 * Store the result of the lookup in the name cache.
2228 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2236 nce->parent_ino = *parent_ino;
2237 nce->parent_gen = *parent_gen;
2238 nce->name_len = fs_path_len(dest);
2240 strcpy(nce->name, dest->start);
2242 if (ino < sctx->send_progress)
2243 nce->need_later_update = 0;
2245 nce->need_later_update = 1;
2247 nce_ret = name_cache_insert(sctx, nce);
2250 name_cache_clean_unused(sctx);
2257 * Magic happens here. This function returns the first ref to an inode as it
2258 * would look like while receiving the stream at this point in time.
2259 * We walk the path up to the root. For every inode in between, we check if it
2260 * was already processed/sent. If yes, we continue with the parent as found
2261 * in send_root. If not, we continue with the parent as found in parent_root.
2262 * If we encounter an inode that was deleted at this point in time, we use the
2263 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2264 * that were not created yet and overwritten inodes/refs.
2266 * When do we have orphan inodes:
2267 * 1. When an inode is freshly created and thus no valid refs are available yet
2268 * 2. When a directory lost all it's refs (deleted) but still has dir items
2269 * inside which were not processed yet (pending for move/delete). If anyone
2270 * tried to get the path to the dir items, it would get a path inside that
2272 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2273 * of an unprocessed inode. If in that case the first ref would be
2274 * overwritten, the overwritten inode gets "orphanized". Later when we
2275 * process this overwritten inode, it is restored at a new place by moving
2278 * sctx->send_progress tells this function at which point in time receiving
2281 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2282 struct fs_path *dest)
2285 struct fs_path *name = NULL;
2286 u64 parent_inode = 0;
2290 name = fs_path_alloc();
2297 fs_path_reset(dest);
2299 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2300 struct waiting_dir_move *wdm;
2302 fs_path_reset(name);
2304 if (is_waiting_for_rm(sctx, ino)) {
2305 ret = gen_unique_name(sctx, ino, gen, name);
2308 ret = fs_path_add_path(dest, name);
2312 wdm = get_waiting_dir_move(sctx, ino);
2313 if (wdm && wdm->orphanized) {
2314 ret = gen_unique_name(sctx, ino, gen, name);
2317 ret = get_first_ref(sctx->parent_root, ino,
2318 &parent_inode, &parent_gen, name);
2320 ret = __get_cur_name_and_parent(sctx, ino, gen,
2330 ret = fs_path_add_path(dest, name);
2341 fs_path_unreverse(dest);
2346 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2348 static int send_subvol_begin(struct send_ctx *sctx)
2351 struct btrfs_root *send_root = sctx->send_root;
2352 struct btrfs_root *parent_root = sctx->parent_root;
2353 struct btrfs_path *path;
2354 struct btrfs_key key;
2355 struct btrfs_root_ref *ref;
2356 struct extent_buffer *leaf;
2360 path = btrfs_alloc_path();
2364 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2366 btrfs_free_path(path);
2370 key.objectid = send_root->root_key.objectid;
2371 key.type = BTRFS_ROOT_BACKREF_KEY;
2374 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2383 leaf = path->nodes[0];
2384 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2385 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2386 key.objectid != send_root->root_key.objectid) {
2390 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2391 namelen = btrfs_root_ref_name_len(leaf, ref);
2392 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2393 btrfs_release_path(path);
2396 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2400 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2405 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2407 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2408 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2409 sctx->send_root->root_item.received_uuid);
2411 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2412 sctx->send_root->root_item.uuid);
2414 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2415 le64_to_cpu(sctx->send_root->root_item.ctransid));
2417 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2418 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2419 parent_root->root_item.received_uuid);
2421 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2422 parent_root->root_item.uuid);
2423 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2424 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2427 ret = send_cmd(sctx);
2431 btrfs_free_path(path);
2436 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2438 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2442 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2444 p = fs_path_alloc();
2448 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2452 ret = get_cur_path(sctx, ino, gen, p);
2455 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2456 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2458 ret = send_cmd(sctx);
2466 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2468 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2472 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2474 p = fs_path_alloc();
2478 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2482 ret = get_cur_path(sctx, ino, gen, p);
2485 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2486 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2488 ret = send_cmd(sctx);
2496 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2498 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2502 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2505 p = fs_path_alloc();
2509 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2513 ret = get_cur_path(sctx, ino, gen, p);
2516 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2517 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2518 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2520 ret = send_cmd(sctx);
2528 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2530 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2532 struct fs_path *p = NULL;
2533 struct btrfs_inode_item *ii;
2534 struct btrfs_path *path = NULL;
2535 struct extent_buffer *eb;
2536 struct btrfs_key key;
2539 btrfs_debug(fs_info, "send_utimes %llu", ino);
2541 p = fs_path_alloc();
2545 path = alloc_path_for_send();
2552 key.type = BTRFS_INODE_ITEM_KEY;
2554 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2560 eb = path->nodes[0];
2561 slot = path->slots[0];
2562 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2564 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2568 ret = get_cur_path(sctx, ino, gen, p);
2571 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2572 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2573 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2574 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2575 /* TODO Add otime support when the otime patches get into upstream */
2577 ret = send_cmd(sctx);
2582 btrfs_free_path(path);
2587 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2588 * a valid path yet because we did not process the refs yet. So, the inode
2589 * is created as orphan.
2591 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2593 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2601 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2603 p = fs_path_alloc();
2607 if (ino != sctx->cur_ino) {
2608 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2613 gen = sctx->cur_inode_gen;
2614 mode = sctx->cur_inode_mode;
2615 rdev = sctx->cur_inode_rdev;
2618 if (S_ISREG(mode)) {
2619 cmd = BTRFS_SEND_C_MKFILE;
2620 } else if (S_ISDIR(mode)) {
2621 cmd = BTRFS_SEND_C_MKDIR;
2622 } else if (S_ISLNK(mode)) {
2623 cmd = BTRFS_SEND_C_SYMLINK;
2624 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2625 cmd = BTRFS_SEND_C_MKNOD;
2626 } else if (S_ISFIFO(mode)) {
2627 cmd = BTRFS_SEND_C_MKFIFO;
2628 } else if (S_ISSOCK(mode)) {
2629 cmd = BTRFS_SEND_C_MKSOCK;
2631 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2632 (int)(mode & S_IFMT));
2637 ret = begin_cmd(sctx, cmd);
2641 ret = gen_unique_name(sctx, ino, gen, p);
2645 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2646 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2648 if (S_ISLNK(mode)) {
2650 ret = read_symlink(sctx->send_root, ino, p);
2653 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2654 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2655 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2656 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2657 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2660 ret = send_cmd(sctx);
2672 * We need some special handling for inodes that get processed before the parent
2673 * directory got created. See process_recorded_refs for details.
2674 * This function does the check if we already created the dir out of order.
2676 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2679 struct btrfs_path *path = NULL;
2680 struct btrfs_key key;
2681 struct btrfs_key found_key;
2682 struct btrfs_key di_key;
2683 struct extent_buffer *eb;
2684 struct btrfs_dir_item *di;
2687 path = alloc_path_for_send();
2694 key.type = BTRFS_DIR_INDEX_KEY;
2696 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2701 eb = path->nodes[0];
2702 slot = path->slots[0];
2703 if (slot >= btrfs_header_nritems(eb)) {
2704 ret = btrfs_next_leaf(sctx->send_root, path);
2707 } else if (ret > 0) {
2714 btrfs_item_key_to_cpu(eb, &found_key, slot);
2715 if (found_key.objectid != key.objectid ||
2716 found_key.type != key.type) {
2721 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2722 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2724 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2725 di_key.objectid < sctx->send_progress) {
2734 btrfs_free_path(path);
2739 * Only creates the inode if it is:
2740 * 1. Not a directory
2741 * 2. Or a directory which was not created already due to out of order
2742 * directories. See did_create_dir and process_recorded_refs for details.
2744 static int send_create_inode_if_needed(struct send_ctx *sctx)
2748 if (S_ISDIR(sctx->cur_inode_mode)) {
2749 ret = did_create_dir(sctx, sctx->cur_ino);
2758 ret = send_create_inode(sctx, sctx->cur_ino);
2766 struct recorded_ref {
2767 struct list_head list;
2769 struct fs_path *full_path;
2775 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2777 ref->full_path = path;
2778 ref->name = (char *)kbasename(ref->full_path->start);
2779 ref->name_len = ref->full_path->end - ref->name;
2783 * We need to process new refs before deleted refs, but compare_tree gives us
2784 * everything mixed. So we first record all refs and later process them.
2785 * This function is a helper to record one ref.
2787 static int __record_ref(struct list_head *head, u64 dir,
2788 u64 dir_gen, struct fs_path *path)
2790 struct recorded_ref *ref;
2792 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2797 ref->dir_gen = dir_gen;
2798 set_ref_path(ref, path);
2799 list_add_tail(&ref->list, head);
2803 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2805 struct recorded_ref *new;
2807 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2811 new->dir = ref->dir;
2812 new->dir_gen = ref->dir_gen;
2813 new->full_path = NULL;
2814 INIT_LIST_HEAD(&new->list);
2815 list_add_tail(&new->list, list);
2819 static void __free_recorded_refs(struct list_head *head)
2821 struct recorded_ref *cur;
2823 while (!list_empty(head)) {
2824 cur = list_entry(head->next, struct recorded_ref, list);
2825 fs_path_free(cur->full_path);
2826 list_del(&cur->list);
2831 static void free_recorded_refs(struct send_ctx *sctx)
2833 __free_recorded_refs(&sctx->new_refs);
2834 __free_recorded_refs(&sctx->deleted_refs);
2838 * Renames/moves a file/dir to its orphan name. Used when the first
2839 * ref of an unprocessed inode gets overwritten and for all non empty
2842 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2843 struct fs_path *path)
2846 struct fs_path *orphan;
2848 orphan = fs_path_alloc();
2852 ret = gen_unique_name(sctx, ino, gen, orphan);
2856 ret = send_rename(sctx, path, orphan);
2859 fs_path_free(orphan);
2863 static struct orphan_dir_info *
2864 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2866 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2867 struct rb_node *parent = NULL;
2868 struct orphan_dir_info *entry, *odi;
2872 entry = rb_entry(parent, struct orphan_dir_info, node);
2873 if (dir_ino < entry->ino) {
2875 } else if (dir_ino > entry->ino) {
2876 p = &(*p)->rb_right;
2882 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2884 return ERR_PTR(-ENOMEM);
2887 odi->last_dir_index_offset = 0;
2889 rb_link_node(&odi->node, parent, p);
2890 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2894 static struct orphan_dir_info *
2895 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2897 struct rb_node *n = sctx->orphan_dirs.rb_node;
2898 struct orphan_dir_info *entry;
2901 entry = rb_entry(n, struct orphan_dir_info, node);
2902 if (dir_ino < entry->ino)
2904 else if (dir_ino > entry->ino)
2912 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2914 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2919 static void free_orphan_dir_info(struct send_ctx *sctx,
2920 struct orphan_dir_info *odi)
2924 rb_erase(&odi->node, &sctx->orphan_dirs);
2929 * Returns 1 if a directory can be removed at this point in time.
2930 * We check this by iterating all dir items and checking if the inode behind
2931 * the dir item was already processed.
2933 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2937 struct btrfs_root *root = sctx->parent_root;
2938 struct btrfs_path *path;
2939 struct btrfs_key key;
2940 struct btrfs_key found_key;
2941 struct btrfs_key loc;
2942 struct btrfs_dir_item *di;
2943 struct orphan_dir_info *odi = NULL;
2946 * Don't try to rmdir the top/root subvolume dir.
2948 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2951 path = alloc_path_for_send();
2956 key.type = BTRFS_DIR_INDEX_KEY;
2959 odi = get_orphan_dir_info(sctx, dir);
2961 key.offset = odi->last_dir_index_offset;
2963 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2968 struct waiting_dir_move *dm;
2970 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2971 ret = btrfs_next_leaf(root, path);
2978 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2980 if (found_key.objectid != key.objectid ||
2981 found_key.type != key.type)
2984 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2985 struct btrfs_dir_item);
2986 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2988 dm = get_waiting_dir_move(sctx, loc.objectid);
2990 odi = add_orphan_dir_info(sctx, dir);
2996 odi->last_dir_index_offset = found_key.offset;
2997 dm->rmdir_ino = dir;
3002 if (loc.objectid > send_progress) {
3003 odi = add_orphan_dir_info(sctx, dir);
3009 odi->last_dir_index_offset = found_key.offset;
3016 free_orphan_dir_info(sctx, odi);
3021 btrfs_free_path(path);
3025 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3027 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3029 return entry != NULL;
3032 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3034 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3035 struct rb_node *parent = NULL;
3036 struct waiting_dir_move *entry, *dm;
3038 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3043 dm->orphanized = orphanized;
3047 entry = rb_entry(parent, struct waiting_dir_move, node);
3048 if (ino < entry->ino) {
3050 } else if (ino > entry->ino) {
3051 p = &(*p)->rb_right;
3058 rb_link_node(&dm->node, parent, p);
3059 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3063 static struct waiting_dir_move *
3064 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3066 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3067 struct waiting_dir_move *entry;
3070 entry = rb_entry(n, struct waiting_dir_move, node);
3071 if (ino < entry->ino)
3073 else if (ino > entry->ino)
3081 static void free_waiting_dir_move(struct send_ctx *sctx,
3082 struct waiting_dir_move *dm)
3086 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3090 static int add_pending_dir_move(struct send_ctx *sctx,
3094 struct list_head *new_refs,
3095 struct list_head *deleted_refs,
3096 const bool is_orphan)
3098 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3099 struct rb_node *parent = NULL;
3100 struct pending_dir_move *entry = NULL, *pm;
3101 struct recorded_ref *cur;
3105 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3108 pm->parent_ino = parent_ino;
3111 INIT_LIST_HEAD(&pm->list);
3112 INIT_LIST_HEAD(&pm->update_refs);
3113 RB_CLEAR_NODE(&pm->node);
3117 entry = rb_entry(parent, struct pending_dir_move, node);
3118 if (parent_ino < entry->parent_ino) {
3120 } else if (parent_ino > entry->parent_ino) {
3121 p = &(*p)->rb_right;
3128 list_for_each_entry(cur, deleted_refs, list) {
3129 ret = dup_ref(cur, &pm->update_refs);
3133 list_for_each_entry(cur, new_refs, list) {
3134 ret = dup_ref(cur, &pm->update_refs);
3139 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3144 list_add_tail(&pm->list, &entry->list);
3146 rb_link_node(&pm->node, parent, p);
3147 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3152 __free_recorded_refs(&pm->update_refs);
3158 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3161 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3162 struct pending_dir_move *entry;
3165 entry = rb_entry(n, struct pending_dir_move, node);
3166 if (parent_ino < entry->parent_ino)
3168 else if (parent_ino > entry->parent_ino)
3176 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3177 u64 ino, u64 gen, u64 *ancestor_ino)
3180 u64 parent_inode = 0;
3182 u64 start_ino = ino;
3185 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3186 fs_path_reset(name);
3188 if (is_waiting_for_rm(sctx, ino))
3190 if (is_waiting_for_move(sctx, ino)) {
3191 if (*ancestor_ino == 0)
3192 *ancestor_ino = ino;
3193 ret = get_first_ref(sctx->parent_root, ino,
3194 &parent_inode, &parent_gen, name);
3196 ret = __get_cur_name_and_parent(sctx, ino, gen,
3206 if (parent_inode == start_ino) {
3208 if (*ancestor_ino == 0)
3209 *ancestor_ino = ino;
3218 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3220 struct fs_path *from_path = NULL;
3221 struct fs_path *to_path = NULL;
3222 struct fs_path *name = NULL;
3223 u64 orig_progress = sctx->send_progress;
3224 struct recorded_ref *cur;
3225 u64 parent_ino, parent_gen;
3226 struct waiting_dir_move *dm = NULL;
3232 name = fs_path_alloc();
3233 from_path = fs_path_alloc();
3234 if (!name || !from_path) {
3239 dm = get_waiting_dir_move(sctx, pm->ino);
3241 rmdir_ino = dm->rmdir_ino;
3242 is_orphan = dm->orphanized;
3243 free_waiting_dir_move(sctx, dm);
3246 ret = gen_unique_name(sctx, pm->ino,
3247 pm->gen, from_path);
3249 ret = get_first_ref(sctx->parent_root, pm->ino,
3250 &parent_ino, &parent_gen, name);
3253 ret = get_cur_path(sctx, parent_ino, parent_gen,
3257 ret = fs_path_add_path(from_path, name);
3262 sctx->send_progress = sctx->cur_ino + 1;
3263 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3267 LIST_HEAD(deleted_refs);
3268 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3269 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3270 &pm->update_refs, &deleted_refs,
3275 dm = get_waiting_dir_move(sctx, pm->ino);
3277 dm->rmdir_ino = rmdir_ino;
3281 fs_path_reset(name);
3284 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3288 ret = send_rename(sctx, from_path, to_path);
3293 struct orphan_dir_info *odi;
3296 odi = get_orphan_dir_info(sctx, rmdir_ino);
3298 /* already deleted */
3303 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3309 name = fs_path_alloc();
3314 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3317 ret = send_rmdir(sctx, name);
3323 ret = send_utimes(sctx, pm->ino, pm->gen);
3328 * After rename/move, need to update the utimes of both new parent(s)
3329 * and old parent(s).
3331 list_for_each_entry(cur, &pm->update_refs, list) {
3333 * The parent inode might have been deleted in the send snapshot
3335 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3336 NULL, NULL, NULL, NULL, NULL);
3337 if (ret == -ENOENT) {
3344 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3351 fs_path_free(from_path);
3352 fs_path_free(to_path);
3353 sctx->send_progress = orig_progress;
3358 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3360 if (!list_empty(&m->list))
3362 if (!RB_EMPTY_NODE(&m->node))
3363 rb_erase(&m->node, &sctx->pending_dir_moves);
3364 __free_recorded_refs(&m->update_refs);
3368 static void tail_append_pending_moves(struct send_ctx *sctx,
3369 struct pending_dir_move *moves,
3370 struct list_head *stack)
3372 if (list_empty(&moves->list)) {
3373 list_add_tail(&moves->list, stack);
3376 list_splice_init(&moves->list, &list);
3377 list_add_tail(&moves->list, stack);
3378 list_splice_tail(&list, stack);
3380 if (!RB_EMPTY_NODE(&moves->node)) {
3381 rb_erase(&moves->node, &sctx->pending_dir_moves);
3382 RB_CLEAR_NODE(&moves->node);
3386 static int apply_children_dir_moves(struct send_ctx *sctx)
3388 struct pending_dir_move *pm;
3389 struct list_head stack;
3390 u64 parent_ino = sctx->cur_ino;
3393 pm = get_pending_dir_moves(sctx, parent_ino);
3397 INIT_LIST_HEAD(&stack);
3398 tail_append_pending_moves(sctx, pm, &stack);
3400 while (!list_empty(&stack)) {
3401 pm = list_first_entry(&stack, struct pending_dir_move, list);
3402 parent_ino = pm->ino;
3403 ret = apply_dir_move(sctx, pm);
3404 free_pending_move(sctx, pm);
3407 pm = get_pending_dir_moves(sctx, parent_ino);
3409 tail_append_pending_moves(sctx, pm, &stack);
3414 while (!list_empty(&stack)) {
3415 pm = list_first_entry(&stack, struct pending_dir_move, list);
3416 free_pending_move(sctx, pm);
3422 * We might need to delay a directory rename even when no ancestor directory
3423 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3424 * renamed. This happens when we rename a directory to the old name (the name
3425 * in the parent root) of some other unrelated directory that got its rename
3426 * delayed due to some ancestor with higher number that got renamed.
3432 * |---- a/ (ino 257)
3433 * | |---- file (ino 260)
3435 * |---- b/ (ino 258)
3436 * |---- c/ (ino 259)
3440 * |---- a/ (ino 258)
3441 * |---- x/ (ino 259)
3442 * |---- y/ (ino 257)
3443 * |----- file (ino 260)
3445 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3446 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3447 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3450 * 1 - rename 259 from 'c' to 'x'
3451 * 2 - rename 257 from 'a' to 'x/y'
3452 * 3 - rename 258 from 'b' to 'a'
3454 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3455 * be done right away and < 0 on error.
3457 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3458 struct recorded_ref *parent_ref,
3459 const bool is_orphan)
3461 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3462 struct btrfs_path *path;
3463 struct btrfs_key key;
3464 struct btrfs_key di_key;
3465 struct btrfs_dir_item *di;
3469 struct waiting_dir_move *wdm;
3471 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3474 path = alloc_path_for_send();
3478 key.objectid = parent_ref->dir;
3479 key.type = BTRFS_DIR_ITEM_KEY;
3480 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3482 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3485 } else if (ret > 0) {
3490 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3491 parent_ref->name_len);
3497 * di_key.objectid has the number of the inode that has a dentry in the
3498 * parent directory with the same name that sctx->cur_ino is being
3499 * renamed to. We need to check if that inode is in the send root as
3500 * well and if it is currently marked as an inode with a pending rename,
3501 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3502 * that it happens after that other inode is renamed.
3504 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3505 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3510 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3511 &left_gen, NULL, NULL, NULL, NULL);
3514 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3515 &right_gen, NULL, NULL, NULL, NULL);
3522 /* Different inode, no need to delay the rename of sctx->cur_ino */
3523 if (right_gen != left_gen) {
3528 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3529 if (wdm && !wdm->orphanized) {
3530 ret = add_pending_dir_move(sctx,
3532 sctx->cur_inode_gen,
3535 &sctx->deleted_refs,
3541 btrfs_free_path(path);
3546 * Check if inode ino2, or any of its ancestors, is inode ino1.
3547 * Return 1 if true, 0 if false and < 0 on error.
3549 static int check_ino_in_path(struct btrfs_root *root,
3554 struct fs_path *fs_path)
3559 return ino1_gen == ino2_gen;
3561 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3566 fs_path_reset(fs_path);
3567 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3571 return parent_gen == ino1_gen;
3578 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3579 * possible path (in case ino2 is not a directory and has multiple hard links).
3580 * Return 1 if true, 0 if false and < 0 on error.
3582 static int is_ancestor(struct btrfs_root *root,
3586 struct fs_path *fs_path)
3588 bool free_fs_path = false;
3590 struct btrfs_path *path = NULL;
3591 struct btrfs_key key;
3594 fs_path = fs_path_alloc();
3597 free_fs_path = true;
3600 path = alloc_path_for_send();
3606 key.objectid = ino2;
3607 key.type = BTRFS_INODE_REF_KEY;
3610 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3615 struct extent_buffer *leaf = path->nodes[0];
3616 int slot = path->slots[0];
3620 if (slot >= btrfs_header_nritems(leaf)) {
3621 ret = btrfs_next_leaf(root, path);
3629 btrfs_item_key_to_cpu(leaf, &key, slot);
3630 if (key.objectid != ino2)
3632 if (key.type != BTRFS_INODE_REF_KEY &&
3633 key.type != BTRFS_INODE_EXTREF_KEY)
3636 item_size = btrfs_item_size_nr(leaf, slot);
3637 while (cur_offset < item_size) {
3641 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3643 struct btrfs_inode_extref *extref;
3645 ptr = btrfs_item_ptr_offset(leaf, slot);
3646 extref = (struct btrfs_inode_extref *)
3648 parent = btrfs_inode_extref_parent(leaf,
3650 cur_offset += sizeof(*extref);
3651 cur_offset += btrfs_inode_extref_name_len(leaf,
3654 parent = key.offset;
3655 cur_offset = item_size;
3658 ret = get_inode_info(root, parent, NULL, &parent_gen,
3659 NULL, NULL, NULL, NULL);
3662 ret = check_ino_in_path(root, ino1, ino1_gen,
3663 parent, parent_gen, fs_path);
3671 btrfs_free_path(path);
3673 fs_path_free(fs_path);
3677 static int wait_for_parent_move(struct send_ctx *sctx,
3678 struct recorded_ref *parent_ref,
3679 const bool is_orphan)
3682 u64 ino = parent_ref->dir;
3683 u64 ino_gen = parent_ref->dir_gen;
3684 u64 parent_ino_before, parent_ino_after;
3685 struct fs_path *path_before = NULL;
3686 struct fs_path *path_after = NULL;
3689 path_after = fs_path_alloc();
3690 path_before = fs_path_alloc();
3691 if (!path_after || !path_before) {
3697 * Our current directory inode may not yet be renamed/moved because some
3698 * ancestor (immediate or not) has to be renamed/moved first. So find if
3699 * such ancestor exists and make sure our own rename/move happens after
3700 * that ancestor is processed to avoid path build infinite loops (done
3701 * at get_cur_path()).
3703 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3704 u64 parent_ino_after_gen;
3706 if (is_waiting_for_move(sctx, ino)) {
3708 * If the current inode is an ancestor of ino in the
3709 * parent root, we need to delay the rename of the
3710 * current inode, otherwise don't delayed the rename
3711 * because we can end up with a circular dependency
3712 * of renames, resulting in some directories never
3713 * getting the respective rename operations issued in
3714 * the send stream or getting into infinite path build
3717 ret = is_ancestor(sctx->parent_root,
3718 sctx->cur_ino, sctx->cur_inode_gen,
3724 fs_path_reset(path_before);
3725 fs_path_reset(path_after);
3727 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3728 &parent_ino_after_gen, path_after);
3731 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3733 if (ret < 0 && ret != -ENOENT) {
3735 } else if (ret == -ENOENT) {
3740 len1 = fs_path_len(path_before);
3741 len2 = fs_path_len(path_after);
3742 if (ino > sctx->cur_ino &&
3743 (parent_ino_before != parent_ino_after || len1 != len2 ||
3744 memcmp(path_before->start, path_after->start, len1))) {
3747 ret = get_inode_info(sctx->parent_root, ino, NULL,
3748 &parent_ino_gen, NULL, NULL, NULL,
3752 if (ino_gen == parent_ino_gen) {
3757 ino = parent_ino_after;
3758 ino_gen = parent_ino_after_gen;
3762 fs_path_free(path_before);
3763 fs_path_free(path_after);
3766 ret = add_pending_dir_move(sctx,
3768 sctx->cur_inode_gen,
3771 &sctx->deleted_refs,
3780 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3783 struct fs_path *new_path;
3786 * Our reference's name member points to its full_path member string, so
3787 * we use here a new path.
3789 new_path = fs_path_alloc();
3793 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3795 fs_path_free(new_path);
3798 ret = fs_path_add(new_path, ref->name, ref->name_len);
3800 fs_path_free(new_path);
3804 fs_path_free(ref->full_path);
3805 set_ref_path(ref, new_path);
3811 * This does all the move/link/unlink/rmdir magic.
3813 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3815 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3817 struct recorded_ref *cur;
3818 struct recorded_ref *cur2;
3819 struct list_head check_dirs;
3820 struct fs_path *valid_path = NULL;
3824 int did_overwrite = 0;
3826 u64 last_dir_ino_rm = 0;
3827 bool can_rename = true;
3828 bool orphanized_dir = false;
3829 bool orphanized_ancestor = false;
3831 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3834 * This should never happen as the root dir always has the same ref
3835 * which is always '..'
3837 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3838 INIT_LIST_HEAD(&check_dirs);
3840 valid_path = fs_path_alloc();
3847 * First, check if the first ref of the current inode was overwritten
3848 * before. If yes, we know that the current inode was already orphanized
3849 * and thus use the orphan name. If not, we can use get_cur_path to
3850 * get the path of the first ref as it would like while receiving at
3851 * this point in time.
3852 * New inodes are always orphan at the beginning, so force to use the
3853 * orphan name in this case.
3854 * The first ref is stored in valid_path and will be updated if it
3855 * gets moved around.
3857 if (!sctx->cur_inode_new) {
3858 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3859 sctx->cur_inode_gen);
3865 if (sctx->cur_inode_new || did_overwrite) {
3866 ret = gen_unique_name(sctx, sctx->cur_ino,
3867 sctx->cur_inode_gen, valid_path);
3872 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3878 list_for_each_entry(cur, &sctx->new_refs, list) {
3880 * We may have refs where the parent directory does not exist
3881 * yet. This happens if the parent directories inum is higher
3882 * than the current inum. To handle this case, we create the
3883 * parent directory out of order. But we need to check if this
3884 * did already happen before due to other refs in the same dir.
3886 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3889 if (ret == inode_state_will_create) {
3892 * First check if any of the current inodes refs did
3893 * already create the dir.
3895 list_for_each_entry(cur2, &sctx->new_refs, list) {
3898 if (cur2->dir == cur->dir) {
3905 * If that did not happen, check if a previous inode
3906 * did already create the dir.
3909 ret = did_create_dir(sctx, cur->dir);
3913 ret = send_create_inode(sctx, cur->dir);
3920 * Check if this new ref would overwrite the first ref of
3921 * another unprocessed inode. If yes, orphanize the
3922 * overwritten inode. If we find an overwritten ref that is
3923 * not the first ref, simply unlink it.
3925 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3926 cur->name, cur->name_len,
3927 &ow_inode, &ow_gen, &ow_mode);
3931 ret = is_first_ref(sctx->parent_root,
3932 ow_inode, cur->dir, cur->name,
3937 struct name_cache_entry *nce;
3938 struct waiting_dir_move *wdm;
3940 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3944 if (S_ISDIR(ow_mode))
3945 orphanized_dir = true;
3948 * If ow_inode has its rename operation delayed
3949 * make sure that its orphanized name is used in
3950 * the source path when performing its rename
3953 if (is_waiting_for_move(sctx, ow_inode)) {
3954 wdm = get_waiting_dir_move(sctx,
3957 wdm->orphanized = true;
3961 * Make sure we clear our orphanized inode's
3962 * name from the name cache. This is because the
3963 * inode ow_inode might be an ancestor of some
3964 * other inode that will be orphanized as well
3965 * later and has an inode number greater than
3966 * sctx->send_progress. We need to prevent
3967 * future name lookups from using the old name
3968 * and get instead the orphan name.
3970 nce = name_cache_search(sctx, ow_inode, ow_gen);
3972 name_cache_delete(sctx, nce);
3977 * ow_inode might currently be an ancestor of
3978 * cur_ino, therefore compute valid_path (the
3979 * current path of cur_ino) again because it
3980 * might contain the pre-orphanization name of
3981 * ow_inode, which is no longer valid.
3983 ret = is_ancestor(sctx->parent_root,
3985 sctx->cur_ino, NULL);
3987 orphanized_ancestor = true;
3988 fs_path_reset(valid_path);
3989 ret = get_cur_path(sctx, sctx->cur_ino,
3990 sctx->cur_inode_gen,
3996 ret = send_unlink(sctx, cur->full_path);
4002 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
4003 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
4012 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
4014 ret = wait_for_parent_move(sctx, cur, is_orphan);
4024 * link/move the ref to the new place. If we have an orphan
4025 * inode, move it and update valid_path. If not, link or move
4026 * it depending on the inode mode.
4028 if (is_orphan && can_rename) {
4029 ret = send_rename(sctx, valid_path, cur->full_path);
4033 ret = fs_path_copy(valid_path, cur->full_path);
4036 } else if (can_rename) {
4037 if (S_ISDIR(sctx->cur_inode_mode)) {
4039 * Dirs can't be linked, so move it. For moved
4040 * dirs, we always have one new and one deleted
4041 * ref. The deleted ref is ignored later.
4043 ret = send_rename(sctx, valid_path,
4046 ret = fs_path_copy(valid_path,
4052 * We might have previously orphanized an inode
4053 * which is an ancestor of our current inode,
4054 * so our reference's full path, which was
4055 * computed before any such orphanizations, must
4058 if (orphanized_dir) {
4059 ret = update_ref_path(sctx, cur);
4063 ret = send_link(sctx, cur->full_path,
4069 ret = dup_ref(cur, &check_dirs);
4074 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4076 * Check if we can already rmdir the directory. If not,
4077 * orphanize it. For every dir item inside that gets deleted
4078 * later, we do this check again and rmdir it then if possible.
4079 * See the use of check_dirs for more details.
4081 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4086 ret = send_rmdir(sctx, valid_path);
4089 } else if (!is_orphan) {
4090 ret = orphanize_inode(sctx, sctx->cur_ino,
4091 sctx->cur_inode_gen, valid_path);
4097 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4098 ret = dup_ref(cur, &check_dirs);
4102 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4103 !list_empty(&sctx->deleted_refs)) {
4105 * We have a moved dir. Add the old parent to check_dirs
4107 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4109 ret = dup_ref(cur, &check_dirs);
4112 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4114 * We have a non dir inode. Go through all deleted refs and
4115 * unlink them if they were not already overwritten by other
4118 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4119 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4120 sctx->cur_ino, sctx->cur_inode_gen,
4121 cur->name, cur->name_len);
4126 * If we orphanized any ancestor before, we need
4127 * to recompute the full path for deleted names,
4128 * since any such path was computed before we
4129 * processed any references and orphanized any
4132 if (orphanized_ancestor) {
4133 ret = update_ref_path(sctx, cur);
4137 ret = send_unlink(sctx, cur->full_path);
4141 ret = dup_ref(cur, &check_dirs);
4146 * If the inode is still orphan, unlink the orphan. This may
4147 * happen when a previous inode did overwrite the first ref
4148 * of this inode and no new refs were added for the current
4149 * inode. Unlinking does not mean that the inode is deleted in
4150 * all cases. There may still be links to this inode in other
4154 ret = send_unlink(sctx, valid_path);
4161 * We did collect all parent dirs where cur_inode was once located. We
4162 * now go through all these dirs and check if they are pending for
4163 * deletion and if it's finally possible to perform the rmdir now.
4164 * We also update the inode stats of the parent dirs here.
4166 list_for_each_entry(cur, &check_dirs, list) {
4168 * In case we had refs into dirs that were not processed yet,
4169 * we don't need to do the utime and rmdir logic for these dirs.
4170 * The dir will be processed later.
4172 if (cur->dir > sctx->cur_ino)
4175 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4179 if (ret == inode_state_did_create ||
4180 ret == inode_state_no_change) {
4181 /* TODO delayed utimes */
4182 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4185 } else if (ret == inode_state_did_delete &&
4186 cur->dir != last_dir_ino_rm) {
4187 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4192 ret = get_cur_path(sctx, cur->dir,
4193 cur->dir_gen, valid_path);
4196 ret = send_rmdir(sctx, valid_path);
4199 last_dir_ino_rm = cur->dir;
4207 __free_recorded_refs(&check_dirs);
4208 free_recorded_refs(sctx);
4209 fs_path_free(valid_path);
4213 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4214 void *ctx, struct list_head *refs)
4217 struct send_ctx *sctx = ctx;
4221 p = fs_path_alloc();
4225 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4230 ret = get_cur_path(sctx, dir, gen, p);
4233 ret = fs_path_add_path(p, name);
4237 ret = __record_ref(refs, dir, gen, p);
4245 static int __record_new_ref(int num, u64 dir, int index,
4246 struct fs_path *name,
4249 struct send_ctx *sctx = ctx;
4250 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4254 static int __record_deleted_ref(int num, u64 dir, int index,
4255 struct fs_path *name,
4258 struct send_ctx *sctx = ctx;
4259 return record_ref(sctx->parent_root, dir, name, ctx,
4260 &sctx->deleted_refs);
4263 static int record_new_ref(struct send_ctx *sctx)
4267 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4268 sctx->cmp_key, 0, __record_new_ref, sctx);
4277 static int record_deleted_ref(struct send_ctx *sctx)
4281 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4282 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4291 struct find_ref_ctx {
4294 struct btrfs_root *root;
4295 struct fs_path *name;
4299 static int __find_iref(int num, u64 dir, int index,
4300 struct fs_path *name,
4303 struct find_ref_ctx *ctx = ctx_;
4307 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4308 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4310 * To avoid doing extra lookups we'll only do this if everything
4313 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4317 if (dir_gen != ctx->dir_gen)
4319 ctx->found_idx = num;
4325 static int find_iref(struct btrfs_root *root,
4326 struct btrfs_path *path,
4327 struct btrfs_key *key,
4328 u64 dir, u64 dir_gen, struct fs_path *name)
4331 struct find_ref_ctx ctx;
4335 ctx.dir_gen = dir_gen;
4339 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4343 if (ctx.found_idx == -1)
4346 return ctx.found_idx;
4349 static int __record_changed_new_ref(int num, u64 dir, int index,
4350 struct fs_path *name,
4355 struct send_ctx *sctx = ctx;
4357 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4362 ret = find_iref(sctx->parent_root, sctx->right_path,
4363 sctx->cmp_key, dir, dir_gen, name);
4365 ret = __record_new_ref(num, dir, index, name, sctx);
4372 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4373 struct fs_path *name,
4378 struct send_ctx *sctx = ctx;
4380 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4385 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4386 dir, dir_gen, name);
4388 ret = __record_deleted_ref(num, dir, index, name, sctx);
4395 static int record_changed_ref(struct send_ctx *sctx)
4399 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4400 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4403 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4404 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4414 * Record and process all refs at once. Needed when an inode changes the
4415 * generation number, which means that it was deleted and recreated.
4417 static int process_all_refs(struct send_ctx *sctx,
4418 enum btrfs_compare_tree_result cmd)
4421 struct btrfs_root *root;
4422 struct btrfs_path *path;
4423 struct btrfs_key key;
4424 struct btrfs_key found_key;
4425 struct extent_buffer *eb;
4427 iterate_inode_ref_t cb;
4428 int pending_move = 0;
4430 path = alloc_path_for_send();
4434 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4435 root = sctx->send_root;
4436 cb = __record_new_ref;
4437 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4438 root = sctx->parent_root;
4439 cb = __record_deleted_ref;
4441 btrfs_err(sctx->send_root->fs_info,
4442 "Wrong command %d in process_all_refs", cmd);
4447 key.objectid = sctx->cmp_key->objectid;
4448 key.type = BTRFS_INODE_REF_KEY;
4450 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4455 eb = path->nodes[0];
4456 slot = path->slots[0];
4457 if (slot >= btrfs_header_nritems(eb)) {
4458 ret = btrfs_next_leaf(root, path);
4466 btrfs_item_key_to_cpu(eb, &found_key, slot);
4468 if (found_key.objectid != key.objectid ||
4469 (found_key.type != BTRFS_INODE_REF_KEY &&
4470 found_key.type != BTRFS_INODE_EXTREF_KEY))
4473 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4479 btrfs_release_path(path);
4482 * We don't actually care about pending_move as we are simply
4483 * re-creating this inode and will be rename'ing it into place once we
4484 * rename the parent directory.
4486 ret = process_recorded_refs(sctx, &pending_move);
4488 btrfs_free_path(path);
4492 static int send_set_xattr(struct send_ctx *sctx,
4493 struct fs_path *path,
4494 const char *name, int name_len,
4495 const char *data, int data_len)
4499 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4503 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4504 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4505 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4507 ret = send_cmd(sctx);
4514 static int send_remove_xattr(struct send_ctx *sctx,
4515 struct fs_path *path,
4516 const char *name, int name_len)
4520 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4524 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4525 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4527 ret = send_cmd(sctx);
4534 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4535 const char *name, int name_len,
4536 const char *data, int data_len,
4540 struct send_ctx *sctx = ctx;
4542 struct posix_acl_xattr_header dummy_acl;
4544 /* Capabilities are emitted by finish_inode_if_needed */
4545 if (!strncmp(name, XATTR_NAME_CAPS, name_len))
4548 p = fs_path_alloc();
4553 * This hack is needed because empty acls are stored as zero byte
4554 * data in xattrs. Problem with that is, that receiving these zero byte
4555 * acls will fail later. To fix this, we send a dummy acl list that
4556 * only contains the version number and no entries.
4558 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4559 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4560 if (data_len == 0) {
4561 dummy_acl.a_version =
4562 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4563 data = (char *)&dummy_acl;
4564 data_len = sizeof(dummy_acl);
4568 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4572 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4579 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4580 const char *name, int name_len,
4581 const char *data, int data_len,
4585 struct send_ctx *sctx = ctx;
4588 p = fs_path_alloc();
4592 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4596 ret = send_remove_xattr(sctx, p, name, name_len);
4603 static int process_new_xattr(struct send_ctx *sctx)
4607 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4608 __process_new_xattr, sctx);
4613 static int process_deleted_xattr(struct send_ctx *sctx)
4615 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4616 __process_deleted_xattr, sctx);
4619 struct find_xattr_ctx {
4627 static int __find_xattr(int num, struct btrfs_key *di_key,
4628 const char *name, int name_len,
4629 const char *data, int data_len,
4630 u8 type, void *vctx)
4632 struct find_xattr_ctx *ctx = vctx;
4634 if (name_len == ctx->name_len &&
4635 strncmp(name, ctx->name, name_len) == 0) {
4636 ctx->found_idx = num;
4637 ctx->found_data_len = data_len;
4638 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4639 if (!ctx->found_data)
4646 static int find_xattr(struct btrfs_root *root,
4647 struct btrfs_path *path,
4648 struct btrfs_key *key,
4649 const char *name, int name_len,
4650 char **data, int *data_len)
4653 struct find_xattr_ctx ctx;
4656 ctx.name_len = name_len;
4658 ctx.found_data = NULL;
4659 ctx.found_data_len = 0;
4661 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4665 if (ctx.found_idx == -1)
4668 *data = ctx.found_data;
4669 *data_len = ctx.found_data_len;
4671 kfree(ctx.found_data);
4673 return ctx.found_idx;
4677 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4678 const char *name, int name_len,
4679 const char *data, int data_len,
4683 struct send_ctx *sctx = ctx;
4684 char *found_data = NULL;
4685 int found_data_len = 0;
4687 ret = find_xattr(sctx->parent_root, sctx->right_path,
4688 sctx->cmp_key, name, name_len, &found_data,
4690 if (ret == -ENOENT) {
4691 ret = __process_new_xattr(num, di_key, name, name_len, data,
4692 data_len, type, ctx);
4693 } else if (ret >= 0) {
4694 if (data_len != found_data_len ||
4695 memcmp(data, found_data, data_len)) {
4696 ret = __process_new_xattr(num, di_key, name, name_len,
4697 data, data_len, type, ctx);
4707 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4708 const char *name, int name_len,
4709 const char *data, int data_len,
4713 struct send_ctx *sctx = ctx;
4715 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4716 name, name_len, NULL, NULL);
4718 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4719 data_len, type, ctx);
4726 static int process_changed_xattr(struct send_ctx *sctx)
4730 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4731 __process_changed_new_xattr, sctx);
4734 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4735 __process_changed_deleted_xattr, sctx);
4741 static int process_all_new_xattrs(struct send_ctx *sctx)
4744 struct btrfs_root *root;
4745 struct btrfs_path *path;
4746 struct btrfs_key key;
4747 struct btrfs_key found_key;
4748 struct extent_buffer *eb;
4751 path = alloc_path_for_send();
4755 root = sctx->send_root;
4757 key.objectid = sctx->cmp_key->objectid;
4758 key.type = BTRFS_XATTR_ITEM_KEY;
4760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4765 eb = path->nodes[0];
4766 slot = path->slots[0];
4767 if (slot >= btrfs_header_nritems(eb)) {
4768 ret = btrfs_next_leaf(root, path);
4771 } else if (ret > 0) {
4778 btrfs_item_key_to_cpu(eb, &found_key, slot);
4779 if (found_key.objectid != key.objectid ||
4780 found_key.type != key.type) {
4785 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4793 btrfs_free_path(path);
4797 static int fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4799 struct btrfs_root *root = sctx->send_root;
4800 struct btrfs_fs_info *fs_info = root->fs_info;
4801 struct inode *inode;
4804 pgoff_t index = offset >> PAGE_SHIFT;
4806 unsigned pg_offset = offset_in_page(offset);
4810 inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root);
4812 return PTR_ERR(inode);
4814 last_index = (offset + len - 1) >> PAGE_SHIFT;
4816 /* initial readahead */
4817 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4818 file_ra_state_init(&sctx->ra, inode->i_mapping);
4820 while (index <= last_index) {
4821 unsigned cur_len = min_t(unsigned, len,
4822 PAGE_SIZE - pg_offset);
4824 page = find_lock_page(inode->i_mapping, index);
4826 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4827 NULL, index, last_index + 1 - index);
4829 page = find_or_create_page(inode->i_mapping, index,
4837 if (PageReadahead(page)) {
4838 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4839 NULL, page, index, last_index + 1 - index);
4842 if (!PageUptodate(page)) {
4843 btrfs_readpage(NULL, page);
4845 if (!PageUptodate(page)) {
4854 memcpy(sctx->read_buf + read, addr + pg_offset, cur_len);
4868 * Read some bytes from the current inode/file and send a write command to
4871 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4873 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4877 p = fs_path_alloc();
4881 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4883 ret = fill_read_buf(sctx, offset, len);
4887 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4891 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4895 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4896 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4897 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4899 ret = send_cmd(sctx);
4908 * Send a clone command to user space.
4910 static int send_clone(struct send_ctx *sctx,
4911 u64 offset, u32 len,
4912 struct clone_root *clone_root)
4918 btrfs_debug(sctx->send_root->fs_info,
4919 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4920 offset, len, clone_root->root->root_key.objectid,
4921 clone_root->ino, clone_root->offset);
4923 p = fs_path_alloc();
4927 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4931 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4935 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4936 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4937 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4939 if (clone_root->root == sctx->send_root) {
4940 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4941 &gen, NULL, NULL, NULL, NULL);
4944 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4946 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4952 * If the parent we're using has a received_uuid set then use that as
4953 * our clone source as that is what we will look for when doing a
4956 * This covers the case that we create a snapshot off of a received
4957 * subvolume and then use that as the parent and try to receive on a
4960 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4961 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4962 clone_root->root->root_item.received_uuid);
4964 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4965 clone_root->root->root_item.uuid);
4966 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4967 le64_to_cpu(clone_root->root->root_item.ctransid));
4968 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4969 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4970 clone_root->offset);
4972 ret = send_cmd(sctx);
4981 * Send an update extent command to user space.
4983 static int send_update_extent(struct send_ctx *sctx,
4984 u64 offset, u32 len)
4989 p = fs_path_alloc();
4993 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4997 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5001 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5002 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5003 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
5005 ret = send_cmd(sctx);
5013 static int send_hole(struct send_ctx *sctx, u64 end)
5015 struct fs_path *p = NULL;
5016 u64 offset = sctx->cur_inode_last_extent;
5021 * A hole that starts at EOF or beyond it. Since we do not yet support
5022 * fallocate (for extent preallocation and hole punching), sending a
5023 * write of zeroes starting at EOF or beyond would later require issuing
5024 * a truncate operation which would undo the write and achieve nothing.
5026 if (offset >= sctx->cur_inode_size)
5030 * Don't go beyond the inode's i_size due to prealloc extents that start
5033 end = min_t(u64, end, sctx->cur_inode_size);
5035 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5036 return send_update_extent(sctx, offset, end - offset);
5038 p = fs_path_alloc();
5041 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5043 goto tlv_put_failure;
5044 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5045 while (offset < end) {
5046 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5048 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5051 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5052 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5053 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5054 ret = send_cmd(sctx);
5059 sctx->cur_inode_next_write_offset = offset;
5065 static int send_extent_data(struct send_ctx *sctx,
5071 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5072 return send_update_extent(sctx, offset, len);
5074 while (sent < len) {
5075 u64 size = len - sent;
5078 if (size > BTRFS_SEND_READ_SIZE)
5079 size = BTRFS_SEND_READ_SIZE;
5080 ret = send_write(sctx, offset + sent, size);
5089 * Search for a capability xattr related to sctx->cur_ino. If the capability is
5090 * found, call send_set_xattr function to emit it.
5092 * Return 0 if there isn't a capability, or when the capability was emitted
5093 * successfully, or < 0 if an error occurred.
5095 static int send_capabilities(struct send_ctx *sctx)
5097 struct fs_path *fspath = NULL;
5098 struct btrfs_path *path;
5099 struct btrfs_dir_item *di;
5100 struct extent_buffer *leaf;
5101 unsigned long data_ptr;
5106 path = alloc_path_for_send();
5110 di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino,
5111 XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0);
5113 /* There is no xattr for this inode */
5115 } else if (IS_ERR(di)) {
5120 leaf = path->nodes[0];
5121 buf_len = btrfs_dir_data_len(leaf, di);
5123 fspath = fs_path_alloc();
5124 buf = kmalloc(buf_len, GFP_KERNEL);
5125 if (!fspath || !buf) {
5130 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
5134 data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di);
5135 read_extent_buffer(leaf, buf, data_ptr, buf_len);
5137 ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS,
5138 strlen(XATTR_NAME_CAPS), buf, buf_len);
5141 fs_path_free(fspath);
5142 btrfs_free_path(path);
5146 static int clone_range(struct send_ctx *sctx,
5147 struct clone_root *clone_root,
5148 const u64 disk_byte,
5153 struct btrfs_path *path;
5154 struct btrfs_key key;
5156 u64 clone_src_i_size = 0;
5159 * Prevent cloning from a zero offset with a length matching the sector
5160 * size because in some scenarios this will make the receiver fail.
5162 * For example, if in the source filesystem the extent at offset 0
5163 * has a length of sectorsize and it was written using direct IO, then
5164 * it can never be an inline extent (even if compression is enabled).
5165 * Then this extent can be cloned in the original filesystem to a non
5166 * zero file offset, but it may not be possible to clone in the
5167 * destination filesystem because it can be inlined due to compression
5168 * on the destination filesystem (as the receiver's write operations are
5169 * always done using buffered IO). The same happens when the original
5170 * filesystem does not have compression enabled but the destination
5173 if (clone_root->offset == 0 &&
5174 len == sctx->send_root->fs_info->sectorsize)
5175 return send_extent_data(sctx, offset, len);
5177 path = alloc_path_for_send();
5182 * There are inodes that have extents that lie behind its i_size. Don't
5183 * accept clones from these extents.
5185 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5186 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5187 btrfs_release_path(path);
5192 * We can't send a clone operation for the entire range if we find
5193 * extent items in the respective range in the source file that
5194 * refer to different extents or if we find holes.
5195 * So check for that and do a mix of clone and regular write/copy
5196 * operations if needed.
5200 * mkfs.btrfs -f /dev/sda
5201 * mount /dev/sda /mnt
5202 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5203 * cp --reflink=always /mnt/foo /mnt/bar
5204 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5205 * btrfs subvolume snapshot -r /mnt /mnt/snap
5207 * If when we send the snapshot and we are processing file bar (which
5208 * has a higher inode number than foo) we blindly send a clone operation
5209 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5210 * a file bar that matches the content of file foo - iow, doesn't match
5211 * the content from bar in the original filesystem.
5213 key.objectid = clone_root->ino;
5214 key.type = BTRFS_EXTENT_DATA_KEY;
5215 key.offset = clone_root->offset;
5216 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5219 if (ret > 0 && path->slots[0] > 0) {
5220 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5221 if (key.objectid == clone_root->ino &&
5222 key.type == BTRFS_EXTENT_DATA_KEY)
5227 struct extent_buffer *leaf = path->nodes[0];
5228 int slot = path->slots[0];
5229 struct btrfs_file_extent_item *ei;
5233 u64 clone_data_offset;
5235 if (slot >= btrfs_header_nritems(leaf)) {
5236 ret = btrfs_next_leaf(clone_root->root, path);
5244 btrfs_item_key_to_cpu(leaf, &key, slot);
5247 * We might have an implicit trailing hole (NO_HOLES feature
5248 * enabled). We deal with it after leaving this loop.
5250 if (key.objectid != clone_root->ino ||
5251 key.type != BTRFS_EXTENT_DATA_KEY)
5254 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5255 type = btrfs_file_extent_type(leaf, ei);
5256 if (type == BTRFS_FILE_EXTENT_INLINE) {
5257 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5258 ext_len = PAGE_ALIGN(ext_len);
5260 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5263 if (key.offset + ext_len <= clone_root->offset)
5266 if (key.offset > clone_root->offset) {
5267 /* Implicit hole, NO_HOLES feature enabled. */
5268 u64 hole_len = key.offset - clone_root->offset;
5272 ret = send_extent_data(sctx, offset, hole_len);
5280 clone_root->offset += hole_len;
5281 data_offset += hole_len;
5284 if (key.offset >= clone_root->offset + len)
5287 if (key.offset >= clone_src_i_size)
5290 if (key.offset + ext_len > clone_src_i_size)
5291 ext_len = clone_src_i_size - key.offset;
5293 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5294 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5295 clone_root->offset = key.offset;
5296 if (clone_data_offset < data_offset &&
5297 clone_data_offset + ext_len > data_offset) {
5300 extent_offset = data_offset - clone_data_offset;
5301 ext_len -= extent_offset;
5302 clone_data_offset += extent_offset;
5303 clone_root->offset += extent_offset;
5307 clone_len = min_t(u64, ext_len, len);
5309 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5310 clone_data_offset == data_offset) {
5311 const u64 src_end = clone_root->offset + clone_len;
5312 const u64 sectorsize = SZ_64K;
5315 * We can't clone the last block, when its size is not
5316 * sector size aligned, into the middle of a file. If we
5317 * do so, the receiver will get a failure (-EINVAL) when
5318 * trying to clone or will silently corrupt the data in
5319 * the destination file if it's on a kernel without the
5320 * fix introduced by commit ac765f83f1397646
5321 * ("Btrfs: fix data corruption due to cloning of eof
5324 * So issue a clone of the aligned down range plus a
5325 * regular write for the eof block, if we hit that case.
5327 * Also, we use the maximum possible sector size, 64K,
5328 * because we don't know what's the sector size of the
5329 * filesystem that receives the stream, so we have to
5330 * assume the largest possible sector size.
5332 if (src_end == clone_src_i_size &&
5333 !IS_ALIGNED(src_end, sectorsize) &&
5334 offset + clone_len < sctx->cur_inode_size) {
5337 slen = ALIGN_DOWN(src_end - clone_root->offset,
5340 ret = send_clone(sctx, offset, slen,
5345 ret = send_extent_data(sctx, offset + slen,
5348 ret = send_clone(sctx, offset, clone_len,
5352 ret = send_extent_data(sctx, offset, clone_len);
5361 offset += clone_len;
5362 clone_root->offset += clone_len;
5363 data_offset += clone_len;
5369 ret = send_extent_data(sctx, offset, len);
5373 btrfs_free_path(path);
5377 static int send_write_or_clone(struct send_ctx *sctx,
5378 struct btrfs_path *path,
5379 struct btrfs_key *key,
5380 struct clone_root *clone_root)
5383 struct btrfs_file_extent_item *ei;
5384 u64 offset = key->offset;
5387 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5389 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5390 struct btrfs_file_extent_item);
5391 type = btrfs_file_extent_type(path->nodes[0], ei);
5392 if (type == BTRFS_FILE_EXTENT_INLINE) {
5393 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5395 * it is possible the inline item won't cover the whole page,
5396 * but there may be items after this page. Make
5397 * sure to send the whole thing
5399 len = PAGE_ALIGN(len);
5401 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5404 if (offset >= sctx->cur_inode_size) {
5408 if (offset + len > sctx->cur_inode_size)
5409 len = sctx->cur_inode_size - offset;
5415 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5419 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5420 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5421 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5424 ret = send_extent_data(sctx, offset, len);
5426 sctx->cur_inode_next_write_offset = offset + len;
5431 static int is_extent_unchanged(struct send_ctx *sctx,
5432 struct btrfs_path *left_path,
5433 struct btrfs_key *ekey)
5436 struct btrfs_key key;
5437 struct btrfs_path *path = NULL;
5438 struct extent_buffer *eb;
5440 struct btrfs_key found_key;
5441 struct btrfs_file_extent_item *ei;
5446 u64 left_offset_fixed;
5454 path = alloc_path_for_send();
5458 eb = left_path->nodes[0];
5459 slot = left_path->slots[0];
5460 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5461 left_type = btrfs_file_extent_type(eb, ei);
5463 if (left_type != BTRFS_FILE_EXTENT_REG) {
5467 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5468 left_len = btrfs_file_extent_num_bytes(eb, ei);
5469 left_offset = btrfs_file_extent_offset(eb, ei);
5470 left_gen = btrfs_file_extent_generation(eb, ei);
5473 * Following comments will refer to these graphics. L is the left
5474 * extents which we are checking at the moment. 1-8 are the right
5475 * extents that we iterate.
5478 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5481 * |--1--|-2b-|...(same as above)
5483 * Alternative situation. Happens on files where extents got split.
5485 * |-----------7-----------|-6-|
5487 * Alternative situation. Happens on files which got larger.
5490 * Nothing follows after 8.
5493 key.objectid = ekey->objectid;
5494 key.type = BTRFS_EXTENT_DATA_KEY;
5495 key.offset = ekey->offset;
5496 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5505 * Handle special case where the right side has no extents at all.
5507 eb = path->nodes[0];
5508 slot = path->slots[0];
5509 btrfs_item_key_to_cpu(eb, &found_key, slot);
5510 if (found_key.objectid != key.objectid ||
5511 found_key.type != key.type) {
5512 /* If we're a hole then just pretend nothing changed */
5513 ret = (left_disknr) ? 0 : 1;
5518 * We're now on 2a, 2b or 7.
5521 while (key.offset < ekey->offset + left_len) {
5522 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5523 right_type = btrfs_file_extent_type(eb, ei);
5524 if (right_type != BTRFS_FILE_EXTENT_REG &&
5525 right_type != BTRFS_FILE_EXTENT_INLINE) {
5530 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5531 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5532 right_len = PAGE_ALIGN(right_len);
5534 right_len = btrfs_file_extent_num_bytes(eb, ei);
5538 * Are we at extent 8? If yes, we know the extent is changed.
5539 * This may only happen on the first iteration.
5541 if (found_key.offset + right_len <= ekey->offset) {
5542 /* If we're a hole just pretend nothing changed */
5543 ret = (left_disknr) ? 0 : 1;
5548 * We just wanted to see if when we have an inline extent, what
5549 * follows it is a regular extent (wanted to check the above
5550 * condition for inline extents too). This should normally not
5551 * happen but it's possible for example when we have an inline
5552 * compressed extent representing data with a size matching
5553 * the page size (currently the same as sector size).
5555 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5560 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5561 right_offset = btrfs_file_extent_offset(eb, ei);
5562 right_gen = btrfs_file_extent_generation(eb, ei);
5564 left_offset_fixed = left_offset;
5565 if (key.offset < ekey->offset) {
5566 /* Fix the right offset for 2a and 7. */
5567 right_offset += ekey->offset - key.offset;
5569 /* Fix the left offset for all behind 2a and 2b */
5570 left_offset_fixed += key.offset - ekey->offset;
5574 * Check if we have the same extent.
5576 if (left_disknr != right_disknr ||
5577 left_offset_fixed != right_offset ||
5578 left_gen != right_gen) {
5584 * Go to the next extent.
5586 ret = btrfs_next_item(sctx->parent_root, path);
5590 eb = path->nodes[0];
5591 slot = path->slots[0];
5592 btrfs_item_key_to_cpu(eb, &found_key, slot);
5594 if (ret || found_key.objectid != key.objectid ||
5595 found_key.type != key.type) {
5596 key.offset += right_len;
5599 if (found_key.offset != key.offset + right_len) {
5607 * We're now behind the left extent (treat as unchanged) or at the end
5608 * of the right side (treat as changed).
5610 if (key.offset >= ekey->offset + left_len)
5617 btrfs_free_path(path);
5621 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5623 struct btrfs_path *path;
5624 struct btrfs_root *root = sctx->send_root;
5625 struct btrfs_key key;
5628 path = alloc_path_for_send();
5632 sctx->cur_inode_last_extent = 0;
5634 key.objectid = sctx->cur_ino;
5635 key.type = BTRFS_EXTENT_DATA_KEY;
5636 key.offset = offset;
5637 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5641 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5642 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5645 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5647 btrfs_free_path(path);
5651 static int range_is_hole_in_parent(struct send_ctx *sctx,
5655 struct btrfs_path *path;
5656 struct btrfs_key key;
5657 struct btrfs_root *root = sctx->parent_root;
5658 u64 search_start = start;
5661 path = alloc_path_for_send();
5665 key.objectid = sctx->cur_ino;
5666 key.type = BTRFS_EXTENT_DATA_KEY;
5667 key.offset = search_start;
5668 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5671 if (ret > 0 && path->slots[0] > 0)
5674 while (search_start < end) {
5675 struct extent_buffer *leaf = path->nodes[0];
5676 int slot = path->slots[0];
5677 struct btrfs_file_extent_item *fi;
5680 if (slot >= btrfs_header_nritems(leaf)) {
5681 ret = btrfs_next_leaf(root, path);
5689 btrfs_item_key_to_cpu(leaf, &key, slot);
5690 if (key.objectid < sctx->cur_ino ||
5691 key.type < BTRFS_EXTENT_DATA_KEY)
5693 if (key.objectid > sctx->cur_ino ||
5694 key.type > BTRFS_EXTENT_DATA_KEY ||
5698 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5699 extent_end = btrfs_file_extent_end(path);
5700 if (extent_end <= start)
5702 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5703 search_start = extent_end;
5713 btrfs_free_path(path);
5717 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5718 struct btrfs_key *key)
5722 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5725 if (sctx->cur_inode_last_extent == (u64)-1) {
5726 ret = get_last_extent(sctx, key->offset - 1);
5731 if (path->slots[0] == 0 &&
5732 sctx->cur_inode_last_extent < key->offset) {
5734 * We might have skipped entire leafs that contained only
5735 * file extent items for our current inode. These leafs have
5736 * a generation number smaller (older) than the one in the
5737 * current leaf and the leaf our last extent came from, and
5738 * are located between these 2 leafs.
5740 ret = get_last_extent(sctx, key->offset - 1);
5745 if (sctx->cur_inode_last_extent < key->offset) {
5746 ret = range_is_hole_in_parent(sctx,
5747 sctx->cur_inode_last_extent,
5752 ret = send_hole(sctx, key->offset);
5756 sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5760 static int process_extent(struct send_ctx *sctx,
5761 struct btrfs_path *path,
5762 struct btrfs_key *key)
5764 struct clone_root *found_clone = NULL;
5767 if (S_ISLNK(sctx->cur_inode_mode))
5770 if (sctx->parent_root && !sctx->cur_inode_new) {
5771 ret = is_extent_unchanged(sctx, path, key);
5779 struct btrfs_file_extent_item *ei;
5782 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5783 struct btrfs_file_extent_item);
5784 type = btrfs_file_extent_type(path->nodes[0], ei);
5785 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5786 type == BTRFS_FILE_EXTENT_REG) {
5788 * The send spec does not have a prealloc command yet,
5789 * so just leave a hole for prealloc'ed extents until
5790 * we have enough commands queued up to justify rev'ing
5793 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5798 /* Have a hole, just skip it. */
5799 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5806 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5807 sctx->cur_inode_size, &found_clone);
5808 if (ret != -ENOENT && ret < 0)
5811 ret = send_write_or_clone(sctx, path, key, found_clone);
5815 ret = maybe_send_hole(sctx, path, key);
5820 static int process_all_extents(struct send_ctx *sctx)
5823 struct btrfs_root *root;
5824 struct btrfs_path *path;
5825 struct btrfs_key key;
5826 struct btrfs_key found_key;
5827 struct extent_buffer *eb;
5830 root = sctx->send_root;
5831 path = alloc_path_for_send();
5835 key.objectid = sctx->cmp_key->objectid;
5836 key.type = BTRFS_EXTENT_DATA_KEY;
5838 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5843 eb = path->nodes[0];
5844 slot = path->slots[0];
5846 if (slot >= btrfs_header_nritems(eb)) {
5847 ret = btrfs_next_leaf(root, path);
5850 } else if (ret > 0) {
5857 btrfs_item_key_to_cpu(eb, &found_key, slot);
5859 if (found_key.objectid != key.objectid ||
5860 found_key.type != key.type) {
5865 ret = process_extent(sctx, path, &found_key);
5873 btrfs_free_path(path);
5877 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5879 int *refs_processed)
5883 if (sctx->cur_ino == 0)
5885 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5886 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5888 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5891 ret = process_recorded_refs(sctx, pending_move);
5895 *refs_processed = 1;
5900 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5911 int need_truncate = 1;
5912 int pending_move = 0;
5913 int refs_processed = 0;
5915 if (sctx->ignore_cur_inode)
5918 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5924 * We have processed the refs and thus need to advance send_progress.
5925 * Now, calls to get_cur_xxx will take the updated refs of the current
5926 * inode into account.
5928 * On the other hand, if our current inode is a directory and couldn't
5929 * be moved/renamed because its parent was renamed/moved too and it has
5930 * a higher inode number, we can only move/rename our current inode
5931 * after we moved/renamed its parent. Therefore in this case operate on
5932 * the old path (pre move/rename) of our current inode, and the
5933 * move/rename will be performed later.
5935 if (refs_processed && !pending_move)
5936 sctx->send_progress = sctx->cur_ino + 1;
5938 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5940 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5943 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5944 &left_mode, &left_uid, &left_gid, NULL);
5948 if (!sctx->parent_root || sctx->cur_inode_new) {
5950 if (!S_ISLNK(sctx->cur_inode_mode))
5952 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5957 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5958 &old_size, NULL, &right_mode, &right_uid,
5963 if (left_uid != right_uid || left_gid != right_gid)
5965 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5967 if ((old_size == sctx->cur_inode_size) ||
5968 (sctx->cur_inode_size > old_size &&
5969 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5973 if (S_ISREG(sctx->cur_inode_mode)) {
5974 if (need_send_hole(sctx)) {
5975 if (sctx->cur_inode_last_extent == (u64)-1 ||
5976 sctx->cur_inode_last_extent <
5977 sctx->cur_inode_size) {
5978 ret = get_last_extent(sctx, (u64)-1);
5982 if (sctx->cur_inode_last_extent <
5983 sctx->cur_inode_size) {
5984 ret = send_hole(sctx, sctx->cur_inode_size);
5989 if (need_truncate) {
5990 ret = send_truncate(sctx, sctx->cur_ino,
5991 sctx->cur_inode_gen,
5992 sctx->cur_inode_size);
5999 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6000 left_uid, left_gid);
6005 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
6011 ret = send_capabilities(sctx);
6016 * If other directory inodes depended on our current directory
6017 * inode's move/rename, now do their move/rename operations.
6019 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
6020 ret = apply_children_dir_moves(sctx);
6024 * Need to send that every time, no matter if it actually
6025 * changed between the two trees as we have done changes to
6026 * the inode before. If our inode is a directory and it's
6027 * waiting to be moved/renamed, we will send its utimes when
6028 * it's moved/renamed, therefore we don't need to do it here.
6030 sctx->send_progress = sctx->cur_ino + 1;
6031 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
6040 struct parent_paths_ctx {
6041 struct list_head *refs;
6042 struct send_ctx *sctx;
6045 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
6048 struct parent_paths_ctx *ppctx = ctx;
6050 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6055 * Issue unlink operations for all paths of the current inode found in the
6058 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6060 LIST_HEAD(deleted_refs);
6061 struct btrfs_path *path;
6062 struct btrfs_key key;
6063 struct parent_paths_ctx ctx;
6066 path = alloc_path_for_send();
6070 key.objectid = sctx->cur_ino;
6071 key.type = BTRFS_INODE_REF_KEY;
6073 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6077 ctx.refs = &deleted_refs;
6081 struct extent_buffer *eb = path->nodes[0];
6082 int slot = path->slots[0];
6084 if (slot >= btrfs_header_nritems(eb)) {
6085 ret = btrfs_next_leaf(sctx->parent_root, path);
6093 btrfs_item_key_to_cpu(eb, &key, slot);
6094 if (key.objectid != sctx->cur_ino)
6096 if (key.type != BTRFS_INODE_REF_KEY &&
6097 key.type != BTRFS_INODE_EXTREF_KEY)
6100 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6101 record_parent_ref, &ctx);
6108 while (!list_empty(&deleted_refs)) {
6109 struct recorded_ref *ref;
6111 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6112 ret = send_unlink(sctx, ref->full_path);
6115 fs_path_free(ref->full_path);
6116 list_del(&ref->list);
6121 btrfs_free_path(path);
6123 __free_recorded_refs(&deleted_refs);
6127 static int changed_inode(struct send_ctx *sctx,
6128 enum btrfs_compare_tree_result result)
6131 struct btrfs_key *key = sctx->cmp_key;
6132 struct btrfs_inode_item *left_ii = NULL;
6133 struct btrfs_inode_item *right_ii = NULL;
6137 sctx->cur_ino = key->objectid;
6138 sctx->cur_inode_new_gen = 0;
6139 sctx->cur_inode_last_extent = (u64)-1;
6140 sctx->cur_inode_next_write_offset = 0;
6141 sctx->ignore_cur_inode = false;
6144 * Set send_progress to current inode. This will tell all get_cur_xxx
6145 * functions that the current inode's refs are not updated yet. Later,
6146 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6148 sctx->send_progress = sctx->cur_ino;
6150 if (result == BTRFS_COMPARE_TREE_NEW ||
6151 result == BTRFS_COMPARE_TREE_CHANGED) {
6152 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6153 sctx->left_path->slots[0],
6154 struct btrfs_inode_item);
6155 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6158 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6159 sctx->right_path->slots[0],
6160 struct btrfs_inode_item);
6161 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6164 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6165 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6166 sctx->right_path->slots[0],
6167 struct btrfs_inode_item);
6169 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6173 * The cur_ino = root dir case is special here. We can't treat
6174 * the inode as deleted+reused because it would generate a
6175 * stream that tries to delete/mkdir the root dir.
6177 if (left_gen != right_gen &&
6178 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6179 sctx->cur_inode_new_gen = 1;
6183 * Normally we do not find inodes with a link count of zero (orphans)
6184 * because the most common case is to create a snapshot and use it
6185 * for a send operation. However other less common use cases involve
6186 * using a subvolume and send it after turning it to RO mode just
6187 * after deleting all hard links of a file while holding an open
6188 * file descriptor against it or turning a RO snapshot into RW mode,
6189 * keep an open file descriptor against a file, delete it and then
6190 * turn the snapshot back to RO mode before using it for a send
6191 * operation. So if we find such cases, ignore the inode and all its
6192 * items completely if it's a new inode, or if it's a changed inode
6193 * make sure all its previous paths (from the parent snapshot) are all
6194 * unlinked and all other the inode items are ignored.
6196 if (result == BTRFS_COMPARE_TREE_NEW ||
6197 result == BTRFS_COMPARE_TREE_CHANGED) {
6200 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6202 sctx->ignore_cur_inode = true;
6203 if (result == BTRFS_COMPARE_TREE_CHANGED)
6204 ret = btrfs_unlink_all_paths(sctx);
6209 if (result == BTRFS_COMPARE_TREE_NEW) {
6210 sctx->cur_inode_gen = left_gen;
6211 sctx->cur_inode_new = 1;
6212 sctx->cur_inode_deleted = 0;
6213 sctx->cur_inode_size = btrfs_inode_size(
6214 sctx->left_path->nodes[0], left_ii);
6215 sctx->cur_inode_mode = btrfs_inode_mode(
6216 sctx->left_path->nodes[0], left_ii);
6217 sctx->cur_inode_rdev = btrfs_inode_rdev(
6218 sctx->left_path->nodes[0], left_ii);
6219 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6220 ret = send_create_inode_if_needed(sctx);
6221 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6222 sctx->cur_inode_gen = right_gen;
6223 sctx->cur_inode_new = 0;
6224 sctx->cur_inode_deleted = 1;
6225 sctx->cur_inode_size = btrfs_inode_size(
6226 sctx->right_path->nodes[0], right_ii);
6227 sctx->cur_inode_mode = btrfs_inode_mode(
6228 sctx->right_path->nodes[0], right_ii);
6229 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6231 * We need to do some special handling in case the inode was
6232 * reported as changed with a changed generation number. This
6233 * means that the original inode was deleted and new inode
6234 * reused the same inum. So we have to treat the old inode as
6235 * deleted and the new one as new.
6237 if (sctx->cur_inode_new_gen) {
6239 * First, process the inode as if it was deleted.
6241 sctx->cur_inode_gen = right_gen;
6242 sctx->cur_inode_new = 0;
6243 sctx->cur_inode_deleted = 1;
6244 sctx->cur_inode_size = btrfs_inode_size(
6245 sctx->right_path->nodes[0], right_ii);
6246 sctx->cur_inode_mode = btrfs_inode_mode(
6247 sctx->right_path->nodes[0], right_ii);
6248 ret = process_all_refs(sctx,
6249 BTRFS_COMPARE_TREE_DELETED);
6254 * Now process the inode as if it was new.
6256 sctx->cur_inode_gen = left_gen;
6257 sctx->cur_inode_new = 1;
6258 sctx->cur_inode_deleted = 0;
6259 sctx->cur_inode_size = btrfs_inode_size(
6260 sctx->left_path->nodes[0], left_ii);
6261 sctx->cur_inode_mode = btrfs_inode_mode(
6262 sctx->left_path->nodes[0], left_ii);
6263 sctx->cur_inode_rdev = btrfs_inode_rdev(
6264 sctx->left_path->nodes[0], left_ii);
6265 ret = send_create_inode_if_needed(sctx);
6269 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6273 * Advance send_progress now as we did not get into
6274 * process_recorded_refs_if_needed in the new_gen case.
6276 sctx->send_progress = sctx->cur_ino + 1;
6279 * Now process all extents and xattrs of the inode as if
6280 * they were all new.
6282 ret = process_all_extents(sctx);
6285 ret = process_all_new_xattrs(sctx);
6289 sctx->cur_inode_gen = left_gen;
6290 sctx->cur_inode_new = 0;
6291 sctx->cur_inode_new_gen = 0;
6292 sctx->cur_inode_deleted = 0;
6293 sctx->cur_inode_size = btrfs_inode_size(
6294 sctx->left_path->nodes[0], left_ii);
6295 sctx->cur_inode_mode = btrfs_inode_mode(
6296 sctx->left_path->nodes[0], left_ii);
6305 * We have to process new refs before deleted refs, but compare_trees gives us
6306 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6307 * first and later process them in process_recorded_refs.
6308 * For the cur_inode_new_gen case, we skip recording completely because
6309 * changed_inode did already initiate processing of refs. The reason for this is
6310 * that in this case, compare_tree actually compares the refs of 2 different
6311 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6312 * refs of the right tree as deleted and all refs of the left tree as new.
6314 static int changed_ref(struct send_ctx *sctx,
6315 enum btrfs_compare_tree_result result)
6319 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6320 inconsistent_snapshot_error(sctx, result, "reference");
6324 if (!sctx->cur_inode_new_gen &&
6325 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6326 if (result == BTRFS_COMPARE_TREE_NEW)
6327 ret = record_new_ref(sctx);
6328 else if (result == BTRFS_COMPARE_TREE_DELETED)
6329 ret = record_deleted_ref(sctx);
6330 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6331 ret = record_changed_ref(sctx);
6338 * Process new/deleted/changed xattrs. We skip processing in the
6339 * cur_inode_new_gen case because changed_inode did already initiate processing
6340 * of xattrs. The reason is the same as in changed_ref
6342 static int changed_xattr(struct send_ctx *sctx,
6343 enum btrfs_compare_tree_result result)
6347 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6348 inconsistent_snapshot_error(sctx, result, "xattr");
6352 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6353 if (result == BTRFS_COMPARE_TREE_NEW)
6354 ret = process_new_xattr(sctx);
6355 else if (result == BTRFS_COMPARE_TREE_DELETED)
6356 ret = process_deleted_xattr(sctx);
6357 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6358 ret = process_changed_xattr(sctx);
6365 * Process new/deleted/changed extents. We skip processing in the
6366 * cur_inode_new_gen case because changed_inode did already initiate processing
6367 * of extents. The reason is the same as in changed_ref
6369 static int changed_extent(struct send_ctx *sctx,
6370 enum btrfs_compare_tree_result result)
6375 * We have found an extent item that changed without the inode item
6376 * having changed. This can happen either after relocation (where the
6377 * disk_bytenr of an extent item is replaced at
6378 * relocation.c:replace_file_extents()) or after deduplication into a
6379 * file in both the parent and send snapshots (where an extent item can
6380 * get modified or replaced with a new one). Note that deduplication
6381 * updates the inode item, but it only changes the iversion (sequence
6382 * field in the inode item) of the inode, so if a file is deduplicated
6383 * the same amount of times in both the parent and send snapshots, its
6384 * iversion becames the same in both snapshots, whence the inode item is
6385 * the same on both snapshots.
6387 if (sctx->cur_ino != sctx->cmp_key->objectid)
6390 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6391 if (result != BTRFS_COMPARE_TREE_DELETED)
6392 ret = process_extent(sctx, sctx->left_path,
6399 static int dir_changed(struct send_ctx *sctx, u64 dir)
6401 u64 orig_gen, new_gen;
6404 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6409 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6414 return (orig_gen != new_gen) ? 1 : 0;
6417 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6418 struct btrfs_key *key)
6420 struct btrfs_inode_extref *extref;
6421 struct extent_buffer *leaf;
6422 u64 dirid = 0, last_dirid = 0;
6429 /* Easy case, just check this one dirid */
6430 if (key->type == BTRFS_INODE_REF_KEY) {
6431 dirid = key->offset;
6433 ret = dir_changed(sctx, dirid);
6437 leaf = path->nodes[0];
6438 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6439 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6440 while (cur_offset < item_size) {
6441 extref = (struct btrfs_inode_extref *)(ptr +
6443 dirid = btrfs_inode_extref_parent(leaf, extref);
6444 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6445 cur_offset += ref_name_len + sizeof(*extref);
6446 if (dirid == last_dirid)
6448 ret = dir_changed(sctx, dirid);
6458 * Updates compare related fields in sctx and simply forwards to the actual
6459 * changed_xxx functions.
6461 static int changed_cb(struct btrfs_path *left_path,
6462 struct btrfs_path *right_path,
6463 struct btrfs_key *key,
6464 enum btrfs_compare_tree_result result,
6468 struct send_ctx *sctx = ctx;
6470 if (result == BTRFS_COMPARE_TREE_SAME) {
6471 if (key->type == BTRFS_INODE_REF_KEY ||
6472 key->type == BTRFS_INODE_EXTREF_KEY) {
6473 ret = compare_refs(sctx, left_path, key);
6478 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6479 return maybe_send_hole(sctx, left_path, key);
6483 result = BTRFS_COMPARE_TREE_CHANGED;
6487 sctx->left_path = left_path;
6488 sctx->right_path = right_path;
6489 sctx->cmp_key = key;
6491 ret = finish_inode_if_needed(sctx, 0);
6495 /* Ignore non-FS objects */
6496 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6497 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6500 if (key->type == BTRFS_INODE_ITEM_KEY) {
6501 ret = changed_inode(sctx, result);
6502 } else if (!sctx->ignore_cur_inode) {
6503 if (key->type == BTRFS_INODE_REF_KEY ||
6504 key->type == BTRFS_INODE_EXTREF_KEY)
6505 ret = changed_ref(sctx, result);
6506 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6507 ret = changed_xattr(sctx, result);
6508 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6509 ret = changed_extent(sctx, result);
6516 static int full_send_tree(struct send_ctx *sctx)
6519 struct btrfs_root *send_root = sctx->send_root;
6520 struct btrfs_key key;
6521 struct btrfs_path *path;
6522 struct extent_buffer *eb;
6525 path = alloc_path_for_send();
6529 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6530 key.type = BTRFS_INODE_ITEM_KEY;
6533 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6540 eb = path->nodes[0];
6541 slot = path->slots[0];
6542 btrfs_item_key_to_cpu(eb, &key, slot);
6544 ret = changed_cb(path, NULL, &key,
6545 BTRFS_COMPARE_TREE_NEW, sctx);
6549 ret = btrfs_next_item(send_root, path);
6559 ret = finish_inode_if_needed(sctx, 1);
6562 btrfs_free_path(path);
6566 static int tree_move_down(struct btrfs_path *path, int *level)
6568 struct extent_buffer *eb;
6570 BUG_ON(*level == 0);
6571 eb = btrfs_read_node_slot(path->nodes[*level], path->slots[*level]);
6575 path->nodes[*level - 1] = eb;
6576 path->slots[*level - 1] = 0;
6581 static int tree_move_next_or_upnext(struct btrfs_path *path,
6582 int *level, int root_level)
6586 nritems = btrfs_header_nritems(path->nodes[*level]);
6588 path->slots[*level]++;
6590 while (path->slots[*level] >= nritems) {
6591 if (*level == root_level)
6595 path->slots[*level] = 0;
6596 free_extent_buffer(path->nodes[*level]);
6597 path->nodes[*level] = NULL;
6599 path->slots[*level]++;
6601 nritems = btrfs_header_nritems(path->nodes[*level]);
6608 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
6611 static int tree_advance(struct btrfs_path *path,
6612 int *level, int root_level,
6614 struct btrfs_key *key)
6618 if (*level == 0 || !allow_down) {
6619 ret = tree_move_next_or_upnext(path, level, root_level);
6621 ret = tree_move_down(path, level);
6625 btrfs_item_key_to_cpu(path->nodes[*level], key,
6626 path->slots[*level]);
6628 btrfs_node_key_to_cpu(path->nodes[*level], key,
6629 path->slots[*level]);
6634 static int tree_compare_item(struct btrfs_path *left_path,
6635 struct btrfs_path *right_path,
6640 unsigned long off1, off2;
6642 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
6643 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
6647 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
6648 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
6649 right_path->slots[0]);
6651 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
6653 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
6660 * This function compares two trees and calls the provided callback for
6661 * every changed/new/deleted item it finds.
6662 * If shared tree blocks are encountered, whole subtrees are skipped, making
6663 * the compare pretty fast on snapshotted subvolumes.
6665 * This currently works on commit roots only. As commit roots are read only,
6666 * we don't do any locking. The commit roots are protected with transactions.
6667 * Transactions are ended and rejoined when a commit is tried in between.
6669 * This function checks for modifications done to the trees while comparing.
6670 * If it detects a change, it aborts immediately.
6672 static int btrfs_compare_trees(struct btrfs_root *left_root,
6673 struct btrfs_root *right_root, void *ctx)
6675 struct btrfs_fs_info *fs_info = left_root->fs_info;
6678 struct btrfs_path *left_path = NULL;
6679 struct btrfs_path *right_path = NULL;
6680 struct btrfs_key left_key;
6681 struct btrfs_key right_key;
6682 char *tmp_buf = NULL;
6683 int left_root_level;
6684 int right_root_level;
6687 int left_end_reached;
6688 int right_end_reached;
6696 left_path = btrfs_alloc_path();
6701 right_path = btrfs_alloc_path();
6707 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
6713 left_path->search_commit_root = 1;
6714 left_path->skip_locking = 1;
6715 right_path->search_commit_root = 1;
6716 right_path->skip_locking = 1;
6719 * Strategy: Go to the first items of both trees. Then do
6721 * If both trees are at level 0
6722 * Compare keys of current items
6723 * If left < right treat left item as new, advance left tree
6725 * If left > right treat right item as deleted, advance right tree
6727 * If left == right do deep compare of items, treat as changed if
6728 * needed, advance both trees and repeat
6729 * If both trees are at the same level but not at level 0
6730 * Compare keys of current nodes/leafs
6731 * If left < right advance left tree and repeat
6732 * If left > right advance right tree and repeat
6733 * If left == right compare blockptrs of the next nodes/leafs
6734 * If they match advance both trees but stay at the same level
6736 * If they don't match advance both trees while allowing to go
6738 * If tree levels are different
6739 * Advance the tree that needs it and repeat
6741 * Advancing a tree means:
6742 * If we are at level 0, try to go to the next slot. If that's not
6743 * possible, go one level up and repeat. Stop when we found a level
6744 * where we could go to the next slot. We may at this point be on a
6747 * If we are not at level 0 and not on shared tree blocks, go one
6750 * If we are not at level 0 and on shared tree blocks, go one slot to
6751 * the right if possible or go up and right.
6754 down_read(&fs_info->commit_root_sem);
6755 left_level = btrfs_header_level(left_root->commit_root);
6756 left_root_level = left_level;
6757 left_path->nodes[left_level] =
6758 btrfs_clone_extent_buffer(left_root->commit_root);
6759 if (!left_path->nodes[left_level]) {
6760 up_read(&fs_info->commit_root_sem);
6765 right_level = btrfs_header_level(right_root->commit_root);
6766 right_root_level = right_level;
6767 right_path->nodes[right_level] =
6768 btrfs_clone_extent_buffer(right_root->commit_root);
6769 if (!right_path->nodes[right_level]) {
6770 up_read(&fs_info->commit_root_sem);
6774 up_read(&fs_info->commit_root_sem);
6776 if (left_level == 0)
6777 btrfs_item_key_to_cpu(left_path->nodes[left_level],
6778 &left_key, left_path->slots[left_level]);
6780 btrfs_node_key_to_cpu(left_path->nodes[left_level],
6781 &left_key, left_path->slots[left_level]);
6782 if (right_level == 0)
6783 btrfs_item_key_to_cpu(right_path->nodes[right_level],
6784 &right_key, right_path->slots[right_level]);
6786 btrfs_node_key_to_cpu(right_path->nodes[right_level],
6787 &right_key, right_path->slots[right_level]);
6789 left_end_reached = right_end_reached = 0;
6790 advance_left = advance_right = 0;
6794 if (advance_left && !left_end_reached) {
6795 ret = tree_advance(left_path, &left_level,
6797 advance_left != ADVANCE_ONLY_NEXT,
6800 left_end_reached = ADVANCE;
6805 if (advance_right && !right_end_reached) {
6806 ret = tree_advance(right_path, &right_level,
6808 advance_right != ADVANCE_ONLY_NEXT,
6811 right_end_reached = ADVANCE;
6817 if (left_end_reached && right_end_reached) {
6820 } else if (left_end_reached) {
6821 if (right_level == 0) {
6822 ret = changed_cb(left_path, right_path,
6824 BTRFS_COMPARE_TREE_DELETED,
6829 advance_right = ADVANCE;
6831 } else if (right_end_reached) {
6832 if (left_level == 0) {
6833 ret = changed_cb(left_path, right_path,
6835 BTRFS_COMPARE_TREE_NEW,
6840 advance_left = ADVANCE;
6844 if (left_level == 0 && right_level == 0) {
6845 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6847 ret = changed_cb(left_path, right_path,
6849 BTRFS_COMPARE_TREE_NEW,
6853 advance_left = ADVANCE;
6854 } else if (cmp > 0) {
6855 ret = changed_cb(left_path, right_path,
6857 BTRFS_COMPARE_TREE_DELETED,
6861 advance_right = ADVANCE;
6863 enum btrfs_compare_tree_result result;
6865 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
6866 ret = tree_compare_item(left_path, right_path,
6869 result = BTRFS_COMPARE_TREE_CHANGED;
6871 result = BTRFS_COMPARE_TREE_SAME;
6872 ret = changed_cb(left_path, right_path,
6873 &left_key, result, ctx);
6876 advance_left = ADVANCE;
6877 advance_right = ADVANCE;
6879 } else if (left_level == right_level) {
6880 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
6882 advance_left = ADVANCE;
6883 } else if (cmp > 0) {
6884 advance_right = ADVANCE;
6886 left_blockptr = btrfs_node_blockptr(
6887 left_path->nodes[left_level],
6888 left_path->slots[left_level]);
6889 right_blockptr = btrfs_node_blockptr(
6890 right_path->nodes[right_level],
6891 right_path->slots[right_level]);
6892 left_gen = btrfs_node_ptr_generation(
6893 left_path->nodes[left_level],
6894 left_path->slots[left_level]);
6895 right_gen = btrfs_node_ptr_generation(
6896 right_path->nodes[right_level],
6897 right_path->slots[right_level]);
6898 if (left_blockptr == right_blockptr &&
6899 left_gen == right_gen) {
6901 * As we're on a shared block, don't
6902 * allow to go deeper.
6904 advance_left = ADVANCE_ONLY_NEXT;
6905 advance_right = ADVANCE_ONLY_NEXT;
6907 advance_left = ADVANCE;
6908 advance_right = ADVANCE;
6911 } else if (left_level < right_level) {
6912 advance_right = ADVANCE;
6914 advance_left = ADVANCE;
6919 btrfs_free_path(left_path);
6920 btrfs_free_path(right_path);
6925 static int send_subvol(struct send_ctx *sctx)
6929 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6930 ret = send_header(sctx);
6935 ret = send_subvol_begin(sctx);
6939 if (sctx->parent_root) {
6940 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx);
6943 ret = finish_inode_if_needed(sctx, 1);
6947 ret = full_send_tree(sctx);
6953 free_recorded_refs(sctx);
6958 * If orphan cleanup did remove any orphans from a root, it means the tree
6959 * was modified and therefore the commit root is not the same as the current
6960 * root anymore. This is a problem, because send uses the commit root and
6961 * therefore can see inode items that don't exist in the current root anymore,
6962 * and for example make calls to btrfs_iget, which will do tree lookups based
6963 * on the current root and not on the commit root. Those lookups will fail,
6964 * returning a -ESTALE error, and making send fail with that error. So make
6965 * sure a send does not see any orphans we have just removed, and that it will
6966 * see the same inodes regardless of whether a transaction commit happened
6967 * before it started (meaning that the commit root will be the same as the
6968 * current root) or not.
6970 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6973 struct btrfs_trans_handle *trans = NULL;
6976 if (sctx->parent_root &&
6977 sctx->parent_root->node != sctx->parent_root->commit_root)
6980 for (i = 0; i < sctx->clone_roots_cnt; i++)
6981 if (sctx->clone_roots[i].root->node !=
6982 sctx->clone_roots[i].root->commit_root)
6986 return btrfs_end_transaction(trans);
6991 /* Use any root, all fs roots will get their commit roots updated. */
6993 trans = btrfs_join_transaction(sctx->send_root);
6995 return PTR_ERR(trans);
6999 return btrfs_commit_transaction(trans);
7003 * Make sure any existing dellaloc is flushed for any root used by a send
7004 * operation so that we do not miss any data and we do not race with writeback
7005 * finishing and changing a tree while send is using the tree. This could
7006 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
7007 * a send operation then uses the subvolume.
7008 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
7010 static int flush_delalloc_roots(struct send_ctx *sctx)
7012 struct btrfs_root *root = sctx->parent_root;
7017 ret = btrfs_start_delalloc_snapshot(root);
7020 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7023 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7024 root = sctx->clone_roots[i].root;
7025 ret = btrfs_start_delalloc_snapshot(root);
7028 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
7034 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
7036 spin_lock(&root->root_item_lock);
7037 root->send_in_progress--;
7039 * Not much left to do, we don't know why it's unbalanced and
7040 * can't blindly reset it to 0.
7042 if (root->send_in_progress < 0)
7043 btrfs_err(root->fs_info,
7044 "send_in_progress unbalanced %d root %llu",
7045 root->send_in_progress, root->root_key.objectid);
7046 spin_unlock(&root->root_item_lock);
7049 static void dedupe_in_progress_warn(const struct btrfs_root *root)
7051 btrfs_warn_rl(root->fs_info,
7052 "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
7053 root->root_key.objectid, root->dedupe_in_progress);
7056 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
7059 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
7060 struct btrfs_fs_info *fs_info = send_root->fs_info;
7061 struct btrfs_root *clone_root;
7062 struct send_ctx *sctx = NULL;
7064 u64 *clone_sources_tmp = NULL;
7065 int clone_sources_to_rollback = 0;
7066 unsigned alloc_size;
7067 int sort_clone_roots = 0;
7069 if (!capable(CAP_SYS_ADMIN))
7073 * The subvolume must remain read-only during send, protect against
7074 * making it RW. This also protects against deletion.
7076 spin_lock(&send_root->root_item_lock);
7077 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
7078 dedupe_in_progress_warn(send_root);
7079 spin_unlock(&send_root->root_item_lock);
7082 send_root->send_in_progress++;
7083 spin_unlock(&send_root->root_item_lock);
7086 * Userspace tools do the checks and warn the user if it's
7089 if (!btrfs_root_readonly(send_root)) {
7095 * Check that we don't overflow at later allocations, we request
7096 * clone_sources_count + 1 items, and compare to unsigned long inside
7099 if (arg->clone_sources_count >
7100 ULONG_MAX / sizeof(struct clone_root) - 1) {
7105 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
7110 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
7116 INIT_LIST_HEAD(&sctx->new_refs);
7117 INIT_LIST_HEAD(&sctx->deleted_refs);
7118 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
7119 INIT_LIST_HEAD(&sctx->name_cache_list);
7121 sctx->flags = arg->flags;
7123 sctx->send_filp = fget(arg->send_fd);
7124 if (!sctx->send_filp) {
7129 sctx->send_root = send_root;
7131 * Unlikely but possible, if the subvolume is marked for deletion but
7132 * is slow to remove the directory entry, send can still be started
7134 if (btrfs_root_dead(sctx->send_root)) {
7139 sctx->clone_roots_cnt = arg->clone_sources_count;
7141 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
7142 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
7143 if (!sctx->send_buf) {
7148 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
7149 if (!sctx->read_buf) {
7154 sctx->pending_dir_moves = RB_ROOT;
7155 sctx->waiting_dir_moves = RB_ROOT;
7156 sctx->orphan_dirs = RB_ROOT;
7158 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
7160 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
7161 if (!sctx->clone_roots) {
7166 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
7168 if (arg->clone_sources_count) {
7169 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
7170 if (!clone_sources_tmp) {
7175 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
7182 for (i = 0; i < arg->clone_sources_count; i++) {
7183 clone_root = btrfs_get_fs_root(fs_info,
7184 clone_sources_tmp[i], true);
7185 if (IS_ERR(clone_root)) {
7186 ret = PTR_ERR(clone_root);
7189 spin_lock(&clone_root->root_item_lock);
7190 if (!btrfs_root_readonly(clone_root) ||
7191 btrfs_root_dead(clone_root)) {
7192 spin_unlock(&clone_root->root_item_lock);
7193 btrfs_put_root(clone_root);
7197 if (clone_root->dedupe_in_progress) {
7198 dedupe_in_progress_warn(clone_root);
7199 spin_unlock(&clone_root->root_item_lock);
7200 btrfs_put_root(clone_root);
7204 clone_root->send_in_progress++;
7205 spin_unlock(&clone_root->root_item_lock);
7207 sctx->clone_roots[i].root = clone_root;
7208 clone_sources_to_rollback = i + 1;
7210 kvfree(clone_sources_tmp);
7211 clone_sources_tmp = NULL;
7214 if (arg->parent_root) {
7215 sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root,
7217 if (IS_ERR(sctx->parent_root)) {
7218 ret = PTR_ERR(sctx->parent_root);
7222 spin_lock(&sctx->parent_root->root_item_lock);
7223 sctx->parent_root->send_in_progress++;
7224 if (!btrfs_root_readonly(sctx->parent_root) ||
7225 btrfs_root_dead(sctx->parent_root)) {
7226 spin_unlock(&sctx->parent_root->root_item_lock);
7230 if (sctx->parent_root->dedupe_in_progress) {
7231 dedupe_in_progress_warn(sctx->parent_root);
7232 spin_unlock(&sctx->parent_root->root_item_lock);
7236 spin_unlock(&sctx->parent_root->root_item_lock);
7240 * Clones from send_root are allowed, but only if the clone source
7241 * is behind the current send position. This is checked while searching
7242 * for possible clone sources.
7244 sctx->clone_roots[sctx->clone_roots_cnt++].root =
7245 btrfs_grab_root(sctx->send_root);
7247 /* We do a bsearch later */
7248 sort(sctx->clone_roots, sctx->clone_roots_cnt,
7249 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
7251 sort_clone_roots = 1;
7253 ret = flush_delalloc_roots(sctx);
7257 ret = ensure_commit_roots_uptodate(sctx);
7261 mutex_lock(&fs_info->balance_mutex);
7262 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
7263 mutex_unlock(&fs_info->balance_mutex);
7264 btrfs_warn_rl(fs_info,
7265 "cannot run send because a balance operation is in progress");
7269 fs_info->send_in_progress++;
7270 mutex_unlock(&fs_info->balance_mutex);
7272 current->journal_info = BTRFS_SEND_TRANS_STUB;
7273 ret = send_subvol(sctx);
7274 current->journal_info = NULL;
7275 mutex_lock(&fs_info->balance_mutex);
7276 fs_info->send_in_progress--;
7277 mutex_unlock(&fs_info->balance_mutex);
7281 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
7282 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
7285 ret = send_cmd(sctx);
7291 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
7292 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
7294 struct pending_dir_move *pm;
7296 n = rb_first(&sctx->pending_dir_moves);
7297 pm = rb_entry(n, struct pending_dir_move, node);
7298 while (!list_empty(&pm->list)) {
7299 struct pending_dir_move *pm2;
7301 pm2 = list_first_entry(&pm->list,
7302 struct pending_dir_move, list);
7303 free_pending_move(sctx, pm2);
7305 free_pending_move(sctx, pm);
7308 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
7309 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
7311 struct waiting_dir_move *dm;
7313 n = rb_first(&sctx->waiting_dir_moves);
7314 dm = rb_entry(n, struct waiting_dir_move, node);
7315 rb_erase(&dm->node, &sctx->waiting_dir_moves);
7319 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
7320 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
7322 struct orphan_dir_info *odi;
7324 n = rb_first(&sctx->orphan_dirs);
7325 odi = rb_entry(n, struct orphan_dir_info, node);
7326 free_orphan_dir_info(sctx, odi);
7329 if (sort_clone_roots) {
7330 for (i = 0; i < sctx->clone_roots_cnt; i++) {
7331 btrfs_root_dec_send_in_progress(
7332 sctx->clone_roots[i].root);
7333 btrfs_put_root(sctx->clone_roots[i].root);
7336 for (i = 0; sctx && i < clone_sources_to_rollback; i++) {
7337 btrfs_root_dec_send_in_progress(
7338 sctx->clone_roots[i].root);
7339 btrfs_put_root(sctx->clone_roots[i].root);
7342 btrfs_root_dec_send_in_progress(send_root);
7344 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) {
7345 btrfs_root_dec_send_in_progress(sctx->parent_root);
7346 btrfs_put_root(sctx->parent_root);
7349 kvfree(clone_sources_tmp);
7352 if (sctx->send_filp)
7353 fput(sctx->send_filp);
7355 kvfree(sctx->clone_roots);
7356 kvfree(sctx->send_buf);
7357 kvfree(sctx->read_buf);
7359 name_cache_free(sctx);