1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2012 Alexander Block. All rights reserved.
6 #include <linux/bsearch.h>
8 #include <linux/file.h>
9 #include <linux/sort.h>
10 #include <linux/mount.h>
11 #include <linux/xattr.h>
12 #include <linux/posix_acl_xattr.h>
13 #include <linux/radix-tree.h>
14 #include <linux/vmalloc.h>
15 #include <linux/string.h>
16 #include <linux/compat.h>
17 #include <linux/crc32c.h>
23 #include "btrfs_inode.h"
24 #include "transaction.h"
25 #include "compression.h"
28 * A fs_path is a helper to dynamically build path names with unknown size.
29 * It reallocates the internal buffer on demand.
30 * It allows fast adding of path elements on the right side (normal path) and
31 * fast adding to the left side (reversed path). A reversed path can also be
32 * unreversed if needed.
41 unsigned short buf_len:15;
42 unsigned short reversed:1;
46 * Average path length does not exceed 200 bytes, we'll have
47 * better packing in the slab and higher chance to satisfy
48 * a allocation later during send.
53 #define FS_PATH_INLINE_SIZE \
54 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
57 /* reused for each extent */
59 struct btrfs_root *root;
66 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
67 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
70 struct file *send_filp;
76 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
77 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
79 struct btrfs_root *send_root;
80 struct btrfs_root *parent_root;
81 struct clone_root *clone_roots;
84 /* current state of the compare_tree call */
85 struct btrfs_path *left_path;
86 struct btrfs_path *right_path;
87 struct btrfs_key *cmp_key;
90 * infos of the currently processed inode. In case of deleted inodes,
91 * these are the values from the deleted inode.
96 int cur_inode_new_gen;
97 int cur_inode_deleted;
101 u64 cur_inode_last_extent;
102 u64 cur_inode_next_write_offset;
103 bool ignore_cur_inode;
107 struct list_head new_refs;
108 struct list_head deleted_refs;
110 struct radix_tree_root name_cache;
111 struct list_head name_cache_list;
114 struct file_ra_state ra;
119 * We process inodes by their increasing order, so if before an
120 * incremental send we reverse the parent/child relationship of
121 * directories such that a directory with a lower inode number was
122 * the parent of a directory with a higher inode number, and the one
123 * becoming the new parent got renamed too, we can't rename/move the
124 * directory with lower inode number when we finish processing it - we
125 * must process the directory with higher inode number first, then
126 * rename/move it and then rename/move the directory with lower inode
127 * number. Example follows.
129 * Tree state when the first send was performed:
141 * Tree state when the second (incremental) send is performed:
150 * The sequence of steps that lead to the second state was:
152 * mv /a/b/c/d /a/b/c2/d2
153 * mv /a/b/c /a/b/c2/d2/cc
155 * "c" has lower inode number, but we can't move it (2nd mv operation)
156 * before we move "d", which has higher inode number.
158 * So we just memorize which move/rename operations must be performed
159 * later when their respective parent is processed and moved/renamed.
162 /* Indexed by parent directory inode number. */
163 struct rb_root pending_dir_moves;
166 * Reverse index, indexed by the inode number of a directory that
167 * is waiting for the move/rename of its immediate parent before its
168 * own move/rename can be performed.
170 struct rb_root waiting_dir_moves;
173 * A directory that is going to be rm'ed might have a child directory
174 * which is in the pending directory moves index above. In this case,
175 * the directory can only be removed after the move/rename of its child
176 * is performed. Example:
196 * Sequence of steps that lead to the send snapshot:
197 * rm -f /a/b/c/foo.txt
199 * mv /a/b/c/x /a/b/YY
202 * When the child is processed, its move/rename is delayed until its
203 * parent is processed (as explained above), but all other operations
204 * like update utimes, chown, chgrp, etc, are performed and the paths
205 * that it uses for those operations must use the orphanized name of
206 * its parent (the directory we're going to rm later), so we need to
207 * memorize that name.
209 * Indexed by the inode number of the directory to be deleted.
211 struct rb_root orphan_dirs;
214 struct pending_dir_move {
216 struct list_head list;
220 struct list_head update_refs;
223 struct waiting_dir_move {
227 * There might be some directory that could not be removed because it
228 * was waiting for this directory inode to be moved first. Therefore
229 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
235 struct orphan_dir_info {
239 u64 last_dir_index_offset;
242 struct name_cache_entry {
243 struct list_head list;
245 * radix_tree has only 32bit entries but we need to handle 64bit inums.
246 * We use the lower 32bit of the 64bit inum to store it in the tree. If
247 * more then one inum would fall into the same entry, we use radix_list
248 * to store the additional entries. radix_list is also used to store
249 * entries where two entries have the same inum but different
252 struct list_head radix_list;
258 int need_later_update;
264 static void inconsistent_snapshot_error(struct send_ctx *sctx,
265 enum btrfs_compare_tree_result result,
268 const char *result_string;
271 case BTRFS_COMPARE_TREE_NEW:
272 result_string = "new";
274 case BTRFS_COMPARE_TREE_DELETED:
275 result_string = "deleted";
277 case BTRFS_COMPARE_TREE_CHANGED:
278 result_string = "updated";
280 case BTRFS_COMPARE_TREE_SAME:
282 result_string = "unchanged";
286 result_string = "unexpected";
289 btrfs_err(sctx->send_root->fs_info,
290 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
291 result_string, what, sctx->cmp_key->objectid,
292 sctx->send_root->root_key.objectid,
294 sctx->parent_root->root_key.objectid : 0));
297 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
299 static struct waiting_dir_move *
300 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
302 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
304 static int need_send_hole(struct send_ctx *sctx)
306 return (sctx->parent_root && !sctx->cur_inode_new &&
307 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
308 S_ISREG(sctx->cur_inode_mode));
311 static void fs_path_reset(struct fs_path *p)
314 p->start = p->buf + p->buf_len - 1;
324 static struct fs_path *fs_path_alloc(void)
328 p = kmalloc(sizeof(*p), GFP_KERNEL);
332 p->buf = p->inline_buf;
333 p->buf_len = FS_PATH_INLINE_SIZE;
338 static struct fs_path *fs_path_alloc_reversed(void)
350 static void fs_path_free(struct fs_path *p)
354 if (p->buf != p->inline_buf)
359 static int fs_path_len(struct fs_path *p)
361 return p->end - p->start;
364 static int fs_path_ensure_buf(struct fs_path *p, int len)
372 if (p->buf_len >= len)
375 if (len > PATH_MAX) {
380 path_len = p->end - p->start;
381 old_buf_len = p->buf_len;
384 * First time the inline_buf does not suffice
386 if (p->buf == p->inline_buf) {
387 tmp_buf = kmalloc(len, GFP_KERNEL);
389 memcpy(tmp_buf, p->buf, old_buf_len);
391 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
397 * The real size of the buffer is bigger, this will let the fast path
398 * happen most of the time
400 p->buf_len = ksize(p->buf);
403 tmp_buf = p->buf + old_buf_len - path_len - 1;
404 p->end = p->buf + p->buf_len - 1;
405 p->start = p->end - path_len;
406 memmove(p->start, tmp_buf, path_len + 1);
409 p->end = p->start + path_len;
414 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
420 new_len = p->end - p->start + name_len;
421 if (p->start != p->end)
423 ret = fs_path_ensure_buf(p, new_len);
428 if (p->start != p->end)
430 p->start -= name_len;
431 *prepared = p->start;
433 if (p->start != p->end)
444 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
449 ret = fs_path_prepare_for_add(p, name_len, &prepared);
452 memcpy(prepared, name, name_len);
458 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
463 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
466 memcpy(prepared, p2->start, p2->end - p2->start);
472 static int fs_path_add_from_extent_buffer(struct fs_path *p,
473 struct extent_buffer *eb,
474 unsigned long off, int len)
479 ret = fs_path_prepare_for_add(p, len, &prepared);
483 read_extent_buffer(eb, prepared, off, len);
489 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
493 p->reversed = from->reversed;
496 ret = fs_path_add_path(p, from);
502 static void fs_path_unreverse(struct fs_path *p)
511 len = p->end - p->start;
513 p->end = p->start + len;
514 memmove(p->start, tmp, len + 1);
518 static struct btrfs_path *alloc_path_for_send(void)
520 struct btrfs_path *path;
522 path = btrfs_alloc_path();
525 path->search_commit_root = 1;
526 path->skip_locking = 1;
527 path->need_commit_sem = 1;
531 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
537 ret = kernel_write(filp, buf + pos, len - pos, off);
538 /* TODO handle that correctly */
539 /*if (ret == -ERESTARTSYS) {
553 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
555 struct btrfs_tlv_header *hdr;
556 int total_len = sizeof(*hdr) + len;
557 int left = sctx->send_max_size - sctx->send_size;
559 if (unlikely(left < total_len))
562 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
563 hdr->tlv_type = cpu_to_le16(attr);
564 hdr->tlv_len = cpu_to_le16(len);
565 memcpy(hdr + 1, data, len);
566 sctx->send_size += total_len;
571 #define TLV_PUT_DEFINE_INT(bits) \
572 static int tlv_put_u##bits(struct send_ctx *sctx, \
573 u##bits attr, u##bits value) \
575 __le##bits __tmp = cpu_to_le##bits(value); \
576 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
579 TLV_PUT_DEFINE_INT(64)
581 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
582 const char *str, int len)
586 return tlv_put(sctx, attr, str, len);
589 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
592 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
595 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
596 struct extent_buffer *eb,
597 struct btrfs_timespec *ts)
599 struct btrfs_timespec bts;
600 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
601 return tlv_put(sctx, attr, &bts, sizeof(bts));
605 #define TLV_PUT(sctx, attrtype, data, attrlen) \
607 ret = tlv_put(sctx, attrtype, data, attrlen); \
609 goto tlv_put_failure; \
612 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
614 ret = tlv_put_u##bits(sctx, attrtype, value); \
616 goto tlv_put_failure; \
619 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
620 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
621 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
622 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
623 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
625 ret = tlv_put_string(sctx, attrtype, str, len); \
627 goto tlv_put_failure; \
629 #define TLV_PUT_PATH(sctx, attrtype, p) \
631 ret = tlv_put_string(sctx, attrtype, p->start, \
632 p->end - p->start); \
634 goto tlv_put_failure; \
636 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
638 ret = tlv_put_uuid(sctx, attrtype, uuid); \
640 goto tlv_put_failure; \
642 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
644 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
646 goto tlv_put_failure; \
649 static int send_header(struct send_ctx *sctx)
651 struct btrfs_stream_header hdr;
653 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
654 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
656 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
661 * For each command/item we want to send to userspace, we call this function.
663 static int begin_cmd(struct send_ctx *sctx, int cmd)
665 struct btrfs_cmd_header *hdr;
667 if (WARN_ON(!sctx->send_buf))
670 BUG_ON(sctx->send_size);
672 sctx->send_size += sizeof(*hdr);
673 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
674 hdr->cmd = cpu_to_le16(cmd);
679 static int send_cmd(struct send_ctx *sctx)
682 struct btrfs_cmd_header *hdr;
685 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
686 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
689 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
690 hdr->crc = cpu_to_le32(crc);
692 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
695 sctx->total_send_size += sctx->send_size;
696 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
703 * Sends a move instruction to user space
705 static int send_rename(struct send_ctx *sctx,
706 struct fs_path *from, struct fs_path *to)
708 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
711 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
713 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
717 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
718 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
720 ret = send_cmd(sctx);
728 * Sends a link instruction to user space
730 static int send_link(struct send_ctx *sctx,
731 struct fs_path *path, struct fs_path *lnk)
733 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
736 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
738 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
742 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
743 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
745 ret = send_cmd(sctx);
753 * Sends an unlink instruction to user space
755 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
757 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
760 btrfs_debug(fs_info, "send_unlink %s", path->start);
762 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
766 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
768 ret = send_cmd(sctx);
776 * Sends a rmdir instruction to user space
778 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
780 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
783 btrfs_debug(fs_info, "send_rmdir %s", path->start);
785 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
789 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
791 ret = send_cmd(sctx);
799 * Helper function to retrieve some fields from an inode item.
801 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
802 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
806 struct btrfs_inode_item *ii;
807 struct btrfs_key key;
810 key.type = BTRFS_INODE_ITEM_KEY;
812 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
819 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
820 struct btrfs_inode_item);
822 *size = btrfs_inode_size(path->nodes[0], ii);
824 *gen = btrfs_inode_generation(path->nodes[0], ii);
826 *mode = btrfs_inode_mode(path->nodes[0], ii);
828 *uid = btrfs_inode_uid(path->nodes[0], ii);
830 *gid = btrfs_inode_gid(path->nodes[0], ii);
832 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
837 static int get_inode_info(struct btrfs_root *root,
838 u64 ino, u64 *size, u64 *gen,
839 u64 *mode, u64 *uid, u64 *gid,
842 struct btrfs_path *path;
845 path = alloc_path_for_send();
848 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
850 btrfs_free_path(path);
854 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
859 * Helper function to iterate the entries in ONE btrfs_inode_ref or
860 * btrfs_inode_extref.
861 * The iterate callback may return a non zero value to stop iteration. This can
862 * be a negative value for error codes or 1 to simply stop it.
864 * path must point to the INODE_REF or INODE_EXTREF when called.
866 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
867 struct btrfs_key *found_key, int resolve,
868 iterate_inode_ref_t iterate, void *ctx)
870 struct extent_buffer *eb = path->nodes[0];
871 struct btrfs_item *item;
872 struct btrfs_inode_ref *iref;
873 struct btrfs_inode_extref *extref;
874 struct btrfs_path *tmp_path;
878 int slot = path->slots[0];
885 unsigned long name_off;
886 unsigned long elem_size;
889 p = fs_path_alloc_reversed();
893 tmp_path = alloc_path_for_send();
900 if (found_key->type == BTRFS_INODE_REF_KEY) {
901 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
902 struct btrfs_inode_ref);
903 item = btrfs_item_nr(slot);
904 total = btrfs_item_size(eb, item);
905 elem_size = sizeof(*iref);
907 ptr = btrfs_item_ptr_offset(eb, slot);
908 total = btrfs_item_size_nr(eb, slot);
909 elem_size = sizeof(*extref);
912 while (cur < total) {
915 if (found_key->type == BTRFS_INODE_REF_KEY) {
916 iref = (struct btrfs_inode_ref *)(ptr + cur);
917 name_len = btrfs_inode_ref_name_len(eb, iref);
918 name_off = (unsigned long)(iref + 1);
919 index = btrfs_inode_ref_index(eb, iref);
920 dir = found_key->offset;
922 extref = (struct btrfs_inode_extref *)(ptr + cur);
923 name_len = btrfs_inode_extref_name_len(eb, extref);
924 name_off = (unsigned long)&extref->name;
925 index = btrfs_inode_extref_index(eb, extref);
926 dir = btrfs_inode_extref_parent(eb, extref);
930 start = btrfs_ref_to_path(root, tmp_path, name_len,
934 ret = PTR_ERR(start);
937 if (start < p->buf) {
938 /* overflow , try again with larger buffer */
939 ret = fs_path_ensure_buf(p,
940 p->buf_len + p->buf - start);
943 start = btrfs_ref_to_path(root, tmp_path,
948 ret = PTR_ERR(start);
951 BUG_ON(start < p->buf);
955 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
961 cur += elem_size + name_len;
962 ret = iterate(num, dir, index, p, ctx);
969 btrfs_free_path(tmp_path);
974 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
975 const char *name, int name_len,
976 const char *data, int data_len,
980 * Helper function to iterate the entries in ONE btrfs_dir_item.
981 * The iterate callback may return a non zero value to stop iteration. This can
982 * be a negative value for error codes or 1 to simply stop it.
984 * path must point to the dir item when called.
986 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
987 iterate_dir_item_t iterate, void *ctx)
990 struct extent_buffer *eb;
991 struct btrfs_item *item;
992 struct btrfs_dir_item *di;
993 struct btrfs_key di_key;
1006 * Start with a small buffer (1 page). If later we end up needing more
1007 * space, which can happen for xattrs on a fs with a leaf size greater
1008 * then the page size, attempt to increase the buffer. Typically xattr
1012 buf = kmalloc(buf_len, GFP_KERNEL);
1018 eb = path->nodes[0];
1019 slot = path->slots[0];
1020 item = btrfs_item_nr(slot);
1021 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1024 total = btrfs_item_size(eb, item);
1027 while (cur < total) {
1028 name_len = btrfs_dir_name_len(eb, di);
1029 data_len = btrfs_dir_data_len(eb, di);
1030 type = btrfs_dir_type(eb, di);
1031 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1033 if (type == BTRFS_FT_XATTR) {
1034 if (name_len > XATTR_NAME_MAX) {
1035 ret = -ENAMETOOLONG;
1038 if (name_len + data_len >
1039 BTRFS_MAX_XATTR_SIZE(root->fs_info)) {
1047 if (name_len + data_len > PATH_MAX) {
1048 ret = -ENAMETOOLONG;
1053 if (name_len + data_len > buf_len) {
1054 buf_len = name_len + data_len;
1055 if (is_vmalloc_addr(buf)) {
1059 char *tmp = krealloc(buf, buf_len,
1060 GFP_KERNEL | __GFP_NOWARN);
1067 buf = kvmalloc(buf_len, GFP_KERNEL);
1075 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1076 name_len + data_len);
1078 len = sizeof(*di) + name_len + data_len;
1079 di = (struct btrfs_dir_item *)((char *)di + len);
1082 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1083 data_len, type, ctx);
1099 static int __copy_first_ref(int num, u64 dir, int index,
1100 struct fs_path *p, void *ctx)
1103 struct fs_path *pt = ctx;
1105 ret = fs_path_copy(pt, p);
1109 /* we want the first only */
1114 * Retrieve the first path of an inode. If an inode has more then one
1115 * ref/hardlink, this is ignored.
1117 static int get_inode_path(struct btrfs_root *root,
1118 u64 ino, struct fs_path *path)
1121 struct btrfs_key key, found_key;
1122 struct btrfs_path *p;
1124 p = alloc_path_for_send();
1128 fs_path_reset(path);
1131 key.type = BTRFS_INODE_REF_KEY;
1134 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1141 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1142 if (found_key.objectid != ino ||
1143 (found_key.type != BTRFS_INODE_REF_KEY &&
1144 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1149 ret = iterate_inode_ref(root, p, &found_key, 1,
1150 __copy_first_ref, path);
1160 struct backref_ctx {
1161 struct send_ctx *sctx;
1163 struct btrfs_path *path;
1164 /* number of total found references */
1168 * used for clones found in send_root. clones found behind cur_objectid
1169 * and cur_offset are not considered as allowed clones.
1174 /* may be truncated in case it's the last extent in a file */
1177 /* data offset in the file extent item */
1180 /* Just to check for bugs in backref resolving */
1184 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1186 u64 root = (u64)(uintptr_t)key;
1187 struct clone_root *cr = (struct clone_root *)elt;
1189 if (root < cr->root->root_key.objectid)
1191 if (root > cr->root->root_key.objectid)
1196 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1198 struct clone_root *cr1 = (struct clone_root *)e1;
1199 struct clone_root *cr2 = (struct clone_root *)e2;
1201 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1203 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1209 * Called for every backref that is found for the current extent.
1210 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1212 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1214 struct backref_ctx *bctx = ctx_;
1215 struct clone_root *found;
1219 /* First check if the root is in the list of accepted clone sources */
1220 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1221 bctx->sctx->clone_roots_cnt,
1222 sizeof(struct clone_root),
1223 __clone_root_cmp_bsearch);
1227 if (found->root == bctx->sctx->send_root &&
1228 ino == bctx->cur_objectid &&
1229 offset == bctx->cur_offset) {
1230 bctx->found_itself = 1;
1234 * There are inodes that have extents that lie behind its i_size. Don't
1235 * accept clones from these extents.
1237 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1239 btrfs_release_path(bctx->path);
1243 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1247 * Make sure we don't consider clones from send_root that are
1248 * behind the current inode/offset.
1250 if (found->root == bctx->sctx->send_root) {
1252 * TODO for the moment we don't accept clones from the inode
1253 * that is currently send. We may change this when
1254 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1257 if (ino >= bctx->cur_objectid)
1262 found->found_refs++;
1263 if (ino < found->ino) {
1265 found->offset = offset;
1266 } else if (found->ino == ino) {
1268 * same extent found more then once in the same file.
1270 if (found->offset > offset + bctx->extent_len)
1271 found->offset = offset;
1278 * Given an inode, offset and extent item, it finds a good clone for a clone
1279 * instruction. Returns -ENOENT when none could be found. The function makes
1280 * sure that the returned clone is usable at the point where sending is at the
1281 * moment. This means, that no clones are accepted which lie behind the current
1284 * path must point to the extent item when called.
1286 static int find_extent_clone(struct send_ctx *sctx,
1287 struct btrfs_path *path,
1288 u64 ino, u64 data_offset,
1290 struct clone_root **found)
1292 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1298 u64 extent_item_pos;
1300 struct btrfs_file_extent_item *fi;
1301 struct extent_buffer *eb = path->nodes[0];
1302 struct backref_ctx *backref_ctx = NULL;
1303 struct clone_root *cur_clone_root;
1304 struct btrfs_key found_key;
1305 struct btrfs_path *tmp_path;
1309 tmp_path = alloc_path_for_send();
1313 /* We only use this path under the commit sem */
1314 tmp_path->need_commit_sem = 0;
1316 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1322 backref_ctx->path = tmp_path;
1324 if (data_offset >= ino_size) {
1326 * There may be extents that lie behind the file's size.
1327 * I at least had this in combination with snapshotting while
1328 * writing large files.
1334 fi = btrfs_item_ptr(eb, path->slots[0],
1335 struct btrfs_file_extent_item);
1336 extent_type = btrfs_file_extent_type(eb, fi);
1337 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1341 compressed = btrfs_file_extent_compression(eb, fi);
1343 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1344 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1345 if (disk_byte == 0) {
1349 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1351 down_read(&fs_info->commit_root_sem);
1352 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1353 &found_key, &flags);
1354 up_read(&fs_info->commit_root_sem);
1355 btrfs_release_path(tmp_path);
1359 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1365 * Setup the clone roots.
1367 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1368 cur_clone_root = sctx->clone_roots + i;
1369 cur_clone_root->ino = (u64)-1;
1370 cur_clone_root->offset = 0;
1371 cur_clone_root->found_refs = 0;
1374 backref_ctx->sctx = sctx;
1375 backref_ctx->found = 0;
1376 backref_ctx->cur_objectid = ino;
1377 backref_ctx->cur_offset = data_offset;
1378 backref_ctx->found_itself = 0;
1379 backref_ctx->extent_len = num_bytes;
1381 * For non-compressed extents iterate_extent_inodes() gives us extent
1382 * offsets that already take into account the data offset, but not for
1383 * compressed extents, since the offset is logical and not relative to
1384 * the physical extent locations. We must take this into account to
1385 * avoid sending clone offsets that go beyond the source file's size,
1386 * which would result in the clone ioctl failing with -EINVAL on the
1389 if (compressed == BTRFS_COMPRESS_NONE)
1390 backref_ctx->data_offset = 0;
1392 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1395 * The last extent of a file may be too large due to page alignment.
1396 * We need to adjust extent_len in this case so that the checks in
1397 * __iterate_backrefs work.
1399 if (data_offset + num_bytes >= ino_size)
1400 backref_ctx->extent_len = ino_size - data_offset;
1403 * Now collect all backrefs.
1405 if (compressed == BTRFS_COMPRESS_NONE)
1406 extent_item_pos = logical - found_key.objectid;
1408 extent_item_pos = 0;
1409 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1410 extent_item_pos, 1, __iterate_backrefs,
1411 backref_ctx, false);
1416 if (!backref_ctx->found_itself) {
1417 /* found a bug in backref code? */
1420 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1421 ino, data_offset, disk_byte, found_key.objectid);
1425 btrfs_debug(fs_info,
1426 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1427 data_offset, ino, num_bytes, logical);
1429 if (!backref_ctx->found)
1430 btrfs_debug(fs_info, "no clones found");
1432 cur_clone_root = NULL;
1433 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1434 if (sctx->clone_roots[i].found_refs) {
1435 if (!cur_clone_root)
1436 cur_clone_root = sctx->clone_roots + i;
1437 else if (sctx->clone_roots[i].root == sctx->send_root)
1438 /* prefer clones from send_root over others */
1439 cur_clone_root = sctx->clone_roots + i;
1444 if (cur_clone_root) {
1445 *found = cur_clone_root;
1452 btrfs_free_path(tmp_path);
1457 static int read_symlink(struct btrfs_root *root,
1459 struct fs_path *dest)
1462 struct btrfs_path *path;
1463 struct btrfs_key key;
1464 struct btrfs_file_extent_item *ei;
1470 path = alloc_path_for_send();
1475 key.type = BTRFS_EXTENT_DATA_KEY;
1477 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1482 * An empty symlink inode. Can happen in rare error paths when
1483 * creating a symlink (transaction committed before the inode
1484 * eviction handler removed the symlink inode items and a crash
1485 * happened in between or the subvol was snapshoted in between).
1486 * Print an informative message to dmesg/syslog so that the user
1487 * can delete the symlink.
1489 btrfs_err(root->fs_info,
1490 "Found empty symlink inode %llu at root %llu",
1491 ino, root->root_key.objectid);
1496 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1497 struct btrfs_file_extent_item);
1498 type = btrfs_file_extent_type(path->nodes[0], ei);
1499 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1500 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1501 BUG_ON(compression);
1503 off = btrfs_file_extent_inline_start(ei);
1504 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1506 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1509 btrfs_free_path(path);
1514 * Helper function to generate a file name that is unique in the root of
1515 * send_root and parent_root. This is used to generate names for orphan inodes.
1517 static int gen_unique_name(struct send_ctx *sctx,
1519 struct fs_path *dest)
1522 struct btrfs_path *path;
1523 struct btrfs_dir_item *di;
1528 path = alloc_path_for_send();
1533 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1535 ASSERT(len < sizeof(tmp));
1537 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1538 path, BTRFS_FIRST_FREE_OBJECTID,
1539 tmp, strlen(tmp), 0);
1540 btrfs_release_path(path);
1546 /* not unique, try again */
1551 if (!sctx->parent_root) {
1557 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1558 path, BTRFS_FIRST_FREE_OBJECTID,
1559 tmp, strlen(tmp), 0);
1560 btrfs_release_path(path);
1566 /* not unique, try again */
1574 ret = fs_path_add(dest, tmp, strlen(tmp));
1577 btrfs_free_path(path);
1582 inode_state_no_change,
1583 inode_state_will_create,
1584 inode_state_did_create,
1585 inode_state_will_delete,
1586 inode_state_did_delete,
1589 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1597 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1599 if (ret < 0 && ret != -ENOENT)
1603 if (!sctx->parent_root) {
1604 right_ret = -ENOENT;
1606 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1607 NULL, NULL, NULL, NULL);
1608 if (ret < 0 && ret != -ENOENT)
1613 if (!left_ret && !right_ret) {
1614 if (left_gen == gen && right_gen == gen) {
1615 ret = inode_state_no_change;
1616 } else if (left_gen == gen) {
1617 if (ino < sctx->send_progress)
1618 ret = inode_state_did_create;
1620 ret = inode_state_will_create;
1621 } else if (right_gen == gen) {
1622 if (ino < sctx->send_progress)
1623 ret = inode_state_did_delete;
1625 ret = inode_state_will_delete;
1629 } else if (!left_ret) {
1630 if (left_gen == gen) {
1631 if (ino < sctx->send_progress)
1632 ret = inode_state_did_create;
1634 ret = inode_state_will_create;
1638 } else if (!right_ret) {
1639 if (right_gen == gen) {
1640 if (ino < sctx->send_progress)
1641 ret = inode_state_did_delete;
1643 ret = inode_state_will_delete;
1655 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1659 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1662 ret = get_cur_inode_state(sctx, ino, gen);
1666 if (ret == inode_state_no_change ||
1667 ret == inode_state_did_create ||
1668 ret == inode_state_will_delete)
1678 * Helper function to lookup a dir item in a dir.
1680 static int lookup_dir_item_inode(struct btrfs_root *root,
1681 u64 dir, const char *name, int name_len,
1686 struct btrfs_dir_item *di;
1687 struct btrfs_key key;
1688 struct btrfs_path *path;
1690 path = alloc_path_for_send();
1694 di = btrfs_lookup_dir_item(NULL, root, path,
1695 dir, name, name_len, 0);
1696 if (IS_ERR_OR_NULL(di)) {
1697 ret = di ? PTR_ERR(di) : -ENOENT;
1700 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1701 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1705 *found_inode = key.objectid;
1706 *found_type = btrfs_dir_type(path->nodes[0], di);
1709 btrfs_free_path(path);
1714 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1715 * generation of the parent dir and the name of the dir entry.
1717 static int get_first_ref(struct btrfs_root *root, u64 ino,
1718 u64 *dir, u64 *dir_gen, struct fs_path *name)
1721 struct btrfs_key key;
1722 struct btrfs_key found_key;
1723 struct btrfs_path *path;
1727 path = alloc_path_for_send();
1732 key.type = BTRFS_INODE_REF_KEY;
1735 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1739 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1741 if (ret || found_key.objectid != ino ||
1742 (found_key.type != BTRFS_INODE_REF_KEY &&
1743 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1748 if (found_key.type == BTRFS_INODE_REF_KEY) {
1749 struct btrfs_inode_ref *iref;
1750 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1751 struct btrfs_inode_ref);
1752 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1753 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1754 (unsigned long)(iref + 1),
1756 parent_dir = found_key.offset;
1758 struct btrfs_inode_extref *extref;
1759 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1760 struct btrfs_inode_extref);
1761 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1762 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1763 (unsigned long)&extref->name, len);
1764 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1768 btrfs_release_path(path);
1771 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1780 btrfs_free_path(path);
1784 static int is_first_ref(struct btrfs_root *root,
1786 const char *name, int name_len)
1789 struct fs_path *tmp_name;
1792 tmp_name = fs_path_alloc();
1796 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1800 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1805 ret = !memcmp(tmp_name->start, name, name_len);
1808 fs_path_free(tmp_name);
1813 * Used by process_recorded_refs to determine if a new ref would overwrite an
1814 * already existing ref. In case it detects an overwrite, it returns the
1815 * inode/gen in who_ino/who_gen.
1816 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1817 * to make sure later references to the overwritten inode are possible.
1818 * Orphanizing is however only required for the first ref of an inode.
1819 * process_recorded_refs does an additional is_first_ref check to see if
1820 * orphanizing is really required.
1822 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1823 const char *name, int name_len,
1824 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1828 u64 other_inode = 0;
1831 if (!sctx->parent_root)
1834 ret = is_inode_existent(sctx, dir, dir_gen);
1839 * If we have a parent root we need to verify that the parent dir was
1840 * not deleted and then re-created, if it was then we have no overwrite
1841 * and we can just unlink this entry.
1843 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1844 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1846 if (ret < 0 && ret != -ENOENT)
1856 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1857 &other_inode, &other_type);
1858 if (ret < 0 && ret != -ENOENT)
1866 * Check if the overwritten ref was already processed. If yes, the ref
1867 * was already unlinked/moved, so we can safely assume that we will not
1868 * overwrite anything at this point in time.
1870 if (other_inode > sctx->send_progress ||
1871 is_waiting_for_move(sctx, other_inode)) {
1872 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1873 who_gen, who_mode, NULL, NULL, NULL);
1878 *who_ino = other_inode;
1888 * Checks if the ref was overwritten by an already processed inode. This is
1889 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1890 * thus the orphan name needs be used.
1891 * process_recorded_refs also uses it to avoid unlinking of refs that were
1894 static int did_overwrite_ref(struct send_ctx *sctx,
1895 u64 dir, u64 dir_gen,
1896 u64 ino, u64 ino_gen,
1897 const char *name, int name_len)
1904 if (!sctx->parent_root)
1907 ret = is_inode_existent(sctx, dir, dir_gen);
1911 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1912 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1914 if (ret < 0 && ret != -ENOENT)
1924 /* check if the ref was overwritten by another ref */
1925 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1926 &ow_inode, &other_type);
1927 if (ret < 0 && ret != -ENOENT)
1930 /* was never and will never be overwritten */
1935 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1940 if (ow_inode == ino && gen == ino_gen) {
1946 * We know that it is or will be overwritten. Check this now.
1947 * The current inode being processed might have been the one that caused
1948 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1949 * the current inode being processed.
1951 if ((ow_inode < sctx->send_progress) ||
1952 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1953 gen == sctx->cur_inode_gen))
1963 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1964 * that got overwritten. This is used by process_recorded_refs to determine
1965 * if it has to use the path as returned by get_cur_path or the orphan name.
1967 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1970 struct fs_path *name = NULL;
1974 if (!sctx->parent_root)
1977 name = fs_path_alloc();
1981 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1985 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1986 name->start, fs_path_len(name));
1994 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1995 * so we need to do some special handling in case we have clashes. This function
1996 * takes care of this with the help of name_cache_entry::radix_list.
1997 * In case of error, nce is kfreed.
1999 static int name_cache_insert(struct send_ctx *sctx,
2000 struct name_cache_entry *nce)
2003 struct list_head *nce_head;
2005 nce_head = radix_tree_lookup(&sctx->name_cache,
2006 (unsigned long)nce->ino);
2008 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2013 INIT_LIST_HEAD(nce_head);
2015 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2022 list_add_tail(&nce->radix_list, nce_head);
2023 list_add_tail(&nce->list, &sctx->name_cache_list);
2024 sctx->name_cache_size++;
2029 static void name_cache_delete(struct send_ctx *sctx,
2030 struct name_cache_entry *nce)
2032 struct list_head *nce_head;
2034 nce_head = radix_tree_lookup(&sctx->name_cache,
2035 (unsigned long)nce->ino);
2037 btrfs_err(sctx->send_root->fs_info,
2038 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2039 nce->ino, sctx->name_cache_size);
2042 list_del(&nce->radix_list);
2043 list_del(&nce->list);
2044 sctx->name_cache_size--;
2047 * We may not get to the final release of nce_head if the lookup fails
2049 if (nce_head && list_empty(nce_head)) {
2050 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2055 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2058 struct list_head *nce_head;
2059 struct name_cache_entry *cur;
2061 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2065 list_for_each_entry(cur, nce_head, radix_list) {
2066 if (cur->ino == ino && cur->gen == gen)
2073 * Removes the entry from the list and adds it back to the end. This marks the
2074 * entry as recently used so that name_cache_clean_unused does not remove it.
2076 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2078 list_del(&nce->list);
2079 list_add_tail(&nce->list, &sctx->name_cache_list);
2083 * Remove some entries from the beginning of name_cache_list.
2085 static void name_cache_clean_unused(struct send_ctx *sctx)
2087 struct name_cache_entry *nce;
2089 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2092 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2093 nce = list_entry(sctx->name_cache_list.next,
2094 struct name_cache_entry, list);
2095 name_cache_delete(sctx, nce);
2100 static void name_cache_free(struct send_ctx *sctx)
2102 struct name_cache_entry *nce;
2104 while (!list_empty(&sctx->name_cache_list)) {
2105 nce = list_entry(sctx->name_cache_list.next,
2106 struct name_cache_entry, list);
2107 name_cache_delete(sctx, nce);
2113 * Used by get_cur_path for each ref up to the root.
2114 * Returns 0 if it succeeded.
2115 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2116 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2117 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2118 * Returns <0 in case of error.
2120 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2124 struct fs_path *dest)
2128 struct name_cache_entry *nce = NULL;
2131 * First check if we already did a call to this function with the same
2132 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2133 * return the cached result.
2135 nce = name_cache_search(sctx, ino, gen);
2137 if (ino < sctx->send_progress && nce->need_later_update) {
2138 name_cache_delete(sctx, nce);
2142 name_cache_used(sctx, nce);
2143 *parent_ino = nce->parent_ino;
2144 *parent_gen = nce->parent_gen;
2145 ret = fs_path_add(dest, nce->name, nce->name_len);
2154 * If the inode is not existent yet, add the orphan name and return 1.
2155 * This should only happen for the parent dir that we determine in
2158 ret = is_inode_existent(sctx, ino, gen);
2163 ret = gen_unique_name(sctx, ino, gen, dest);
2171 * Depending on whether the inode was already processed or not, use
2172 * send_root or parent_root for ref lookup.
2174 if (ino < sctx->send_progress)
2175 ret = get_first_ref(sctx->send_root, ino,
2176 parent_ino, parent_gen, dest);
2178 ret = get_first_ref(sctx->parent_root, ino,
2179 parent_ino, parent_gen, dest);
2184 * Check if the ref was overwritten by an inode's ref that was processed
2185 * earlier. If yes, treat as orphan and return 1.
2187 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2188 dest->start, dest->end - dest->start);
2192 fs_path_reset(dest);
2193 ret = gen_unique_name(sctx, ino, gen, dest);
2201 * Store the result of the lookup in the name cache.
2203 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2211 nce->parent_ino = *parent_ino;
2212 nce->parent_gen = *parent_gen;
2213 nce->name_len = fs_path_len(dest);
2215 strcpy(nce->name, dest->start);
2217 if (ino < sctx->send_progress)
2218 nce->need_later_update = 0;
2220 nce->need_later_update = 1;
2222 nce_ret = name_cache_insert(sctx, nce);
2225 name_cache_clean_unused(sctx);
2232 * Magic happens here. This function returns the first ref to an inode as it
2233 * would look like while receiving the stream at this point in time.
2234 * We walk the path up to the root. For every inode in between, we check if it
2235 * was already processed/sent. If yes, we continue with the parent as found
2236 * in send_root. If not, we continue with the parent as found in parent_root.
2237 * If we encounter an inode that was deleted at this point in time, we use the
2238 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2239 * that were not created yet and overwritten inodes/refs.
2241 * When do we have have orphan inodes:
2242 * 1. When an inode is freshly created and thus no valid refs are available yet
2243 * 2. When a directory lost all it's refs (deleted) but still has dir items
2244 * inside which were not processed yet (pending for move/delete). If anyone
2245 * tried to get the path to the dir items, it would get a path inside that
2247 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2248 * of an unprocessed inode. If in that case the first ref would be
2249 * overwritten, the overwritten inode gets "orphanized". Later when we
2250 * process this overwritten inode, it is restored at a new place by moving
2253 * sctx->send_progress tells this function at which point in time receiving
2256 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2257 struct fs_path *dest)
2260 struct fs_path *name = NULL;
2261 u64 parent_inode = 0;
2265 name = fs_path_alloc();
2272 fs_path_reset(dest);
2274 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2275 struct waiting_dir_move *wdm;
2277 fs_path_reset(name);
2279 if (is_waiting_for_rm(sctx, ino)) {
2280 ret = gen_unique_name(sctx, ino, gen, name);
2283 ret = fs_path_add_path(dest, name);
2287 wdm = get_waiting_dir_move(sctx, ino);
2288 if (wdm && wdm->orphanized) {
2289 ret = gen_unique_name(sctx, ino, gen, name);
2292 ret = get_first_ref(sctx->parent_root, ino,
2293 &parent_inode, &parent_gen, name);
2295 ret = __get_cur_name_and_parent(sctx, ino, gen,
2305 ret = fs_path_add_path(dest, name);
2316 fs_path_unreverse(dest);
2321 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2323 static int send_subvol_begin(struct send_ctx *sctx)
2326 struct btrfs_root *send_root = sctx->send_root;
2327 struct btrfs_root *parent_root = sctx->parent_root;
2328 struct btrfs_path *path;
2329 struct btrfs_key key;
2330 struct btrfs_root_ref *ref;
2331 struct extent_buffer *leaf;
2335 path = btrfs_alloc_path();
2339 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2341 btrfs_free_path(path);
2345 key.objectid = send_root->root_key.objectid;
2346 key.type = BTRFS_ROOT_BACKREF_KEY;
2349 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2358 leaf = path->nodes[0];
2359 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2360 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2361 key.objectid != send_root->root_key.objectid) {
2365 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2366 namelen = btrfs_root_ref_name_len(leaf, ref);
2367 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2368 btrfs_release_path(path);
2371 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2375 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2380 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2382 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2383 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2384 sctx->send_root->root_item.received_uuid);
2386 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2387 sctx->send_root->root_item.uuid);
2389 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2390 le64_to_cpu(sctx->send_root->root_item.ctransid));
2392 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2393 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2394 parent_root->root_item.received_uuid);
2396 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2397 parent_root->root_item.uuid);
2398 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2399 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2402 ret = send_cmd(sctx);
2406 btrfs_free_path(path);
2411 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2413 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2417 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2419 p = fs_path_alloc();
2423 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2427 ret = get_cur_path(sctx, ino, gen, p);
2430 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2431 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2433 ret = send_cmd(sctx);
2441 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2443 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2447 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2449 p = fs_path_alloc();
2453 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2457 ret = get_cur_path(sctx, ino, gen, p);
2460 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2461 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2463 ret = send_cmd(sctx);
2471 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2473 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2477 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2480 p = fs_path_alloc();
2484 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2488 ret = get_cur_path(sctx, ino, gen, p);
2491 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2492 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2493 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2495 ret = send_cmd(sctx);
2503 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2505 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2507 struct fs_path *p = NULL;
2508 struct btrfs_inode_item *ii;
2509 struct btrfs_path *path = NULL;
2510 struct extent_buffer *eb;
2511 struct btrfs_key key;
2514 btrfs_debug(fs_info, "send_utimes %llu", ino);
2516 p = fs_path_alloc();
2520 path = alloc_path_for_send();
2527 key.type = BTRFS_INODE_ITEM_KEY;
2529 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2535 eb = path->nodes[0];
2536 slot = path->slots[0];
2537 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2539 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2543 ret = get_cur_path(sctx, ino, gen, p);
2546 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2547 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2548 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2549 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2550 /* TODO Add otime support when the otime patches get into upstream */
2552 ret = send_cmd(sctx);
2557 btrfs_free_path(path);
2562 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2563 * a valid path yet because we did not process the refs yet. So, the inode
2564 * is created as orphan.
2566 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2568 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2576 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2578 p = fs_path_alloc();
2582 if (ino != sctx->cur_ino) {
2583 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2588 gen = sctx->cur_inode_gen;
2589 mode = sctx->cur_inode_mode;
2590 rdev = sctx->cur_inode_rdev;
2593 if (S_ISREG(mode)) {
2594 cmd = BTRFS_SEND_C_MKFILE;
2595 } else if (S_ISDIR(mode)) {
2596 cmd = BTRFS_SEND_C_MKDIR;
2597 } else if (S_ISLNK(mode)) {
2598 cmd = BTRFS_SEND_C_SYMLINK;
2599 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2600 cmd = BTRFS_SEND_C_MKNOD;
2601 } else if (S_ISFIFO(mode)) {
2602 cmd = BTRFS_SEND_C_MKFIFO;
2603 } else if (S_ISSOCK(mode)) {
2604 cmd = BTRFS_SEND_C_MKSOCK;
2606 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2607 (int)(mode & S_IFMT));
2612 ret = begin_cmd(sctx, cmd);
2616 ret = gen_unique_name(sctx, ino, gen, p);
2620 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2621 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2623 if (S_ISLNK(mode)) {
2625 ret = read_symlink(sctx->send_root, ino, p);
2628 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2629 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2630 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2631 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2632 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2635 ret = send_cmd(sctx);
2647 * We need some special handling for inodes that get processed before the parent
2648 * directory got created. See process_recorded_refs for details.
2649 * This function does the check if we already created the dir out of order.
2651 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2654 struct btrfs_path *path = NULL;
2655 struct btrfs_key key;
2656 struct btrfs_key found_key;
2657 struct btrfs_key di_key;
2658 struct extent_buffer *eb;
2659 struct btrfs_dir_item *di;
2662 path = alloc_path_for_send();
2669 key.type = BTRFS_DIR_INDEX_KEY;
2671 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2676 eb = path->nodes[0];
2677 slot = path->slots[0];
2678 if (slot >= btrfs_header_nritems(eb)) {
2679 ret = btrfs_next_leaf(sctx->send_root, path);
2682 } else if (ret > 0) {
2689 btrfs_item_key_to_cpu(eb, &found_key, slot);
2690 if (found_key.objectid != key.objectid ||
2691 found_key.type != key.type) {
2696 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2697 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2699 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2700 di_key.objectid < sctx->send_progress) {
2709 btrfs_free_path(path);
2714 * Only creates the inode if it is:
2715 * 1. Not a directory
2716 * 2. Or a directory which was not created already due to out of order
2717 * directories. See did_create_dir and process_recorded_refs for details.
2719 static int send_create_inode_if_needed(struct send_ctx *sctx)
2723 if (S_ISDIR(sctx->cur_inode_mode)) {
2724 ret = did_create_dir(sctx, sctx->cur_ino);
2733 ret = send_create_inode(sctx, sctx->cur_ino);
2741 struct recorded_ref {
2742 struct list_head list;
2744 struct fs_path *full_path;
2750 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2752 ref->full_path = path;
2753 ref->name = (char *)kbasename(ref->full_path->start);
2754 ref->name_len = ref->full_path->end - ref->name;
2758 * We need to process new refs before deleted refs, but compare_tree gives us
2759 * everything mixed. So we first record all refs and later process them.
2760 * This function is a helper to record one ref.
2762 static int __record_ref(struct list_head *head, u64 dir,
2763 u64 dir_gen, struct fs_path *path)
2765 struct recorded_ref *ref;
2767 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2772 ref->dir_gen = dir_gen;
2773 set_ref_path(ref, path);
2774 list_add_tail(&ref->list, head);
2778 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2780 struct recorded_ref *new;
2782 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2786 new->dir = ref->dir;
2787 new->dir_gen = ref->dir_gen;
2788 new->full_path = NULL;
2789 INIT_LIST_HEAD(&new->list);
2790 list_add_tail(&new->list, list);
2794 static void __free_recorded_refs(struct list_head *head)
2796 struct recorded_ref *cur;
2798 while (!list_empty(head)) {
2799 cur = list_entry(head->next, struct recorded_ref, list);
2800 fs_path_free(cur->full_path);
2801 list_del(&cur->list);
2806 static void free_recorded_refs(struct send_ctx *sctx)
2808 __free_recorded_refs(&sctx->new_refs);
2809 __free_recorded_refs(&sctx->deleted_refs);
2813 * Renames/moves a file/dir to its orphan name. Used when the first
2814 * ref of an unprocessed inode gets overwritten and for all non empty
2817 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2818 struct fs_path *path)
2821 struct fs_path *orphan;
2823 orphan = fs_path_alloc();
2827 ret = gen_unique_name(sctx, ino, gen, orphan);
2831 ret = send_rename(sctx, path, orphan);
2834 fs_path_free(orphan);
2838 static struct orphan_dir_info *
2839 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2841 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2842 struct rb_node *parent = NULL;
2843 struct orphan_dir_info *entry, *odi;
2847 entry = rb_entry(parent, struct orphan_dir_info, node);
2848 if (dir_ino < entry->ino) {
2850 } else if (dir_ino > entry->ino) {
2851 p = &(*p)->rb_right;
2857 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2859 return ERR_PTR(-ENOMEM);
2862 odi->last_dir_index_offset = 0;
2864 rb_link_node(&odi->node, parent, p);
2865 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2869 static struct orphan_dir_info *
2870 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2872 struct rb_node *n = sctx->orphan_dirs.rb_node;
2873 struct orphan_dir_info *entry;
2876 entry = rb_entry(n, struct orphan_dir_info, node);
2877 if (dir_ino < entry->ino)
2879 else if (dir_ino > entry->ino)
2887 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2889 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2894 static void free_orphan_dir_info(struct send_ctx *sctx,
2895 struct orphan_dir_info *odi)
2899 rb_erase(&odi->node, &sctx->orphan_dirs);
2904 * Returns 1 if a directory can be removed at this point in time.
2905 * We check this by iterating all dir items and checking if the inode behind
2906 * the dir item was already processed.
2908 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2912 struct btrfs_root *root = sctx->parent_root;
2913 struct btrfs_path *path;
2914 struct btrfs_key key;
2915 struct btrfs_key found_key;
2916 struct btrfs_key loc;
2917 struct btrfs_dir_item *di;
2918 struct orphan_dir_info *odi = NULL;
2921 * Don't try to rmdir the top/root subvolume dir.
2923 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2926 path = alloc_path_for_send();
2931 key.type = BTRFS_DIR_INDEX_KEY;
2934 odi = get_orphan_dir_info(sctx, dir);
2936 key.offset = odi->last_dir_index_offset;
2938 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2943 struct waiting_dir_move *dm;
2945 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2946 ret = btrfs_next_leaf(root, path);
2953 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2955 if (found_key.objectid != key.objectid ||
2956 found_key.type != key.type)
2959 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2960 struct btrfs_dir_item);
2961 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2963 dm = get_waiting_dir_move(sctx, loc.objectid);
2965 odi = add_orphan_dir_info(sctx, dir);
2971 odi->last_dir_index_offset = found_key.offset;
2972 dm->rmdir_ino = dir;
2977 if (loc.objectid > send_progress) {
2978 odi = add_orphan_dir_info(sctx, dir);
2984 odi->last_dir_index_offset = found_key.offset;
2991 free_orphan_dir_info(sctx, odi);
2996 btrfs_free_path(path);
3000 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3002 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3004 return entry != NULL;
3007 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3009 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3010 struct rb_node *parent = NULL;
3011 struct waiting_dir_move *entry, *dm;
3013 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3018 dm->orphanized = orphanized;
3022 entry = rb_entry(parent, struct waiting_dir_move, node);
3023 if (ino < entry->ino) {
3025 } else if (ino > entry->ino) {
3026 p = &(*p)->rb_right;
3033 rb_link_node(&dm->node, parent, p);
3034 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3038 static struct waiting_dir_move *
3039 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3041 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3042 struct waiting_dir_move *entry;
3045 entry = rb_entry(n, struct waiting_dir_move, node);
3046 if (ino < entry->ino)
3048 else if (ino > entry->ino)
3056 static void free_waiting_dir_move(struct send_ctx *sctx,
3057 struct waiting_dir_move *dm)
3061 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3065 static int add_pending_dir_move(struct send_ctx *sctx,
3069 struct list_head *new_refs,
3070 struct list_head *deleted_refs,
3071 const bool is_orphan)
3073 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3074 struct rb_node *parent = NULL;
3075 struct pending_dir_move *entry = NULL, *pm;
3076 struct recorded_ref *cur;
3080 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3083 pm->parent_ino = parent_ino;
3086 INIT_LIST_HEAD(&pm->list);
3087 INIT_LIST_HEAD(&pm->update_refs);
3088 RB_CLEAR_NODE(&pm->node);
3092 entry = rb_entry(parent, struct pending_dir_move, node);
3093 if (parent_ino < entry->parent_ino) {
3095 } else if (parent_ino > entry->parent_ino) {
3096 p = &(*p)->rb_right;
3103 list_for_each_entry(cur, deleted_refs, list) {
3104 ret = dup_ref(cur, &pm->update_refs);
3108 list_for_each_entry(cur, new_refs, list) {
3109 ret = dup_ref(cur, &pm->update_refs);
3114 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3119 list_add_tail(&pm->list, &entry->list);
3121 rb_link_node(&pm->node, parent, p);
3122 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3127 __free_recorded_refs(&pm->update_refs);
3133 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3136 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3137 struct pending_dir_move *entry;
3140 entry = rb_entry(n, struct pending_dir_move, node);
3141 if (parent_ino < entry->parent_ino)
3143 else if (parent_ino > entry->parent_ino)
3151 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3152 u64 ino, u64 gen, u64 *ancestor_ino)
3155 u64 parent_inode = 0;
3157 u64 start_ino = ino;
3160 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3161 fs_path_reset(name);
3163 if (is_waiting_for_rm(sctx, ino))
3165 if (is_waiting_for_move(sctx, ino)) {
3166 if (*ancestor_ino == 0)
3167 *ancestor_ino = ino;
3168 ret = get_first_ref(sctx->parent_root, ino,
3169 &parent_inode, &parent_gen, name);
3171 ret = __get_cur_name_and_parent(sctx, ino, gen,
3181 if (parent_inode == start_ino) {
3183 if (*ancestor_ino == 0)
3184 *ancestor_ino = ino;
3193 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3195 struct fs_path *from_path = NULL;
3196 struct fs_path *to_path = NULL;
3197 struct fs_path *name = NULL;
3198 u64 orig_progress = sctx->send_progress;
3199 struct recorded_ref *cur;
3200 u64 parent_ino, parent_gen;
3201 struct waiting_dir_move *dm = NULL;
3207 name = fs_path_alloc();
3208 from_path = fs_path_alloc();
3209 if (!name || !from_path) {
3214 dm = get_waiting_dir_move(sctx, pm->ino);
3216 rmdir_ino = dm->rmdir_ino;
3217 is_orphan = dm->orphanized;
3218 free_waiting_dir_move(sctx, dm);
3221 ret = gen_unique_name(sctx, pm->ino,
3222 pm->gen, from_path);
3224 ret = get_first_ref(sctx->parent_root, pm->ino,
3225 &parent_ino, &parent_gen, name);
3228 ret = get_cur_path(sctx, parent_ino, parent_gen,
3232 ret = fs_path_add_path(from_path, name);
3237 sctx->send_progress = sctx->cur_ino + 1;
3238 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3242 LIST_HEAD(deleted_refs);
3243 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3244 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3245 &pm->update_refs, &deleted_refs,
3250 dm = get_waiting_dir_move(sctx, pm->ino);
3252 dm->rmdir_ino = rmdir_ino;
3256 fs_path_reset(name);
3259 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3263 ret = send_rename(sctx, from_path, to_path);
3268 struct orphan_dir_info *odi;
3271 odi = get_orphan_dir_info(sctx, rmdir_ino);
3273 /* already deleted */
3278 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3284 name = fs_path_alloc();
3289 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3292 ret = send_rmdir(sctx, name);
3298 ret = send_utimes(sctx, pm->ino, pm->gen);
3303 * After rename/move, need to update the utimes of both new parent(s)
3304 * and old parent(s).
3306 list_for_each_entry(cur, &pm->update_refs, list) {
3308 * The parent inode might have been deleted in the send snapshot
3310 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3311 NULL, NULL, NULL, NULL, NULL);
3312 if (ret == -ENOENT) {
3319 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3326 fs_path_free(from_path);
3327 fs_path_free(to_path);
3328 sctx->send_progress = orig_progress;
3333 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3335 if (!list_empty(&m->list))
3337 if (!RB_EMPTY_NODE(&m->node))
3338 rb_erase(&m->node, &sctx->pending_dir_moves);
3339 __free_recorded_refs(&m->update_refs);
3343 static void tail_append_pending_moves(struct send_ctx *sctx,
3344 struct pending_dir_move *moves,
3345 struct list_head *stack)
3347 if (list_empty(&moves->list)) {
3348 list_add_tail(&moves->list, stack);
3351 list_splice_init(&moves->list, &list);
3352 list_add_tail(&moves->list, stack);
3353 list_splice_tail(&list, stack);
3355 if (!RB_EMPTY_NODE(&moves->node)) {
3356 rb_erase(&moves->node, &sctx->pending_dir_moves);
3357 RB_CLEAR_NODE(&moves->node);
3361 static int apply_children_dir_moves(struct send_ctx *sctx)
3363 struct pending_dir_move *pm;
3364 struct list_head stack;
3365 u64 parent_ino = sctx->cur_ino;
3368 pm = get_pending_dir_moves(sctx, parent_ino);
3372 INIT_LIST_HEAD(&stack);
3373 tail_append_pending_moves(sctx, pm, &stack);
3375 while (!list_empty(&stack)) {
3376 pm = list_first_entry(&stack, struct pending_dir_move, list);
3377 parent_ino = pm->ino;
3378 ret = apply_dir_move(sctx, pm);
3379 free_pending_move(sctx, pm);
3382 pm = get_pending_dir_moves(sctx, parent_ino);
3384 tail_append_pending_moves(sctx, pm, &stack);
3389 while (!list_empty(&stack)) {
3390 pm = list_first_entry(&stack, struct pending_dir_move, list);
3391 free_pending_move(sctx, pm);
3397 * We might need to delay a directory rename even when no ancestor directory
3398 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3399 * renamed. This happens when we rename a directory to the old name (the name
3400 * in the parent root) of some other unrelated directory that got its rename
3401 * delayed due to some ancestor with higher number that got renamed.
3407 * |---- a/ (ino 257)
3408 * | |---- file (ino 260)
3410 * |---- b/ (ino 258)
3411 * |---- c/ (ino 259)
3415 * |---- a/ (ino 258)
3416 * |---- x/ (ino 259)
3417 * |---- y/ (ino 257)
3418 * |----- file (ino 260)
3420 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3421 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3422 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3425 * 1 - rename 259 from 'c' to 'x'
3426 * 2 - rename 257 from 'a' to 'x/y'
3427 * 3 - rename 258 from 'b' to 'a'
3429 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3430 * be done right away and < 0 on error.
3432 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3433 struct recorded_ref *parent_ref,
3434 const bool is_orphan)
3436 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3437 struct btrfs_path *path;
3438 struct btrfs_key key;
3439 struct btrfs_key di_key;
3440 struct btrfs_dir_item *di;
3444 struct waiting_dir_move *wdm;
3446 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3449 path = alloc_path_for_send();
3453 key.objectid = parent_ref->dir;
3454 key.type = BTRFS_DIR_ITEM_KEY;
3455 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3457 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3460 } else if (ret > 0) {
3465 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3466 parent_ref->name_len);
3472 * di_key.objectid has the number of the inode that has a dentry in the
3473 * parent directory with the same name that sctx->cur_ino is being
3474 * renamed to. We need to check if that inode is in the send root as
3475 * well and if it is currently marked as an inode with a pending rename,
3476 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3477 * that it happens after that other inode is renamed.
3479 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3480 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3485 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3486 &left_gen, NULL, NULL, NULL, NULL);
3489 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3490 &right_gen, NULL, NULL, NULL, NULL);
3497 /* Different inode, no need to delay the rename of sctx->cur_ino */
3498 if (right_gen != left_gen) {
3503 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3504 if (wdm && !wdm->orphanized) {
3505 ret = add_pending_dir_move(sctx,
3507 sctx->cur_inode_gen,
3510 &sctx->deleted_refs,
3516 btrfs_free_path(path);
3521 * Check if inode ino2, or any of its ancestors, is inode ino1.
3522 * Return 1 if true, 0 if false and < 0 on error.
3524 static int check_ino_in_path(struct btrfs_root *root,
3529 struct fs_path *fs_path)
3534 return ino1_gen == ino2_gen;
3536 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3541 fs_path_reset(fs_path);
3542 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3546 return parent_gen == ino1_gen;
3553 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3554 * possible path (in case ino2 is not a directory and has multiple hard links).
3555 * Return 1 if true, 0 if false and < 0 on error.
3557 static int is_ancestor(struct btrfs_root *root,
3561 struct fs_path *fs_path)
3563 bool free_fs_path = false;
3565 struct btrfs_path *path = NULL;
3566 struct btrfs_key key;
3569 fs_path = fs_path_alloc();
3572 free_fs_path = true;
3575 path = alloc_path_for_send();
3581 key.objectid = ino2;
3582 key.type = BTRFS_INODE_REF_KEY;
3585 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3590 struct extent_buffer *leaf = path->nodes[0];
3591 int slot = path->slots[0];
3595 if (slot >= btrfs_header_nritems(leaf)) {
3596 ret = btrfs_next_leaf(root, path);
3604 btrfs_item_key_to_cpu(leaf, &key, slot);
3605 if (key.objectid != ino2)
3607 if (key.type != BTRFS_INODE_REF_KEY &&
3608 key.type != BTRFS_INODE_EXTREF_KEY)
3611 item_size = btrfs_item_size_nr(leaf, slot);
3612 while (cur_offset < item_size) {
3616 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3618 struct btrfs_inode_extref *extref;
3620 ptr = btrfs_item_ptr_offset(leaf, slot);
3621 extref = (struct btrfs_inode_extref *)
3623 parent = btrfs_inode_extref_parent(leaf,
3625 cur_offset += sizeof(*extref);
3626 cur_offset += btrfs_inode_extref_name_len(leaf,
3629 parent = key.offset;
3630 cur_offset = item_size;
3633 ret = get_inode_info(root, parent, NULL, &parent_gen,
3634 NULL, NULL, NULL, NULL);
3637 ret = check_ino_in_path(root, ino1, ino1_gen,
3638 parent, parent_gen, fs_path);
3646 btrfs_free_path(path);
3648 fs_path_free(fs_path);
3652 static int wait_for_parent_move(struct send_ctx *sctx,
3653 struct recorded_ref *parent_ref,
3654 const bool is_orphan)
3657 u64 ino = parent_ref->dir;
3658 u64 ino_gen = parent_ref->dir_gen;
3659 u64 parent_ino_before, parent_ino_after;
3660 struct fs_path *path_before = NULL;
3661 struct fs_path *path_after = NULL;
3664 path_after = fs_path_alloc();
3665 path_before = fs_path_alloc();
3666 if (!path_after || !path_before) {
3672 * Our current directory inode may not yet be renamed/moved because some
3673 * ancestor (immediate or not) has to be renamed/moved first. So find if
3674 * such ancestor exists and make sure our own rename/move happens after
3675 * that ancestor is processed to avoid path build infinite loops (done
3676 * at get_cur_path()).
3678 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3679 u64 parent_ino_after_gen;
3681 if (is_waiting_for_move(sctx, ino)) {
3683 * If the current inode is an ancestor of ino in the
3684 * parent root, we need to delay the rename of the
3685 * current inode, otherwise don't delayed the rename
3686 * because we can end up with a circular dependency
3687 * of renames, resulting in some directories never
3688 * getting the respective rename operations issued in
3689 * the send stream or getting into infinite path build
3692 ret = is_ancestor(sctx->parent_root,
3693 sctx->cur_ino, sctx->cur_inode_gen,
3699 fs_path_reset(path_before);
3700 fs_path_reset(path_after);
3702 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3703 &parent_ino_after_gen, path_after);
3706 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3708 if (ret < 0 && ret != -ENOENT) {
3710 } else if (ret == -ENOENT) {
3715 len1 = fs_path_len(path_before);
3716 len2 = fs_path_len(path_after);
3717 if (ino > sctx->cur_ino &&
3718 (parent_ino_before != parent_ino_after || len1 != len2 ||
3719 memcmp(path_before->start, path_after->start, len1))) {
3722 ret = get_inode_info(sctx->parent_root, ino, NULL,
3723 &parent_ino_gen, NULL, NULL, NULL,
3727 if (ino_gen == parent_ino_gen) {
3732 ino = parent_ino_after;
3733 ino_gen = parent_ino_after_gen;
3737 fs_path_free(path_before);
3738 fs_path_free(path_after);
3741 ret = add_pending_dir_move(sctx,
3743 sctx->cur_inode_gen,
3746 &sctx->deleted_refs,
3755 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3758 struct fs_path *new_path;
3761 * Our reference's name member points to its full_path member string, so
3762 * we use here a new path.
3764 new_path = fs_path_alloc();
3768 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3770 fs_path_free(new_path);
3773 ret = fs_path_add(new_path, ref->name, ref->name_len);
3775 fs_path_free(new_path);
3779 fs_path_free(ref->full_path);
3780 set_ref_path(ref, new_path);
3786 * This does all the move/link/unlink/rmdir magic.
3788 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3790 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3792 struct recorded_ref *cur;
3793 struct recorded_ref *cur2;
3794 struct list_head check_dirs;
3795 struct fs_path *valid_path = NULL;
3799 int did_overwrite = 0;
3801 u64 last_dir_ino_rm = 0;
3802 bool can_rename = true;
3803 bool orphanized_dir = false;
3804 bool orphanized_ancestor = false;
3806 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3809 * This should never happen as the root dir always has the same ref
3810 * which is always '..'
3812 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3813 INIT_LIST_HEAD(&check_dirs);
3815 valid_path = fs_path_alloc();
3822 * First, check if the first ref of the current inode was overwritten
3823 * before. If yes, we know that the current inode was already orphanized
3824 * and thus use the orphan name. If not, we can use get_cur_path to
3825 * get the path of the first ref as it would like while receiving at
3826 * this point in time.
3827 * New inodes are always orphan at the beginning, so force to use the
3828 * orphan name in this case.
3829 * The first ref is stored in valid_path and will be updated if it
3830 * gets moved around.
3832 if (!sctx->cur_inode_new) {
3833 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3834 sctx->cur_inode_gen);
3840 if (sctx->cur_inode_new || did_overwrite) {
3841 ret = gen_unique_name(sctx, sctx->cur_ino,
3842 sctx->cur_inode_gen, valid_path);
3847 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3853 list_for_each_entry(cur, &sctx->new_refs, list) {
3855 * We may have refs where the parent directory does not exist
3856 * yet. This happens if the parent directories inum is higher
3857 * the the current inum. To handle this case, we create the
3858 * parent directory out of order. But we need to check if this
3859 * did already happen before due to other refs in the same dir.
3861 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3864 if (ret == inode_state_will_create) {
3867 * First check if any of the current inodes refs did
3868 * already create the dir.
3870 list_for_each_entry(cur2, &sctx->new_refs, list) {
3873 if (cur2->dir == cur->dir) {
3880 * If that did not happen, check if a previous inode
3881 * did already create the dir.
3884 ret = did_create_dir(sctx, cur->dir);
3888 ret = send_create_inode(sctx, cur->dir);
3895 * Check if this new ref would overwrite the first ref of
3896 * another unprocessed inode. If yes, orphanize the
3897 * overwritten inode. If we find an overwritten ref that is
3898 * not the first ref, simply unlink it.
3900 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3901 cur->name, cur->name_len,
3902 &ow_inode, &ow_gen, &ow_mode);
3906 ret = is_first_ref(sctx->parent_root,
3907 ow_inode, cur->dir, cur->name,
3912 struct name_cache_entry *nce;
3913 struct waiting_dir_move *wdm;
3915 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3919 if (S_ISDIR(ow_mode))
3920 orphanized_dir = true;
3923 * If ow_inode has its rename operation delayed
3924 * make sure that its orphanized name is used in
3925 * the source path when performing its rename
3928 if (is_waiting_for_move(sctx, ow_inode)) {
3929 wdm = get_waiting_dir_move(sctx,
3932 wdm->orphanized = true;
3936 * Make sure we clear our orphanized inode's
3937 * name from the name cache. This is because the
3938 * inode ow_inode might be an ancestor of some
3939 * other inode that will be orphanized as well
3940 * later and has an inode number greater than
3941 * sctx->send_progress. We need to prevent
3942 * future name lookups from using the old name
3943 * and get instead the orphan name.
3945 nce = name_cache_search(sctx, ow_inode, ow_gen);
3947 name_cache_delete(sctx, nce);
3952 * ow_inode might currently be an ancestor of
3953 * cur_ino, therefore compute valid_path (the
3954 * current path of cur_ino) again because it
3955 * might contain the pre-orphanization name of
3956 * ow_inode, which is no longer valid.
3958 ret = is_ancestor(sctx->parent_root,
3960 sctx->cur_ino, NULL);
3962 orphanized_ancestor = true;
3963 fs_path_reset(valid_path);
3964 ret = get_cur_path(sctx, sctx->cur_ino,
3965 sctx->cur_inode_gen,
3971 ret = send_unlink(sctx, cur->full_path);
3977 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3978 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3987 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3989 ret = wait_for_parent_move(sctx, cur, is_orphan);
3999 * link/move the ref to the new place. If we have an orphan
4000 * inode, move it and update valid_path. If not, link or move
4001 * it depending on the inode mode.
4003 if (is_orphan && can_rename) {
4004 ret = send_rename(sctx, valid_path, cur->full_path);
4008 ret = fs_path_copy(valid_path, cur->full_path);
4011 } else if (can_rename) {
4012 if (S_ISDIR(sctx->cur_inode_mode)) {
4014 * Dirs can't be linked, so move it. For moved
4015 * dirs, we always have one new and one deleted
4016 * ref. The deleted ref is ignored later.
4018 ret = send_rename(sctx, valid_path,
4021 ret = fs_path_copy(valid_path,
4027 * We might have previously orphanized an inode
4028 * which is an ancestor of our current inode,
4029 * so our reference's full path, which was
4030 * computed before any such orphanizations, must
4033 if (orphanized_dir) {
4034 ret = update_ref_path(sctx, cur);
4038 ret = send_link(sctx, cur->full_path,
4044 ret = dup_ref(cur, &check_dirs);
4049 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4051 * Check if we can already rmdir the directory. If not,
4052 * orphanize it. For every dir item inside that gets deleted
4053 * later, we do this check again and rmdir it then if possible.
4054 * See the use of check_dirs for more details.
4056 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4061 ret = send_rmdir(sctx, valid_path);
4064 } else if (!is_orphan) {
4065 ret = orphanize_inode(sctx, sctx->cur_ino,
4066 sctx->cur_inode_gen, valid_path);
4072 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4073 ret = dup_ref(cur, &check_dirs);
4077 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4078 !list_empty(&sctx->deleted_refs)) {
4080 * We have a moved dir. Add the old parent to check_dirs
4082 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4084 ret = dup_ref(cur, &check_dirs);
4087 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4089 * We have a non dir inode. Go through all deleted refs and
4090 * unlink them if they were not already overwritten by other
4093 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4094 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4095 sctx->cur_ino, sctx->cur_inode_gen,
4096 cur->name, cur->name_len);
4101 * If we orphanized any ancestor before, we need
4102 * to recompute the full path for deleted names,
4103 * since any such path was computed before we
4104 * processed any references and orphanized any
4107 if (orphanized_ancestor) {
4108 ret = update_ref_path(sctx, cur);
4112 ret = send_unlink(sctx, cur->full_path);
4116 ret = dup_ref(cur, &check_dirs);
4121 * If the inode is still orphan, unlink the orphan. This may
4122 * happen when a previous inode did overwrite the first ref
4123 * of this inode and no new refs were added for the current
4124 * inode. Unlinking does not mean that the inode is deleted in
4125 * all cases. There may still be links to this inode in other
4129 ret = send_unlink(sctx, valid_path);
4136 * We did collect all parent dirs where cur_inode was once located. We
4137 * now go through all these dirs and check if they are pending for
4138 * deletion and if it's finally possible to perform the rmdir now.
4139 * We also update the inode stats of the parent dirs here.
4141 list_for_each_entry(cur, &check_dirs, list) {
4143 * In case we had refs into dirs that were not processed yet,
4144 * we don't need to do the utime and rmdir logic for these dirs.
4145 * The dir will be processed later.
4147 if (cur->dir > sctx->cur_ino)
4150 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4154 if (ret == inode_state_did_create ||
4155 ret == inode_state_no_change) {
4156 /* TODO delayed utimes */
4157 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4160 } else if (ret == inode_state_did_delete &&
4161 cur->dir != last_dir_ino_rm) {
4162 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4167 ret = get_cur_path(sctx, cur->dir,
4168 cur->dir_gen, valid_path);
4171 ret = send_rmdir(sctx, valid_path);
4174 last_dir_ino_rm = cur->dir;
4182 __free_recorded_refs(&check_dirs);
4183 free_recorded_refs(sctx);
4184 fs_path_free(valid_path);
4188 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4189 void *ctx, struct list_head *refs)
4192 struct send_ctx *sctx = ctx;
4196 p = fs_path_alloc();
4200 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4205 ret = get_cur_path(sctx, dir, gen, p);
4208 ret = fs_path_add_path(p, name);
4212 ret = __record_ref(refs, dir, gen, p);
4220 static int __record_new_ref(int num, u64 dir, int index,
4221 struct fs_path *name,
4224 struct send_ctx *sctx = ctx;
4225 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4229 static int __record_deleted_ref(int num, u64 dir, int index,
4230 struct fs_path *name,
4233 struct send_ctx *sctx = ctx;
4234 return record_ref(sctx->parent_root, dir, name, ctx,
4235 &sctx->deleted_refs);
4238 static int record_new_ref(struct send_ctx *sctx)
4242 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4243 sctx->cmp_key, 0, __record_new_ref, sctx);
4252 static int record_deleted_ref(struct send_ctx *sctx)
4256 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4257 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4266 struct find_ref_ctx {
4269 struct btrfs_root *root;
4270 struct fs_path *name;
4274 static int __find_iref(int num, u64 dir, int index,
4275 struct fs_path *name,
4278 struct find_ref_ctx *ctx = ctx_;
4282 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4283 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4285 * To avoid doing extra lookups we'll only do this if everything
4288 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4292 if (dir_gen != ctx->dir_gen)
4294 ctx->found_idx = num;
4300 static int find_iref(struct btrfs_root *root,
4301 struct btrfs_path *path,
4302 struct btrfs_key *key,
4303 u64 dir, u64 dir_gen, struct fs_path *name)
4306 struct find_ref_ctx ctx;
4310 ctx.dir_gen = dir_gen;
4314 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4318 if (ctx.found_idx == -1)
4321 return ctx.found_idx;
4324 static int __record_changed_new_ref(int num, u64 dir, int index,
4325 struct fs_path *name,
4330 struct send_ctx *sctx = ctx;
4332 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4337 ret = find_iref(sctx->parent_root, sctx->right_path,
4338 sctx->cmp_key, dir, dir_gen, name);
4340 ret = __record_new_ref(num, dir, index, name, sctx);
4347 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4348 struct fs_path *name,
4353 struct send_ctx *sctx = ctx;
4355 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4360 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4361 dir, dir_gen, name);
4363 ret = __record_deleted_ref(num, dir, index, name, sctx);
4370 static int record_changed_ref(struct send_ctx *sctx)
4374 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4375 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4378 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4379 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4389 * Record and process all refs at once. Needed when an inode changes the
4390 * generation number, which means that it was deleted and recreated.
4392 static int process_all_refs(struct send_ctx *sctx,
4393 enum btrfs_compare_tree_result cmd)
4396 struct btrfs_root *root;
4397 struct btrfs_path *path;
4398 struct btrfs_key key;
4399 struct btrfs_key found_key;
4400 struct extent_buffer *eb;
4402 iterate_inode_ref_t cb;
4403 int pending_move = 0;
4405 path = alloc_path_for_send();
4409 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4410 root = sctx->send_root;
4411 cb = __record_new_ref;
4412 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4413 root = sctx->parent_root;
4414 cb = __record_deleted_ref;
4416 btrfs_err(sctx->send_root->fs_info,
4417 "Wrong command %d in process_all_refs", cmd);
4422 key.objectid = sctx->cmp_key->objectid;
4423 key.type = BTRFS_INODE_REF_KEY;
4425 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4430 eb = path->nodes[0];
4431 slot = path->slots[0];
4432 if (slot >= btrfs_header_nritems(eb)) {
4433 ret = btrfs_next_leaf(root, path);
4441 btrfs_item_key_to_cpu(eb, &found_key, slot);
4443 if (found_key.objectid != key.objectid ||
4444 (found_key.type != BTRFS_INODE_REF_KEY &&
4445 found_key.type != BTRFS_INODE_EXTREF_KEY))
4448 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4454 btrfs_release_path(path);
4457 * We don't actually care about pending_move as we are simply
4458 * re-creating this inode and will be rename'ing it into place once we
4459 * rename the parent directory.
4461 ret = process_recorded_refs(sctx, &pending_move);
4463 btrfs_free_path(path);
4467 static int send_set_xattr(struct send_ctx *sctx,
4468 struct fs_path *path,
4469 const char *name, int name_len,
4470 const char *data, int data_len)
4474 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4478 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4479 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4480 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4482 ret = send_cmd(sctx);
4489 static int send_remove_xattr(struct send_ctx *sctx,
4490 struct fs_path *path,
4491 const char *name, int name_len)
4495 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4499 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4500 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4502 ret = send_cmd(sctx);
4509 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4510 const char *name, int name_len,
4511 const char *data, int data_len,
4515 struct send_ctx *sctx = ctx;
4517 struct posix_acl_xattr_header dummy_acl;
4519 p = fs_path_alloc();
4524 * This hack is needed because empty acls are stored as zero byte
4525 * data in xattrs. Problem with that is, that receiving these zero byte
4526 * acls will fail later. To fix this, we send a dummy acl list that
4527 * only contains the version number and no entries.
4529 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4530 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4531 if (data_len == 0) {
4532 dummy_acl.a_version =
4533 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4534 data = (char *)&dummy_acl;
4535 data_len = sizeof(dummy_acl);
4539 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4543 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4550 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4551 const char *name, int name_len,
4552 const char *data, int data_len,
4556 struct send_ctx *sctx = ctx;
4559 p = fs_path_alloc();
4563 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4567 ret = send_remove_xattr(sctx, p, name, name_len);
4574 static int process_new_xattr(struct send_ctx *sctx)
4578 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4579 __process_new_xattr, sctx);
4584 static int process_deleted_xattr(struct send_ctx *sctx)
4586 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4587 __process_deleted_xattr, sctx);
4590 struct find_xattr_ctx {
4598 static int __find_xattr(int num, struct btrfs_key *di_key,
4599 const char *name, int name_len,
4600 const char *data, int data_len,
4601 u8 type, void *vctx)
4603 struct find_xattr_ctx *ctx = vctx;
4605 if (name_len == ctx->name_len &&
4606 strncmp(name, ctx->name, name_len) == 0) {
4607 ctx->found_idx = num;
4608 ctx->found_data_len = data_len;
4609 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4610 if (!ctx->found_data)
4617 static int find_xattr(struct btrfs_root *root,
4618 struct btrfs_path *path,
4619 struct btrfs_key *key,
4620 const char *name, int name_len,
4621 char **data, int *data_len)
4624 struct find_xattr_ctx ctx;
4627 ctx.name_len = name_len;
4629 ctx.found_data = NULL;
4630 ctx.found_data_len = 0;
4632 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4636 if (ctx.found_idx == -1)
4639 *data = ctx.found_data;
4640 *data_len = ctx.found_data_len;
4642 kfree(ctx.found_data);
4644 return ctx.found_idx;
4648 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4649 const char *name, int name_len,
4650 const char *data, int data_len,
4654 struct send_ctx *sctx = ctx;
4655 char *found_data = NULL;
4656 int found_data_len = 0;
4658 ret = find_xattr(sctx->parent_root, sctx->right_path,
4659 sctx->cmp_key, name, name_len, &found_data,
4661 if (ret == -ENOENT) {
4662 ret = __process_new_xattr(num, di_key, name, name_len, data,
4663 data_len, type, ctx);
4664 } else if (ret >= 0) {
4665 if (data_len != found_data_len ||
4666 memcmp(data, found_data, data_len)) {
4667 ret = __process_new_xattr(num, di_key, name, name_len,
4668 data, data_len, type, ctx);
4678 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4679 const char *name, int name_len,
4680 const char *data, int data_len,
4684 struct send_ctx *sctx = ctx;
4686 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4687 name, name_len, NULL, NULL);
4689 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4690 data_len, type, ctx);
4697 static int process_changed_xattr(struct send_ctx *sctx)
4701 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4702 __process_changed_new_xattr, sctx);
4705 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4706 __process_changed_deleted_xattr, sctx);
4712 static int process_all_new_xattrs(struct send_ctx *sctx)
4715 struct btrfs_root *root;
4716 struct btrfs_path *path;
4717 struct btrfs_key key;
4718 struct btrfs_key found_key;
4719 struct extent_buffer *eb;
4722 path = alloc_path_for_send();
4726 root = sctx->send_root;
4728 key.objectid = sctx->cmp_key->objectid;
4729 key.type = BTRFS_XATTR_ITEM_KEY;
4731 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4736 eb = path->nodes[0];
4737 slot = path->slots[0];
4738 if (slot >= btrfs_header_nritems(eb)) {
4739 ret = btrfs_next_leaf(root, path);
4742 } else if (ret > 0) {
4749 btrfs_item_key_to_cpu(eb, &found_key, slot);
4750 if (found_key.objectid != key.objectid ||
4751 found_key.type != key.type) {
4756 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4764 btrfs_free_path(path);
4768 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4770 struct btrfs_root *root = sctx->send_root;
4771 struct btrfs_fs_info *fs_info = root->fs_info;
4772 struct inode *inode;
4775 struct btrfs_key key;
4776 pgoff_t index = offset >> PAGE_SHIFT;
4778 unsigned pg_offset = offset & ~PAGE_MASK;
4781 key.objectid = sctx->cur_ino;
4782 key.type = BTRFS_INODE_ITEM_KEY;
4785 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4787 return PTR_ERR(inode);
4789 if (offset + len > i_size_read(inode)) {
4790 if (offset > i_size_read(inode))
4793 len = offset - i_size_read(inode);
4798 last_index = (offset + len - 1) >> PAGE_SHIFT;
4800 /* initial readahead */
4801 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4802 file_ra_state_init(&sctx->ra, inode->i_mapping);
4804 while (index <= last_index) {
4805 unsigned cur_len = min_t(unsigned, len,
4806 PAGE_SIZE - pg_offset);
4808 page = find_lock_page(inode->i_mapping, index);
4810 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4811 NULL, index, last_index + 1 - index);
4813 page = find_or_create_page(inode->i_mapping, index,
4821 if (PageReadahead(page)) {
4822 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4823 NULL, page, index, last_index + 1 - index);
4826 if (!PageUptodate(page)) {
4827 btrfs_readpage(NULL, page);
4829 if (!PageUptodate(page)) {
4838 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4853 * Read some bytes from the current inode/file and send a write command to
4856 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4858 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4861 ssize_t num_read = 0;
4863 p = fs_path_alloc();
4867 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4869 num_read = fill_read_buf(sctx, offset, len);
4870 if (num_read <= 0) {
4876 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4880 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4884 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4885 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4886 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4888 ret = send_cmd(sctx);
4899 * Send a clone command to user space.
4901 static int send_clone(struct send_ctx *sctx,
4902 u64 offset, u32 len,
4903 struct clone_root *clone_root)
4909 btrfs_debug(sctx->send_root->fs_info,
4910 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4911 offset, len, clone_root->root->root_key.objectid,
4912 clone_root->ino, clone_root->offset);
4914 p = fs_path_alloc();
4918 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4922 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4926 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4927 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4928 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4930 if (clone_root->root == sctx->send_root) {
4931 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4932 &gen, NULL, NULL, NULL, NULL);
4935 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4937 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4943 * If the parent we're using has a received_uuid set then use that as
4944 * our clone source as that is what we will look for when doing a
4947 * This covers the case that we create a snapshot off of a received
4948 * subvolume and then use that as the parent and try to receive on a
4951 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4952 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4953 clone_root->root->root_item.received_uuid);
4955 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4956 clone_root->root->root_item.uuid);
4957 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4958 le64_to_cpu(clone_root->root->root_item.ctransid));
4959 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4960 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4961 clone_root->offset);
4963 ret = send_cmd(sctx);
4972 * Send an update extent command to user space.
4974 static int send_update_extent(struct send_ctx *sctx,
4975 u64 offset, u32 len)
4980 p = fs_path_alloc();
4984 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4988 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4992 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4993 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4994 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4996 ret = send_cmd(sctx);
5004 static int send_hole(struct send_ctx *sctx, u64 end)
5006 struct fs_path *p = NULL;
5007 u64 offset = sctx->cur_inode_last_extent;
5012 * A hole that starts at EOF or beyond it. Since we do not yet support
5013 * fallocate (for extent preallocation and hole punching), sending a
5014 * write of zeroes starting at EOF or beyond would later require issuing
5015 * a truncate operation which would undo the write and achieve nothing.
5017 if (offset >= sctx->cur_inode_size)
5020 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5021 return send_update_extent(sctx, offset, end - offset);
5023 p = fs_path_alloc();
5026 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5028 goto tlv_put_failure;
5029 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5030 while (offset < end) {
5031 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5033 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5036 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5037 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5038 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5039 ret = send_cmd(sctx);
5044 sctx->cur_inode_next_write_offset = offset;
5050 static int send_extent_data(struct send_ctx *sctx,
5056 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5057 return send_update_extent(sctx, offset, len);
5059 while (sent < len) {
5060 u64 size = len - sent;
5063 if (size > BTRFS_SEND_READ_SIZE)
5064 size = BTRFS_SEND_READ_SIZE;
5065 ret = send_write(sctx, offset + sent, size);
5075 static int clone_range(struct send_ctx *sctx,
5076 struct clone_root *clone_root,
5077 const u64 disk_byte,
5082 struct btrfs_path *path;
5083 struct btrfs_key key;
5087 * Prevent cloning from a zero offset with a length matching the sector
5088 * size because in some scenarios this will make the receiver fail.
5090 * For example, if in the source filesystem the extent at offset 0
5091 * has a length of sectorsize and it was written using direct IO, then
5092 * it can never be an inline extent (even if compression is enabled).
5093 * Then this extent can be cloned in the original filesystem to a non
5094 * zero file offset, but it may not be possible to clone in the
5095 * destination filesystem because it can be inlined due to compression
5096 * on the destination filesystem (as the receiver's write operations are
5097 * always done using buffered IO). The same happens when the original
5098 * filesystem does not have compression enabled but the destination
5101 if (clone_root->offset == 0 &&
5102 len == sctx->send_root->fs_info->sectorsize)
5103 return send_extent_data(sctx, offset, len);
5105 path = alloc_path_for_send();
5110 * We can't send a clone operation for the entire range if we find
5111 * extent items in the respective range in the source file that
5112 * refer to different extents or if we find holes.
5113 * So check for that and do a mix of clone and regular write/copy
5114 * operations if needed.
5118 * mkfs.btrfs -f /dev/sda
5119 * mount /dev/sda /mnt
5120 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5121 * cp --reflink=always /mnt/foo /mnt/bar
5122 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5123 * btrfs subvolume snapshot -r /mnt /mnt/snap
5125 * If when we send the snapshot and we are processing file bar (which
5126 * has a higher inode number than foo) we blindly send a clone operation
5127 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5128 * a file bar that matches the content of file foo - iow, doesn't match
5129 * the content from bar in the original filesystem.
5131 key.objectid = clone_root->ino;
5132 key.type = BTRFS_EXTENT_DATA_KEY;
5133 key.offset = clone_root->offset;
5134 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5137 if (ret > 0 && path->slots[0] > 0) {
5138 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5139 if (key.objectid == clone_root->ino &&
5140 key.type == BTRFS_EXTENT_DATA_KEY)
5145 struct extent_buffer *leaf = path->nodes[0];
5146 int slot = path->slots[0];
5147 struct btrfs_file_extent_item *ei;
5152 if (slot >= btrfs_header_nritems(leaf)) {
5153 ret = btrfs_next_leaf(clone_root->root, path);
5161 btrfs_item_key_to_cpu(leaf, &key, slot);
5164 * We might have an implicit trailing hole (NO_HOLES feature
5165 * enabled). We deal with it after leaving this loop.
5167 if (key.objectid != clone_root->ino ||
5168 key.type != BTRFS_EXTENT_DATA_KEY)
5171 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5172 type = btrfs_file_extent_type(leaf, ei);
5173 if (type == BTRFS_FILE_EXTENT_INLINE) {
5174 ext_len = btrfs_file_extent_ram_bytes(leaf, ei);
5175 ext_len = PAGE_ALIGN(ext_len);
5177 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
5180 if (key.offset + ext_len <= clone_root->offset)
5183 if (key.offset > clone_root->offset) {
5184 /* Implicit hole, NO_HOLES feature enabled. */
5185 u64 hole_len = key.offset - clone_root->offset;
5189 ret = send_extent_data(sctx, offset, hole_len);
5197 clone_root->offset += hole_len;
5198 data_offset += hole_len;
5201 if (key.offset >= clone_root->offset + len)
5204 clone_len = min_t(u64, ext_len, len);
5206 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5207 btrfs_file_extent_offset(leaf, ei) == data_offset)
5208 ret = send_clone(sctx, offset, clone_len, clone_root);
5210 ret = send_extent_data(sctx, offset, clone_len);
5218 offset += clone_len;
5219 clone_root->offset += clone_len;
5220 data_offset += clone_len;
5226 ret = send_extent_data(sctx, offset, len);
5230 btrfs_free_path(path);
5234 static int send_write_or_clone(struct send_ctx *sctx,
5235 struct btrfs_path *path,
5236 struct btrfs_key *key,
5237 struct clone_root *clone_root)
5240 struct btrfs_file_extent_item *ei;
5241 u64 offset = key->offset;
5244 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5246 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5247 struct btrfs_file_extent_item);
5248 type = btrfs_file_extent_type(path->nodes[0], ei);
5249 if (type == BTRFS_FILE_EXTENT_INLINE) {
5250 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5252 * it is possible the inline item won't cover the whole page,
5253 * but there may be items after this page. Make
5254 * sure to send the whole thing
5256 len = PAGE_ALIGN(len);
5258 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5261 if (offset >= sctx->cur_inode_size) {
5265 if (offset + len > sctx->cur_inode_size)
5266 len = sctx->cur_inode_size - offset;
5272 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5276 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5277 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5278 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5281 ret = send_extent_data(sctx, offset, len);
5283 sctx->cur_inode_next_write_offset = offset + len;
5288 static int is_extent_unchanged(struct send_ctx *sctx,
5289 struct btrfs_path *left_path,
5290 struct btrfs_key *ekey)
5293 struct btrfs_key key;
5294 struct btrfs_path *path = NULL;
5295 struct extent_buffer *eb;
5297 struct btrfs_key found_key;
5298 struct btrfs_file_extent_item *ei;
5303 u64 left_offset_fixed;
5311 path = alloc_path_for_send();
5315 eb = left_path->nodes[0];
5316 slot = left_path->slots[0];
5317 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5318 left_type = btrfs_file_extent_type(eb, ei);
5320 if (left_type != BTRFS_FILE_EXTENT_REG) {
5324 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5325 left_len = btrfs_file_extent_num_bytes(eb, ei);
5326 left_offset = btrfs_file_extent_offset(eb, ei);
5327 left_gen = btrfs_file_extent_generation(eb, ei);
5330 * Following comments will refer to these graphics. L is the left
5331 * extents which we are checking at the moment. 1-8 are the right
5332 * extents that we iterate.
5335 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5338 * |--1--|-2b-|...(same as above)
5340 * Alternative situation. Happens on files where extents got split.
5342 * |-----------7-----------|-6-|
5344 * Alternative situation. Happens on files which got larger.
5347 * Nothing follows after 8.
5350 key.objectid = ekey->objectid;
5351 key.type = BTRFS_EXTENT_DATA_KEY;
5352 key.offset = ekey->offset;
5353 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5362 * Handle special case where the right side has no extents at all.
5364 eb = path->nodes[0];
5365 slot = path->slots[0];
5366 btrfs_item_key_to_cpu(eb, &found_key, slot);
5367 if (found_key.objectid != key.objectid ||
5368 found_key.type != key.type) {
5369 /* If we're a hole then just pretend nothing changed */
5370 ret = (left_disknr) ? 0 : 1;
5375 * We're now on 2a, 2b or 7.
5378 while (key.offset < ekey->offset + left_len) {
5379 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5380 right_type = btrfs_file_extent_type(eb, ei);
5381 if (right_type != BTRFS_FILE_EXTENT_REG &&
5382 right_type != BTRFS_FILE_EXTENT_INLINE) {
5387 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5388 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5389 right_len = PAGE_ALIGN(right_len);
5391 right_len = btrfs_file_extent_num_bytes(eb, ei);
5395 * Are we at extent 8? If yes, we know the extent is changed.
5396 * This may only happen on the first iteration.
5398 if (found_key.offset + right_len <= ekey->offset) {
5399 /* If we're a hole just pretend nothing changed */
5400 ret = (left_disknr) ? 0 : 1;
5405 * We just wanted to see if when we have an inline extent, what
5406 * follows it is a regular extent (wanted to check the above
5407 * condition for inline extents too). This should normally not
5408 * happen but it's possible for example when we have an inline
5409 * compressed extent representing data with a size matching
5410 * the page size (currently the same as sector size).
5412 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5417 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5418 right_offset = btrfs_file_extent_offset(eb, ei);
5419 right_gen = btrfs_file_extent_generation(eb, ei);
5421 left_offset_fixed = left_offset;
5422 if (key.offset < ekey->offset) {
5423 /* Fix the right offset for 2a and 7. */
5424 right_offset += ekey->offset - key.offset;
5426 /* Fix the left offset for all behind 2a and 2b */
5427 left_offset_fixed += key.offset - ekey->offset;
5431 * Check if we have the same extent.
5433 if (left_disknr != right_disknr ||
5434 left_offset_fixed != right_offset ||
5435 left_gen != right_gen) {
5441 * Go to the next extent.
5443 ret = btrfs_next_item(sctx->parent_root, path);
5447 eb = path->nodes[0];
5448 slot = path->slots[0];
5449 btrfs_item_key_to_cpu(eb, &found_key, slot);
5451 if (ret || found_key.objectid != key.objectid ||
5452 found_key.type != key.type) {
5453 key.offset += right_len;
5456 if (found_key.offset != key.offset + right_len) {
5464 * We're now behind the left extent (treat as unchanged) or at the end
5465 * of the right side (treat as changed).
5467 if (key.offset >= ekey->offset + left_len)
5474 btrfs_free_path(path);
5478 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5480 struct btrfs_path *path;
5481 struct btrfs_root *root = sctx->send_root;
5482 struct btrfs_file_extent_item *fi;
5483 struct btrfs_key key;
5488 path = alloc_path_for_send();
5492 sctx->cur_inode_last_extent = 0;
5494 key.objectid = sctx->cur_ino;
5495 key.type = BTRFS_EXTENT_DATA_KEY;
5496 key.offset = offset;
5497 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5501 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5502 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5505 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5506 struct btrfs_file_extent_item);
5507 type = btrfs_file_extent_type(path->nodes[0], fi);
5508 if (type == BTRFS_FILE_EXTENT_INLINE) {
5509 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5510 extent_end = ALIGN(key.offset + size,
5511 sctx->send_root->fs_info->sectorsize);
5513 extent_end = key.offset +
5514 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5516 sctx->cur_inode_last_extent = extent_end;
5518 btrfs_free_path(path);
5522 static int range_is_hole_in_parent(struct send_ctx *sctx,
5526 struct btrfs_path *path;
5527 struct btrfs_key key;
5528 struct btrfs_root *root = sctx->parent_root;
5529 u64 search_start = start;
5532 path = alloc_path_for_send();
5536 key.objectid = sctx->cur_ino;
5537 key.type = BTRFS_EXTENT_DATA_KEY;
5538 key.offset = search_start;
5539 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5542 if (ret > 0 && path->slots[0] > 0)
5545 while (search_start < end) {
5546 struct extent_buffer *leaf = path->nodes[0];
5547 int slot = path->slots[0];
5548 struct btrfs_file_extent_item *fi;
5551 if (slot >= btrfs_header_nritems(leaf)) {
5552 ret = btrfs_next_leaf(root, path);
5560 btrfs_item_key_to_cpu(leaf, &key, slot);
5561 if (key.objectid < sctx->cur_ino ||
5562 key.type < BTRFS_EXTENT_DATA_KEY)
5564 if (key.objectid > sctx->cur_ino ||
5565 key.type > BTRFS_EXTENT_DATA_KEY ||
5569 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5570 if (btrfs_file_extent_type(leaf, fi) ==
5571 BTRFS_FILE_EXTENT_INLINE) {
5572 u64 size = btrfs_file_extent_ram_bytes(leaf, fi);
5574 extent_end = ALIGN(key.offset + size,
5575 root->fs_info->sectorsize);
5577 extent_end = key.offset +
5578 btrfs_file_extent_num_bytes(leaf, fi);
5580 if (extent_end <= start)
5582 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5583 search_start = extent_end;
5593 btrfs_free_path(path);
5597 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5598 struct btrfs_key *key)
5600 struct btrfs_file_extent_item *fi;
5605 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5608 if (sctx->cur_inode_last_extent == (u64)-1) {
5609 ret = get_last_extent(sctx, key->offset - 1);
5614 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5615 struct btrfs_file_extent_item);
5616 type = btrfs_file_extent_type(path->nodes[0], fi);
5617 if (type == BTRFS_FILE_EXTENT_INLINE) {
5618 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5619 extent_end = ALIGN(key->offset + size,
5620 sctx->send_root->fs_info->sectorsize);
5622 extent_end = key->offset +
5623 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5626 if (path->slots[0] == 0 &&
5627 sctx->cur_inode_last_extent < key->offset) {
5629 * We might have skipped entire leafs that contained only
5630 * file extent items for our current inode. These leafs have
5631 * a generation number smaller (older) than the one in the
5632 * current leaf and the leaf our last extent came from, and
5633 * are located between these 2 leafs.
5635 ret = get_last_extent(sctx, key->offset - 1);
5640 if (sctx->cur_inode_last_extent < key->offset) {
5641 ret = range_is_hole_in_parent(sctx,
5642 sctx->cur_inode_last_extent,
5647 ret = send_hole(sctx, key->offset);
5651 sctx->cur_inode_last_extent = extent_end;
5655 static int process_extent(struct send_ctx *sctx,
5656 struct btrfs_path *path,
5657 struct btrfs_key *key)
5659 struct clone_root *found_clone = NULL;
5662 if (S_ISLNK(sctx->cur_inode_mode))
5665 if (sctx->parent_root && !sctx->cur_inode_new) {
5666 ret = is_extent_unchanged(sctx, path, key);
5674 struct btrfs_file_extent_item *ei;
5677 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5678 struct btrfs_file_extent_item);
5679 type = btrfs_file_extent_type(path->nodes[0], ei);
5680 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5681 type == BTRFS_FILE_EXTENT_REG) {
5683 * The send spec does not have a prealloc command yet,
5684 * so just leave a hole for prealloc'ed extents until
5685 * we have enough commands queued up to justify rev'ing
5688 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5693 /* Have a hole, just skip it. */
5694 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5701 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5702 sctx->cur_inode_size, &found_clone);
5703 if (ret != -ENOENT && ret < 0)
5706 ret = send_write_or_clone(sctx, path, key, found_clone);
5710 ret = maybe_send_hole(sctx, path, key);
5715 static int process_all_extents(struct send_ctx *sctx)
5718 struct btrfs_root *root;
5719 struct btrfs_path *path;
5720 struct btrfs_key key;
5721 struct btrfs_key found_key;
5722 struct extent_buffer *eb;
5725 root = sctx->send_root;
5726 path = alloc_path_for_send();
5730 key.objectid = sctx->cmp_key->objectid;
5731 key.type = BTRFS_EXTENT_DATA_KEY;
5733 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5738 eb = path->nodes[0];
5739 slot = path->slots[0];
5741 if (slot >= btrfs_header_nritems(eb)) {
5742 ret = btrfs_next_leaf(root, path);
5745 } else if (ret > 0) {
5752 btrfs_item_key_to_cpu(eb, &found_key, slot);
5754 if (found_key.objectid != key.objectid ||
5755 found_key.type != key.type) {
5760 ret = process_extent(sctx, path, &found_key);
5768 btrfs_free_path(path);
5772 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5774 int *refs_processed)
5778 if (sctx->cur_ino == 0)
5780 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5781 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5783 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5786 ret = process_recorded_refs(sctx, pending_move);
5790 *refs_processed = 1;
5795 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5806 int need_truncate = 1;
5807 int pending_move = 0;
5808 int refs_processed = 0;
5810 if (sctx->ignore_cur_inode)
5813 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5819 * We have processed the refs and thus need to advance send_progress.
5820 * Now, calls to get_cur_xxx will take the updated refs of the current
5821 * inode into account.
5823 * On the other hand, if our current inode is a directory and couldn't
5824 * be moved/renamed because its parent was renamed/moved too and it has
5825 * a higher inode number, we can only move/rename our current inode
5826 * after we moved/renamed its parent. Therefore in this case operate on
5827 * the old path (pre move/rename) of our current inode, and the
5828 * move/rename will be performed later.
5830 if (refs_processed && !pending_move)
5831 sctx->send_progress = sctx->cur_ino + 1;
5833 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5835 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5838 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5839 &left_mode, &left_uid, &left_gid, NULL);
5843 if (!sctx->parent_root || sctx->cur_inode_new) {
5845 if (!S_ISLNK(sctx->cur_inode_mode))
5847 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5852 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5853 &old_size, NULL, &right_mode, &right_uid,
5858 if (left_uid != right_uid || left_gid != right_gid)
5860 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5862 if ((old_size == sctx->cur_inode_size) ||
5863 (sctx->cur_inode_size > old_size &&
5864 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5868 if (S_ISREG(sctx->cur_inode_mode)) {
5869 if (need_send_hole(sctx)) {
5870 if (sctx->cur_inode_last_extent == (u64)-1 ||
5871 sctx->cur_inode_last_extent <
5872 sctx->cur_inode_size) {
5873 ret = get_last_extent(sctx, (u64)-1);
5877 if (sctx->cur_inode_last_extent <
5878 sctx->cur_inode_size) {
5879 ret = send_hole(sctx, sctx->cur_inode_size);
5884 if (need_truncate) {
5885 ret = send_truncate(sctx, sctx->cur_ino,
5886 sctx->cur_inode_gen,
5887 sctx->cur_inode_size);
5894 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5895 left_uid, left_gid);
5900 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5907 * If other directory inodes depended on our current directory
5908 * inode's move/rename, now do their move/rename operations.
5910 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5911 ret = apply_children_dir_moves(sctx);
5915 * Need to send that every time, no matter if it actually
5916 * changed between the two trees as we have done changes to
5917 * the inode before. If our inode is a directory and it's
5918 * waiting to be moved/renamed, we will send its utimes when
5919 * it's moved/renamed, therefore we don't need to do it here.
5921 sctx->send_progress = sctx->cur_ino + 1;
5922 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5931 struct parent_paths_ctx {
5932 struct list_head *refs;
5933 struct send_ctx *sctx;
5936 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
5939 struct parent_paths_ctx *ppctx = ctx;
5941 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
5946 * Issue unlink operations for all paths of the current inode found in the
5949 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
5951 LIST_HEAD(deleted_refs);
5952 struct btrfs_path *path;
5953 struct btrfs_key key;
5954 struct parent_paths_ctx ctx;
5957 path = alloc_path_for_send();
5961 key.objectid = sctx->cur_ino;
5962 key.type = BTRFS_INODE_REF_KEY;
5964 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
5968 ctx.refs = &deleted_refs;
5972 struct extent_buffer *eb = path->nodes[0];
5973 int slot = path->slots[0];
5975 if (slot >= btrfs_header_nritems(eb)) {
5976 ret = btrfs_next_leaf(sctx->parent_root, path);
5984 btrfs_item_key_to_cpu(eb, &key, slot);
5985 if (key.objectid != sctx->cur_ino)
5987 if (key.type != BTRFS_INODE_REF_KEY &&
5988 key.type != BTRFS_INODE_EXTREF_KEY)
5991 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
5992 record_parent_ref, &ctx);
5999 while (!list_empty(&deleted_refs)) {
6000 struct recorded_ref *ref;
6002 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6003 ret = send_unlink(sctx, ref->full_path);
6006 fs_path_free(ref->full_path);
6007 list_del(&ref->list);
6012 btrfs_free_path(path);
6014 __free_recorded_refs(&deleted_refs);
6018 static int changed_inode(struct send_ctx *sctx,
6019 enum btrfs_compare_tree_result result)
6022 struct btrfs_key *key = sctx->cmp_key;
6023 struct btrfs_inode_item *left_ii = NULL;
6024 struct btrfs_inode_item *right_ii = NULL;
6028 sctx->cur_ino = key->objectid;
6029 sctx->cur_inode_new_gen = 0;
6030 sctx->cur_inode_last_extent = (u64)-1;
6031 sctx->cur_inode_next_write_offset = 0;
6032 sctx->ignore_cur_inode = false;
6035 * Set send_progress to current inode. This will tell all get_cur_xxx
6036 * functions that the current inode's refs are not updated yet. Later,
6037 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6039 sctx->send_progress = sctx->cur_ino;
6041 if (result == BTRFS_COMPARE_TREE_NEW ||
6042 result == BTRFS_COMPARE_TREE_CHANGED) {
6043 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6044 sctx->left_path->slots[0],
6045 struct btrfs_inode_item);
6046 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6049 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6050 sctx->right_path->slots[0],
6051 struct btrfs_inode_item);
6052 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6055 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6056 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6057 sctx->right_path->slots[0],
6058 struct btrfs_inode_item);
6060 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6064 * The cur_ino = root dir case is special here. We can't treat
6065 * the inode as deleted+reused because it would generate a
6066 * stream that tries to delete/mkdir the root dir.
6068 if (left_gen != right_gen &&
6069 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6070 sctx->cur_inode_new_gen = 1;
6074 * Normally we do not find inodes with a link count of zero (orphans)
6075 * because the most common case is to create a snapshot and use it
6076 * for a send operation. However other less common use cases involve
6077 * using a subvolume and send it after turning it to RO mode just
6078 * after deleting all hard links of a file while holding an open
6079 * file descriptor against it or turning a RO snapshot into RW mode,
6080 * keep an open file descriptor against a file, delete it and then
6081 * turn the snapshot back to RO mode before using it for a send
6082 * operation. So if we find such cases, ignore the inode and all its
6083 * items completely if it's a new inode, or if it's a changed inode
6084 * make sure all its previous paths (from the parent snapshot) are all
6085 * unlinked and all other the inode items are ignored.
6087 if (result == BTRFS_COMPARE_TREE_NEW ||
6088 result == BTRFS_COMPARE_TREE_CHANGED) {
6091 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6093 sctx->ignore_cur_inode = true;
6094 if (result == BTRFS_COMPARE_TREE_CHANGED)
6095 ret = btrfs_unlink_all_paths(sctx);
6100 if (result == BTRFS_COMPARE_TREE_NEW) {
6101 sctx->cur_inode_gen = left_gen;
6102 sctx->cur_inode_new = 1;
6103 sctx->cur_inode_deleted = 0;
6104 sctx->cur_inode_size = btrfs_inode_size(
6105 sctx->left_path->nodes[0], left_ii);
6106 sctx->cur_inode_mode = btrfs_inode_mode(
6107 sctx->left_path->nodes[0], left_ii);
6108 sctx->cur_inode_rdev = btrfs_inode_rdev(
6109 sctx->left_path->nodes[0], left_ii);
6110 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6111 ret = send_create_inode_if_needed(sctx);
6112 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6113 sctx->cur_inode_gen = right_gen;
6114 sctx->cur_inode_new = 0;
6115 sctx->cur_inode_deleted = 1;
6116 sctx->cur_inode_size = btrfs_inode_size(
6117 sctx->right_path->nodes[0], right_ii);
6118 sctx->cur_inode_mode = btrfs_inode_mode(
6119 sctx->right_path->nodes[0], right_ii);
6120 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6122 * We need to do some special handling in case the inode was
6123 * reported as changed with a changed generation number. This
6124 * means that the original inode was deleted and new inode
6125 * reused the same inum. So we have to treat the old inode as
6126 * deleted and the new one as new.
6128 if (sctx->cur_inode_new_gen) {
6130 * First, process the inode as if it was deleted.
6132 sctx->cur_inode_gen = right_gen;
6133 sctx->cur_inode_new = 0;
6134 sctx->cur_inode_deleted = 1;
6135 sctx->cur_inode_size = btrfs_inode_size(
6136 sctx->right_path->nodes[0], right_ii);
6137 sctx->cur_inode_mode = btrfs_inode_mode(
6138 sctx->right_path->nodes[0], right_ii);
6139 ret = process_all_refs(sctx,
6140 BTRFS_COMPARE_TREE_DELETED);
6145 * Now process the inode as if it was new.
6147 sctx->cur_inode_gen = left_gen;
6148 sctx->cur_inode_new = 1;
6149 sctx->cur_inode_deleted = 0;
6150 sctx->cur_inode_size = btrfs_inode_size(
6151 sctx->left_path->nodes[0], left_ii);
6152 sctx->cur_inode_mode = btrfs_inode_mode(
6153 sctx->left_path->nodes[0], left_ii);
6154 sctx->cur_inode_rdev = btrfs_inode_rdev(
6155 sctx->left_path->nodes[0], left_ii);
6156 ret = send_create_inode_if_needed(sctx);
6160 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6164 * Advance send_progress now as we did not get into
6165 * process_recorded_refs_if_needed in the new_gen case.
6167 sctx->send_progress = sctx->cur_ino + 1;
6170 * Now process all extents and xattrs of the inode as if
6171 * they were all new.
6173 ret = process_all_extents(sctx);
6176 ret = process_all_new_xattrs(sctx);
6180 sctx->cur_inode_gen = left_gen;
6181 sctx->cur_inode_new = 0;
6182 sctx->cur_inode_new_gen = 0;
6183 sctx->cur_inode_deleted = 0;
6184 sctx->cur_inode_size = btrfs_inode_size(
6185 sctx->left_path->nodes[0], left_ii);
6186 sctx->cur_inode_mode = btrfs_inode_mode(
6187 sctx->left_path->nodes[0], left_ii);
6196 * We have to process new refs before deleted refs, but compare_trees gives us
6197 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6198 * first and later process them in process_recorded_refs.
6199 * For the cur_inode_new_gen case, we skip recording completely because
6200 * changed_inode did already initiate processing of refs. The reason for this is
6201 * that in this case, compare_tree actually compares the refs of 2 different
6202 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6203 * refs of the right tree as deleted and all refs of the left tree as new.
6205 static int changed_ref(struct send_ctx *sctx,
6206 enum btrfs_compare_tree_result result)
6210 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6211 inconsistent_snapshot_error(sctx, result, "reference");
6215 if (!sctx->cur_inode_new_gen &&
6216 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6217 if (result == BTRFS_COMPARE_TREE_NEW)
6218 ret = record_new_ref(sctx);
6219 else if (result == BTRFS_COMPARE_TREE_DELETED)
6220 ret = record_deleted_ref(sctx);
6221 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6222 ret = record_changed_ref(sctx);
6229 * Process new/deleted/changed xattrs. We skip processing in the
6230 * cur_inode_new_gen case because changed_inode did already initiate processing
6231 * of xattrs. The reason is the same as in changed_ref
6233 static int changed_xattr(struct send_ctx *sctx,
6234 enum btrfs_compare_tree_result result)
6238 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6239 inconsistent_snapshot_error(sctx, result, "xattr");
6243 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6244 if (result == BTRFS_COMPARE_TREE_NEW)
6245 ret = process_new_xattr(sctx);
6246 else if (result == BTRFS_COMPARE_TREE_DELETED)
6247 ret = process_deleted_xattr(sctx);
6248 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6249 ret = process_changed_xattr(sctx);
6256 * Process new/deleted/changed extents. We skip processing in the
6257 * cur_inode_new_gen case because changed_inode did already initiate processing
6258 * of extents. The reason is the same as in changed_ref
6260 static int changed_extent(struct send_ctx *sctx,
6261 enum btrfs_compare_tree_result result)
6265 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6267 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6268 struct extent_buffer *leaf_l;
6269 struct extent_buffer *leaf_r;
6270 struct btrfs_file_extent_item *ei_l;
6271 struct btrfs_file_extent_item *ei_r;
6273 leaf_l = sctx->left_path->nodes[0];
6274 leaf_r = sctx->right_path->nodes[0];
6275 ei_l = btrfs_item_ptr(leaf_l,
6276 sctx->left_path->slots[0],
6277 struct btrfs_file_extent_item);
6278 ei_r = btrfs_item_ptr(leaf_r,
6279 sctx->right_path->slots[0],
6280 struct btrfs_file_extent_item);
6283 * We may have found an extent item that has changed
6284 * only its disk_bytenr field and the corresponding
6285 * inode item was not updated. This case happens due to
6286 * very specific timings during relocation when a leaf
6287 * that contains file extent items is COWed while
6288 * relocation is ongoing and its in the stage where it
6289 * updates data pointers. So when this happens we can
6290 * safely ignore it since we know it's the same extent,
6291 * but just at different logical and physical locations
6292 * (when an extent is fully replaced with a new one, we
6293 * know the generation number must have changed too,
6294 * since snapshot creation implies committing the current
6295 * transaction, and the inode item must have been updated
6297 * This replacement of the disk_bytenr happens at
6298 * relocation.c:replace_file_extents() through
6299 * relocation.c:btrfs_reloc_cow_block().
6301 if (btrfs_file_extent_generation(leaf_l, ei_l) ==
6302 btrfs_file_extent_generation(leaf_r, ei_r) &&
6303 btrfs_file_extent_ram_bytes(leaf_l, ei_l) ==
6304 btrfs_file_extent_ram_bytes(leaf_r, ei_r) &&
6305 btrfs_file_extent_compression(leaf_l, ei_l) ==
6306 btrfs_file_extent_compression(leaf_r, ei_r) &&
6307 btrfs_file_extent_encryption(leaf_l, ei_l) ==
6308 btrfs_file_extent_encryption(leaf_r, ei_r) &&
6309 btrfs_file_extent_other_encoding(leaf_l, ei_l) ==
6310 btrfs_file_extent_other_encoding(leaf_r, ei_r) &&
6311 btrfs_file_extent_type(leaf_l, ei_l) ==
6312 btrfs_file_extent_type(leaf_r, ei_r) &&
6313 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) !=
6314 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) &&
6315 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) ==
6316 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) &&
6317 btrfs_file_extent_offset(leaf_l, ei_l) ==
6318 btrfs_file_extent_offset(leaf_r, ei_r) &&
6319 btrfs_file_extent_num_bytes(leaf_l, ei_l) ==
6320 btrfs_file_extent_num_bytes(leaf_r, ei_r))
6324 inconsistent_snapshot_error(sctx, result, "extent");
6328 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6329 if (result != BTRFS_COMPARE_TREE_DELETED)
6330 ret = process_extent(sctx, sctx->left_path,
6337 static int dir_changed(struct send_ctx *sctx, u64 dir)
6339 u64 orig_gen, new_gen;
6342 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6347 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6352 return (orig_gen != new_gen) ? 1 : 0;
6355 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6356 struct btrfs_key *key)
6358 struct btrfs_inode_extref *extref;
6359 struct extent_buffer *leaf;
6360 u64 dirid = 0, last_dirid = 0;
6367 /* Easy case, just check this one dirid */
6368 if (key->type == BTRFS_INODE_REF_KEY) {
6369 dirid = key->offset;
6371 ret = dir_changed(sctx, dirid);
6375 leaf = path->nodes[0];
6376 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6377 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6378 while (cur_offset < item_size) {
6379 extref = (struct btrfs_inode_extref *)(ptr +
6381 dirid = btrfs_inode_extref_parent(leaf, extref);
6382 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6383 cur_offset += ref_name_len + sizeof(*extref);
6384 if (dirid == last_dirid)
6386 ret = dir_changed(sctx, dirid);
6396 * Updates compare related fields in sctx and simply forwards to the actual
6397 * changed_xxx functions.
6399 static int changed_cb(struct btrfs_path *left_path,
6400 struct btrfs_path *right_path,
6401 struct btrfs_key *key,
6402 enum btrfs_compare_tree_result result,
6406 struct send_ctx *sctx = ctx;
6408 if (result == BTRFS_COMPARE_TREE_SAME) {
6409 if (key->type == BTRFS_INODE_REF_KEY ||
6410 key->type == BTRFS_INODE_EXTREF_KEY) {
6411 ret = compare_refs(sctx, left_path, key);
6416 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6417 return maybe_send_hole(sctx, left_path, key);
6421 result = BTRFS_COMPARE_TREE_CHANGED;
6425 sctx->left_path = left_path;
6426 sctx->right_path = right_path;
6427 sctx->cmp_key = key;
6429 ret = finish_inode_if_needed(sctx, 0);
6433 /* Ignore non-FS objects */
6434 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6435 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6438 if (key->type == BTRFS_INODE_ITEM_KEY) {
6439 ret = changed_inode(sctx, result);
6440 } else if (!sctx->ignore_cur_inode) {
6441 if (key->type == BTRFS_INODE_REF_KEY ||
6442 key->type == BTRFS_INODE_EXTREF_KEY)
6443 ret = changed_ref(sctx, result);
6444 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6445 ret = changed_xattr(sctx, result);
6446 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6447 ret = changed_extent(sctx, result);
6454 static int full_send_tree(struct send_ctx *sctx)
6457 struct btrfs_root *send_root = sctx->send_root;
6458 struct btrfs_key key;
6459 struct btrfs_path *path;
6460 struct extent_buffer *eb;
6463 path = alloc_path_for_send();
6467 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6468 key.type = BTRFS_INODE_ITEM_KEY;
6471 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6478 eb = path->nodes[0];
6479 slot = path->slots[0];
6480 btrfs_item_key_to_cpu(eb, &key, slot);
6482 ret = changed_cb(path, NULL, &key,
6483 BTRFS_COMPARE_TREE_NEW, sctx);
6487 ret = btrfs_next_item(send_root, path);
6497 ret = finish_inode_if_needed(sctx, 1);
6500 btrfs_free_path(path);
6504 static int send_subvol(struct send_ctx *sctx)
6508 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6509 ret = send_header(sctx);
6514 ret = send_subvol_begin(sctx);
6518 if (sctx->parent_root) {
6519 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6523 ret = finish_inode_if_needed(sctx, 1);
6527 ret = full_send_tree(sctx);
6533 free_recorded_refs(sctx);
6538 * If orphan cleanup did remove any orphans from a root, it means the tree
6539 * was modified and therefore the commit root is not the same as the current
6540 * root anymore. This is a problem, because send uses the commit root and
6541 * therefore can see inode items that don't exist in the current root anymore,
6542 * and for example make calls to btrfs_iget, which will do tree lookups based
6543 * on the current root and not on the commit root. Those lookups will fail,
6544 * returning a -ESTALE error, and making send fail with that error. So make
6545 * sure a send does not see any orphans we have just removed, and that it will
6546 * see the same inodes regardless of whether a transaction commit happened
6547 * before it started (meaning that the commit root will be the same as the
6548 * current root) or not.
6550 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6553 struct btrfs_trans_handle *trans = NULL;
6556 if (sctx->parent_root &&
6557 sctx->parent_root->node != sctx->parent_root->commit_root)
6560 for (i = 0; i < sctx->clone_roots_cnt; i++)
6561 if (sctx->clone_roots[i].root->node !=
6562 sctx->clone_roots[i].root->commit_root)
6566 return btrfs_end_transaction(trans);
6571 /* Use any root, all fs roots will get their commit roots updated. */
6573 trans = btrfs_join_transaction(sctx->send_root);
6575 return PTR_ERR(trans);
6579 return btrfs_commit_transaction(trans);
6582 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6584 spin_lock(&root->root_item_lock);
6585 root->send_in_progress--;
6587 * Not much left to do, we don't know why it's unbalanced and
6588 * can't blindly reset it to 0.
6590 if (root->send_in_progress < 0)
6591 btrfs_err(root->fs_info,
6592 "send_in_progress unbalanced %d root %llu",
6593 root->send_in_progress, root->root_key.objectid);
6594 spin_unlock(&root->root_item_lock);
6597 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
6600 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
6601 struct btrfs_fs_info *fs_info = send_root->fs_info;
6602 struct btrfs_root *clone_root;
6603 struct btrfs_key key;
6604 struct send_ctx *sctx = NULL;
6606 u64 *clone_sources_tmp = NULL;
6607 int clone_sources_to_rollback = 0;
6608 unsigned alloc_size;
6609 int sort_clone_roots = 0;
6612 if (!capable(CAP_SYS_ADMIN))
6616 * The subvolume must remain read-only during send, protect against
6617 * making it RW. This also protects against deletion.
6619 spin_lock(&send_root->root_item_lock);
6620 send_root->send_in_progress++;
6621 spin_unlock(&send_root->root_item_lock);
6624 * This is done when we lookup the root, it should already be complete
6625 * by the time we get here.
6627 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6630 * Userspace tools do the checks and warn the user if it's
6633 if (!btrfs_root_readonly(send_root)) {
6639 * Check that we don't overflow at later allocations, we request
6640 * clone_sources_count + 1 items, and compare to unsigned long inside
6643 if (arg->clone_sources_count >
6644 ULONG_MAX / sizeof(struct clone_root) - 1) {
6649 if (!access_ok(VERIFY_READ, arg->clone_sources,
6650 sizeof(*arg->clone_sources) *
6651 arg->clone_sources_count)) {
6656 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6661 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
6667 INIT_LIST_HEAD(&sctx->new_refs);
6668 INIT_LIST_HEAD(&sctx->deleted_refs);
6669 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
6670 INIT_LIST_HEAD(&sctx->name_cache_list);
6672 sctx->flags = arg->flags;
6674 sctx->send_filp = fget(arg->send_fd);
6675 if (!sctx->send_filp) {
6680 sctx->send_root = send_root;
6682 * Unlikely but possible, if the subvolume is marked for deletion but
6683 * is slow to remove the directory entry, send can still be started
6685 if (btrfs_root_dead(sctx->send_root)) {
6690 sctx->clone_roots_cnt = arg->clone_sources_count;
6692 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6693 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
6694 if (!sctx->send_buf) {
6699 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
6700 if (!sctx->read_buf) {
6705 sctx->pending_dir_moves = RB_ROOT;
6706 sctx->waiting_dir_moves = RB_ROOT;
6707 sctx->orphan_dirs = RB_ROOT;
6709 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6711 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
6712 if (!sctx->clone_roots) {
6717 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6719 if (arg->clone_sources_count) {
6720 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
6721 if (!clone_sources_tmp) {
6726 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6733 for (i = 0; i < arg->clone_sources_count; i++) {
6734 key.objectid = clone_sources_tmp[i];
6735 key.type = BTRFS_ROOT_ITEM_KEY;
6736 key.offset = (u64)-1;
6738 index = srcu_read_lock(&fs_info->subvol_srcu);
6740 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6741 if (IS_ERR(clone_root)) {
6742 srcu_read_unlock(&fs_info->subvol_srcu, index);
6743 ret = PTR_ERR(clone_root);
6746 spin_lock(&clone_root->root_item_lock);
6747 if (!btrfs_root_readonly(clone_root) ||
6748 btrfs_root_dead(clone_root)) {
6749 spin_unlock(&clone_root->root_item_lock);
6750 srcu_read_unlock(&fs_info->subvol_srcu, index);
6754 clone_root->send_in_progress++;
6755 spin_unlock(&clone_root->root_item_lock);
6756 srcu_read_unlock(&fs_info->subvol_srcu, index);
6758 sctx->clone_roots[i].root = clone_root;
6759 clone_sources_to_rollback = i + 1;
6761 kvfree(clone_sources_tmp);
6762 clone_sources_tmp = NULL;
6765 if (arg->parent_root) {
6766 key.objectid = arg->parent_root;
6767 key.type = BTRFS_ROOT_ITEM_KEY;
6768 key.offset = (u64)-1;
6770 index = srcu_read_lock(&fs_info->subvol_srcu);
6772 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6773 if (IS_ERR(sctx->parent_root)) {
6774 srcu_read_unlock(&fs_info->subvol_srcu, index);
6775 ret = PTR_ERR(sctx->parent_root);
6779 spin_lock(&sctx->parent_root->root_item_lock);
6780 sctx->parent_root->send_in_progress++;
6781 if (!btrfs_root_readonly(sctx->parent_root) ||
6782 btrfs_root_dead(sctx->parent_root)) {
6783 spin_unlock(&sctx->parent_root->root_item_lock);
6784 srcu_read_unlock(&fs_info->subvol_srcu, index);
6788 spin_unlock(&sctx->parent_root->root_item_lock);
6790 srcu_read_unlock(&fs_info->subvol_srcu, index);
6794 * Clones from send_root are allowed, but only if the clone source
6795 * is behind the current send position. This is checked while searching
6796 * for possible clone sources.
6798 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6800 /* We do a bsearch later */
6801 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6802 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6804 sort_clone_roots = 1;
6806 ret = ensure_commit_roots_uptodate(sctx);
6810 current->journal_info = BTRFS_SEND_TRANS_STUB;
6811 ret = send_subvol(sctx);
6812 current->journal_info = NULL;
6816 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6817 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6820 ret = send_cmd(sctx);
6826 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6827 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6829 struct pending_dir_move *pm;
6831 n = rb_first(&sctx->pending_dir_moves);
6832 pm = rb_entry(n, struct pending_dir_move, node);
6833 while (!list_empty(&pm->list)) {
6834 struct pending_dir_move *pm2;
6836 pm2 = list_first_entry(&pm->list,
6837 struct pending_dir_move, list);
6838 free_pending_move(sctx, pm2);
6840 free_pending_move(sctx, pm);
6843 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6844 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6846 struct waiting_dir_move *dm;
6848 n = rb_first(&sctx->waiting_dir_moves);
6849 dm = rb_entry(n, struct waiting_dir_move, node);
6850 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6854 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6855 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6857 struct orphan_dir_info *odi;
6859 n = rb_first(&sctx->orphan_dirs);
6860 odi = rb_entry(n, struct orphan_dir_info, node);
6861 free_orphan_dir_info(sctx, odi);
6864 if (sort_clone_roots) {
6865 for (i = 0; i < sctx->clone_roots_cnt; i++)
6866 btrfs_root_dec_send_in_progress(
6867 sctx->clone_roots[i].root);
6869 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6870 btrfs_root_dec_send_in_progress(
6871 sctx->clone_roots[i].root);
6873 btrfs_root_dec_send_in_progress(send_root);
6875 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6876 btrfs_root_dec_send_in_progress(sctx->parent_root);
6878 kvfree(clone_sources_tmp);
6881 if (sctx->send_filp)
6882 fput(sctx->send_filp);
6884 kvfree(sctx->clone_roots);
6885 kvfree(sctx->send_buf);
6886 kvfree(sctx->read_buf);
6888 name_cache_free(sctx);