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 = btrfs_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 /* number of total found references */
1167 * used for clones found in send_root. clones found behind cur_objectid
1168 * and cur_offset are not considered as allowed clones.
1173 /* may be truncated in case it's the last extent in a file */
1176 /* data offset in the file extent item */
1179 /* Just to check for bugs in backref resolving */
1183 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1185 u64 root = (u64)(uintptr_t)key;
1186 struct clone_root *cr = (struct clone_root *)elt;
1188 if (root < cr->root->root_key.objectid)
1190 if (root > cr->root->root_key.objectid)
1195 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1197 struct clone_root *cr1 = (struct clone_root *)e1;
1198 struct clone_root *cr2 = (struct clone_root *)e2;
1200 if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1202 if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1208 * Called for every backref that is found for the current extent.
1209 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1211 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1213 struct backref_ctx *bctx = ctx_;
1214 struct clone_root *found;
1216 /* First check if the root is in the list of accepted clone sources */
1217 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1218 bctx->sctx->clone_roots_cnt,
1219 sizeof(struct clone_root),
1220 __clone_root_cmp_bsearch);
1224 if (found->root == bctx->sctx->send_root &&
1225 ino == bctx->cur_objectid &&
1226 offset == bctx->cur_offset) {
1227 bctx->found_itself = 1;
1231 * Make sure we don't consider clones from send_root that are
1232 * behind the current inode/offset.
1234 if (found->root == bctx->sctx->send_root) {
1236 * TODO for the moment we don't accept clones from the inode
1237 * that is currently send. We may change this when
1238 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1241 if (ino >= bctx->cur_objectid)
1246 found->found_refs++;
1247 if (ino < found->ino) {
1249 found->offset = offset;
1250 } else if (found->ino == ino) {
1252 * same extent found more then once in the same file.
1254 if (found->offset > offset + bctx->extent_len)
1255 found->offset = offset;
1262 * Given an inode, offset and extent item, it finds a good clone for a clone
1263 * instruction. Returns -ENOENT when none could be found. The function makes
1264 * sure that the returned clone is usable at the point where sending is at the
1265 * moment. This means, that no clones are accepted which lie behind the current
1268 * path must point to the extent item when called.
1270 static int find_extent_clone(struct send_ctx *sctx,
1271 struct btrfs_path *path,
1272 u64 ino, u64 data_offset,
1274 struct clone_root **found)
1276 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1282 u64 extent_item_pos;
1284 struct btrfs_file_extent_item *fi;
1285 struct extent_buffer *eb = path->nodes[0];
1286 struct backref_ctx *backref_ctx = NULL;
1287 struct clone_root *cur_clone_root;
1288 struct btrfs_key found_key;
1289 struct btrfs_path *tmp_path;
1293 tmp_path = alloc_path_for_send();
1297 /* We only use this path under the commit sem */
1298 tmp_path->need_commit_sem = 0;
1300 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1306 if (data_offset >= ino_size) {
1308 * There may be extents that lie behind the file's size.
1309 * I at least had this in combination with snapshotting while
1310 * writing large files.
1316 fi = btrfs_item_ptr(eb, path->slots[0],
1317 struct btrfs_file_extent_item);
1318 extent_type = btrfs_file_extent_type(eb, fi);
1319 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1323 compressed = btrfs_file_extent_compression(eb, fi);
1325 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1326 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1327 if (disk_byte == 0) {
1331 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1333 down_read(&fs_info->commit_root_sem);
1334 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1335 &found_key, &flags);
1336 up_read(&fs_info->commit_root_sem);
1337 btrfs_release_path(tmp_path);
1341 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1347 * Setup the clone roots.
1349 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1350 cur_clone_root = sctx->clone_roots + i;
1351 cur_clone_root->ino = (u64)-1;
1352 cur_clone_root->offset = 0;
1353 cur_clone_root->found_refs = 0;
1356 backref_ctx->sctx = sctx;
1357 backref_ctx->found = 0;
1358 backref_ctx->cur_objectid = ino;
1359 backref_ctx->cur_offset = data_offset;
1360 backref_ctx->found_itself = 0;
1361 backref_ctx->extent_len = num_bytes;
1363 * For non-compressed extents iterate_extent_inodes() gives us extent
1364 * offsets that already take into account the data offset, but not for
1365 * compressed extents, since the offset is logical and not relative to
1366 * the physical extent locations. We must take this into account to
1367 * avoid sending clone offsets that go beyond the source file's size,
1368 * which would result in the clone ioctl failing with -EINVAL on the
1371 if (compressed == BTRFS_COMPRESS_NONE)
1372 backref_ctx->data_offset = 0;
1374 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1377 * The last extent of a file may be too large due to page alignment.
1378 * We need to adjust extent_len in this case so that the checks in
1379 * __iterate_backrefs work.
1381 if (data_offset + num_bytes >= ino_size)
1382 backref_ctx->extent_len = ino_size - data_offset;
1385 * Now collect all backrefs.
1387 if (compressed == BTRFS_COMPRESS_NONE)
1388 extent_item_pos = logical - found_key.objectid;
1390 extent_item_pos = 0;
1391 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1392 extent_item_pos, 1, __iterate_backrefs,
1393 backref_ctx, false);
1398 if (!backref_ctx->found_itself) {
1399 /* found a bug in backref code? */
1402 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1403 ino, data_offset, disk_byte, found_key.objectid);
1407 btrfs_debug(fs_info,
1408 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1409 data_offset, ino, num_bytes, logical);
1411 if (!backref_ctx->found)
1412 btrfs_debug(fs_info, "no clones found");
1414 cur_clone_root = NULL;
1415 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1416 if (sctx->clone_roots[i].found_refs) {
1417 if (!cur_clone_root)
1418 cur_clone_root = sctx->clone_roots + i;
1419 else if (sctx->clone_roots[i].root == sctx->send_root)
1420 /* prefer clones from send_root over others */
1421 cur_clone_root = sctx->clone_roots + i;
1426 if (cur_clone_root) {
1427 *found = cur_clone_root;
1434 btrfs_free_path(tmp_path);
1439 static int read_symlink(struct btrfs_root *root,
1441 struct fs_path *dest)
1444 struct btrfs_path *path;
1445 struct btrfs_key key;
1446 struct btrfs_file_extent_item *ei;
1452 path = alloc_path_for_send();
1457 key.type = BTRFS_EXTENT_DATA_KEY;
1459 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1464 * An empty symlink inode. Can happen in rare error paths when
1465 * creating a symlink (transaction committed before the inode
1466 * eviction handler removed the symlink inode items and a crash
1467 * happened in between or the subvol was snapshoted in between).
1468 * Print an informative message to dmesg/syslog so that the user
1469 * can delete the symlink.
1471 btrfs_err(root->fs_info,
1472 "Found empty symlink inode %llu at root %llu",
1473 ino, root->root_key.objectid);
1478 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1479 struct btrfs_file_extent_item);
1480 type = btrfs_file_extent_type(path->nodes[0], ei);
1481 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1482 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1483 BUG_ON(compression);
1485 off = btrfs_file_extent_inline_start(ei);
1486 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
1488 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1491 btrfs_free_path(path);
1496 * Helper function to generate a file name that is unique in the root of
1497 * send_root and parent_root. This is used to generate names for orphan inodes.
1499 static int gen_unique_name(struct send_ctx *sctx,
1501 struct fs_path *dest)
1504 struct btrfs_path *path;
1505 struct btrfs_dir_item *di;
1510 path = alloc_path_for_send();
1515 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1517 ASSERT(len < sizeof(tmp));
1519 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1520 path, BTRFS_FIRST_FREE_OBJECTID,
1521 tmp, strlen(tmp), 0);
1522 btrfs_release_path(path);
1528 /* not unique, try again */
1533 if (!sctx->parent_root) {
1539 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1540 path, BTRFS_FIRST_FREE_OBJECTID,
1541 tmp, strlen(tmp), 0);
1542 btrfs_release_path(path);
1548 /* not unique, try again */
1556 ret = fs_path_add(dest, tmp, strlen(tmp));
1559 btrfs_free_path(path);
1564 inode_state_no_change,
1565 inode_state_will_create,
1566 inode_state_did_create,
1567 inode_state_will_delete,
1568 inode_state_did_delete,
1571 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1579 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1581 if (ret < 0 && ret != -ENOENT)
1585 if (!sctx->parent_root) {
1586 right_ret = -ENOENT;
1588 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1589 NULL, NULL, NULL, NULL);
1590 if (ret < 0 && ret != -ENOENT)
1595 if (!left_ret && !right_ret) {
1596 if (left_gen == gen && right_gen == gen) {
1597 ret = inode_state_no_change;
1598 } else if (left_gen == gen) {
1599 if (ino < sctx->send_progress)
1600 ret = inode_state_did_create;
1602 ret = inode_state_will_create;
1603 } else if (right_gen == gen) {
1604 if (ino < sctx->send_progress)
1605 ret = inode_state_did_delete;
1607 ret = inode_state_will_delete;
1611 } else if (!left_ret) {
1612 if (left_gen == gen) {
1613 if (ino < sctx->send_progress)
1614 ret = inode_state_did_create;
1616 ret = inode_state_will_create;
1620 } else if (!right_ret) {
1621 if (right_gen == gen) {
1622 if (ino < sctx->send_progress)
1623 ret = inode_state_did_delete;
1625 ret = inode_state_will_delete;
1637 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1641 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1644 ret = get_cur_inode_state(sctx, ino, gen);
1648 if (ret == inode_state_no_change ||
1649 ret == inode_state_did_create ||
1650 ret == inode_state_will_delete)
1660 * Helper function to lookup a dir item in a dir.
1662 static int lookup_dir_item_inode(struct btrfs_root *root,
1663 u64 dir, const char *name, int name_len,
1668 struct btrfs_dir_item *di;
1669 struct btrfs_key key;
1670 struct btrfs_path *path;
1672 path = alloc_path_for_send();
1676 di = btrfs_lookup_dir_item(NULL, root, path,
1677 dir, name, name_len, 0);
1678 if (IS_ERR_OR_NULL(di)) {
1679 ret = di ? PTR_ERR(di) : -ENOENT;
1682 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1683 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1687 *found_inode = key.objectid;
1688 *found_type = btrfs_dir_type(path->nodes[0], di);
1691 btrfs_free_path(path);
1696 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1697 * generation of the parent dir and the name of the dir entry.
1699 static int get_first_ref(struct btrfs_root *root, u64 ino,
1700 u64 *dir, u64 *dir_gen, struct fs_path *name)
1703 struct btrfs_key key;
1704 struct btrfs_key found_key;
1705 struct btrfs_path *path;
1709 path = alloc_path_for_send();
1714 key.type = BTRFS_INODE_REF_KEY;
1717 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1721 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1723 if (ret || found_key.objectid != ino ||
1724 (found_key.type != BTRFS_INODE_REF_KEY &&
1725 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1730 if (found_key.type == BTRFS_INODE_REF_KEY) {
1731 struct btrfs_inode_ref *iref;
1732 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1733 struct btrfs_inode_ref);
1734 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1735 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1736 (unsigned long)(iref + 1),
1738 parent_dir = found_key.offset;
1740 struct btrfs_inode_extref *extref;
1741 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1742 struct btrfs_inode_extref);
1743 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1744 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1745 (unsigned long)&extref->name, len);
1746 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1750 btrfs_release_path(path);
1753 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1762 btrfs_free_path(path);
1766 static int is_first_ref(struct btrfs_root *root,
1768 const char *name, int name_len)
1771 struct fs_path *tmp_name;
1774 tmp_name = fs_path_alloc();
1778 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1782 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1787 ret = !memcmp(tmp_name->start, name, name_len);
1790 fs_path_free(tmp_name);
1795 * Used by process_recorded_refs to determine if a new ref would overwrite an
1796 * already existing ref. In case it detects an overwrite, it returns the
1797 * inode/gen in who_ino/who_gen.
1798 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1799 * to make sure later references to the overwritten inode are possible.
1800 * Orphanizing is however only required for the first ref of an inode.
1801 * process_recorded_refs does an additional is_first_ref check to see if
1802 * orphanizing is really required.
1804 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1805 const char *name, int name_len,
1806 u64 *who_ino, u64 *who_gen, u64 *who_mode)
1810 u64 other_inode = 0;
1813 if (!sctx->parent_root)
1816 ret = is_inode_existent(sctx, dir, dir_gen);
1821 * If we have a parent root we need to verify that the parent dir was
1822 * not deleted and then re-created, if it was then we have no overwrite
1823 * and we can just unlink this entry.
1825 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1826 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1828 if (ret < 0 && ret != -ENOENT)
1838 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1839 &other_inode, &other_type);
1840 if (ret < 0 && ret != -ENOENT)
1848 * Check if the overwritten ref was already processed. If yes, the ref
1849 * was already unlinked/moved, so we can safely assume that we will not
1850 * overwrite anything at this point in time.
1852 if (other_inode > sctx->send_progress ||
1853 is_waiting_for_move(sctx, other_inode)) {
1854 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1855 who_gen, who_mode, NULL, NULL, NULL);
1860 *who_ino = other_inode;
1870 * Checks if the ref was overwritten by an already processed inode. This is
1871 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1872 * thus the orphan name needs be used.
1873 * process_recorded_refs also uses it to avoid unlinking of refs that were
1876 static int did_overwrite_ref(struct send_ctx *sctx,
1877 u64 dir, u64 dir_gen,
1878 u64 ino, u64 ino_gen,
1879 const char *name, int name_len)
1886 if (!sctx->parent_root)
1889 ret = is_inode_existent(sctx, dir, dir_gen);
1893 if (dir != BTRFS_FIRST_FREE_OBJECTID) {
1894 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL,
1896 if (ret < 0 && ret != -ENOENT)
1906 /* check if the ref was overwritten by another ref */
1907 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1908 &ow_inode, &other_type);
1909 if (ret < 0 && ret != -ENOENT)
1912 /* was never and will never be overwritten */
1917 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1922 if (ow_inode == ino && gen == ino_gen) {
1928 * We know that it is or will be overwritten. Check this now.
1929 * The current inode being processed might have been the one that caused
1930 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1931 * the current inode being processed.
1933 if ((ow_inode < sctx->send_progress) ||
1934 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1935 gen == sctx->cur_inode_gen))
1945 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1946 * that got overwritten. This is used by process_recorded_refs to determine
1947 * if it has to use the path as returned by get_cur_path or the orphan name.
1949 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1952 struct fs_path *name = NULL;
1956 if (!sctx->parent_root)
1959 name = fs_path_alloc();
1963 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1967 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1968 name->start, fs_path_len(name));
1976 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1977 * so we need to do some special handling in case we have clashes. This function
1978 * takes care of this with the help of name_cache_entry::radix_list.
1979 * In case of error, nce is kfreed.
1981 static int name_cache_insert(struct send_ctx *sctx,
1982 struct name_cache_entry *nce)
1985 struct list_head *nce_head;
1987 nce_head = radix_tree_lookup(&sctx->name_cache,
1988 (unsigned long)nce->ino);
1990 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
1995 INIT_LIST_HEAD(nce_head);
1997 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2004 list_add_tail(&nce->radix_list, nce_head);
2005 list_add_tail(&nce->list, &sctx->name_cache_list);
2006 sctx->name_cache_size++;
2011 static void name_cache_delete(struct send_ctx *sctx,
2012 struct name_cache_entry *nce)
2014 struct list_head *nce_head;
2016 nce_head = radix_tree_lookup(&sctx->name_cache,
2017 (unsigned long)nce->ino);
2019 btrfs_err(sctx->send_root->fs_info,
2020 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2021 nce->ino, sctx->name_cache_size);
2024 list_del(&nce->radix_list);
2025 list_del(&nce->list);
2026 sctx->name_cache_size--;
2029 * We may not get to the final release of nce_head if the lookup fails
2031 if (nce_head && list_empty(nce_head)) {
2032 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2037 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2040 struct list_head *nce_head;
2041 struct name_cache_entry *cur;
2043 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2047 list_for_each_entry(cur, nce_head, radix_list) {
2048 if (cur->ino == ino && cur->gen == gen)
2055 * Removes the entry from the list and adds it back to the end. This marks the
2056 * entry as recently used so that name_cache_clean_unused does not remove it.
2058 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2060 list_del(&nce->list);
2061 list_add_tail(&nce->list, &sctx->name_cache_list);
2065 * Remove some entries from the beginning of name_cache_list.
2067 static void name_cache_clean_unused(struct send_ctx *sctx)
2069 struct name_cache_entry *nce;
2071 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2074 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2075 nce = list_entry(sctx->name_cache_list.next,
2076 struct name_cache_entry, list);
2077 name_cache_delete(sctx, nce);
2082 static void name_cache_free(struct send_ctx *sctx)
2084 struct name_cache_entry *nce;
2086 while (!list_empty(&sctx->name_cache_list)) {
2087 nce = list_entry(sctx->name_cache_list.next,
2088 struct name_cache_entry, list);
2089 name_cache_delete(sctx, nce);
2095 * Used by get_cur_path for each ref up to the root.
2096 * Returns 0 if it succeeded.
2097 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2098 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2099 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2100 * Returns <0 in case of error.
2102 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2106 struct fs_path *dest)
2110 struct name_cache_entry *nce = NULL;
2113 * First check if we already did a call to this function with the same
2114 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2115 * return the cached result.
2117 nce = name_cache_search(sctx, ino, gen);
2119 if (ino < sctx->send_progress && nce->need_later_update) {
2120 name_cache_delete(sctx, nce);
2124 name_cache_used(sctx, nce);
2125 *parent_ino = nce->parent_ino;
2126 *parent_gen = nce->parent_gen;
2127 ret = fs_path_add(dest, nce->name, nce->name_len);
2136 * If the inode is not existent yet, add the orphan name and return 1.
2137 * This should only happen for the parent dir that we determine in
2140 ret = is_inode_existent(sctx, ino, gen);
2145 ret = gen_unique_name(sctx, ino, gen, dest);
2153 * Depending on whether the inode was already processed or not, use
2154 * send_root or parent_root for ref lookup.
2156 if (ino < sctx->send_progress)
2157 ret = get_first_ref(sctx->send_root, ino,
2158 parent_ino, parent_gen, dest);
2160 ret = get_first_ref(sctx->parent_root, ino,
2161 parent_ino, parent_gen, dest);
2166 * Check if the ref was overwritten by an inode's ref that was processed
2167 * earlier. If yes, treat as orphan and return 1.
2169 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2170 dest->start, dest->end - dest->start);
2174 fs_path_reset(dest);
2175 ret = gen_unique_name(sctx, ino, gen, dest);
2183 * Store the result of the lookup in the name cache.
2185 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2193 nce->parent_ino = *parent_ino;
2194 nce->parent_gen = *parent_gen;
2195 nce->name_len = fs_path_len(dest);
2197 strcpy(nce->name, dest->start);
2199 if (ino < sctx->send_progress)
2200 nce->need_later_update = 0;
2202 nce->need_later_update = 1;
2204 nce_ret = name_cache_insert(sctx, nce);
2207 name_cache_clean_unused(sctx);
2214 * Magic happens here. This function returns the first ref to an inode as it
2215 * would look like while receiving the stream at this point in time.
2216 * We walk the path up to the root. For every inode in between, we check if it
2217 * was already processed/sent. If yes, we continue with the parent as found
2218 * in send_root. If not, we continue with the parent as found in parent_root.
2219 * If we encounter an inode that was deleted at this point in time, we use the
2220 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2221 * that were not created yet and overwritten inodes/refs.
2223 * When do we have orphan inodes:
2224 * 1. When an inode is freshly created and thus no valid refs are available yet
2225 * 2. When a directory lost all it's refs (deleted) but still has dir items
2226 * inside which were not processed yet (pending for move/delete). If anyone
2227 * tried to get the path to the dir items, it would get a path inside that
2229 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2230 * of an unprocessed inode. If in that case the first ref would be
2231 * overwritten, the overwritten inode gets "orphanized". Later when we
2232 * process this overwritten inode, it is restored at a new place by moving
2235 * sctx->send_progress tells this function at which point in time receiving
2238 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2239 struct fs_path *dest)
2242 struct fs_path *name = NULL;
2243 u64 parent_inode = 0;
2247 name = fs_path_alloc();
2254 fs_path_reset(dest);
2256 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2257 struct waiting_dir_move *wdm;
2259 fs_path_reset(name);
2261 if (is_waiting_for_rm(sctx, ino)) {
2262 ret = gen_unique_name(sctx, ino, gen, name);
2265 ret = fs_path_add_path(dest, name);
2269 wdm = get_waiting_dir_move(sctx, ino);
2270 if (wdm && wdm->orphanized) {
2271 ret = gen_unique_name(sctx, ino, gen, name);
2274 ret = get_first_ref(sctx->parent_root, ino,
2275 &parent_inode, &parent_gen, name);
2277 ret = __get_cur_name_and_parent(sctx, ino, gen,
2287 ret = fs_path_add_path(dest, name);
2298 fs_path_unreverse(dest);
2303 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2305 static int send_subvol_begin(struct send_ctx *sctx)
2308 struct btrfs_root *send_root = sctx->send_root;
2309 struct btrfs_root *parent_root = sctx->parent_root;
2310 struct btrfs_path *path;
2311 struct btrfs_key key;
2312 struct btrfs_root_ref *ref;
2313 struct extent_buffer *leaf;
2317 path = btrfs_alloc_path();
2321 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2323 btrfs_free_path(path);
2327 key.objectid = send_root->root_key.objectid;
2328 key.type = BTRFS_ROOT_BACKREF_KEY;
2331 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2340 leaf = path->nodes[0];
2341 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2342 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2343 key.objectid != send_root->root_key.objectid) {
2347 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2348 namelen = btrfs_root_ref_name_len(leaf, ref);
2349 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2350 btrfs_release_path(path);
2353 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2357 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2362 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2364 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2365 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2366 sctx->send_root->root_item.received_uuid);
2368 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2369 sctx->send_root->root_item.uuid);
2371 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2372 le64_to_cpu(sctx->send_root->root_item.ctransid));
2374 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2375 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2376 parent_root->root_item.received_uuid);
2378 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2379 parent_root->root_item.uuid);
2380 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2381 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2384 ret = send_cmd(sctx);
2388 btrfs_free_path(path);
2393 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2395 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2399 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2401 p = fs_path_alloc();
2405 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2409 ret = get_cur_path(sctx, ino, gen, p);
2412 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2413 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2415 ret = send_cmd(sctx);
2423 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2425 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2429 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2431 p = fs_path_alloc();
2435 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2439 ret = get_cur_path(sctx, ino, gen, p);
2442 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2443 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2445 ret = send_cmd(sctx);
2453 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2455 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2459 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2462 p = fs_path_alloc();
2466 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2470 ret = get_cur_path(sctx, ino, gen, p);
2473 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2474 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2475 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2477 ret = send_cmd(sctx);
2485 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2487 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2489 struct fs_path *p = NULL;
2490 struct btrfs_inode_item *ii;
2491 struct btrfs_path *path = NULL;
2492 struct extent_buffer *eb;
2493 struct btrfs_key key;
2496 btrfs_debug(fs_info, "send_utimes %llu", ino);
2498 p = fs_path_alloc();
2502 path = alloc_path_for_send();
2509 key.type = BTRFS_INODE_ITEM_KEY;
2511 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2517 eb = path->nodes[0];
2518 slot = path->slots[0];
2519 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2521 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2525 ret = get_cur_path(sctx, ino, gen, p);
2528 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2529 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2530 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2531 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2532 /* TODO Add otime support when the otime patches get into upstream */
2534 ret = send_cmd(sctx);
2539 btrfs_free_path(path);
2544 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2545 * a valid path yet because we did not process the refs yet. So, the inode
2546 * is created as orphan.
2548 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2550 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2558 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2560 p = fs_path_alloc();
2564 if (ino != sctx->cur_ino) {
2565 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2570 gen = sctx->cur_inode_gen;
2571 mode = sctx->cur_inode_mode;
2572 rdev = sctx->cur_inode_rdev;
2575 if (S_ISREG(mode)) {
2576 cmd = BTRFS_SEND_C_MKFILE;
2577 } else if (S_ISDIR(mode)) {
2578 cmd = BTRFS_SEND_C_MKDIR;
2579 } else if (S_ISLNK(mode)) {
2580 cmd = BTRFS_SEND_C_SYMLINK;
2581 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2582 cmd = BTRFS_SEND_C_MKNOD;
2583 } else if (S_ISFIFO(mode)) {
2584 cmd = BTRFS_SEND_C_MKFIFO;
2585 } else if (S_ISSOCK(mode)) {
2586 cmd = BTRFS_SEND_C_MKSOCK;
2588 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2589 (int)(mode & S_IFMT));
2594 ret = begin_cmd(sctx, cmd);
2598 ret = gen_unique_name(sctx, ino, gen, p);
2602 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2603 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2605 if (S_ISLNK(mode)) {
2607 ret = read_symlink(sctx->send_root, ino, p);
2610 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2611 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2612 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2613 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2614 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2617 ret = send_cmd(sctx);
2629 * We need some special handling for inodes that get processed before the parent
2630 * directory got created. See process_recorded_refs for details.
2631 * This function does the check if we already created the dir out of order.
2633 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2636 struct btrfs_path *path = NULL;
2637 struct btrfs_key key;
2638 struct btrfs_key found_key;
2639 struct btrfs_key di_key;
2640 struct extent_buffer *eb;
2641 struct btrfs_dir_item *di;
2644 path = alloc_path_for_send();
2651 key.type = BTRFS_DIR_INDEX_KEY;
2653 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2658 eb = path->nodes[0];
2659 slot = path->slots[0];
2660 if (slot >= btrfs_header_nritems(eb)) {
2661 ret = btrfs_next_leaf(sctx->send_root, path);
2664 } else if (ret > 0) {
2671 btrfs_item_key_to_cpu(eb, &found_key, slot);
2672 if (found_key.objectid != key.objectid ||
2673 found_key.type != key.type) {
2678 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2679 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2681 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2682 di_key.objectid < sctx->send_progress) {
2691 btrfs_free_path(path);
2696 * Only creates the inode if it is:
2697 * 1. Not a directory
2698 * 2. Or a directory which was not created already due to out of order
2699 * directories. See did_create_dir and process_recorded_refs for details.
2701 static int send_create_inode_if_needed(struct send_ctx *sctx)
2705 if (S_ISDIR(sctx->cur_inode_mode)) {
2706 ret = did_create_dir(sctx, sctx->cur_ino);
2715 ret = send_create_inode(sctx, sctx->cur_ino);
2723 struct recorded_ref {
2724 struct list_head list;
2726 struct fs_path *full_path;
2732 static void set_ref_path(struct recorded_ref *ref, struct fs_path *path)
2734 ref->full_path = path;
2735 ref->name = (char *)kbasename(ref->full_path->start);
2736 ref->name_len = ref->full_path->end - ref->name;
2740 * We need to process new refs before deleted refs, but compare_tree gives us
2741 * everything mixed. So we first record all refs and later process them.
2742 * This function is a helper to record one ref.
2744 static int __record_ref(struct list_head *head, u64 dir,
2745 u64 dir_gen, struct fs_path *path)
2747 struct recorded_ref *ref;
2749 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2754 ref->dir_gen = dir_gen;
2755 set_ref_path(ref, path);
2756 list_add_tail(&ref->list, head);
2760 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2762 struct recorded_ref *new;
2764 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2768 new->dir = ref->dir;
2769 new->dir_gen = ref->dir_gen;
2770 new->full_path = NULL;
2771 INIT_LIST_HEAD(&new->list);
2772 list_add_tail(&new->list, list);
2776 static void __free_recorded_refs(struct list_head *head)
2778 struct recorded_ref *cur;
2780 while (!list_empty(head)) {
2781 cur = list_entry(head->next, struct recorded_ref, list);
2782 fs_path_free(cur->full_path);
2783 list_del(&cur->list);
2788 static void free_recorded_refs(struct send_ctx *sctx)
2790 __free_recorded_refs(&sctx->new_refs);
2791 __free_recorded_refs(&sctx->deleted_refs);
2795 * Renames/moves a file/dir to its orphan name. Used when the first
2796 * ref of an unprocessed inode gets overwritten and for all non empty
2799 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2800 struct fs_path *path)
2803 struct fs_path *orphan;
2805 orphan = fs_path_alloc();
2809 ret = gen_unique_name(sctx, ino, gen, orphan);
2813 ret = send_rename(sctx, path, orphan);
2816 fs_path_free(orphan);
2820 static struct orphan_dir_info *
2821 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2823 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2824 struct rb_node *parent = NULL;
2825 struct orphan_dir_info *entry, *odi;
2829 entry = rb_entry(parent, struct orphan_dir_info, node);
2830 if (dir_ino < entry->ino) {
2832 } else if (dir_ino > entry->ino) {
2833 p = &(*p)->rb_right;
2839 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2841 return ERR_PTR(-ENOMEM);
2844 odi->last_dir_index_offset = 0;
2846 rb_link_node(&odi->node, parent, p);
2847 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2851 static struct orphan_dir_info *
2852 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2854 struct rb_node *n = sctx->orphan_dirs.rb_node;
2855 struct orphan_dir_info *entry;
2858 entry = rb_entry(n, struct orphan_dir_info, node);
2859 if (dir_ino < entry->ino)
2861 else if (dir_ino > entry->ino)
2869 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2871 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2876 static void free_orphan_dir_info(struct send_ctx *sctx,
2877 struct orphan_dir_info *odi)
2881 rb_erase(&odi->node, &sctx->orphan_dirs);
2886 * Returns 1 if a directory can be removed at this point in time.
2887 * We check this by iterating all dir items and checking if the inode behind
2888 * the dir item was already processed.
2890 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2894 struct btrfs_root *root = sctx->parent_root;
2895 struct btrfs_path *path;
2896 struct btrfs_key key;
2897 struct btrfs_key found_key;
2898 struct btrfs_key loc;
2899 struct btrfs_dir_item *di;
2900 struct orphan_dir_info *odi = NULL;
2903 * Don't try to rmdir the top/root subvolume dir.
2905 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2908 path = alloc_path_for_send();
2913 key.type = BTRFS_DIR_INDEX_KEY;
2916 odi = get_orphan_dir_info(sctx, dir);
2918 key.offset = odi->last_dir_index_offset;
2920 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2925 struct waiting_dir_move *dm;
2927 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2928 ret = btrfs_next_leaf(root, path);
2935 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2937 if (found_key.objectid != key.objectid ||
2938 found_key.type != key.type)
2941 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2942 struct btrfs_dir_item);
2943 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2945 dm = get_waiting_dir_move(sctx, loc.objectid);
2947 odi = add_orphan_dir_info(sctx, dir);
2953 odi->last_dir_index_offset = found_key.offset;
2954 dm->rmdir_ino = dir;
2959 if (loc.objectid > send_progress) {
2960 odi = add_orphan_dir_info(sctx, dir);
2966 odi->last_dir_index_offset = found_key.offset;
2973 free_orphan_dir_info(sctx, odi);
2978 btrfs_free_path(path);
2982 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2984 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2986 return entry != NULL;
2989 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
2991 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2992 struct rb_node *parent = NULL;
2993 struct waiting_dir_move *entry, *dm;
2995 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3000 dm->orphanized = orphanized;
3004 entry = rb_entry(parent, struct waiting_dir_move, node);
3005 if (ino < entry->ino) {
3007 } else if (ino > entry->ino) {
3008 p = &(*p)->rb_right;
3015 rb_link_node(&dm->node, parent, p);
3016 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3020 static struct waiting_dir_move *
3021 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3023 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3024 struct waiting_dir_move *entry;
3027 entry = rb_entry(n, struct waiting_dir_move, node);
3028 if (ino < entry->ino)
3030 else if (ino > entry->ino)
3038 static void free_waiting_dir_move(struct send_ctx *sctx,
3039 struct waiting_dir_move *dm)
3043 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3047 static int add_pending_dir_move(struct send_ctx *sctx,
3051 struct list_head *new_refs,
3052 struct list_head *deleted_refs,
3053 const bool is_orphan)
3055 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3056 struct rb_node *parent = NULL;
3057 struct pending_dir_move *entry = NULL, *pm;
3058 struct recorded_ref *cur;
3062 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3065 pm->parent_ino = parent_ino;
3068 INIT_LIST_HEAD(&pm->list);
3069 INIT_LIST_HEAD(&pm->update_refs);
3070 RB_CLEAR_NODE(&pm->node);
3074 entry = rb_entry(parent, struct pending_dir_move, node);
3075 if (parent_ino < entry->parent_ino) {
3077 } else if (parent_ino > entry->parent_ino) {
3078 p = &(*p)->rb_right;
3085 list_for_each_entry(cur, deleted_refs, list) {
3086 ret = dup_ref(cur, &pm->update_refs);
3090 list_for_each_entry(cur, new_refs, list) {
3091 ret = dup_ref(cur, &pm->update_refs);
3096 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3101 list_add_tail(&pm->list, &entry->list);
3103 rb_link_node(&pm->node, parent, p);
3104 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3109 __free_recorded_refs(&pm->update_refs);
3115 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3118 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3119 struct pending_dir_move *entry;
3122 entry = rb_entry(n, struct pending_dir_move, node);
3123 if (parent_ino < entry->parent_ino)
3125 else if (parent_ino > entry->parent_ino)
3133 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3134 u64 ino, u64 gen, u64 *ancestor_ino)
3137 u64 parent_inode = 0;
3139 u64 start_ino = ino;
3142 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3143 fs_path_reset(name);
3145 if (is_waiting_for_rm(sctx, ino))
3147 if (is_waiting_for_move(sctx, ino)) {
3148 if (*ancestor_ino == 0)
3149 *ancestor_ino = ino;
3150 ret = get_first_ref(sctx->parent_root, ino,
3151 &parent_inode, &parent_gen, name);
3153 ret = __get_cur_name_and_parent(sctx, ino, gen,
3163 if (parent_inode == start_ino) {
3165 if (*ancestor_ino == 0)
3166 *ancestor_ino = ino;
3175 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3177 struct fs_path *from_path = NULL;
3178 struct fs_path *to_path = NULL;
3179 struct fs_path *name = NULL;
3180 u64 orig_progress = sctx->send_progress;
3181 struct recorded_ref *cur;
3182 u64 parent_ino, parent_gen;
3183 struct waiting_dir_move *dm = NULL;
3189 name = fs_path_alloc();
3190 from_path = fs_path_alloc();
3191 if (!name || !from_path) {
3196 dm = get_waiting_dir_move(sctx, pm->ino);
3198 rmdir_ino = dm->rmdir_ino;
3199 is_orphan = dm->orphanized;
3200 free_waiting_dir_move(sctx, dm);
3203 ret = gen_unique_name(sctx, pm->ino,
3204 pm->gen, from_path);
3206 ret = get_first_ref(sctx->parent_root, pm->ino,
3207 &parent_ino, &parent_gen, name);
3210 ret = get_cur_path(sctx, parent_ino, parent_gen,
3214 ret = fs_path_add_path(from_path, name);
3219 sctx->send_progress = sctx->cur_ino + 1;
3220 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3224 LIST_HEAD(deleted_refs);
3225 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3226 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3227 &pm->update_refs, &deleted_refs,
3232 dm = get_waiting_dir_move(sctx, pm->ino);
3234 dm->rmdir_ino = rmdir_ino;
3238 fs_path_reset(name);
3241 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3245 ret = send_rename(sctx, from_path, to_path);
3250 struct orphan_dir_info *odi;
3253 odi = get_orphan_dir_info(sctx, rmdir_ino);
3255 /* already deleted */
3260 ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino);
3266 name = fs_path_alloc();
3271 ret = get_cur_path(sctx, rmdir_ino, gen, name);
3274 ret = send_rmdir(sctx, name);
3280 ret = send_utimes(sctx, pm->ino, pm->gen);
3285 * After rename/move, need to update the utimes of both new parent(s)
3286 * and old parent(s).
3288 list_for_each_entry(cur, &pm->update_refs, list) {
3290 * The parent inode might have been deleted in the send snapshot
3292 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3293 NULL, NULL, NULL, NULL, NULL);
3294 if (ret == -ENOENT) {
3301 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3308 fs_path_free(from_path);
3309 fs_path_free(to_path);
3310 sctx->send_progress = orig_progress;
3315 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3317 if (!list_empty(&m->list))
3319 if (!RB_EMPTY_NODE(&m->node))
3320 rb_erase(&m->node, &sctx->pending_dir_moves);
3321 __free_recorded_refs(&m->update_refs);
3325 static void tail_append_pending_moves(struct send_ctx *sctx,
3326 struct pending_dir_move *moves,
3327 struct list_head *stack)
3329 if (list_empty(&moves->list)) {
3330 list_add_tail(&moves->list, stack);
3333 list_splice_init(&moves->list, &list);
3334 list_add_tail(&moves->list, stack);
3335 list_splice_tail(&list, stack);
3337 if (!RB_EMPTY_NODE(&moves->node)) {
3338 rb_erase(&moves->node, &sctx->pending_dir_moves);
3339 RB_CLEAR_NODE(&moves->node);
3343 static int apply_children_dir_moves(struct send_ctx *sctx)
3345 struct pending_dir_move *pm;
3346 struct list_head stack;
3347 u64 parent_ino = sctx->cur_ino;
3350 pm = get_pending_dir_moves(sctx, parent_ino);
3354 INIT_LIST_HEAD(&stack);
3355 tail_append_pending_moves(sctx, pm, &stack);
3357 while (!list_empty(&stack)) {
3358 pm = list_first_entry(&stack, struct pending_dir_move, list);
3359 parent_ino = pm->ino;
3360 ret = apply_dir_move(sctx, pm);
3361 free_pending_move(sctx, pm);
3364 pm = get_pending_dir_moves(sctx, parent_ino);
3366 tail_append_pending_moves(sctx, pm, &stack);
3371 while (!list_empty(&stack)) {
3372 pm = list_first_entry(&stack, struct pending_dir_move, list);
3373 free_pending_move(sctx, pm);
3379 * We might need to delay a directory rename even when no ancestor directory
3380 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3381 * renamed. This happens when we rename a directory to the old name (the name
3382 * in the parent root) of some other unrelated directory that got its rename
3383 * delayed due to some ancestor with higher number that got renamed.
3389 * |---- a/ (ino 257)
3390 * | |---- file (ino 260)
3392 * |---- b/ (ino 258)
3393 * |---- c/ (ino 259)
3397 * |---- a/ (ino 258)
3398 * |---- x/ (ino 259)
3399 * |---- y/ (ino 257)
3400 * |----- file (ino 260)
3402 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3403 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3404 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3407 * 1 - rename 259 from 'c' to 'x'
3408 * 2 - rename 257 from 'a' to 'x/y'
3409 * 3 - rename 258 from 'b' to 'a'
3411 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3412 * be done right away and < 0 on error.
3414 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3415 struct recorded_ref *parent_ref,
3416 const bool is_orphan)
3418 struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info;
3419 struct btrfs_path *path;
3420 struct btrfs_key key;
3421 struct btrfs_key di_key;
3422 struct btrfs_dir_item *di;
3426 struct waiting_dir_move *wdm;
3428 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3431 path = alloc_path_for_send();
3435 key.objectid = parent_ref->dir;
3436 key.type = BTRFS_DIR_ITEM_KEY;
3437 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3439 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3442 } else if (ret > 0) {
3447 di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name,
3448 parent_ref->name_len);
3454 * di_key.objectid has the number of the inode that has a dentry in the
3455 * parent directory with the same name that sctx->cur_ino is being
3456 * renamed to. We need to check if that inode is in the send root as
3457 * well and if it is currently marked as an inode with a pending rename,
3458 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3459 * that it happens after that other inode is renamed.
3461 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3462 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3467 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3468 &left_gen, NULL, NULL, NULL, NULL);
3471 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3472 &right_gen, NULL, NULL, NULL, NULL);
3479 /* Different inode, no need to delay the rename of sctx->cur_ino */
3480 if (right_gen != left_gen) {
3485 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3486 if (wdm && !wdm->orphanized) {
3487 ret = add_pending_dir_move(sctx,
3489 sctx->cur_inode_gen,
3492 &sctx->deleted_refs,
3498 btrfs_free_path(path);
3503 * Check if inode ino2, or any of its ancestors, is inode ino1.
3504 * Return 1 if true, 0 if false and < 0 on error.
3506 static int check_ino_in_path(struct btrfs_root *root,
3511 struct fs_path *fs_path)
3516 return ino1_gen == ino2_gen;
3518 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3523 fs_path_reset(fs_path);
3524 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3528 return parent_gen == ino1_gen;
3535 * Check if ino ino1 is an ancestor of inode ino2 in the given root for any
3536 * possible path (in case ino2 is not a directory and has multiple hard links).
3537 * Return 1 if true, 0 if false and < 0 on error.
3539 static int is_ancestor(struct btrfs_root *root,
3543 struct fs_path *fs_path)
3545 bool free_fs_path = false;
3547 struct btrfs_path *path = NULL;
3548 struct btrfs_key key;
3551 fs_path = fs_path_alloc();
3554 free_fs_path = true;
3557 path = alloc_path_for_send();
3563 key.objectid = ino2;
3564 key.type = BTRFS_INODE_REF_KEY;
3567 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3572 struct extent_buffer *leaf = path->nodes[0];
3573 int slot = path->slots[0];
3577 if (slot >= btrfs_header_nritems(leaf)) {
3578 ret = btrfs_next_leaf(root, path);
3586 btrfs_item_key_to_cpu(leaf, &key, slot);
3587 if (key.objectid != ino2)
3589 if (key.type != BTRFS_INODE_REF_KEY &&
3590 key.type != BTRFS_INODE_EXTREF_KEY)
3593 item_size = btrfs_item_size_nr(leaf, slot);
3594 while (cur_offset < item_size) {
3598 if (key.type == BTRFS_INODE_EXTREF_KEY) {
3600 struct btrfs_inode_extref *extref;
3602 ptr = btrfs_item_ptr_offset(leaf, slot);
3603 extref = (struct btrfs_inode_extref *)
3605 parent = btrfs_inode_extref_parent(leaf,
3607 cur_offset += sizeof(*extref);
3608 cur_offset += btrfs_inode_extref_name_len(leaf,
3611 parent = key.offset;
3612 cur_offset = item_size;
3615 ret = get_inode_info(root, parent, NULL, &parent_gen,
3616 NULL, NULL, NULL, NULL);
3619 ret = check_ino_in_path(root, ino1, ino1_gen,
3620 parent, parent_gen, fs_path);
3628 btrfs_free_path(path);
3630 fs_path_free(fs_path);
3634 static int wait_for_parent_move(struct send_ctx *sctx,
3635 struct recorded_ref *parent_ref,
3636 const bool is_orphan)
3639 u64 ino = parent_ref->dir;
3640 u64 ino_gen = parent_ref->dir_gen;
3641 u64 parent_ino_before, parent_ino_after;
3642 struct fs_path *path_before = NULL;
3643 struct fs_path *path_after = NULL;
3646 path_after = fs_path_alloc();
3647 path_before = fs_path_alloc();
3648 if (!path_after || !path_before) {
3654 * Our current directory inode may not yet be renamed/moved because some
3655 * ancestor (immediate or not) has to be renamed/moved first. So find if
3656 * such ancestor exists and make sure our own rename/move happens after
3657 * that ancestor is processed to avoid path build infinite loops (done
3658 * at get_cur_path()).
3660 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3661 u64 parent_ino_after_gen;
3663 if (is_waiting_for_move(sctx, ino)) {
3665 * If the current inode is an ancestor of ino in the
3666 * parent root, we need to delay the rename of the
3667 * current inode, otherwise don't delayed the rename
3668 * because we can end up with a circular dependency
3669 * of renames, resulting in some directories never
3670 * getting the respective rename operations issued in
3671 * the send stream or getting into infinite path build
3674 ret = is_ancestor(sctx->parent_root,
3675 sctx->cur_ino, sctx->cur_inode_gen,
3681 fs_path_reset(path_before);
3682 fs_path_reset(path_after);
3684 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3685 &parent_ino_after_gen, path_after);
3688 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3690 if (ret < 0 && ret != -ENOENT) {
3692 } else if (ret == -ENOENT) {
3697 len1 = fs_path_len(path_before);
3698 len2 = fs_path_len(path_after);
3699 if (ino > sctx->cur_ino &&
3700 (parent_ino_before != parent_ino_after || len1 != len2 ||
3701 memcmp(path_before->start, path_after->start, len1))) {
3704 ret = get_inode_info(sctx->parent_root, ino, NULL,
3705 &parent_ino_gen, NULL, NULL, NULL,
3709 if (ino_gen == parent_ino_gen) {
3714 ino = parent_ino_after;
3715 ino_gen = parent_ino_after_gen;
3719 fs_path_free(path_before);
3720 fs_path_free(path_after);
3723 ret = add_pending_dir_move(sctx,
3725 sctx->cur_inode_gen,
3728 &sctx->deleted_refs,
3737 static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref)
3740 struct fs_path *new_path;
3743 * Our reference's name member points to its full_path member string, so
3744 * we use here a new path.
3746 new_path = fs_path_alloc();
3750 ret = get_cur_path(sctx, ref->dir, ref->dir_gen, new_path);
3752 fs_path_free(new_path);
3755 ret = fs_path_add(new_path, ref->name, ref->name_len);
3757 fs_path_free(new_path);
3761 fs_path_free(ref->full_path);
3762 set_ref_path(ref, new_path);
3768 * This does all the move/link/unlink/rmdir magic.
3770 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3772 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3774 struct recorded_ref *cur;
3775 struct recorded_ref *cur2;
3776 struct list_head check_dirs;
3777 struct fs_path *valid_path = NULL;
3781 int did_overwrite = 0;
3783 u64 last_dir_ino_rm = 0;
3784 bool can_rename = true;
3785 bool orphanized_dir = false;
3786 bool orphanized_ancestor = false;
3788 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3791 * This should never happen as the root dir always has the same ref
3792 * which is always '..'
3794 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3795 INIT_LIST_HEAD(&check_dirs);
3797 valid_path = fs_path_alloc();
3804 * First, check if the first ref of the current inode was overwritten
3805 * before. If yes, we know that the current inode was already orphanized
3806 * and thus use the orphan name. If not, we can use get_cur_path to
3807 * get the path of the first ref as it would like while receiving at
3808 * this point in time.
3809 * New inodes are always orphan at the beginning, so force to use the
3810 * orphan name in this case.
3811 * The first ref is stored in valid_path and will be updated if it
3812 * gets moved around.
3814 if (!sctx->cur_inode_new) {
3815 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3816 sctx->cur_inode_gen);
3822 if (sctx->cur_inode_new || did_overwrite) {
3823 ret = gen_unique_name(sctx, sctx->cur_ino,
3824 sctx->cur_inode_gen, valid_path);
3829 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3835 list_for_each_entry(cur, &sctx->new_refs, list) {
3837 * We may have refs where the parent directory does not exist
3838 * yet. This happens if the parent directories inum is higher
3839 * than the current inum. To handle this case, we create the
3840 * parent directory out of order. But we need to check if this
3841 * did already happen before due to other refs in the same dir.
3843 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3846 if (ret == inode_state_will_create) {
3849 * First check if any of the current inodes refs did
3850 * already create the dir.
3852 list_for_each_entry(cur2, &sctx->new_refs, list) {
3855 if (cur2->dir == cur->dir) {
3862 * If that did not happen, check if a previous inode
3863 * did already create the dir.
3866 ret = did_create_dir(sctx, cur->dir);
3870 ret = send_create_inode(sctx, cur->dir);
3877 * Check if this new ref would overwrite the first ref of
3878 * another unprocessed inode. If yes, orphanize the
3879 * overwritten inode. If we find an overwritten ref that is
3880 * not the first ref, simply unlink it.
3882 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3883 cur->name, cur->name_len,
3884 &ow_inode, &ow_gen, &ow_mode);
3888 ret = is_first_ref(sctx->parent_root,
3889 ow_inode, cur->dir, cur->name,
3894 struct name_cache_entry *nce;
3895 struct waiting_dir_move *wdm;
3897 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3901 if (S_ISDIR(ow_mode))
3902 orphanized_dir = true;
3905 * If ow_inode has its rename operation delayed
3906 * make sure that its orphanized name is used in
3907 * the source path when performing its rename
3910 if (is_waiting_for_move(sctx, ow_inode)) {
3911 wdm = get_waiting_dir_move(sctx,
3914 wdm->orphanized = true;
3918 * Make sure we clear our orphanized inode's
3919 * name from the name cache. This is because the
3920 * inode ow_inode might be an ancestor of some
3921 * other inode that will be orphanized as well
3922 * later and has an inode number greater than
3923 * sctx->send_progress. We need to prevent
3924 * future name lookups from using the old name
3925 * and get instead the orphan name.
3927 nce = name_cache_search(sctx, ow_inode, ow_gen);
3929 name_cache_delete(sctx, nce);
3934 * ow_inode might currently be an ancestor of
3935 * cur_ino, therefore compute valid_path (the
3936 * current path of cur_ino) again because it
3937 * might contain the pre-orphanization name of
3938 * ow_inode, which is no longer valid.
3940 ret = is_ancestor(sctx->parent_root,
3942 sctx->cur_ino, NULL);
3944 orphanized_ancestor = true;
3945 fs_path_reset(valid_path);
3946 ret = get_cur_path(sctx, sctx->cur_ino,
3947 sctx->cur_inode_gen,
3953 ret = send_unlink(sctx, cur->full_path);
3959 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3960 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3969 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3971 ret = wait_for_parent_move(sctx, cur, is_orphan);
3981 * link/move the ref to the new place. If we have an orphan
3982 * inode, move it and update valid_path. If not, link or move
3983 * it depending on the inode mode.
3985 if (is_orphan && can_rename) {
3986 ret = send_rename(sctx, valid_path, cur->full_path);
3990 ret = fs_path_copy(valid_path, cur->full_path);
3993 } else if (can_rename) {
3994 if (S_ISDIR(sctx->cur_inode_mode)) {
3996 * Dirs can't be linked, so move it. For moved
3997 * dirs, we always have one new and one deleted
3998 * ref. The deleted ref is ignored later.
4000 ret = send_rename(sctx, valid_path,
4003 ret = fs_path_copy(valid_path,
4009 * We might have previously orphanized an inode
4010 * which is an ancestor of our current inode,
4011 * so our reference's full path, which was
4012 * computed before any such orphanizations, must
4015 if (orphanized_dir) {
4016 ret = update_ref_path(sctx, cur);
4020 ret = send_link(sctx, cur->full_path,
4026 ret = dup_ref(cur, &check_dirs);
4031 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
4033 * Check if we can already rmdir the directory. If not,
4034 * orphanize it. For every dir item inside that gets deleted
4035 * later, we do this check again and rmdir it then if possible.
4036 * See the use of check_dirs for more details.
4038 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4043 ret = send_rmdir(sctx, valid_path);
4046 } else if (!is_orphan) {
4047 ret = orphanize_inode(sctx, sctx->cur_ino,
4048 sctx->cur_inode_gen, valid_path);
4054 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4055 ret = dup_ref(cur, &check_dirs);
4059 } else if (S_ISDIR(sctx->cur_inode_mode) &&
4060 !list_empty(&sctx->deleted_refs)) {
4062 * We have a moved dir. Add the old parent to check_dirs
4064 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
4066 ret = dup_ref(cur, &check_dirs);
4069 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
4071 * We have a non dir inode. Go through all deleted refs and
4072 * unlink them if they were not already overwritten by other
4075 list_for_each_entry(cur, &sctx->deleted_refs, list) {
4076 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4077 sctx->cur_ino, sctx->cur_inode_gen,
4078 cur->name, cur->name_len);
4083 * If we orphanized any ancestor before, we need
4084 * to recompute the full path for deleted names,
4085 * since any such path was computed before we
4086 * processed any references and orphanized any
4089 if (orphanized_ancestor) {
4090 ret = update_ref_path(sctx, cur);
4094 ret = send_unlink(sctx, cur->full_path);
4098 ret = dup_ref(cur, &check_dirs);
4103 * If the inode is still orphan, unlink the orphan. This may
4104 * happen when a previous inode did overwrite the first ref
4105 * of this inode and no new refs were added for the current
4106 * inode. Unlinking does not mean that the inode is deleted in
4107 * all cases. There may still be links to this inode in other
4111 ret = send_unlink(sctx, valid_path);
4118 * We did collect all parent dirs where cur_inode was once located. We
4119 * now go through all these dirs and check if they are pending for
4120 * deletion and if it's finally possible to perform the rmdir now.
4121 * We also update the inode stats of the parent dirs here.
4123 list_for_each_entry(cur, &check_dirs, list) {
4125 * In case we had refs into dirs that were not processed yet,
4126 * we don't need to do the utime and rmdir logic for these dirs.
4127 * The dir will be processed later.
4129 if (cur->dir > sctx->cur_ino)
4132 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
4136 if (ret == inode_state_did_create ||
4137 ret == inode_state_no_change) {
4138 /* TODO delayed utimes */
4139 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
4142 } else if (ret == inode_state_did_delete &&
4143 cur->dir != last_dir_ino_rm) {
4144 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
4149 ret = get_cur_path(sctx, cur->dir,
4150 cur->dir_gen, valid_path);
4153 ret = send_rmdir(sctx, valid_path);
4156 last_dir_ino_rm = cur->dir;
4164 __free_recorded_refs(&check_dirs);
4165 free_recorded_refs(sctx);
4166 fs_path_free(valid_path);
4170 static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name,
4171 void *ctx, struct list_head *refs)
4174 struct send_ctx *sctx = ctx;
4178 p = fs_path_alloc();
4182 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4187 ret = get_cur_path(sctx, dir, gen, p);
4190 ret = fs_path_add_path(p, name);
4194 ret = __record_ref(refs, dir, gen, p);
4202 static int __record_new_ref(int num, u64 dir, int index,
4203 struct fs_path *name,
4206 struct send_ctx *sctx = ctx;
4207 return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs);
4211 static int __record_deleted_ref(int num, u64 dir, int index,
4212 struct fs_path *name,
4215 struct send_ctx *sctx = ctx;
4216 return record_ref(sctx->parent_root, dir, name, ctx,
4217 &sctx->deleted_refs);
4220 static int record_new_ref(struct send_ctx *sctx)
4224 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4225 sctx->cmp_key, 0, __record_new_ref, sctx);
4234 static int record_deleted_ref(struct send_ctx *sctx)
4238 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4239 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4248 struct find_ref_ctx {
4251 struct btrfs_root *root;
4252 struct fs_path *name;
4256 static int __find_iref(int num, u64 dir, int index,
4257 struct fs_path *name,
4260 struct find_ref_ctx *ctx = ctx_;
4264 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4265 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4267 * To avoid doing extra lookups we'll only do this if everything
4270 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4274 if (dir_gen != ctx->dir_gen)
4276 ctx->found_idx = num;
4282 static int find_iref(struct btrfs_root *root,
4283 struct btrfs_path *path,
4284 struct btrfs_key *key,
4285 u64 dir, u64 dir_gen, struct fs_path *name)
4288 struct find_ref_ctx ctx;
4292 ctx.dir_gen = dir_gen;
4296 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4300 if (ctx.found_idx == -1)
4303 return ctx.found_idx;
4306 static int __record_changed_new_ref(int num, u64 dir, int index,
4307 struct fs_path *name,
4312 struct send_ctx *sctx = ctx;
4314 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4319 ret = find_iref(sctx->parent_root, sctx->right_path,
4320 sctx->cmp_key, dir, dir_gen, name);
4322 ret = __record_new_ref(num, dir, index, name, sctx);
4329 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4330 struct fs_path *name,
4335 struct send_ctx *sctx = ctx;
4337 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4342 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4343 dir, dir_gen, name);
4345 ret = __record_deleted_ref(num, dir, index, name, sctx);
4352 static int record_changed_ref(struct send_ctx *sctx)
4356 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4357 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4360 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4361 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4371 * Record and process all refs at once. Needed when an inode changes the
4372 * generation number, which means that it was deleted and recreated.
4374 static int process_all_refs(struct send_ctx *sctx,
4375 enum btrfs_compare_tree_result cmd)
4378 struct btrfs_root *root;
4379 struct btrfs_path *path;
4380 struct btrfs_key key;
4381 struct btrfs_key found_key;
4382 struct extent_buffer *eb;
4384 iterate_inode_ref_t cb;
4385 int pending_move = 0;
4387 path = alloc_path_for_send();
4391 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4392 root = sctx->send_root;
4393 cb = __record_new_ref;
4394 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4395 root = sctx->parent_root;
4396 cb = __record_deleted_ref;
4398 btrfs_err(sctx->send_root->fs_info,
4399 "Wrong command %d in process_all_refs", cmd);
4404 key.objectid = sctx->cmp_key->objectid;
4405 key.type = BTRFS_INODE_REF_KEY;
4407 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4412 eb = path->nodes[0];
4413 slot = path->slots[0];
4414 if (slot >= btrfs_header_nritems(eb)) {
4415 ret = btrfs_next_leaf(root, path);
4423 btrfs_item_key_to_cpu(eb, &found_key, slot);
4425 if (found_key.objectid != key.objectid ||
4426 (found_key.type != BTRFS_INODE_REF_KEY &&
4427 found_key.type != BTRFS_INODE_EXTREF_KEY))
4430 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4436 btrfs_release_path(path);
4439 * We don't actually care about pending_move as we are simply
4440 * re-creating this inode and will be rename'ing it into place once we
4441 * rename the parent directory.
4443 ret = process_recorded_refs(sctx, &pending_move);
4445 btrfs_free_path(path);
4449 static int send_set_xattr(struct send_ctx *sctx,
4450 struct fs_path *path,
4451 const char *name, int name_len,
4452 const char *data, int data_len)
4456 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4460 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4461 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4462 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4464 ret = send_cmd(sctx);
4471 static int send_remove_xattr(struct send_ctx *sctx,
4472 struct fs_path *path,
4473 const char *name, int name_len)
4477 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4481 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4482 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4484 ret = send_cmd(sctx);
4491 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4492 const char *name, int name_len,
4493 const char *data, int data_len,
4497 struct send_ctx *sctx = ctx;
4499 struct posix_acl_xattr_header dummy_acl;
4501 p = fs_path_alloc();
4506 * This hack is needed because empty acls are stored as zero byte
4507 * data in xattrs. Problem with that is, that receiving these zero byte
4508 * acls will fail later. To fix this, we send a dummy acl list that
4509 * only contains the version number and no entries.
4511 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4512 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4513 if (data_len == 0) {
4514 dummy_acl.a_version =
4515 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4516 data = (char *)&dummy_acl;
4517 data_len = sizeof(dummy_acl);
4521 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4525 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4532 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4533 const char *name, int name_len,
4534 const char *data, int data_len,
4538 struct send_ctx *sctx = ctx;
4541 p = fs_path_alloc();
4545 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4549 ret = send_remove_xattr(sctx, p, name, name_len);
4556 static int process_new_xattr(struct send_ctx *sctx)
4560 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4561 __process_new_xattr, sctx);
4566 static int process_deleted_xattr(struct send_ctx *sctx)
4568 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4569 __process_deleted_xattr, sctx);
4572 struct find_xattr_ctx {
4580 static int __find_xattr(int num, struct btrfs_key *di_key,
4581 const char *name, int name_len,
4582 const char *data, int data_len,
4583 u8 type, void *vctx)
4585 struct find_xattr_ctx *ctx = vctx;
4587 if (name_len == ctx->name_len &&
4588 strncmp(name, ctx->name, name_len) == 0) {
4589 ctx->found_idx = num;
4590 ctx->found_data_len = data_len;
4591 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4592 if (!ctx->found_data)
4599 static int find_xattr(struct btrfs_root *root,
4600 struct btrfs_path *path,
4601 struct btrfs_key *key,
4602 const char *name, int name_len,
4603 char **data, int *data_len)
4606 struct find_xattr_ctx ctx;
4609 ctx.name_len = name_len;
4611 ctx.found_data = NULL;
4612 ctx.found_data_len = 0;
4614 ret = iterate_dir_item(root, path, __find_xattr, &ctx);
4618 if (ctx.found_idx == -1)
4621 *data = ctx.found_data;
4622 *data_len = ctx.found_data_len;
4624 kfree(ctx.found_data);
4626 return ctx.found_idx;
4630 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4631 const char *name, int name_len,
4632 const char *data, int data_len,
4636 struct send_ctx *sctx = ctx;
4637 char *found_data = NULL;
4638 int found_data_len = 0;
4640 ret = find_xattr(sctx->parent_root, sctx->right_path,
4641 sctx->cmp_key, name, name_len, &found_data,
4643 if (ret == -ENOENT) {
4644 ret = __process_new_xattr(num, di_key, name, name_len, data,
4645 data_len, type, ctx);
4646 } else if (ret >= 0) {
4647 if (data_len != found_data_len ||
4648 memcmp(data, found_data, data_len)) {
4649 ret = __process_new_xattr(num, di_key, name, name_len,
4650 data, data_len, type, ctx);
4660 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4661 const char *name, int name_len,
4662 const char *data, int data_len,
4666 struct send_ctx *sctx = ctx;
4668 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4669 name, name_len, NULL, NULL);
4671 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4672 data_len, type, ctx);
4679 static int process_changed_xattr(struct send_ctx *sctx)
4683 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4684 __process_changed_new_xattr, sctx);
4687 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4688 __process_changed_deleted_xattr, sctx);
4694 static int process_all_new_xattrs(struct send_ctx *sctx)
4697 struct btrfs_root *root;
4698 struct btrfs_path *path;
4699 struct btrfs_key key;
4700 struct btrfs_key found_key;
4701 struct extent_buffer *eb;
4704 path = alloc_path_for_send();
4708 root = sctx->send_root;
4710 key.objectid = sctx->cmp_key->objectid;
4711 key.type = BTRFS_XATTR_ITEM_KEY;
4713 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4718 eb = path->nodes[0];
4719 slot = path->slots[0];
4720 if (slot >= btrfs_header_nritems(eb)) {
4721 ret = btrfs_next_leaf(root, path);
4724 } else if (ret > 0) {
4731 btrfs_item_key_to_cpu(eb, &found_key, slot);
4732 if (found_key.objectid != key.objectid ||
4733 found_key.type != key.type) {
4738 ret = iterate_dir_item(root, path, __process_new_xattr, sctx);
4746 btrfs_free_path(path);
4750 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4752 struct btrfs_root *root = sctx->send_root;
4753 struct btrfs_fs_info *fs_info = root->fs_info;
4754 struct inode *inode;
4757 struct btrfs_key key;
4758 pgoff_t index = offset >> PAGE_SHIFT;
4760 unsigned pg_offset = offset_in_page(offset);
4763 key.objectid = sctx->cur_ino;
4764 key.type = BTRFS_INODE_ITEM_KEY;
4767 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4769 return PTR_ERR(inode);
4771 if (offset + len > i_size_read(inode)) {
4772 if (offset > i_size_read(inode))
4775 len = offset - i_size_read(inode);
4780 last_index = (offset + len - 1) >> PAGE_SHIFT;
4782 /* initial readahead */
4783 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4784 file_ra_state_init(&sctx->ra, inode->i_mapping);
4786 while (index <= last_index) {
4787 unsigned cur_len = min_t(unsigned, len,
4788 PAGE_SIZE - pg_offset);
4790 page = find_lock_page(inode->i_mapping, index);
4792 page_cache_sync_readahead(inode->i_mapping, &sctx->ra,
4793 NULL, index, last_index + 1 - index);
4795 page = find_or_create_page(inode->i_mapping, index,
4803 if (PageReadahead(page)) {
4804 page_cache_async_readahead(inode->i_mapping, &sctx->ra,
4805 NULL, page, index, last_index + 1 - index);
4808 if (!PageUptodate(page)) {
4809 btrfs_readpage(NULL, page);
4811 if (!PageUptodate(page)) {
4820 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4835 * Read some bytes from the current inode/file and send a write command to
4838 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4840 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4843 ssize_t num_read = 0;
4845 p = fs_path_alloc();
4849 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4851 num_read = fill_read_buf(sctx, offset, len);
4852 if (num_read <= 0) {
4858 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4862 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4866 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4867 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4868 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4870 ret = send_cmd(sctx);
4881 * Send a clone command to user space.
4883 static int send_clone(struct send_ctx *sctx,
4884 u64 offset, u32 len,
4885 struct clone_root *clone_root)
4891 btrfs_debug(sctx->send_root->fs_info,
4892 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4893 offset, len, clone_root->root->root_key.objectid,
4894 clone_root->ino, clone_root->offset);
4896 p = fs_path_alloc();
4900 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4904 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4908 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4909 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4910 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4912 if (clone_root->root == sctx->send_root) {
4913 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4914 &gen, NULL, NULL, NULL, NULL);
4917 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4919 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4925 * If the parent we're using has a received_uuid set then use that as
4926 * our clone source as that is what we will look for when doing a
4929 * This covers the case that we create a snapshot off of a received
4930 * subvolume and then use that as the parent and try to receive on a
4933 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4934 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4935 clone_root->root->root_item.received_uuid);
4937 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4938 clone_root->root->root_item.uuid);
4939 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4940 le64_to_cpu(clone_root->root->root_item.ctransid));
4941 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4942 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4943 clone_root->offset);
4945 ret = send_cmd(sctx);
4954 * Send an update extent command to user space.
4956 static int send_update_extent(struct send_ctx *sctx,
4957 u64 offset, u32 len)
4962 p = fs_path_alloc();
4966 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4970 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4974 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4975 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4976 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4978 ret = send_cmd(sctx);
4986 static int send_hole(struct send_ctx *sctx, u64 end)
4988 struct fs_path *p = NULL;
4989 u64 offset = sctx->cur_inode_last_extent;
4994 * A hole that starts at EOF or beyond it. Since we do not yet support
4995 * fallocate (for extent preallocation and hole punching), sending a
4996 * write of zeroes starting at EOF or beyond would later require issuing
4997 * a truncate operation which would undo the write and achieve nothing.
4999 if (offset >= sctx->cur_inode_size)
5003 * Don't go beyond the inode's i_size due to prealloc extents that start
5006 end = min_t(u64, end, sctx->cur_inode_size);
5008 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5009 return send_update_extent(sctx, offset, end - offset);
5011 p = fs_path_alloc();
5014 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5016 goto tlv_put_failure;
5017 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5018 while (offset < end) {
5019 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
5021 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5024 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5025 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5026 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
5027 ret = send_cmd(sctx);
5032 sctx->cur_inode_next_write_offset = offset;
5038 static int send_extent_data(struct send_ctx *sctx,
5044 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5045 return send_update_extent(sctx, offset, len);
5047 while (sent < len) {
5048 u64 size = len - sent;
5051 if (size > BTRFS_SEND_READ_SIZE)
5052 size = BTRFS_SEND_READ_SIZE;
5053 ret = send_write(sctx, offset + sent, size);
5063 static int clone_range(struct send_ctx *sctx,
5064 struct clone_root *clone_root,
5065 const u64 disk_byte,
5070 struct btrfs_path *path;
5071 struct btrfs_key key;
5073 u64 clone_src_i_size;
5076 * Prevent cloning from a zero offset with a length matching the sector
5077 * size because in some scenarios this will make the receiver fail.
5079 * For example, if in the source filesystem the extent at offset 0
5080 * has a length of sectorsize and it was written using direct IO, then
5081 * it can never be an inline extent (even if compression is enabled).
5082 * Then this extent can be cloned in the original filesystem to a non
5083 * zero file offset, but it may not be possible to clone in the
5084 * destination filesystem because it can be inlined due to compression
5085 * on the destination filesystem (as the receiver's write operations are
5086 * always done using buffered IO). The same happens when the original
5087 * filesystem does not have compression enabled but the destination
5090 if (clone_root->offset == 0 &&
5091 len == sctx->send_root->fs_info->sectorsize)
5092 return send_extent_data(sctx, offset, len);
5094 path = alloc_path_for_send();
5099 * There are inodes that have extents that lie behind its i_size. Don't
5100 * accept clones from these extents.
5102 ret = __get_inode_info(clone_root->root, path, clone_root->ino,
5103 &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
5104 btrfs_release_path(path);
5109 * We can't send a clone operation for the entire range if we find
5110 * extent items in the respective range in the source file that
5111 * refer to different extents or if we find holes.
5112 * So check for that and do a mix of clone and regular write/copy
5113 * operations if needed.
5117 * mkfs.btrfs -f /dev/sda
5118 * mount /dev/sda /mnt
5119 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
5120 * cp --reflink=always /mnt/foo /mnt/bar
5121 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
5122 * btrfs subvolume snapshot -r /mnt /mnt/snap
5124 * If when we send the snapshot and we are processing file bar (which
5125 * has a higher inode number than foo) we blindly send a clone operation
5126 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
5127 * a file bar that matches the content of file foo - iow, doesn't match
5128 * the content from bar in the original filesystem.
5130 key.objectid = clone_root->ino;
5131 key.type = BTRFS_EXTENT_DATA_KEY;
5132 key.offset = clone_root->offset;
5133 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
5136 if (ret > 0 && path->slots[0] > 0) {
5137 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
5138 if (key.objectid == clone_root->ino &&
5139 key.type == BTRFS_EXTENT_DATA_KEY)
5144 struct extent_buffer *leaf = path->nodes[0];
5145 int slot = path->slots[0];
5146 struct btrfs_file_extent_item *ei;
5150 u64 clone_data_offset;
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 if (key.offset >= clone_src_i_size)
5207 if (key.offset + ext_len > clone_src_i_size)
5208 ext_len = clone_src_i_size - key.offset;
5210 clone_data_offset = btrfs_file_extent_offset(leaf, ei);
5211 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
5212 clone_root->offset = key.offset;
5213 if (clone_data_offset < data_offset &&
5214 clone_data_offset + ext_len > data_offset) {
5217 extent_offset = data_offset - clone_data_offset;
5218 ext_len -= extent_offset;
5219 clone_data_offset += extent_offset;
5220 clone_root->offset += extent_offset;
5224 clone_len = min_t(u64, ext_len, len);
5226 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5227 clone_data_offset == data_offset) {
5228 const u64 src_end = clone_root->offset + clone_len;
5229 const u64 sectorsize = SZ_64K;
5232 * We can't clone the last block, when its size is not
5233 * sector size aligned, into the middle of a file. If we
5234 * do so, the receiver will get a failure (-EINVAL) when
5235 * trying to clone or will silently corrupt the data in
5236 * the destination file if it's on a kernel without the
5237 * fix introduced by commit ac765f83f1397646
5238 * ("Btrfs: fix data corruption due to cloning of eof
5241 * So issue a clone of the aligned down range plus a
5242 * regular write for the eof block, if we hit that case.
5244 * Also, we use the maximum possible sector size, 64K,
5245 * because we don't know what's the sector size of the
5246 * filesystem that receives the stream, so we have to
5247 * assume the largest possible sector size.
5249 if (src_end == clone_src_i_size &&
5250 !IS_ALIGNED(src_end, sectorsize) &&
5251 offset + clone_len < sctx->cur_inode_size) {
5254 slen = ALIGN_DOWN(src_end - clone_root->offset,
5257 ret = send_clone(sctx, offset, slen,
5262 ret = send_extent_data(sctx, offset + slen,
5265 ret = send_clone(sctx, offset, clone_len,
5269 ret = send_extent_data(sctx, offset, clone_len);
5278 offset += clone_len;
5279 clone_root->offset += clone_len;
5280 data_offset += clone_len;
5286 ret = send_extent_data(sctx, offset, len);
5290 btrfs_free_path(path);
5294 static int send_write_or_clone(struct send_ctx *sctx,
5295 struct btrfs_path *path,
5296 struct btrfs_key *key,
5297 struct clone_root *clone_root)
5300 struct btrfs_file_extent_item *ei;
5301 u64 offset = key->offset;
5304 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5306 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5307 struct btrfs_file_extent_item);
5308 type = btrfs_file_extent_type(path->nodes[0], ei);
5309 if (type == BTRFS_FILE_EXTENT_INLINE) {
5310 len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5312 * it is possible the inline item won't cover the whole page,
5313 * but there may be items after this page. Make
5314 * sure to send the whole thing
5316 len = PAGE_ALIGN(len);
5318 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5321 if (offset >= sctx->cur_inode_size) {
5325 if (offset + len > sctx->cur_inode_size)
5326 len = sctx->cur_inode_size - offset;
5332 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5336 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5337 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5338 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5341 ret = send_extent_data(sctx, offset, len);
5343 sctx->cur_inode_next_write_offset = offset + len;
5348 static int is_extent_unchanged(struct send_ctx *sctx,
5349 struct btrfs_path *left_path,
5350 struct btrfs_key *ekey)
5353 struct btrfs_key key;
5354 struct btrfs_path *path = NULL;
5355 struct extent_buffer *eb;
5357 struct btrfs_key found_key;
5358 struct btrfs_file_extent_item *ei;
5363 u64 left_offset_fixed;
5371 path = alloc_path_for_send();
5375 eb = left_path->nodes[0];
5376 slot = left_path->slots[0];
5377 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5378 left_type = btrfs_file_extent_type(eb, ei);
5380 if (left_type != BTRFS_FILE_EXTENT_REG) {
5384 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5385 left_len = btrfs_file_extent_num_bytes(eb, ei);
5386 left_offset = btrfs_file_extent_offset(eb, ei);
5387 left_gen = btrfs_file_extent_generation(eb, ei);
5390 * Following comments will refer to these graphics. L is the left
5391 * extents which we are checking at the moment. 1-8 are the right
5392 * extents that we iterate.
5395 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5398 * |--1--|-2b-|...(same as above)
5400 * Alternative situation. Happens on files where extents got split.
5402 * |-----------7-----------|-6-|
5404 * Alternative situation. Happens on files which got larger.
5407 * Nothing follows after 8.
5410 key.objectid = ekey->objectid;
5411 key.type = BTRFS_EXTENT_DATA_KEY;
5412 key.offset = ekey->offset;
5413 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5422 * Handle special case where the right side has no extents at all.
5424 eb = path->nodes[0];
5425 slot = path->slots[0];
5426 btrfs_item_key_to_cpu(eb, &found_key, slot);
5427 if (found_key.objectid != key.objectid ||
5428 found_key.type != key.type) {
5429 /* If we're a hole then just pretend nothing changed */
5430 ret = (left_disknr) ? 0 : 1;
5435 * We're now on 2a, 2b or 7.
5438 while (key.offset < ekey->offset + left_len) {
5439 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5440 right_type = btrfs_file_extent_type(eb, ei);
5441 if (right_type != BTRFS_FILE_EXTENT_REG &&
5442 right_type != BTRFS_FILE_EXTENT_INLINE) {
5447 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5448 right_len = btrfs_file_extent_ram_bytes(eb, ei);
5449 right_len = PAGE_ALIGN(right_len);
5451 right_len = btrfs_file_extent_num_bytes(eb, ei);
5455 * Are we at extent 8? If yes, we know the extent is changed.
5456 * This may only happen on the first iteration.
5458 if (found_key.offset + right_len <= ekey->offset) {
5459 /* If we're a hole just pretend nothing changed */
5460 ret = (left_disknr) ? 0 : 1;
5465 * We just wanted to see if when we have an inline extent, what
5466 * follows it is a regular extent (wanted to check the above
5467 * condition for inline extents too). This should normally not
5468 * happen but it's possible for example when we have an inline
5469 * compressed extent representing data with a size matching
5470 * the page size (currently the same as sector size).
5472 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5477 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5478 right_offset = btrfs_file_extent_offset(eb, ei);
5479 right_gen = btrfs_file_extent_generation(eb, ei);
5481 left_offset_fixed = left_offset;
5482 if (key.offset < ekey->offset) {
5483 /* Fix the right offset for 2a and 7. */
5484 right_offset += ekey->offset - key.offset;
5486 /* Fix the left offset for all behind 2a and 2b */
5487 left_offset_fixed += key.offset - ekey->offset;
5491 * Check if we have the same extent.
5493 if (left_disknr != right_disknr ||
5494 left_offset_fixed != right_offset ||
5495 left_gen != right_gen) {
5501 * Go to the next extent.
5503 ret = btrfs_next_item(sctx->parent_root, path);
5507 eb = path->nodes[0];
5508 slot = path->slots[0];
5509 btrfs_item_key_to_cpu(eb, &found_key, slot);
5511 if (ret || found_key.objectid != key.objectid ||
5512 found_key.type != key.type) {
5513 key.offset += right_len;
5516 if (found_key.offset != key.offset + right_len) {
5524 * We're now behind the left extent (treat as unchanged) or at the end
5525 * of the right side (treat as changed).
5527 if (key.offset >= ekey->offset + left_len)
5534 btrfs_free_path(path);
5538 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5540 struct btrfs_path *path;
5541 struct btrfs_root *root = sctx->send_root;
5542 struct btrfs_file_extent_item *fi;
5543 struct btrfs_key key;
5548 path = alloc_path_for_send();
5552 sctx->cur_inode_last_extent = 0;
5554 key.objectid = sctx->cur_ino;
5555 key.type = BTRFS_EXTENT_DATA_KEY;
5556 key.offset = offset;
5557 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5561 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5562 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5565 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5566 struct btrfs_file_extent_item);
5567 type = btrfs_file_extent_type(path->nodes[0], fi);
5568 if (type == BTRFS_FILE_EXTENT_INLINE) {
5569 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5570 extent_end = ALIGN(key.offset + size,
5571 sctx->send_root->fs_info->sectorsize);
5573 extent_end = key.offset +
5574 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5576 sctx->cur_inode_last_extent = extent_end;
5578 btrfs_free_path(path);
5582 static int range_is_hole_in_parent(struct send_ctx *sctx,
5586 struct btrfs_path *path;
5587 struct btrfs_key key;
5588 struct btrfs_root *root = sctx->parent_root;
5589 u64 search_start = start;
5592 path = alloc_path_for_send();
5596 key.objectid = sctx->cur_ino;
5597 key.type = BTRFS_EXTENT_DATA_KEY;
5598 key.offset = search_start;
5599 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5602 if (ret > 0 && path->slots[0] > 0)
5605 while (search_start < end) {
5606 struct extent_buffer *leaf = path->nodes[0];
5607 int slot = path->slots[0];
5608 struct btrfs_file_extent_item *fi;
5611 if (slot >= btrfs_header_nritems(leaf)) {
5612 ret = btrfs_next_leaf(root, path);
5620 btrfs_item_key_to_cpu(leaf, &key, slot);
5621 if (key.objectid < sctx->cur_ino ||
5622 key.type < BTRFS_EXTENT_DATA_KEY)
5624 if (key.objectid > sctx->cur_ino ||
5625 key.type > BTRFS_EXTENT_DATA_KEY ||
5629 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5630 if (btrfs_file_extent_type(leaf, fi) ==
5631 BTRFS_FILE_EXTENT_INLINE) {
5632 u64 size = btrfs_file_extent_ram_bytes(leaf, fi);
5634 extent_end = ALIGN(key.offset + size,
5635 root->fs_info->sectorsize);
5637 extent_end = key.offset +
5638 btrfs_file_extent_num_bytes(leaf, fi);
5640 if (extent_end <= start)
5642 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
5643 search_start = extent_end;
5653 btrfs_free_path(path);
5657 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5658 struct btrfs_key *key)
5660 struct btrfs_file_extent_item *fi;
5665 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5668 if (sctx->cur_inode_last_extent == (u64)-1) {
5669 ret = get_last_extent(sctx, key->offset - 1);
5674 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5675 struct btrfs_file_extent_item);
5676 type = btrfs_file_extent_type(path->nodes[0], fi);
5677 if (type == BTRFS_FILE_EXTENT_INLINE) {
5678 u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5679 extent_end = ALIGN(key->offset + size,
5680 sctx->send_root->fs_info->sectorsize);
5682 extent_end = key->offset +
5683 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5686 if (path->slots[0] == 0 &&
5687 sctx->cur_inode_last_extent < key->offset) {
5689 * We might have skipped entire leafs that contained only
5690 * file extent items for our current inode. These leafs have
5691 * a generation number smaller (older) than the one in the
5692 * current leaf and the leaf our last extent came from, and
5693 * are located between these 2 leafs.
5695 ret = get_last_extent(sctx, key->offset - 1);
5700 if (sctx->cur_inode_last_extent < key->offset) {
5701 ret = range_is_hole_in_parent(sctx,
5702 sctx->cur_inode_last_extent,
5707 ret = send_hole(sctx, key->offset);
5711 sctx->cur_inode_last_extent = extent_end;
5715 static int process_extent(struct send_ctx *sctx,
5716 struct btrfs_path *path,
5717 struct btrfs_key *key)
5719 struct clone_root *found_clone = NULL;
5722 if (S_ISLNK(sctx->cur_inode_mode))
5725 if (sctx->parent_root && !sctx->cur_inode_new) {
5726 ret = is_extent_unchanged(sctx, path, key);
5734 struct btrfs_file_extent_item *ei;
5737 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5738 struct btrfs_file_extent_item);
5739 type = btrfs_file_extent_type(path->nodes[0], ei);
5740 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5741 type == BTRFS_FILE_EXTENT_REG) {
5743 * The send spec does not have a prealloc command yet,
5744 * so just leave a hole for prealloc'ed extents until
5745 * we have enough commands queued up to justify rev'ing
5748 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5753 /* Have a hole, just skip it. */
5754 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5761 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5762 sctx->cur_inode_size, &found_clone);
5763 if (ret != -ENOENT && ret < 0)
5766 ret = send_write_or_clone(sctx, path, key, found_clone);
5770 ret = maybe_send_hole(sctx, path, key);
5775 static int process_all_extents(struct send_ctx *sctx)
5778 struct btrfs_root *root;
5779 struct btrfs_path *path;
5780 struct btrfs_key key;
5781 struct btrfs_key found_key;
5782 struct extent_buffer *eb;
5785 root = sctx->send_root;
5786 path = alloc_path_for_send();
5790 key.objectid = sctx->cmp_key->objectid;
5791 key.type = BTRFS_EXTENT_DATA_KEY;
5793 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5798 eb = path->nodes[0];
5799 slot = path->slots[0];
5801 if (slot >= btrfs_header_nritems(eb)) {
5802 ret = btrfs_next_leaf(root, path);
5805 } else if (ret > 0) {
5812 btrfs_item_key_to_cpu(eb, &found_key, slot);
5814 if (found_key.objectid != key.objectid ||
5815 found_key.type != key.type) {
5820 ret = process_extent(sctx, path, &found_key);
5828 btrfs_free_path(path);
5832 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5834 int *refs_processed)
5838 if (sctx->cur_ino == 0)
5840 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5841 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5843 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5846 ret = process_recorded_refs(sctx, pending_move);
5850 *refs_processed = 1;
5855 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5866 int need_truncate = 1;
5867 int pending_move = 0;
5868 int refs_processed = 0;
5870 if (sctx->ignore_cur_inode)
5873 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5879 * We have processed the refs and thus need to advance send_progress.
5880 * Now, calls to get_cur_xxx will take the updated refs of the current
5881 * inode into account.
5883 * On the other hand, if our current inode is a directory and couldn't
5884 * be moved/renamed because its parent was renamed/moved too and it has
5885 * a higher inode number, we can only move/rename our current inode
5886 * after we moved/renamed its parent. Therefore in this case operate on
5887 * the old path (pre move/rename) of our current inode, and the
5888 * move/rename will be performed later.
5890 if (refs_processed && !pending_move)
5891 sctx->send_progress = sctx->cur_ino + 1;
5893 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5895 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5898 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5899 &left_mode, &left_uid, &left_gid, NULL);
5903 if (!sctx->parent_root || sctx->cur_inode_new) {
5905 if (!S_ISLNK(sctx->cur_inode_mode))
5907 if (sctx->cur_inode_next_write_offset == sctx->cur_inode_size)
5912 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5913 &old_size, NULL, &right_mode, &right_uid,
5918 if (left_uid != right_uid || left_gid != right_gid)
5920 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5922 if ((old_size == sctx->cur_inode_size) ||
5923 (sctx->cur_inode_size > old_size &&
5924 sctx->cur_inode_next_write_offset == sctx->cur_inode_size))
5928 if (S_ISREG(sctx->cur_inode_mode)) {
5929 if (need_send_hole(sctx)) {
5930 if (sctx->cur_inode_last_extent == (u64)-1 ||
5931 sctx->cur_inode_last_extent <
5932 sctx->cur_inode_size) {
5933 ret = get_last_extent(sctx, (u64)-1);
5937 if (sctx->cur_inode_last_extent <
5938 sctx->cur_inode_size) {
5939 ret = send_hole(sctx, sctx->cur_inode_size);
5944 if (need_truncate) {
5945 ret = send_truncate(sctx, sctx->cur_ino,
5946 sctx->cur_inode_gen,
5947 sctx->cur_inode_size);
5954 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5955 left_uid, left_gid);
5960 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5967 * If other directory inodes depended on our current directory
5968 * inode's move/rename, now do their move/rename operations.
5970 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5971 ret = apply_children_dir_moves(sctx);
5975 * Need to send that every time, no matter if it actually
5976 * changed between the two trees as we have done changes to
5977 * the inode before. If our inode is a directory and it's
5978 * waiting to be moved/renamed, we will send its utimes when
5979 * it's moved/renamed, therefore we don't need to do it here.
5981 sctx->send_progress = sctx->cur_ino + 1;
5982 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5991 struct parent_paths_ctx {
5992 struct list_head *refs;
5993 struct send_ctx *sctx;
5996 static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name,
5999 struct parent_paths_ctx *ppctx = ctx;
6001 return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx,
6006 * Issue unlink operations for all paths of the current inode found in the
6009 static int btrfs_unlink_all_paths(struct send_ctx *sctx)
6011 LIST_HEAD(deleted_refs);
6012 struct btrfs_path *path;
6013 struct btrfs_key key;
6014 struct parent_paths_ctx ctx;
6017 path = alloc_path_for_send();
6021 key.objectid = sctx->cur_ino;
6022 key.type = BTRFS_INODE_REF_KEY;
6024 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
6028 ctx.refs = &deleted_refs;
6032 struct extent_buffer *eb = path->nodes[0];
6033 int slot = path->slots[0];
6035 if (slot >= btrfs_header_nritems(eb)) {
6036 ret = btrfs_next_leaf(sctx->parent_root, path);
6044 btrfs_item_key_to_cpu(eb, &key, slot);
6045 if (key.objectid != sctx->cur_ino)
6047 if (key.type != BTRFS_INODE_REF_KEY &&
6048 key.type != BTRFS_INODE_EXTREF_KEY)
6051 ret = iterate_inode_ref(sctx->parent_root, path, &key, 1,
6052 record_parent_ref, &ctx);
6059 while (!list_empty(&deleted_refs)) {
6060 struct recorded_ref *ref;
6062 ref = list_first_entry(&deleted_refs, struct recorded_ref, list);
6063 ret = send_unlink(sctx, ref->full_path);
6066 fs_path_free(ref->full_path);
6067 list_del(&ref->list);
6072 btrfs_free_path(path);
6074 __free_recorded_refs(&deleted_refs);
6078 static int changed_inode(struct send_ctx *sctx,
6079 enum btrfs_compare_tree_result result)
6082 struct btrfs_key *key = sctx->cmp_key;
6083 struct btrfs_inode_item *left_ii = NULL;
6084 struct btrfs_inode_item *right_ii = NULL;
6088 sctx->cur_ino = key->objectid;
6089 sctx->cur_inode_new_gen = 0;
6090 sctx->cur_inode_last_extent = (u64)-1;
6091 sctx->cur_inode_next_write_offset = 0;
6092 sctx->ignore_cur_inode = false;
6095 * Set send_progress to current inode. This will tell all get_cur_xxx
6096 * functions that the current inode's refs are not updated yet. Later,
6097 * when process_recorded_refs is finished, it is set to cur_ino + 1.
6099 sctx->send_progress = sctx->cur_ino;
6101 if (result == BTRFS_COMPARE_TREE_NEW ||
6102 result == BTRFS_COMPARE_TREE_CHANGED) {
6103 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
6104 sctx->left_path->slots[0],
6105 struct btrfs_inode_item);
6106 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
6109 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6110 sctx->right_path->slots[0],
6111 struct btrfs_inode_item);
6112 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6115 if (result == BTRFS_COMPARE_TREE_CHANGED) {
6116 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
6117 sctx->right_path->slots[0],
6118 struct btrfs_inode_item);
6120 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
6124 * The cur_ino = root dir case is special here. We can't treat
6125 * the inode as deleted+reused because it would generate a
6126 * stream that tries to delete/mkdir the root dir.
6128 if (left_gen != right_gen &&
6129 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6130 sctx->cur_inode_new_gen = 1;
6134 * Normally we do not find inodes with a link count of zero (orphans)
6135 * because the most common case is to create a snapshot and use it
6136 * for a send operation. However other less common use cases involve
6137 * using a subvolume and send it after turning it to RO mode just
6138 * after deleting all hard links of a file while holding an open
6139 * file descriptor against it or turning a RO snapshot into RW mode,
6140 * keep an open file descriptor against a file, delete it and then
6141 * turn the snapshot back to RO mode before using it for a send
6142 * operation. So if we find such cases, ignore the inode and all its
6143 * items completely if it's a new inode, or if it's a changed inode
6144 * make sure all its previous paths (from the parent snapshot) are all
6145 * unlinked and all other the inode items are ignored.
6147 if (result == BTRFS_COMPARE_TREE_NEW ||
6148 result == BTRFS_COMPARE_TREE_CHANGED) {
6151 nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii);
6153 sctx->ignore_cur_inode = true;
6154 if (result == BTRFS_COMPARE_TREE_CHANGED)
6155 ret = btrfs_unlink_all_paths(sctx);
6160 if (result == BTRFS_COMPARE_TREE_NEW) {
6161 sctx->cur_inode_gen = left_gen;
6162 sctx->cur_inode_new = 1;
6163 sctx->cur_inode_deleted = 0;
6164 sctx->cur_inode_size = btrfs_inode_size(
6165 sctx->left_path->nodes[0], left_ii);
6166 sctx->cur_inode_mode = btrfs_inode_mode(
6167 sctx->left_path->nodes[0], left_ii);
6168 sctx->cur_inode_rdev = btrfs_inode_rdev(
6169 sctx->left_path->nodes[0], left_ii);
6170 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
6171 ret = send_create_inode_if_needed(sctx);
6172 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
6173 sctx->cur_inode_gen = right_gen;
6174 sctx->cur_inode_new = 0;
6175 sctx->cur_inode_deleted = 1;
6176 sctx->cur_inode_size = btrfs_inode_size(
6177 sctx->right_path->nodes[0], right_ii);
6178 sctx->cur_inode_mode = btrfs_inode_mode(
6179 sctx->right_path->nodes[0], right_ii);
6180 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
6182 * We need to do some special handling in case the inode was
6183 * reported as changed with a changed generation number. This
6184 * means that the original inode was deleted and new inode
6185 * reused the same inum. So we have to treat the old inode as
6186 * deleted and the new one as new.
6188 if (sctx->cur_inode_new_gen) {
6190 * First, process the inode as if it was deleted.
6192 sctx->cur_inode_gen = right_gen;
6193 sctx->cur_inode_new = 0;
6194 sctx->cur_inode_deleted = 1;
6195 sctx->cur_inode_size = btrfs_inode_size(
6196 sctx->right_path->nodes[0], right_ii);
6197 sctx->cur_inode_mode = btrfs_inode_mode(
6198 sctx->right_path->nodes[0], right_ii);
6199 ret = process_all_refs(sctx,
6200 BTRFS_COMPARE_TREE_DELETED);
6205 * Now process the inode as if it was new.
6207 sctx->cur_inode_gen = left_gen;
6208 sctx->cur_inode_new = 1;
6209 sctx->cur_inode_deleted = 0;
6210 sctx->cur_inode_size = btrfs_inode_size(
6211 sctx->left_path->nodes[0], left_ii);
6212 sctx->cur_inode_mode = btrfs_inode_mode(
6213 sctx->left_path->nodes[0], left_ii);
6214 sctx->cur_inode_rdev = btrfs_inode_rdev(
6215 sctx->left_path->nodes[0], left_ii);
6216 ret = send_create_inode_if_needed(sctx);
6220 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
6224 * Advance send_progress now as we did not get into
6225 * process_recorded_refs_if_needed in the new_gen case.
6227 sctx->send_progress = sctx->cur_ino + 1;
6230 * Now process all extents and xattrs of the inode as if
6231 * they were all new.
6233 ret = process_all_extents(sctx);
6236 ret = process_all_new_xattrs(sctx);
6240 sctx->cur_inode_gen = left_gen;
6241 sctx->cur_inode_new = 0;
6242 sctx->cur_inode_new_gen = 0;
6243 sctx->cur_inode_deleted = 0;
6244 sctx->cur_inode_size = btrfs_inode_size(
6245 sctx->left_path->nodes[0], left_ii);
6246 sctx->cur_inode_mode = btrfs_inode_mode(
6247 sctx->left_path->nodes[0], left_ii);
6256 * We have to process new refs before deleted refs, but compare_trees gives us
6257 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
6258 * first and later process them in process_recorded_refs.
6259 * For the cur_inode_new_gen case, we skip recording completely because
6260 * changed_inode did already initiate processing of refs. The reason for this is
6261 * that in this case, compare_tree actually compares the refs of 2 different
6262 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
6263 * refs of the right tree as deleted and all refs of the left tree as new.
6265 static int changed_ref(struct send_ctx *sctx,
6266 enum btrfs_compare_tree_result result)
6270 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6271 inconsistent_snapshot_error(sctx, result, "reference");
6275 if (!sctx->cur_inode_new_gen &&
6276 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
6277 if (result == BTRFS_COMPARE_TREE_NEW)
6278 ret = record_new_ref(sctx);
6279 else if (result == BTRFS_COMPARE_TREE_DELETED)
6280 ret = record_deleted_ref(sctx);
6281 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6282 ret = record_changed_ref(sctx);
6289 * Process new/deleted/changed xattrs. We skip processing in the
6290 * cur_inode_new_gen case because changed_inode did already initiate processing
6291 * of xattrs. The reason is the same as in changed_ref
6293 static int changed_xattr(struct send_ctx *sctx,
6294 enum btrfs_compare_tree_result result)
6298 if (sctx->cur_ino != sctx->cmp_key->objectid) {
6299 inconsistent_snapshot_error(sctx, result, "xattr");
6303 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6304 if (result == BTRFS_COMPARE_TREE_NEW)
6305 ret = process_new_xattr(sctx);
6306 else if (result == BTRFS_COMPARE_TREE_DELETED)
6307 ret = process_deleted_xattr(sctx);
6308 else if (result == BTRFS_COMPARE_TREE_CHANGED)
6309 ret = process_changed_xattr(sctx);
6316 * Process new/deleted/changed extents. We skip processing in the
6317 * cur_inode_new_gen case because changed_inode did already initiate processing
6318 * of extents. The reason is the same as in changed_ref
6320 static int changed_extent(struct send_ctx *sctx,
6321 enum btrfs_compare_tree_result result)
6326 * We have found an extent item that changed without the inode item
6327 * having changed. This can happen either after relocation (where the
6328 * disk_bytenr of an extent item is replaced at
6329 * relocation.c:replace_file_extents()) or after deduplication into a
6330 * file in both the parent and send snapshots (where an extent item can
6331 * get modified or replaced with a new one). Note that deduplication
6332 * updates the inode item, but it only changes the iversion (sequence
6333 * field in the inode item) of the inode, so if a file is deduplicated
6334 * the same amount of times in both the parent and send snapshots, its
6335 * iversion becames the same in both snapshots, whence the inode item is
6336 * the same on both snapshots.
6338 if (sctx->cur_ino != sctx->cmp_key->objectid)
6341 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
6342 if (result != BTRFS_COMPARE_TREE_DELETED)
6343 ret = process_extent(sctx, sctx->left_path,
6350 static int dir_changed(struct send_ctx *sctx, u64 dir)
6352 u64 orig_gen, new_gen;
6355 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
6360 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
6365 return (orig_gen != new_gen) ? 1 : 0;
6368 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
6369 struct btrfs_key *key)
6371 struct btrfs_inode_extref *extref;
6372 struct extent_buffer *leaf;
6373 u64 dirid = 0, last_dirid = 0;
6380 /* Easy case, just check this one dirid */
6381 if (key->type == BTRFS_INODE_REF_KEY) {
6382 dirid = key->offset;
6384 ret = dir_changed(sctx, dirid);
6388 leaf = path->nodes[0];
6389 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
6390 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
6391 while (cur_offset < item_size) {
6392 extref = (struct btrfs_inode_extref *)(ptr +
6394 dirid = btrfs_inode_extref_parent(leaf, extref);
6395 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
6396 cur_offset += ref_name_len + sizeof(*extref);
6397 if (dirid == last_dirid)
6399 ret = dir_changed(sctx, dirid);
6409 * Updates compare related fields in sctx and simply forwards to the actual
6410 * changed_xxx functions.
6412 static int changed_cb(struct btrfs_path *left_path,
6413 struct btrfs_path *right_path,
6414 struct btrfs_key *key,
6415 enum btrfs_compare_tree_result result,
6419 struct send_ctx *sctx = ctx;
6421 if (result == BTRFS_COMPARE_TREE_SAME) {
6422 if (key->type == BTRFS_INODE_REF_KEY ||
6423 key->type == BTRFS_INODE_EXTREF_KEY) {
6424 ret = compare_refs(sctx, left_path, key);
6429 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
6430 return maybe_send_hole(sctx, left_path, key);
6434 result = BTRFS_COMPARE_TREE_CHANGED;
6438 sctx->left_path = left_path;
6439 sctx->right_path = right_path;
6440 sctx->cmp_key = key;
6442 ret = finish_inode_if_needed(sctx, 0);
6446 /* Ignore non-FS objects */
6447 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
6448 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
6451 if (key->type == BTRFS_INODE_ITEM_KEY) {
6452 ret = changed_inode(sctx, result);
6453 } else if (!sctx->ignore_cur_inode) {
6454 if (key->type == BTRFS_INODE_REF_KEY ||
6455 key->type == BTRFS_INODE_EXTREF_KEY)
6456 ret = changed_ref(sctx, result);
6457 else if (key->type == BTRFS_XATTR_ITEM_KEY)
6458 ret = changed_xattr(sctx, result);
6459 else if (key->type == BTRFS_EXTENT_DATA_KEY)
6460 ret = changed_extent(sctx, result);
6467 static int full_send_tree(struct send_ctx *sctx)
6470 struct btrfs_root *send_root = sctx->send_root;
6471 struct btrfs_key key;
6472 struct btrfs_path *path;
6473 struct extent_buffer *eb;
6476 path = alloc_path_for_send();
6480 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6481 key.type = BTRFS_INODE_ITEM_KEY;
6484 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6491 eb = path->nodes[0];
6492 slot = path->slots[0];
6493 btrfs_item_key_to_cpu(eb, &key, slot);
6495 ret = changed_cb(path, NULL, &key,
6496 BTRFS_COMPARE_TREE_NEW, sctx);
6500 ret = btrfs_next_item(send_root, path);
6510 ret = finish_inode_if_needed(sctx, 1);
6513 btrfs_free_path(path);
6517 static int send_subvol(struct send_ctx *sctx)
6521 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6522 ret = send_header(sctx);
6527 ret = send_subvol_begin(sctx);
6531 if (sctx->parent_root) {
6532 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6536 ret = finish_inode_if_needed(sctx, 1);
6540 ret = full_send_tree(sctx);
6546 free_recorded_refs(sctx);
6551 * If orphan cleanup did remove any orphans from a root, it means the tree
6552 * was modified and therefore the commit root is not the same as the current
6553 * root anymore. This is a problem, because send uses the commit root and
6554 * therefore can see inode items that don't exist in the current root anymore,
6555 * and for example make calls to btrfs_iget, which will do tree lookups based
6556 * on the current root and not on the commit root. Those lookups will fail,
6557 * returning a -ESTALE error, and making send fail with that error. So make
6558 * sure a send does not see any orphans we have just removed, and that it will
6559 * see the same inodes regardless of whether a transaction commit happened
6560 * before it started (meaning that the commit root will be the same as the
6561 * current root) or not.
6563 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6566 struct btrfs_trans_handle *trans = NULL;
6569 if (sctx->parent_root &&
6570 sctx->parent_root->node != sctx->parent_root->commit_root)
6573 for (i = 0; i < sctx->clone_roots_cnt; i++)
6574 if (sctx->clone_roots[i].root->node !=
6575 sctx->clone_roots[i].root->commit_root)
6579 return btrfs_end_transaction(trans);
6584 /* Use any root, all fs roots will get their commit roots updated. */
6586 trans = btrfs_join_transaction(sctx->send_root);
6588 return PTR_ERR(trans);
6592 return btrfs_commit_transaction(trans);
6596 * Make sure any existing dellaloc is flushed for any root used by a send
6597 * operation so that we do not miss any data and we do not race with writeback
6598 * finishing and changing a tree while send is using the tree. This could
6599 * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and
6600 * a send operation then uses the subvolume.
6601 * After flushing delalloc ensure_commit_roots_uptodate() must be called.
6603 static int flush_delalloc_roots(struct send_ctx *sctx)
6605 struct btrfs_root *root = sctx->parent_root;
6610 ret = btrfs_start_delalloc_snapshot(root);
6613 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
6616 for (i = 0; i < sctx->clone_roots_cnt; i++) {
6617 root = sctx->clone_roots[i].root;
6618 ret = btrfs_start_delalloc_snapshot(root);
6621 btrfs_wait_ordered_extents(root, U64_MAX, 0, U64_MAX);
6627 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6629 spin_lock(&root->root_item_lock);
6630 root->send_in_progress--;
6632 * Not much left to do, we don't know why it's unbalanced and
6633 * can't blindly reset it to 0.
6635 if (root->send_in_progress < 0)
6636 btrfs_err(root->fs_info,
6637 "send_in_progress unbalanced %d root %llu",
6638 root->send_in_progress, root->root_key.objectid);
6639 spin_unlock(&root->root_item_lock);
6642 static void dedupe_in_progress_warn(const struct btrfs_root *root)
6644 btrfs_warn_rl(root->fs_info,
6645 "cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
6646 root->root_key.objectid, root->dedupe_in_progress);
6649 long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg)
6652 struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
6653 struct btrfs_fs_info *fs_info = send_root->fs_info;
6654 struct btrfs_root *clone_root;
6655 struct btrfs_key key;
6656 struct send_ctx *sctx = NULL;
6658 u64 *clone_sources_tmp = NULL;
6659 int clone_sources_to_rollback = 0;
6660 unsigned alloc_size;
6661 int sort_clone_roots = 0;
6664 if (!capable(CAP_SYS_ADMIN))
6668 * The subvolume must remain read-only during send, protect against
6669 * making it RW. This also protects against deletion.
6671 spin_lock(&send_root->root_item_lock);
6672 if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
6673 dedupe_in_progress_warn(send_root);
6674 spin_unlock(&send_root->root_item_lock);
6677 send_root->send_in_progress++;
6678 spin_unlock(&send_root->root_item_lock);
6681 * This is done when we lookup the root, it should already be complete
6682 * by the time we get here.
6684 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6687 * Userspace tools do the checks and warn the user if it's
6690 if (!btrfs_root_readonly(send_root)) {
6696 * Check that we don't overflow at later allocations, we request
6697 * clone_sources_count + 1 items, and compare to unsigned long inside
6700 if (arg->clone_sources_count >
6701 ULONG_MAX / sizeof(struct clone_root) - 1) {
6706 if (!access_ok(arg->clone_sources,
6707 sizeof(*arg->clone_sources) *
6708 arg->clone_sources_count)) {
6713 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6718 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
6724 INIT_LIST_HEAD(&sctx->new_refs);
6725 INIT_LIST_HEAD(&sctx->deleted_refs);
6726 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
6727 INIT_LIST_HEAD(&sctx->name_cache_list);
6729 sctx->flags = arg->flags;
6731 sctx->send_filp = fget(arg->send_fd);
6732 if (!sctx->send_filp) {
6737 sctx->send_root = send_root;
6739 * Unlikely but possible, if the subvolume is marked for deletion but
6740 * is slow to remove the directory entry, send can still be started
6742 if (btrfs_root_dead(sctx->send_root)) {
6747 sctx->clone_roots_cnt = arg->clone_sources_count;
6749 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6750 sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL);
6751 if (!sctx->send_buf) {
6756 sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
6757 if (!sctx->read_buf) {
6762 sctx->pending_dir_moves = RB_ROOT;
6763 sctx->waiting_dir_moves = RB_ROOT;
6764 sctx->orphan_dirs = RB_ROOT;
6766 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6768 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL);
6769 if (!sctx->clone_roots) {
6774 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6776 if (arg->clone_sources_count) {
6777 clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
6778 if (!clone_sources_tmp) {
6783 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6790 for (i = 0; i < arg->clone_sources_count; i++) {
6791 key.objectid = clone_sources_tmp[i];
6792 key.type = BTRFS_ROOT_ITEM_KEY;
6793 key.offset = (u64)-1;
6795 index = srcu_read_lock(&fs_info->subvol_srcu);
6797 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6798 if (IS_ERR(clone_root)) {
6799 srcu_read_unlock(&fs_info->subvol_srcu, index);
6800 ret = PTR_ERR(clone_root);
6803 spin_lock(&clone_root->root_item_lock);
6804 if (!btrfs_root_readonly(clone_root) ||
6805 btrfs_root_dead(clone_root)) {
6806 spin_unlock(&clone_root->root_item_lock);
6807 srcu_read_unlock(&fs_info->subvol_srcu, index);
6811 if (clone_root->dedupe_in_progress) {
6812 dedupe_in_progress_warn(clone_root);
6813 spin_unlock(&clone_root->root_item_lock);
6814 srcu_read_unlock(&fs_info->subvol_srcu, index);
6818 clone_root->send_in_progress++;
6819 spin_unlock(&clone_root->root_item_lock);
6820 srcu_read_unlock(&fs_info->subvol_srcu, index);
6822 sctx->clone_roots[i].root = clone_root;
6823 clone_sources_to_rollback = i + 1;
6825 kvfree(clone_sources_tmp);
6826 clone_sources_tmp = NULL;
6829 if (arg->parent_root) {
6830 key.objectid = arg->parent_root;
6831 key.type = BTRFS_ROOT_ITEM_KEY;
6832 key.offset = (u64)-1;
6834 index = srcu_read_lock(&fs_info->subvol_srcu);
6836 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6837 if (IS_ERR(sctx->parent_root)) {
6838 srcu_read_unlock(&fs_info->subvol_srcu, index);
6839 ret = PTR_ERR(sctx->parent_root);
6843 spin_lock(&sctx->parent_root->root_item_lock);
6844 sctx->parent_root->send_in_progress++;
6845 if (!btrfs_root_readonly(sctx->parent_root) ||
6846 btrfs_root_dead(sctx->parent_root)) {
6847 spin_unlock(&sctx->parent_root->root_item_lock);
6848 srcu_read_unlock(&fs_info->subvol_srcu, index);
6852 if (sctx->parent_root->dedupe_in_progress) {
6853 dedupe_in_progress_warn(sctx->parent_root);
6854 spin_unlock(&sctx->parent_root->root_item_lock);
6855 srcu_read_unlock(&fs_info->subvol_srcu, index);
6859 spin_unlock(&sctx->parent_root->root_item_lock);
6861 srcu_read_unlock(&fs_info->subvol_srcu, index);
6865 * Clones from send_root are allowed, but only if the clone source
6866 * is behind the current send position. This is checked while searching
6867 * for possible clone sources.
6869 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6871 /* We do a bsearch later */
6872 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6873 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6875 sort_clone_roots = 1;
6877 ret = flush_delalloc_roots(sctx);
6881 ret = ensure_commit_roots_uptodate(sctx);
6885 mutex_lock(&fs_info->balance_mutex);
6886 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
6887 mutex_unlock(&fs_info->balance_mutex);
6888 btrfs_warn_rl(fs_info,
6889 "cannot run send because a balance operation is in progress");
6893 fs_info->send_in_progress++;
6894 mutex_unlock(&fs_info->balance_mutex);
6896 current->journal_info = BTRFS_SEND_TRANS_STUB;
6897 ret = send_subvol(sctx);
6898 current->journal_info = NULL;
6899 mutex_lock(&fs_info->balance_mutex);
6900 fs_info->send_in_progress--;
6901 mutex_unlock(&fs_info->balance_mutex);
6905 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6906 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6909 ret = send_cmd(sctx);
6915 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6916 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6918 struct pending_dir_move *pm;
6920 n = rb_first(&sctx->pending_dir_moves);
6921 pm = rb_entry(n, struct pending_dir_move, node);
6922 while (!list_empty(&pm->list)) {
6923 struct pending_dir_move *pm2;
6925 pm2 = list_first_entry(&pm->list,
6926 struct pending_dir_move, list);
6927 free_pending_move(sctx, pm2);
6929 free_pending_move(sctx, pm);
6932 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6933 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6935 struct waiting_dir_move *dm;
6937 n = rb_first(&sctx->waiting_dir_moves);
6938 dm = rb_entry(n, struct waiting_dir_move, node);
6939 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6943 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6944 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6946 struct orphan_dir_info *odi;
6948 n = rb_first(&sctx->orphan_dirs);
6949 odi = rb_entry(n, struct orphan_dir_info, node);
6950 free_orphan_dir_info(sctx, odi);
6953 if (sort_clone_roots) {
6954 for (i = 0; i < sctx->clone_roots_cnt; i++)
6955 btrfs_root_dec_send_in_progress(
6956 sctx->clone_roots[i].root);
6958 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6959 btrfs_root_dec_send_in_progress(
6960 sctx->clone_roots[i].root);
6962 btrfs_root_dec_send_in_progress(send_root);
6964 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6965 btrfs_root_dec_send_in_progress(sctx->parent_root);
6967 kvfree(clone_sources_tmp);
6970 if (sctx->send_filp)
6971 fput(sctx->send_filp);
6973 kvfree(sctx->clone_roots);
6974 kvfree(sctx->send_buf);
6975 kvfree(sctx->read_buf);
6977 name_cache_free(sctx);