2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
37 #include "compression.h"
40 * A fs_path is a helper to dynamically build path names with unknown size.
41 * It reallocates the internal buffer on demand.
42 * It allows fast adding of path elements on the right side (normal path) and
43 * fast adding to the left side (reversed path). A reversed path can also be
44 * unreversed if needed.
53 unsigned short buf_len:15;
54 unsigned short reversed:1;
58 * Average path length does not exceed 200 bytes, we'll have
59 * better packing in the slab and higher chance to satisfy
60 * a allocation later during send.
65 #define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
69 /* reused for each extent */
71 struct btrfs_root *root;
78 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
82 struct file *send_filp;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
113 u64 cur_inode_last_extent;
117 struct list_head new_refs;
118 struct list_head deleted_refs;
120 struct radix_tree_root name_cache;
121 struct list_head name_cache_list;
124 struct file_ra_state ra;
129 * We process inodes by their increasing order, so if before an
130 * incremental send we reverse the parent/child relationship of
131 * directories such that a directory with a lower inode number was
132 * the parent of a directory with a higher inode number, and the one
133 * becoming the new parent got renamed too, we can't rename/move the
134 * directory with lower inode number when we finish processing it - we
135 * must process the directory with higher inode number first, then
136 * rename/move it and then rename/move the directory with lower inode
137 * number. Example follows.
139 * Tree state when the first send was performed:
151 * Tree state when the second (incremental) send is performed:
160 * The sequence of steps that lead to the second state was:
162 * mv /a/b/c/d /a/b/c2/d2
163 * mv /a/b/c /a/b/c2/d2/cc
165 * "c" has lower inode number, but we can't move it (2nd mv operation)
166 * before we move "d", which has higher inode number.
168 * So we just memorize which move/rename operations must be performed
169 * later when their respective parent is processed and moved/renamed.
172 /* Indexed by parent directory inode number. */
173 struct rb_root pending_dir_moves;
176 * Reverse index, indexed by the inode number of a directory that
177 * is waiting for the move/rename of its immediate parent before its
178 * own move/rename can be performed.
180 struct rb_root waiting_dir_moves;
183 * A directory that is going to be rm'ed might have a child directory
184 * which is in the pending directory moves index above. In this case,
185 * the directory can only be removed after the move/rename of its child
186 * is performed. Example:
206 * Sequence of steps that lead to the send snapshot:
207 * rm -f /a/b/c/foo.txt
209 * mv /a/b/c/x /a/b/YY
212 * When the child is processed, its move/rename is delayed until its
213 * parent is processed (as explained above), but all other operations
214 * like update utimes, chown, chgrp, etc, are performed and the paths
215 * that it uses for those operations must use the orphanized name of
216 * its parent (the directory we're going to rm later), so we need to
217 * memorize that name.
219 * Indexed by the inode number of the directory to be deleted.
221 struct rb_root orphan_dirs;
224 struct pending_dir_move {
226 struct list_head list;
230 struct list_head update_refs;
233 struct waiting_dir_move {
237 * There might be some directory that could not be removed because it
238 * was waiting for this directory inode to be moved first. Therefore
239 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
245 struct orphan_dir_info {
251 struct name_cache_entry {
252 struct list_head list;
254 * radix_tree has only 32bit entries but we need to handle 64bit inums.
255 * We use the lower 32bit of the 64bit inum to store it in the tree. If
256 * more then one inum would fall into the same entry, we use radix_list
257 * to store the additional entries. radix_list is also used to store
258 * entries where two entries have the same inum but different
261 struct list_head radix_list;
267 int need_later_update;
272 static void inconsistent_snapshot_error(struct send_ctx *sctx,
273 enum btrfs_compare_tree_result result,
276 const char *result_string;
279 case BTRFS_COMPARE_TREE_NEW:
280 result_string = "new";
282 case BTRFS_COMPARE_TREE_DELETED:
283 result_string = "deleted";
285 case BTRFS_COMPARE_TREE_CHANGED:
286 result_string = "updated";
288 case BTRFS_COMPARE_TREE_SAME:
290 result_string = "unchanged";
294 result_string = "unexpected";
297 btrfs_err(sctx->send_root->fs_info,
298 "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu",
299 result_string, what, sctx->cmp_key->objectid,
300 sctx->send_root->root_key.objectid,
302 sctx->parent_root->root_key.objectid : 0));
305 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
307 static struct waiting_dir_move *
308 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
310 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
312 static int need_send_hole(struct send_ctx *sctx)
314 return (sctx->parent_root && !sctx->cur_inode_new &&
315 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
316 S_ISREG(sctx->cur_inode_mode));
319 static void fs_path_reset(struct fs_path *p)
322 p->start = p->buf + p->buf_len - 1;
332 static struct fs_path *fs_path_alloc(void)
336 p = kmalloc(sizeof(*p), GFP_KERNEL);
340 p->buf = p->inline_buf;
341 p->buf_len = FS_PATH_INLINE_SIZE;
346 static struct fs_path *fs_path_alloc_reversed(void)
358 static void fs_path_free(struct fs_path *p)
362 if (p->buf != p->inline_buf)
367 static int fs_path_len(struct fs_path *p)
369 return p->end - p->start;
372 static int fs_path_ensure_buf(struct fs_path *p, int len)
380 if (p->buf_len >= len)
383 if (len > PATH_MAX) {
388 path_len = p->end - p->start;
389 old_buf_len = p->buf_len;
392 * First time the inline_buf does not suffice
394 if (p->buf == p->inline_buf) {
395 tmp_buf = kmalloc(len, GFP_KERNEL);
397 memcpy(tmp_buf, p->buf, old_buf_len);
399 tmp_buf = krealloc(p->buf, len, GFP_KERNEL);
405 * The real size of the buffer is bigger, this will let the fast path
406 * happen most of the time
408 p->buf_len = ksize(p->buf);
411 tmp_buf = p->buf + old_buf_len - path_len - 1;
412 p->end = p->buf + p->buf_len - 1;
413 p->start = p->end - path_len;
414 memmove(p->start, tmp_buf, path_len + 1);
417 p->end = p->start + path_len;
422 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
428 new_len = p->end - p->start + name_len;
429 if (p->start != p->end)
431 ret = fs_path_ensure_buf(p, new_len);
436 if (p->start != p->end)
438 p->start -= name_len;
439 *prepared = p->start;
441 if (p->start != p->end)
452 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
457 ret = fs_path_prepare_for_add(p, name_len, &prepared);
460 memcpy(prepared, name, name_len);
466 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
471 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
474 memcpy(prepared, p2->start, p2->end - p2->start);
480 static int fs_path_add_from_extent_buffer(struct fs_path *p,
481 struct extent_buffer *eb,
482 unsigned long off, int len)
487 ret = fs_path_prepare_for_add(p, len, &prepared);
491 read_extent_buffer(eb, prepared, off, len);
497 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
501 p->reversed = from->reversed;
504 ret = fs_path_add_path(p, from);
510 static void fs_path_unreverse(struct fs_path *p)
519 len = p->end - p->start;
521 p->end = p->start + len;
522 memmove(p->start, tmp, len + 1);
526 static struct btrfs_path *alloc_path_for_send(void)
528 struct btrfs_path *path;
530 path = btrfs_alloc_path();
533 path->search_commit_root = 1;
534 path->skip_locking = 1;
535 path->need_commit_sem = 1;
539 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
549 ret = vfs_write(filp, (__force const char __user *)buf + pos,
551 /* TODO handle that correctly */
552 /*if (ret == -ERESTARTSYS) {
571 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
573 struct btrfs_tlv_header *hdr;
574 int total_len = sizeof(*hdr) + len;
575 int left = sctx->send_max_size - sctx->send_size;
577 if (unlikely(left < total_len))
580 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
581 hdr->tlv_type = cpu_to_le16(attr);
582 hdr->tlv_len = cpu_to_le16(len);
583 memcpy(hdr + 1, data, len);
584 sctx->send_size += total_len;
589 #define TLV_PUT_DEFINE_INT(bits) \
590 static int tlv_put_u##bits(struct send_ctx *sctx, \
591 u##bits attr, u##bits value) \
593 __le##bits __tmp = cpu_to_le##bits(value); \
594 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
597 TLV_PUT_DEFINE_INT(64)
599 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
600 const char *str, int len)
604 return tlv_put(sctx, attr, str, len);
607 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
610 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
613 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
614 struct extent_buffer *eb,
615 struct btrfs_timespec *ts)
617 struct btrfs_timespec bts;
618 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
619 return tlv_put(sctx, attr, &bts, sizeof(bts));
623 #define TLV_PUT(sctx, attrtype, attrlen, data) \
625 ret = tlv_put(sctx, attrtype, attrlen, data); \
627 goto tlv_put_failure; \
630 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
632 ret = tlv_put_u##bits(sctx, attrtype, value); \
634 goto tlv_put_failure; \
637 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
638 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
639 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
640 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
641 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
643 ret = tlv_put_string(sctx, attrtype, str, len); \
645 goto tlv_put_failure; \
647 #define TLV_PUT_PATH(sctx, attrtype, p) \
649 ret = tlv_put_string(sctx, attrtype, p->start, \
650 p->end - p->start); \
652 goto tlv_put_failure; \
654 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
656 ret = tlv_put_uuid(sctx, attrtype, uuid); \
658 goto tlv_put_failure; \
660 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
662 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
664 goto tlv_put_failure; \
667 static int send_header(struct send_ctx *sctx)
669 struct btrfs_stream_header hdr;
671 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
672 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
674 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
679 * For each command/item we want to send to userspace, we call this function.
681 static int begin_cmd(struct send_ctx *sctx, int cmd)
683 struct btrfs_cmd_header *hdr;
685 if (WARN_ON(!sctx->send_buf))
688 BUG_ON(sctx->send_size);
690 sctx->send_size += sizeof(*hdr);
691 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
692 hdr->cmd = cpu_to_le16(cmd);
697 static int send_cmd(struct send_ctx *sctx)
700 struct btrfs_cmd_header *hdr;
703 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
704 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
707 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
708 hdr->crc = cpu_to_le32(crc);
710 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
713 sctx->total_send_size += sctx->send_size;
714 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
721 * Sends a move instruction to user space
723 static int send_rename(struct send_ctx *sctx,
724 struct fs_path *from, struct fs_path *to)
726 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
729 btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start);
731 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
735 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
736 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
738 ret = send_cmd(sctx);
746 * Sends a link instruction to user space
748 static int send_link(struct send_ctx *sctx,
749 struct fs_path *path, struct fs_path *lnk)
751 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
754 btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start);
756 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
760 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
761 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
763 ret = send_cmd(sctx);
771 * Sends an unlink instruction to user space
773 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
775 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
778 btrfs_debug(fs_info, "send_unlink %s", path->start);
780 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
784 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
786 ret = send_cmd(sctx);
794 * Sends a rmdir instruction to user space
796 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
798 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
801 btrfs_debug(fs_info, "send_rmdir %s", path->start);
803 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
807 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
809 ret = send_cmd(sctx);
817 * Helper function to retrieve some fields from an inode item.
819 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
820 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
824 struct btrfs_inode_item *ii;
825 struct btrfs_key key;
828 key.type = BTRFS_INODE_ITEM_KEY;
830 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
837 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
838 struct btrfs_inode_item);
840 *size = btrfs_inode_size(path->nodes[0], ii);
842 *gen = btrfs_inode_generation(path->nodes[0], ii);
844 *mode = btrfs_inode_mode(path->nodes[0], ii);
846 *uid = btrfs_inode_uid(path->nodes[0], ii);
848 *gid = btrfs_inode_gid(path->nodes[0], ii);
850 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
855 static int get_inode_info(struct btrfs_root *root,
856 u64 ino, u64 *size, u64 *gen,
857 u64 *mode, u64 *uid, u64 *gid,
860 struct btrfs_path *path;
863 path = alloc_path_for_send();
866 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
868 btrfs_free_path(path);
872 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
877 * Helper function to iterate the entries in ONE btrfs_inode_ref or
878 * btrfs_inode_extref.
879 * The iterate callback may return a non zero value to stop iteration. This can
880 * be a negative value for error codes or 1 to simply stop it.
882 * path must point to the INODE_REF or INODE_EXTREF when called.
884 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
885 struct btrfs_key *found_key, int resolve,
886 iterate_inode_ref_t iterate, void *ctx)
888 struct extent_buffer *eb = path->nodes[0];
889 struct btrfs_item *item;
890 struct btrfs_inode_ref *iref;
891 struct btrfs_inode_extref *extref;
892 struct btrfs_path *tmp_path;
896 int slot = path->slots[0];
903 unsigned long name_off;
904 unsigned long elem_size;
907 p = fs_path_alloc_reversed();
911 tmp_path = alloc_path_for_send();
918 if (found_key->type == BTRFS_INODE_REF_KEY) {
919 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
920 struct btrfs_inode_ref);
921 item = btrfs_item_nr(slot);
922 total = btrfs_item_size(eb, item);
923 elem_size = sizeof(*iref);
925 ptr = btrfs_item_ptr_offset(eb, slot);
926 total = btrfs_item_size_nr(eb, slot);
927 elem_size = sizeof(*extref);
930 while (cur < total) {
933 if (found_key->type == BTRFS_INODE_REF_KEY) {
934 iref = (struct btrfs_inode_ref *)(ptr + cur);
935 name_len = btrfs_inode_ref_name_len(eb, iref);
936 name_off = (unsigned long)(iref + 1);
937 index = btrfs_inode_ref_index(eb, iref);
938 dir = found_key->offset;
940 extref = (struct btrfs_inode_extref *)(ptr + cur);
941 name_len = btrfs_inode_extref_name_len(eb, extref);
942 name_off = (unsigned long)&extref->name;
943 index = btrfs_inode_extref_index(eb, extref);
944 dir = btrfs_inode_extref_parent(eb, extref);
948 start = btrfs_ref_to_path(root, tmp_path, name_len,
952 ret = PTR_ERR(start);
955 if (start < p->buf) {
956 /* overflow , try again with larger buffer */
957 ret = fs_path_ensure_buf(p,
958 p->buf_len + p->buf - start);
961 start = btrfs_ref_to_path(root, tmp_path,
966 ret = PTR_ERR(start);
969 BUG_ON(start < p->buf);
973 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
979 cur += elem_size + name_len;
980 ret = iterate(num, dir, index, p, ctx);
987 btrfs_free_path(tmp_path);
992 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
993 const char *name, int name_len,
994 const char *data, int data_len,
998 * Helper function to iterate the entries in ONE btrfs_dir_item.
999 * The iterate callback may return a non zero value to stop iteration. This can
1000 * be a negative value for error codes or 1 to simply stop it.
1002 * path must point to the dir item when called.
1004 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
1005 struct btrfs_key *found_key,
1006 iterate_dir_item_t iterate, void *ctx)
1009 struct extent_buffer *eb;
1010 struct btrfs_item *item;
1011 struct btrfs_dir_item *di;
1012 struct btrfs_key di_key;
1025 * Start with a small buffer (1 page). If later we end up needing more
1026 * space, which can happen for xattrs on a fs with a leaf size greater
1027 * then the page size, attempt to increase the buffer. Typically xattr
1031 buf = kmalloc(buf_len, GFP_KERNEL);
1037 eb = path->nodes[0];
1038 slot = path->slots[0];
1039 item = btrfs_item_nr(slot);
1040 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1043 total = btrfs_item_size(eb, item);
1046 while (cur < total) {
1047 name_len = btrfs_dir_name_len(eb, di);
1048 data_len = btrfs_dir_data_len(eb, di);
1049 type = btrfs_dir_type(eb, di);
1050 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1052 if (type == BTRFS_FT_XATTR) {
1053 if (name_len > XATTR_NAME_MAX) {
1054 ret = -ENAMETOOLONG;
1057 if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(root)) {
1065 if (name_len + data_len > PATH_MAX) {
1066 ret = -ENAMETOOLONG;
1071 if (name_len + data_len > buf_len) {
1072 buf_len = name_len + data_len;
1073 if (is_vmalloc_addr(buf)) {
1077 char *tmp = krealloc(buf, buf_len,
1078 GFP_KERNEL | __GFP_NOWARN);
1085 buf = vmalloc(buf_len);
1093 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1094 name_len + data_len);
1096 len = sizeof(*di) + name_len + data_len;
1097 di = (struct btrfs_dir_item *)((char *)di + len);
1100 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1101 data_len, type, ctx);
1117 static int __copy_first_ref(int num, u64 dir, int index,
1118 struct fs_path *p, void *ctx)
1121 struct fs_path *pt = ctx;
1123 ret = fs_path_copy(pt, p);
1127 /* we want the first only */
1132 * Retrieve the first path of an inode. If an inode has more then one
1133 * ref/hardlink, this is ignored.
1135 static int get_inode_path(struct btrfs_root *root,
1136 u64 ino, struct fs_path *path)
1139 struct btrfs_key key, found_key;
1140 struct btrfs_path *p;
1142 p = alloc_path_for_send();
1146 fs_path_reset(path);
1149 key.type = BTRFS_INODE_REF_KEY;
1152 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1159 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1160 if (found_key.objectid != ino ||
1161 (found_key.type != BTRFS_INODE_REF_KEY &&
1162 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1167 ret = iterate_inode_ref(root, p, &found_key, 1,
1168 __copy_first_ref, path);
1178 struct backref_ctx {
1179 struct send_ctx *sctx;
1181 struct btrfs_path *path;
1182 /* number of total found references */
1186 * used for clones found in send_root. clones found behind cur_objectid
1187 * and cur_offset are not considered as allowed clones.
1192 /* may be truncated in case it's the last extent in a file */
1195 /* data offset in the file extent item */
1198 /* Just to check for bugs in backref resolving */
1202 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1204 u64 root = (u64)(uintptr_t)key;
1205 struct clone_root *cr = (struct clone_root *)elt;
1207 if (root < cr->root->objectid)
1209 if (root > cr->root->objectid)
1214 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1216 struct clone_root *cr1 = (struct clone_root *)e1;
1217 struct clone_root *cr2 = (struct clone_root *)e2;
1219 if (cr1->root->objectid < cr2->root->objectid)
1221 if (cr1->root->objectid > cr2->root->objectid)
1227 * Called for every backref that is found for the current extent.
1228 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1230 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1232 struct backref_ctx *bctx = ctx_;
1233 struct clone_root *found;
1237 /* First check if the root is in the list of accepted clone sources */
1238 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1239 bctx->sctx->clone_roots_cnt,
1240 sizeof(struct clone_root),
1241 __clone_root_cmp_bsearch);
1245 if (found->root == bctx->sctx->send_root &&
1246 ino == bctx->cur_objectid &&
1247 offset == bctx->cur_offset) {
1248 bctx->found_itself = 1;
1252 * There are inodes that have extents that lie behind its i_size. Don't
1253 * accept clones from these extents.
1255 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1257 btrfs_release_path(bctx->path);
1261 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1265 * Make sure we don't consider clones from send_root that are
1266 * behind the current inode/offset.
1268 if (found->root == bctx->sctx->send_root) {
1270 * TODO for the moment we don't accept clones from the inode
1271 * that is currently send. We may change this when
1272 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1275 if (ino >= bctx->cur_objectid)
1278 if (ino > bctx->cur_objectid)
1280 if (offset + bctx->extent_len > bctx->cur_offset)
1286 found->found_refs++;
1287 if (ino < found->ino) {
1289 found->offset = offset;
1290 } else if (found->ino == ino) {
1292 * same extent found more then once in the same file.
1294 if (found->offset > offset + bctx->extent_len)
1295 found->offset = offset;
1302 * Given an inode, offset and extent item, it finds a good clone for a clone
1303 * instruction. Returns -ENOENT when none could be found. The function makes
1304 * sure that the returned clone is usable at the point where sending is at the
1305 * moment. This means, that no clones are accepted which lie behind the current
1308 * path must point to the extent item when called.
1310 static int find_extent_clone(struct send_ctx *sctx,
1311 struct btrfs_path *path,
1312 u64 ino, u64 data_offset,
1314 struct clone_root **found)
1316 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
1322 u64 extent_item_pos;
1324 struct btrfs_file_extent_item *fi;
1325 struct extent_buffer *eb = path->nodes[0];
1326 struct backref_ctx *backref_ctx = NULL;
1327 struct clone_root *cur_clone_root;
1328 struct btrfs_key found_key;
1329 struct btrfs_path *tmp_path;
1333 tmp_path = alloc_path_for_send();
1337 /* We only use this path under the commit sem */
1338 tmp_path->need_commit_sem = 0;
1340 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL);
1346 backref_ctx->path = tmp_path;
1348 if (data_offset >= ino_size) {
1350 * There may be extents that lie behind the file's size.
1351 * I at least had this in combination with snapshotting while
1352 * writing large files.
1358 fi = btrfs_item_ptr(eb, path->slots[0],
1359 struct btrfs_file_extent_item);
1360 extent_type = btrfs_file_extent_type(eb, fi);
1361 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1365 compressed = btrfs_file_extent_compression(eb, fi);
1367 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1368 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1369 if (disk_byte == 0) {
1373 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1375 down_read(&fs_info->commit_root_sem);
1376 ret = extent_from_logical(fs_info, disk_byte, tmp_path,
1377 &found_key, &flags);
1378 up_read(&fs_info->commit_root_sem);
1379 btrfs_release_path(tmp_path);
1383 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1389 * Setup the clone roots.
1391 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1392 cur_clone_root = sctx->clone_roots + i;
1393 cur_clone_root->ino = (u64)-1;
1394 cur_clone_root->offset = 0;
1395 cur_clone_root->found_refs = 0;
1398 backref_ctx->sctx = sctx;
1399 backref_ctx->found = 0;
1400 backref_ctx->cur_objectid = ino;
1401 backref_ctx->cur_offset = data_offset;
1402 backref_ctx->found_itself = 0;
1403 backref_ctx->extent_len = num_bytes;
1405 * For non-compressed extents iterate_extent_inodes() gives us extent
1406 * offsets that already take into account the data offset, but not for
1407 * compressed extents, since the offset is logical and not relative to
1408 * the physical extent locations. We must take this into account to
1409 * avoid sending clone offsets that go beyond the source file's size,
1410 * which would result in the clone ioctl failing with -EINVAL on the
1413 if (compressed == BTRFS_COMPRESS_NONE)
1414 backref_ctx->data_offset = 0;
1416 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1419 * The last extent of a file may be too large due to page alignment.
1420 * We need to adjust extent_len in this case so that the checks in
1421 * __iterate_backrefs work.
1423 if (data_offset + num_bytes >= ino_size)
1424 backref_ctx->extent_len = ino_size - data_offset;
1427 * Now collect all backrefs.
1429 if (compressed == BTRFS_COMPRESS_NONE)
1430 extent_item_pos = logical - found_key.objectid;
1432 extent_item_pos = 0;
1433 ret = iterate_extent_inodes(fs_info,
1434 found_key.objectid, extent_item_pos, 1,
1435 __iterate_backrefs, backref_ctx);
1440 if (!backref_ctx->found_itself) {
1441 /* found a bug in backref code? */
1444 "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu",
1445 ino, data_offset, disk_byte, found_key.objectid);
1449 btrfs_debug(fs_info,
1450 "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu",
1451 data_offset, ino, num_bytes, logical);
1453 if (!backref_ctx->found)
1454 btrfs_debug(fs_info, "no clones found");
1456 cur_clone_root = NULL;
1457 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1458 if (sctx->clone_roots[i].found_refs) {
1459 if (!cur_clone_root)
1460 cur_clone_root = sctx->clone_roots + i;
1461 else if (sctx->clone_roots[i].root == sctx->send_root)
1462 /* prefer clones from send_root over others */
1463 cur_clone_root = sctx->clone_roots + i;
1468 if (cur_clone_root) {
1469 *found = cur_clone_root;
1476 btrfs_free_path(tmp_path);
1481 static int read_symlink(struct btrfs_root *root,
1483 struct fs_path *dest)
1486 struct btrfs_path *path;
1487 struct btrfs_key key;
1488 struct btrfs_file_extent_item *ei;
1494 path = alloc_path_for_send();
1499 key.type = BTRFS_EXTENT_DATA_KEY;
1501 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1506 * An empty symlink inode. Can happen in rare error paths when
1507 * creating a symlink (transaction committed before the inode
1508 * eviction handler removed the symlink inode items and a crash
1509 * happened in between or the subvol was snapshoted in between).
1510 * Print an informative message to dmesg/syslog so that the user
1511 * can delete the symlink.
1513 btrfs_err(root->fs_info,
1514 "Found empty symlink inode %llu at root %llu",
1515 ino, root->root_key.objectid);
1520 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1521 struct btrfs_file_extent_item);
1522 type = btrfs_file_extent_type(path->nodes[0], ei);
1523 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1524 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1525 BUG_ON(compression);
1527 off = btrfs_file_extent_inline_start(ei);
1528 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1530 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1533 btrfs_free_path(path);
1538 * Helper function to generate a file name that is unique in the root of
1539 * send_root and parent_root. This is used to generate names for orphan inodes.
1541 static int gen_unique_name(struct send_ctx *sctx,
1543 struct fs_path *dest)
1546 struct btrfs_path *path;
1547 struct btrfs_dir_item *di;
1552 path = alloc_path_for_send();
1557 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1559 ASSERT(len < sizeof(tmp));
1561 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1562 path, BTRFS_FIRST_FREE_OBJECTID,
1563 tmp, strlen(tmp), 0);
1564 btrfs_release_path(path);
1570 /* not unique, try again */
1575 if (!sctx->parent_root) {
1581 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1582 path, BTRFS_FIRST_FREE_OBJECTID,
1583 tmp, strlen(tmp), 0);
1584 btrfs_release_path(path);
1590 /* not unique, try again */
1598 ret = fs_path_add(dest, tmp, strlen(tmp));
1601 btrfs_free_path(path);
1606 inode_state_no_change,
1607 inode_state_will_create,
1608 inode_state_did_create,
1609 inode_state_will_delete,
1610 inode_state_did_delete,
1613 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1621 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1623 if (ret < 0 && ret != -ENOENT)
1627 if (!sctx->parent_root) {
1628 right_ret = -ENOENT;
1630 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1631 NULL, NULL, NULL, NULL);
1632 if (ret < 0 && ret != -ENOENT)
1637 if (!left_ret && !right_ret) {
1638 if (left_gen == gen && right_gen == gen) {
1639 ret = inode_state_no_change;
1640 } else if (left_gen == gen) {
1641 if (ino < sctx->send_progress)
1642 ret = inode_state_did_create;
1644 ret = inode_state_will_create;
1645 } else if (right_gen == gen) {
1646 if (ino < sctx->send_progress)
1647 ret = inode_state_did_delete;
1649 ret = inode_state_will_delete;
1653 } else if (!left_ret) {
1654 if (left_gen == gen) {
1655 if (ino < sctx->send_progress)
1656 ret = inode_state_did_create;
1658 ret = inode_state_will_create;
1662 } else if (!right_ret) {
1663 if (right_gen == gen) {
1664 if (ino < sctx->send_progress)
1665 ret = inode_state_did_delete;
1667 ret = inode_state_will_delete;
1679 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1683 ret = get_cur_inode_state(sctx, ino, gen);
1687 if (ret == inode_state_no_change ||
1688 ret == inode_state_did_create ||
1689 ret == inode_state_will_delete)
1699 * Helper function to lookup a dir item in a dir.
1701 static int lookup_dir_item_inode(struct btrfs_root *root,
1702 u64 dir, const char *name, int name_len,
1707 struct btrfs_dir_item *di;
1708 struct btrfs_key key;
1709 struct btrfs_path *path;
1711 path = alloc_path_for_send();
1715 di = btrfs_lookup_dir_item(NULL, root, path,
1716 dir, name, name_len, 0);
1725 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1726 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1730 *found_inode = key.objectid;
1731 *found_type = btrfs_dir_type(path->nodes[0], di);
1734 btrfs_free_path(path);
1739 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1740 * generation of the parent dir and the name of the dir entry.
1742 static int get_first_ref(struct btrfs_root *root, u64 ino,
1743 u64 *dir, u64 *dir_gen, struct fs_path *name)
1746 struct btrfs_key key;
1747 struct btrfs_key found_key;
1748 struct btrfs_path *path;
1752 path = alloc_path_for_send();
1757 key.type = BTRFS_INODE_REF_KEY;
1760 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1764 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1766 if (ret || found_key.objectid != ino ||
1767 (found_key.type != BTRFS_INODE_REF_KEY &&
1768 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1773 if (found_key.type == BTRFS_INODE_REF_KEY) {
1774 struct btrfs_inode_ref *iref;
1775 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1776 struct btrfs_inode_ref);
1777 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1778 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1779 (unsigned long)(iref + 1),
1781 parent_dir = found_key.offset;
1783 struct btrfs_inode_extref *extref;
1784 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1785 struct btrfs_inode_extref);
1786 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1787 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1788 (unsigned long)&extref->name, len);
1789 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1793 btrfs_release_path(path);
1796 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1805 btrfs_free_path(path);
1809 static int is_first_ref(struct btrfs_root *root,
1811 const char *name, int name_len)
1814 struct fs_path *tmp_name;
1817 tmp_name = fs_path_alloc();
1821 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1825 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1830 ret = !memcmp(tmp_name->start, name, name_len);
1833 fs_path_free(tmp_name);
1838 * Used by process_recorded_refs to determine if a new ref would overwrite an
1839 * already existing ref. In case it detects an overwrite, it returns the
1840 * inode/gen in who_ino/who_gen.
1841 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1842 * to make sure later references to the overwritten inode are possible.
1843 * Orphanizing is however only required for the first ref of an inode.
1844 * process_recorded_refs does an additional is_first_ref check to see if
1845 * orphanizing is really required.
1847 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1848 const char *name, int name_len,
1849 u64 *who_ino, u64 *who_gen)
1853 u64 other_inode = 0;
1856 if (!sctx->parent_root)
1859 ret = is_inode_existent(sctx, dir, dir_gen);
1864 * If we have a parent root we need to verify that the parent dir was
1865 * not deleted and then re-created, if it was then we have no overwrite
1866 * and we can just unlink this entry.
1868 if (sctx->parent_root) {
1869 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1871 if (ret < 0 && ret != -ENOENT)
1881 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1882 &other_inode, &other_type);
1883 if (ret < 0 && ret != -ENOENT)
1891 * Check if the overwritten ref was already processed. If yes, the ref
1892 * was already unlinked/moved, so we can safely assume that we will not
1893 * overwrite anything at this point in time.
1895 if (other_inode > sctx->send_progress ||
1896 is_waiting_for_move(sctx, other_inode)) {
1897 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1898 who_gen, NULL, NULL, NULL, NULL);
1903 *who_ino = other_inode;
1913 * Checks if the ref was overwritten by an already processed inode. This is
1914 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1915 * thus the orphan name needs be used.
1916 * process_recorded_refs also uses it to avoid unlinking of refs that were
1919 static int did_overwrite_ref(struct send_ctx *sctx,
1920 u64 dir, u64 dir_gen,
1921 u64 ino, u64 ino_gen,
1922 const char *name, int name_len)
1929 if (!sctx->parent_root)
1932 ret = is_inode_existent(sctx, dir, dir_gen);
1936 /* check if the ref was overwritten by another ref */
1937 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1938 &ow_inode, &other_type);
1939 if (ret < 0 && ret != -ENOENT)
1942 /* was never and will never be overwritten */
1947 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1952 if (ow_inode == ino && gen == ino_gen) {
1958 * We know that it is or will be overwritten. Check this now.
1959 * The current inode being processed might have been the one that caused
1960 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1961 * the current inode being processed.
1963 if ((ow_inode < sctx->send_progress) ||
1964 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1965 gen == sctx->cur_inode_gen))
1975 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1976 * that got overwritten. This is used by process_recorded_refs to determine
1977 * if it has to use the path as returned by get_cur_path or the orphan name.
1979 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1982 struct fs_path *name = NULL;
1986 if (!sctx->parent_root)
1989 name = fs_path_alloc();
1993 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1997 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1998 name->start, fs_path_len(name));
2006 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
2007 * so we need to do some special handling in case we have clashes. This function
2008 * takes care of this with the help of name_cache_entry::radix_list.
2009 * In case of error, nce is kfreed.
2011 static int name_cache_insert(struct send_ctx *sctx,
2012 struct name_cache_entry *nce)
2015 struct list_head *nce_head;
2017 nce_head = radix_tree_lookup(&sctx->name_cache,
2018 (unsigned long)nce->ino);
2020 nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL);
2025 INIT_LIST_HEAD(nce_head);
2027 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2034 list_add_tail(&nce->radix_list, nce_head);
2035 list_add_tail(&nce->list, &sctx->name_cache_list);
2036 sctx->name_cache_size++;
2041 static void name_cache_delete(struct send_ctx *sctx,
2042 struct name_cache_entry *nce)
2044 struct list_head *nce_head;
2046 nce_head = radix_tree_lookup(&sctx->name_cache,
2047 (unsigned long)nce->ino);
2049 btrfs_err(sctx->send_root->fs_info,
2050 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2051 nce->ino, sctx->name_cache_size);
2054 list_del(&nce->radix_list);
2055 list_del(&nce->list);
2056 sctx->name_cache_size--;
2059 * We may not get to the final release of nce_head if the lookup fails
2061 if (nce_head && list_empty(nce_head)) {
2062 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2067 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2070 struct list_head *nce_head;
2071 struct name_cache_entry *cur;
2073 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2077 list_for_each_entry(cur, nce_head, radix_list) {
2078 if (cur->ino == ino && cur->gen == gen)
2085 * Removes the entry from the list and adds it back to the end. This marks the
2086 * entry as recently used so that name_cache_clean_unused does not remove it.
2088 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2090 list_del(&nce->list);
2091 list_add_tail(&nce->list, &sctx->name_cache_list);
2095 * Remove some entries from the beginning of name_cache_list.
2097 static void name_cache_clean_unused(struct send_ctx *sctx)
2099 struct name_cache_entry *nce;
2101 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2104 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2105 nce = list_entry(sctx->name_cache_list.next,
2106 struct name_cache_entry, list);
2107 name_cache_delete(sctx, nce);
2112 static void name_cache_free(struct send_ctx *sctx)
2114 struct name_cache_entry *nce;
2116 while (!list_empty(&sctx->name_cache_list)) {
2117 nce = list_entry(sctx->name_cache_list.next,
2118 struct name_cache_entry, list);
2119 name_cache_delete(sctx, nce);
2125 * Used by get_cur_path for each ref up to the root.
2126 * Returns 0 if it succeeded.
2127 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2128 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2129 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2130 * Returns <0 in case of error.
2132 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2136 struct fs_path *dest)
2140 struct name_cache_entry *nce = NULL;
2143 * First check if we already did a call to this function with the same
2144 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2145 * return the cached result.
2147 nce = name_cache_search(sctx, ino, gen);
2149 if (ino < sctx->send_progress && nce->need_later_update) {
2150 name_cache_delete(sctx, nce);
2154 name_cache_used(sctx, nce);
2155 *parent_ino = nce->parent_ino;
2156 *parent_gen = nce->parent_gen;
2157 ret = fs_path_add(dest, nce->name, nce->name_len);
2166 * If the inode is not existent yet, add the orphan name and return 1.
2167 * This should only happen for the parent dir that we determine in
2170 ret = is_inode_existent(sctx, ino, gen);
2175 ret = gen_unique_name(sctx, ino, gen, dest);
2183 * Depending on whether the inode was already processed or not, use
2184 * send_root or parent_root for ref lookup.
2186 if (ino < sctx->send_progress)
2187 ret = get_first_ref(sctx->send_root, ino,
2188 parent_ino, parent_gen, dest);
2190 ret = get_first_ref(sctx->parent_root, ino,
2191 parent_ino, parent_gen, dest);
2196 * Check if the ref was overwritten by an inode's ref that was processed
2197 * earlier. If yes, treat as orphan and return 1.
2199 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2200 dest->start, dest->end - dest->start);
2204 fs_path_reset(dest);
2205 ret = gen_unique_name(sctx, ino, gen, dest);
2213 * Store the result of the lookup in the name cache.
2215 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL);
2223 nce->parent_ino = *parent_ino;
2224 nce->parent_gen = *parent_gen;
2225 nce->name_len = fs_path_len(dest);
2227 strcpy(nce->name, dest->start);
2229 if (ino < sctx->send_progress)
2230 nce->need_later_update = 0;
2232 nce->need_later_update = 1;
2234 nce_ret = name_cache_insert(sctx, nce);
2237 name_cache_clean_unused(sctx);
2244 * Magic happens here. This function returns the first ref to an inode as it
2245 * would look like while receiving the stream at this point in time.
2246 * We walk the path up to the root. For every inode in between, we check if it
2247 * was already processed/sent. If yes, we continue with the parent as found
2248 * in send_root. If not, we continue with the parent as found in parent_root.
2249 * If we encounter an inode that was deleted at this point in time, we use the
2250 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2251 * that were not created yet and overwritten inodes/refs.
2253 * When do we have have orphan inodes:
2254 * 1. When an inode is freshly created and thus no valid refs are available yet
2255 * 2. When a directory lost all it's refs (deleted) but still has dir items
2256 * inside which were not processed yet (pending for move/delete). If anyone
2257 * tried to get the path to the dir items, it would get a path inside that
2259 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2260 * of an unprocessed inode. If in that case the first ref would be
2261 * overwritten, the overwritten inode gets "orphanized". Later when we
2262 * process this overwritten inode, it is restored at a new place by moving
2265 * sctx->send_progress tells this function at which point in time receiving
2268 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2269 struct fs_path *dest)
2272 struct fs_path *name = NULL;
2273 u64 parent_inode = 0;
2277 name = fs_path_alloc();
2284 fs_path_reset(dest);
2286 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2287 struct waiting_dir_move *wdm;
2289 fs_path_reset(name);
2291 if (is_waiting_for_rm(sctx, ino)) {
2292 ret = gen_unique_name(sctx, ino, gen, name);
2295 ret = fs_path_add_path(dest, name);
2299 wdm = get_waiting_dir_move(sctx, ino);
2300 if (wdm && wdm->orphanized) {
2301 ret = gen_unique_name(sctx, ino, gen, name);
2304 ret = get_first_ref(sctx->parent_root, ino,
2305 &parent_inode, &parent_gen, name);
2307 ret = __get_cur_name_and_parent(sctx, ino, gen,
2317 ret = fs_path_add_path(dest, name);
2328 fs_path_unreverse(dest);
2333 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2335 static int send_subvol_begin(struct send_ctx *sctx)
2338 struct btrfs_root *send_root = sctx->send_root;
2339 struct btrfs_root *parent_root = sctx->parent_root;
2340 struct btrfs_path *path;
2341 struct btrfs_key key;
2342 struct btrfs_root_ref *ref;
2343 struct extent_buffer *leaf;
2347 path = btrfs_alloc_path();
2351 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
2353 btrfs_free_path(path);
2357 key.objectid = send_root->objectid;
2358 key.type = BTRFS_ROOT_BACKREF_KEY;
2361 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2370 leaf = path->nodes[0];
2371 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2372 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2373 key.objectid != send_root->objectid) {
2377 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2378 namelen = btrfs_root_ref_name_len(leaf, ref);
2379 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2380 btrfs_release_path(path);
2383 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2387 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2392 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2394 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2395 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2396 sctx->send_root->root_item.received_uuid);
2398 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2399 sctx->send_root->root_item.uuid);
2401 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2402 le64_to_cpu(sctx->send_root->root_item.ctransid));
2404 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2405 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2406 parent_root->root_item.received_uuid);
2408 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2409 parent_root->root_item.uuid);
2410 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2411 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2414 ret = send_cmd(sctx);
2418 btrfs_free_path(path);
2423 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2425 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2429 btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size);
2431 p = fs_path_alloc();
2435 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
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_SIZE, size);
2445 ret = send_cmd(sctx);
2453 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2455 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2459 btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode);
2461 p = fs_path_alloc();
2465 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2469 ret = get_cur_path(sctx, ino, gen, p);
2472 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2473 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2475 ret = send_cmd(sctx);
2483 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2485 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2489 btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu",
2492 p = fs_path_alloc();
2496 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2500 ret = get_cur_path(sctx, ino, gen, p);
2503 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2504 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2505 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2507 ret = send_cmd(sctx);
2515 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2517 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2519 struct fs_path *p = NULL;
2520 struct btrfs_inode_item *ii;
2521 struct btrfs_path *path = NULL;
2522 struct extent_buffer *eb;
2523 struct btrfs_key key;
2526 btrfs_debug(fs_info, "send_utimes %llu", ino);
2528 p = fs_path_alloc();
2532 path = alloc_path_for_send();
2539 key.type = BTRFS_INODE_ITEM_KEY;
2541 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2547 eb = path->nodes[0];
2548 slot = path->slots[0];
2549 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2551 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2555 ret = get_cur_path(sctx, ino, gen, p);
2558 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2559 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2560 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2561 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2562 /* TODO Add otime support when the otime patches get into upstream */
2564 ret = send_cmd(sctx);
2569 btrfs_free_path(path);
2574 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2575 * a valid path yet because we did not process the refs yet. So, the inode
2576 * is created as orphan.
2578 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2580 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
2588 btrfs_debug(fs_info, "send_create_inode %llu", ino);
2590 p = fs_path_alloc();
2594 if (ino != sctx->cur_ino) {
2595 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2600 gen = sctx->cur_inode_gen;
2601 mode = sctx->cur_inode_mode;
2602 rdev = sctx->cur_inode_rdev;
2605 if (S_ISREG(mode)) {
2606 cmd = BTRFS_SEND_C_MKFILE;
2607 } else if (S_ISDIR(mode)) {
2608 cmd = BTRFS_SEND_C_MKDIR;
2609 } else if (S_ISLNK(mode)) {
2610 cmd = BTRFS_SEND_C_SYMLINK;
2611 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2612 cmd = BTRFS_SEND_C_MKNOD;
2613 } else if (S_ISFIFO(mode)) {
2614 cmd = BTRFS_SEND_C_MKFIFO;
2615 } else if (S_ISSOCK(mode)) {
2616 cmd = BTRFS_SEND_C_MKSOCK;
2618 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2619 (int)(mode & S_IFMT));
2624 ret = begin_cmd(sctx, cmd);
2628 ret = gen_unique_name(sctx, ino, gen, p);
2632 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2633 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2635 if (S_ISLNK(mode)) {
2637 ret = read_symlink(sctx->send_root, ino, p);
2640 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2641 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2642 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2643 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2644 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2647 ret = send_cmd(sctx);
2659 * We need some special handling for inodes that get processed before the parent
2660 * directory got created. See process_recorded_refs for details.
2661 * This function does the check if we already created the dir out of order.
2663 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2666 struct btrfs_path *path = NULL;
2667 struct btrfs_key key;
2668 struct btrfs_key found_key;
2669 struct btrfs_key di_key;
2670 struct extent_buffer *eb;
2671 struct btrfs_dir_item *di;
2674 path = alloc_path_for_send();
2681 key.type = BTRFS_DIR_INDEX_KEY;
2683 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2688 eb = path->nodes[0];
2689 slot = path->slots[0];
2690 if (slot >= btrfs_header_nritems(eb)) {
2691 ret = btrfs_next_leaf(sctx->send_root, path);
2694 } else if (ret > 0) {
2701 btrfs_item_key_to_cpu(eb, &found_key, slot);
2702 if (found_key.objectid != key.objectid ||
2703 found_key.type != key.type) {
2708 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2709 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2711 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2712 di_key.objectid < sctx->send_progress) {
2721 btrfs_free_path(path);
2726 * Only creates the inode if it is:
2727 * 1. Not a directory
2728 * 2. Or a directory which was not created already due to out of order
2729 * directories. See did_create_dir and process_recorded_refs for details.
2731 static int send_create_inode_if_needed(struct send_ctx *sctx)
2735 if (S_ISDIR(sctx->cur_inode_mode)) {
2736 ret = did_create_dir(sctx, sctx->cur_ino);
2745 ret = send_create_inode(sctx, sctx->cur_ino);
2753 struct recorded_ref {
2754 struct list_head list;
2757 struct fs_path *full_path;
2765 * We need to process new refs before deleted refs, but compare_tree gives us
2766 * everything mixed. So we first record all refs and later process them.
2767 * This function is a helper to record one ref.
2769 static int __record_ref(struct list_head *head, u64 dir,
2770 u64 dir_gen, struct fs_path *path)
2772 struct recorded_ref *ref;
2774 ref = kmalloc(sizeof(*ref), GFP_KERNEL);
2779 ref->dir_gen = dir_gen;
2780 ref->full_path = path;
2782 ref->name = (char *)kbasename(ref->full_path->start);
2783 ref->name_len = ref->full_path->end - ref->name;
2784 ref->dir_path = ref->full_path->start;
2785 if (ref->name == ref->full_path->start)
2786 ref->dir_path_len = 0;
2788 ref->dir_path_len = ref->full_path->end -
2789 ref->full_path->start - 1 - ref->name_len;
2791 list_add_tail(&ref->list, head);
2795 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2797 struct recorded_ref *new;
2799 new = kmalloc(sizeof(*ref), GFP_KERNEL);
2803 new->dir = ref->dir;
2804 new->dir_gen = ref->dir_gen;
2805 new->full_path = NULL;
2806 INIT_LIST_HEAD(&new->list);
2807 list_add_tail(&new->list, list);
2811 static void __free_recorded_refs(struct list_head *head)
2813 struct recorded_ref *cur;
2815 while (!list_empty(head)) {
2816 cur = list_entry(head->next, struct recorded_ref, list);
2817 fs_path_free(cur->full_path);
2818 list_del(&cur->list);
2823 static void free_recorded_refs(struct send_ctx *sctx)
2825 __free_recorded_refs(&sctx->new_refs);
2826 __free_recorded_refs(&sctx->deleted_refs);
2830 * Renames/moves a file/dir to its orphan name. Used when the first
2831 * ref of an unprocessed inode gets overwritten and for all non empty
2834 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2835 struct fs_path *path)
2838 struct fs_path *orphan;
2840 orphan = fs_path_alloc();
2844 ret = gen_unique_name(sctx, ino, gen, orphan);
2848 ret = send_rename(sctx, path, orphan);
2851 fs_path_free(orphan);
2855 static struct orphan_dir_info *
2856 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2858 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2859 struct rb_node *parent = NULL;
2860 struct orphan_dir_info *entry, *odi;
2862 odi = kmalloc(sizeof(*odi), GFP_KERNEL);
2864 return ERR_PTR(-ENOMEM);
2870 entry = rb_entry(parent, struct orphan_dir_info, node);
2871 if (dir_ino < entry->ino) {
2873 } else if (dir_ino > entry->ino) {
2874 p = &(*p)->rb_right;
2881 rb_link_node(&odi->node, parent, p);
2882 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2886 static struct orphan_dir_info *
2887 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2889 struct rb_node *n = sctx->orphan_dirs.rb_node;
2890 struct orphan_dir_info *entry;
2893 entry = rb_entry(n, struct orphan_dir_info, node);
2894 if (dir_ino < entry->ino)
2896 else if (dir_ino > entry->ino)
2904 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2906 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2911 static void free_orphan_dir_info(struct send_ctx *sctx,
2912 struct orphan_dir_info *odi)
2916 rb_erase(&odi->node, &sctx->orphan_dirs);
2921 * Returns 1 if a directory can be removed at this point in time.
2922 * We check this by iterating all dir items and checking if the inode behind
2923 * the dir item was already processed.
2925 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2929 struct btrfs_root *root = sctx->parent_root;
2930 struct btrfs_path *path;
2931 struct btrfs_key key;
2932 struct btrfs_key found_key;
2933 struct btrfs_key loc;
2934 struct btrfs_dir_item *di;
2937 * Don't try to rmdir the top/root subvolume dir.
2939 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2942 path = alloc_path_for_send();
2947 key.type = BTRFS_DIR_INDEX_KEY;
2949 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2954 struct waiting_dir_move *dm;
2956 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2957 ret = btrfs_next_leaf(root, path);
2964 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2966 if (found_key.objectid != key.objectid ||
2967 found_key.type != key.type)
2970 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2971 struct btrfs_dir_item);
2972 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2974 dm = get_waiting_dir_move(sctx, loc.objectid);
2976 struct orphan_dir_info *odi;
2978 odi = add_orphan_dir_info(sctx, dir);
2984 dm->rmdir_ino = dir;
2989 if (loc.objectid > send_progress) {
2990 struct orphan_dir_info *odi;
2992 odi = get_orphan_dir_info(sctx, dir);
2993 free_orphan_dir_info(sctx, odi);
3004 btrfs_free_path(path);
3008 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
3010 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
3012 return entry != NULL;
3015 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
3017 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
3018 struct rb_node *parent = NULL;
3019 struct waiting_dir_move *entry, *dm;
3021 dm = kmalloc(sizeof(*dm), GFP_KERNEL);
3026 dm->orphanized = orphanized;
3030 entry = rb_entry(parent, struct waiting_dir_move, node);
3031 if (ino < entry->ino) {
3033 } else if (ino > entry->ino) {
3034 p = &(*p)->rb_right;
3041 rb_link_node(&dm->node, parent, p);
3042 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3046 static struct waiting_dir_move *
3047 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3049 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3050 struct waiting_dir_move *entry;
3053 entry = rb_entry(n, struct waiting_dir_move, node);
3054 if (ino < entry->ino)
3056 else if (ino > entry->ino)
3064 static void free_waiting_dir_move(struct send_ctx *sctx,
3065 struct waiting_dir_move *dm)
3069 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3073 static int add_pending_dir_move(struct send_ctx *sctx,
3077 struct list_head *new_refs,
3078 struct list_head *deleted_refs,
3079 const bool is_orphan)
3081 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3082 struct rb_node *parent = NULL;
3083 struct pending_dir_move *entry = NULL, *pm;
3084 struct recorded_ref *cur;
3088 pm = kmalloc(sizeof(*pm), GFP_KERNEL);
3091 pm->parent_ino = parent_ino;
3094 INIT_LIST_HEAD(&pm->list);
3095 INIT_LIST_HEAD(&pm->update_refs);
3096 RB_CLEAR_NODE(&pm->node);
3100 entry = rb_entry(parent, struct pending_dir_move, node);
3101 if (parent_ino < entry->parent_ino) {
3103 } else if (parent_ino > entry->parent_ino) {
3104 p = &(*p)->rb_right;
3111 list_for_each_entry(cur, deleted_refs, list) {
3112 ret = dup_ref(cur, &pm->update_refs);
3116 list_for_each_entry(cur, new_refs, list) {
3117 ret = dup_ref(cur, &pm->update_refs);
3122 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3127 list_add_tail(&pm->list, &entry->list);
3129 rb_link_node(&pm->node, parent, p);
3130 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3135 __free_recorded_refs(&pm->update_refs);
3141 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3144 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3145 struct pending_dir_move *entry;
3148 entry = rb_entry(n, struct pending_dir_move, node);
3149 if (parent_ino < entry->parent_ino)
3151 else if (parent_ino > entry->parent_ino)
3159 static int path_loop(struct send_ctx *sctx, struct fs_path *name,
3160 u64 ino, u64 gen, u64 *ancestor_ino)
3163 u64 parent_inode = 0;
3165 u64 start_ino = ino;
3168 while (ino != BTRFS_FIRST_FREE_OBJECTID) {
3169 fs_path_reset(name);
3171 if (is_waiting_for_rm(sctx, ino))
3173 if (is_waiting_for_move(sctx, ino)) {
3174 if (*ancestor_ino == 0)
3175 *ancestor_ino = ino;
3176 ret = get_first_ref(sctx->parent_root, ino,
3177 &parent_inode, &parent_gen, name);
3179 ret = __get_cur_name_and_parent(sctx, ino, gen,
3189 if (parent_inode == start_ino) {
3191 if (*ancestor_ino == 0)
3192 *ancestor_ino = ino;
3201 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3203 struct fs_path *from_path = NULL;
3204 struct fs_path *to_path = NULL;
3205 struct fs_path *name = NULL;
3206 u64 orig_progress = sctx->send_progress;
3207 struct recorded_ref *cur;
3208 u64 parent_ino, parent_gen;
3209 struct waiting_dir_move *dm = NULL;
3215 name = fs_path_alloc();
3216 from_path = fs_path_alloc();
3217 if (!name || !from_path) {
3222 dm = get_waiting_dir_move(sctx, pm->ino);
3224 rmdir_ino = dm->rmdir_ino;
3225 is_orphan = dm->orphanized;
3226 free_waiting_dir_move(sctx, dm);
3229 ret = gen_unique_name(sctx, pm->ino,
3230 pm->gen, from_path);
3232 ret = get_first_ref(sctx->parent_root, pm->ino,
3233 &parent_ino, &parent_gen, name);
3236 ret = get_cur_path(sctx, parent_ino, parent_gen,
3240 ret = fs_path_add_path(from_path, name);
3245 sctx->send_progress = sctx->cur_ino + 1;
3246 ret = path_loop(sctx, name, pm->ino, pm->gen, &ancestor);
3250 LIST_HEAD(deleted_refs);
3251 ASSERT(ancestor > BTRFS_FIRST_FREE_OBJECTID);
3252 ret = add_pending_dir_move(sctx, pm->ino, pm->gen, ancestor,
3253 &pm->update_refs, &deleted_refs,
3258 dm = get_waiting_dir_move(sctx, pm->ino);
3260 dm->rmdir_ino = rmdir_ino;
3264 fs_path_reset(name);
3267 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3271 ret = send_rename(sctx, from_path, to_path);
3276 struct orphan_dir_info *odi;
3278 odi = get_orphan_dir_info(sctx, rmdir_ino);
3280 /* already deleted */
3283 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino);
3289 name = fs_path_alloc();
3294 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3297 ret = send_rmdir(sctx, name);
3300 free_orphan_dir_info(sctx, odi);
3304 ret = send_utimes(sctx, pm->ino, pm->gen);
3309 * After rename/move, need to update the utimes of both new parent(s)
3310 * and old parent(s).
3312 list_for_each_entry(cur, &pm->update_refs, list) {
3314 * The parent inode might have been deleted in the send snapshot
3316 ret = get_inode_info(sctx->send_root, cur->dir, NULL,
3317 NULL, NULL, NULL, NULL, NULL);
3318 if (ret == -ENOENT) {
3325 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3332 fs_path_free(from_path);
3333 fs_path_free(to_path);
3334 sctx->send_progress = orig_progress;
3339 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3341 if (!list_empty(&m->list))
3343 if (!RB_EMPTY_NODE(&m->node))
3344 rb_erase(&m->node, &sctx->pending_dir_moves);
3345 __free_recorded_refs(&m->update_refs);
3349 static void tail_append_pending_moves(struct pending_dir_move *moves,
3350 struct list_head *stack)
3352 if (list_empty(&moves->list)) {
3353 list_add_tail(&moves->list, stack);
3356 list_splice_init(&moves->list, &list);
3357 list_add_tail(&moves->list, stack);
3358 list_splice_tail(&list, stack);
3362 static int apply_children_dir_moves(struct send_ctx *sctx)
3364 struct pending_dir_move *pm;
3365 struct list_head stack;
3366 u64 parent_ino = sctx->cur_ino;
3369 pm = get_pending_dir_moves(sctx, parent_ino);
3373 INIT_LIST_HEAD(&stack);
3374 tail_append_pending_moves(pm, &stack);
3376 while (!list_empty(&stack)) {
3377 pm = list_first_entry(&stack, struct pending_dir_move, list);
3378 parent_ino = pm->ino;
3379 ret = apply_dir_move(sctx, pm);
3380 free_pending_move(sctx, pm);
3383 pm = get_pending_dir_moves(sctx, parent_ino);
3385 tail_append_pending_moves(pm, &stack);
3390 while (!list_empty(&stack)) {
3391 pm = list_first_entry(&stack, struct pending_dir_move, list);
3392 free_pending_move(sctx, pm);
3398 * We might need to delay a directory rename even when no ancestor directory
3399 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3400 * renamed. This happens when we rename a directory to the old name (the name
3401 * in the parent root) of some other unrelated directory that got its rename
3402 * delayed due to some ancestor with higher number that got renamed.
3408 * |---- a/ (ino 257)
3409 * | |---- file (ino 260)
3411 * |---- b/ (ino 258)
3412 * |---- c/ (ino 259)
3416 * |---- a/ (ino 258)
3417 * |---- x/ (ino 259)
3418 * |---- y/ (ino 257)
3419 * |----- file (ino 260)
3421 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3422 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3423 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3426 * 1 - rename 259 from 'c' to 'x'
3427 * 2 - rename 257 from 'a' to 'x/y'
3428 * 3 - rename 258 from 'b' to 'a'
3430 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3431 * be done right away and < 0 on error.
3433 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3434 struct recorded_ref *parent_ref,
3435 const bool is_orphan)
3437 struct btrfs_path *path;
3438 struct btrfs_key key;
3439 struct btrfs_key di_key;
3440 struct btrfs_dir_item *di;
3444 struct waiting_dir_move *wdm;
3446 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3449 path = alloc_path_for_send();
3453 key.objectid = parent_ref->dir;
3454 key.type = BTRFS_DIR_ITEM_KEY;
3455 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3457 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3460 } else if (ret > 0) {
3465 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3466 parent_ref->name, parent_ref->name_len);
3472 * di_key.objectid has the number of the inode that has a dentry in the
3473 * parent directory with the same name that sctx->cur_ino is being
3474 * renamed to. We need to check if that inode is in the send root as
3475 * well and if it is currently marked as an inode with a pending rename,
3476 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3477 * that it happens after that other inode is renamed.
3479 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3480 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3485 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3486 &left_gen, NULL, NULL, NULL, NULL);
3489 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3490 &right_gen, NULL, NULL, NULL, NULL);
3497 /* Different inode, no need to delay the rename of sctx->cur_ino */
3498 if (right_gen != left_gen) {
3503 wdm = get_waiting_dir_move(sctx, di_key.objectid);
3504 if (wdm && !wdm->orphanized) {
3505 ret = add_pending_dir_move(sctx,
3507 sctx->cur_inode_gen,
3510 &sctx->deleted_refs,
3516 btrfs_free_path(path);
3521 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3522 * Return 1 if true, 0 if false and < 0 on error.
3524 static int is_ancestor(struct btrfs_root *root,
3528 struct fs_path *fs_path)
3532 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3537 fs_path_reset(fs_path);
3538 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3540 if (ret == -ENOENT && ino == ino2)
3545 return parent_gen == ino1_gen ? 1 : 0;
3551 static int wait_for_parent_move(struct send_ctx *sctx,
3552 struct recorded_ref *parent_ref,
3553 const bool is_orphan)
3556 u64 ino = parent_ref->dir;
3557 u64 parent_ino_before, parent_ino_after;
3558 struct fs_path *path_before = NULL;
3559 struct fs_path *path_after = NULL;
3562 path_after = fs_path_alloc();
3563 path_before = fs_path_alloc();
3564 if (!path_after || !path_before) {
3570 * Our current directory inode may not yet be renamed/moved because some
3571 * ancestor (immediate or not) has to be renamed/moved first. So find if
3572 * such ancestor exists and make sure our own rename/move happens after
3573 * that ancestor is processed to avoid path build infinite loops (done
3574 * at get_cur_path()).
3576 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3577 if (is_waiting_for_move(sctx, ino)) {
3579 * If the current inode is an ancestor of ino in the
3580 * parent root, we need to delay the rename of the
3581 * current inode, otherwise don't delayed the rename
3582 * because we can end up with a circular dependency
3583 * of renames, resulting in some directories never
3584 * getting the respective rename operations issued in
3585 * the send stream or getting into infinite path build
3588 ret = is_ancestor(sctx->parent_root,
3589 sctx->cur_ino, sctx->cur_inode_gen,
3595 fs_path_reset(path_before);
3596 fs_path_reset(path_after);
3598 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3602 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3604 if (ret < 0 && ret != -ENOENT) {
3606 } else if (ret == -ENOENT) {
3611 len1 = fs_path_len(path_before);
3612 len2 = fs_path_len(path_after);
3613 if (ino > sctx->cur_ino &&
3614 (parent_ino_before != parent_ino_after || len1 != len2 ||
3615 memcmp(path_before->start, path_after->start, len1))) {
3619 ino = parent_ino_after;
3623 fs_path_free(path_before);
3624 fs_path_free(path_after);
3627 ret = add_pending_dir_move(sctx,
3629 sctx->cur_inode_gen,
3632 &sctx->deleted_refs,
3642 * This does all the move/link/unlink/rmdir magic.
3644 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3646 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
3648 struct recorded_ref *cur;
3649 struct recorded_ref *cur2;
3650 struct list_head check_dirs;
3651 struct fs_path *valid_path = NULL;
3654 int did_overwrite = 0;
3656 u64 last_dir_ino_rm = 0;
3657 bool can_rename = true;
3659 btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino);
3662 * This should never happen as the root dir always has the same ref
3663 * which is always '..'
3665 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3666 INIT_LIST_HEAD(&check_dirs);
3668 valid_path = fs_path_alloc();
3675 * First, check if the first ref of the current inode was overwritten
3676 * before. If yes, we know that the current inode was already orphanized
3677 * and thus use the orphan name. If not, we can use get_cur_path to
3678 * get the path of the first ref as it would like while receiving at
3679 * this point in time.
3680 * New inodes are always orphan at the beginning, so force to use the
3681 * orphan name in this case.
3682 * The first ref is stored in valid_path and will be updated if it
3683 * gets moved around.
3685 if (!sctx->cur_inode_new) {
3686 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3687 sctx->cur_inode_gen);
3693 if (sctx->cur_inode_new || did_overwrite) {
3694 ret = gen_unique_name(sctx, sctx->cur_ino,
3695 sctx->cur_inode_gen, valid_path);
3700 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3706 list_for_each_entry(cur, &sctx->new_refs, list) {
3708 * We may have refs where the parent directory does not exist
3709 * yet. This happens if the parent directories inum is higher
3710 * the the current inum. To handle this case, we create the
3711 * parent directory out of order. But we need to check if this
3712 * did already happen before due to other refs in the same dir.
3714 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3717 if (ret == inode_state_will_create) {
3720 * First check if any of the current inodes refs did
3721 * already create the dir.
3723 list_for_each_entry(cur2, &sctx->new_refs, list) {
3726 if (cur2->dir == cur->dir) {
3733 * If that did not happen, check if a previous inode
3734 * did already create the dir.
3737 ret = did_create_dir(sctx, cur->dir);
3741 ret = send_create_inode(sctx, cur->dir);
3748 * Check if this new ref would overwrite the first ref of
3749 * another unprocessed inode. If yes, orphanize the
3750 * overwritten inode. If we find an overwritten ref that is
3751 * not the first ref, simply unlink it.
3753 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3754 cur->name, cur->name_len,
3755 &ow_inode, &ow_gen);
3759 ret = is_first_ref(sctx->parent_root,
3760 ow_inode, cur->dir, cur->name,
3765 struct name_cache_entry *nce;
3766 struct waiting_dir_move *wdm;
3768 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3774 * If ow_inode has its rename operation delayed
3775 * make sure that its orphanized name is used in
3776 * the source path when performing its rename
3779 if (is_waiting_for_move(sctx, ow_inode)) {
3780 wdm = get_waiting_dir_move(sctx,
3783 wdm->orphanized = true;
3787 * Make sure we clear our orphanized inode's
3788 * name from the name cache. This is because the
3789 * inode ow_inode might be an ancestor of some
3790 * other inode that will be orphanized as well
3791 * later and has an inode number greater than
3792 * sctx->send_progress. We need to prevent
3793 * future name lookups from using the old name
3794 * and get instead the orphan name.
3796 nce = name_cache_search(sctx, ow_inode, ow_gen);
3798 name_cache_delete(sctx, nce);
3803 * ow_inode might currently be an ancestor of
3804 * cur_ino, therefore compute valid_path (the
3805 * current path of cur_ino) again because it
3806 * might contain the pre-orphanization name of
3807 * ow_inode, which is no longer valid.
3809 fs_path_reset(valid_path);
3810 ret = get_cur_path(sctx, sctx->cur_ino,
3811 sctx->cur_inode_gen, valid_path);
3815 ret = send_unlink(sctx, cur->full_path);
3821 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3822 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3831 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3833 ret = wait_for_parent_move(sctx, cur, is_orphan);
3843 * link/move the ref to the new place. If we have an orphan
3844 * inode, move it and update valid_path. If not, link or move
3845 * it depending on the inode mode.
3847 if (is_orphan && can_rename) {
3848 ret = send_rename(sctx, valid_path, cur->full_path);
3852 ret = fs_path_copy(valid_path, cur->full_path);
3855 } else if (can_rename) {
3856 if (S_ISDIR(sctx->cur_inode_mode)) {
3858 * Dirs can't be linked, so move it. For moved
3859 * dirs, we always have one new and one deleted
3860 * ref. The deleted ref is ignored later.
3862 ret = send_rename(sctx, valid_path,
3865 ret = fs_path_copy(valid_path,
3870 ret = send_link(sctx, cur->full_path,
3876 ret = dup_ref(cur, &check_dirs);
3881 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3883 * Check if we can already rmdir the directory. If not,
3884 * orphanize it. For every dir item inside that gets deleted
3885 * later, we do this check again and rmdir it then if possible.
3886 * See the use of check_dirs for more details.
3888 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3893 ret = send_rmdir(sctx, valid_path);
3896 } else if (!is_orphan) {
3897 ret = orphanize_inode(sctx, sctx->cur_ino,
3898 sctx->cur_inode_gen, valid_path);
3904 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3905 ret = dup_ref(cur, &check_dirs);
3909 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3910 !list_empty(&sctx->deleted_refs)) {
3912 * We have a moved dir. Add the old parent to check_dirs
3914 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3916 ret = dup_ref(cur, &check_dirs);
3919 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3921 * We have a non dir inode. Go through all deleted refs and
3922 * unlink them if they were not already overwritten by other
3925 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3926 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3927 sctx->cur_ino, sctx->cur_inode_gen,
3928 cur->name, cur->name_len);
3932 ret = send_unlink(sctx, cur->full_path);
3936 ret = dup_ref(cur, &check_dirs);
3941 * If the inode is still orphan, unlink the orphan. This may
3942 * happen when a previous inode did overwrite the first ref
3943 * of this inode and no new refs were added for the current
3944 * inode. Unlinking does not mean that the inode is deleted in
3945 * all cases. There may still be links to this inode in other
3949 ret = send_unlink(sctx, valid_path);
3956 * We did collect all parent dirs where cur_inode was once located. We
3957 * now go through all these dirs and check if they are pending for
3958 * deletion and if it's finally possible to perform the rmdir now.
3959 * We also update the inode stats of the parent dirs here.
3961 list_for_each_entry(cur, &check_dirs, list) {
3963 * In case we had refs into dirs that were not processed yet,
3964 * we don't need to do the utime and rmdir logic for these dirs.
3965 * The dir will be processed later.
3967 if (cur->dir > sctx->cur_ino)
3970 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3974 if (ret == inode_state_did_create ||
3975 ret == inode_state_no_change) {
3976 /* TODO delayed utimes */
3977 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3980 } else if (ret == inode_state_did_delete &&
3981 cur->dir != last_dir_ino_rm) {
3982 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3987 ret = get_cur_path(sctx, cur->dir,
3988 cur->dir_gen, valid_path);
3991 ret = send_rmdir(sctx, valid_path);
3994 last_dir_ino_rm = cur->dir;
4002 __free_recorded_refs(&check_dirs);
4003 free_recorded_refs(sctx);
4004 fs_path_free(valid_path);
4008 static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
4009 struct fs_path *name, void *ctx, struct list_head *refs)
4012 struct send_ctx *sctx = ctx;
4016 p = fs_path_alloc();
4020 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
4025 ret = get_cur_path(sctx, dir, gen, p);
4028 ret = fs_path_add_path(p, name);
4032 ret = __record_ref(refs, dir, gen, p);
4040 static int __record_new_ref(int num, u64 dir, int index,
4041 struct fs_path *name,
4044 struct send_ctx *sctx = ctx;
4045 return record_ref(sctx->send_root, num, dir, index, name,
4046 ctx, &sctx->new_refs);
4050 static int __record_deleted_ref(int num, u64 dir, int index,
4051 struct fs_path *name,
4054 struct send_ctx *sctx = ctx;
4055 return record_ref(sctx->parent_root, num, dir, index, name,
4056 ctx, &sctx->deleted_refs);
4059 static int record_new_ref(struct send_ctx *sctx)
4063 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4064 sctx->cmp_key, 0, __record_new_ref, sctx);
4073 static int record_deleted_ref(struct send_ctx *sctx)
4077 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4078 sctx->cmp_key, 0, __record_deleted_ref, sctx);
4087 struct find_ref_ctx {
4090 struct btrfs_root *root;
4091 struct fs_path *name;
4095 static int __find_iref(int num, u64 dir, int index,
4096 struct fs_path *name,
4099 struct find_ref_ctx *ctx = ctx_;
4103 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
4104 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
4106 * To avoid doing extra lookups we'll only do this if everything
4109 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
4113 if (dir_gen != ctx->dir_gen)
4115 ctx->found_idx = num;
4121 static int find_iref(struct btrfs_root *root,
4122 struct btrfs_path *path,
4123 struct btrfs_key *key,
4124 u64 dir, u64 dir_gen, struct fs_path *name)
4127 struct find_ref_ctx ctx;
4131 ctx.dir_gen = dir_gen;
4135 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
4139 if (ctx.found_idx == -1)
4142 return ctx.found_idx;
4145 static int __record_changed_new_ref(int num, u64 dir, int index,
4146 struct fs_path *name,
4151 struct send_ctx *sctx = ctx;
4153 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4158 ret = find_iref(sctx->parent_root, sctx->right_path,
4159 sctx->cmp_key, dir, dir_gen, name);
4161 ret = __record_new_ref(num, dir, index, name, sctx);
4168 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4169 struct fs_path *name,
4174 struct send_ctx *sctx = ctx;
4176 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4181 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4182 dir, dir_gen, name);
4184 ret = __record_deleted_ref(num, dir, index, name, sctx);
4191 static int record_changed_ref(struct send_ctx *sctx)
4195 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4196 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4199 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4200 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4210 * Record and process all refs at once. Needed when an inode changes the
4211 * generation number, which means that it was deleted and recreated.
4213 static int process_all_refs(struct send_ctx *sctx,
4214 enum btrfs_compare_tree_result cmd)
4217 struct btrfs_root *root;
4218 struct btrfs_path *path;
4219 struct btrfs_key key;
4220 struct btrfs_key found_key;
4221 struct extent_buffer *eb;
4223 iterate_inode_ref_t cb;
4224 int pending_move = 0;
4226 path = alloc_path_for_send();
4230 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4231 root = sctx->send_root;
4232 cb = __record_new_ref;
4233 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4234 root = sctx->parent_root;
4235 cb = __record_deleted_ref;
4237 btrfs_err(sctx->send_root->fs_info,
4238 "Wrong command %d in process_all_refs", cmd);
4243 key.objectid = sctx->cmp_key->objectid;
4244 key.type = BTRFS_INODE_REF_KEY;
4246 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4251 eb = path->nodes[0];
4252 slot = path->slots[0];
4253 if (slot >= btrfs_header_nritems(eb)) {
4254 ret = btrfs_next_leaf(root, path);
4262 btrfs_item_key_to_cpu(eb, &found_key, slot);
4264 if (found_key.objectid != key.objectid ||
4265 (found_key.type != BTRFS_INODE_REF_KEY &&
4266 found_key.type != BTRFS_INODE_EXTREF_KEY))
4269 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4275 btrfs_release_path(path);
4278 * We don't actually care about pending_move as we are simply
4279 * re-creating this inode and will be rename'ing it into place once we
4280 * rename the parent directory.
4282 ret = process_recorded_refs(sctx, &pending_move);
4284 btrfs_free_path(path);
4288 static int send_set_xattr(struct send_ctx *sctx,
4289 struct fs_path *path,
4290 const char *name, int name_len,
4291 const char *data, int data_len)
4295 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4299 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4300 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4301 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4303 ret = send_cmd(sctx);
4310 static int send_remove_xattr(struct send_ctx *sctx,
4311 struct fs_path *path,
4312 const char *name, int name_len)
4316 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4320 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4321 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4323 ret = send_cmd(sctx);
4330 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4331 const char *name, int name_len,
4332 const char *data, int data_len,
4336 struct send_ctx *sctx = ctx;
4338 struct posix_acl_xattr_header dummy_acl;
4340 p = fs_path_alloc();
4345 * This hack is needed because empty acls are stored as zero byte
4346 * data in xattrs. Problem with that is, that receiving these zero byte
4347 * acls will fail later. To fix this, we send a dummy acl list that
4348 * only contains the version number and no entries.
4350 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4351 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4352 if (data_len == 0) {
4353 dummy_acl.a_version =
4354 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4355 data = (char *)&dummy_acl;
4356 data_len = sizeof(dummy_acl);
4360 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4364 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4371 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4372 const char *name, int name_len,
4373 const char *data, int data_len,
4377 struct send_ctx *sctx = ctx;
4380 p = fs_path_alloc();
4384 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4388 ret = send_remove_xattr(sctx, p, name, name_len);
4395 static int process_new_xattr(struct send_ctx *sctx)
4399 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4400 sctx->cmp_key, __process_new_xattr, sctx);
4405 static int process_deleted_xattr(struct send_ctx *sctx)
4407 return iterate_dir_item(sctx->parent_root, sctx->right_path,
4408 sctx->cmp_key, __process_deleted_xattr, sctx);
4411 struct find_xattr_ctx {
4419 static int __find_xattr(int num, struct btrfs_key *di_key,
4420 const char *name, int name_len,
4421 const char *data, int data_len,
4422 u8 type, void *vctx)
4424 struct find_xattr_ctx *ctx = vctx;
4426 if (name_len == ctx->name_len &&
4427 strncmp(name, ctx->name, name_len) == 0) {
4428 ctx->found_idx = num;
4429 ctx->found_data_len = data_len;
4430 ctx->found_data = kmemdup(data, data_len, GFP_KERNEL);
4431 if (!ctx->found_data)
4438 static int find_xattr(struct btrfs_root *root,
4439 struct btrfs_path *path,
4440 struct btrfs_key *key,
4441 const char *name, int name_len,
4442 char **data, int *data_len)
4445 struct find_xattr_ctx ctx;
4448 ctx.name_len = name_len;
4450 ctx.found_data = NULL;
4451 ctx.found_data_len = 0;
4453 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4457 if (ctx.found_idx == -1)
4460 *data = ctx.found_data;
4461 *data_len = ctx.found_data_len;
4463 kfree(ctx.found_data);
4465 return ctx.found_idx;
4469 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4470 const char *name, int name_len,
4471 const char *data, int data_len,
4475 struct send_ctx *sctx = ctx;
4476 char *found_data = NULL;
4477 int found_data_len = 0;
4479 ret = find_xattr(sctx->parent_root, sctx->right_path,
4480 sctx->cmp_key, name, name_len, &found_data,
4482 if (ret == -ENOENT) {
4483 ret = __process_new_xattr(num, di_key, name, name_len, data,
4484 data_len, type, ctx);
4485 } else if (ret >= 0) {
4486 if (data_len != found_data_len ||
4487 memcmp(data, found_data, data_len)) {
4488 ret = __process_new_xattr(num, di_key, name, name_len,
4489 data, data_len, type, ctx);
4499 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4500 const char *name, int name_len,
4501 const char *data, int data_len,
4505 struct send_ctx *sctx = ctx;
4507 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4508 name, name_len, NULL, NULL);
4510 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4511 data_len, type, ctx);
4518 static int process_changed_xattr(struct send_ctx *sctx)
4522 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4523 sctx->cmp_key, __process_changed_new_xattr, sctx);
4526 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4527 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4533 static int process_all_new_xattrs(struct send_ctx *sctx)
4536 struct btrfs_root *root;
4537 struct btrfs_path *path;
4538 struct btrfs_key key;
4539 struct btrfs_key found_key;
4540 struct extent_buffer *eb;
4543 path = alloc_path_for_send();
4547 root = sctx->send_root;
4549 key.objectid = sctx->cmp_key->objectid;
4550 key.type = BTRFS_XATTR_ITEM_KEY;
4552 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4557 eb = path->nodes[0];
4558 slot = path->slots[0];
4559 if (slot >= btrfs_header_nritems(eb)) {
4560 ret = btrfs_next_leaf(root, path);
4563 } else if (ret > 0) {
4570 btrfs_item_key_to_cpu(eb, &found_key, slot);
4571 if (found_key.objectid != key.objectid ||
4572 found_key.type != key.type) {
4577 ret = iterate_dir_item(root, path, &found_key,
4578 __process_new_xattr, sctx);
4586 btrfs_free_path(path);
4590 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4592 struct btrfs_root *root = sctx->send_root;
4593 struct btrfs_fs_info *fs_info = root->fs_info;
4594 struct inode *inode;
4597 struct btrfs_key key;
4598 pgoff_t index = offset >> PAGE_SHIFT;
4600 unsigned pg_offset = offset & ~PAGE_MASK;
4603 key.objectid = sctx->cur_ino;
4604 key.type = BTRFS_INODE_ITEM_KEY;
4607 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4609 return PTR_ERR(inode);
4611 if (offset + len > i_size_read(inode)) {
4612 if (offset > i_size_read(inode))
4615 len = offset - i_size_read(inode);
4620 last_index = (offset + len - 1) >> PAGE_SHIFT;
4622 /* initial readahead */
4623 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4624 file_ra_state_init(&sctx->ra, inode->i_mapping);
4625 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4626 last_index - index + 1);
4628 while (index <= last_index) {
4629 unsigned cur_len = min_t(unsigned, len,
4630 PAGE_SIZE - pg_offset);
4631 page = find_or_create_page(inode->i_mapping, index, GFP_KERNEL);
4637 if (!PageUptodate(page)) {
4638 btrfs_readpage(NULL, page);
4640 if (!PageUptodate(page)) {
4649 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4664 * Read some bytes from the current inode/file and send a write command to
4667 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4669 struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4672 ssize_t num_read = 0;
4674 p = fs_path_alloc();
4678 btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
4680 num_read = fill_read_buf(sctx, offset, len);
4681 if (num_read <= 0) {
4687 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4691 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4695 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4696 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4697 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4699 ret = send_cmd(sctx);
4710 * Send a clone command to user space.
4712 static int send_clone(struct send_ctx *sctx,
4713 u64 offset, u32 len,
4714 struct clone_root *clone_root)
4720 btrfs_debug(sctx->send_root->fs_info,
4721 "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
4722 offset, len, clone_root->root->objectid, clone_root->ino,
4723 clone_root->offset);
4725 p = fs_path_alloc();
4729 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4733 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4737 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4738 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4739 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4741 if (clone_root->root == sctx->send_root) {
4742 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4743 &gen, NULL, NULL, NULL, NULL);
4746 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4748 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4754 * If the parent we're using has a received_uuid set then use that as
4755 * our clone source as that is what we will look for when doing a
4758 * This covers the case that we create a snapshot off of a received
4759 * subvolume and then use that as the parent and try to receive on a
4762 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4763 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4764 clone_root->root->root_item.received_uuid);
4766 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4767 clone_root->root->root_item.uuid);
4768 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4769 le64_to_cpu(clone_root->root->root_item.ctransid));
4770 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4771 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4772 clone_root->offset);
4774 ret = send_cmd(sctx);
4783 * Send an update extent command to user space.
4785 static int send_update_extent(struct send_ctx *sctx,
4786 u64 offset, u32 len)
4791 p = fs_path_alloc();
4795 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4799 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4803 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4804 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4805 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4807 ret = send_cmd(sctx);
4815 static int send_hole(struct send_ctx *sctx, u64 end)
4817 struct fs_path *p = NULL;
4818 u64 offset = sctx->cur_inode_last_extent;
4822 p = fs_path_alloc();
4825 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4827 goto tlv_put_failure;
4828 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4829 while (offset < end) {
4830 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4832 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4835 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4836 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4837 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4838 ret = send_cmd(sctx);
4848 static int send_extent_data(struct send_ctx *sctx,
4854 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4855 return send_update_extent(sctx, offset, len);
4857 while (sent < len) {
4858 u64 size = len - sent;
4861 if (size > BTRFS_SEND_READ_SIZE)
4862 size = BTRFS_SEND_READ_SIZE;
4863 ret = send_write(sctx, offset + sent, size);
4873 static int clone_range(struct send_ctx *sctx,
4874 struct clone_root *clone_root,
4875 const u64 disk_byte,
4880 struct btrfs_path *path;
4881 struct btrfs_key key;
4884 path = alloc_path_for_send();
4889 * We can't send a clone operation for the entire range if we find
4890 * extent items in the respective range in the source file that
4891 * refer to different extents or if we find holes.
4892 * So check for that and do a mix of clone and regular write/copy
4893 * operations if needed.
4897 * mkfs.btrfs -f /dev/sda
4898 * mount /dev/sda /mnt
4899 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4900 * cp --reflink=always /mnt/foo /mnt/bar
4901 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4902 * btrfs subvolume snapshot -r /mnt /mnt/snap
4904 * If when we send the snapshot and we are processing file bar (which
4905 * has a higher inode number than foo) we blindly send a clone operation
4906 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4907 * a file bar that matches the content of file foo - iow, doesn't match
4908 * the content from bar in the original filesystem.
4910 key.objectid = clone_root->ino;
4911 key.type = BTRFS_EXTENT_DATA_KEY;
4912 key.offset = clone_root->offset;
4913 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4916 if (ret > 0 && path->slots[0] > 0) {
4917 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4918 if (key.objectid == clone_root->ino &&
4919 key.type == BTRFS_EXTENT_DATA_KEY)
4924 struct extent_buffer *leaf = path->nodes[0];
4925 int slot = path->slots[0];
4926 struct btrfs_file_extent_item *ei;
4931 if (slot >= btrfs_header_nritems(leaf)) {
4932 ret = btrfs_next_leaf(clone_root->root, path);
4940 btrfs_item_key_to_cpu(leaf, &key, slot);
4943 * We might have an implicit trailing hole (NO_HOLES feature
4944 * enabled). We deal with it after leaving this loop.
4946 if (key.objectid != clone_root->ino ||
4947 key.type != BTRFS_EXTENT_DATA_KEY)
4950 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4951 type = btrfs_file_extent_type(leaf, ei);
4952 if (type == BTRFS_FILE_EXTENT_INLINE) {
4953 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
4954 ext_len = PAGE_ALIGN(ext_len);
4956 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4959 if (key.offset + ext_len <= clone_root->offset)
4962 if (key.offset > clone_root->offset) {
4963 /* Implicit hole, NO_HOLES feature enabled. */
4964 u64 hole_len = key.offset - clone_root->offset;
4968 ret = send_extent_data(sctx, offset, hole_len);
4976 clone_root->offset += hole_len;
4977 data_offset += hole_len;
4980 if (key.offset >= clone_root->offset + len)
4983 clone_len = min_t(u64, ext_len, len);
4985 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4986 btrfs_file_extent_offset(leaf, ei) == data_offset)
4987 ret = send_clone(sctx, offset, clone_len, clone_root);
4989 ret = send_extent_data(sctx, offset, clone_len);
4997 offset += clone_len;
4998 clone_root->offset += clone_len;
4999 data_offset += clone_len;
5005 ret = send_extent_data(sctx, offset, len);
5009 btrfs_free_path(path);
5013 static int send_write_or_clone(struct send_ctx *sctx,
5014 struct btrfs_path *path,
5015 struct btrfs_key *key,
5016 struct clone_root *clone_root)
5019 struct btrfs_file_extent_item *ei;
5020 u64 offset = key->offset;
5023 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5025 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5026 struct btrfs_file_extent_item);
5027 type = btrfs_file_extent_type(path->nodes[0], ei);
5028 if (type == BTRFS_FILE_EXTENT_INLINE) {
5029 len = btrfs_file_extent_inline_len(path->nodes[0],
5030 path->slots[0], ei);
5032 * it is possible the inline item won't cover the whole page,
5033 * but there may be items after this page. Make
5034 * sure to send the whole thing
5036 len = PAGE_ALIGN(len);
5038 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5041 if (offset + len > sctx->cur_inode_size)
5042 len = sctx->cur_inode_size - offset;
5048 if (clone_root && IS_ALIGNED(offset + len, bs)) {
5052 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5053 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5054 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5057 ret = send_extent_data(sctx, offset, len);
5063 static int is_extent_unchanged(struct send_ctx *sctx,
5064 struct btrfs_path *left_path,
5065 struct btrfs_key *ekey)
5068 struct btrfs_key key;
5069 struct btrfs_path *path = NULL;
5070 struct extent_buffer *eb;
5072 struct btrfs_key found_key;
5073 struct btrfs_file_extent_item *ei;
5078 u64 left_offset_fixed;
5086 path = alloc_path_for_send();
5090 eb = left_path->nodes[0];
5091 slot = left_path->slots[0];
5092 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5093 left_type = btrfs_file_extent_type(eb, ei);
5095 if (left_type != BTRFS_FILE_EXTENT_REG) {
5099 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5100 left_len = btrfs_file_extent_num_bytes(eb, ei);
5101 left_offset = btrfs_file_extent_offset(eb, ei);
5102 left_gen = btrfs_file_extent_generation(eb, ei);
5105 * Following comments will refer to these graphics. L is the left
5106 * extents which we are checking at the moment. 1-8 are the right
5107 * extents that we iterate.
5110 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5113 * |--1--|-2b-|...(same as above)
5115 * Alternative situation. Happens on files where extents got split.
5117 * |-----------7-----------|-6-|
5119 * Alternative situation. Happens on files which got larger.
5122 * Nothing follows after 8.
5125 key.objectid = ekey->objectid;
5126 key.type = BTRFS_EXTENT_DATA_KEY;
5127 key.offset = ekey->offset;
5128 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5137 * Handle special case where the right side has no extents at all.
5139 eb = path->nodes[0];
5140 slot = path->slots[0];
5141 btrfs_item_key_to_cpu(eb, &found_key, slot);
5142 if (found_key.objectid != key.objectid ||
5143 found_key.type != key.type) {
5144 /* If we're a hole then just pretend nothing changed */
5145 ret = (left_disknr) ? 0 : 1;
5150 * We're now on 2a, 2b or 7.
5153 while (key.offset < ekey->offset + left_len) {
5154 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5155 right_type = btrfs_file_extent_type(eb, ei);
5156 if (right_type != BTRFS_FILE_EXTENT_REG) {
5161 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5162 right_len = btrfs_file_extent_num_bytes(eb, ei);
5163 right_offset = btrfs_file_extent_offset(eb, ei);
5164 right_gen = btrfs_file_extent_generation(eb, ei);
5167 * Are we at extent 8? If yes, we know the extent is changed.
5168 * This may only happen on the first iteration.
5170 if (found_key.offset + right_len <= ekey->offset) {
5171 /* If we're a hole just pretend nothing changed */
5172 ret = (left_disknr) ? 0 : 1;
5176 left_offset_fixed = left_offset;
5177 if (key.offset < ekey->offset) {
5178 /* Fix the right offset for 2a and 7. */
5179 right_offset += ekey->offset - key.offset;
5181 /* Fix the left offset for all behind 2a and 2b */
5182 left_offset_fixed += key.offset - ekey->offset;
5186 * Check if we have the same extent.
5188 if (left_disknr != right_disknr ||
5189 left_offset_fixed != right_offset ||
5190 left_gen != right_gen) {
5196 * Go to the next extent.
5198 ret = btrfs_next_item(sctx->parent_root, path);
5202 eb = path->nodes[0];
5203 slot = path->slots[0];
5204 btrfs_item_key_to_cpu(eb, &found_key, slot);
5206 if (ret || found_key.objectid != key.objectid ||
5207 found_key.type != key.type) {
5208 key.offset += right_len;
5211 if (found_key.offset != key.offset + right_len) {
5219 * We're now behind the left extent (treat as unchanged) or at the end
5220 * of the right side (treat as changed).
5222 if (key.offset >= ekey->offset + left_len)
5229 btrfs_free_path(path);
5233 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5235 struct btrfs_path *path;
5236 struct btrfs_root *root = sctx->send_root;
5237 struct btrfs_file_extent_item *fi;
5238 struct btrfs_key key;
5243 path = alloc_path_for_send();
5247 sctx->cur_inode_last_extent = 0;
5249 key.objectid = sctx->cur_ino;
5250 key.type = BTRFS_EXTENT_DATA_KEY;
5251 key.offset = offset;
5252 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5256 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5257 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5260 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5261 struct btrfs_file_extent_item);
5262 type = btrfs_file_extent_type(path->nodes[0], fi);
5263 if (type == BTRFS_FILE_EXTENT_INLINE) {
5264 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5265 path->slots[0], fi);
5266 extent_end = ALIGN(key.offset + size,
5267 sctx->send_root->sectorsize);
5269 extent_end = key.offset +
5270 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5272 sctx->cur_inode_last_extent = extent_end;
5274 btrfs_free_path(path);
5278 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5279 struct btrfs_key *key)
5281 struct btrfs_file_extent_item *fi;
5286 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5289 if (sctx->cur_inode_last_extent == (u64)-1) {
5290 ret = get_last_extent(sctx, key->offset - 1);
5295 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5296 struct btrfs_file_extent_item);
5297 type = btrfs_file_extent_type(path->nodes[0], fi);
5298 if (type == BTRFS_FILE_EXTENT_INLINE) {
5299 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5300 path->slots[0], fi);
5301 extent_end = ALIGN(key->offset + size,
5302 sctx->send_root->sectorsize);
5304 extent_end = key->offset +
5305 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5308 if (path->slots[0] == 0 &&
5309 sctx->cur_inode_last_extent < key->offset) {
5311 * We might have skipped entire leafs that contained only
5312 * file extent items for our current inode. These leafs have
5313 * a generation number smaller (older) than the one in the
5314 * current leaf and the leaf our last extent came from, and
5315 * are located between these 2 leafs.
5317 ret = get_last_extent(sctx, key->offset - 1);
5322 if (sctx->cur_inode_last_extent < key->offset)
5323 ret = send_hole(sctx, key->offset);
5324 sctx->cur_inode_last_extent = extent_end;
5328 static int process_extent(struct send_ctx *sctx,
5329 struct btrfs_path *path,
5330 struct btrfs_key *key)
5332 struct clone_root *found_clone = NULL;
5335 if (S_ISLNK(sctx->cur_inode_mode))
5338 if (sctx->parent_root && !sctx->cur_inode_new) {
5339 ret = is_extent_unchanged(sctx, path, key);
5347 struct btrfs_file_extent_item *ei;
5350 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5351 struct btrfs_file_extent_item);
5352 type = btrfs_file_extent_type(path->nodes[0], ei);
5353 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5354 type == BTRFS_FILE_EXTENT_REG) {
5356 * The send spec does not have a prealloc command yet,
5357 * so just leave a hole for prealloc'ed extents until
5358 * we have enough commands queued up to justify rev'ing
5361 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5366 /* Have a hole, just skip it. */
5367 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5374 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5375 sctx->cur_inode_size, &found_clone);
5376 if (ret != -ENOENT && ret < 0)
5379 ret = send_write_or_clone(sctx, path, key, found_clone);
5383 ret = maybe_send_hole(sctx, path, key);
5388 static int process_all_extents(struct send_ctx *sctx)
5391 struct btrfs_root *root;
5392 struct btrfs_path *path;
5393 struct btrfs_key key;
5394 struct btrfs_key found_key;
5395 struct extent_buffer *eb;
5398 root = sctx->send_root;
5399 path = alloc_path_for_send();
5403 key.objectid = sctx->cmp_key->objectid;
5404 key.type = BTRFS_EXTENT_DATA_KEY;
5406 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5411 eb = path->nodes[0];
5412 slot = path->slots[0];
5414 if (slot >= btrfs_header_nritems(eb)) {
5415 ret = btrfs_next_leaf(root, path);
5418 } else if (ret > 0) {
5425 btrfs_item_key_to_cpu(eb, &found_key, slot);
5427 if (found_key.objectid != key.objectid ||
5428 found_key.type != key.type) {
5433 ret = process_extent(sctx, path, &found_key);
5441 btrfs_free_path(path);
5445 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5447 int *refs_processed)
5451 if (sctx->cur_ino == 0)
5453 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5454 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5456 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5459 ret = process_recorded_refs(sctx, pending_move);
5463 *refs_processed = 1;
5468 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5479 int pending_move = 0;
5480 int refs_processed = 0;
5482 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5488 * We have processed the refs and thus need to advance send_progress.
5489 * Now, calls to get_cur_xxx will take the updated refs of the current
5490 * inode into account.
5492 * On the other hand, if our current inode is a directory and couldn't
5493 * be moved/renamed because its parent was renamed/moved too and it has
5494 * a higher inode number, we can only move/rename our current inode
5495 * after we moved/renamed its parent. Therefore in this case operate on
5496 * the old path (pre move/rename) of our current inode, and the
5497 * move/rename will be performed later.
5499 if (refs_processed && !pending_move)
5500 sctx->send_progress = sctx->cur_ino + 1;
5502 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5504 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5507 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5508 &left_mode, &left_uid, &left_gid, NULL);
5512 if (!sctx->parent_root || sctx->cur_inode_new) {
5514 if (!S_ISLNK(sctx->cur_inode_mode))
5517 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5518 NULL, NULL, &right_mode, &right_uid,
5523 if (left_uid != right_uid || left_gid != right_gid)
5525 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5529 if (S_ISREG(sctx->cur_inode_mode)) {
5530 if (need_send_hole(sctx)) {
5531 if (sctx->cur_inode_last_extent == (u64)-1 ||
5532 sctx->cur_inode_last_extent <
5533 sctx->cur_inode_size) {
5534 ret = get_last_extent(sctx, (u64)-1);
5538 if (sctx->cur_inode_last_extent <
5539 sctx->cur_inode_size) {
5540 ret = send_hole(sctx, sctx->cur_inode_size);
5545 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5546 sctx->cur_inode_size);
5552 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5553 left_uid, left_gid);
5558 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5565 * If other directory inodes depended on our current directory
5566 * inode's move/rename, now do their move/rename operations.
5568 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5569 ret = apply_children_dir_moves(sctx);
5573 * Need to send that every time, no matter if it actually
5574 * changed between the two trees as we have done changes to
5575 * the inode before. If our inode is a directory and it's
5576 * waiting to be moved/renamed, we will send its utimes when
5577 * it's moved/renamed, therefore we don't need to do it here.
5579 sctx->send_progress = sctx->cur_ino + 1;
5580 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5589 static int changed_inode(struct send_ctx *sctx,
5590 enum btrfs_compare_tree_result result)
5593 struct btrfs_key *key = sctx->cmp_key;
5594 struct btrfs_inode_item *left_ii = NULL;
5595 struct btrfs_inode_item *right_ii = NULL;
5599 sctx->cur_ino = key->objectid;
5600 sctx->cur_inode_new_gen = 0;
5601 sctx->cur_inode_last_extent = (u64)-1;
5604 * Set send_progress to current inode. This will tell all get_cur_xxx
5605 * functions that the current inode's refs are not updated yet. Later,
5606 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5608 sctx->send_progress = sctx->cur_ino;
5610 if (result == BTRFS_COMPARE_TREE_NEW ||
5611 result == BTRFS_COMPARE_TREE_CHANGED) {
5612 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5613 sctx->left_path->slots[0],
5614 struct btrfs_inode_item);
5615 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5618 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5619 sctx->right_path->slots[0],
5620 struct btrfs_inode_item);
5621 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5624 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5625 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5626 sctx->right_path->slots[0],
5627 struct btrfs_inode_item);
5629 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5633 * The cur_ino = root dir case is special here. We can't treat
5634 * the inode as deleted+reused because it would generate a
5635 * stream that tries to delete/mkdir the root dir.
5637 if (left_gen != right_gen &&
5638 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5639 sctx->cur_inode_new_gen = 1;
5642 if (result == BTRFS_COMPARE_TREE_NEW) {
5643 sctx->cur_inode_gen = left_gen;
5644 sctx->cur_inode_new = 1;
5645 sctx->cur_inode_deleted = 0;
5646 sctx->cur_inode_size = btrfs_inode_size(
5647 sctx->left_path->nodes[0], left_ii);
5648 sctx->cur_inode_mode = btrfs_inode_mode(
5649 sctx->left_path->nodes[0], left_ii);
5650 sctx->cur_inode_rdev = btrfs_inode_rdev(
5651 sctx->left_path->nodes[0], left_ii);
5652 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5653 ret = send_create_inode_if_needed(sctx);
5654 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5655 sctx->cur_inode_gen = right_gen;
5656 sctx->cur_inode_new = 0;
5657 sctx->cur_inode_deleted = 1;
5658 sctx->cur_inode_size = btrfs_inode_size(
5659 sctx->right_path->nodes[0], right_ii);
5660 sctx->cur_inode_mode = btrfs_inode_mode(
5661 sctx->right_path->nodes[0], right_ii);
5662 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5664 * We need to do some special handling in case the inode was
5665 * reported as changed with a changed generation number. This
5666 * means that the original inode was deleted and new inode
5667 * reused the same inum. So we have to treat the old inode as
5668 * deleted and the new one as new.
5670 if (sctx->cur_inode_new_gen) {
5672 * First, process the inode as if it was deleted.
5674 sctx->cur_inode_gen = right_gen;
5675 sctx->cur_inode_new = 0;
5676 sctx->cur_inode_deleted = 1;
5677 sctx->cur_inode_size = btrfs_inode_size(
5678 sctx->right_path->nodes[0], right_ii);
5679 sctx->cur_inode_mode = btrfs_inode_mode(
5680 sctx->right_path->nodes[0], right_ii);
5681 ret = process_all_refs(sctx,
5682 BTRFS_COMPARE_TREE_DELETED);
5687 * Now process the inode as if it was new.
5689 sctx->cur_inode_gen = left_gen;
5690 sctx->cur_inode_new = 1;
5691 sctx->cur_inode_deleted = 0;
5692 sctx->cur_inode_size = btrfs_inode_size(
5693 sctx->left_path->nodes[0], left_ii);
5694 sctx->cur_inode_mode = btrfs_inode_mode(
5695 sctx->left_path->nodes[0], left_ii);
5696 sctx->cur_inode_rdev = btrfs_inode_rdev(
5697 sctx->left_path->nodes[0], left_ii);
5698 ret = send_create_inode_if_needed(sctx);
5702 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5706 * Advance send_progress now as we did not get into
5707 * process_recorded_refs_if_needed in the new_gen case.
5709 sctx->send_progress = sctx->cur_ino + 1;
5712 * Now process all extents and xattrs of the inode as if
5713 * they were all new.
5715 ret = process_all_extents(sctx);
5718 ret = process_all_new_xattrs(sctx);
5722 sctx->cur_inode_gen = left_gen;
5723 sctx->cur_inode_new = 0;
5724 sctx->cur_inode_new_gen = 0;
5725 sctx->cur_inode_deleted = 0;
5726 sctx->cur_inode_size = btrfs_inode_size(
5727 sctx->left_path->nodes[0], left_ii);
5728 sctx->cur_inode_mode = btrfs_inode_mode(
5729 sctx->left_path->nodes[0], left_ii);
5738 * We have to process new refs before deleted refs, but compare_trees gives us
5739 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5740 * first and later process them in process_recorded_refs.
5741 * For the cur_inode_new_gen case, we skip recording completely because
5742 * changed_inode did already initiate processing of refs. The reason for this is
5743 * that in this case, compare_tree actually compares the refs of 2 different
5744 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5745 * refs of the right tree as deleted and all refs of the left tree as new.
5747 static int changed_ref(struct send_ctx *sctx,
5748 enum btrfs_compare_tree_result result)
5752 if (sctx->cur_ino != sctx->cmp_key->objectid) {
5753 inconsistent_snapshot_error(sctx, result, "reference");
5757 if (!sctx->cur_inode_new_gen &&
5758 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5759 if (result == BTRFS_COMPARE_TREE_NEW)
5760 ret = record_new_ref(sctx);
5761 else if (result == BTRFS_COMPARE_TREE_DELETED)
5762 ret = record_deleted_ref(sctx);
5763 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5764 ret = record_changed_ref(sctx);
5771 * Process new/deleted/changed xattrs. We skip processing in the
5772 * cur_inode_new_gen case because changed_inode did already initiate processing
5773 * of xattrs. The reason is the same as in changed_ref
5775 static int changed_xattr(struct send_ctx *sctx,
5776 enum btrfs_compare_tree_result result)
5780 if (sctx->cur_ino != sctx->cmp_key->objectid) {
5781 inconsistent_snapshot_error(sctx, result, "xattr");
5785 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5786 if (result == BTRFS_COMPARE_TREE_NEW)
5787 ret = process_new_xattr(sctx);
5788 else if (result == BTRFS_COMPARE_TREE_DELETED)
5789 ret = process_deleted_xattr(sctx);
5790 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5791 ret = process_changed_xattr(sctx);
5798 * Process new/deleted/changed extents. We skip processing in the
5799 * cur_inode_new_gen case because changed_inode did already initiate processing
5800 * of extents. The reason is the same as in changed_ref
5802 static int changed_extent(struct send_ctx *sctx,
5803 enum btrfs_compare_tree_result result)
5807 if (sctx->cur_ino != sctx->cmp_key->objectid) {
5809 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5810 struct extent_buffer *leaf_l;
5811 struct extent_buffer *leaf_r;
5812 struct btrfs_file_extent_item *ei_l;
5813 struct btrfs_file_extent_item *ei_r;
5815 leaf_l = sctx->left_path->nodes[0];
5816 leaf_r = sctx->right_path->nodes[0];
5817 ei_l = btrfs_item_ptr(leaf_l,
5818 sctx->left_path->slots[0],
5819 struct btrfs_file_extent_item);
5820 ei_r = btrfs_item_ptr(leaf_r,
5821 sctx->right_path->slots[0],
5822 struct btrfs_file_extent_item);
5825 * We may have found an extent item that has changed
5826 * only its disk_bytenr field and the corresponding
5827 * inode item was not updated. This case happens due to
5828 * very specific timings during relocation when a leaf
5829 * that contains file extent items is COWed while
5830 * relocation is ongoing and its in the stage where it
5831 * updates data pointers. So when this happens we can
5832 * safely ignore it since we know it's the same extent,
5833 * but just at different logical and physical locations
5834 * (when an extent is fully replaced with a new one, we
5835 * know the generation number must have changed too,
5836 * since snapshot creation implies committing the current
5837 * transaction, and the inode item must have been updated
5839 * This replacement of the disk_bytenr happens at
5840 * relocation.c:replace_file_extents() through
5841 * relocation.c:btrfs_reloc_cow_block().
5843 if (btrfs_file_extent_generation(leaf_l, ei_l) ==
5844 btrfs_file_extent_generation(leaf_r, ei_r) &&
5845 btrfs_file_extent_ram_bytes(leaf_l, ei_l) ==
5846 btrfs_file_extent_ram_bytes(leaf_r, ei_r) &&
5847 btrfs_file_extent_compression(leaf_l, ei_l) ==
5848 btrfs_file_extent_compression(leaf_r, ei_r) &&
5849 btrfs_file_extent_encryption(leaf_l, ei_l) ==
5850 btrfs_file_extent_encryption(leaf_r, ei_r) &&
5851 btrfs_file_extent_other_encoding(leaf_l, ei_l) ==
5852 btrfs_file_extent_other_encoding(leaf_r, ei_r) &&
5853 btrfs_file_extent_type(leaf_l, ei_l) ==
5854 btrfs_file_extent_type(leaf_r, ei_r) &&
5855 btrfs_file_extent_disk_bytenr(leaf_l, ei_l) !=
5856 btrfs_file_extent_disk_bytenr(leaf_r, ei_r) &&
5857 btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) ==
5858 btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) &&
5859 btrfs_file_extent_offset(leaf_l, ei_l) ==
5860 btrfs_file_extent_offset(leaf_r, ei_r) &&
5861 btrfs_file_extent_num_bytes(leaf_l, ei_l) ==
5862 btrfs_file_extent_num_bytes(leaf_r, ei_r))
5866 inconsistent_snapshot_error(sctx, result, "extent");
5870 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5871 if (result != BTRFS_COMPARE_TREE_DELETED)
5872 ret = process_extent(sctx, sctx->left_path,
5879 static int dir_changed(struct send_ctx *sctx, u64 dir)
5881 u64 orig_gen, new_gen;
5884 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5889 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5894 return (orig_gen != new_gen) ? 1 : 0;
5897 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5898 struct btrfs_key *key)
5900 struct btrfs_inode_extref *extref;
5901 struct extent_buffer *leaf;
5902 u64 dirid = 0, last_dirid = 0;
5909 /* Easy case, just check this one dirid */
5910 if (key->type == BTRFS_INODE_REF_KEY) {
5911 dirid = key->offset;
5913 ret = dir_changed(sctx, dirid);
5917 leaf = path->nodes[0];
5918 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5919 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5920 while (cur_offset < item_size) {
5921 extref = (struct btrfs_inode_extref *)(ptr +
5923 dirid = btrfs_inode_extref_parent(leaf, extref);
5924 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5925 cur_offset += ref_name_len + sizeof(*extref);
5926 if (dirid == last_dirid)
5928 ret = dir_changed(sctx, dirid);
5938 * Updates compare related fields in sctx and simply forwards to the actual
5939 * changed_xxx functions.
5941 static int changed_cb(struct btrfs_root *left_root,
5942 struct btrfs_root *right_root,
5943 struct btrfs_path *left_path,
5944 struct btrfs_path *right_path,
5945 struct btrfs_key *key,
5946 enum btrfs_compare_tree_result result,
5950 struct send_ctx *sctx = ctx;
5952 if (result == BTRFS_COMPARE_TREE_SAME) {
5953 if (key->type == BTRFS_INODE_REF_KEY ||
5954 key->type == BTRFS_INODE_EXTREF_KEY) {
5955 ret = compare_refs(sctx, left_path, key);
5960 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5961 return maybe_send_hole(sctx, left_path, key);
5965 result = BTRFS_COMPARE_TREE_CHANGED;
5969 sctx->left_path = left_path;
5970 sctx->right_path = right_path;
5971 sctx->cmp_key = key;
5973 ret = finish_inode_if_needed(sctx, 0);
5977 /* Ignore non-FS objects */
5978 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5979 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5982 if (key->type == BTRFS_INODE_ITEM_KEY)
5983 ret = changed_inode(sctx, result);
5984 else if (key->type == BTRFS_INODE_REF_KEY ||
5985 key->type == BTRFS_INODE_EXTREF_KEY)
5986 ret = changed_ref(sctx, result);
5987 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5988 ret = changed_xattr(sctx, result);
5989 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5990 ret = changed_extent(sctx, result);
5996 static int full_send_tree(struct send_ctx *sctx)
5999 struct btrfs_root *send_root = sctx->send_root;
6000 struct btrfs_key key;
6001 struct btrfs_key found_key;
6002 struct btrfs_path *path;
6003 struct extent_buffer *eb;
6006 path = alloc_path_for_send();
6010 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
6011 key.type = BTRFS_INODE_ITEM_KEY;
6014 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
6021 eb = path->nodes[0];
6022 slot = path->slots[0];
6023 btrfs_item_key_to_cpu(eb, &found_key, slot);
6025 ret = changed_cb(send_root, NULL, path, NULL,
6026 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
6030 key.objectid = found_key.objectid;
6031 key.type = found_key.type;
6032 key.offset = found_key.offset + 1;
6034 ret = btrfs_next_item(send_root, path);
6044 ret = finish_inode_if_needed(sctx, 1);
6047 btrfs_free_path(path);
6051 static int send_subvol(struct send_ctx *sctx)
6055 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
6056 ret = send_header(sctx);
6061 ret = send_subvol_begin(sctx);
6065 if (sctx->parent_root) {
6066 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6070 ret = finish_inode_if_needed(sctx, 1);
6074 ret = full_send_tree(sctx);
6080 free_recorded_refs(sctx);
6085 * If orphan cleanup did remove any orphans from a root, it means the tree
6086 * was modified and therefore the commit root is not the same as the current
6087 * root anymore. This is a problem, because send uses the commit root and
6088 * therefore can see inode items that don't exist in the current root anymore,
6089 * and for example make calls to btrfs_iget, which will do tree lookups based
6090 * on the current root and not on the commit root. Those lookups will fail,
6091 * returning a -ESTALE error, and making send fail with that error. So make
6092 * sure a send does not see any orphans we have just removed, and that it will
6093 * see the same inodes regardless of whether a transaction commit happened
6094 * before it started (meaning that the commit root will be the same as the
6095 * current root) or not.
6097 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
6100 struct btrfs_trans_handle *trans = NULL;
6103 if (sctx->parent_root &&
6104 sctx->parent_root->node != sctx->parent_root->commit_root)
6107 for (i = 0; i < sctx->clone_roots_cnt; i++)
6108 if (sctx->clone_roots[i].root->node !=
6109 sctx->clone_roots[i].root->commit_root)
6113 return btrfs_end_transaction(trans, sctx->send_root);
6118 /* Use any root, all fs roots will get their commit roots updated. */
6120 trans = btrfs_join_transaction(sctx->send_root);
6122 return PTR_ERR(trans);
6126 return btrfs_commit_transaction(trans, sctx->send_root);
6129 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6131 spin_lock(&root->root_item_lock);
6132 root->send_in_progress--;
6134 * Not much left to do, we don't know why it's unbalanced and
6135 * can't blindly reset it to 0.
6137 if (root->send_in_progress < 0)
6138 btrfs_err(root->fs_info,
6139 "send_in_progres unbalanced %d root %llu",
6140 root->send_in_progress, root->root_key.objectid);
6141 spin_unlock(&root->root_item_lock);
6144 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
6147 struct btrfs_root *send_root;
6148 struct btrfs_root *clone_root;
6149 struct btrfs_fs_info *fs_info;
6150 struct btrfs_ioctl_send_args *arg = NULL;
6151 struct btrfs_key key;
6152 struct send_ctx *sctx = NULL;
6154 u64 *clone_sources_tmp = NULL;
6155 int clone_sources_to_rollback = 0;
6156 unsigned alloc_size;
6157 int sort_clone_roots = 0;
6160 if (!capable(CAP_SYS_ADMIN))
6163 send_root = BTRFS_I(file_inode(mnt_file))->root;
6164 fs_info = send_root->fs_info;
6167 * The subvolume must remain read-only during send, protect against
6168 * making it RW. This also protects against deletion.
6170 spin_lock(&send_root->root_item_lock);
6171 send_root->send_in_progress++;
6172 spin_unlock(&send_root->root_item_lock);
6175 * This is done when we lookup the root, it should already be complete
6176 * by the time we get here.
6178 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6181 * Userspace tools do the checks and warn the user if it's
6184 if (!btrfs_root_readonly(send_root)) {
6189 arg = memdup_user(arg_, sizeof(*arg));
6196 if (arg->clone_sources_count >
6197 ULLONG_MAX / sizeof(*arg->clone_sources)) {
6202 if (!access_ok(VERIFY_READ, arg->clone_sources,
6203 sizeof(*arg->clone_sources) *
6204 arg->clone_sources_count)) {
6209 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6214 sctx = kzalloc(sizeof(struct send_ctx), GFP_KERNEL);
6220 INIT_LIST_HEAD(&sctx->new_refs);
6221 INIT_LIST_HEAD(&sctx->deleted_refs);
6222 INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL);
6223 INIT_LIST_HEAD(&sctx->name_cache_list);
6225 sctx->flags = arg->flags;
6227 sctx->send_filp = fget(arg->send_fd);
6228 if (!sctx->send_filp) {
6233 sctx->send_root = send_root;
6235 * Unlikely but possible, if the subvolume is marked for deletion but
6236 * is slow to remove the directory entry, send can still be started
6238 if (btrfs_root_dead(sctx->send_root)) {
6243 sctx->clone_roots_cnt = arg->clone_sources_count;
6245 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6246 sctx->send_buf = kmalloc(sctx->send_max_size, GFP_KERNEL | __GFP_NOWARN);
6247 if (!sctx->send_buf) {
6248 sctx->send_buf = vmalloc(sctx->send_max_size);
6249 if (!sctx->send_buf) {
6255 sctx->read_buf = kmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL | __GFP_NOWARN);
6256 if (!sctx->read_buf) {
6257 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6258 if (!sctx->read_buf) {
6264 sctx->pending_dir_moves = RB_ROOT;
6265 sctx->waiting_dir_moves = RB_ROOT;
6266 sctx->orphan_dirs = RB_ROOT;
6268 alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6270 sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
6271 if (!sctx->clone_roots) {
6272 sctx->clone_roots = vzalloc(alloc_size);
6273 if (!sctx->clone_roots) {
6279 alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
6281 if (arg->clone_sources_count) {
6282 clone_sources_tmp = kmalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN);
6283 if (!clone_sources_tmp) {
6284 clone_sources_tmp = vmalloc(alloc_size);
6285 if (!clone_sources_tmp) {
6291 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6298 for (i = 0; i < arg->clone_sources_count; i++) {
6299 key.objectid = clone_sources_tmp[i];
6300 key.type = BTRFS_ROOT_ITEM_KEY;
6301 key.offset = (u64)-1;
6303 index = srcu_read_lock(&fs_info->subvol_srcu);
6305 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6306 if (IS_ERR(clone_root)) {
6307 srcu_read_unlock(&fs_info->subvol_srcu, index);
6308 ret = PTR_ERR(clone_root);
6311 spin_lock(&clone_root->root_item_lock);
6312 if (!btrfs_root_readonly(clone_root) ||
6313 btrfs_root_dead(clone_root)) {
6314 spin_unlock(&clone_root->root_item_lock);
6315 srcu_read_unlock(&fs_info->subvol_srcu, index);
6319 clone_root->send_in_progress++;
6320 spin_unlock(&clone_root->root_item_lock);
6321 srcu_read_unlock(&fs_info->subvol_srcu, index);
6323 sctx->clone_roots[i].root = clone_root;
6324 clone_sources_to_rollback = i + 1;
6326 kvfree(clone_sources_tmp);
6327 clone_sources_tmp = NULL;
6330 if (arg->parent_root) {
6331 key.objectid = arg->parent_root;
6332 key.type = BTRFS_ROOT_ITEM_KEY;
6333 key.offset = (u64)-1;
6335 index = srcu_read_lock(&fs_info->subvol_srcu);
6337 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6338 if (IS_ERR(sctx->parent_root)) {
6339 srcu_read_unlock(&fs_info->subvol_srcu, index);
6340 ret = PTR_ERR(sctx->parent_root);
6344 spin_lock(&sctx->parent_root->root_item_lock);
6345 sctx->parent_root->send_in_progress++;
6346 if (!btrfs_root_readonly(sctx->parent_root) ||
6347 btrfs_root_dead(sctx->parent_root)) {
6348 spin_unlock(&sctx->parent_root->root_item_lock);
6349 srcu_read_unlock(&fs_info->subvol_srcu, index);
6353 spin_unlock(&sctx->parent_root->root_item_lock);
6355 srcu_read_unlock(&fs_info->subvol_srcu, index);
6359 * Clones from send_root are allowed, but only if the clone source
6360 * is behind the current send position. This is checked while searching
6361 * for possible clone sources.
6363 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6365 /* We do a bsearch later */
6366 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6367 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6369 sort_clone_roots = 1;
6371 ret = ensure_commit_roots_uptodate(sctx);
6375 current->journal_info = BTRFS_SEND_TRANS_STUB;
6376 ret = send_subvol(sctx);
6377 current->journal_info = NULL;
6381 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6382 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6385 ret = send_cmd(sctx);
6391 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6392 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6394 struct pending_dir_move *pm;
6396 n = rb_first(&sctx->pending_dir_moves);
6397 pm = rb_entry(n, struct pending_dir_move, node);
6398 while (!list_empty(&pm->list)) {
6399 struct pending_dir_move *pm2;
6401 pm2 = list_first_entry(&pm->list,
6402 struct pending_dir_move, list);
6403 free_pending_move(sctx, pm2);
6405 free_pending_move(sctx, pm);
6408 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6409 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6411 struct waiting_dir_move *dm;
6413 n = rb_first(&sctx->waiting_dir_moves);
6414 dm = rb_entry(n, struct waiting_dir_move, node);
6415 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6419 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6420 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6422 struct orphan_dir_info *odi;
6424 n = rb_first(&sctx->orphan_dirs);
6425 odi = rb_entry(n, struct orphan_dir_info, node);
6426 free_orphan_dir_info(sctx, odi);
6429 if (sort_clone_roots) {
6430 for (i = 0; i < sctx->clone_roots_cnt; i++)
6431 btrfs_root_dec_send_in_progress(
6432 sctx->clone_roots[i].root);
6434 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6435 btrfs_root_dec_send_in_progress(
6436 sctx->clone_roots[i].root);
6438 btrfs_root_dec_send_in_progress(send_root);
6440 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6441 btrfs_root_dec_send_in_progress(sctx->parent_root);
6444 kvfree(clone_sources_tmp);
6447 if (sctx->send_filp)
6448 fput(sctx->send_filp);
6450 kvfree(sctx->clone_roots);
6451 kvfree(sctx->send_buf);
6452 kvfree(sctx->read_buf);
6454 name_cache_free(sctx);