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"
39 static int g_verbose = 0;
41 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
44 * A fs_path is a helper to dynamically build path names with unknown size.
45 * It reallocates the internal buffer on demand.
46 * It allows fast adding of path elements on the right side (normal path) and
47 * fast adding to the left side (reversed path). A reversed path can also be
48 * unreversed if needed.
57 unsigned short buf_len:15;
58 unsigned short reversed:1;
62 * Average path length does not exceed 200 bytes, we'll have
63 * better packing in the slab and higher chance to satisfy
64 * a allocation later during send.
69 #define FS_PATH_INLINE_SIZE \
70 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
73 /* reused for each extent */
75 struct btrfs_root *root;
82 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
83 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
86 struct file *send_filp;
92 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
93 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
95 struct btrfs_root *send_root;
96 struct btrfs_root *parent_root;
97 struct clone_root *clone_roots;
100 /* current state of the compare_tree call */
101 struct btrfs_path *left_path;
102 struct btrfs_path *right_path;
103 struct btrfs_key *cmp_key;
106 * infos of the currently processed inode. In case of deleted inodes,
107 * these are the values from the deleted inode.
112 int cur_inode_new_gen;
113 int cur_inode_deleted;
117 u64 cur_inode_last_extent;
121 struct list_head new_refs;
122 struct list_head deleted_refs;
124 struct radix_tree_root name_cache;
125 struct list_head name_cache_list;
128 struct file_ra_state ra;
133 * We process inodes by their increasing order, so if before an
134 * incremental send we reverse the parent/child relationship of
135 * directories such that a directory with a lower inode number was
136 * the parent of a directory with a higher inode number, and the one
137 * becoming the new parent got renamed too, we can't rename/move the
138 * directory with lower inode number when we finish processing it - we
139 * must process the directory with higher inode number first, then
140 * rename/move it and then rename/move the directory with lower inode
141 * number. Example follows.
143 * Tree state when the first send was performed:
155 * Tree state when the second (incremental) send is performed:
164 * The sequence of steps that lead to the second state was:
166 * mv /a/b/c/d /a/b/c2/d2
167 * mv /a/b/c /a/b/c2/d2/cc
169 * "c" has lower inode number, but we can't move it (2nd mv operation)
170 * before we move "d", which has higher inode number.
172 * So we just memorize which move/rename operations must be performed
173 * later when their respective parent is processed and moved/renamed.
176 /* Indexed by parent directory inode number. */
177 struct rb_root pending_dir_moves;
180 * Reverse index, indexed by the inode number of a directory that
181 * is waiting for the move/rename of its immediate parent before its
182 * own move/rename can be performed.
184 struct rb_root waiting_dir_moves;
187 * A directory that is going to be rm'ed might have a child directory
188 * which is in the pending directory moves index above. In this case,
189 * the directory can only be removed after the move/rename of its child
190 * is performed. Example:
210 * Sequence of steps that lead to the send snapshot:
211 * rm -f /a/b/c/foo.txt
213 * mv /a/b/c/x /a/b/YY
216 * When the child is processed, its move/rename is delayed until its
217 * parent is processed (as explained above), but all other operations
218 * like update utimes, chown, chgrp, etc, are performed and the paths
219 * that it uses for those operations must use the orphanized name of
220 * its parent (the directory we're going to rm later), so we need to
221 * memorize that name.
223 * Indexed by the inode number of the directory to be deleted.
225 struct rb_root orphan_dirs;
228 struct pending_dir_move {
230 struct list_head list;
235 struct list_head update_refs;
238 struct waiting_dir_move {
242 * There might be some directory that could not be removed because it
243 * was waiting for this directory inode to be moved first. Therefore
244 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
250 struct orphan_dir_info {
256 struct name_cache_entry {
257 struct list_head list;
259 * radix_tree has only 32bit entries but we need to handle 64bit inums.
260 * We use the lower 32bit of the 64bit inum to store it in the tree. If
261 * more then one inum would fall into the same entry, we use radix_list
262 * to store the additional entries. radix_list is also used to store
263 * entries where two entries have the same inum but different
266 struct list_head radix_list;
272 int need_later_update;
277 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
279 static struct waiting_dir_move *
280 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
282 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
284 static int need_send_hole(struct send_ctx *sctx)
286 return (sctx->parent_root && !sctx->cur_inode_new &&
287 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
288 S_ISREG(sctx->cur_inode_mode));
291 static void fs_path_reset(struct fs_path *p)
294 p->start = p->buf + p->buf_len - 1;
304 static struct fs_path *fs_path_alloc(void)
308 p = kmalloc(sizeof(*p), GFP_NOFS);
312 p->buf = p->inline_buf;
313 p->buf_len = FS_PATH_INLINE_SIZE;
318 static struct fs_path *fs_path_alloc_reversed(void)
330 static void fs_path_free(struct fs_path *p)
334 if (p->buf != p->inline_buf)
339 static int fs_path_len(struct fs_path *p)
341 return p->end - p->start;
344 static int fs_path_ensure_buf(struct fs_path *p, int len)
352 if (p->buf_len >= len)
355 if (len > PATH_MAX) {
360 path_len = p->end - p->start;
361 old_buf_len = p->buf_len;
364 * First time the inline_buf does not suffice
366 if (p->buf == p->inline_buf) {
367 tmp_buf = kmalloc(len, GFP_NOFS);
369 memcpy(tmp_buf, p->buf, old_buf_len);
371 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
377 * The real size of the buffer is bigger, this will let the fast path
378 * happen most of the time
380 p->buf_len = ksize(p->buf);
383 tmp_buf = p->buf + old_buf_len - path_len - 1;
384 p->end = p->buf + p->buf_len - 1;
385 p->start = p->end - path_len;
386 memmove(p->start, tmp_buf, path_len + 1);
389 p->end = p->start + path_len;
394 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
400 new_len = p->end - p->start + name_len;
401 if (p->start != p->end)
403 ret = fs_path_ensure_buf(p, new_len);
408 if (p->start != p->end)
410 p->start -= name_len;
411 *prepared = p->start;
413 if (p->start != p->end)
424 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
429 ret = fs_path_prepare_for_add(p, name_len, &prepared);
432 memcpy(prepared, name, name_len);
438 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
443 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
446 memcpy(prepared, p2->start, p2->end - p2->start);
452 static int fs_path_add_from_extent_buffer(struct fs_path *p,
453 struct extent_buffer *eb,
454 unsigned long off, int len)
459 ret = fs_path_prepare_for_add(p, len, &prepared);
463 read_extent_buffer(eb, prepared, off, len);
469 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
473 p->reversed = from->reversed;
476 ret = fs_path_add_path(p, from);
482 static void fs_path_unreverse(struct fs_path *p)
491 len = p->end - p->start;
493 p->end = p->start + len;
494 memmove(p->start, tmp, len + 1);
498 static struct btrfs_path *alloc_path_for_send(void)
500 struct btrfs_path *path;
502 path = btrfs_alloc_path();
505 path->search_commit_root = 1;
506 path->skip_locking = 1;
507 path->need_commit_sem = 1;
511 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
521 ret = vfs_write(filp, (__force const char __user *)buf + pos,
523 /* TODO handle that correctly */
524 /*if (ret == -ERESTARTSYS) {
543 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
545 struct btrfs_tlv_header *hdr;
546 int total_len = sizeof(*hdr) + len;
547 int left = sctx->send_max_size - sctx->send_size;
549 if (unlikely(left < total_len))
552 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
553 hdr->tlv_type = cpu_to_le16(attr);
554 hdr->tlv_len = cpu_to_le16(len);
555 memcpy(hdr + 1, data, len);
556 sctx->send_size += total_len;
561 #define TLV_PUT_DEFINE_INT(bits) \
562 static int tlv_put_u##bits(struct send_ctx *sctx, \
563 u##bits attr, u##bits value) \
565 __le##bits __tmp = cpu_to_le##bits(value); \
566 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
569 TLV_PUT_DEFINE_INT(64)
571 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
572 const char *str, int len)
576 return tlv_put(sctx, attr, str, len);
579 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
582 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
585 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
586 struct extent_buffer *eb,
587 struct btrfs_timespec *ts)
589 struct btrfs_timespec bts;
590 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
591 return tlv_put(sctx, attr, &bts, sizeof(bts));
595 #define TLV_PUT(sctx, attrtype, attrlen, data) \
597 ret = tlv_put(sctx, attrtype, attrlen, data); \
599 goto tlv_put_failure; \
602 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
604 ret = tlv_put_u##bits(sctx, attrtype, value); \
606 goto tlv_put_failure; \
609 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
610 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
611 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
612 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
613 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
615 ret = tlv_put_string(sctx, attrtype, str, len); \
617 goto tlv_put_failure; \
619 #define TLV_PUT_PATH(sctx, attrtype, p) \
621 ret = tlv_put_string(sctx, attrtype, p->start, \
622 p->end - p->start); \
624 goto tlv_put_failure; \
626 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
628 ret = tlv_put_uuid(sctx, attrtype, uuid); \
630 goto tlv_put_failure; \
632 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
634 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
636 goto tlv_put_failure; \
639 static int send_header(struct send_ctx *sctx)
641 struct btrfs_stream_header hdr;
643 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
644 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
646 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
651 * For each command/item we want to send to userspace, we call this function.
653 static int begin_cmd(struct send_ctx *sctx, int cmd)
655 struct btrfs_cmd_header *hdr;
657 if (WARN_ON(!sctx->send_buf))
660 BUG_ON(sctx->send_size);
662 sctx->send_size += sizeof(*hdr);
663 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
664 hdr->cmd = cpu_to_le16(cmd);
669 static int send_cmd(struct send_ctx *sctx)
672 struct btrfs_cmd_header *hdr;
675 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
676 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
679 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
680 hdr->crc = cpu_to_le32(crc);
682 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
685 sctx->total_send_size += sctx->send_size;
686 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
693 * Sends a move instruction to user space
695 static int send_rename(struct send_ctx *sctx,
696 struct fs_path *from, struct fs_path *to)
700 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
702 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
706 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
707 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
709 ret = send_cmd(sctx);
717 * Sends a link instruction to user space
719 static int send_link(struct send_ctx *sctx,
720 struct fs_path *path, struct fs_path *lnk)
724 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
726 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
730 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
731 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
733 ret = send_cmd(sctx);
741 * Sends an unlink instruction to user space
743 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
747 verbose_printk("btrfs: send_unlink %s\n", path->start);
749 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
753 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
755 ret = send_cmd(sctx);
763 * Sends a rmdir instruction to user space
765 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
769 verbose_printk("btrfs: send_rmdir %s\n", path->start);
771 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
775 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
777 ret = send_cmd(sctx);
785 * Helper function to retrieve some fields from an inode item.
787 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
788 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
792 struct btrfs_inode_item *ii;
793 struct btrfs_key key;
796 key.type = BTRFS_INODE_ITEM_KEY;
798 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
805 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
806 struct btrfs_inode_item);
808 *size = btrfs_inode_size(path->nodes[0], ii);
810 *gen = btrfs_inode_generation(path->nodes[0], ii);
812 *mode = btrfs_inode_mode(path->nodes[0], ii);
814 *uid = btrfs_inode_uid(path->nodes[0], ii);
816 *gid = btrfs_inode_gid(path->nodes[0], ii);
818 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
823 static int get_inode_info(struct btrfs_root *root,
824 u64 ino, u64 *size, u64 *gen,
825 u64 *mode, u64 *uid, u64 *gid,
828 struct btrfs_path *path;
831 path = alloc_path_for_send();
834 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
836 btrfs_free_path(path);
840 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
845 * Helper function to iterate the entries in ONE btrfs_inode_ref or
846 * btrfs_inode_extref.
847 * The iterate callback may return a non zero value to stop iteration. This can
848 * be a negative value for error codes or 1 to simply stop it.
850 * path must point to the INODE_REF or INODE_EXTREF when called.
852 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
853 struct btrfs_key *found_key, int resolve,
854 iterate_inode_ref_t iterate, void *ctx)
856 struct extent_buffer *eb = path->nodes[0];
857 struct btrfs_item *item;
858 struct btrfs_inode_ref *iref;
859 struct btrfs_inode_extref *extref;
860 struct btrfs_path *tmp_path;
864 int slot = path->slots[0];
871 unsigned long name_off;
872 unsigned long elem_size;
875 p = fs_path_alloc_reversed();
879 tmp_path = alloc_path_for_send();
886 if (found_key->type == BTRFS_INODE_REF_KEY) {
887 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
888 struct btrfs_inode_ref);
889 item = btrfs_item_nr(slot);
890 total = btrfs_item_size(eb, item);
891 elem_size = sizeof(*iref);
893 ptr = btrfs_item_ptr_offset(eb, slot);
894 total = btrfs_item_size_nr(eb, slot);
895 elem_size = sizeof(*extref);
898 while (cur < total) {
901 if (found_key->type == BTRFS_INODE_REF_KEY) {
902 iref = (struct btrfs_inode_ref *)(ptr + cur);
903 name_len = btrfs_inode_ref_name_len(eb, iref);
904 name_off = (unsigned long)(iref + 1);
905 index = btrfs_inode_ref_index(eb, iref);
906 dir = found_key->offset;
908 extref = (struct btrfs_inode_extref *)(ptr + cur);
909 name_len = btrfs_inode_extref_name_len(eb, extref);
910 name_off = (unsigned long)&extref->name;
911 index = btrfs_inode_extref_index(eb, extref);
912 dir = btrfs_inode_extref_parent(eb, extref);
916 start = btrfs_ref_to_path(root, tmp_path, name_len,
920 ret = PTR_ERR(start);
923 if (start < p->buf) {
924 /* overflow , try again with larger buffer */
925 ret = fs_path_ensure_buf(p,
926 p->buf_len + p->buf - start);
929 start = btrfs_ref_to_path(root, tmp_path,
934 ret = PTR_ERR(start);
937 BUG_ON(start < p->buf);
941 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
947 cur += elem_size + name_len;
948 ret = iterate(num, dir, index, p, ctx);
955 btrfs_free_path(tmp_path);
960 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
961 const char *name, int name_len,
962 const char *data, int data_len,
966 * Helper function to iterate the entries in ONE btrfs_dir_item.
967 * The iterate callback may return a non zero value to stop iteration. This can
968 * be a negative value for error codes or 1 to simply stop it.
970 * path must point to the dir item when called.
972 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
973 struct btrfs_key *found_key,
974 iterate_dir_item_t iterate, void *ctx)
977 struct extent_buffer *eb;
978 struct btrfs_item *item;
979 struct btrfs_dir_item *di;
980 struct btrfs_key di_key;
993 * Start with a small buffer (1 page). If later we end up needing more
994 * space, which can happen for xattrs on a fs with a leaf size greater
995 * then the page size, attempt to increase the buffer. Typically xattr
999 buf = kmalloc(buf_len, GFP_NOFS);
1005 eb = path->nodes[0];
1006 slot = path->slots[0];
1007 item = btrfs_item_nr(slot);
1008 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1011 total = btrfs_item_size(eb, item);
1014 while (cur < total) {
1015 name_len = btrfs_dir_name_len(eb, di);
1016 data_len = btrfs_dir_data_len(eb, di);
1017 type = btrfs_dir_type(eb, di);
1018 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1020 if (type == BTRFS_FT_XATTR) {
1021 if (name_len > XATTR_NAME_MAX) {
1022 ret = -ENAMETOOLONG;
1025 if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(root)) {
1033 if (name_len + data_len > PATH_MAX) {
1034 ret = -ENAMETOOLONG;
1039 if (name_len + data_len > buf_len) {
1040 buf_len = name_len + data_len;
1041 if (is_vmalloc_addr(buf)) {
1045 char *tmp = krealloc(buf, buf_len,
1046 GFP_NOFS | __GFP_NOWARN);
1053 buf = vmalloc(buf_len);
1061 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1062 name_len + data_len);
1064 len = sizeof(*di) + name_len + data_len;
1065 di = (struct btrfs_dir_item *)((char *)di + len);
1068 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1069 data_len, type, ctx);
1085 static int __copy_first_ref(int num, u64 dir, int index,
1086 struct fs_path *p, void *ctx)
1089 struct fs_path *pt = ctx;
1091 ret = fs_path_copy(pt, p);
1095 /* we want the first only */
1100 * Retrieve the first path of an inode. If an inode has more then one
1101 * ref/hardlink, this is ignored.
1103 static int get_inode_path(struct btrfs_root *root,
1104 u64 ino, struct fs_path *path)
1107 struct btrfs_key key, found_key;
1108 struct btrfs_path *p;
1110 p = alloc_path_for_send();
1114 fs_path_reset(path);
1117 key.type = BTRFS_INODE_REF_KEY;
1120 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1127 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1128 if (found_key.objectid != ino ||
1129 (found_key.type != BTRFS_INODE_REF_KEY &&
1130 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1135 ret = iterate_inode_ref(root, p, &found_key, 1,
1136 __copy_first_ref, path);
1146 struct backref_ctx {
1147 struct send_ctx *sctx;
1149 struct btrfs_path *path;
1150 /* number of total found references */
1154 * used for clones found in send_root. clones found behind cur_objectid
1155 * and cur_offset are not considered as allowed clones.
1160 /* may be truncated in case it's the last extent in a file */
1163 /* data offset in the file extent item */
1166 /* Just to check for bugs in backref resolving */
1170 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1172 u64 root = (u64)(uintptr_t)key;
1173 struct clone_root *cr = (struct clone_root *)elt;
1175 if (root < cr->root->objectid)
1177 if (root > cr->root->objectid)
1182 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1184 struct clone_root *cr1 = (struct clone_root *)e1;
1185 struct clone_root *cr2 = (struct clone_root *)e2;
1187 if (cr1->root->objectid < cr2->root->objectid)
1189 if (cr1->root->objectid > cr2->root->objectid)
1195 * Called for every backref that is found for the current extent.
1196 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1198 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1200 struct backref_ctx *bctx = ctx_;
1201 struct clone_root *found;
1205 /* First check if the root is in the list of accepted clone sources */
1206 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1207 bctx->sctx->clone_roots_cnt,
1208 sizeof(struct clone_root),
1209 __clone_root_cmp_bsearch);
1213 if (found->root == bctx->sctx->send_root &&
1214 ino == bctx->cur_objectid &&
1215 offset == bctx->cur_offset) {
1216 bctx->found_itself = 1;
1220 * There are inodes that have extents that lie behind its i_size. Don't
1221 * accept clones from these extents.
1223 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1225 btrfs_release_path(bctx->path);
1229 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1233 * Make sure we don't consider clones from send_root that are
1234 * behind the current inode/offset.
1236 if (found->root == bctx->sctx->send_root) {
1238 * TODO for the moment we don't accept clones from the inode
1239 * that is currently send. We may change this when
1240 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1243 if (ino >= bctx->cur_objectid)
1246 if (ino > bctx->cur_objectid)
1248 if (offset + bctx->extent_len > bctx->cur_offset)
1254 found->found_refs++;
1255 if (ino < found->ino) {
1257 found->offset = offset;
1258 } else if (found->ino == ino) {
1260 * same extent found more then once in the same file.
1262 if (found->offset > offset + bctx->extent_len)
1263 found->offset = offset;
1270 * Given an inode, offset and extent item, it finds a good clone for a clone
1271 * instruction. Returns -ENOENT when none could be found. The function makes
1272 * sure that the returned clone is usable at the point where sending is at the
1273 * moment. This means, that no clones are accepted which lie behind the current
1276 * path must point to the extent item when called.
1278 static int find_extent_clone(struct send_ctx *sctx,
1279 struct btrfs_path *path,
1280 u64 ino, u64 data_offset,
1282 struct clone_root **found)
1289 u64 extent_item_pos;
1291 struct btrfs_file_extent_item *fi;
1292 struct extent_buffer *eb = path->nodes[0];
1293 struct backref_ctx *backref_ctx = NULL;
1294 struct clone_root *cur_clone_root;
1295 struct btrfs_key found_key;
1296 struct btrfs_path *tmp_path;
1300 tmp_path = alloc_path_for_send();
1304 /* We only use this path under the commit sem */
1305 tmp_path->need_commit_sem = 0;
1307 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1313 backref_ctx->path = tmp_path;
1315 if (data_offset >= ino_size) {
1317 * There may be extents that lie behind the file's size.
1318 * I at least had this in combination with snapshotting while
1319 * writing large files.
1325 fi = btrfs_item_ptr(eb, path->slots[0],
1326 struct btrfs_file_extent_item);
1327 extent_type = btrfs_file_extent_type(eb, fi);
1328 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1332 compressed = btrfs_file_extent_compression(eb, fi);
1334 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1335 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1336 if (disk_byte == 0) {
1340 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1342 down_read(&sctx->send_root->fs_info->commit_root_sem);
1343 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1344 &found_key, &flags);
1345 up_read(&sctx->send_root->fs_info->commit_root_sem);
1346 btrfs_release_path(tmp_path);
1350 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1356 * Setup the clone roots.
1358 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1359 cur_clone_root = sctx->clone_roots + i;
1360 cur_clone_root->ino = (u64)-1;
1361 cur_clone_root->offset = 0;
1362 cur_clone_root->found_refs = 0;
1365 backref_ctx->sctx = sctx;
1366 backref_ctx->found = 0;
1367 backref_ctx->cur_objectid = ino;
1368 backref_ctx->cur_offset = data_offset;
1369 backref_ctx->found_itself = 0;
1370 backref_ctx->extent_len = num_bytes;
1372 * For non-compressed extents iterate_extent_inodes() gives us extent
1373 * offsets that already take into account the data offset, but not for
1374 * compressed extents, since the offset is logical and not relative to
1375 * the physical extent locations. We must take this into account to
1376 * avoid sending clone offsets that go beyond the source file's size,
1377 * which would result in the clone ioctl failing with -EINVAL on the
1380 if (compressed == BTRFS_COMPRESS_NONE)
1381 backref_ctx->data_offset = 0;
1383 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1386 * The last extent of a file may be too large due to page alignment.
1387 * We need to adjust extent_len in this case so that the checks in
1388 * __iterate_backrefs work.
1390 if (data_offset + num_bytes >= ino_size)
1391 backref_ctx->extent_len = ino_size - data_offset;
1394 * Now collect all backrefs.
1396 if (compressed == BTRFS_COMPRESS_NONE)
1397 extent_item_pos = logical - found_key.objectid;
1399 extent_item_pos = 0;
1400 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1401 found_key.objectid, extent_item_pos, 1,
1402 __iterate_backrefs, backref_ctx);
1407 if (!backref_ctx->found_itself) {
1408 /* found a bug in backref code? */
1410 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1411 "send_root. inode=%llu, offset=%llu, "
1412 "disk_byte=%llu found extent=%llu",
1413 ino, data_offset, disk_byte, found_key.objectid);
1417 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1419 "num_bytes=%llu, logical=%llu\n",
1420 data_offset, ino, num_bytes, logical);
1422 if (!backref_ctx->found)
1423 verbose_printk("btrfs: no clones found\n");
1425 cur_clone_root = NULL;
1426 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1427 if (sctx->clone_roots[i].found_refs) {
1428 if (!cur_clone_root)
1429 cur_clone_root = sctx->clone_roots + i;
1430 else if (sctx->clone_roots[i].root == sctx->send_root)
1431 /* prefer clones from send_root over others */
1432 cur_clone_root = sctx->clone_roots + i;
1437 if (cur_clone_root) {
1438 *found = cur_clone_root;
1445 btrfs_free_path(tmp_path);
1450 static int read_symlink(struct btrfs_root *root,
1452 struct fs_path *dest)
1455 struct btrfs_path *path;
1456 struct btrfs_key key;
1457 struct btrfs_file_extent_item *ei;
1463 path = alloc_path_for_send();
1468 key.type = BTRFS_EXTENT_DATA_KEY;
1470 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1475 * An empty symlink inode. Can happen in rare error paths when
1476 * creating a symlink (transaction committed before the inode
1477 * eviction handler removed the symlink inode items and a crash
1478 * happened in between or the subvol was snapshoted in between).
1479 * Print an informative message to dmesg/syslog so that the user
1480 * can delete the symlink.
1482 btrfs_err(root->fs_info,
1483 "Found empty symlink inode %llu at root %llu",
1484 ino, root->root_key.objectid);
1489 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1490 struct btrfs_file_extent_item);
1491 type = btrfs_file_extent_type(path->nodes[0], ei);
1492 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1493 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1494 BUG_ON(compression);
1496 off = btrfs_file_extent_inline_start(ei);
1497 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1499 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1502 btrfs_free_path(path);
1507 * Helper function to generate a file name that is unique in the root of
1508 * send_root and parent_root. This is used to generate names for orphan inodes.
1510 static int gen_unique_name(struct send_ctx *sctx,
1512 struct fs_path *dest)
1515 struct btrfs_path *path;
1516 struct btrfs_dir_item *di;
1521 path = alloc_path_for_send();
1526 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1528 ASSERT(len < sizeof(tmp));
1530 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1531 path, BTRFS_FIRST_FREE_OBJECTID,
1532 tmp, strlen(tmp), 0);
1533 btrfs_release_path(path);
1539 /* not unique, try again */
1544 if (!sctx->parent_root) {
1550 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1551 path, BTRFS_FIRST_FREE_OBJECTID,
1552 tmp, strlen(tmp), 0);
1553 btrfs_release_path(path);
1559 /* not unique, try again */
1567 ret = fs_path_add(dest, tmp, strlen(tmp));
1570 btrfs_free_path(path);
1575 inode_state_no_change,
1576 inode_state_will_create,
1577 inode_state_did_create,
1578 inode_state_will_delete,
1579 inode_state_did_delete,
1582 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1590 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1592 if (ret < 0 && ret != -ENOENT)
1596 if (!sctx->parent_root) {
1597 right_ret = -ENOENT;
1599 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1600 NULL, NULL, NULL, NULL);
1601 if (ret < 0 && ret != -ENOENT)
1606 if (!left_ret && !right_ret) {
1607 if (left_gen == gen && right_gen == gen) {
1608 ret = inode_state_no_change;
1609 } else if (left_gen == gen) {
1610 if (ino < sctx->send_progress)
1611 ret = inode_state_did_create;
1613 ret = inode_state_will_create;
1614 } else if (right_gen == gen) {
1615 if (ino < sctx->send_progress)
1616 ret = inode_state_did_delete;
1618 ret = inode_state_will_delete;
1622 } else if (!left_ret) {
1623 if (left_gen == gen) {
1624 if (ino < sctx->send_progress)
1625 ret = inode_state_did_create;
1627 ret = inode_state_will_create;
1631 } else if (!right_ret) {
1632 if (right_gen == gen) {
1633 if (ino < sctx->send_progress)
1634 ret = inode_state_did_delete;
1636 ret = inode_state_will_delete;
1648 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1652 if (ino == BTRFS_FIRST_FREE_OBJECTID)
1655 ret = get_cur_inode_state(sctx, ino, gen);
1659 if (ret == inode_state_no_change ||
1660 ret == inode_state_did_create ||
1661 ret == inode_state_will_delete)
1671 * Helper function to lookup a dir item in a dir.
1673 static int lookup_dir_item_inode(struct btrfs_root *root,
1674 u64 dir, const char *name, int name_len,
1679 struct btrfs_dir_item *di;
1680 struct btrfs_key key;
1681 struct btrfs_path *path;
1683 path = alloc_path_for_send();
1687 di = btrfs_lookup_dir_item(NULL, root, path,
1688 dir, name, name_len, 0);
1697 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1698 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1702 *found_inode = key.objectid;
1703 *found_type = btrfs_dir_type(path->nodes[0], di);
1706 btrfs_free_path(path);
1711 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1712 * generation of the parent dir and the name of the dir entry.
1714 static int get_first_ref(struct btrfs_root *root, u64 ino,
1715 u64 *dir, u64 *dir_gen, struct fs_path *name)
1718 struct btrfs_key key;
1719 struct btrfs_key found_key;
1720 struct btrfs_path *path;
1724 path = alloc_path_for_send();
1729 key.type = BTRFS_INODE_REF_KEY;
1732 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1736 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1738 if (ret || found_key.objectid != ino ||
1739 (found_key.type != BTRFS_INODE_REF_KEY &&
1740 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1745 if (found_key.type == BTRFS_INODE_REF_KEY) {
1746 struct btrfs_inode_ref *iref;
1747 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1748 struct btrfs_inode_ref);
1749 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1750 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1751 (unsigned long)(iref + 1),
1753 parent_dir = found_key.offset;
1755 struct btrfs_inode_extref *extref;
1756 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1757 struct btrfs_inode_extref);
1758 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1759 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1760 (unsigned long)&extref->name, len);
1761 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1765 btrfs_release_path(path);
1768 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1777 btrfs_free_path(path);
1781 static int is_first_ref(struct btrfs_root *root,
1783 const char *name, int name_len)
1786 struct fs_path *tmp_name;
1789 tmp_name = fs_path_alloc();
1793 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1797 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1802 ret = !memcmp(tmp_name->start, name, name_len);
1805 fs_path_free(tmp_name);
1810 * Used by process_recorded_refs to determine if a new ref would overwrite an
1811 * already existing ref. In case it detects an overwrite, it returns the
1812 * inode/gen in who_ino/who_gen.
1813 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1814 * to make sure later references to the overwritten inode are possible.
1815 * Orphanizing is however only required for the first ref of an inode.
1816 * process_recorded_refs does an additional is_first_ref check to see if
1817 * orphanizing is really required.
1819 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1820 const char *name, int name_len,
1821 u64 *who_ino, u64 *who_gen)
1825 u64 other_inode = 0;
1828 if (!sctx->parent_root)
1831 ret = is_inode_existent(sctx, dir, dir_gen);
1836 * If we have a parent root we need to verify that the parent dir was
1837 * not delted and then re-created, if it was then we have no overwrite
1838 * and we can just unlink this entry.
1840 if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) {
1841 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1843 if (ret < 0 && ret != -ENOENT)
1853 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1854 &other_inode, &other_type);
1855 if (ret < 0 && ret != -ENOENT)
1863 * Check if the overwritten ref was already processed. If yes, the ref
1864 * was already unlinked/moved, so we can safely assume that we will not
1865 * overwrite anything at this point in time.
1867 if (other_inode > sctx->send_progress) {
1868 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1869 who_gen, NULL, NULL, NULL, NULL);
1874 *who_ino = other_inode;
1884 * Checks if the ref was overwritten by an already processed inode. This is
1885 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1886 * thus the orphan name needs be used.
1887 * process_recorded_refs also uses it to avoid unlinking of refs that were
1890 static int did_overwrite_ref(struct send_ctx *sctx,
1891 u64 dir, u64 dir_gen,
1892 u64 ino, u64 ino_gen,
1893 const char *name, int name_len)
1900 if (!sctx->parent_root)
1903 ret = is_inode_existent(sctx, dir, dir_gen);
1907 /* check if the ref was overwritten by another ref */
1908 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1909 &ow_inode, &other_type);
1910 if (ret < 0 && ret != -ENOENT)
1913 /* was never and will never be overwritten */
1918 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1923 if (ow_inode == ino && gen == ino_gen) {
1929 * We know that it is or will be overwritten. Check this now.
1930 * The current inode being processed might have been the one that caused
1931 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1932 * the current inode being processed.
1934 if ((ow_inode < sctx->send_progress) ||
1935 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1936 gen == sctx->cur_inode_gen))
1946 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1947 * that got overwritten. This is used by process_recorded_refs to determine
1948 * if it has to use the path as returned by get_cur_path or the orphan name.
1950 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1953 struct fs_path *name = NULL;
1957 if (!sctx->parent_root)
1960 name = fs_path_alloc();
1964 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1968 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1969 name->start, fs_path_len(name));
1977 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1978 * so we need to do some special handling in case we have clashes. This function
1979 * takes care of this with the help of name_cache_entry::radix_list.
1980 * In case of error, nce is kfreed.
1982 static int name_cache_insert(struct send_ctx *sctx,
1983 struct name_cache_entry *nce)
1986 struct list_head *nce_head;
1988 nce_head = radix_tree_lookup(&sctx->name_cache,
1989 (unsigned long)nce->ino);
1991 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1996 INIT_LIST_HEAD(nce_head);
1998 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
2005 list_add_tail(&nce->radix_list, nce_head);
2006 list_add_tail(&nce->list, &sctx->name_cache_list);
2007 sctx->name_cache_size++;
2012 static void name_cache_delete(struct send_ctx *sctx,
2013 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 btrfs_err(sctx->send_root->fs_info,
2021 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2022 nce->ino, sctx->name_cache_size);
2025 list_del(&nce->radix_list);
2026 list_del(&nce->list);
2027 sctx->name_cache_size--;
2030 * We may not get to the final release of nce_head if the lookup fails
2032 if (nce_head && list_empty(nce_head)) {
2033 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2038 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2041 struct list_head *nce_head;
2042 struct name_cache_entry *cur;
2044 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2048 list_for_each_entry(cur, nce_head, radix_list) {
2049 if (cur->ino == ino && cur->gen == gen)
2056 * Removes the entry from the list and adds it back to the end. This marks the
2057 * entry as recently used so that name_cache_clean_unused does not remove it.
2059 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2061 list_del(&nce->list);
2062 list_add_tail(&nce->list, &sctx->name_cache_list);
2066 * Remove some entries from the beginning of name_cache_list.
2068 static void name_cache_clean_unused(struct send_ctx *sctx)
2070 struct name_cache_entry *nce;
2072 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2075 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2076 nce = list_entry(sctx->name_cache_list.next,
2077 struct name_cache_entry, list);
2078 name_cache_delete(sctx, nce);
2083 static void name_cache_free(struct send_ctx *sctx)
2085 struct name_cache_entry *nce;
2087 while (!list_empty(&sctx->name_cache_list)) {
2088 nce = list_entry(sctx->name_cache_list.next,
2089 struct name_cache_entry, list);
2090 name_cache_delete(sctx, nce);
2096 * Used by get_cur_path for each ref up to the root.
2097 * Returns 0 if it succeeded.
2098 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2099 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2100 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2101 * Returns <0 in case of error.
2103 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2107 struct fs_path *dest)
2111 struct name_cache_entry *nce = NULL;
2114 * First check if we already did a call to this function with the same
2115 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2116 * return the cached result.
2118 nce = name_cache_search(sctx, ino, gen);
2120 if (ino < sctx->send_progress && nce->need_later_update) {
2121 name_cache_delete(sctx, nce);
2125 name_cache_used(sctx, nce);
2126 *parent_ino = nce->parent_ino;
2127 *parent_gen = nce->parent_gen;
2128 ret = fs_path_add(dest, nce->name, nce->name_len);
2137 * If the inode is not existent yet, add the orphan name and return 1.
2138 * This should only happen for the parent dir that we determine in
2141 ret = is_inode_existent(sctx, ino, gen);
2146 ret = gen_unique_name(sctx, ino, gen, dest);
2154 * Depending on whether the inode was already processed or not, use
2155 * send_root or parent_root for ref lookup.
2157 if (ino < sctx->send_progress)
2158 ret = get_first_ref(sctx->send_root, ino,
2159 parent_ino, parent_gen, dest);
2161 ret = get_first_ref(sctx->parent_root, ino,
2162 parent_ino, parent_gen, dest);
2167 * Check if the ref was overwritten by an inode's ref that was processed
2168 * earlier. If yes, treat as orphan and return 1.
2170 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2171 dest->start, dest->end - dest->start);
2175 fs_path_reset(dest);
2176 ret = gen_unique_name(sctx, ino, gen, dest);
2184 * Store the result of the lookup in the name cache.
2186 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2194 nce->parent_ino = *parent_ino;
2195 nce->parent_gen = *parent_gen;
2196 nce->name_len = fs_path_len(dest);
2198 strcpy(nce->name, dest->start);
2200 if (ino < sctx->send_progress)
2201 nce->need_later_update = 0;
2203 nce->need_later_update = 1;
2205 nce_ret = name_cache_insert(sctx, nce);
2208 name_cache_clean_unused(sctx);
2215 * Magic happens here. This function returns the first ref to an inode as it
2216 * would look like while receiving the stream at this point in time.
2217 * We walk the path up to the root. For every inode in between, we check if it
2218 * was already processed/sent. If yes, we continue with the parent as found
2219 * in send_root. If not, we continue with the parent as found in parent_root.
2220 * If we encounter an inode that was deleted at this point in time, we use the
2221 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2222 * that were not created yet and overwritten inodes/refs.
2224 * When do we have have orphan inodes:
2225 * 1. When an inode is freshly created and thus no valid refs are available yet
2226 * 2. When a directory lost all it's refs (deleted) but still has dir items
2227 * inside which were not processed yet (pending for move/delete). If anyone
2228 * tried to get the path to the dir items, it would get a path inside that
2230 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2231 * of an unprocessed inode. If in that case the first ref would be
2232 * overwritten, the overwritten inode gets "orphanized". Later when we
2233 * process this overwritten inode, it is restored at a new place by moving
2236 * sctx->send_progress tells this function at which point in time receiving
2239 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2240 struct fs_path *dest)
2243 struct fs_path *name = NULL;
2244 u64 parent_inode = 0;
2248 name = fs_path_alloc();
2255 fs_path_reset(dest);
2257 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2258 struct waiting_dir_move *wdm;
2260 fs_path_reset(name);
2262 if (is_waiting_for_rm(sctx, ino)) {
2263 ret = gen_unique_name(sctx, ino, gen, name);
2266 ret = fs_path_add_path(dest, name);
2270 wdm = get_waiting_dir_move(sctx, ino);
2271 if (wdm && wdm->orphanized) {
2272 ret = gen_unique_name(sctx, ino, gen, name);
2275 ret = get_first_ref(sctx->parent_root, ino,
2276 &parent_inode, &parent_gen, name);
2278 ret = __get_cur_name_and_parent(sctx, ino, gen,
2288 ret = fs_path_add_path(dest, name);
2299 fs_path_unreverse(dest);
2304 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2306 static int send_subvol_begin(struct send_ctx *sctx)
2309 struct btrfs_root *send_root = sctx->send_root;
2310 struct btrfs_root *parent_root = sctx->parent_root;
2311 struct btrfs_path *path;
2312 struct btrfs_key key;
2313 struct btrfs_root_ref *ref;
2314 struct extent_buffer *leaf;
2318 path = btrfs_alloc_path();
2322 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2324 btrfs_free_path(path);
2328 key.objectid = send_root->objectid;
2329 key.type = BTRFS_ROOT_BACKREF_KEY;
2332 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2341 leaf = path->nodes[0];
2342 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2343 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2344 key.objectid != send_root->objectid) {
2348 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2349 namelen = btrfs_root_ref_name_len(leaf, ref);
2350 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2351 btrfs_release_path(path);
2354 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2358 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2363 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2365 if (!btrfs_is_empty_uuid(sctx->send_root->root_item.received_uuid))
2366 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2367 sctx->send_root->root_item.received_uuid);
2369 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2370 sctx->send_root->root_item.uuid);
2372 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2373 le64_to_cpu(sctx->send_root->root_item.ctransid));
2375 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2376 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2377 parent_root->root_item.received_uuid);
2379 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2380 parent_root->root_item.uuid);
2381 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2382 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2385 ret = send_cmd(sctx);
2389 btrfs_free_path(path);
2394 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2399 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2401 p = fs_path_alloc();
2405 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2409 ret = get_cur_path(sctx, ino, gen, p);
2412 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2413 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2415 ret = send_cmd(sctx);
2423 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2428 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2430 p = fs_path_alloc();
2434 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2438 ret = get_cur_path(sctx, ino, gen, p);
2441 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2442 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2444 ret = send_cmd(sctx);
2452 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2457 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2459 p = fs_path_alloc();
2463 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2467 ret = get_cur_path(sctx, ino, gen, p);
2470 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2471 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2472 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2474 ret = send_cmd(sctx);
2482 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2485 struct fs_path *p = NULL;
2486 struct btrfs_inode_item *ii;
2487 struct btrfs_path *path = NULL;
2488 struct extent_buffer *eb;
2489 struct btrfs_key key;
2492 verbose_printk("btrfs: send_utimes %llu\n", ino);
2494 p = fs_path_alloc();
2498 path = alloc_path_for_send();
2505 key.type = BTRFS_INODE_ITEM_KEY;
2507 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2511 eb = path->nodes[0];
2512 slot = path->slots[0];
2513 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2515 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2519 ret = get_cur_path(sctx, ino, gen, p);
2522 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2523 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2524 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2525 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2526 /* TODO Add otime support when the otime patches get into upstream */
2528 ret = send_cmd(sctx);
2533 btrfs_free_path(path);
2538 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2539 * a valid path yet because we did not process the refs yet. So, the inode
2540 * is created as orphan.
2542 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2551 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2553 p = fs_path_alloc();
2557 if (ino != sctx->cur_ino) {
2558 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2563 gen = sctx->cur_inode_gen;
2564 mode = sctx->cur_inode_mode;
2565 rdev = sctx->cur_inode_rdev;
2568 if (S_ISREG(mode)) {
2569 cmd = BTRFS_SEND_C_MKFILE;
2570 } else if (S_ISDIR(mode)) {
2571 cmd = BTRFS_SEND_C_MKDIR;
2572 } else if (S_ISLNK(mode)) {
2573 cmd = BTRFS_SEND_C_SYMLINK;
2574 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2575 cmd = BTRFS_SEND_C_MKNOD;
2576 } else if (S_ISFIFO(mode)) {
2577 cmd = BTRFS_SEND_C_MKFIFO;
2578 } else if (S_ISSOCK(mode)) {
2579 cmd = BTRFS_SEND_C_MKSOCK;
2581 btrfs_warn(sctx->send_root->fs_info, "unexpected inode type %o",
2582 (int)(mode & S_IFMT));
2587 ret = begin_cmd(sctx, cmd);
2591 ret = gen_unique_name(sctx, ino, gen, p);
2595 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2596 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2598 if (S_ISLNK(mode)) {
2600 ret = read_symlink(sctx->send_root, ino, p);
2603 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2604 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2605 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2606 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2607 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2610 ret = send_cmd(sctx);
2622 * We need some special handling for inodes that get processed before the parent
2623 * directory got created. See process_recorded_refs for details.
2624 * This function does the check if we already created the dir out of order.
2626 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2629 struct btrfs_path *path = NULL;
2630 struct btrfs_key key;
2631 struct btrfs_key found_key;
2632 struct btrfs_key di_key;
2633 struct extent_buffer *eb;
2634 struct btrfs_dir_item *di;
2637 path = alloc_path_for_send();
2644 key.type = BTRFS_DIR_INDEX_KEY;
2646 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2651 eb = path->nodes[0];
2652 slot = path->slots[0];
2653 if (slot >= btrfs_header_nritems(eb)) {
2654 ret = btrfs_next_leaf(sctx->send_root, path);
2657 } else if (ret > 0) {
2664 btrfs_item_key_to_cpu(eb, &found_key, slot);
2665 if (found_key.objectid != key.objectid ||
2666 found_key.type != key.type) {
2671 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2672 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2674 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2675 di_key.objectid < sctx->send_progress) {
2684 btrfs_free_path(path);
2689 * Only creates the inode if it is:
2690 * 1. Not a directory
2691 * 2. Or a directory which was not created already due to out of order
2692 * directories. See did_create_dir and process_recorded_refs for details.
2694 static int send_create_inode_if_needed(struct send_ctx *sctx)
2698 if (S_ISDIR(sctx->cur_inode_mode)) {
2699 ret = did_create_dir(sctx, sctx->cur_ino);
2708 ret = send_create_inode(sctx, sctx->cur_ino);
2716 struct recorded_ref {
2717 struct list_head list;
2720 struct fs_path *full_path;
2728 * We need to process new refs before deleted refs, but compare_tree gives us
2729 * everything mixed. So we first record all refs and later process them.
2730 * This function is a helper to record one ref.
2732 static int __record_ref(struct list_head *head, u64 dir,
2733 u64 dir_gen, struct fs_path *path)
2735 struct recorded_ref *ref;
2737 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2742 ref->dir_gen = dir_gen;
2743 ref->full_path = path;
2745 ref->name = (char *)kbasename(ref->full_path->start);
2746 ref->name_len = ref->full_path->end - ref->name;
2747 ref->dir_path = ref->full_path->start;
2748 if (ref->name == ref->full_path->start)
2749 ref->dir_path_len = 0;
2751 ref->dir_path_len = ref->full_path->end -
2752 ref->full_path->start - 1 - ref->name_len;
2754 list_add_tail(&ref->list, head);
2758 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2760 struct recorded_ref *new;
2762 new = kmalloc(sizeof(*ref), GFP_NOFS);
2766 new->dir = ref->dir;
2767 new->dir_gen = ref->dir_gen;
2768 new->full_path = NULL;
2769 INIT_LIST_HEAD(&new->list);
2770 list_add_tail(&new->list, list);
2774 static void __free_recorded_refs(struct list_head *head)
2776 struct recorded_ref *cur;
2778 while (!list_empty(head)) {
2779 cur = list_entry(head->next, struct recorded_ref, list);
2780 fs_path_free(cur->full_path);
2781 list_del(&cur->list);
2786 static void free_recorded_refs(struct send_ctx *sctx)
2788 __free_recorded_refs(&sctx->new_refs);
2789 __free_recorded_refs(&sctx->deleted_refs);
2793 * Renames/moves a file/dir to its orphan name. Used when the first
2794 * ref of an unprocessed inode gets overwritten and for all non empty
2797 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2798 struct fs_path *path)
2801 struct fs_path *orphan;
2803 orphan = fs_path_alloc();
2807 ret = gen_unique_name(sctx, ino, gen, orphan);
2811 ret = send_rename(sctx, path, orphan);
2814 fs_path_free(orphan);
2818 static struct orphan_dir_info *
2819 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2821 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2822 struct rb_node *parent = NULL;
2823 struct orphan_dir_info *entry, *odi;
2825 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2827 return ERR_PTR(-ENOMEM);
2833 entry = rb_entry(parent, struct orphan_dir_info, node);
2834 if (dir_ino < entry->ino) {
2836 } else if (dir_ino > entry->ino) {
2837 p = &(*p)->rb_right;
2844 rb_link_node(&odi->node, parent, p);
2845 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2849 static struct orphan_dir_info *
2850 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2852 struct rb_node *n = sctx->orphan_dirs.rb_node;
2853 struct orphan_dir_info *entry;
2856 entry = rb_entry(n, struct orphan_dir_info, node);
2857 if (dir_ino < entry->ino)
2859 else if (dir_ino > entry->ino)
2867 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2869 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2874 static void free_orphan_dir_info(struct send_ctx *sctx,
2875 struct orphan_dir_info *odi)
2879 rb_erase(&odi->node, &sctx->orphan_dirs);
2884 * Returns 1 if a directory can be removed at this point in time.
2885 * We check this by iterating all dir items and checking if the inode behind
2886 * the dir item was already processed.
2888 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2892 struct btrfs_root *root = sctx->parent_root;
2893 struct btrfs_path *path;
2894 struct btrfs_key key;
2895 struct btrfs_key found_key;
2896 struct btrfs_key loc;
2897 struct btrfs_dir_item *di;
2900 * Don't try to rmdir the top/root subvolume dir.
2902 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2905 path = alloc_path_for_send();
2910 key.type = BTRFS_DIR_INDEX_KEY;
2912 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2917 struct waiting_dir_move *dm;
2919 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2920 ret = btrfs_next_leaf(root, path);
2927 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2929 if (found_key.objectid != key.objectid ||
2930 found_key.type != key.type)
2933 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2934 struct btrfs_dir_item);
2935 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2937 dm = get_waiting_dir_move(sctx, loc.objectid);
2939 struct orphan_dir_info *odi;
2941 odi = add_orphan_dir_info(sctx, dir);
2947 dm->rmdir_ino = dir;
2952 if (loc.objectid > send_progress) {
2963 btrfs_free_path(path);
2967 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2969 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2971 return entry != NULL;
2974 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
2976 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2977 struct rb_node *parent = NULL;
2978 struct waiting_dir_move *entry, *dm;
2980 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2985 dm->orphanized = orphanized;
2989 entry = rb_entry(parent, struct waiting_dir_move, node);
2990 if (ino < entry->ino) {
2992 } else if (ino > entry->ino) {
2993 p = &(*p)->rb_right;
3000 rb_link_node(&dm->node, parent, p);
3001 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
3005 static struct waiting_dir_move *
3006 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
3008 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
3009 struct waiting_dir_move *entry;
3012 entry = rb_entry(n, struct waiting_dir_move, node);
3013 if (ino < entry->ino)
3015 else if (ino > entry->ino)
3023 static void free_waiting_dir_move(struct send_ctx *sctx,
3024 struct waiting_dir_move *dm)
3028 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3032 static int add_pending_dir_move(struct send_ctx *sctx,
3036 struct list_head *new_refs,
3037 struct list_head *deleted_refs,
3038 const bool is_orphan)
3040 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3041 struct rb_node *parent = NULL;
3042 struct pending_dir_move *entry = NULL, *pm;
3043 struct recorded_ref *cur;
3047 pm = kmalloc(sizeof(*pm), GFP_NOFS);
3050 pm->parent_ino = parent_ino;
3053 pm->is_orphan = is_orphan;
3054 INIT_LIST_HEAD(&pm->list);
3055 INIT_LIST_HEAD(&pm->update_refs);
3056 RB_CLEAR_NODE(&pm->node);
3060 entry = rb_entry(parent, struct pending_dir_move, node);
3061 if (parent_ino < entry->parent_ino) {
3063 } else if (parent_ino > entry->parent_ino) {
3064 p = &(*p)->rb_right;
3071 list_for_each_entry(cur, deleted_refs, list) {
3072 ret = dup_ref(cur, &pm->update_refs);
3076 list_for_each_entry(cur, new_refs, list) {
3077 ret = dup_ref(cur, &pm->update_refs);
3082 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3087 list_add_tail(&pm->list, &entry->list);
3089 rb_link_node(&pm->node, parent, p);
3090 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3095 __free_recorded_refs(&pm->update_refs);
3101 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3104 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3105 struct pending_dir_move *entry;
3108 entry = rb_entry(n, struct pending_dir_move, node);
3109 if (parent_ino < entry->parent_ino)
3111 else if (parent_ino > entry->parent_ino)
3119 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3121 struct fs_path *from_path = NULL;
3122 struct fs_path *to_path = NULL;
3123 struct fs_path *name = NULL;
3124 u64 orig_progress = sctx->send_progress;
3125 struct recorded_ref *cur;
3126 u64 parent_ino, parent_gen;
3127 struct waiting_dir_move *dm = NULL;
3131 name = fs_path_alloc();
3132 from_path = fs_path_alloc();
3133 if (!name || !from_path) {
3138 dm = get_waiting_dir_move(sctx, pm->ino);
3140 rmdir_ino = dm->rmdir_ino;
3141 free_waiting_dir_move(sctx, dm);
3143 if (pm->is_orphan) {
3144 ret = gen_unique_name(sctx, pm->ino,
3145 pm->gen, from_path);
3147 ret = get_first_ref(sctx->parent_root, pm->ino,
3148 &parent_ino, &parent_gen, name);
3151 ret = get_cur_path(sctx, parent_ino, parent_gen,
3155 ret = fs_path_add_path(from_path, name);
3160 sctx->send_progress = sctx->cur_ino + 1;
3161 fs_path_reset(name);
3164 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3168 ret = send_rename(sctx, from_path, to_path);
3173 struct orphan_dir_info *odi;
3175 odi = get_orphan_dir_info(sctx, rmdir_ino);
3177 /* already deleted */
3180 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3186 name = fs_path_alloc();
3191 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3194 ret = send_rmdir(sctx, name);
3197 free_orphan_dir_info(sctx, odi);
3201 ret = send_utimes(sctx, pm->ino, pm->gen);
3206 * After rename/move, need to update the utimes of both new parent(s)
3207 * and old parent(s).
3209 list_for_each_entry(cur, &pm->update_refs, list) {
3210 if (cur->dir == rmdir_ino)
3212 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3219 fs_path_free(from_path);
3220 fs_path_free(to_path);
3221 sctx->send_progress = orig_progress;
3226 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3228 if (!list_empty(&m->list))
3230 if (!RB_EMPTY_NODE(&m->node))
3231 rb_erase(&m->node, &sctx->pending_dir_moves);
3232 __free_recorded_refs(&m->update_refs);
3236 static void tail_append_pending_moves(struct send_ctx *sctx,
3237 struct pending_dir_move *moves,
3238 struct list_head *stack)
3240 if (list_empty(&moves->list)) {
3241 list_add_tail(&moves->list, stack);
3244 list_splice_init(&moves->list, &list);
3245 list_add_tail(&moves->list, stack);
3246 list_splice_tail(&list, stack);
3248 if (!RB_EMPTY_NODE(&moves->node)) {
3249 rb_erase(&moves->node, &sctx->pending_dir_moves);
3250 RB_CLEAR_NODE(&moves->node);
3254 static int apply_children_dir_moves(struct send_ctx *sctx)
3256 struct pending_dir_move *pm;
3257 struct list_head stack;
3258 u64 parent_ino = sctx->cur_ino;
3261 pm = get_pending_dir_moves(sctx, parent_ino);
3265 INIT_LIST_HEAD(&stack);
3266 tail_append_pending_moves(sctx, pm, &stack);
3268 while (!list_empty(&stack)) {
3269 pm = list_first_entry(&stack, struct pending_dir_move, list);
3270 parent_ino = pm->ino;
3271 ret = apply_dir_move(sctx, pm);
3272 free_pending_move(sctx, pm);
3275 pm = get_pending_dir_moves(sctx, parent_ino);
3277 tail_append_pending_moves(sctx, pm, &stack);
3282 while (!list_empty(&stack)) {
3283 pm = list_first_entry(&stack, struct pending_dir_move, list);
3284 free_pending_move(sctx, pm);
3290 * We might need to delay a directory rename even when no ancestor directory
3291 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3292 * renamed. This happens when we rename a directory to the old name (the name
3293 * in the parent root) of some other unrelated directory that got its rename
3294 * delayed due to some ancestor with higher number that got renamed.
3300 * |---- a/ (ino 257)
3301 * | |---- file (ino 260)
3303 * |---- b/ (ino 258)
3304 * |---- c/ (ino 259)
3308 * |---- a/ (ino 258)
3309 * |---- x/ (ino 259)
3310 * |---- y/ (ino 257)
3311 * |----- file (ino 260)
3313 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3314 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3315 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3318 * 1 - rename 259 from 'c' to 'x'
3319 * 2 - rename 257 from 'a' to 'x/y'
3320 * 3 - rename 258 from 'b' to 'a'
3322 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3323 * be done right away and < 0 on error.
3325 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3326 struct recorded_ref *parent_ref,
3327 const bool is_orphan)
3329 struct btrfs_path *path;
3330 struct btrfs_key key;
3331 struct btrfs_key di_key;
3332 struct btrfs_dir_item *di;
3337 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3340 path = alloc_path_for_send();
3344 key.objectid = parent_ref->dir;
3345 key.type = BTRFS_DIR_ITEM_KEY;
3346 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3348 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3351 } else if (ret > 0) {
3356 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3357 parent_ref->name, parent_ref->name_len);
3363 * di_key.objectid has the number of the inode that has a dentry in the
3364 * parent directory with the same name that sctx->cur_ino is being
3365 * renamed to. We need to check if that inode is in the send root as
3366 * well and if it is currently marked as an inode with a pending rename,
3367 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3368 * that it happens after that other inode is renamed.
3370 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3371 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3376 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3377 &left_gen, NULL, NULL, NULL, NULL);
3380 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3381 &right_gen, NULL, NULL, NULL, NULL);
3388 /* Different inode, no need to delay the rename of sctx->cur_ino */
3389 if (right_gen != left_gen) {
3394 if (is_waiting_for_move(sctx, di_key.objectid)) {
3395 ret = add_pending_dir_move(sctx,
3397 sctx->cur_inode_gen,
3400 &sctx->deleted_refs,
3406 btrfs_free_path(path);
3411 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3412 * Return 1 if true, 0 if false and < 0 on error.
3414 static int is_ancestor(struct btrfs_root *root,
3418 struct fs_path *fs_path)
3422 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3427 fs_path_reset(fs_path);
3428 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3430 if (ret == -ENOENT && ino == ino2)
3435 return parent_gen == ino1_gen ? 1 : 0;
3441 static int wait_for_parent_move(struct send_ctx *sctx,
3442 struct recorded_ref *parent_ref,
3443 const bool is_orphan)
3446 u64 ino = parent_ref->dir;
3447 u64 parent_ino_before, parent_ino_after;
3448 struct fs_path *path_before = NULL;
3449 struct fs_path *path_after = NULL;
3452 path_after = fs_path_alloc();
3453 path_before = fs_path_alloc();
3454 if (!path_after || !path_before) {
3460 * Our current directory inode may not yet be renamed/moved because some
3461 * ancestor (immediate or not) has to be renamed/moved first. So find if
3462 * such ancestor exists and make sure our own rename/move happens after
3463 * that ancestor is processed to avoid path build infinite loops (done
3464 * at get_cur_path()).
3466 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3467 if (is_waiting_for_move(sctx, ino)) {
3469 * If the current inode is an ancestor of ino in the
3470 * parent root, we need to delay the rename of the
3471 * current inode, otherwise don't delayed the rename
3472 * because we can end up with a circular dependency
3473 * of renames, resulting in some directories never
3474 * getting the respective rename operations issued in
3475 * the send stream or getting into infinite path build
3478 ret = is_ancestor(sctx->parent_root,
3479 sctx->cur_ino, sctx->cur_inode_gen,
3484 fs_path_reset(path_before);
3485 fs_path_reset(path_after);
3487 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3491 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3493 if (ret < 0 && ret != -ENOENT) {
3495 } else if (ret == -ENOENT) {
3500 len1 = fs_path_len(path_before);
3501 len2 = fs_path_len(path_after);
3502 if (ino > sctx->cur_ino &&
3503 (parent_ino_before != parent_ino_after || len1 != len2 ||
3504 memcmp(path_before->start, path_after->start, len1))) {
3508 ino = parent_ino_after;
3512 fs_path_free(path_before);
3513 fs_path_free(path_after);
3516 ret = add_pending_dir_move(sctx,
3518 sctx->cur_inode_gen,
3521 &sctx->deleted_refs,
3531 * This does all the move/link/unlink/rmdir magic.
3533 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3536 struct recorded_ref *cur;
3537 struct recorded_ref *cur2;
3538 struct list_head check_dirs;
3539 struct fs_path *valid_path = NULL;
3542 int did_overwrite = 0;
3544 u64 last_dir_ino_rm = 0;
3545 bool can_rename = true;
3547 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3550 * This should never happen as the root dir always has the same ref
3551 * which is always '..'
3553 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3554 INIT_LIST_HEAD(&check_dirs);
3556 valid_path = fs_path_alloc();
3563 * First, check if the first ref of the current inode was overwritten
3564 * before. If yes, we know that the current inode was already orphanized
3565 * and thus use the orphan name. If not, we can use get_cur_path to
3566 * get the path of the first ref as it would like while receiving at
3567 * this point in time.
3568 * New inodes are always orphan at the beginning, so force to use the
3569 * orphan name in this case.
3570 * The first ref is stored in valid_path and will be updated if it
3571 * gets moved around.
3573 if (!sctx->cur_inode_new) {
3574 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3575 sctx->cur_inode_gen);
3581 if (sctx->cur_inode_new || did_overwrite) {
3582 ret = gen_unique_name(sctx, sctx->cur_ino,
3583 sctx->cur_inode_gen, valid_path);
3588 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3594 list_for_each_entry(cur, &sctx->new_refs, list) {
3596 * We may have refs where the parent directory does not exist
3597 * yet. This happens if the parent directories inum is higher
3598 * the the current inum. To handle this case, we create the
3599 * parent directory out of order. But we need to check if this
3600 * did already happen before due to other refs in the same dir.
3602 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3605 if (ret == inode_state_will_create) {
3608 * First check if any of the current inodes refs did
3609 * already create the dir.
3611 list_for_each_entry(cur2, &sctx->new_refs, list) {
3614 if (cur2->dir == cur->dir) {
3621 * If that did not happen, check if a previous inode
3622 * did already create the dir.
3625 ret = did_create_dir(sctx, cur->dir);
3629 ret = send_create_inode(sctx, cur->dir);
3636 * Check if this new ref would overwrite the first ref of
3637 * another unprocessed inode. If yes, orphanize the
3638 * overwritten inode. If we find an overwritten ref that is
3639 * not the first ref, simply unlink it.
3641 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3642 cur->name, cur->name_len,
3643 &ow_inode, &ow_gen);
3647 ret = is_first_ref(sctx->parent_root,
3648 ow_inode, cur->dir, cur->name,
3653 struct name_cache_entry *nce;
3655 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3660 * Make sure we clear our orphanized inode's
3661 * name from the name cache. This is because the
3662 * inode ow_inode might be an ancestor of some
3663 * other inode that will be orphanized as well
3664 * later and has an inode number greater than
3665 * sctx->send_progress. We need to prevent
3666 * future name lookups from using the old name
3667 * and get instead the orphan name.
3669 nce = name_cache_search(sctx, ow_inode, ow_gen);
3671 name_cache_delete(sctx, nce);
3675 ret = send_unlink(sctx, cur->full_path);
3681 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3682 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3691 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3693 ret = wait_for_parent_move(sctx, cur, is_orphan);
3703 * link/move the ref to the new place. If we have an orphan
3704 * inode, move it and update valid_path. If not, link or move
3705 * it depending on the inode mode.
3707 if (is_orphan && can_rename) {
3708 ret = send_rename(sctx, valid_path, cur->full_path);
3712 ret = fs_path_copy(valid_path, cur->full_path);
3715 } else if (can_rename) {
3716 if (S_ISDIR(sctx->cur_inode_mode)) {
3718 * Dirs can't be linked, so move it. For moved
3719 * dirs, we always have one new and one deleted
3720 * ref. The deleted ref is ignored later.
3722 ret = send_rename(sctx, valid_path,
3725 ret = fs_path_copy(valid_path,
3730 ret = send_link(sctx, cur->full_path,
3736 ret = dup_ref(cur, &check_dirs);
3741 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3743 * Check if we can already rmdir the directory. If not,
3744 * orphanize it. For every dir item inside that gets deleted
3745 * later, we do this check again and rmdir it then if possible.
3746 * See the use of check_dirs for more details.
3748 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3753 ret = send_rmdir(sctx, valid_path);
3756 } else if (!is_orphan) {
3757 ret = orphanize_inode(sctx, sctx->cur_ino,
3758 sctx->cur_inode_gen, valid_path);
3764 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3765 ret = dup_ref(cur, &check_dirs);
3769 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3770 !list_empty(&sctx->deleted_refs)) {
3772 * We have a moved dir. Add the old parent to check_dirs
3774 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3776 ret = dup_ref(cur, &check_dirs);
3779 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3781 * We have a non dir inode. Go through all deleted refs and
3782 * unlink them if they were not already overwritten by other
3785 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3786 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3787 sctx->cur_ino, sctx->cur_inode_gen,
3788 cur->name, cur->name_len);
3792 ret = send_unlink(sctx, cur->full_path);
3796 ret = dup_ref(cur, &check_dirs);
3801 * If the inode is still orphan, unlink the orphan. This may
3802 * happen when a previous inode did overwrite the first ref
3803 * of this inode and no new refs were added for the current
3804 * inode. Unlinking does not mean that the inode is deleted in
3805 * all cases. There may still be links to this inode in other
3809 ret = send_unlink(sctx, valid_path);
3816 * We did collect all parent dirs where cur_inode was once located. We
3817 * now go through all these dirs and check if they are pending for
3818 * deletion and if it's finally possible to perform the rmdir now.
3819 * We also update the inode stats of the parent dirs here.
3821 list_for_each_entry(cur, &check_dirs, list) {
3823 * In case we had refs into dirs that were not processed yet,
3824 * we don't need to do the utime and rmdir logic for these dirs.
3825 * The dir will be processed later.
3827 if (cur->dir > sctx->cur_ino)
3830 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3834 if (ret == inode_state_did_create ||
3835 ret == inode_state_no_change) {
3836 /* TODO delayed utimes */
3837 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3840 } else if (ret == inode_state_did_delete &&
3841 cur->dir != last_dir_ino_rm) {
3842 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3847 ret = get_cur_path(sctx, cur->dir,
3848 cur->dir_gen, valid_path);
3851 ret = send_rmdir(sctx, valid_path);
3854 last_dir_ino_rm = cur->dir;
3862 __free_recorded_refs(&check_dirs);
3863 free_recorded_refs(sctx);
3864 fs_path_free(valid_path);
3868 static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3869 struct fs_path *name, void *ctx, struct list_head *refs)
3872 struct send_ctx *sctx = ctx;
3876 p = fs_path_alloc();
3880 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
3885 ret = get_cur_path(sctx, dir, gen, p);
3888 ret = fs_path_add_path(p, name);
3892 ret = __record_ref(refs, dir, gen, p);
3900 static int __record_new_ref(int num, u64 dir, int index,
3901 struct fs_path *name,
3904 struct send_ctx *sctx = ctx;
3905 return record_ref(sctx->send_root, num, dir, index, name,
3906 ctx, &sctx->new_refs);
3910 static int __record_deleted_ref(int num, u64 dir, int index,
3911 struct fs_path *name,
3914 struct send_ctx *sctx = ctx;
3915 return record_ref(sctx->parent_root, num, dir, index, name,
3916 ctx, &sctx->deleted_refs);
3919 static int record_new_ref(struct send_ctx *sctx)
3923 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3924 sctx->cmp_key, 0, __record_new_ref, sctx);
3933 static int record_deleted_ref(struct send_ctx *sctx)
3937 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3938 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3947 struct find_ref_ctx {
3950 struct btrfs_root *root;
3951 struct fs_path *name;
3955 static int __find_iref(int num, u64 dir, int index,
3956 struct fs_path *name,
3959 struct find_ref_ctx *ctx = ctx_;
3963 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3964 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3966 * To avoid doing extra lookups we'll only do this if everything
3969 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3973 if (dir_gen != ctx->dir_gen)
3975 ctx->found_idx = num;
3981 static int find_iref(struct btrfs_root *root,
3982 struct btrfs_path *path,
3983 struct btrfs_key *key,
3984 u64 dir, u64 dir_gen, struct fs_path *name)
3987 struct find_ref_ctx ctx;
3991 ctx.dir_gen = dir_gen;
3995 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3999 if (ctx.found_idx == -1)
4002 return ctx.found_idx;
4005 static int __record_changed_new_ref(int num, u64 dir, int index,
4006 struct fs_path *name,
4011 struct send_ctx *sctx = ctx;
4013 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
4018 ret = find_iref(sctx->parent_root, sctx->right_path,
4019 sctx->cmp_key, dir, dir_gen, name);
4021 ret = __record_new_ref(num, dir, index, name, sctx);
4028 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4029 struct fs_path *name,
4034 struct send_ctx *sctx = ctx;
4036 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4041 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4042 dir, dir_gen, name);
4044 ret = __record_deleted_ref(num, dir, index, name, sctx);
4051 static int record_changed_ref(struct send_ctx *sctx)
4055 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4056 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4059 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4060 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4070 * Record and process all refs at once. Needed when an inode changes the
4071 * generation number, which means that it was deleted and recreated.
4073 static int process_all_refs(struct send_ctx *sctx,
4074 enum btrfs_compare_tree_result cmd)
4077 struct btrfs_root *root;
4078 struct btrfs_path *path;
4079 struct btrfs_key key;
4080 struct btrfs_key found_key;
4081 struct extent_buffer *eb;
4083 iterate_inode_ref_t cb;
4084 int pending_move = 0;
4086 path = alloc_path_for_send();
4090 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4091 root = sctx->send_root;
4092 cb = __record_new_ref;
4093 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4094 root = sctx->parent_root;
4095 cb = __record_deleted_ref;
4097 btrfs_err(sctx->send_root->fs_info,
4098 "Wrong command %d in process_all_refs", cmd);
4103 key.objectid = sctx->cmp_key->objectid;
4104 key.type = BTRFS_INODE_REF_KEY;
4106 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4111 eb = path->nodes[0];
4112 slot = path->slots[0];
4113 if (slot >= btrfs_header_nritems(eb)) {
4114 ret = btrfs_next_leaf(root, path);
4122 btrfs_item_key_to_cpu(eb, &found_key, slot);
4124 if (found_key.objectid != key.objectid ||
4125 (found_key.type != BTRFS_INODE_REF_KEY &&
4126 found_key.type != BTRFS_INODE_EXTREF_KEY))
4129 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4135 btrfs_release_path(path);
4137 ret = process_recorded_refs(sctx, &pending_move);
4138 /* Only applicable to an incremental send. */
4139 ASSERT(pending_move == 0);
4142 btrfs_free_path(path);
4146 static int send_set_xattr(struct send_ctx *sctx,
4147 struct fs_path *path,
4148 const char *name, int name_len,
4149 const char *data, int data_len)
4153 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4157 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4158 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4159 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4161 ret = send_cmd(sctx);
4168 static int send_remove_xattr(struct send_ctx *sctx,
4169 struct fs_path *path,
4170 const char *name, int name_len)
4174 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4178 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4179 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4181 ret = send_cmd(sctx);
4188 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4189 const char *name, int name_len,
4190 const char *data, int data_len,
4194 struct send_ctx *sctx = ctx;
4196 posix_acl_xattr_header dummy_acl;
4198 /* Capabilities are emitted by finish_inode_if_needed */
4199 if (!strncmp(name, XATTR_NAME_CAPS, name_len))
4202 p = fs_path_alloc();
4207 * This hack is needed because empty acl's are stored as zero byte
4208 * data in xattrs. Problem with that is, that receiving these zero byte
4209 * acl's will fail later. To fix this, we send a dummy acl list that
4210 * only contains the version number and no entries.
4212 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4213 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4214 if (data_len == 0) {
4215 dummy_acl.a_version =
4216 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4217 data = (char *)&dummy_acl;
4218 data_len = sizeof(dummy_acl);
4222 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4226 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4233 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4234 const char *name, int name_len,
4235 const char *data, int data_len,
4239 struct send_ctx *sctx = ctx;
4242 p = fs_path_alloc();
4246 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4250 ret = send_remove_xattr(sctx, p, name, name_len);
4257 static int process_new_xattr(struct send_ctx *sctx)
4261 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4262 sctx->cmp_key, __process_new_xattr, sctx);
4267 static int process_deleted_xattr(struct send_ctx *sctx)
4271 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4272 sctx->cmp_key, __process_deleted_xattr, sctx);
4277 struct find_xattr_ctx {
4285 static int __find_xattr(int num, struct btrfs_key *di_key,
4286 const char *name, int name_len,
4287 const char *data, int data_len,
4288 u8 type, void *vctx)
4290 struct find_xattr_ctx *ctx = vctx;
4292 if (name_len == ctx->name_len &&
4293 strncmp(name, ctx->name, name_len) == 0) {
4294 ctx->found_idx = num;
4295 ctx->found_data_len = data_len;
4296 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
4297 if (!ctx->found_data)
4304 static int find_xattr(struct btrfs_root *root,
4305 struct btrfs_path *path,
4306 struct btrfs_key *key,
4307 const char *name, int name_len,
4308 char **data, int *data_len)
4311 struct find_xattr_ctx ctx;
4314 ctx.name_len = name_len;
4316 ctx.found_data = NULL;
4317 ctx.found_data_len = 0;
4319 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4323 if (ctx.found_idx == -1)
4326 *data = ctx.found_data;
4327 *data_len = ctx.found_data_len;
4329 kfree(ctx.found_data);
4331 return ctx.found_idx;
4335 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4336 const char *name, int name_len,
4337 const char *data, int data_len,
4341 struct send_ctx *sctx = ctx;
4342 char *found_data = NULL;
4343 int found_data_len = 0;
4345 ret = find_xattr(sctx->parent_root, sctx->right_path,
4346 sctx->cmp_key, name, name_len, &found_data,
4348 if (ret == -ENOENT) {
4349 ret = __process_new_xattr(num, di_key, name, name_len, data,
4350 data_len, type, ctx);
4351 } else if (ret >= 0) {
4352 if (data_len != found_data_len ||
4353 memcmp(data, found_data, data_len)) {
4354 ret = __process_new_xattr(num, di_key, name, name_len,
4355 data, data_len, type, ctx);
4365 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4366 const char *name, int name_len,
4367 const char *data, int data_len,
4371 struct send_ctx *sctx = ctx;
4373 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4374 name, name_len, NULL, NULL);
4376 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4377 data_len, type, ctx);
4384 static int process_changed_xattr(struct send_ctx *sctx)
4388 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4389 sctx->cmp_key, __process_changed_new_xattr, sctx);
4392 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4393 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4399 static int process_all_new_xattrs(struct send_ctx *sctx)
4402 struct btrfs_root *root;
4403 struct btrfs_path *path;
4404 struct btrfs_key key;
4405 struct btrfs_key found_key;
4406 struct extent_buffer *eb;
4409 path = alloc_path_for_send();
4413 root = sctx->send_root;
4415 key.objectid = sctx->cmp_key->objectid;
4416 key.type = BTRFS_XATTR_ITEM_KEY;
4418 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4423 eb = path->nodes[0];
4424 slot = path->slots[0];
4425 if (slot >= btrfs_header_nritems(eb)) {
4426 ret = btrfs_next_leaf(root, path);
4429 } else if (ret > 0) {
4436 btrfs_item_key_to_cpu(eb, &found_key, slot);
4437 if (found_key.objectid != key.objectid ||
4438 found_key.type != key.type) {
4443 ret = iterate_dir_item(root, path, &found_key,
4444 __process_new_xattr, sctx);
4452 btrfs_free_path(path);
4456 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4458 struct btrfs_root *root = sctx->send_root;
4459 struct btrfs_fs_info *fs_info = root->fs_info;
4460 struct inode *inode;
4463 struct btrfs_key key;
4464 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4466 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4469 key.objectid = sctx->cur_ino;
4470 key.type = BTRFS_INODE_ITEM_KEY;
4473 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4475 return PTR_ERR(inode);
4477 if (offset + len > i_size_read(inode)) {
4478 if (offset > i_size_read(inode))
4481 len = offset - i_size_read(inode);
4486 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4488 /* initial readahead */
4489 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4490 file_ra_state_init(&sctx->ra, inode->i_mapping);
4491 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4492 last_index - index + 1);
4494 while (index <= last_index) {
4495 unsigned cur_len = min_t(unsigned, len,
4496 PAGE_CACHE_SIZE - pg_offset);
4497 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4503 if (!PageUptodate(page)) {
4504 btrfs_readpage(NULL, page);
4506 if (!PageUptodate(page)) {
4508 page_cache_release(page);
4515 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4518 page_cache_release(page);
4530 * Read some bytes from the current inode/file and send a write command to
4533 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4537 ssize_t num_read = 0;
4539 p = fs_path_alloc();
4543 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4545 num_read = fill_read_buf(sctx, offset, len);
4546 if (num_read <= 0) {
4552 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4556 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4560 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4561 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4562 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4564 ret = send_cmd(sctx);
4575 * Send a clone command to user space.
4577 static int send_clone(struct send_ctx *sctx,
4578 u64 offset, u32 len,
4579 struct clone_root *clone_root)
4585 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4586 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4587 clone_root->root->objectid, clone_root->ino,
4588 clone_root->offset);
4590 p = fs_path_alloc();
4594 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4598 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4602 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4603 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4604 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4606 if (clone_root->root == sctx->send_root) {
4607 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4608 &gen, NULL, NULL, NULL, NULL);
4611 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4613 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4619 * If the parent we're using has a received_uuid set then use that as
4620 * our clone source as that is what we will look for when doing a
4623 * This covers the case that we create a snapshot off of a received
4624 * subvolume and then use that as the parent and try to receive on a
4627 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4628 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4629 clone_root->root->root_item.received_uuid);
4631 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4632 clone_root->root->root_item.uuid);
4633 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4634 le64_to_cpu(clone_root->root->root_item.ctransid));
4635 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4636 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4637 clone_root->offset);
4639 ret = send_cmd(sctx);
4648 * Send an update extent command to user space.
4650 static int send_update_extent(struct send_ctx *sctx,
4651 u64 offset, u32 len)
4656 p = fs_path_alloc();
4660 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4664 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4668 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4669 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4670 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4672 ret = send_cmd(sctx);
4680 static int send_hole(struct send_ctx *sctx, u64 end)
4682 struct fs_path *p = NULL;
4683 u64 offset = sctx->cur_inode_last_extent;
4687 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4688 return send_update_extent(sctx, offset, end - offset);
4690 p = fs_path_alloc();
4693 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4695 goto tlv_put_failure;
4696 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4697 while (offset < end) {
4698 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4700 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4703 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4704 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4705 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4706 ret = send_cmd(sctx);
4716 static int send_extent_data(struct send_ctx *sctx,
4722 if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
4723 return send_update_extent(sctx, offset, len);
4725 while (sent < len) {
4726 u64 size = len - sent;
4729 if (size > BTRFS_SEND_READ_SIZE)
4730 size = BTRFS_SEND_READ_SIZE;
4731 ret = send_write(sctx, offset + sent, size);
4742 * Search for a capability xattr related to sctx->cur_ino. If the capability is
4743 * found, call send_set_xattr function to emit it.
4745 * Return 0 if there isn't a capability, or when the capability was emitted
4746 * successfully, or < 0 if an error occurred.
4748 static int send_capabilities(struct send_ctx *sctx)
4750 struct fs_path *fspath = NULL;
4751 struct btrfs_path *path;
4752 struct btrfs_dir_item *di;
4753 struct extent_buffer *leaf;
4754 unsigned long data_ptr;
4759 path = alloc_path_for_send();
4763 di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino,
4764 XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0);
4766 /* There is no xattr for this inode */
4768 } else if (IS_ERR(di)) {
4773 leaf = path->nodes[0];
4774 buf_len = btrfs_dir_data_len(leaf, di);
4776 fspath = fs_path_alloc();
4777 buf = kmalloc(buf_len, GFP_KERNEL);
4778 if (!fspath || !buf) {
4783 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, fspath);
4787 data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di);
4788 read_extent_buffer(leaf, buf, data_ptr, buf_len);
4790 ret = send_set_xattr(sctx, fspath, XATTR_NAME_CAPS,
4791 strlen(XATTR_NAME_CAPS), buf, buf_len);
4794 fs_path_free(fspath);
4795 btrfs_free_path(path);
4799 static int clone_range(struct send_ctx *sctx,
4800 struct clone_root *clone_root,
4801 const u64 disk_byte,
4806 struct btrfs_path *path;
4807 struct btrfs_key key;
4810 path = alloc_path_for_send();
4815 * We can't send a clone operation for the entire range if we find
4816 * extent items in the respective range in the source file that
4817 * refer to different extents or if we find holes.
4818 * So check for that and do a mix of clone and regular write/copy
4819 * operations if needed.
4823 * mkfs.btrfs -f /dev/sda
4824 * mount /dev/sda /mnt
4825 * xfs_io -f -c "pwrite -S 0xaa 0K 100K" /mnt/foo
4826 * cp --reflink=always /mnt/foo /mnt/bar
4827 * xfs_io -c "pwrite -S 0xbb 50K 50K" /mnt/foo
4828 * btrfs subvolume snapshot -r /mnt /mnt/snap
4830 * If when we send the snapshot and we are processing file bar (which
4831 * has a higher inode number than foo) we blindly send a clone operation
4832 * for the [0, 100K[ range from foo to bar, the receiver ends up getting
4833 * a file bar that matches the content of file foo - iow, doesn't match
4834 * the content from bar in the original filesystem.
4836 key.objectid = clone_root->ino;
4837 key.type = BTRFS_EXTENT_DATA_KEY;
4838 key.offset = clone_root->offset;
4839 ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0);
4842 if (ret > 0 && path->slots[0] > 0) {
4843 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
4844 if (key.objectid == clone_root->ino &&
4845 key.type == BTRFS_EXTENT_DATA_KEY)
4850 struct extent_buffer *leaf = path->nodes[0];
4851 int slot = path->slots[0];
4852 struct btrfs_file_extent_item *ei;
4857 if (slot >= btrfs_header_nritems(leaf)) {
4858 ret = btrfs_next_leaf(clone_root->root, path);
4866 btrfs_item_key_to_cpu(leaf, &key, slot);
4869 * We might have an implicit trailing hole (NO_HOLES feature
4870 * enabled). We deal with it after leaving this loop.
4872 if (key.objectid != clone_root->ino ||
4873 key.type != BTRFS_EXTENT_DATA_KEY)
4876 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4877 type = btrfs_file_extent_type(leaf, ei);
4878 if (type == BTRFS_FILE_EXTENT_INLINE) {
4879 ext_len = btrfs_file_extent_inline_len(leaf, slot, ei);
4880 ext_len = PAGE_CACHE_ALIGN(ext_len);
4882 ext_len = btrfs_file_extent_num_bytes(leaf, ei);
4885 if (key.offset + ext_len <= clone_root->offset)
4888 if (key.offset > clone_root->offset) {
4889 /* Implicit hole, NO_HOLES feature enabled. */
4890 u64 hole_len = key.offset - clone_root->offset;
4894 ret = send_extent_data(sctx, offset, hole_len);
4902 clone_root->offset += hole_len;
4903 data_offset += hole_len;
4906 if (key.offset >= clone_root->offset + len)
4909 clone_len = min_t(u64, ext_len, len);
4911 if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
4912 btrfs_file_extent_offset(leaf, ei) == data_offset)
4913 ret = send_clone(sctx, offset, clone_len, clone_root);
4915 ret = send_extent_data(sctx, offset, clone_len);
4923 offset += clone_len;
4924 clone_root->offset += clone_len;
4925 data_offset += clone_len;
4931 ret = send_extent_data(sctx, offset, len);
4935 btrfs_free_path(path);
4939 static int send_write_or_clone(struct send_ctx *sctx,
4940 struct btrfs_path *path,
4941 struct btrfs_key *key,
4942 struct clone_root *clone_root)
4945 struct btrfs_file_extent_item *ei;
4946 u64 offset = key->offset;
4949 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4951 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4952 struct btrfs_file_extent_item);
4953 type = btrfs_file_extent_type(path->nodes[0], ei);
4954 if (type == BTRFS_FILE_EXTENT_INLINE) {
4955 len = btrfs_file_extent_inline_len(path->nodes[0],
4956 path->slots[0], ei);
4958 * it is possible the inline item won't cover the whole page,
4959 * but there may be items after this page. Make
4960 * sure to send the whole thing
4962 len = PAGE_CACHE_ALIGN(len);
4964 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4967 if (offset + len > sctx->cur_inode_size)
4968 len = sctx->cur_inode_size - offset;
4974 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4978 disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
4979 data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
4980 ret = clone_range(sctx, clone_root, disk_byte, data_offset,
4983 ret = send_extent_data(sctx, offset, len);
4989 static int is_extent_unchanged(struct send_ctx *sctx,
4990 struct btrfs_path *left_path,
4991 struct btrfs_key *ekey)
4994 struct btrfs_key key;
4995 struct btrfs_path *path = NULL;
4996 struct extent_buffer *eb;
4998 struct btrfs_key found_key;
4999 struct btrfs_file_extent_item *ei;
5004 u64 left_offset_fixed;
5012 path = alloc_path_for_send();
5016 eb = left_path->nodes[0];
5017 slot = left_path->slots[0];
5018 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5019 left_type = btrfs_file_extent_type(eb, ei);
5021 if (left_type != BTRFS_FILE_EXTENT_REG) {
5025 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5026 left_len = btrfs_file_extent_num_bytes(eb, ei);
5027 left_offset = btrfs_file_extent_offset(eb, ei);
5028 left_gen = btrfs_file_extent_generation(eb, ei);
5031 * Following comments will refer to these graphics. L is the left
5032 * extents which we are checking at the moment. 1-8 are the right
5033 * extents that we iterate.
5036 * |-1-|-2a-|-3-|-4-|-5-|-6-|
5039 * |--1--|-2b-|...(same as above)
5041 * Alternative situation. Happens on files where extents got split.
5043 * |-----------7-----------|-6-|
5045 * Alternative situation. Happens on files which got larger.
5048 * Nothing follows after 8.
5051 key.objectid = ekey->objectid;
5052 key.type = BTRFS_EXTENT_DATA_KEY;
5053 key.offset = ekey->offset;
5054 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
5063 * Handle special case where the right side has no extents at all.
5065 eb = path->nodes[0];
5066 slot = path->slots[0];
5067 btrfs_item_key_to_cpu(eb, &found_key, slot);
5068 if (found_key.objectid != key.objectid ||
5069 found_key.type != key.type) {
5070 /* If we're a hole then just pretend nothing changed */
5071 ret = (left_disknr) ? 0 : 1;
5076 * We're now on 2a, 2b or 7.
5079 while (key.offset < ekey->offset + left_len) {
5080 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
5081 right_type = btrfs_file_extent_type(eb, ei);
5082 if (right_type != BTRFS_FILE_EXTENT_REG &&
5083 right_type != BTRFS_FILE_EXTENT_INLINE) {
5088 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5089 right_len = btrfs_file_extent_inline_len(eb, slot, ei);
5090 right_len = PAGE_ALIGN(right_len);
5092 right_len = btrfs_file_extent_num_bytes(eb, ei);
5096 * Are we at extent 8? If yes, we know the extent is changed.
5097 * This may only happen on the first iteration.
5099 if (found_key.offset + right_len <= ekey->offset) {
5100 /* If we're a hole just pretend nothing changed */
5101 ret = (left_disknr) ? 0 : 1;
5106 * We just wanted to see if when we have an inline extent, what
5107 * follows it is a regular extent (wanted to check the above
5108 * condition for inline extents too). This should normally not
5109 * happen but it's possible for example when we have an inline
5110 * compressed extent representing data with a size matching
5111 * the page size (currently the same as sector size).
5113 if (right_type == BTRFS_FILE_EXTENT_INLINE) {
5118 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
5119 right_offset = btrfs_file_extent_offset(eb, ei);
5120 right_gen = btrfs_file_extent_generation(eb, ei);
5122 left_offset_fixed = left_offset;
5123 if (key.offset < ekey->offset) {
5124 /* Fix the right offset for 2a and 7. */
5125 right_offset += ekey->offset - key.offset;
5127 /* Fix the left offset for all behind 2a and 2b */
5128 left_offset_fixed += key.offset - ekey->offset;
5132 * Check if we have the same extent.
5134 if (left_disknr != right_disknr ||
5135 left_offset_fixed != right_offset ||
5136 left_gen != right_gen) {
5142 * Go to the next extent.
5144 ret = btrfs_next_item(sctx->parent_root, path);
5148 eb = path->nodes[0];
5149 slot = path->slots[0];
5150 btrfs_item_key_to_cpu(eb, &found_key, slot);
5152 if (ret || found_key.objectid != key.objectid ||
5153 found_key.type != key.type) {
5154 key.offset += right_len;
5157 if (found_key.offset != key.offset + right_len) {
5165 * We're now behind the left extent (treat as unchanged) or at the end
5166 * of the right side (treat as changed).
5168 if (key.offset >= ekey->offset + left_len)
5175 btrfs_free_path(path);
5179 static int get_last_extent(struct send_ctx *sctx, u64 offset)
5181 struct btrfs_path *path;
5182 struct btrfs_root *root = sctx->send_root;
5183 struct btrfs_file_extent_item *fi;
5184 struct btrfs_key key;
5189 path = alloc_path_for_send();
5193 sctx->cur_inode_last_extent = 0;
5195 key.objectid = sctx->cur_ino;
5196 key.type = BTRFS_EXTENT_DATA_KEY;
5197 key.offset = offset;
5198 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
5202 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5203 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
5206 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5207 struct btrfs_file_extent_item);
5208 type = btrfs_file_extent_type(path->nodes[0], fi);
5209 if (type == BTRFS_FILE_EXTENT_INLINE) {
5210 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5211 path->slots[0], fi);
5212 extent_end = ALIGN(key.offset + size,
5213 sctx->send_root->sectorsize);
5215 extent_end = key.offset +
5216 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5218 sctx->cur_inode_last_extent = extent_end;
5220 btrfs_free_path(path);
5224 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
5225 struct btrfs_key *key)
5227 struct btrfs_file_extent_item *fi;
5232 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
5235 if (sctx->cur_inode_last_extent == (u64)-1) {
5236 ret = get_last_extent(sctx, key->offset - 1);
5241 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5242 struct btrfs_file_extent_item);
5243 type = btrfs_file_extent_type(path->nodes[0], fi);
5244 if (type == BTRFS_FILE_EXTENT_INLINE) {
5245 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
5246 path->slots[0], fi);
5247 extent_end = ALIGN(key->offset + size,
5248 sctx->send_root->sectorsize);
5250 extent_end = key->offset +
5251 btrfs_file_extent_num_bytes(path->nodes[0], fi);
5254 if (path->slots[0] == 0 &&
5255 sctx->cur_inode_last_extent < key->offset) {
5257 * We might have skipped entire leafs that contained only
5258 * file extent items for our current inode. These leafs have
5259 * a generation number smaller (older) than the one in the
5260 * current leaf and the leaf our last extent came from, and
5261 * are located between these 2 leafs.
5263 ret = get_last_extent(sctx, key->offset - 1);
5268 if (sctx->cur_inode_last_extent < key->offset)
5269 ret = send_hole(sctx, key->offset);
5270 sctx->cur_inode_last_extent = extent_end;
5274 static int process_extent(struct send_ctx *sctx,
5275 struct btrfs_path *path,
5276 struct btrfs_key *key)
5278 struct clone_root *found_clone = NULL;
5281 if (S_ISLNK(sctx->cur_inode_mode))
5284 if (sctx->parent_root && !sctx->cur_inode_new) {
5285 ret = is_extent_unchanged(sctx, path, key);
5293 struct btrfs_file_extent_item *ei;
5296 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5297 struct btrfs_file_extent_item);
5298 type = btrfs_file_extent_type(path->nodes[0], ei);
5299 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5300 type == BTRFS_FILE_EXTENT_REG) {
5302 * The send spec does not have a prealloc command yet,
5303 * so just leave a hole for prealloc'ed extents until
5304 * we have enough commands queued up to justify rev'ing
5307 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5312 /* Have a hole, just skip it. */
5313 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5320 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5321 sctx->cur_inode_size, &found_clone);
5322 if (ret != -ENOENT && ret < 0)
5325 ret = send_write_or_clone(sctx, path, key, found_clone);
5329 ret = maybe_send_hole(sctx, path, key);
5334 static int process_all_extents(struct send_ctx *sctx)
5337 struct btrfs_root *root;
5338 struct btrfs_path *path;
5339 struct btrfs_key key;
5340 struct btrfs_key found_key;
5341 struct extent_buffer *eb;
5344 root = sctx->send_root;
5345 path = alloc_path_for_send();
5349 key.objectid = sctx->cmp_key->objectid;
5350 key.type = BTRFS_EXTENT_DATA_KEY;
5352 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5357 eb = path->nodes[0];
5358 slot = path->slots[0];
5360 if (slot >= btrfs_header_nritems(eb)) {
5361 ret = btrfs_next_leaf(root, path);
5364 } else if (ret > 0) {
5371 btrfs_item_key_to_cpu(eb, &found_key, slot);
5373 if (found_key.objectid != key.objectid ||
5374 found_key.type != key.type) {
5379 ret = process_extent(sctx, path, &found_key);
5387 btrfs_free_path(path);
5391 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5393 int *refs_processed)
5397 if (sctx->cur_ino == 0)
5399 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5400 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5402 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5405 ret = process_recorded_refs(sctx, pending_move);
5409 *refs_processed = 1;
5414 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5425 int pending_move = 0;
5426 int refs_processed = 0;
5428 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5434 * We have processed the refs and thus need to advance send_progress.
5435 * Now, calls to get_cur_xxx will take the updated refs of the current
5436 * inode into account.
5438 * On the other hand, if our current inode is a directory and couldn't
5439 * be moved/renamed because its parent was renamed/moved too and it has
5440 * a higher inode number, we can only move/rename our current inode
5441 * after we moved/renamed its parent. Therefore in this case operate on
5442 * the old path (pre move/rename) of our current inode, and the
5443 * move/rename will be performed later.
5445 if (refs_processed && !pending_move)
5446 sctx->send_progress = sctx->cur_ino + 1;
5448 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5450 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5453 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5454 &left_mode, &left_uid, &left_gid, NULL);
5458 if (!sctx->parent_root || sctx->cur_inode_new) {
5460 if (!S_ISLNK(sctx->cur_inode_mode))
5463 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5464 NULL, NULL, &right_mode, &right_uid,
5469 if (left_uid != right_uid || left_gid != right_gid)
5471 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5475 if (S_ISREG(sctx->cur_inode_mode)) {
5476 if (need_send_hole(sctx)) {
5477 if (sctx->cur_inode_last_extent == (u64)-1 ||
5478 sctx->cur_inode_last_extent <
5479 sctx->cur_inode_size) {
5480 ret = get_last_extent(sctx, (u64)-1);
5484 if (sctx->cur_inode_last_extent <
5485 sctx->cur_inode_size) {
5486 ret = send_hole(sctx, sctx->cur_inode_size);
5491 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5492 sctx->cur_inode_size);
5498 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5499 left_uid, left_gid);
5504 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5510 ret = send_capabilities(sctx);
5515 * If other directory inodes depended on our current directory
5516 * inode's move/rename, now do their move/rename operations.
5518 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5519 ret = apply_children_dir_moves(sctx);
5523 * Need to send that every time, no matter if it actually
5524 * changed between the two trees as we have done changes to
5525 * the inode before. If our inode is a directory and it's
5526 * waiting to be moved/renamed, we will send its utimes when
5527 * it's moved/renamed, therefore we don't need to do it here.
5529 sctx->send_progress = sctx->cur_ino + 1;
5530 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5539 static int changed_inode(struct send_ctx *sctx,
5540 enum btrfs_compare_tree_result result)
5543 struct btrfs_key *key = sctx->cmp_key;
5544 struct btrfs_inode_item *left_ii = NULL;
5545 struct btrfs_inode_item *right_ii = NULL;
5549 sctx->cur_ino = key->objectid;
5550 sctx->cur_inode_new_gen = 0;
5551 sctx->cur_inode_last_extent = (u64)-1;
5554 * Set send_progress to current inode. This will tell all get_cur_xxx
5555 * functions that the current inode's refs are not updated yet. Later,
5556 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5558 sctx->send_progress = sctx->cur_ino;
5560 if (result == BTRFS_COMPARE_TREE_NEW ||
5561 result == BTRFS_COMPARE_TREE_CHANGED) {
5562 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5563 sctx->left_path->slots[0],
5564 struct btrfs_inode_item);
5565 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5568 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5569 sctx->right_path->slots[0],
5570 struct btrfs_inode_item);
5571 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5574 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5575 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5576 sctx->right_path->slots[0],
5577 struct btrfs_inode_item);
5579 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5583 * The cur_ino = root dir case is special here. We can't treat
5584 * the inode as deleted+reused because it would generate a
5585 * stream that tries to delete/mkdir the root dir.
5587 if (left_gen != right_gen &&
5588 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5589 sctx->cur_inode_new_gen = 1;
5592 if (result == BTRFS_COMPARE_TREE_NEW) {
5593 sctx->cur_inode_gen = left_gen;
5594 sctx->cur_inode_new = 1;
5595 sctx->cur_inode_deleted = 0;
5596 sctx->cur_inode_size = btrfs_inode_size(
5597 sctx->left_path->nodes[0], left_ii);
5598 sctx->cur_inode_mode = btrfs_inode_mode(
5599 sctx->left_path->nodes[0], left_ii);
5600 sctx->cur_inode_rdev = btrfs_inode_rdev(
5601 sctx->left_path->nodes[0], left_ii);
5602 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5603 ret = send_create_inode_if_needed(sctx);
5604 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5605 sctx->cur_inode_gen = right_gen;
5606 sctx->cur_inode_new = 0;
5607 sctx->cur_inode_deleted = 1;
5608 sctx->cur_inode_size = btrfs_inode_size(
5609 sctx->right_path->nodes[0], right_ii);
5610 sctx->cur_inode_mode = btrfs_inode_mode(
5611 sctx->right_path->nodes[0], right_ii);
5612 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5614 * We need to do some special handling in case the inode was
5615 * reported as changed with a changed generation number. This
5616 * means that the original inode was deleted and new inode
5617 * reused the same inum. So we have to treat the old inode as
5618 * deleted and the new one as new.
5620 if (sctx->cur_inode_new_gen) {
5622 * First, process the inode as if it was deleted.
5624 sctx->cur_inode_gen = right_gen;
5625 sctx->cur_inode_new = 0;
5626 sctx->cur_inode_deleted = 1;
5627 sctx->cur_inode_size = btrfs_inode_size(
5628 sctx->right_path->nodes[0], right_ii);
5629 sctx->cur_inode_mode = btrfs_inode_mode(
5630 sctx->right_path->nodes[0], right_ii);
5631 ret = process_all_refs(sctx,
5632 BTRFS_COMPARE_TREE_DELETED);
5637 * Now process the inode as if it was new.
5639 sctx->cur_inode_gen = left_gen;
5640 sctx->cur_inode_new = 1;
5641 sctx->cur_inode_deleted = 0;
5642 sctx->cur_inode_size = btrfs_inode_size(
5643 sctx->left_path->nodes[0], left_ii);
5644 sctx->cur_inode_mode = btrfs_inode_mode(
5645 sctx->left_path->nodes[0], left_ii);
5646 sctx->cur_inode_rdev = btrfs_inode_rdev(
5647 sctx->left_path->nodes[0], left_ii);
5648 ret = send_create_inode_if_needed(sctx);
5652 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5656 * Advance send_progress now as we did not get into
5657 * process_recorded_refs_if_needed in the new_gen case.
5659 sctx->send_progress = sctx->cur_ino + 1;
5662 * Now process all extents and xattrs of the inode as if
5663 * they were all new.
5665 ret = process_all_extents(sctx);
5668 ret = process_all_new_xattrs(sctx);
5672 sctx->cur_inode_gen = left_gen;
5673 sctx->cur_inode_new = 0;
5674 sctx->cur_inode_new_gen = 0;
5675 sctx->cur_inode_deleted = 0;
5676 sctx->cur_inode_size = btrfs_inode_size(
5677 sctx->left_path->nodes[0], left_ii);
5678 sctx->cur_inode_mode = btrfs_inode_mode(
5679 sctx->left_path->nodes[0], left_ii);
5688 * We have to process new refs before deleted refs, but compare_trees gives us
5689 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5690 * first and later process them in process_recorded_refs.
5691 * For the cur_inode_new_gen case, we skip recording completely because
5692 * changed_inode did already initiate processing of refs. The reason for this is
5693 * that in this case, compare_tree actually compares the refs of 2 different
5694 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5695 * refs of the right tree as deleted and all refs of the left tree as new.
5697 static int changed_ref(struct send_ctx *sctx,
5698 enum btrfs_compare_tree_result result)
5702 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5704 if (!sctx->cur_inode_new_gen &&
5705 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5706 if (result == BTRFS_COMPARE_TREE_NEW)
5707 ret = record_new_ref(sctx);
5708 else if (result == BTRFS_COMPARE_TREE_DELETED)
5709 ret = record_deleted_ref(sctx);
5710 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5711 ret = record_changed_ref(sctx);
5718 * Process new/deleted/changed xattrs. We skip processing in the
5719 * cur_inode_new_gen case because changed_inode did already initiate processing
5720 * of xattrs. The reason is the same as in changed_ref
5722 static int changed_xattr(struct send_ctx *sctx,
5723 enum btrfs_compare_tree_result result)
5727 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5729 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5730 if (result == BTRFS_COMPARE_TREE_NEW)
5731 ret = process_new_xattr(sctx);
5732 else if (result == BTRFS_COMPARE_TREE_DELETED)
5733 ret = process_deleted_xattr(sctx);
5734 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5735 ret = process_changed_xattr(sctx);
5742 * Process new/deleted/changed extents. We skip processing in the
5743 * cur_inode_new_gen case because changed_inode did already initiate processing
5744 * of extents. The reason is the same as in changed_ref
5746 static int changed_extent(struct send_ctx *sctx,
5747 enum btrfs_compare_tree_result result)
5751 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5753 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5754 if (result != BTRFS_COMPARE_TREE_DELETED)
5755 ret = process_extent(sctx, sctx->left_path,
5762 static int dir_changed(struct send_ctx *sctx, u64 dir)
5764 u64 orig_gen, new_gen;
5767 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5772 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5777 return (orig_gen != new_gen) ? 1 : 0;
5780 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5781 struct btrfs_key *key)
5783 struct btrfs_inode_extref *extref;
5784 struct extent_buffer *leaf;
5785 u64 dirid = 0, last_dirid = 0;
5792 /* Easy case, just check this one dirid */
5793 if (key->type == BTRFS_INODE_REF_KEY) {
5794 dirid = key->offset;
5796 ret = dir_changed(sctx, dirid);
5800 leaf = path->nodes[0];
5801 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5802 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5803 while (cur_offset < item_size) {
5804 extref = (struct btrfs_inode_extref *)(ptr +
5806 dirid = btrfs_inode_extref_parent(leaf, extref);
5807 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5808 cur_offset += ref_name_len + sizeof(*extref);
5809 if (dirid == last_dirid)
5811 ret = dir_changed(sctx, dirid);
5821 * Updates compare related fields in sctx and simply forwards to the actual
5822 * changed_xxx functions.
5824 static int changed_cb(struct btrfs_root *left_root,
5825 struct btrfs_root *right_root,
5826 struct btrfs_path *left_path,
5827 struct btrfs_path *right_path,
5828 struct btrfs_key *key,
5829 enum btrfs_compare_tree_result result,
5833 struct send_ctx *sctx = ctx;
5835 if (result == BTRFS_COMPARE_TREE_SAME) {
5836 if (key->type == BTRFS_INODE_REF_KEY ||
5837 key->type == BTRFS_INODE_EXTREF_KEY) {
5838 ret = compare_refs(sctx, left_path, key);
5843 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5844 return maybe_send_hole(sctx, left_path, key);
5848 result = BTRFS_COMPARE_TREE_CHANGED;
5852 sctx->left_path = left_path;
5853 sctx->right_path = right_path;
5854 sctx->cmp_key = key;
5856 ret = finish_inode_if_needed(sctx, 0);
5860 /* Ignore non-FS objects */
5861 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5862 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5865 if (key->type == BTRFS_INODE_ITEM_KEY)
5866 ret = changed_inode(sctx, result);
5867 else if (key->type == BTRFS_INODE_REF_KEY ||
5868 key->type == BTRFS_INODE_EXTREF_KEY)
5869 ret = changed_ref(sctx, result);
5870 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5871 ret = changed_xattr(sctx, result);
5872 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5873 ret = changed_extent(sctx, result);
5879 static int full_send_tree(struct send_ctx *sctx)
5882 struct btrfs_root *send_root = sctx->send_root;
5883 struct btrfs_key key;
5884 struct btrfs_key found_key;
5885 struct btrfs_path *path;
5886 struct extent_buffer *eb;
5889 path = alloc_path_for_send();
5893 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5894 key.type = BTRFS_INODE_ITEM_KEY;
5897 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5904 eb = path->nodes[0];
5905 slot = path->slots[0];
5906 btrfs_item_key_to_cpu(eb, &found_key, slot);
5908 ret = changed_cb(send_root, NULL, path, NULL,
5909 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5913 key.objectid = found_key.objectid;
5914 key.type = found_key.type;
5915 key.offset = found_key.offset + 1;
5917 ret = btrfs_next_item(send_root, path);
5927 ret = finish_inode_if_needed(sctx, 1);
5930 btrfs_free_path(path);
5934 static int send_subvol(struct send_ctx *sctx)
5938 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5939 ret = send_header(sctx);
5944 ret = send_subvol_begin(sctx);
5948 if (sctx->parent_root) {
5949 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5953 ret = finish_inode_if_needed(sctx, 1);
5957 ret = full_send_tree(sctx);
5963 free_recorded_refs(sctx);
5968 * If orphan cleanup did remove any orphans from a root, it means the tree
5969 * was modified and therefore the commit root is not the same as the current
5970 * root anymore. This is a problem, because send uses the commit root and
5971 * therefore can see inode items that don't exist in the current root anymore,
5972 * and for example make calls to btrfs_iget, which will do tree lookups based
5973 * on the current root and not on the commit root. Those lookups will fail,
5974 * returning a -ESTALE error, and making send fail with that error. So make
5975 * sure a send does not see any orphans we have just removed, and that it will
5976 * see the same inodes regardless of whether a transaction commit happened
5977 * before it started (meaning that the commit root will be the same as the
5978 * current root) or not.
5980 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5983 struct btrfs_trans_handle *trans = NULL;
5986 if (sctx->parent_root &&
5987 sctx->parent_root->node != sctx->parent_root->commit_root)
5990 for (i = 0; i < sctx->clone_roots_cnt; i++)
5991 if (sctx->clone_roots[i].root->node !=
5992 sctx->clone_roots[i].root->commit_root)
5996 return btrfs_end_transaction(trans, sctx->send_root);
6001 /* Use any root, all fs roots will get their commit roots updated. */
6003 trans = btrfs_join_transaction(sctx->send_root);
6005 return PTR_ERR(trans);
6009 return btrfs_commit_transaction(trans, sctx->send_root);
6012 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
6014 spin_lock(&root->root_item_lock);
6015 root->send_in_progress--;
6017 * Not much left to do, we don't know why it's unbalanced and
6018 * can't blindly reset it to 0.
6020 if (root->send_in_progress < 0)
6021 btrfs_err(root->fs_info,
6022 "send_in_progres unbalanced %d root %llu",
6023 root->send_in_progress, root->root_key.objectid);
6024 spin_unlock(&root->root_item_lock);
6027 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
6030 struct btrfs_root *send_root;
6031 struct btrfs_root *clone_root;
6032 struct btrfs_fs_info *fs_info;
6033 struct btrfs_ioctl_send_args *arg = NULL;
6034 struct btrfs_key key;
6035 struct send_ctx *sctx = NULL;
6037 u64 *clone_sources_tmp = NULL;
6038 int clone_sources_to_rollback = 0;
6039 int sort_clone_roots = 0;
6042 if (!capable(CAP_SYS_ADMIN))
6045 send_root = BTRFS_I(file_inode(mnt_file))->root;
6046 fs_info = send_root->fs_info;
6049 * The subvolume must remain read-only during send, protect against
6050 * making it RW. This also protects against deletion.
6052 spin_lock(&send_root->root_item_lock);
6053 send_root->send_in_progress++;
6054 spin_unlock(&send_root->root_item_lock);
6057 * This is done when we lookup the root, it should already be complete
6058 * by the time we get here.
6060 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
6063 * Userspace tools do the checks and warn the user if it's
6066 if (!btrfs_root_readonly(send_root)) {
6071 arg = memdup_user(arg_, sizeof(*arg));
6078 if (!access_ok(VERIFY_READ, arg->clone_sources,
6079 sizeof(*arg->clone_sources) *
6080 arg->clone_sources_count)) {
6085 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
6090 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
6096 INIT_LIST_HEAD(&sctx->new_refs);
6097 INIT_LIST_HEAD(&sctx->deleted_refs);
6098 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
6099 INIT_LIST_HEAD(&sctx->name_cache_list);
6101 sctx->flags = arg->flags;
6103 sctx->send_filp = fget(arg->send_fd);
6104 if (!sctx->send_filp) {
6109 sctx->send_root = send_root;
6111 * Unlikely but possible, if the subvolume is marked for deletion but
6112 * is slow to remove the directory entry, send can still be started
6114 if (btrfs_root_dead(sctx->send_root)) {
6119 sctx->clone_roots_cnt = arg->clone_sources_count;
6121 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
6122 sctx->send_buf = vmalloc(sctx->send_max_size);
6123 if (!sctx->send_buf) {
6128 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
6129 if (!sctx->read_buf) {
6134 sctx->pending_dir_moves = RB_ROOT;
6135 sctx->waiting_dir_moves = RB_ROOT;
6136 sctx->orphan_dirs = RB_ROOT;
6138 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
6139 (arg->clone_sources_count + 1));
6140 if (!sctx->clone_roots) {
6145 if (arg->clone_sources_count) {
6146 clone_sources_tmp = vmalloc(arg->clone_sources_count *
6147 sizeof(*arg->clone_sources));
6148 if (!clone_sources_tmp) {
6153 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
6154 arg->clone_sources_count *
6155 sizeof(*arg->clone_sources));
6161 for (i = 0; i < arg->clone_sources_count; i++) {
6162 key.objectid = clone_sources_tmp[i];
6163 key.type = BTRFS_ROOT_ITEM_KEY;
6164 key.offset = (u64)-1;
6166 index = srcu_read_lock(&fs_info->subvol_srcu);
6168 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
6169 if (IS_ERR(clone_root)) {
6170 srcu_read_unlock(&fs_info->subvol_srcu, index);
6171 ret = PTR_ERR(clone_root);
6174 spin_lock(&clone_root->root_item_lock);
6175 if (!btrfs_root_readonly(clone_root) ||
6176 btrfs_root_dead(clone_root)) {
6177 spin_unlock(&clone_root->root_item_lock);
6178 srcu_read_unlock(&fs_info->subvol_srcu, index);
6182 clone_root->send_in_progress++;
6183 spin_unlock(&clone_root->root_item_lock);
6184 srcu_read_unlock(&fs_info->subvol_srcu, index);
6186 sctx->clone_roots[i].root = clone_root;
6187 clone_sources_to_rollback = i + 1;
6189 vfree(clone_sources_tmp);
6190 clone_sources_tmp = NULL;
6193 if (arg->parent_root) {
6194 key.objectid = arg->parent_root;
6195 key.type = BTRFS_ROOT_ITEM_KEY;
6196 key.offset = (u64)-1;
6198 index = srcu_read_lock(&fs_info->subvol_srcu);
6200 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
6201 if (IS_ERR(sctx->parent_root)) {
6202 srcu_read_unlock(&fs_info->subvol_srcu, index);
6203 ret = PTR_ERR(sctx->parent_root);
6207 spin_lock(&sctx->parent_root->root_item_lock);
6208 sctx->parent_root->send_in_progress++;
6209 if (!btrfs_root_readonly(sctx->parent_root) ||
6210 btrfs_root_dead(sctx->parent_root)) {
6211 spin_unlock(&sctx->parent_root->root_item_lock);
6212 srcu_read_unlock(&fs_info->subvol_srcu, index);
6216 spin_unlock(&sctx->parent_root->root_item_lock);
6218 srcu_read_unlock(&fs_info->subvol_srcu, index);
6222 * Clones from send_root are allowed, but only if the clone source
6223 * is behind the current send position. This is checked while searching
6224 * for possible clone sources.
6226 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
6228 /* We do a bsearch later */
6229 sort(sctx->clone_roots, sctx->clone_roots_cnt,
6230 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
6232 sort_clone_roots = 1;
6234 ret = ensure_commit_roots_uptodate(sctx);
6238 current->journal_info = BTRFS_SEND_TRANS_STUB;
6239 ret = send_subvol(sctx);
6240 current->journal_info = NULL;
6244 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
6245 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
6248 ret = send_cmd(sctx);
6254 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
6255 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
6257 struct pending_dir_move *pm;
6259 n = rb_first(&sctx->pending_dir_moves);
6260 pm = rb_entry(n, struct pending_dir_move, node);
6261 while (!list_empty(&pm->list)) {
6262 struct pending_dir_move *pm2;
6264 pm2 = list_first_entry(&pm->list,
6265 struct pending_dir_move, list);
6266 free_pending_move(sctx, pm2);
6268 free_pending_move(sctx, pm);
6271 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6272 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6274 struct waiting_dir_move *dm;
6276 n = rb_first(&sctx->waiting_dir_moves);
6277 dm = rb_entry(n, struct waiting_dir_move, node);
6278 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6282 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6283 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6285 struct orphan_dir_info *odi;
6287 n = rb_first(&sctx->orphan_dirs);
6288 odi = rb_entry(n, struct orphan_dir_info, node);
6289 free_orphan_dir_info(sctx, odi);
6292 if (sort_clone_roots) {
6293 for (i = 0; i < sctx->clone_roots_cnt; i++)
6294 btrfs_root_dec_send_in_progress(
6295 sctx->clone_roots[i].root);
6297 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6298 btrfs_root_dec_send_in_progress(
6299 sctx->clone_roots[i].root);
6301 btrfs_root_dec_send_in_progress(send_root);
6303 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6304 btrfs_root_dec_send_in_progress(sctx->parent_root);
6307 vfree(clone_sources_tmp);
6310 if (sctx->send_filp)
6311 fput(sctx->send_filp);
6313 vfree(sctx->clone_roots);
6314 vfree(sctx->send_buf);
6315 vfree(sctx->read_buf);
6317 name_cache_free(sctx);