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"
38 static int g_verbose = 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
58 unsigned int reversed:1;
59 unsigned int virtual_mem:1;
65 #define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
69 /* reused for each extent */
71 struct btrfs_root *root;
78 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
82 struct file *send_filp;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
112 u64 cur_inode_last_extent;
116 struct list_head new_refs;
117 struct list_head deleted_refs;
119 struct radix_tree_root name_cache;
120 struct list_head name_cache_list;
126 * We process inodes by their increasing order, so if before an
127 * incremental send we reverse the parent/child relationship of
128 * directories such that a directory with a lower inode number was
129 * the parent of a directory with a higher inode number, and the one
130 * becoming the new parent got renamed too, we can't rename/move the
131 * directory with lower inode number when we finish processing it - we
132 * must process the directory with higher inode number first, then
133 * rename/move it and then rename/move the directory with lower inode
134 * number. Example follows.
136 * Tree state when the first send was performed:
148 * Tree state when the second (incremental) send is performed:
157 * The sequence of steps that lead to the second state was:
159 * mv /a/b/c/d /a/b/c2/d2
160 * mv /a/b/c /a/b/c2/d2/cc
162 * "c" has lower inode number, but we can't move it (2nd mv operation)
163 * before we move "d", which has higher inode number.
165 * So we just memorize which move/rename operations must be performed
166 * later when their respective parent is processed and moved/renamed.
169 /* Indexed by parent directory inode number. */
170 struct rb_root pending_dir_moves;
173 * Reverse index, indexed by the inode number of a directory that
174 * is waiting for the move/rename of its immediate parent before its
175 * own move/rename can be performed.
177 struct rb_root waiting_dir_moves;
180 struct pending_dir_move {
182 struct list_head list;
186 struct list_head update_refs;
189 struct waiting_dir_move {
194 struct name_cache_entry {
195 struct list_head list;
197 * radix_tree has only 32bit entries but we need to handle 64bit inums.
198 * We use the lower 32bit of the 64bit inum to store it in the tree. If
199 * more then one inum would fall into the same entry, we use radix_list
200 * to store the additional entries. radix_list is also used to store
201 * entries where two entries have the same inum but different
204 struct list_head radix_list;
210 int need_later_update;
215 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
217 static int need_send_hole(struct send_ctx *sctx)
219 return (sctx->parent_root && !sctx->cur_inode_new &&
220 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
221 S_ISREG(sctx->cur_inode_mode));
224 static void fs_path_reset(struct fs_path *p)
227 p->start = p->buf + p->buf_len - 1;
237 static struct fs_path *fs_path_alloc(void)
241 p = kmalloc(sizeof(*p), GFP_NOFS);
246 p->buf = p->inline_buf;
247 p->buf_len = FS_PATH_INLINE_SIZE;
252 static struct fs_path *fs_path_alloc_reversed(void)
264 static void fs_path_free(struct fs_path *p)
268 if (p->buf != p->inline_buf) {
277 static int fs_path_len(struct fs_path *p)
279 return p->end - p->start;
282 static int fs_path_ensure_buf(struct fs_path *p, int len)
290 if (p->buf_len >= len)
293 path_len = p->end - p->start;
294 old_buf_len = p->buf_len;
295 len = PAGE_ALIGN(len);
297 if (p->buf == p->inline_buf) {
298 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
300 tmp_buf = vmalloc(len);
305 memcpy(tmp_buf, p->buf, p->buf_len);
309 if (p->virtual_mem) {
310 tmp_buf = vmalloc(len);
313 memcpy(tmp_buf, p->buf, p->buf_len);
316 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
318 tmp_buf = vmalloc(len);
321 memcpy(tmp_buf, p->buf, p->buf_len);
330 tmp_buf = p->buf + old_buf_len - path_len - 1;
331 p->end = p->buf + p->buf_len - 1;
332 p->start = p->end - path_len;
333 memmove(p->start, tmp_buf, path_len + 1);
336 p->end = p->start + path_len;
341 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
346 new_len = p->end - p->start + name_len;
347 if (p->start != p->end)
349 ret = fs_path_ensure_buf(p, new_len);
354 if (p->start != p->end)
356 p->start -= name_len;
357 p->prepared = p->start;
359 if (p->start != p->end)
361 p->prepared = p->end;
370 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
374 ret = fs_path_prepare_for_add(p, name_len);
377 memcpy(p->prepared, name, name_len);
384 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
388 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
391 memcpy(p->prepared, p2->start, p2->end - p2->start);
398 static int fs_path_add_from_extent_buffer(struct fs_path *p,
399 struct extent_buffer *eb,
400 unsigned long off, int len)
404 ret = fs_path_prepare_for_add(p, len);
408 read_extent_buffer(eb, p->prepared, off, len);
415 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
419 p->reversed = from->reversed;
422 ret = fs_path_add_path(p, from);
428 static void fs_path_unreverse(struct fs_path *p)
437 len = p->end - p->start;
439 p->end = p->start + len;
440 memmove(p->start, tmp, len + 1);
444 static struct btrfs_path *alloc_path_for_send(void)
446 struct btrfs_path *path;
448 path = btrfs_alloc_path();
451 path->search_commit_root = 1;
452 path->skip_locking = 1;
456 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
466 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
467 /* TODO handle that correctly */
468 /*if (ret == -ERESTARTSYS) {
487 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
489 struct btrfs_tlv_header *hdr;
490 int total_len = sizeof(*hdr) + len;
491 int left = sctx->send_max_size - sctx->send_size;
493 if (unlikely(left < total_len))
496 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
497 hdr->tlv_type = cpu_to_le16(attr);
498 hdr->tlv_len = cpu_to_le16(len);
499 memcpy(hdr + 1, data, len);
500 sctx->send_size += total_len;
505 #define TLV_PUT_DEFINE_INT(bits) \
506 static int tlv_put_u##bits(struct send_ctx *sctx, \
507 u##bits attr, u##bits value) \
509 __le##bits __tmp = cpu_to_le##bits(value); \
510 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
513 TLV_PUT_DEFINE_INT(64)
515 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
516 const char *str, int len)
520 return tlv_put(sctx, attr, str, len);
523 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
526 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
529 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
530 struct extent_buffer *eb,
531 struct btrfs_timespec *ts)
533 struct btrfs_timespec bts;
534 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
535 return tlv_put(sctx, attr, &bts, sizeof(bts));
539 #define TLV_PUT(sctx, attrtype, attrlen, data) \
541 ret = tlv_put(sctx, attrtype, attrlen, data); \
543 goto tlv_put_failure; \
546 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
548 ret = tlv_put_u##bits(sctx, attrtype, value); \
550 goto tlv_put_failure; \
553 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
554 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
555 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
556 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
557 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
559 ret = tlv_put_string(sctx, attrtype, str, len); \
561 goto tlv_put_failure; \
563 #define TLV_PUT_PATH(sctx, attrtype, p) \
565 ret = tlv_put_string(sctx, attrtype, p->start, \
566 p->end - p->start); \
568 goto tlv_put_failure; \
570 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
572 ret = tlv_put_uuid(sctx, attrtype, uuid); \
574 goto tlv_put_failure; \
576 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
578 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
580 goto tlv_put_failure; \
583 static int send_header(struct send_ctx *sctx)
585 struct btrfs_stream_header hdr;
587 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
588 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
590 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
595 * For each command/item we want to send to userspace, we call this function.
597 static int begin_cmd(struct send_ctx *sctx, int cmd)
599 struct btrfs_cmd_header *hdr;
601 if (WARN_ON(!sctx->send_buf))
604 BUG_ON(sctx->send_size);
606 sctx->send_size += sizeof(*hdr);
607 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
608 hdr->cmd = cpu_to_le16(cmd);
613 static int send_cmd(struct send_ctx *sctx)
616 struct btrfs_cmd_header *hdr;
619 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
620 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
623 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
624 hdr->crc = cpu_to_le32(crc);
626 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
629 sctx->total_send_size += sctx->send_size;
630 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
637 * Sends a move instruction to user space
639 static int send_rename(struct send_ctx *sctx,
640 struct fs_path *from, struct fs_path *to)
644 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
646 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
650 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
651 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
653 ret = send_cmd(sctx);
661 * Sends a link instruction to user space
663 static int send_link(struct send_ctx *sctx,
664 struct fs_path *path, struct fs_path *lnk)
668 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
670 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
674 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
675 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
677 ret = send_cmd(sctx);
685 * Sends an unlink instruction to user space
687 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
691 verbose_printk("btrfs: send_unlink %s\n", path->start);
693 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
697 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
699 ret = send_cmd(sctx);
707 * Sends a rmdir instruction to user space
709 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
713 verbose_printk("btrfs: send_rmdir %s\n", path->start);
715 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
719 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
721 ret = send_cmd(sctx);
729 * Helper function to retrieve some fields from an inode item.
731 static int get_inode_info(struct btrfs_root *root,
732 u64 ino, u64 *size, u64 *gen,
733 u64 *mode, u64 *uid, u64 *gid,
737 struct btrfs_inode_item *ii;
738 struct btrfs_key key;
739 struct btrfs_path *path;
741 path = alloc_path_for_send();
746 key.type = BTRFS_INODE_ITEM_KEY;
748 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
756 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
757 struct btrfs_inode_item);
759 *size = btrfs_inode_size(path->nodes[0], ii);
761 *gen = btrfs_inode_generation(path->nodes[0], ii);
763 *mode = btrfs_inode_mode(path->nodes[0], ii);
765 *uid = btrfs_inode_uid(path->nodes[0], ii);
767 *gid = btrfs_inode_gid(path->nodes[0], ii);
769 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
772 btrfs_free_path(path);
776 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
781 * Helper function to iterate the entries in ONE btrfs_inode_ref or
782 * btrfs_inode_extref.
783 * The iterate callback may return a non zero value to stop iteration. This can
784 * be a negative value for error codes or 1 to simply stop it.
786 * path must point to the INODE_REF or INODE_EXTREF when called.
788 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
789 struct btrfs_key *found_key, int resolve,
790 iterate_inode_ref_t iterate, void *ctx)
792 struct extent_buffer *eb = path->nodes[0];
793 struct btrfs_item *item;
794 struct btrfs_inode_ref *iref;
795 struct btrfs_inode_extref *extref;
796 struct btrfs_path *tmp_path;
800 int slot = path->slots[0];
807 unsigned long name_off;
808 unsigned long elem_size;
811 p = fs_path_alloc_reversed();
815 tmp_path = alloc_path_for_send();
822 if (found_key->type == BTRFS_INODE_REF_KEY) {
823 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
824 struct btrfs_inode_ref);
825 item = btrfs_item_nr(slot);
826 total = btrfs_item_size(eb, item);
827 elem_size = sizeof(*iref);
829 ptr = btrfs_item_ptr_offset(eb, slot);
830 total = btrfs_item_size_nr(eb, slot);
831 elem_size = sizeof(*extref);
834 while (cur < total) {
837 if (found_key->type == BTRFS_INODE_REF_KEY) {
838 iref = (struct btrfs_inode_ref *)(ptr + cur);
839 name_len = btrfs_inode_ref_name_len(eb, iref);
840 name_off = (unsigned long)(iref + 1);
841 index = btrfs_inode_ref_index(eb, iref);
842 dir = found_key->offset;
844 extref = (struct btrfs_inode_extref *)(ptr + cur);
845 name_len = btrfs_inode_extref_name_len(eb, extref);
846 name_off = (unsigned long)&extref->name;
847 index = btrfs_inode_extref_index(eb, extref);
848 dir = btrfs_inode_extref_parent(eb, extref);
852 start = btrfs_ref_to_path(root, tmp_path, name_len,
856 ret = PTR_ERR(start);
859 if (start < p->buf) {
860 /* overflow , try again with larger buffer */
861 ret = fs_path_ensure_buf(p,
862 p->buf_len + p->buf - start);
865 start = btrfs_ref_to_path(root, tmp_path,
870 ret = PTR_ERR(start);
873 BUG_ON(start < p->buf);
877 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
883 cur += elem_size + name_len;
884 ret = iterate(num, dir, index, p, ctx);
891 btrfs_free_path(tmp_path);
896 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
897 const char *name, int name_len,
898 const char *data, int data_len,
902 * Helper function to iterate the entries in ONE btrfs_dir_item.
903 * The iterate callback may return a non zero value to stop iteration. This can
904 * be a negative value for error codes or 1 to simply stop it.
906 * path must point to the dir item when called.
908 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
909 struct btrfs_key *found_key,
910 iterate_dir_item_t iterate, void *ctx)
913 struct extent_buffer *eb;
914 struct btrfs_item *item;
915 struct btrfs_dir_item *di;
916 struct btrfs_key di_key;
931 buf = kmalloc(buf_len, GFP_NOFS);
938 slot = path->slots[0];
939 item = btrfs_item_nr(slot);
940 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
943 total = btrfs_item_size(eb, item);
946 while (cur < total) {
947 name_len = btrfs_dir_name_len(eb, di);
948 data_len = btrfs_dir_data_len(eb, di);
949 type = btrfs_dir_type(eb, di);
950 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
952 if (name_len + data_len > buf_len) {
953 buf_len = PAGE_ALIGN(name_len + data_len);
955 buf2 = vmalloc(buf_len);
962 buf2 = krealloc(buf, buf_len, GFP_NOFS);
964 buf2 = vmalloc(buf_len);
978 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
979 name_len + data_len);
981 len = sizeof(*di) + name_len + data_len;
982 di = (struct btrfs_dir_item *)((char *)di + len);
985 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
986 data_len, type, ctx);
1005 static int __copy_first_ref(int num, u64 dir, int index,
1006 struct fs_path *p, void *ctx)
1009 struct fs_path *pt = ctx;
1011 ret = fs_path_copy(pt, p);
1015 /* we want the first only */
1020 * Retrieve the first path of an inode. If an inode has more then one
1021 * ref/hardlink, this is ignored.
1023 static int get_inode_path(struct btrfs_root *root,
1024 u64 ino, struct fs_path *path)
1027 struct btrfs_key key, found_key;
1028 struct btrfs_path *p;
1030 p = alloc_path_for_send();
1034 fs_path_reset(path);
1037 key.type = BTRFS_INODE_REF_KEY;
1040 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1047 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1048 if (found_key.objectid != ino ||
1049 (found_key.type != BTRFS_INODE_REF_KEY &&
1050 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1055 ret = iterate_inode_ref(root, p, &found_key, 1,
1056 __copy_first_ref, path);
1066 struct backref_ctx {
1067 struct send_ctx *sctx;
1069 /* number of total found references */
1073 * used for clones found in send_root. clones found behind cur_objectid
1074 * and cur_offset are not considered as allowed clones.
1079 /* may be truncated in case it's the last extent in a file */
1082 /* Just to check for bugs in backref resolving */
1086 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1088 u64 root = (u64)(uintptr_t)key;
1089 struct clone_root *cr = (struct clone_root *)elt;
1091 if (root < cr->root->objectid)
1093 if (root > cr->root->objectid)
1098 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1100 struct clone_root *cr1 = (struct clone_root *)e1;
1101 struct clone_root *cr2 = (struct clone_root *)e2;
1103 if (cr1->root->objectid < cr2->root->objectid)
1105 if (cr1->root->objectid > cr2->root->objectid)
1111 * Called for every backref that is found for the current extent.
1112 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1114 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1116 struct backref_ctx *bctx = ctx_;
1117 struct clone_root *found;
1121 /* First check if the root is in the list of accepted clone sources */
1122 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1123 bctx->sctx->clone_roots_cnt,
1124 sizeof(struct clone_root),
1125 __clone_root_cmp_bsearch);
1129 if (found->root == bctx->sctx->send_root &&
1130 ino == bctx->cur_objectid &&
1131 offset == bctx->cur_offset) {
1132 bctx->found_itself = 1;
1136 * There are inodes that have extents that lie behind its i_size. Don't
1137 * accept clones from these extents.
1139 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1144 if (offset + bctx->extent_len > i_size)
1148 * Make sure we don't consider clones from send_root that are
1149 * behind the current inode/offset.
1151 if (found->root == bctx->sctx->send_root) {
1153 * TODO for the moment we don't accept clones from the inode
1154 * that is currently send. We may change this when
1155 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1158 if (ino >= bctx->cur_objectid)
1161 if (ino > bctx->cur_objectid)
1163 if (offset + bctx->extent_len > bctx->cur_offset)
1169 found->found_refs++;
1170 if (ino < found->ino) {
1172 found->offset = offset;
1173 } else if (found->ino == ino) {
1175 * same extent found more then once in the same file.
1177 if (found->offset > offset + bctx->extent_len)
1178 found->offset = offset;
1185 * Given an inode, offset and extent item, it finds a good clone for a clone
1186 * instruction. Returns -ENOENT when none could be found. The function makes
1187 * sure that the returned clone is usable at the point where sending is at the
1188 * moment. This means, that no clones are accepted which lie behind the current
1191 * path must point to the extent item when called.
1193 static int find_extent_clone(struct send_ctx *sctx,
1194 struct btrfs_path *path,
1195 u64 ino, u64 data_offset,
1197 struct clone_root **found)
1204 u64 extent_item_pos;
1206 struct btrfs_file_extent_item *fi;
1207 struct extent_buffer *eb = path->nodes[0];
1208 struct backref_ctx *backref_ctx = NULL;
1209 struct clone_root *cur_clone_root;
1210 struct btrfs_key found_key;
1211 struct btrfs_path *tmp_path;
1215 tmp_path = alloc_path_for_send();
1219 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1225 if (data_offset >= ino_size) {
1227 * There may be extents that lie behind the file's size.
1228 * I at least had this in combination with snapshotting while
1229 * writing large files.
1235 fi = btrfs_item_ptr(eb, path->slots[0],
1236 struct btrfs_file_extent_item);
1237 extent_type = btrfs_file_extent_type(eb, fi);
1238 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1242 compressed = btrfs_file_extent_compression(eb, fi);
1244 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1245 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1246 if (disk_byte == 0) {
1250 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1252 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1253 &found_key, &flags);
1254 btrfs_release_path(tmp_path);
1258 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1264 * Setup the clone roots.
1266 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1267 cur_clone_root = sctx->clone_roots + i;
1268 cur_clone_root->ino = (u64)-1;
1269 cur_clone_root->offset = 0;
1270 cur_clone_root->found_refs = 0;
1273 backref_ctx->sctx = sctx;
1274 backref_ctx->found = 0;
1275 backref_ctx->cur_objectid = ino;
1276 backref_ctx->cur_offset = data_offset;
1277 backref_ctx->found_itself = 0;
1278 backref_ctx->extent_len = num_bytes;
1281 * The last extent of a file may be too large due to page alignment.
1282 * We need to adjust extent_len in this case so that the checks in
1283 * __iterate_backrefs work.
1285 if (data_offset + num_bytes >= ino_size)
1286 backref_ctx->extent_len = ino_size - data_offset;
1289 * Now collect all backrefs.
1291 if (compressed == BTRFS_COMPRESS_NONE)
1292 extent_item_pos = logical - found_key.objectid;
1294 extent_item_pos = 0;
1296 extent_item_pos = logical - found_key.objectid;
1297 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1298 found_key.objectid, extent_item_pos, 1,
1299 __iterate_backrefs, backref_ctx);
1304 if (!backref_ctx->found_itself) {
1305 /* found a bug in backref code? */
1307 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1308 "send_root. inode=%llu, offset=%llu, "
1309 "disk_byte=%llu found extent=%llu\n",
1310 ino, data_offset, disk_byte, found_key.objectid);
1314 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1316 "num_bytes=%llu, logical=%llu\n",
1317 data_offset, ino, num_bytes, logical);
1319 if (!backref_ctx->found)
1320 verbose_printk("btrfs: no clones found\n");
1322 cur_clone_root = NULL;
1323 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1324 if (sctx->clone_roots[i].found_refs) {
1325 if (!cur_clone_root)
1326 cur_clone_root = sctx->clone_roots + i;
1327 else if (sctx->clone_roots[i].root == sctx->send_root)
1328 /* prefer clones from send_root over others */
1329 cur_clone_root = sctx->clone_roots + i;
1334 if (cur_clone_root) {
1335 if (compressed != BTRFS_COMPRESS_NONE) {
1337 * Offsets given by iterate_extent_inodes() are relative
1338 * to the start of the extent, we need to add logical
1339 * offset from the file extent item.
1340 * (See why at backref.c:check_extent_in_eb())
1342 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1345 *found = cur_clone_root;
1352 btrfs_free_path(tmp_path);
1357 static int read_symlink(struct btrfs_root *root,
1359 struct fs_path *dest)
1362 struct btrfs_path *path;
1363 struct btrfs_key key;
1364 struct btrfs_file_extent_item *ei;
1370 path = alloc_path_for_send();
1375 key.type = BTRFS_EXTENT_DATA_KEY;
1377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1382 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1383 struct btrfs_file_extent_item);
1384 type = btrfs_file_extent_type(path->nodes[0], ei);
1385 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1386 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1387 BUG_ON(compression);
1389 off = btrfs_file_extent_inline_start(ei);
1390 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1392 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1395 btrfs_free_path(path);
1400 * Helper function to generate a file name that is unique in the root of
1401 * send_root and parent_root. This is used to generate names for orphan inodes.
1403 static int gen_unique_name(struct send_ctx *sctx,
1405 struct fs_path *dest)
1408 struct btrfs_path *path;
1409 struct btrfs_dir_item *di;
1414 path = alloc_path_for_send();
1419 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1421 if (len >= sizeof(tmp)) {
1422 /* should really not happen */
1427 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1428 path, BTRFS_FIRST_FREE_OBJECTID,
1429 tmp, strlen(tmp), 0);
1430 btrfs_release_path(path);
1436 /* not unique, try again */
1441 if (!sctx->parent_root) {
1447 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1448 path, BTRFS_FIRST_FREE_OBJECTID,
1449 tmp, strlen(tmp), 0);
1450 btrfs_release_path(path);
1456 /* not unique, try again */
1464 ret = fs_path_add(dest, tmp, strlen(tmp));
1467 btrfs_free_path(path);
1472 inode_state_no_change,
1473 inode_state_will_create,
1474 inode_state_did_create,
1475 inode_state_will_delete,
1476 inode_state_did_delete,
1479 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1487 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1489 if (ret < 0 && ret != -ENOENT)
1493 if (!sctx->parent_root) {
1494 right_ret = -ENOENT;
1496 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1497 NULL, NULL, NULL, NULL);
1498 if (ret < 0 && ret != -ENOENT)
1503 if (!left_ret && !right_ret) {
1504 if (left_gen == gen && right_gen == gen) {
1505 ret = inode_state_no_change;
1506 } else if (left_gen == gen) {
1507 if (ino < sctx->send_progress)
1508 ret = inode_state_did_create;
1510 ret = inode_state_will_create;
1511 } else if (right_gen == gen) {
1512 if (ino < sctx->send_progress)
1513 ret = inode_state_did_delete;
1515 ret = inode_state_will_delete;
1519 } else if (!left_ret) {
1520 if (left_gen == gen) {
1521 if (ino < sctx->send_progress)
1522 ret = inode_state_did_create;
1524 ret = inode_state_will_create;
1528 } else if (!right_ret) {
1529 if (right_gen == gen) {
1530 if (ino < sctx->send_progress)
1531 ret = inode_state_did_delete;
1533 ret = inode_state_will_delete;
1545 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1549 ret = get_cur_inode_state(sctx, ino, gen);
1553 if (ret == inode_state_no_change ||
1554 ret == inode_state_did_create ||
1555 ret == inode_state_will_delete)
1565 * Helper function to lookup a dir item in a dir.
1567 static int lookup_dir_item_inode(struct btrfs_root *root,
1568 u64 dir, const char *name, int name_len,
1573 struct btrfs_dir_item *di;
1574 struct btrfs_key key;
1575 struct btrfs_path *path;
1577 path = alloc_path_for_send();
1581 di = btrfs_lookup_dir_item(NULL, root, path,
1582 dir, name, name_len, 0);
1591 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1592 *found_inode = key.objectid;
1593 *found_type = btrfs_dir_type(path->nodes[0], di);
1596 btrfs_free_path(path);
1601 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1602 * generation of the parent dir and the name of the dir entry.
1604 static int get_first_ref(struct btrfs_root *root, u64 ino,
1605 u64 *dir, u64 *dir_gen, struct fs_path *name)
1608 struct btrfs_key key;
1609 struct btrfs_key found_key;
1610 struct btrfs_path *path;
1614 path = alloc_path_for_send();
1619 key.type = BTRFS_INODE_REF_KEY;
1622 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1626 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1628 if (ret || found_key.objectid != ino ||
1629 (found_key.type != BTRFS_INODE_REF_KEY &&
1630 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1635 if (key.type == BTRFS_INODE_REF_KEY) {
1636 struct btrfs_inode_ref *iref;
1637 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1638 struct btrfs_inode_ref);
1639 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1640 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1641 (unsigned long)(iref + 1),
1643 parent_dir = found_key.offset;
1645 struct btrfs_inode_extref *extref;
1646 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1647 struct btrfs_inode_extref);
1648 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1649 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1650 (unsigned long)&extref->name, len);
1651 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1655 btrfs_release_path(path);
1657 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1665 btrfs_free_path(path);
1669 static int is_first_ref(struct btrfs_root *root,
1671 const char *name, int name_len)
1674 struct fs_path *tmp_name;
1678 tmp_name = fs_path_alloc();
1682 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1686 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1691 ret = !memcmp(tmp_name->start, name, name_len);
1694 fs_path_free(tmp_name);
1699 * Used by process_recorded_refs to determine if a new ref would overwrite an
1700 * already existing ref. In case it detects an overwrite, it returns the
1701 * inode/gen in who_ino/who_gen.
1702 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1703 * to make sure later references to the overwritten inode are possible.
1704 * Orphanizing is however only required for the first ref of an inode.
1705 * process_recorded_refs does an additional is_first_ref check to see if
1706 * orphanizing is really required.
1708 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1709 const char *name, int name_len,
1710 u64 *who_ino, u64 *who_gen)
1714 u64 other_inode = 0;
1717 if (!sctx->parent_root)
1720 ret = is_inode_existent(sctx, dir, dir_gen);
1725 * If we have a parent root we need to verify that the parent dir was
1726 * not delted and then re-created, if it was then we have no overwrite
1727 * and we can just unlink this entry.
1729 if (sctx->parent_root) {
1730 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1732 if (ret < 0 && ret != -ENOENT)
1742 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1743 &other_inode, &other_type);
1744 if (ret < 0 && ret != -ENOENT)
1752 * Check if the overwritten ref was already processed. If yes, the ref
1753 * was already unlinked/moved, so we can safely assume that we will not
1754 * overwrite anything at this point in time.
1756 if (other_inode > sctx->send_progress) {
1757 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1758 who_gen, NULL, NULL, NULL, NULL);
1763 *who_ino = other_inode;
1773 * Checks if the ref was overwritten by an already processed inode. This is
1774 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1775 * thus the orphan name needs be used.
1776 * process_recorded_refs also uses it to avoid unlinking of refs that were
1779 static int did_overwrite_ref(struct send_ctx *sctx,
1780 u64 dir, u64 dir_gen,
1781 u64 ino, u64 ino_gen,
1782 const char *name, int name_len)
1789 if (!sctx->parent_root)
1792 ret = is_inode_existent(sctx, dir, dir_gen);
1796 /* check if the ref was overwritten by another ref */
1797 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1798 &ow_inode, &other_type);
1799 if (ret < 0 && ret != -ENOENT)
1802 /* was never and will never be overwritten */
1807 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1812 if (ow_inode == ino && gen == ino_gen) {
1817 /* we know that it is or will be overwritten. check this now */
1818 if (ow_inode < sctx->send_progress)
1828 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1829 * that got overwritten. This is used by process_recorded_refs to determine
1830 * if it has to use the path as returned by get_cur_path or the orphan name.
1832 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1835 struct fs_path *name = NULL;
1839 if (!sctx->parent_root)
1842 name = fs_path_alloc();
1846 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1850 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1851 name->start, fs_path_len(name));
1859 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1860 * so we need to do some special handling in case we have clashes. This function
1861 * takes care of this with the help of name_cache_entry::radix_list.
1862 * In case of error, nce is kfreed.
1864 static int name_cache_insert(struct send_ctx *sctx,
1865 struct name_cache_entry *nce)
1868 struct list_head *nce_head;
1870 nce_head = radix_tree_lookup(&sctx->name_cache,
1871 (unsigned long)nce->ino);
1873 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1878 INIT_LIST_HEAD(nce_head);
1880 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1887 list_add_tail(&nce->radix_list, nce_head);
1888 list_add_tail(&nce->list, &sctx->name_cache_list);
1889 sctx->name_cache_size++;
1894 static void name_cache_delete(struct send_ctx *sctx,
1895 struct name_cache_entry *nce)
1897 struct list_head *nce_head;
1899 nce_head = radix_tree_lookup(&sctx->name_cache,
1900 (unsigned long)nce->ino);
1903 list_del(&nce->radix_list);
1904 list_del(&nce->list);
1905 sctx->name_cache_size--;
1907 if (list_empty(nce_head)) {
1908 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1913 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1916 struct list_head *nce_head;
1917 struct name_cache_entry *cur;
1919 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1923 list_for_each_entry(cur, nce_head, radix_list) {
1924 if (cur->ino == ino && cur->gen == gen)
1931 * Removes the entry from the list and adds it back to the end. This marks the
1932 * entry as recently used so that name_cache_clean_unused does not remove it.
1934 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1936 list_del(&nce->list);
1937 list_add_tail(&nce->list, &sctx->name_cache_list);
1941 * Remove some entries from the beginning of name_cache_list.
1943 static void name_cache_clean_unused(struct send_ctx *sctx)
1945 struct name_cache_entry *nce;
1947 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1950 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1951 nce = list_entry(sctx->name_cache_list.next,
1952 struct name_cache_entry, list);
1953 name_cache_delete(sctx, nce);
1958 static void name_cache_free(struct send_ctx *sctx)
1960 struct name_cache_entry *nce;
1962 while (!list_empty(&sctx->name_cache_list)) {
1963 nce = list_entry(sctx->name_cache_list.next,
1964 struct name_cache_entry, list);
1965 name_cache_delete(sctx, nce);
1971 * Used by get_cur_path for each ref up to the root.
1972 * Returns 0 if it succeeded.
1973 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1974 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1975 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1976 * Returns <0 in case of error.
1978 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1980 int skip_name_cache,
1983 struct fs_path *dest)
1987 struct btrfs_path *path = NULL;
1988 struct name_cache_entry *nce = NULL;
1990 if (skip_name_cache)
1993 * First check if we already did a call to this function with the same
1994 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1995 * return the cached result.
1997 nce = name_cache_search(sctx, ino, gen);
1999 if (ino < sctx->send_progress && nce->need_later_update) {
2000 name_cache_delete(sctx, nce);
2004 name_cache_used(sctx, nce);
2005 *parent_ino = nce->parent_ino;
2006 *parent_gen = nce->parent_gen;
2007 ret = fs_path_add(dest, nce->name, nce->name_len);
2015 path = alloc_path_for_send();
2020 * If the inode is not existent yet, add the orphan name and return 1.
2021 * This should only happen for the parent dir that we determine in
2024 ret = is_inode_existent(sctx, ino, gen);
2029 ret = gen_unique_name(sctx, ino, gen, dest);
2038 * Depending on whether the inode was already processed or not, use
2039 * send_root or parent_root for ref lookup.
2041 if (ino < sctx->send_progress && !skip_name_cache)
2042 ret = get_first_ref(sctx->send_root, ino,
2043 parent_ino, parent_gen, dest);
2045 ret = get_first_ref(sctx->parent_root, ino,
2046 parent_ino, parent_gen, dest);
2051 * Check if the ref was overwritten by an inode's ref that was processed
2052 * earlier. If yes, treat as orphan and return 1.
2054 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2055 dest->start, dest->end - dest->start);
2059 fs_path_reset(dest);
2060 ret = gen_unique_name(sctx, ino, gen, dest);
2065 if (skip_name_cache)
2070 * Store the result of the lookup in the name cache.
2072 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2080 nce->parent_ino = *parent_ino;
2081 nce->parent_gen = *parent_gen;
2082 nce->name_len = fs_path_len(dest);
2084 strcpy(nce->name, dest->start);
2086 if (ino < sctx->send_progress)
2087 nce->need_later_update = 0;
2089 nce->need_later_update = 1;
2091 nce_ret = name_cache_insert(sctx, nce);
2094 name_cache_clean_unused(sctx);
2097 btrfs_free_path(path);
2102 * Magic happens here. This function returns the first ref to an inode as it
2103 * would look like while receiving the stream at this point in time.
2104 * We walk the path up to the root. For every inode in between, we check if it
2105 * was already processed/sent. If yes, we continue with the parent as found
2106 * in send_root. If not, we continue with the parent as found in parent_root.
2107 * If we encounter an inode that was deleted at this point in time, we use the
2108 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2109 * that were not created yet and overwritten inodes/refs.
2111 * When do we have have orphan inodes:
2112 * 1. When an inode is freshly created and thus no valid refs are available yet
2113 * 2. When a directory lost all it's refs (deleted) but still has dir items
2114 * inside which were not processed yet (pending for move/delete). If anyone
2115 * tried to get the path to the dir items, it would get a path inside that
2117 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2118 * of an unprocessed inode. If in that case the first ref would be
2119 * overwritten, the overwritten inode gets "orphanized". Later when we
2120 * process this overwritten inode, it is restored at a new place by moving
2123 * sctx->send_progress tells this function at which point in time receiving
2126 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2127 struct fs_path *dest)
2130 struct fs_path *name = NULL;
2131 u64 parent_inode = 0;
2134 int skip_name_cache = 0;
2136 name = fs_path_alloc();
2142 if (is_waiting_for_move(sctx, ino))
2143 skip_name_cache = 1;
2146 fs_path_reset(dest);
2148 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2149 fs_path_reset(name);
2151 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
2152 &parent_inode, &parent_gen, name);
2158 if (!skip_name_cache &&
2159 is_waiting_for_move(sctx, parent_inode))
2160 skip_name_cache = 1;
2162 ret = fs_path_add_path(dest, name);
2173 fs_path_unreverse(dest);
2178 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2180 static int send_subvol_begin(struct send_ctx *sctx)
2183 struct btrfs_root *send_root = sctx->send_root;
2184 struct btrfs_root *parent_root = sctx->parent_root;
2185 struct btrfs_path *path;
2186 struct btrfs_key key;
2187 struct btrfs_root_ref *ref;
2188 struct extent_buffer *leaf;
2192 path = btrfs_alloc_path();
2196 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2198 btrfs_free_path(path);
2202 key.objectid = send_root->objectid;
2203 key.type = BTRFS_ROOT_BACKREF_KEY;
2206 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2215 leaf = path->nodes[0];
2216 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2217 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2218 key.objectid != send_root->objectid) {
2222 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2223 namelen = btrfs_root_ref_name_len(leaf, ref);
2224 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2225 btrfs_release_path(path);
2228 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2232 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2237 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2238 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2239 sctx->send_root->root_item.uuid);
2240 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2241 le64_to_cpu(sctx->send_root->root_item.ctransid));
2243 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2244 sctx->parent_root->root_item.uuid);
2245 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2246 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2249 ret = send_cmd(sctx);
2253 btrfs_free_path(path);
2258 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2263 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2265 p = fs_path_alloc();
2269 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2273 ret = get_cur_path(sctx, ino, gen, p);
2276 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2277 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2279 ret = send_cmd(sctx);
2287 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2292 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2294 p = fs_path_alloc();
2298 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2302 ret = get_cur_path(sctx, ino, gen, p);
2305 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2306 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2308 ret = send_cmd(sctx);
2316 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2321 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2323 p = fs_path_alloc();
2327 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2331 ret = get_cur_path(sctx, ino, gen, p);
2334 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2335 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2336 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2338 ret = send_cmd(sctx);
2346 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2349 struct fs_path *p = NULL;
2350 struct btrfs_inode_item *ii;
2351 struct btrfs_path *path = NULL;
2352 struct extent_buffer *eb;
2353 struct btrfs_key key;
2356 verbose_printk("btrfs: send_utimes %llu\n", ino);
2358 p = fs_path_alloc();
2362 path = alloc_path_for_send();
2369 key.type = BTRFS_INODE_ITEM_KEY;
2371 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2375 eb = path->nodes[0];
2376 slot = path->slots[0];
2377 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2379 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2383 ret = get_cur_path(sctx, ino, gen, p);
2386 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2387 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2388 btrfs_inode_atime(ii));
2389 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2390 btrfs_inode_mtime(ii));
2391 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2392 btrfs_inode_ctime(ii));
2393 /* TODO Add otime support when the otime patches get into upstream */
2395 ret = send_cmd(sctx);
2400 btrfs_free_path(path);
2405 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2406 * a valid path yet because we did not process the refs yet. So, the inode
2407 * is created as orphan.
2409 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2418 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2420 p = fs_path_alloc();
2424 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2429 if (S_ISREG(mode)) {
2430 cmd = BTRFS_SEND_C_MKFILE;
2431 } else if (S_ISDIR(mode)) {
2432 cmd = BTRFS_SEND_C_MKDIR;
2433 } else if (S_ISLNK(mode)) {
2434 cmd = BTRFS_SEND_C_SYMLINK;
2435 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2436 cmd = BTRFS_SEND_C_MKNOD;
2437 } else if (S_ISFIFO(mode)) {
2438 cmd = BTRFS_SEND_C_MKFIFO;
2439 } else if (S_ISSOCK(mode)) {
2440 cmd = BTRFS_SEND_C_MKSOCK;
2442 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2443 (int)(mode & S_IFMT));
2448 ret = begin_cmd(sctx, cmd);
2452 ret = gen_unique_name(sctx, ino, gen, p);
2456 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2457 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2459 if (S_ISLNK(mode)) {
2461 ret = read_symlink(sctx->send_root, ino, p);
2464 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2465 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2466 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2467 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2468 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2471 ret = send_cmd(sctx);
2483 * We need some special handling for inodes that get processed before the parent
2484 * directory got created. See process_recorded_refs for details.
2485 * This function does the check if we already created the dir out of order.
2487 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2490 struct btrfs_path *path = NULL;
2491 struct btrfs_key key;
2492 struct btrfs_key found_key;
2493 struct btrfs_key di_key;
2494 struct extent_buffer *eb;
2495 struct btrfs_dir_item *di;
2498 path = alloc_path_for_send();
2505 key.type = BTRFS_DIR_INDEX_KEY;
2508 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2513 eb = path->nodes[0];
2514 slot = path->slots[0];
2515 btrfs_item_key_to_cpu(eb, &found_key, slot);
2517 if (ret || found_key.objectid != key.objectid ||
2518 found_key.type != key.type) {
2523 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2524 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2526 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2527 di_key.objectid < sctx->send_progress) {
2532 key.offset = found_key.offset + 1;
2533 btrfs_release_path(path);
2537 btrfs_free_path(path);
2542 * Only creates the inode if it is:
2543 * 1. Not a directory
2544 * 2. Or a directory which was not created already due to out of order
2545 * directories. See did_create_dir and process_recorded_refs for details.
2547 static int send_create_inode_if_needed(struct send_ctx *sctx)
2551 if (S_ISDIR(sctx->cur_inode_mode)) {
2552 ret = did_create_dir(sctx, sctx->cur_ino);
2561 ret = send_create_inode(sctx, sctx->cur_ino);
2569 struct recorded_ref {
2570 struct list_head list;
2573 struct fs_path *full_path;
2581 * We need to process new refs before deleted refs, but compare_tree gives us
2582 * everything mixed. So we first record all refs and later process them.
2583 * This function is a helper to record one ref.
2585 static int record_ref(struct list_head *head, u64 dir,
2586 u64 dir_gen, struct fs_path *path)
2588 struct recorded_ref *ref;
2590 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2595 ref->dir_gen = dir_gen;
2596 ref->full_path = path;
2598 ref->name = (char *)kbasename(ref->full_path->start);
2599 ref->name_len = ref->full_path->end - ref->name;
2600 ref->dir_path = ref->full_path->start;
2601 if (ref->name == ref->full_path->start)
2602 ref->dir_path_len = 0;
2604 ref->dir_path_len = ref->full_path->end -
2605 ref->full_path->start - 1 - ref->name_len;
2607 list_add_tail(&ref->list, head);
2611 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2613 struct recorded_ref *new;
2615 new = kmalloc(sizeof(*ref), GFP_NOFS);
2619 new->dir = ref->dir;
2620 new->dir_gen = ref->dir_gen;
2621 new->full_path = NULL;
2622 INIT_LIST_HEAD(&new->list);
2623 list_add_tail(&new->list, list);
2627 static void __free_recorded_refs(struct list_head *head)
2629 struct recorded_ref *cur;
2631 while (!list_empty(head)) {
2632 cur = list_entry(head->next, struct recorded_ref, list);
2633 fs_path_free(cur->full_path);
2634 list_del(&cur->list);
2639 static void free_recorded_refs(struct send_ctx *sctx)
2641 __free_recorded_refs(&sctx->new_refs);
2642 __free_recorded_refs(&sctx->deleted_refs);
2646 * Renames/moves a file/dir to its orphan name. Used when the first
2647 * ref of an unprocessed inode gets overwritten and for all non empty
2650 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2651 struct fs_path *path)
2654 struct fs_path *orphan;
2656 orphan = fs_path_alloc();
2660 ret = gen_unique_name(sctx, ino, gen, orphan);
2664 ret = send_rename(sctx, path, orphan);
2667 fs_path_free(orphan);
2672 * Returns 1 if a directory can be removed at this point in time.
2673 * We check this by iterating all dir items and checking if the inode behind
2674 * the dir item was already processed.
2676 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2679 struct btrfs_root *root = sctx->parent_root;
2680 struct btrfs_path *path;
2681 struct btrfs_key key;
2682 struct btrfs_key found_key;
2683 struct btrfs_key loc;
2684 struct btrfs_dir_item *di;
2687 * Don't try to rmdir the top/root subvolume dir.
2689 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2692 path = alloc_path_for_send();
2697 key.type = BTRFS_DIR_INDEX_KEY;
2701 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2705 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2708 if (ret || found_key.objectid != key.objectid ||
2709 found_key.type != key.type) {
2713 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2714 struct btrfs_dir_item);
2715 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2717 if (loc.objectid > send_progress) {
2722 btrfs_release_path(path);
2723 key.offset = found_key.offset + 1;
2729 btrfs_free_path(path);
2733 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2735 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2736 struct waiting_dir_move *entry;
2739 entry = rb_entry(n, struct waiting_dir_move, node);
2740 if (ino < entry->ino)
2742 else if (ino > entry->ino)
2750 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2752 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2753 struct rb_node *parent = NULL;
2754 struct waiting_dir_move *entry, *dm;
2756 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2763 entry = rb_entry(parent, struct waiting_dir_move, node);
2764 if (ino < entry->ino) {
2766 } else if (ino > entry->ino) {
2767 p = &(*p)->rb_right;
2774 rb_link_node(&dm->node, parent, p);
2775 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2779 static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2781 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2782 struct waiting_dir_move *entry;
2785 entry = rb_entry(n, struct waiting_dir_move, node);
2786 if (ino < entry->ino) {
2788 } else if (ino > entry->ino) {
2791 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2799 static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2801 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2802 struct rb_node *parent = NULL;
2803 struct pending_dir_move *entry, *pm;
2804 struct recorded_ref *cur;
2808 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2811 pm->parent_ino = parent_ino;
2812 pm->ino = sctx->cur_ino;
2813 pm->gen = sctx->cur_inode_gen;
2814 INIT_LIST_HEAD(&pm->list);
2815 INIT_LIST_HEAD(&pm->update_refs);
2816 RB_CLEAR_NODE(&pm->node);
2820 entry = rb_entry(parent, struct pending_dir_move, node);
2821 if (parent_ino < entry->parent_ino) {
2823 } else if (parent_ino > entry->parent_ino) {
2824 p = &(*p)->rb_right;
2831 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2832 ret = dup_ref(cur, &pm->update_refs);
2836 list_for_each_entry(cur, &sctx->new_refs, list) {
2837 ret = dup_ref(cur, &pm->update_refs);
2842 ret = add_waiting_dir_move(sctx, pm->ino);
2847 list_add_tail(&pm->list, &entry->list);
2849 rb_link_node(&pm->node, parent, p);
2850 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2855 __free_recorded_refs(&pm->update_refs);
2861 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2864 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2865 struct pending_dir_move *entry;
2868 entry = rb_entry(n, struct pending_dir_move, node);
2869 if (parent_ino < entry->parent_ino)
2871 else if (parent_ino > entry->parent_ino)
2879 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2881 struct fs_path *from_path = NULL;
2882 struct fs_path *to_path = NULL;
2883 u64 orig_progress = sctx->send_progress;
2884 struct recorded_ref *cur;
2887 from_path = fs_path_alloc();
2891 sctx->send_progress = pm->ino;
2892 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2896 to_path = fs_path_alloc();
2902 sctx->send_progress = sctx->cur_ino + 1;
2903 ret = del_waiting_dir_move(sctx, pm->ino);
2906 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2910 ret = send_rename(sctx, from_path, to_path);
2914 ret = send_utimes(sctx, pm->ino, pm->gen);
2919 * After rename/move, need to update the utimes of both new parent(s)
2920 * and old parent(s).
2922 list_for_each_entry(cur, &pm->update_refs, list) {
2923 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2929 fs_path_free(from_path);
2930 fs_path_free(to_path);
2931 sctx->send_progress = orig_progress;
2936 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2938 if (!list_empty(&m->list))
2940 if (!RB_EMPTY_NODE(&m->node))
2941 rb_erase(&m->node, &sctx->pending_dir_moves);
2942 __free_recorded_refs(&m->update_refs);
2946 static void tail_append_pending_moves(struct pending_dir_move *moves,
2947 struct list_head *stack)
2949 if (list_empty(&moves->list)) {
2950 list_add_tail(&moves->list, stack);
2953 list_splice_init(&moves->list, &list);
2954 list_add_tail(&moves->list, stack);
2955 list_splice_tail(&list, stack);
2959 static int apply_children_dir_moves(struct send_ctx *sctx)
2961 struct pending_dir_move *pm;
2962 struct list_head stack;
2963 u64 parent_ino = sctx->cur_ino;
2966 pm = get_pending_dir_moves(sctx, parent_ino);
2970 INIT_LIST_HEAD(&stack);
2971 tail_append_pending_moves(pm, &stack);
2973 while (!list_empty(&stack)) {
2974 pm = list_first_entry(&stack, struct pending_dir_move, list);
2975 parent_ino = pm->ino;
2976 ret = apply_dir_move(sctx, pm);
2977 free_pending_move(sctx, pm);
2980 pm = get_pending_dir_moves(sctx, parent_ino);
2982 tail_append_pending_moves(pm, &stack);
2987 while (!list_empty(&stack)) {
2988 pm = list_first_entry(&stack, struct pending_dir_move, list);
2989 free_pending_move(sctx, pm);
2994 static int wait_for_parent_move(struct send_ctx *sctx,
2995 struct recorded_ref *parent_ref)
2998 u64 ino = parent_ref->dir;
2999 u64 parent_ino_before, parent_ino_after;
3000 u64 new_gen, old_gen;
3001 struct fs_path *path_before = NULL;
3002 struct fs_path *path_after = NULL;
3005 if (parent_ref->dir <= sctx->cur_ino)
3008 if (is_waiting_for_move(sctx, ino))
3011 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3012 NULL, NULL, NULL, NULL);
3018 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3019 NULL, NULL, NULL, NULL);
3023 if (new_gen != old_gen)
3026 path_before = fs_path_alloc();
3030 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3032 if (ret == -ENOENT) {
3035 } else if (ret < 0) {
3039 path_after = fs_path_alloc();
3045 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3047 if (ret == -ENOENT) {
3050 } else if (ret < 0) {
3054 len1 = fs_path_len(path_before);
3055 len2 = fs_path_len(path_after);
3056 if ((parent_ino_before != parent_ino_after) && (len1 != len2 ||
3057 memcmp(path_before->start, path_after->start, len1))) {
3064 fs_path_free(path_before);
3065 fs_path_free(path_after);
3071 * This does all the move/link/unlink/rmdir magic.
3073 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3076 struct recorded_ref *cur;
3077 struct recorded_ref *cur2;
3078 struct list_head check_dirs;
3079 struct fs_path *valid_path = NULL;
3082 int did_overwrite = 0;
3085 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3088 * This should never happen as the root dir always has the same ref
3089 * which is always '..'
3091 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3092 INIT_LIST_HEAD(&check_dirs);
3094 valid_path = fs_path_alloc();
3101 * First, check if the first ref of the current inode was overwritten
3102 * before. If yes, we know that the current inode was already orphanized
3103 * and thus use the orphan name. If not, we can use get_cur_path to
3104 * get the path of the first ref as it would like while receiving at
3105 * this point in time.
3106 * New inodes are always orphan at the beginning, so force to use the
3107 * orphan name in this case.
3108 * The first ref is stored in valid_path and will be updated if it
3109 * gets moved around.
3111 if (!sctx->cur_inode_new) {
3112 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3113 sctx->cur_inode_gen);
3119 if (sctx->cur_inode_new || did_overwrite) {
3120 ret = gen_unique_name(sctx, sctx->cur_ino,
3121 sctx->cur_inode_gen, valid_path);
3126 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3132 list_for_each_entry(cur, &sctx->new_refs, list) {
3134 * We may have refs where the parent directory does not exist
3135 * yet. This happens if the parent directories inum is higher
3136 * the the current inum. To handle this case, we create the
3137 * parent directory out of order. But we need to check if this
3138 * did already happen before due to other refs in the same dir.
3140 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3143 if (ret == inode_state_will_create) {
3146 * First check if any of the current inodes refs did
3147 * already create the dir.
3149 list_for_each_entry(cur2, &sctx->new_refs, list) {
3152 if (cur2->dir == cur->dir) {
3159 * If that did not happen, check if a previous inode
3160 * did already create the dir.
3163 ret = did_create_dir(sctx, cur->dir);
3167 ret = send_create_inode(sctx, cur->dir);
3174 * Check if this new ref would overwrite the first ref of
3175 * another unprocessed inode. If yes, orphanize the
3176 * overwritten inode. If we find an overwritten ref that is
3177 * not the first ref, simply unlink it.
3179 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3180 cur->name, cur->name_len,
3181 &ow_inode, &ow_gen);
3185 ret = is_first_ref(sctx->parent_root,
3186 ow_inode, cur->dir, cur->name,
3191 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3196 ret = send_unlink(sctx, cur->full_path);
3203 * link/move the ref to the new place. If we have an orphan
3204 * inode, move it and update valid_path. If not, link or move
3205 * it depending on the inode mode.
3208 ret = send_rename(sctx, valid_path, cur->full_path);
3212 ret = fs_path_copy(valid_path, cur->full_path);
3216 if (S_ISDIR(sctx->cur_inode_mode)) {
3218 * Dirs can't be linked, so move it. For moved
3219 * dirs, we always have one new and one deleted
3220 * ref. The deleted ref is ignored later.
3222 ret = wait_for_parent_move(sctx, cur);
3226 ret = add_pending_dir_move(sctx,
3230 ret = send_rename(sctx, valid_path,
3233 ret = fs_path_copy(valid_path,
3239 ret = send_link(sctx, cur->full_path,
3245 ret = dup_ref(cur, &check_dirs);
3250 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3252 * Check if we can already rmdir the directory. If not,
3253 * orphanize it. For every dir item inside that gets deleted
3254 * later, we do this check again and rmdir it then if possible.
3255 * See the use of check_dirs for more details.
3257 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3261 ret = send_rmdir(sctx, valid_path);
3264 } else if (!is_orphan) {
3265 ret = orphanize_inode(sctx, sctx->cur_ino,
3266 sctx->cur_inode_gen, valid_path);
3272 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3273 ret = dup_ref(cur, &check_dirs);
3277 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3278 !list_empty(&sctx->deleted_refs)) {
3280 * We have a moved dir. Add the old parent to check_dirs
3282 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3284 ret = dup_ref(cur, &check_dirs);
3287 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3289 * We have a non dir inode. Go through all deleted refs and
3290 * unlink them if they were not already overwritten by other
3293 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3294 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3295 sctx->cur_ino, sctx->cur_inode_gen,
3296 cur->name, cur->name_len);
3300 ret = send_unlink(sctx, cur->full_path);
3304 ret = dup_ref(cur, &check_dirs);
3309 * If the inode is still orphan, unlink the orphan. This may
3310 * happen when a previous inode did overwrite the first ref
3311 * of this inode and no new refs were added for the current
3312 * inode. Unlinking does not mean that the inode is deleted in
3313 * all cases. There may still be links to this inode in other
3317 ret = send_unlink(sctx, valid_path);
3324 * We did collect all parent dirs where cur_inode was once located. We
3325 * now go through all these dirs and check if they are pending for
3326 * deletion and if it's finally possible to perform the rmdir now.
3327 * We also update the inode stats of the parent dirs here.
3329 list_for_each_entry(cur, &check_dirs, list) {
3331 * In case we had refs into dirs that were not processed yet,
3332 * we don't need to do the utime and rmdir logic for these dirs.
3333 * The dir will be processed later.
3335 if (cur->dir > sctx->cur_ino)
3338 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3342 if (ret == inode_state_did_create ||
3343 ret == inode_state_no_change) {
3344 /* TODO delayed utimes */
3345 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3348 } else if (ret == inode_state_did_delete) {
3349 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
3353 ret = get_cur_path(sctx, cur->dir,
3354 cur->dir_gen, valid_path);
3357 ret = send_rmdir(sctx, valid_path);
3367 __free_recorded_refs(&check_dirs);
3368 free_recorded_refs(sctx);
3369 fs_path_free(valid_path);
3373 static int __record_new_ref(int num, u64 dir, int index,
3374 struct fs_path *name,
3378 struct send_ctx *sctx = ctx;
3382 p = fs_path_alloc();
3386 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3391 ret = get_cur_path(sctx, dir, gen, p);
3394 ret = fs_path_add_path(p, name);
3398 ret = record_ref(&sctx->new_refs, dir, gen, p);
3406 static int __record_deleted_ref(int num, u64 dir, int index,
3407 struct fs_path *name,
3411 struct send_ctx *sctx = ctx;
3415 p = fs_path_alloc();
3419 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3424 ret = get_cur_path(sctx, dir, gen, p);
3427 ret = fs_path_add_path(p, name);
3431 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3439 static int record_new_ref(struct send_ctx *sctx)
3443 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3444 sctx->cmp_key, 0, __record_new_ref, sctx);
3453 static int record_deleted_ref(struct send_ctx *sctx)
3457 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3458 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3467 struct find_ref_ctx {
3470 struct btrfs_root *root;
3471 struct fs_path *name;
3475 static int __find_iref(int num, u64 dir, int index,
3476 struct fs_path *name,
3479 struct find_ref_ctx *ctx = ctx_;
3483 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3484 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3486 * To avoid doing extra lookups we'll only do this if everything
3489 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3493 if (dir_gen != ctx->dir_gen)
3495 ctx->found_idx = num;
3501 static int find_iref(struct btrfs_root *root,
3502 struct btrfs_path *path,
3503 struct btrfs_key *key,
3504 u64 dir, u64 dir_gen, struct fs_path *name)
3507 struct find_ref_ctx ctx;
3511 ctx.dir_gen = dir_gen;
3515 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3519 if (ctx.found_idx == -1)
3522 return ctx.found_idx;
3525 static int __record_changed_new_ref(int num, u64 dir, int index,
3526 struct fs_path *name,
3531 struct send_ctx *sctx = ctx;
3533 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3538 ret = find_iref(sctx->parent_root, sctx->right_path,
3539 sctx->cmp_key, dir, dir_gen, name);
3541 ret = __record_new_ref(num, dir, index, name, sctx);
3548 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3549 struct fs_path *name,
3554 struct send_ctx *sctx = ctx;
3556 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3561 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3562 dir, dir_gen, name);
3564 ret = __record_deleted_ref(num, dir, index, name, sctx);
3571 static int record_changed_ref(struct send_ctx *sctx)
3575 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3576 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3579 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3580 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3590 * Record and process all refs at once. Needed when an inode changes the
3591 * generation number, which means that it was deleted and recreated.
3593 static int process_all_refs(struct send_ctx *sctx,
3594 enum btrfs_compare_tree_result cmd)
3597 struct btrfs_root *root;
3598 struct btrfs_path *path;
3599 struct btrfs_key key;
3600 struct btrfs_key found_key;
3601 struct extent_buffer *eb;
3603 iterate_inode_ref_t cb;
3604 int pending_move = 0;
3606 path = alloc_path_for_send();
3610 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3611 root = sctx->send_root;
3612 cb = __record_new_ref;
3613 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3614 root = sctx->parent_root;
3615 cb = __record_deleted_ref;
3620 key.objectid = sctx->cmp_key->objectid;
3621 key.type = BTRFS_INODE_REF_KEY;
3624 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3630 eb = path->nodes[0];
3631 slot = path->slots[0];
3632 btrfs_item_key_to_cpu(eb, &found_key, slot);
3634 if (found_key.objectid != key.objectid ||
3635 (found_key.type != BTRFS_INODE_REF_KEY &&
3636 found_key.type != BTRFS_INODE_EXTREF_KEY))
3639 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3640 btrfs_release_path(path);
3644 key.offset = found_key.offset + 1;
3646 btrfs_release_path(path);
3648 ret = process_recorded_refs(sctx, &pending_move);
3649 /* Only applicable to an incremental send. */
3650 ASSERT(pending_move == 0);
3653 btrfs_free_path(path);
3657 static int send_set_xattr(struct send_ctx *sctx,
3658 struct fs_path *path,
3659 const char *name, int name_len,
3660 const char *data, int data_len)
3664 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3668 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3669 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3670 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3672 ret = send_cmd(sctx);
3679 static int send_remove_xattr(struct send_ctx *sctx,
3680 struct fs_path *path,
3681 const char *name, int name_len)
3685 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3689 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3690 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3692 ret = send_cmd(sctx);
3699 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3700 const char *name, int name_len,
3701 const char *data, int data_len,
3705 struct send_ctx *sctx = ctx;
3707 posix_acl_xattr_header dummy_acl;
3709 p = fs_path_alloc();
3714 * This hack is needed because empty acl's are stored as zero byte
3715 * data in xattrs. Problem with that is, that receiving these zero byte
3716 * acl's will fail later. To fix this, we send a dummy acl list that
3717 * only contains the version number and no entries.
3719 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3720 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3721 if (data_len == 0) {
3722 dummy_acl.a_version =
3723 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3724 data = (char *)&dummy_acl;
3725 data_len = sizeof(dummy_acl);
3729 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3733 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3740 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3741 const char *name, int name_len,
3742 const char *data, int data_len,
3746 struct send_ctx *sctx = ctx;
3749 p = fs_path_alloc();
3753 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3757 ret = send_remove_xattr(sctx, p, name, name_len);
3764 static int process_new_xattr(struct send_ctx *sctx)
3768 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3769 sctx->cmp_key, __process_new_xattr, sctx);
3774 static int process_deleted_xattr(struct send_ctx *sctx)
3778 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3779 sctx->cmp_key, __process_deleted_xattr, sctx);
3784 struct find_xattr_ctx {
3792 static int __find_xattr(int num, struct btrfs_key *di_key,
3793 const char *name, int name_len,
3794 const char *data, int data_len,
3795 u8 type, void *vctx)
3797 struct find_xattr_ctx *ctx = vctx;
3799 if (name_len == ctx->name_len &&
3800 strncmp(name, ctx->name, name_len) == 0) {
3801 ctx->found_idx = num;
3802 ctx->found_data_len = data_len;
3803 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
3804 if (!ctx->found_data)
3811 static int find_xattr(struct btrfs_root *root,
3812 struct btrfs_path *path,
3813 struct btrfs_key *key,
3814 const char *name, int name_len,
3815 char **data, int *data_len)
3818 struct find_xattr_ctx ctx;
3821 ctx.name_len = name_len;
3823 ctx.found_data = NULL;
3824 ctx.found_data_len = 0;
3826 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
3830 if (ctx.found_idx == -1)
3833 *data = ctx.found_data;
3834 *data_len = ctx.found_data_len;
3836 kfree(ctx.found_data);
3838 return ctx.found_idx;
3842 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3843 const char *name, int name_len,
3844 const char *data, int data_len,
3848 struct send_ctx *sctx = ctx;
3849 char *found_data = NULL;
3850 int found_data_len = 0;
3852 ret = find_xattr(sctx->parent_root, sctx->right_path,
3853 sctx->cmp_key, name, name_len, &found_data,
3855 if (ret == -ENOENT) {
3856 ret = __process_new_xattr(num, di_key, name, name_len, data,
3857 data_len, type, ctx);
3858 } else if (ret >= 0) {
3859 if (data_len != found_data_len ||
3860 memcmp(data, found_data, data_len)) {
3861 ret = __process_new_xattr(num, di_key, name, name_len,
3862 data, data_len, type, ctx);
3872 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3873 const char *name, int name_len,
3874 const char *data, int data_len,
3878 struct send_ctx *sctx = ctx;
3880 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3881 name, name_len, NULL, NULL);
3883 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3884 data_len, type, ctx);
3891 static int process_changed_xattr(struct send_ctx *sctx)
3895 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3896 sctx->cmp_key, __process_changed_new_xattr, sctx);
3899 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3900 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3906 static int process_all_new_xattrs(struct send_ctx *sctx)
3909 struct btrfs_root *root;
3910 struct btrfs_path *path;
3911 struct btrfs_key key;
3912 struct btrfs_key found_key;
3913 struct extent_buffer *eb;
3916 path = alloc_path_for_send();
3920 root = sctx->send_root;
3922 key.objectid = sctx->cmp_key->objectid;
3923 key.type = BTRFS_XATTR_ITEM_KEY;
3926 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3934 eb = path->nodes[0];
3935 slot = path->slots[0];
3936 btrfs_item_key_to_cpu(eb, &found_key, slot);
3938 if (found_key.objectid != key.objectid ||
3939 found_key.type != key.type) {
3944 ret = iterate_dir_item(root, path, &found_key,
3945 __process_new_xattr, sctx);
3949 btrfs_release_path(path);
3950 key.offset = found_key.offset + 1;
3954 btrfs_free_path(path);
3958 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3960 struct btrfs_root *root = sctx->send_root;
3961 struct btrfs_fs_info *fs_info = root->fs_info;
3962 struct inode *inode;
3965 struct btrfs_key key;
3966 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3968 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3971 key.objectid = sctx->cur_ino;
3972 key.type = BTRFS_INODE_ITEM_KEY;
3975 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3977 return PTR_ERR(inode);
3979 if (offset + len > i_size_read(inode)) {
3980 if (offset > i_size_read(inode))
3983 len = offset - i_size_read(inode);
3988 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3989 while (index <= last_index) {
3990 unsigned cur_len = min_t(unsigned, len,
3991 PAGE_CACHE_SIZE - pg_offset);
3992 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3998 if (!PageUptodate(page)) {
3999 btrfs_readpage(NULL, page);
4001 if (!PageUptodate(page)) {
4003 page_cache_release(page);
4010 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4013 page_cache_release(page);
4025 * Read some bytes from the current inode/file and send a write command to
4028 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4032 ssize_t num_read = 0;
4034 p = fs_path_alloc();
4038 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4040 num_read = fill_read_buf(sctx, offset, len);
4041 if (num_read <= 0) {
4047 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4051 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4055 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4056 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4057 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4059 ret = send_cmd(sctx);
4070 * Send a clone command to user space.
4072 static int send_clone(struct send_ctx *sctx,
4073 u64 offset, u32 len,
4074 struct clone_root *clone_root)
4080 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4081 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4082 clone_root->root->objectid, clone_root->ino,
4083 clone_root->offset);
4085 p = fs_path_alloc();
4089 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4093 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4097 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4098 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4099 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4101 if (clone_root->root == sctx->send_root) {
4102 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4103 &gen, NULL, NULL, NULL, NULL);
4106 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4108 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4113 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4114 clone_root->root->root_item.uuid);
4115 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4116 le64_to_cpu(clone_root->root->root_item.ctransid));
4117 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4118 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4119 clone_root->offset);
4121 ret = send_cmd(sctx);
4130 * Send an update extent command to user space.
4132 static int send_update_extent(struct send_ctx *sctx,
4133 u64 offset, u32 len)
4138 p = fs_path_alloc();
4142 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4146 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4150 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4151 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4152 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4154 ret = send_cmd(sctx);
4162 static int send_hole(struct send_ctx *sctx, u64 end)
4164 struct fs_path *p = NULL;
4165 u64 offset = sctx->cur_inode_last_extent;
4169 p = fs_path_alloc();
4172 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4173 while (offset < end) {
4174 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4176 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4179 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4182 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4183 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4184 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4185 ret = send_cmd(sctx);
4195 static int send_write_or_clone(struct send_ctx *sctx,
4196 struct btrfs_path *path,
4197 struct btrfs_key *key,
4198 struct clone_root *clone_root)
4201 struct btrfs_file_extent_item *ei;
4202 u64 offset = key->offset;
4207 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4209 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4210 struct btrfs_file_extent_item);
4211 type = btrfs_file_extent_type(path->nodes[0], ei);
4212 if (type == BTRFS_FILE_EXTENT_INLINE) {
4213 len = btrfs_file_extent_inline_len(path->nodes[0],
4214 path->slots[0], ei);
4216 * it is possible the inline item won't cover the whole page,
4217 * but there may be items after this page. Make
4218 * sure to send the whole thing
4220 len = PAGE_CACHE_ALIGN(len);
4222 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4225 if (offset + len > sctx->cur_inode_size)
4226 len = sctx->cur_inode_size - offset;
4232 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4233 ret = send_clone(sctx, offset, len, clone_root);
4234 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4235 ret = send_update_extent(sctx, offset, len);
4239 if (l > BTRFS_SEND_READ_SIZE)
4240 l = BTRFS_SEND_READ_SIZE;
4241 ret = send_write(sctx, pos + offset, l);
4254 static int is_extent_unchanged(struct send_ctx *sctx,
4255 struct btrfs_path *left_path,
4256 struct btrfs_key *ekey)
4259 struct btrfs_key key;
4260 struct btrfs_path *path = NULL;
4261 struct extent_buffer *eb;
4263 struct btrfs_key found_key;
4264 struct btrfs_file_extent_item *ei;
4269 u64 left_offset_fixed;
4277 path = alloc_path_for_send();
4281 eb = left_path->nodes[0];
4282 slot = left_path->slots[0];
4283 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4284 left_type = btrfs_file_extent_type(eb, ei);
4286 if (left_type != BTRFS_FILE_EXTENT_REG) {
4290 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4291 left_len = btrfs_file_extent_num_bytes(eb, ei);
4292 left_offset = btrfs_file_extent_offset(eb, ei);
4293 left_gen = btrfs_file_extent_generation(eb, ei);
4296 * Following comments will refer to these graphics. L is the left
4297 * extents which we are checking at the moment. 1-8 are the right
4298 * extents that we iterate.
4301 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4304 * |--1--|-2b-|...(same as above)
4306 * Alternative situation. Happens on files where extents got split.
4308 * |-----------7-----------|-6-|
4310 * Alternative situation. Happens on files which got larger.
4313 * Nothing follows after 8.
4316 key.objectid = ekey->objectid;
4317 key.type = BTRFS_EXTENT_DATA_KEY;
4318 key.offset = ekey->offset;
4319 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4328 * Handle special case where the right side has no extents at all.
4330 eb = path->nodes[0];
4331 slot = path->slots[0];
4332 btrfs_item_key_to_cpu(eb, &found_key, slot);
4333 if (found_key.objectid != key.objectid ||
4334 found_key.type != key.type) {
4335 /* If we're a hole then just pretend nothing changed */
4336 ret = (left_disknr) ? 0 : 1;
4341 * We're now on 2a, 2b or 7.
4344 while (key.offset < ekey->offset + left_len) {
4345 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4346 right_type = btrfs_file_extent_type(eb, ei);
4347 if (right_type != BTRFS_FILE_EXTENT_REG) {
4352 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4353 right_len = btrfs_file_extent_num_bytes(eb, ei);
4354 right_offset = btrfs_file_extent_offset(eb, ei);
4355 right_gen = btrfs_file_extent_generation(eb, ei);
4358 * Are we at extent 8? If yes, we know the extent is changed.
4359 * This may only happen on the first iteration.
4361 if (found_key.offset + right_len <= ekey->offset) {
4362 /* If we're a hole just pretend nothing changed */
4363 ret = (left_disknr) ? 0 : 1;
4367 left_offset_fixed = left_offset;
4368 if (key.offset < ekey->offset) {
4369 /* Fix the right offset for 2a and 7. */
4370 right_offset += ekey->offset - key.offset;
4372 /* Fix the left offset for all behind 2a and 2b */
4373 left_offset_fixed += key.offset - ekey->offset;
4377 * Check if we have the same extent.
4379 if (left_disknr != right_disknr ||
4380 left_offset_fixed != right_offset ||
4381 left_gen != right_gen) {
4387 * Go to the next extent.
4389 ret = btrfs_next_item(sctx->parent_root, path);
4393 eb = path->nodes[0];
4394 slot = path->slots[0];
4395 btrfs_item_key_to_cpu(eb, &found_key, slot);
4397 if (ret || found_key.objectid != key.objectid ||
4398 found_key.type != key.type) {
4399 key.offset += right_len;
4402 if (found_key.offset != key.offset + right_len) {
4410 * We're now behind the left extent (treat as unchanged) or at the end
4411 * of the right side (treat as changed).
4413 if (key.offset >= ekey->offset + left_len)
4420 btrfs_free_path(path);
4424 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4426 struct btrfs_path *path;
4427 struct btrfs_root *root = sctx->send_root;
4428 struct btrfs_file_extent_item *fi;
4429 struct btrfs_key key;
4434 path = alloc_path_for_send();
4438 sctx->cur_inode_last_extent = 0;
4440 key.objectid = sctx->cur_ino;
4441 key.type = BTRFS_EXTENT_DATA_KEY;
4442 key.offset = offset;
4443 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4447 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4448 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4451 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4452 struct btrfs_file_extent_item);
4453 type = btrfs_file_extent_type(path->nodes[0], fi);
4454 if (type == BTRFS_FILE_EXTENT_INLINE) {
4455 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4456 path->slots[0], fi);
4457 extent_end = ALIGN(key.offset + size,
4458 sctx->send_root->sectorsize);
4460 extent_end = key.offset +
4461 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4463 sctx->cur_inode_last_extent = extent_end;
4465 btrfs_free_path(path);
4469 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4470 struct btrfs_key *key)
4472 struct btrfs_file_extent_item *fi;
4477 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4480 if (sctx->cur_inode_last_extent == (u64)-1) {
4481 ret = get_last_extent(sctx, key->offset - 1);
4486 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4487 struct btrfs_file_extent_item);
4488 type = btrfs_file_extent_type(path->nodes[0], fi);
4489 if (type == BTRFS_FILE_EXTENT_INLINE) {
4490 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4491 path->slots[0], fi);
4492 extent_end = ALIGN(key->offset + size,
4493 sctx->send_root->sectorsize);
4495 extent_end = key->offset +
4496 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4499 if (path->slots[0] == 0 &&
4500 sctx->cur_inode_last_extent < key->offset) {
4502 * We might have skipped entire leafs that contained only
4503 * file extent items for our current inode. These leafs have
4504 * a generation number smaller (older) than the one in the
4505 * current leaf and the leaf our last extent came from, and
4506 * are located between these 2 leafs.
4508 ret = get_last_extent(sctx, key->offset - 1);
4513 if (sctx->cur_inode_last_extent < key->offset)
4514 ret = send_hole(sctx, key->offset);
4515 sctx->cur_inode_last_extent = extent_end;
4519 static int process_extent(struct send_ctx *sctx,
4520 struct btrfs_path *path,
4521 struct btrfs_key *key)
4523 struct clone_root *found_clone = NULL;
4526 if (S_ISLNK(sctx->cur_inode_mode))
4529 if (sctx->parent_root && !sctx->cur_inode_new) {
4530 ret = is_extent_unchanged(sctx, path, key);
4538 struct btrfs_file_extent_item *ei;
4541 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4542 struct btrfs_file_extent_item);
4543 type = btrfs_file_extent_type(path->nodes[0], ei);
4544 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4545 type == BTRFS_FILE_EXTENT_REG) {
4547 * The send spec does not have a prealloc command yet,
4548 * so just leave a hole for prealloc'ed extents until
4549 * we have enough commands queued up to justify rev'ing
4552 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4557 /* Have a hole, just skip it. */
4558 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4565 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4566 sctx->cur_inode_size, &found_clone);
4567 if (ret != -ENOENT && ret < 0)
4570 ret = send_write_or_clone(sctx, path, key, found_clone);
4574 ret = maybe_send_hole(sctx, path, key);
4579 static int process_all_extents(struct send_ctx *sctx)
4582 struct btrfs_root *root;
4583 struct btrfs_path *path;
4584 struct btrfs_key key;
4585 struct btrfs_key found_key;
4586 struct extent_buffer *eb;
4589 root = sctx->send_root;
4590 path = alloc_path_for_send();
4594 key.objectid = sctx->cmp_key->objectid;
4595 key.type = BTRFS_EXTENT_DATA_KEY;
4597 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4602 eb = path->nodes[0];
4603 slot = path->slots[0];
4605 if (slot >= btrfs_header_nritems(eb)) {
4606 ret = btrfs_next_leaf(root, path);
4609 } else if (ret > 0) {
4616 btrfs_item_key_to_cpu(eb, &found_key, slot);
4618 if (found_key.objectid != key.objectid ||
4619 found_key.type != key.type) {
4624 ret = process_extent(sctx, path, &found_key);
4632 btrfs_free_path(path);
4636 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4638 int *refs_processed)
4642 if (sctx->cur_ino == 0)
4644 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4645 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4647 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4650 ret = process_recorded_refs(sctx, pending_move);
4654 *refs_processed = 1;
4659 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4670 int pending_move = 0;
4671 int refs_processed = 0;
4673 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4679 * We have processed the refs and thus need to advance send_progress.
4680 * Now, calls to get_cur_xxx will take the updated refs of the current
4681 * inode into account.
4683 * On the other hand, if our current inode is a directory and couldn't
4684 * be moved/renamed because its parent was renamed/moved too and it has
4685 * a higher inode number, we can only move/rename our current inode
4686 * after we moved/renamed its parent. Therefore in this case operate on
4687 * the old path (pre move/rename) of our current inode, and the
4688 * move/rename will be performed later.
4690 if (refs_processed && !pending_move)
4691 sctx->send_progress = sctx->cur_ino + 1;
4693 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4695 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4698 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4699 &left_mode, &left_uid, &left_gid, NULL);
4703 if (!sctx->parent_root || sctx->cur_inode_new) {
4705 if (!S_ISLNK(sctx->cur_inode_mode))
4708 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4709 NULL, NULL, &right_mode, &right_uid,
4714 if (left_uid != right_uid || left_gid != right_gid)
4716 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4720 if (S_ISREG(sctx->cur_inode_mode)) {
4721 if (need_send_hole(sctx)) {
4722 if (sctx->cur_inode_last_extent == (u64)-1) {
4723 ret = get_last_extent(sctx, (u64)-1);
4727 if (sctx->cur_inode_last_extent <
4728 sctx->cur_inode_size) {
4729 ret = send_hole(sctx, sctx->cur_inode_size);
4734 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4735 sctx->cur_inode_size);
4741 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4742 left_uid, left_gid);
4747 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4754 * If other directory inodes depended on our current directory
4755 * inode's move/rename, now do their move/rename operations.
4757 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4758 ret = apply_children_dir_moves(sctx);
4764 * Need to send that every time, no matter if it actually
4765 * changed between the two trees as we have done changes to
4768 sctx->send_progress = sctx->cur_ino + 1;
4769 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4777 static int changed_inode(struct send_ctx *sctx,
4778 enum btrfs_compare_tree_result result)
4781 struct btrfs_key *key = sctx->cmp_key;
4782 struct btrfs_inode_item *left_ii = NULL;
4783 struct btrfs_inode_item *right_ii = NULL;
4787 sctx->cur_ino = key->objectid;
4788 sctx->cur_inode_new_gen = 0;
4789 sctx->cur_inode_last_extent = (u64)-1;
4792 * Set send_progress to current inode. This will tell all get_cur_xxx
4793 * functions that the current inode's refs are not updated yet. Later,
4794 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4796 sctx->send_progress = sctx->cur_ino;
4798 if (result == BTRFS_COMPARE_TREE_NEW ||
4799 result == BTRFS_COMPARE_TREE_CHANGED) {
4800 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4801 sctx->left_path->slots[0],
4802 struct btrfs_inode_item);
4803 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4806 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4807 sctx->right_path->slots[0],
4808 struct btrfs_inode_item);
4809 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4812 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4813 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4814 sctx->right_path->slots[0],
4815 struct btrfs_inode_item);
4817 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4821 * The cur_ino = root dir case is special here. We can't treat
4822 * the inode as deleted+reused because it would generate a
4823 * stream that tries to delete/mkdir the root dir.
4825 if (left_gen != right_gen &&
4826 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4827 sctx->cur_inode_new_gen = 1;
4830 if (result == BTRFS_COMPARE_TREE_NEW) {
4831 sctx->cur_inode_gen = left_gen;
4832 sctx->cur_inode_new = 1;
4833 sctx->cur_inode_deleted = 0;
4834 sctx->cur_inode_size = btrfs_inode_size(
4835 sctx->left_path->nodes[0], left_ii);
4836 sctx->cur_inode_mode = btrfs_inode_mode(
4837 sctx->left_path->nodes[0], left_ii);
4838 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4839 ret = send_create_inode_if_needed(sctx);
4840 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4841 sctx->cur_inode_gen = right_gen;
4842 sctx->cur_inode_new = 0;
4843 sctx->cur_inode_deleted = 1;
4844 sctx->cur_inode_size = btrfs_inode_size(
4845 sctx->right_path->nodes[0], right_ii);
4846 sctx->cur_inode_mode = btrfs_inode_mode(
4847 sctx->right_path->nodes[0], right_ii);
4848 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4850 * We need to do some special handling in case the inode was
4851 * reported as changed with a changed generation number. This
4852 * means that the original inode was deleted and new inode
4853 * reused the same inum. So we have to treat the old inode as
4854 * deleted and the new one as new.
4856 if (sctx->cur_inode_new_gen) {
4858 * First, process the inode as if it was deleted.
4860 sctx->cur_inode_gen = right_gen;
4861 sctx->cur_inode_new = 0;
4862 sctx->cur_inode_deleted = 1;
4863 sctx->cur_inode_size = btrfs_inode_size(
4864 sctx->right_path->nodes[0], right_ii);
4865 sctx->cur_inode_mode = btrfs_inode_mode(
4866 sctx->right_path->nodes[0], right_ii);
4867 ret = process_all_refs(sctx,
4868 BTRFS_COMPARE_TREE_DELETED);
4873 * Now process the inode as if it was new.
4875 sctx->cur_inode_gen = left_gen;
4876 sctx->cur_inode_new = 1;
4877 sctx->cur_inode_deleted = 0;
4878 sctx->cur_inode_size = btrfs_inode_size(
4879 sctx->left_path->nodes[0], left_ii);
4880 sctx->cur_inode_mode = btrfs_inode_mode(
4881 sctx->left_path->nodes[0], left_ii);
4882 ret = send_create_inode_if_needed(sctx);
4886 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4890 * Advance send_progress now as we did not get into
4891 * process_recorded_refs_if_needed in the new_gen case.
4893 sctx->send_progress = sctx->cur_ino + 1;
4896 * Now process all extents and xattrs of the inode as if
4897 * they were all new.
4899 ret = process_all_extents(sctx);
4902 ret = process_all_new_xattrs(sctx);
4906 sctx->cur_inode_gen = left_gen;
4907 sctx->cur_inode_new = 0;
4908 sctx->cur_inode_new_gen = 0;
4909 sctx->cur_inode_deleted = 0;
4910 sctx->cur_inode_size = btrfs_inode_size(
4911 sctx->left_path->nodes[0], left_ii);
4912 sctx->cur_inode_mode = btrfs_inode_mode(
4913 sctx->left_path->nodes[0], left_ii);
4922 * We have to process new refs before deleted refs, but compare_trees gives us
4923 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4924 * first and later process them in process_recorded_refs.
4925 * For the cur_inode_new_gen case, we skip recording completely because
4926 * changed_inode did already initiate processing of refs. The reason for this is
4927 * that in this case, compare_tree actually compares the refs of 2 different
4928 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4929 * refs of the right tree as deleted and all refs of the left tree as new.
4931 static int changed_ref(struct send_ctx *sctx,
4932 enum btrfs_compare_tree_result result)
4936 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4938 if (!sctx->cur_inode_new_gen &&
4939 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4940 if (result == BTRFS_COMPARE_TREE_NEW)
4941 ret = record_new_ref(sctx);
4942 else if (result == BTRFS_COMPARE_TREE_DELETED)
4943 ret = record_deleted_ref(sctx);
4944 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4945 ret = record_changed_ref(sctx);
4952 * Process new/deleted/changed xattrs. We skip processing in the
4953 * cur_inode_new_gen case because changed_inode did already initiate processing
4954 * of xattrs. The reason is the same as in changed_ref
4956 static int changed_xattr(struct send_ctx *sctx,
4957 enum btrfs_compare_tree_result result)
4961 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4963 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4964 if (result == BTRFS_COMPARE_TREE_NEW)
4965 ret = process_new_xattr(sctx);
4966 else if (result == BTRFS_COMPARE_TREE_DELETED)
4967 ret = process_deleted_xattr(sctx);
4968 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4969 ret = process_changed_xattr(sctx);
4976 * Process new/deleted/changed extents. We skip processing in the
4977 * cur_inode_new_gen case because changed_inode did already initiate processing
4978 * of extents. The reason is the same as in changed_ref
4980 static int changed_extent(struct send_ctx *sctx,
4981 enum btrfs_compare_tree_result result)
4985 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4987 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4988 if (result != BTRFS_COMPARE_TREE_DELETED)
4989 ret = process_extent(sctx, sctx->left_path,
4996 static int dir_changed(struct send_ctx *sctx, u64 dir)
4998 u64 orig_gen, new_gen;
5001 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5006 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5011 return (orig_gen != new_gen) ? 1 : 0;
5014 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5015 struct btrfs_key *key)
5017 struct btrfs_inode_extref *extref;
5018 struct extent_buffer *leaf;
5019 u64 dirid = 0, last_dirid = 0;
5026 /* Easy case, just check this one dirid */
5027 if (key->type == BTRFS_INODE_REF_KEY) {
5028 dirid = key->offset;
5030 ret = dir_changed(sctx, dirid);
5034 leaf = path->nodes[0];
5035 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5036 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5037 while (cur_offset < item_size) {
5038 extref = (struct btrfs_inode_extref *)(ptr +
5040 dirid = btrfs_inode_extref_parent(leaf, extref);
5041 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5042 cur_offset += ref_name_len + sizeof(*extref);
5043 if (dirid == last_dirid)
5045 ret = dir_changed(sctx, dirid);
5055 * Updates compare related fields in sctx and simply forwards to the actual
5056 * changed_xxx functions.
5058 static int changed_cb(struct btrfs_root *left_root,
5059 struct btrfs_root *right_root,
5060 struct btrfs_path *left_path,
5061 struct btrfs_path *right_path,
5062 struct btrfs_key *key,
5063 enum btrfs_compare_tree_result result,
5067 struct send_ctx *sctx = ctx;
5069 if (result == BTRFS_COMPARE_TREE_SAME) {
5070 if (key->type == BTRFS_INODE_REF_KEY ||
5071 key->type == BTRFS_INODE_EXTREF_KEY) {
5072 ret = compare_refs(sctx, left_path, key);
5077 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5078 return maybe_send_hole(sctx, left_path, key);
5082 result = BTRFS_COMPARE_TREE_CHANGED;
5086 sctx->left_path = left_path;
5087 sctx->right_path = right_path;
5088 sctx->cmp_key = key;
5090 ret = finish_inode_if_needed(sctx, 0);
5094 /* Ignore non-FS objects */
5095 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5096 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5099 if (key->type == BTRFS_INODE_ITEM_KEY)
5100 ret = changed_inode(sctx, result);
5101 else if (key->type == BTRFS_INODE_REF_KEY ||
5102 key->type == BTRFS_INODE_EXTREF_KEY)
5103 ret = changed_ref(sctx, result);
5104 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5105 ret = changed_xattr(sctx, result);
5106 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5107 ret = changed_extent(sctx, result);
5113 static int full_send_tree(struct send_ctx *sctx)
5116 struct btrfs_root *send_root = sctx->send_root;
5117 struct btrfs_key key;
5118 struct btrfs_key found_key;
5119 struct btrfs_path *path;
5120 struct extent_buffer *eb;
5125 path = alloc_path_for_send();
5129 spin_lock(&send_root->root_item_lock);
5130 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5131 spin_unlock(&send_root->root_item_lock);
5133 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5134 key.type = BTRFS_INODE_ITEM_KEY;
5138 * Make sure the tree has not changed after re-joining. We detect this
5139 * by comparing start_ctransid and ctransid. They should always match.
5141 spin_lock(&send_root->root_item_lock);
5142 ctransid = btrfs_root_ctransid(&send_root->root_item);
5143 spin_unlock(&send_root->root_item_lock);
5145 if (ctransid != start_ctransid) {
5146 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
5147 "send was modified in between. This is "
5148 "probably a bug.\n");
5153 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5160 eb = path->nodes[0];
5161 slot = path->slots[0];
5162 btrfs_item_key_to_cpu(eb, &found_key, slot);
5164 ret = changed_cb(send_root, NULL, path, NULL,
5165 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5169 key.objectid = found_key.objectid;
5170 key.type = found_key.type;
5171 key.offset = found_key.offset + 1;
5173 ret = btrfs_next_item(send_root, path);
5183 ret = finish_inode_if_needed(sctx, 1);
5186 btrfs_free_path(path);
5190 static int send_subvol(struct send_ctx *sctx)
5194 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5195 ret = send_header(sctx);
5200 ret = send_subvol_begin(sctx);
5204 if (sctx->parent_root) {
5205 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5209 ret = finish_inode_if_needed(sctx, 1);
5213 ret = full_send_tree(sctx);
5219 free_recorded_refs(sctx);
5223 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5225 spin_lock(&root->root_item_lock);
5226 root->send_in_progress--;
5228 * Not much left to do, we don't know why it's unbalanced and
5229 * can't blindly reset it to 0.
5231 if (root->send_in_progress < 0)
5232 btrfs_err(root->fs_info,
5233 "send_in_progres unbalanced %d root %llu\n",
5234 root->send_in_progress, root->root_key.objectid);
5235 spin_unlock(&root->root_item_lock);
5238 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5241 struct btrfs_root *send_root;
5242 struct btrfs_root *clone_root;
5243 struct btrfs_fs_info *fs_info;
5244 struct btrfs_ioctl_send_args *arg = NULL;
5245 struct btrfs_key key;
5246 struct send_ctx *sctx = NULL;
5248 u64 *clone_sources_tmp = NULL;
5249 int clone_sources_to_rollback = 0;
5250 int sort_clone_roots = 0;
5253 if (!capable(CAP_SYS_ADMIN))
5256 send_root = BTRFS_I(file_inode(mnt_file))->root;
5257 fs_info = send_root->fs_info;
5260 * The subvolume must remain read-only during send, protect against
5263 spin_lock(&send_root->root_item_lock);
5264 send_root->send_in_progress++;
5265 spin_unlock(&send_root->root_item_lock);
5268 * This is done when we lookup the root, it should already be complete
5269 * by the time we get here.
5271 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5274 * Userspace tools do the checks and warn the user if it's
5277 if (!btrfs_root_readonly(send_root)) {
5282 arg = memdup_user(arg_, sizeof(*arg));
5289 if (!access_ok(VERIFY_READ, arg->clone_sources,
5290 sizeof(*arg->clone_sources) *
5291 arg->clone_sources_count)) {
5296 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5301 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5307 INIT_LIST_HEAD(&sctx->new_refs);
5308 INIT_LIST_HEAD(&sctx->deleted_refs);
5309 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5310 INIT_LIST_HEAD(&sctx->name_cache_list);
5312 sctx->flags = arg->flags;
5314 sctx->send_filp = fget(arg->send_fd);
5315 if (!sctx->send_filp) {
5320 sctx->send_root = send_root;
5321 sctx->clone_roots_cnt = arg->clone_sources_count;
5323 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5324 sctx->send_buf = vmalloc(sctx->send_max_size);
5325 if (!sctx->send_buf) {
5330 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5331 if (!sctx->read_buf) {
5336 sctx->pending_dir_moves = RB_ROOT;
5337 sctx->waiting_dir_moves = RB_ROOT;
5339 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5340 (arg->clone_sources_count + 1));
5341 if (!sctx->clone_roots) {
5346 if (arg->clone_sources_count) {
5347 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5348 sizeof(*arg->clone_sources));
5349 if (!clone_sources_tmp) {
5354 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5355 arg->clone_sources_count *
5356 sizeof(*arg->clone_sources));
5362 for (i = 0; i < arg->clone_sources_count; i++) {
5363 key.objectid = clone_sources_tmp[i];
5364 key.type = BTRFS_ROOT_ITEM_KEY;
5365 key.offset = (u64)-1;
5367 index = srcu_read_lock(&fs_info->subvol_srcu);
5369 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5370 if (IS_ERR(clone_root)) {
5371 srcu_read_unlock(&fs_info->subvol_srcu, index);
5372 ret = PTR_ERR(clone_root);
5375 clone_sources_to_rollback = i + 1;
5376 spin_lock(&clone_root->root_item_lock);
5377 clone_root->send_in_progress++;
5378 if (!btrfs_root_readonly(clone_root)) {
5379 spin_unlock(&clone_root->root_item_lock);
5380 srcu_read_unlock(&fs_info->subvol_srcu, index);
5384 spin_unlock(&clone_root->root_item_lock);
5385 srcu_read_unlock(&fs_info->subvol_srcu, index);
5387 sctx->clone_roots[i].root = clone_root;
5389 vfree(clone_sources_tmp);
5390 clone_sources_tmp = NULL;
5393 if (arg->parent_root) {
5394 key.objectid = arg->parent_root;
5395 key.type = BTRFS_ROOT_ITEM_KEY;
5396 key.offset = (u64)-1;
5398 index = srcu_read_lock(&fs_info->subvol_srcu);
5400 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5401 if (IS_ERR(sctx->parent_root)) {
5402 srcu_read_unlock(&fs_info->subvol_srcu, index);
5403 ret = PTR_ERR(sctx->parent_root);
5407 spin_lock(&sctx->parent_root->root_item_lock);
5408 sctx->parent_root->send_in_progress++;
5409 if (!btrfs_root_readonly(sctx->parent_root)) {
5410 spin_unlock(&sctx->parent_root->root_item_lock);
5411 srcu_read_unlock(&fs_info->subvol_srcu, index);
5415 spin_unlock(&sctx->parent_root->root_item_lock);
5417 srcu_read_unlock(&fs_info->subvol_srcu, index);
5421 * Clones from send_root are allowed, but only if the clone source
5422 * is behind the current send position. This is checked while searching
5423 * for possible clone sources.
5425 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5427 /* We do a bsearch later */
5428 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5429 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5431 sort_clone_roots = 1;
5433 ret = send_subvol(sctx);
5437 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5438 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5441 ret = send_cmd(sctx);
5447 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5448 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5450 struct pending_dir_move *pm;
5452 n = rb_first(&sctx->pending_dir_moves);
5453 pm = rb_entry(n, struct pending_dir_move, node);
5454 while (!list_empty(&pm->list)) {
5455 struct pending_dir_move *pm2;
5457 pm2 = list_first_entry(&pm->list,
5458 struct pending_dir_move, list);
5459 free_pending_move(sctx, pm2);
5461 free_pending_move(sctx, pm);
5464 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5465 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5467 struct waiting_dir_move *dm;
5469 n = rb_first(&sctx->waiting_dir_moves);
5470 dm = rb_entry(n, struct waiting_dir_move, node);
5471 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5475 if (sort_clone_roots) {
5476 for (i = 0; i < sctx->clone_roots_cnt; i++)
5477 btrfs_root_dec_send_in_progress(
5478 sctx->clone_roots[i].root);
5480 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5481 btrfs_root_dec_send_in_progress(
5482 sctx->clone_roots[i].root);
5484 btrfs_root_dec_send_in_progress(send_root);
5486 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5487 btrfs_root_dec_send_in_progress(sctx->parent_root);
5490 vfree(clone_sources_tmp);
5493 if (sctx->send_filp)
5494 fput(sctx->send_filp);
5496 vfree(sctx->clone_roots);
5497 vfree(sctx->send_buf);
5498 vfree(sctx->read_buf);
5500 name_cache_free(sctx);