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.
56 unsigned short buf_len:15;
57 unsigned short reversed:1;
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
68 #define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
72 /* reused for each extent */
74 struct btrfs_root *root;
81 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
85 struct file *send_filp;
91 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
92 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
94 struct btrfs_root *send_root;
95 struct btrfs_root *parent_root;
96 struct clone_root *clone_roots;
99 /* current state of the compare_tree call */
100 struct btrfs_path *left_path;
101 struct btrfs_path *right_path;
102 struct btrfs_key *cmp_key;
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
111 int cur_inode_new_gen;
112 int cur_inode_deleted;
115 u64 cur_inode_last_extent;
119 struct list_head new_refs;
120 struct list_head deleted_refs;
122 struct radix_tree_root name_cache;
123 struct list_head name_cache_list;
129 * We process inodes by their increasing order, so if before an
130 * incremental send we reverse the parent/child relationship of
131 * directories such that a directory with a lower inode number was
132 * the parent of a directory with a higher inode number, and the one
133 * becoming the new parent got renamed too, we can't rename/move the
134 * directory with lower inode number when we finish processing it - we
135 * must process the directory with higher inode number first, then
136 * rename/move it and then rename/move the directory with lower inode
137 * number. Example follows.
139 * Tree state when the first send was performed:
151 * Tree state when the second (incremental) send is performed:
160 * The sequence of steps that lead to the second state was:
162 * mv /a/b/c/d /a/b/c2/d2
163 * mv /a/b/c /a/b/c2/d2/cc
165 * "c" has lower inode number, but we can't move it (2nd mv operation)
166 * before we move "d", which has higher inode number.
168 * So we just memorize which move/rename operations must be performed
169 * later when their respective parent is processed and moved/renamed.
172 /* Indexed by parent directory inode number. */
173 struct rb_root pending_dir_moves;
176 * Reverse index, indexed by the inode number of a directory that
177 * is waiting for the move/rename of its immediate parent before its
178 * own move/rename can be performed.
180 struct rb_root waiting_dir_moves;
183 struct pending_dir_move {
185 struct list_head list;
189 struct list_head update_refs;
192 struct waiting_dir_move {
197 struct name_cache_entry {
198 struct list_head list;
200 * radix_tree has only 32bit entries but we need to handle 64bit inums.
201 * We use the lower 32bit of the 64bit inum to store it in the tree. If
202 * more then one inum would fall into the same entry, we use radix_list
203 * to store the additional entries. radix_list is also used to store
204 * entries where two entries have the same inum but different
207 struct list_head radix_list;
213 int need_later_update;
218 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
220 static int need_send_hole(struct send_ctx *sctx)
222 return (sctx->parent_root && !sctx->cur_inode_new &&
223 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
224 S_ISREG(sctx->cur_inode_mode));
227 static void fs_path_reset(struct fs_path *p)
230 p->start = p->buf + p->buf_len - 1;
240 static struct fs_path *fs_path_alloc(void)
244 p = kmalloc(sizeof(*p), GFP_NOFS);
248 p->buf = p->inline_buf;
249 p->buf_len = FS_PATH_INLINE_SIZE;
254 static struct fs_path *fs_path_alloc_reversed(void)
266 static void fs_path_free(struct fs_path *p)
270 if (p->buf != p->inline_buf)
275 static int fs_path_len(struct fs_path *p)
277 return p->end - p->start;
280 static int fs_path_ensure_buf(struct fs_path *p, int len)
288 if (p->buf_len >= len)
292 * First time the inline_buf does not suffice
294 if (p->buf == p->inline_buf) {
295 p->buf = kmalloc(len, GFP_NOFS);
299 * The real size of the buffer is bigger, this will let the
300 * fast path happen most of the time
302 p->buf_len = ksize(p->buf);
306 tmp = krealloc(p->buf, len, GFP_NOFS);
310 p->buf_len = ksize(p->buf);
313 path_len = p->end - p->start;
314 old_buf_len = p->buf_len;
317 tmp_buf = p->buf + old_buf_len - path_len - 1;
318 p->end = p->buf + p->buf_len - 1;
319 p->start = p->end - path_len;
320 memmove(p->start, tmp_buf, path_len + 1);
323 p->end = p->start + path_len;
328 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
334 new_len = p->end - p->start + name_len;
335 if (p->start != p->end)
337 ret = fs_path_ensure_buf(p, new_len);
342 if (p->start != p->end)
344 p->start -= name_len;
345 *prepared = p->start;
347 if (p->start != p->end)
358 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
363 ret = fs_path_prepare_for_add(p, name_len, &prepared);
366 memcpy(prepared, name, name_len);
372 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
377 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
380 memcpy(prepared, p2->start, p2->end - p2->start);
386 static int fs_path_add_from_extent_buffer(struct fs_path *p,
387 struct extent_buffer *eb,
388 unsigned long off, int len)
393 ret = fs_path_prepare_for_add(p, len, &prepared);
397 read_extent_buffer(eb, prepared, off, len);
403 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
407 p->reversed = from->reversed;
410 ret = fs_path_add_path(p, from);
416 static void fs_path_unreverse(struct fs_path *p)
425 len = p->end - p->start;
427 p->end = p->start + len;
428 memmove(p->start, tmp, len + 1);
432 static struct btrfs_path *alloc_path_for_send(void)
434 struct btrfs_path *path;
436 path = btrfs_alloc_path();
439 path->search_commit_root = 1;
440 path->skip_locking = 1;
444 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
454 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
455 /* TODO handle that correctly */
456 /*if (ret == -ERESTARTSYS) {
475 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
477 struct btrfs_tlv_header *hdr;
478 int total_len = sizeof(*hdr) + len;
479 int left = sctx->send_max_size - sctx->send_size;
481 if (unlikely(left < total_len))
484 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
485 hdr->tlv_type = cpu_to_le16(attr);
486 hdr->tlv_len = cpu_to_le16(len);
487 memcpy(hdr + 1, data, len);
488 sctx->send_size += total_len;
493 #define TLV_PUT_DEFINE_INT(bits) \
494 static int tlv_put_u##bits(struct send_ctx *sctx, \
495 u##bits attr, u##bits value) \
497 __le##bits __tmp = cpu_to_le##bits(value); \
498 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
501 TLV_PUT_DEFINE_INT(64)
503 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
504 const char *str, int len)
508 return tlv_put(sctx, attr, str, len);
511 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
514 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
517 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
518 struct extent_buffer *eb,
519 struct btrfs_timespec *ts)
521 struct btrfs_timespec bts;
522 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
523 return tlv_put(sctx, attr, &bts, sizeof(bts));
527 #define TLV_PUT(sctx, attrtype, attrlen, data) \
529 ret = tlv_put(sctx, attrtype, attrlen, data); \
531 goto tlv_put_failure; \
534 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
536 ret = tlv_put_u##bits(sctx, attrtype, value); \
538 goto tlv_put_failure; \
541 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
542 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
543 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
544 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
545 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
547 ret = tlv_put_string(sctx, attrtype, str, len); \
549 goto tlv_put_failure; \
551 #define TLV_PUT_PATH(sctx, attrtype, p) \
553 ret = tlv_put_string(sctx, attrtype, p->start, \
554 p->end - p->start); \
556 goto tlv_put_failure; \
558 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
560 ret = tlv_put_uuid(sctx, attrtype, uuid); \
562 goto tlv_put_failure; \
564 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
566 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
568 goto tlv_put_failure; \
571 static int send_header(struct send_ctx *sctx)
573 struct btrfs_stream_header hdr;
575 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
576 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
578 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
583 * For each command/item we want to send to userspace, we call this function.
585 static int begin_cmd(struct send_ctx *sctx, int cmd)
587 struct btrfs_cmd_header *hdr;
589 if (WARN_ON(!sctx->send_buf))
592 BUG_ON(sctx->send_size);
594 sctx->send_size += sizeof(*hdr);
595 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
596 hdr->cmd = cpu_to_le16(cmd);
601 static int send_cmd(struct send_ctx *sctx)
604 struct btrfs_cmd_header *hdr;
607 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
608 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
611 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
612 hdr->crc = cpu_to_le32(crc);
614 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
617 sctx->total_send_size += sctx->send_size;
618 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
625 * Sends a move instruction to user space
627 static int send_rename(struct send_ctx *sctx,
628 struct fs_path *from, struct fs_path *to)
632 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
634 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
638 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
639 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
641 ret = send_cmd(sctx);
649 * Sends a link instruction to user space
651 static int send_link(struct send_ctx *sctx,
652 struct fs_path *path, struct fs_path *lnk)
656 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
658 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
662 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
663 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
665 ret = send_cmd(sctx);
673 * Sends an unlink instruction to user space
675 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
679 verbose_printk("btrfs: send_unlink %s\n", path->start);
681 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
685 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
687 ret = send_cmd(sctx);
695 * Sends a rmdir instruction to user space
697 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
701 verbose_printk("btrfs: send_rmdir %s\n", path->start);
703 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
707 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
709 ret = send_cmd(sctx);
717 * Helper function to retrieve some fields from an inode item.
719 static int get_inode_info(struct btrfs_root *root,
720 u64 ino, u64 *size, u64 *gen,
721 u64 *mode, u64 *uid, u64 *gid,
725 struct btrfs_inode_item *ii;
726 struct btrfs_key key;
727 struct btrfs_path *path;
729 path = alloc_path_for_send();
734 key.type = BTRFS_INODE_ITEM_KEY;
736 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
744 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
745 struct btrfs_inode_item);
747 *size = btrfs_inode_size(path->nodes[0], ii);
749 *gen = btrfs_inode_generation(path->nodes[0], ii);
751 *mode = btrfs_inode_mode(path->nodes[0], ii);
753 *uid = btrfs_inode_uid(path->nodes[0], ii);
755 *gid = btrfs_inode_gid(path->nodes[0], ii);
757 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
760 btrfs_free_path(path);
764 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
769 * Helper function to iterate the entries in ONE btrfs_inode_ref or
770 * btrfs_inode_extref.
771 * The iterate callback may return a non zero value to stop iteration. This can
772 * be a negative value for error codes or 1 to simply stop it.
774 * path must point to the INODE_REF or INODE_EXTREF when called.
776 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
777 struct btrfs_key *found_key, int resolve,
778 iterate_inode_ref_t iterate, void *ctx)
780 struct extent_buffer *eb = path->nodes[0];
781 struct btrfs_item *item;
782 struct btrfs_inode_ref *iref;
783 struct btrfs_inode_extref *extref;
784 struct btrfs_path *tmp_path;
788 int slot = path->slots[0];
795 unsigned long name_off;
796 unsigned long elem_size;
799 p = fs_path_alloc_reversed();
803 tmp_path = alloc_path_for_send();
810 if (found_key->type == BTRFS_INODE_REF_KEY) {
811 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
812 struct btrfs_inode_ref);
813 item = btrfs_item_nr(slot);
814 total = btrfs_item_size(eb, item);
815 elem_size = sizeof(*iref);
817 ptr = btrfs_item_ptr_offset(eb, slot);
818 total = btrfs_item_size_nr(eb, slot);
819 elem_size = sizeof(*extref);
822 while (cur < total) {
825 if (found_key->type == BTRFS_INODE_REF_KEY) {
826 iref = (struct btrfs_inode_ref *)(ptr + cur);
827 name_len = btrfs_inode_ref_name_len(eb, iref);
828 name_off = (unsigned long)(iref + 1);
829 index = btrfs_inode_ref_index(eb, iref);
830 dir = found_key->offset;
832 extref = (struct btrfs_inode_extref *)(ptr + cur);
833 name_len = btrfs_inode_extref_name_len(eb, extref);
834 name_off = (unsigned long)&extref->name;
835 index = btrfs_inode_extref_index(eb, extref);
836 dir = btrfs_inode_extref_parent(eb, extref);
840 start = btrfs_ref_to_path(root, tmp_path, name_len,
844 ret = PTR_ERR(start);
847 if (start < p->buf) {
848 /* overflow , try again with larger buffer */
849 ret = fs_path_ensure_buf(p,
850 p->buf_len + p->buf - start);
853 start = btrfs_ref_to_path(root, tmp_path,
858 ret = PTR_ERR(start);
861 BUG_ON(start < p->buf);
865 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
871 cur += elem_size + name_len;
872 ret = iterate(num, dir, index, p, ctx);
879 btrfs_free_path(tmp_path);
884 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
885 const char *name, int name_len,
886 const char *data, int data_len,
890 * Helper function to iterate the entries in ONE btrfs_dir_item.
891 * The iterate callback may return a non zero value to stop iteration. This can
892 * be a negative value for error codes or 1 to simply stop it.
894 * path must point to the dir item when called.
896 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
897 struct btrfs_key *found_key,
898 iterate_dir_item_t iterate, void *ctx)
901 struct extent_buffer *eb;
902 struct btrfs_item *item;
903 struct btrfs_dir_item *di;
904 struct btrfs_key di_key;
906 const int buf_len = PATH_MAX;
916 buf = kmalloc(buf_len, GFP_NOFS);
923 slot = path->slots[0];
924 item = btrfs_item_nr(slot);
925 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
928 total = btrfs_item_size(eb, item);
931 while (cur < total) {
932 name_len = btrfs_dir_name_len(eb, di);
933 data_len = btrfs_dir_data_len(eb, di);
934 type = btrfs_dir_type(eb, di);
935 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
940 if (name_len + data_len > buf_len) {
945 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
946 name_len + data_len);
948 len = sizeof(*di) + name_len + data_len;
949 di = (struct btrfs_dir_item *)((char *)di + len);
952 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
953 data_len, type, ctx);
969 static int __copy_first_ref(int num, u64 dir, int index,
970 struct fs_path *p, void *ctx)
973 struct fs_path *pt = ctx;
975 ret = fs_path_copy(pt, p);
979 /* we want the first only */
984 * Retrieve the first path of an inode. If an inode has more then one
985 * ref/hardlink, this is ignored.
987 static int get_inode_path(struct btrfs_root *root,
988 u64 ino, struct fs_path *path)
991 struct btrfs_key key, found_key;
992 struct btrfs_path *p;
994 p = alloc_path_for_send();
1001 key.type = BTRFS_INODE_REF_KEY;
1004 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1011 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1012 if (found_key.objectid != ino ||
1013 (found_key.type != BTRFS_INODE_REF_KEY &&
1014 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1019 ret = iterate_inode_ref(root, p, &found_key, 1,
1020 __copy_first_ref, path);
1030 struct backref_ctx {
1031 struct send_ctx *sctx;
1033 /* number of total found references */
1037 * used for clones found in send_root. clones found behind cur_objectid
1038 * and cur_offset are not considered as allowed clones.
1043 /* may be truncated in case it's the last extent in a file */
1046 /* Just to check for bugs in backref resolving */
1050 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1052 u64 root = (u64)(uintptr_t)key;
1053 struct clone_root *cr = (struct clone_root *)elt;
1055 if (root < cr->root->objectid)
1057 if (root > cr->root->objectid)
1062 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1064 struct clone_root *cr1 = (struct clone_root *)e1;
1065 struct clone_root *cr2 = (struct clone_root *)e2;
1067 if (cr1->root->objectid < cr2->root->objectid)
1069 if (cr1->root->objectid > cr2->root->objectid)
1075 * Called for every backref that is found for the current extent.
1076 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1078 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1080 struct backref_ctx *bctx = ctx_;
1081 struct clone_root *found;
1085 /* First check if the root is in the list of accepted clone sources */
1086 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1087 bctx->sctx->clone_roots_cnt,
1088 sizeof(struct clone_root),
1089 __clone_root_cmp_bsearch);
1093 if (found->root == bctx->sctx->send_root &&
1094 ino == bctx->cur_objectid &&
1095 offset == bctx->cur_offset) {
1096 bctx->found_itself = 1;
1100 * There are inodes that have extents that lie behind its i_size. Don't
1101 * accept clones from these extents.
1103 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1108 if (offset + bctx->extent_len > i_size)
1112 * Make sure we don't consider clones from send_root that are
1113 * behind the current inode/offset.
1115 if (found->root == bctx->sctx->send_root) {
1117 * TODO for the moment we don't accept clones from the inode
1118 * that is currently send. We may change this when
1119 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1122 if (ino >= bctx->cur_objectid)
1125 if (ino > bctx->cur_objectid)
1127 if (offset + bctx->extent_len > bctx->cur_offset)
1133 found->found_refs++;
1134 if (ino < found->ino) {
1136 found->offset = offset;
1137 } else if (found->ino == ino) {
1139 * same extent found more then once in the same file.
1141 if (found->offset > offset + bctx->extent_len)
1142 found->offset = offset;
1149 * Given an inode, offset and extent item, it finds a good clone for a clone
1150 * instruction. Returns -ENOENT when none could be found. The function makes
1151 * sure that the returned clone is usable at the point where sending is at the
1152 * moment. This means, that no clones are accepted which lie behind the current
1155 * path must point to the extent item when called.
1157 static int find_extent_clone(struct send_ctx *sctx,
1158 struct btrfs_path *path,
1159 u64 ino, u64 data_offset,
1161 struct clone_root **found)
1168 u64 extent_item_pos;
1170 struct btrfs_file_extent_item *fi;
1171 struct extent_buffer *eb = path->nodes[0];
1172 struct backref_ctx *backref_ctx = NULL;
1173 struct clone_root *cur_clone_root;
1174 struct btrfs_key found_key;
1175 struct btrfs_path *tmp_path;
1179 tmp_path = alloc_path_for_send();
1183 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1189 if (data_offset >= ino_size) {
1191 * There may be extents that lie behind the file's size.
1192 * I at least had this in combination with snapshotting while
1193 * writing large files.
1199 fi = btrfs_item_ptr(eb, path->slots[0],
1200 struct btrfs_file_extent_item);
1201 extent_type = btrfs_file_extent_type(eb, fi);
1202 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1206 compressed = btrfs_file_extent_compression(eb, fi);
1208 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1209 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1210 if (disk_byte == 0) {
1214 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1216 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1217 &found_key, &flags);
1218 btrfs_release_path(tmp_path);
1222 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1228 * Setup the clone roots.
1230 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1231 cur_clone_root = sctx->clone_roots + i;
1232 cur_clone_root->ino = (u64)-1;
1233 cur_clone_root->offset = 0;
1234 cur_clone_root->found_refs = 0;
1237 backref_ctx->sctx = sctx;
1238 backref_ctx->found = 0;
1239 backref_ctx->cur_objectid = ino;
1240 backref_ctx->cur_offset = data_offset;
1241 backref_ctx->found_itself = 0;
1242 backref_ctx->extent_len = num_bytes;
1245 * The last extent of a file may be too large due to page alignment.
1246 * We need to adjust extent_len in this case so that the checks in
1247 * __iterate_backrefs work.
1249 if (data_offset + num_bytes >= ino_size)
1250 backref_ctx->extent_len = ino_size - data_offset;
1253 * Now collect all backrefs.
1255 if (compressed == BTRFS_COMPRESS_NONE)
1256 extent_item_pos = logical - found_key.objectid;
1258 extent_item_pos = 0;
1259 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1260 found_key.objectid, extent_item_pos, 1,
1261 __iterate_backrefs, backref_ctx);
1266 if (!backref_ctx->found_itself) {
1267 /* found a bug in backref code? */
1269 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1270 "send_root. inode=%llu, offset=%llu, "
1271 "disk_byte=%llu found extent=%llu\n",
1272 ino, data_offset, disk_byte, found_key.objectid);
1276 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1278 "num_bytes=%llu, logical=%llu\n",
1279 data_offset, ino, num_bytes, logical);
1281 if (!backref_ctx->found)
1282 verbose_printk("btrfs: no clones found\n");
1284 cur_clone_root = NULL;
1285 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1286 if (sctx->clone_roots[i].found_refs) {
1287 if (!cur_clone_root)
1288 cur_clone_root = sctx->clone_roots + i;
1289 else if (sctx->clone_roots[i].root == sctx->send_root)
1290 /* prefer clones from send_root over others */
1291 cur_clone_root = sctx->clone_roots + i;
1296 if (cur_clone_root) {
1297 if (compressed != BTRFS_COMPRESS_NONE) {
1299 * Offsets given by iterate_extent_inodes() are relative
1300 * to the start of the extent, we need to add logical
1301 * offset from the file extent item.
1302 * (See why at backref.c:check_extent_in_eb())
1304 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1307 *found = cur_clone_root;
1314 btrfs_free_path(tmp_path);
1319 static int read_symlink(struct btrfs_root *root,
1321 struct fs_path *dest)
1324 struct btrfs_path *path;
1325 struct btrfs_key key;
1326 struct btrfs_file_extent_item *ei;
1332 path = alloc_path_for_send();
1337 key.type = BTRFS_EXTENT_DATA_KEY;
1339 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1344 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1345 struct btrfs_file_extent_item);
1346 type = btrfs_file_extent_type(path->nodes[0], ei);
1347 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1348 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1349 BUG_ON(compression);
1351 off = btrfs_file_extent_inline_start(ei);
1352 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1354 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1357 btrfs_free_path(path);
1362 * Helper function to generate a file name that is unique in the root of
1363 * send_root and parent_root. This is used to generate names for orphan inodes.
1365 static int gen_unique_name(struct send_ctx *sctx,
1367 struct fs_path *dest)
1370 struct btrfs_path *path;
1371 struct btrfs_dir_item *di;
1376 path = alloc_path_for_send();
1381 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1383 ASSERT(len < sizeof(tmp));
1385 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1386 path, BTRFS_FIRST_FREE_OBJECTID,
1387 tmp, strlen(tmp), 0);
1388 btrfs_release_path(path);
1394 /* not unique, try again */
1399 if (!sctx->parent_root) {
1405 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1406 path, BTRFS_FIRST_FREE_OBJECTID,
1407 tmp, strlen(tmp), 0);
1408 btrfs_release_path(path);
1414 /* not unique, try again */
1422 ret = fs_path_add(dest, tmp, strlen(tmp));
1425 btrfs_free_path(path);
1430 inode_state_no_change,
1431 inode_state_will_create,
1432 inode_state_did_create,
1433 inode_state_will_delete,
1434 inode_state_did_delete,
1437 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1445 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1447 if (ret < 0 && ret != -ENOENT)
1451 if (!sctx->parent_root) {
1452 right_ret = -ENOENT;
1454 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1455 NULL, NULL, NULL, NULL);
1456 if (ret < 0 && ret != -ENOENT)
1461 if (!left_ret && !right_ret) {
1462 if (left_gen == gen && right_gen == gen) {
1463 ret = inode_state_no_change;
1464 } else if (left_gen == gen) {
1465 if (ino < sctx->send_progress)
1466 ret = inode_state_did_create;
1468 ret = inode_state_will_create;
1469 } else if (right_gen == gen) {
1470 if (ino < sctx->send_progress)
1471 ret = inode_state_did_delete;
1473 ret = inode_state_will_delete;
1477 } else if (!left_ret) {
1478 if (left_gen == gen) {
1479 if (ino < sctx->send_progress)
1480 ret = inode_state_did_create;
1482 ret = inode_state_will_create;
1486 } else if (!right_ret) {
1487 if (right_gen == gen) {
1488 if (ino < sctx->send_progress)
1489 ret = inode_state_did_delete;
1491 ret = inode_state_will_delete;
1503 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1507 ret = get_cur_inode_state(sctx, ino, gen);
1511 if (ret == inode_state_no_change ||
1512 ret == inode_state_did_create ||
1513 ret == inode_state_will_delete)
1523 * Helper function to lookup a dir item in a dir.
1525 static int lookup_dir_item_inode(struct btrfs_root *root,
1526 u64 dir, const char *name, int name_len,
1531 struct btrfs_dir_item *di;
1532 struct btrfs_key key;
1533 struct btrfs_path *path;
1535 path = alloc_path_for_send();
1539 di = btrfs_lookup_dir_item(NULL, root, path,
1540 dir, name, name_len, 0);
1549 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1550 *found_inode = key.objectid;
1551 *found_type = btrfs_dir_type(path->nodes[0], di);
1554 btrfs_free_path(path);
1559 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1560 * generation of the parent dir and the name of the dir entry.
1562 static int get_first_ref(struct btrfs_root *root, u64 ino,
1563 u64 *dir, u64 *dir_gen, struct fs_path *name)
1566 struct btrfs_key key;
1567 struct btrfs_key found_key;
1568 struct btrfs_path *path;
1572 path = alloc_path_for_send();
1577 key.type = BTRFS_INODE_REF_KEY;
1580 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1584 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1586 if (ret || found_key.objectid != ino ||
1587 (found_key.type != BTRFS_INODE_REF_KEY &&
1588 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1593 if (key.type == BTRFS_INODE_REF_KEY) {
1594 struct btrfs_inode_ref *iref;
1595 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1596 struct btrfs_inode_ref);
1597 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1598 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1599 (unsigned long)(iref + 1),
1601 parent_dir = found_key.offset;
1603 struct btrfs_inode_extref *extref;
1604 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1605 struct btrfs_inode_extref);
1606 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1607 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1608 (unsigned long)&extref->name, len);
1609 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1613 btrfs_release_path(path);
1615 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1623 btrfs_free_path(path);
1627 static int is_first_ref(struct btrfs_root *root,
1629 const char *name, int name_len)
1632 struct fs_path *tmp_name;
1636 tmp_name = fs_path_alloc();
1640 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1644 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1649 ret = !memcmp(tmp_name->start, name, name_len);
1652 fs_path_free(tmp_name);
1657 * Used by process_recorded_refs to determine if a new ref would overwrite an
1658 * already existing ref. In case it detects an overwrite, it returns the
1659 * inode/gen in who_ino/who_gen.
1660 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1661 * to make sure later references to the overwritten inode are possible.
1662 * Orphanizing is however only required for the first ref of an inode.
1663 * process_recorded_refs does an additional is_first_ref check to see if
1664 * orphanizing is really required.
1666 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1667 const char *name, int name_len,
1668 u64 *who_ino, u64 *who_gen)
1672 u64 other_inode = 0;
1675 if (!sctx->parent_root)
1678 ret = is_inode_existent(sctx, dir, dir_gen);
1683 * If we have a parent root we need to verify that the parent dir was
1684 * not delted and then re-created, if it was then we have no overwrite
1685 * and we can just unlink this entry.
1687 if (sctx->parent_root) {
1688 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1690 if (ret < 0 && ret != -ENOENT)
1700 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1701 &other_inode, &other_type);
1702 if (ret < 0 && ret != -ENOENT)
1710 * Check if the overwritten ref was already processed. If yes, the ref
1711 * was already unlinked/moved, so we can safely assume that we will not
1712 * overwrite anything at this point in time.
1714 if (other_inode > sctx->send_progress) {
1715 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1716 who_gen, NULL, NULL, NULL, NULL);
1721 *who_ino = other_inode;
1731 * Checks if the ref was overwritten by an already processed inode. This is
1732 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1733 * thus the orphan name needs be used.
1734 * process_recorded_refs also uses it to avoid unlinking of refs that were
1737 static int did_overwrite_ref(struct send_ctx *sctx,
1738 u64 dir, u64 dir_gen,
1739 u64 ino, u64 ino_gen,
1740 const char *name, int name_len)
1747 if (!sctx->parent_root)
1750 ret = is_inode_existent(sctx, dir, dir_gen);
1754 /* check if the ref was overwritten by another ref */
1755 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1756 &ow_inode, &other_type);
1757 if (ret < 0 && ret != -ENOENT)
1760 /* was never and will never be overwritten */
1765 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1770 if (ow_inode == ino && gen == ino_gen) {
1775 /* we know that it is or will be overwritten. check this now */
1776 if (ow_inode < sctx->send_progress)
1786 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1787 * that got overwritten. This is used by process_recorded_refs to determine
1788 * if it has to use the path as returned by get_cur_path or the orphan name.
1790 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1793 struct fs_path *name = NULL;
1797 if (!sctx->parent_root)
1800 name = fs_path_alloc();
1804 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1808 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1809 name->start, fs_path_len(name));
1817 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1818 * so we need to do some special handling in case we have clashes. This function
1819 * takes care of this with the help of name_cache_entry::radix_list.
1820 * In case of error, nce is kfreed.
1822 static int name_cache_insert(struct send_ctx *sctx,
1823 struct name_cache_entry *nce)
1826 struct list_head *nce_head;
1828 nce_head = radix_tree_lookup(&sctx->name_cache,
1829 (unsigned long)nce->ino);
1831 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1836 INIT_LIST_HEAD(nce_head);
1838 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1845 list_add_tail(&nce->radix_list, nce_head);
1846 list_add_tail(&nce->list, &sctx->name_cache_list);
1847 sctx->name_cache_size++;
1852 static void name_cache_delete(struct send_ctx *sctx,
1853 struct name_cache_entry *nce)
1855 struct list_head *nce_head;
1857 nce_head = radix_tree_lookup(&sctx->name_cache,
1858 (unsigned long)nce->ino);
1860 btrfs_err(sctx->send_root->fs_info,
1861 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1862 nce->ino, sctx->name_cache_size);
1865 list_del(&nce->radix_list);
1866 list_del(&nce->list);
1867 sctx->name_cache_size--;
1870 * We may not get to the final release of nce_head if the lookup fails
1872 if (nce_head && list_empty(nce_head)) {
1873 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1878 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1881 struct list_head *nce_head;
1882 struct name_cache_entry *cur;
1884 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1888 list_for_each_entry(cur, nce_head, radix_list) {
1889 if (cur->ino == ino && cur->gen == gen)
1896 * Removes the entry from the list and adds it back to the end. This marks the
1897 * entry as recently used so that name_cache_clean_unused does not remove it.
1899 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1901 list_del(&nce->list);
1902 list_add_tail(&nce->list, &sctx->name_cache_list);
1906 * Remove some entries from the beginning of name_cache_list.
1908 static void name_cache_clean_unused(struct send_ctx *sctx)
1910 struct name_cache_entry *nce;
1912 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1915 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1916 nce = list_entry(sctx->name_cache_list.next,
1917 struct name_cache_entry, list);
1918 name_cache_delete(sctx, nce);
1923 static void name_cache_free(struct send_ctx *sctx)
1925 struct name_cache_entry *nce;
1927 while (!list_empty(&sctx->name_cache_list)) {
1928 nce = list_entry(sctx->name_cache_list.next,
1929 struct name_cache_entry, list);
1930 name_cache_delete(sctx, nce);
1936 * Used by get_cur_path for each ref up to the root.
1937 * Returns 0 if it succeeded.
1938 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1939 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1940 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1941 * Returns <0 in case of error.
1943 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1945 int skip_name_cache,
1948 struct fs_path *dest)
1952 struct btrfs_path *path = NULL;
1953 struct name_cache_entry *nce = NULL;
1955 if (skip_name_cache)
1958 * First check if we already did a call to this function with the same
1959 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1960 * return the cached result.
1962 nce = name_cache_search(sctx, ino, gen);
1964 if (ino < sctx->send_progress && nce->need_later_update) {
1965 name_cache_delete(sctx, nce);
1969 name_cache_used(sctx, nce);
1970 *parent_ino = nce->parent_ino;
1971 *parent_gen = nce->parent_gen;
1972 ret = fs_path_add(dest, nce->name, nce->name_len);
1980 path = alloc_path_for_send();
1985 * If the inode is not existent yet, add the orphan name and return 1.
1986 * This should only happen for the parent dir that we determine in
1989 ret = is_inode_existent(sctx, ino, gen);
1994 ret = gen_unique_name(sctx, ino, gen, dest);
2003 * Depending on whether the inode was already processed or not, use
2004 * send_root or parent_root for ref lookup.
2006 if (ino < sctx->send_progress && !skip_name_cache)
2007 ret = get_first_ref(sctx->send_root, ino,
2008 parent_ino, parent_gen, dest);
2010 ret = get_first_ref(sctx->parent_root, ino,
2011 parent_ino, parent_gen, dest);
2016 * Check if the ref was overwritten by an inode's ref that was processed
2017 * earlier. If yes, treat as orphan and return 1.
2019 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2020 dest->start, dest->end - dest->start);
2024 fs_path_reset(dest);
2025 ret = gen_unique_name(sctx, ino, gen, dest);
2030 if (skip_name_cache)
2035 * Store the result of the lookup in the name cache.
2037 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2045 nce->parent_ino = *parent_ino;
2046 nce->parent_gen = *parent_gen;
2047 nce->name_len = fs_path_len(dest);
2049 strcpy(nce->name, dest->start);
2051 if (ino < sctx->send_progress)
2052 nce->need_later_update = 0;
2054 nce->need_later_update = 1;
2056 nce_ret = name_cache_insert(sctx, nce);
2059 name_cache_clean_unused(sctx);
2062 btrfs_free_path(path);
2067 * Magic happens here. This function returns the first ref to an inode as it
2068 * would look like while receiving the stream at this point in time.
2069 * We walk the path up to the root. For every inode in between, we check if it
2070 * was already processed/sent. If yes, we continue with the parent as found
2071 * in send_root. If not, we continue with the parent as found in parent_root.
2072 * If we encounter an inode that was deleted at this point in time, we use the
2073 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2074 * that were not created yet and overwritten inodes/refs.
2076 * When do we have have orphan inodes:
2077 * 1. When an inode is freshly created and thus no valid refs are available yet
2078 * 2. When a directory lost all it's refs (deleted) but still has dir items
2079 * inside which were not processed yet (pending for move/delete). If anyone
2080 * tried to get the path to the dir items, it would get a path inside that
2082 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2083 * of an unprocessed inode. If in that case the first ref would be
2084 * overwritten, the overwritten inode gets "orphanized". Later when we
2085 * process this overwritten inode, it is restored at a new place by moving
2088 * sctx->send_progress tells this function at which point in time receiving
2091 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2092 struct fs_path *dest)
2095 struct fs_path *name = NULL;
2096 u64 parent_inode = 0;
2099 int skip_name_cache = 0;
2101 name = fs_path_alloc();
2107 if (is_waiting_for_move(sctx, ino))
2108 skip_name_cache = 1;
2111 fs_path_reset(dest);
2113 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2114 fs_path_reset(name);
2116 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
2117 &parent_inode, &parent_gen, name);
2123 if (!skip_name_cache &&
2124 is_waiting_for_move(sctx, parent_inode))
2125 skip_name_cache = 1;
2127 ret = fs_path_add_path(dest, name);
2138 fs_path_unreverse(dest);
2143 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2145 static int send_subvol_begin(struct send_ctx *sctx)
2148 struct btrfs_root *send_root = sctx->send_root;
2149 struct btrfs_root *parent_root = sctx->parent_root;
2150 struct btrfs_path *path;
2151 struct btrfs_key key;
2152 struct btrfs_root_ref *ref;
2153 struct extent_buffer *leaf;
2157 path = btrfs_alloc_path();
2161 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2163 btrfs_free_path(path);
2167 key.objectid = send_root->objectid;
2168 key.type = BTRFS_ROOT_BACKREF_KEY;
2171 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2180 leaf = path->nodes[0];
2181 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2182 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2183 key.objectid != send_root->objectid) {
2187 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2188 namelen = btrfs_root_ref_name_len(leaf, ref);
2189 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2190 btrfs_release_path(path);
2193 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2197 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2202 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2203 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2204 sctx->send_root->root_item.uuid);
2205 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2206 le64_to_cpu(sctx->send_root->root_item.ctransid));
2208 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2209 sctx->parent_root->root_item.uuid);
2210 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2211 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2214 ret = send_cmd(sctx);
2218 btrfs_free_path(path);
2223 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2228 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2230 p = fs_path_alloc();
2234 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2238 ret = get_cur_path(sctx, ino, gen, p);
2241 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2242 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2244 ret = send_cmd(sctx);
2252 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2257 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2259 p = fs_path_alloc();
2263 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2267 ret = get_cur_path(sctx, ino, gen, p);
2270 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2271 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2273 ret = send_cmd(sctx);
2281 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2286 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2288 p = fs_path_alloc();
2292 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2296 ret = get_cur_path(sctx, ino, gen, p);
2299 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2300 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2301 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2303 ret = send_cmd(sctx);
2311 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2314 struct fs_path *p = NULL;
2315 struct btrfs_inode_item *ii;
2316 struct btrfs_path *path = NULL;
2317 struct extent_buffer *eb;
2318 struct btrfs_key key;
2321 verbose_printk("btrfs: send_utimes %llu\n", ino);
2323 p = fs_path_alloc();
2327 path = alloc_path_for_send();
2334 key.type = BTRFS_INODE_ITEM_KEY;
2336 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2340 eb = path->nodes[0];
2341 slot = path->slots[0];
2342 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2344 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2348 ret = get_cur_path(sctx, ino, gen, p);
2351 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2352 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2353 btrfs_inode_atime(ii));
2354 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2355 btrfs_inode_mtime(ii));
2356 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2357 btrfs_inode_ctime(ii));
2358 /* TODO Add otime support when the otime patches get into upstream */
2360 ret = send_cmd(sctx);
2365 btrfs_free_path(path);
2370 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2371 * a valid path yet because we did not process the refs yet. So, the inode
2372 * is created as orphan.
2374 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2383 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2385 p = fs_path_alloc();
2389 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2394 if (S_ISREG(mode)) {
2395 cmd = BTRFS_SEND_C_MKFILE;
2396 } else if (S_ISDIR(mode)) {
2397 cmd = BTRFS_SEND_C_MKDIR;
2398 } else if (S_ISLNK(mode)) {
2399 cmd = BTRFS_SEND_C_SYMLINK;
2400 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2401 cmd = BTRFS_SEND_C_MKNOD;
2402 } else if (S_ISFIFO(mode)) {
2403 cmd = BTRFS_SEND_C_MKFIFO;
2404 } else if (S_ISSOCK(mode)) {
2405 cmd = BTRFS_SEND_C_MKSOCK;
2407 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2408 (int)(mode & S_IFMT));
2413 ret = begin_cmd(sctx, cmd);
2417 ret = gen_unique_name(sctx, ino, gen, p);
2421 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2422 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2424 if (S_ISLNK(mode)) {
2426 ret = read_symlink(sctx->send_root, ino, p);
2429 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2430 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2431 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2432 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2433 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2436 ret = send_cmd(sctx);
2448 * We need some special handling for inodes that get processed before the parent
2449 * directory got created. See process_recorded_refs for details.
2450 * This function does the check if we already created the dir out of order.
2452 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2455 struct btrfs_path *path = NULL;
2456 struct btrfs_key key;
2457 struct btrfs_key found_key;
2458 struct btrfs_key di_key;
2459 struct extent_buffer *eb;
2460 struct btrfs_dir_item *di;
2463 path = alloc_path_for_send();
2470 key.type = BTRFS_DIR_INDEX_KEY;
2472 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2477 eb = path->nodes[0];
2478 slot = path->slots[0];
2479 if (slot >= btrfs_header_nritems(eb)) {
2480 ret = btrfs_next_leaf(sctx->send_root, path);
2483 } else if (ret > 0) {
2490 btrfs_item_key_to_cpu(eb, &found_key, slot);
2491 if (found_key.objectid != key.objectid ||
2492 found_key.type != key.type) {
2497 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2498 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2500 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2501 di_key.objectid < sctx->send_progress) {
2510 btrfs_free_path(path);
2515 * Only creates the inode if it is:
2516 * 1. Not a directory
2517 * 2. Or a directory which was not created already due to out of order
2518 * directories. See did_create_dir and process_recorded_refs for details.
2520 static int send_create_inode_if_needed(struct send_ctx *sctx)
2524 if (S_ISDIR(sctx->cur_inode_mode)) {
2525 ret = did_create_dir(sctx, sctx->cur_ino);
2534 ret = send_create_inode(sctx, sctx->cur_ino);
2542 struct recorded_ref {
2543 struct list_head list;
2546 struct fs_path *full_path;
2554 * We need to process new refs before deleted refs, but compare_tree gives us
2555 * everything mixed. So we first record all refs and later process them.
2556 * This function is a helper to record one ref.
2558 static int record_ref(struct list_head *head, u64 dir,
2559 u64 dir_gen, struct fs_path *path)
2561 struct recorded_ref *ref;
2563 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2568 ref->dir_gen = dir_gen;
2569 ref->full_path = path;
2571 ref->name = (char *)kbasename(ref->full_path->start);
2572 ref->name_len = ref->full_path->end - ref->name;
2573 ref->dir_path = ref->full_path->start;
2574 if (ref->name == ref->full_path->start)
2575 ref->dir_path_len = 0;
2577 ref->dir_path_len = ref->full_path->end -
2578 ref->full_path->start - 1 - ref->name_len;
2580 list_add_tail(&ref->list, head);
2584 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2586 struct recorded_ref *new;
2588 new = kmalloc(sizeof(*ref), GFP_NOFS);
2592 new->dir = ref->dir;
2593 new->dir_gen = ref->dir_gen;
2594 new->full_path = NULL;
2595 INIT_LIST_HEAD(&new->list);
2596 list_add_tail(&new->list, list);
2600 static void __free_recorded_refs(struct list_head *head)
2602 struct recorded_ref *cur;
2604 while (!list_empty(head)) {
2605 cur = list_entry(head->next, struct recorded_ref, list);
2606 fs_path_free(cur->full_path);
2607 list_del(&cur->list);
2612 static void free_recorded_refs(struct send_ctx *sctx)
2614 __free_recorded_refs(&sctx->new_refs);
2615 __free_recorded_refs(&sctx->deleted_refs);
2619 * Renames/moves a file/dir to its orphan name. Used when the first
2620 * ref of an unprocessed inode gets overwritten and for all non empty
2623 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2624 struct fs_path *path)
2627 struct fs_path *orphan;
2629 orphan = fs_path_alloc();
2633 ret = gen_unique_name(sctx, ino, gen, orphan);
2637 ret = send_rename(sctx, path, orphan);
2640 fs_path_free(orphan);
2645 * Returns 1 if a directory can be removed at this point in time.
2646 * We check this by iterating all dir items and checking if the inode behind
2647 * the dir item was already processed.
2649 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2652 struct btrfs_root *root = sctx->parent_root;
2653 struct btrfs_path *path;
2654 struct btrfs_key key;
2655 struct btrfs_key found_key;
2656 struct btrfs_key loc;
2657 struct btrfs_dir_item *di;
2660 * Don't try to rmdir the top/root subvolume dir.
2662 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2665 path = alloc_path_for_send();
2670 key.type = BTRFS_DIR_INDEX_KEY;
2672 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2677 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2678 ret = btrfs_next_leaf(root, path);
2685 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2687 if (found_key.objectid != key.objectid ||
2688 found_key.type != key.type)
2691 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2692 struct btrfs_dir_item);
2693 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2695 if (loc.objectid > send_progress) {
2706 btrfs_free_path(path);
2710 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2712 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2713 struct waiting_dir_move *entry;
2716 entry = rb_entry(n, struct waiting_dir_move, node);
2717 if (ino < entry->ino)
2719 else if (ino > entry->ino)
2727 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2729 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2730 struct rb_node *parent = NULL;
2731 struct waiting_dir_move *entry, *dm;
2733 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2740 entry = rb_entry(parent, struct waiting_dir_move, node);
2741 if (ino < entry->ino) {
2743 } else if (ino > entry->ino) {
2744 p = &(*p)->rb_right;
2751 rb_link_node(&dm->node, parent, p);
2752 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2756 static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2758 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2759 struct waiting_dir_move *entry;
2762 entry = rb_entry(n, struct waiting_dir_move, node);
2763 if (ino < entry->ino) {
2765 } else if (ino > entry->ino) {
2768 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2776 static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2778 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2779 struct rb_node *parent = NULL;
2780 struct pending_dir_move *entry, *pm;
2781 struct recorded_ref *cur;
2785 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2788 pm->parent_ino = parent_ino;
2789 pm->ino = sctx->cur_ino;
2790 pm->gen = sctx->cur_inode_gen;
2791 INIT_LIST_HEAD(&pm->list);
2792 INIT_LIST_HEAD(&pm->update_refs);
2793 RB_CLEAR_NODE(&pm->node);
2797 entry = rb_entry(parent, struct pending_dir_move, node);
2798 if (parent_ino < entry->parent_ino) {
2800 } else if (parent_ino > entry->parent_ino) {
2801 p = &(*p)->rb_right;
2808 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2809 ret = dup_ref(cur, &pm->update_refs);
2813 list_for_each_entry(cur, &sctx->new_refs, list) {
2814 ret = dup_ref(cur, &pm->update_refs);
2819 ret = add_waiting_dir_move(sctx, pm->ino);
2824 list_add_tail(&pm->list, &entry->list);
2826 rb_link_node(&pm->node, parent, p);
2827 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2832 __free_recorded_refs(&pm->update_refs);
2838 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2841 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2842 struct pending_dir_move *entry;
2845 entry = rb_entry(n, struct pending_dir_move, node);
2846 if (parent_ino < entry->parent_ino)
2848 else if (parent_ino > entry->parent_ino)
2856 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2858 struct fs_path *from_path = NULL;
2859 struct fs_path *to_path = NULL;
2860 struct fs_path *name = NULL;
2861 u64 orig_progress = sctx->send_progress;
2862 struct recorded_ref *cur;
2863 u64 parent_ino, parent_gen;
2866 name = fs_path_alloc();
2867 from_path = fs_path_alloc();
2868 if (!name || !from_path) {
2873 ret = del_waiting_dir_move(sctx, pm->ino);
2876 ret = get_first_ref(sctx->parent_root, pm->ino,
2877 &parent_ino, &parent_gen, name);
2881 if (parent_ino == sctx->cur_ino) {
2882 /* child only renamed, not moved */
2883 ASSERT(parent_gen == sctx->cur_inode_gen);
2884 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2888 ret = fs_path_add_path(from_path, name);
2892 /* child moved and maybe renamed too */
2893 sctx->send_progress = pm->ino;
2894 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2902 to_path = fs_path_alloc();
2908 sctx->send_progress = sctx->cur_ino + 1;
2909 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2913 ret = send_rename(sctx, from_path, to_path);
2917 ret = send_utimes(sctx, pm->ino, pm->gen);
2922 * After rename/move, need to update the utimes of both new parent(s)
2923 * and old parent(s).
2925 list_for_each_entry(cur, &pm->update_refs, list) {
2926 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2933 fs_path_free(from_path);
2934 fs_path_free(to_path);
2935 sctx->send_progress = orig_progress;
2940 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2942 if (!list_empty(&m->list))
2944 if (!RB_EMPTY_NODE(&m->node))
2945 rb_erase(&m->node, &sctx->pending_dir_moves);
2946 __free_recorded_refs(&m->update_refs);
2950 static void tail_append_pending_moves(struct pending_dir_move *moves,
2951 struct list_head *stack)
2953 if (list_empty(&moves->list)) {
2954 list_add_tail(&moves->list, stack);
2957 list_splice_init(&moves->list, &list);
2958 list_add_tail(&moves->list, stack);
2959 list_splice_tail(&list, stack);
2963 static int apply_children_dir_moves(struct send_ctx *sctx)
2965 struct pending_dir_move *pm;
2966 struct list_head stack;
2967 u64 parent_ino = sctx->cur_ino;
2970 pm = get_pending_dir_moves(sctx, parent_ino);
2974 INIT_LIST_HEAD(&stack);
2975 tail_append_pending_moves(pm, &stack);
2977 while (!list_empty(&stack)) {
2978 pm = list_first_entry(&stack, struct pending_dir_move, list);
2979 parent_ino = pm->ino;
2980 ret = apply_dir_move(sctx, pm);
2981 free_pending_move(sctx, pm);
2984 pm = get_pending_dir_moves(sctx, parent_ino);
2986 tail_append_pending_moves(pm, &stack);
2991 while (!list_empty(&stack)) {
2992 pm = list_first_entry(&stack, struct pending_dir_move, list);
2993 free_pending_move(sctx, pm);
2998 static int wait_for_parent_move(struct send_ctx *sctx,
2999 struct recorded_ref *parent_ref)
3002 u64 ino = parent_ref->dir;
3003 u64 parent_ino_before, parent_ino_after;
3004 u64 new_gen, old_gen;
3005 struct fs_path *path_before = NULL;
3006 struct fs_path *path_after = NULL;
3009 if (parent_ref->dir <= sctx->cur_ino)
3012 if (is_waiting_for_move(sctx, ino))
3015 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3016 NULL, NULL, NULL, NULL);
3022 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3023 NULL, NULL, NULL, NULL);
3027 if (new_gen != old_gen)
3030 path_before = fs_path_alloc();
3034 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3036 if (ret == -ENOENT) {
3039 } else if (ret < 0) {
3043 path_after = fs_path_alloc();
3049 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3051 if (ret == -ENOENT) {
3054 } else if (ret < 0) {
3058 len1 = fs_path_len(path_before);
3059 len2 = fs_path_len(path_after);
3060 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3061 memcmp(path_before->start, path_after->start, len1)) {
3068 fs_path_free(path_before);
3069 fs_path_free(path_after);
3075 * This does all the move/link/unlink/rmdir magic.
3077 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3080 struct recorded_ref *cur;
3081 struct recorded_ref *cur2;
3082 struct list_head check_dirs;
3083 struct fs_path *valid_path = NULL;
3086 int did_overwrite = 0;
3088 u64 last_dir_ino_rm = 0;
3090 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3093 * This should never happen as the root dir always has the same ref
3094 * which is always '..'
3096 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3097 INIT_LIST_HEAD(&check_dirs);
3099 valid_path = fs_path_alloc();
3106 * First, check if the first ref of the current inode was overwritten
3107 * before. If yes, we know that the current inode was already orphanized
3108 * and thus use the orphan name. If not, we can use get_cur_path to
3109 * get the path of the first ref as it would like while receiving at
3110 * this point in time.
3111 * New inodes are always orphan at the beginning, so force to use the
3112 * orphan name in this case.
3113 * The first ref is stored in valid_path and will be updated if it
3114 * gets moved around.
3116 if (!sctx->cur_inode_new) {
3117 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3118 sctx->cur_inode_gen);
3124 if (sctx->cur_inode_new || did_overwrite) {
3125 ret = gen_unique_name(sctx, sctx->cur_ino,
3126 sctx->cur_inode_gen, valid_path);
3131 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3137 list_for_each_entry(cur, &sctx->new_refs, list) {
3139 * We may have refs where the parent directory does not exist
3140 * yet. This happens if the parent directories inum is higher
3141 * the the current inum. To handle this case, we create the
3142 * parent directory out of order. But we need to check if this
3143 * did already happen before due to other refs in the same dir.
3145 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3148 if (ret == inode_state_will_create) {
3151 * First check if any of the current inodes refs did
3152 * already create the dir.
3154 list_for_each_entry(cur2, &sctx->new_refs, list) {
3157 if (cur2->dir == cur->dir) {
3164 * If that did not happen, check if a previous inode
3165 * did already create the dir.
3168 ret = did_create_dir(sctx, cur->dir);
3172 ret = send_create_inode(sctx, cur->dir);
3179 * Check if this new ref would overwrite the first ref of
3180 * another unprocessed inode. If yes, orphanize the
3181 * overwritten inode. If we find an overwritten ref that is
3182 * not the first ref, simply unlink it.
3184 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3185 cur->name, cur->name_len,
3186 &ow_inode, &ow_gen);
3190 ret = is_first_ref(sctx->parent_root,
3191 ow_inode, cur->dir, cur->name,
3196 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3201 ret = send_unlink(sctx, cur->full_path);
3208 * link/move the ref to the new place. If we have an orphan
3209 * inode, move it and update valid_path. If not, link or move
3210 * it depending on the inode mode.
3213 ret = send_rename(sctx, valid_path, cur->full_path);
3217 ret = fs_path_copy(valid_path, cur->full_path);
3221 if (S_ISDIR(sctx->cur_inode_mode)) {
3223 * Dirs can't be linked, so move it. For moved
3224 * dirs, we always have one new and one deleted
3225 * ref. The deleted ref is ignored later.
3227 ret = wait_for_parent_move(sctx, cur);
3231 ret = add_pending_dir_move(sctx,
3235 ret = send_rename(sctx, valid_path,
3238 ret = fs_path_copy(valid_path,
3244 ret = send_link(sctx, cur->full_path,
3250 ret = dup_ref(cur, &check_dirs);
3255 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3257 * Check if we can already rmdir the directory. If not,
3258 * orphanize it. For every dir item inside that gets deleted
3259 * later, we do this check again and rmdir it then if possible.
3260 * See the use of check_dirs for more details.
3262 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3266 ret = send_rmdir(sctx, valid_path);
3269 } else if (!is_orphan) {
3270 ret = orphanize_inode(sctx, sctx->cur_ino,
3271 sctx->cur_inode_gen, valid_path);
3277 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3278 ret = dup_ref(cur, &check_dirs);
3282 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3283 !list_empty(&sctx->deleted_refs)) {
3285 * We have a moved dir. Add the old parent to check_dirs
3287 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3289 ret = dup_ref(cur, &check_dirs);
3292 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3294 * We have a non dir inode. Go through all deleted refs and
3295 * unlink them if they were not already overwritten by other
3298 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3299 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3300 sctx->cur_ino, sctx->cur_inode_gen,
3301 cur->name, cur->name_len);
3305 ret = send_unlink(sctx, cur->full_path);
3309 ret = dup_ref(cur, &check_dirs);
3314 * If the inode is still orphan, unlink the orphan. This may
3315 * happen when a previous inode did overwrite the first ref
3316 * of this inode and no new refs were added for the current
3317 * inode. Unlinking does not mean that the inode is deleted in
3318 * all cases. There may still be links to this inode in other
3322 ret = send_unlink(sctx, valid_path);
3329 * We did collect all parent dirs where cur_inode was once located. We
3330 * now go through all these dirs and check if they are pending for
3331 * deletion and if it's finally possible to perform the rmdir now.
3332 * We also update the inode stats of the parent dirs here.
3334 list_for_each_entry(cur, &check_dirs, list) {
3336 * In case we had refs into dirs that were not processed yet,
3337 * we don't need to do the utime and rmdir logic for these dirs.
3338 * The dir will be processed later.
3340 if (cur->dir > sctx->cur_ino)
3343 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3347 if (ret == inode_state_did_create ||
3348 ret == inode_state_no_change) {
3349 /* TODO delayed utimes */
3350 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3353 } else if (ret == inode_state_did_delete &&
3354 cur->dir != last_dir_ino_rm) {
3355 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
3359 ret = get_cur_path(sctx, cur->dir,
3360 cur->dir_gen, valid_path);
3363 ret = send_rmdir(sctx, valid_path);
3366 last_dir_ino_rm = cur->dir;
3374 __free_recorded_refs(&check_dirs);
3375 free_recorded_refs(sctx);
3376 fs_path_free(valid_path);
3380 static int __record_new_ref(int num, u64 dir, int index,
3381 struct fs_path *name,
3385 struct send_ctx *sctx = ctx;
3389 p = fs_path_alloc();
3393 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3398 ret = get_cur_path(sctx, dir, gen, p);
3401 ret = fs_path_add_path(p, name);
3405 ret = record_ref(&sctx->new_refs, dir, gen, p);
3413 static int __record_deleted_ref(int num, u64 dir, int index,
3414 struct fs_path *name,
3418 struct send_ctx *sctx = ctx;
3422 p = fs_path_alloc();
3426 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3431 ret = get_cur_path(sctx, dir, gen, p);
3434 ret = fs_path_add_path(p, name);
3438 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3446 static int record_new_ref(struct send_ctx *sctx)
3450 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3451 sctx->cmp_key, 0, __record_new_ref, sctx);
3460 static int record_deleted_ref(struct send_ctx *sctx)
3464 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3465 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3474 struct find_ref_ctx {
3477 struct btrfs_root *root;
3478 struct fs_path *name;
3482 static int __find_iref(int num, u64 dir, int index,
3483 struct fs_path *name,
3486 struct find_ref_ctx *ctx = ctx_;
3490 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3491 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3493 * To avoid doing extra lookups we'll only do this if everything
3496 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3500 if (dir_gen != ctx->dir_gen)
3502 ctx->found_idx = num;
3508 static int find_iref(struct btrfs_root *root,
3509 struct btrfs_path *path,
3510 struct btrfs_key *key,
3511 u64 dir, u64 dir_gen, struct fs_path *name)
3514 struct find_ref_ctx ctx;
3518 ctx.dir_gen = dir_gen;
3522 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3526 if (ctx.found_idx == -1)
3529 return ctx.found_idx;
3532 static int __record_changed_new_ref(int num, u64 dir, int index,
3533 struct fs_path *name,
3538 struct send_ctx *sctx = ctx;
3540 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3545 ret = find_iref(sctx->parent_root, sctx->right_path,
3546 sctx->cmp_key, dir, dir_gen, name);
3548 ret = __record_new_ref(num, dir, index, name, sctx);
3555 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3556 struct fs_path *name,
3561 struct send_ctx *sctx = ctx;
3563 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3568 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3569 dir, dir_gen, name);
3571 ret = __record_deleted_ref(num, dir, index, name, sctx);
3578 static int record_changed_ref(struct send_ctx *sctx)
3582 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3583 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3586 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3587 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3597 * Record and process all refs at once. Needed when an inode changes the
3598 * generation number, which means that it was deleted and recreated.
3600 static int process_all_refs(struct send_ctx *sctx,
3601 enum btrfs_compare_tree_result cmd)
3604 struct btrfs_root *root;
3605 struct btrfs_path *path;
3606 struct btrfs_key key;
3607 struct btrfs_key found_key;
3608 struct extent_buffer *eb;
3610 iterate_inode_ref_t cb;
3611 int pending_move = 0;
3613 path = alloc_path_for_send();
3617 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3618 root = sctx->send_root;
3619 cb = __record_new_ref;
3620 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3621 root = sctx->parent_root;
3622 cb = __record_deleted_ref;
3624 btrfs_err(sctx->send_root->fs_info,
3625 "Wrong command %d in process_all_refs", cmd);
3630 key.objectid = sctx->cmp_key->objectid;
3631 key.type = BTRFS_INODE_REF_KEY;
3633 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3638 eb = path->nodes[0];
3639 slot = path->slots[0];
3640 if (slot >= btrfs_header_nritems(eb)) {
3641 ret = btrfs_next_leaf(root, path);
3649 btrfs_item_key_to_cpu(eb, &found_key, slot);
3651 if (found_key.objectid != key.objectid ||
3652 (found_key.type != BTRFS_INODE_REF_KEY &&
3653 found_key.type != BTRFS_INODE_EXTREF_KEY))
3656 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3662 btrfs_release_path(path);
3664 ret = process_recorded_refs(sctx, &pending_move);
3665 /* Only applicable to an incremental send. */
3666 ASSERT(pending_move == 0);
3669 btrfs_free_path(path);
3673 static int send_set_xattr(struct send_ctx *sctx,
3674 struct fs_path *path,
3675 const char *name, int name_len,
3676 const char *data, int data_len)
3680 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3684 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3685 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3686 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3688 ret = send_cmd(sctx);
3695 static int send_remove_xattr(struct send_ctx *sctx,
3696 struct fs_path *path,
3697 const char *name, int name_len)
3701 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3705 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3706 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3708 ret = send_cmd(sctx);
3715 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3716 const char *name, int name_len,
3717 const char *data, int data_len,
3721 struct send_ctx *sctx = ctx;
3723 posix_acl_xattr_header dummy_acl;
3725 p = fs_path_alloc();
3730 * This hack is needed because empty acl's are stored as zero byte
3731 * data in xattrs. Problem with that is, that receiving these zero byte
3732 * acl's will fail later. To fix this, we send a dummy acl list that
3733 * only contains the version number and no entries.
3735 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3736 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3737 if (data_len == 0) {
3738 dummy_acl.a_version =
3739 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3740 data = (char *)&dummy_acl;
3741 data_len = sizeof(dummy_acl);
3745 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3749 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3756 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3757 const char *name, int name_len,
3758 const char *data, int data_len,
3762 struct send_ctx *sctx = ctx;
3765 p = fs_path_alloc();
3769 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3773 ret = send_remove_xattr(sctx, p, name, name_len);
3780 static int process_new_xattr(struct send_ctx *sctx)
3784 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3785 sctx->cmp_key, __process_new_xattr, sctx);
3790 static int process_deleted_xattr(struct send_ctx *sctx)
3794 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3795 sctx->cmp_key, __process_deleted_xattr, sctx);
3800 struct find_xattr_ctx {
3808 static int __find_xattr(int num, struct btrfs_key *di_key,
3809 const char *name, int name_len,
3810 const char *data, int data_len,
3811 u8 type, void *vctx)
3813 struct find_xattr_ctx *ctx = vctx;
3815 if (name_len == ctx->name_len &&
3816 strncmp(name, ctx->name, name_len) == 0) {
3817 ctx->found_idx = num;
3818 ctx->found_data_len = data_len;
3819 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
3820 if (!ctx->found_data)
3827 static int find_xattr(struct btrfs_root *root,
3828 struct btrfs_path *path,
3829 struct btrfs_key *key,
3830 const char *name, int name_len,
3831 char **data, int *data_len)
3834 struct find_xattr_ctx ctx;
3837 ctx.name_len = name_len;
3839 ctx.found_data = NULL;
3840 ctx.found_data_len = 0;
3842 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
3846 if (ctx.found_idx == -1)
3849 *data = ctx.found_data;
3850 *data_len = ctx.found_data_len;
3852 kfree(ctx.found_data);
3854 return ctx.found_idx;
3858 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3859 const char *name, int name_len,
3860 const char *data, int data_len,
3864 struct send_ctx *sctx = ctx;
3865 char *found_data = NULL;
3866 int found_data_len = 0;
3868 ret = find_xattr(sctx->parent_root, sctx->right_path,
3869 sctx->cmp_key, name, name_len, &found_data,
3871 if (ret == -ENOENT) {
3872 ret = __process_new_xattr(num, di_key, name, name_len, data,
3873 data_len, type, ctx);
3874 } else if (ret >= 0) {
3875 if (data_len != found_data_len ||
3876 memcmp(data, found_data, data_len)) {
3877 ret = __process_new_xattr(num, di_key, name, name_len,
3878 data, data_len, type, ctx);
3888 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3889 const char *name, int name_len,
3890 const char *data, int data_len,
3894 struct send_ctx *sctx = ctx;
3896 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3897 name, name_len, NULL, NULL);
3899 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3900 data_len, type, ctx);
3907 static int process_changed_xattr(struct send_ctx *sctx)
3911 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3912 sctx->cmp_key, __process_changed_new_xattr, sctx);
3915 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3916 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3922 static int process_all_new_xattrs(struct send_ctx *sctx)
3925 struct btrfs_root *root;
3926 struct btrfs_path *path;
3927 struct btrfs_key key;
3928 struct btrfs_key found_key;
3929 struct extent_buffer *eb;
3932 path = alloc_path_for_send();
3936 root = sctx->send_root;
3938 key.objectid = sctx->cmp_key->objectid;
3939 key.type = BTRFS_XATTR_ITEM_KEY;
3941 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3946 eb = path->nodes[0];
3947 slot = path->slots[0];
3948 if (slot >= btrfs_header_nritems(eb)) {
3949 ret = btrfs_next_leaf(root, path);
3952 } else if (ret > 0) {
3959 btrfs_item_key_to_cpu(eb, &found_key, slot);
3960 if (found_key.objectid != key.objectid ||
3961 found_key.type != key.type) {
3966 ret = iterate_dir_item(root, path, &found_key,
3967 __process_new_xattr, sctx);
3975 btrfs_free_path(path);
3979 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3981 struct btrfs_root *root = sctx->send_root;
3982 struct btrfs_fs_info *fs_info = root->fs_info;
3983 struct inode *inode;
3986 struct btrfs_key key;
3987 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3989 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3992 key.objectid = sctx->cur_ino;
3993 key.type = BTRFS_INODE_ITEM_KEY;
3996 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3998 return PTR_ERR(inode);
4000 if (offset + len > i_size_read(inode)) {
4001 if (offset > i_size_read(inode))
4004 len = offset - i_size_read(inode);
4009 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4010 while (index <= last_index) {
4011 unsigned cur_len = min_t(unsigned, len,
4012 PAGE_CACHE_SIZE - pg_offset);
4013 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4019 if (!PageUptodate(page)) {
4020 btrfs_readpage(NULL, page);
4022 if (!PageUptodate(page)) {
4024 page_cache_release(page);
4031 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4034 page_cache_release(page);
4046 * Read some bytes from the current inode/file and send a write command to
4049 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4053 ssize_t num_read = 0;
4055 p = fs_path_alloc();
4059 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4061 num_read = fill_read_buf(sctx, offset, len);
4062 if (num_read <= 0) {
4068 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4072 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4076 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4077 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4078 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4080 ret = send_cmd(sctx);
4091 * Send a clone command to user space.
4093 static int send_clone(struct send_ctx *sctx,
4094 u64 offset, u32 len,
4095 struct clone_root *clone_root)
4101 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4102 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4103 clone_root->root->objectid, clone_root->ino,
4104 clone_root->offset);
4106 p = fs_path_alloc();
4110 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4114 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4118 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4119 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4120 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4122 if (clone_root->root == sctx->send_root) {
4123 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4124 &gen, NULL, NULL, NULL, NULL);
4127 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4129 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4134 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4135 clone_root->root->root_item.uuid);
4136 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4137 le64_to_cpu(clone_root->root->root_item.ctransid));
4138 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4139 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4140 clone_root->offset);
4142 ret = send_cmd(sctx);
4151 * Send an update extent command to user space.
4153 static int send_update_extent(struct send_ctx *sctx,
4154 u64 offset, u32 len)
4159 p = fs_path_alloc();
4163 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4167 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4171 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4172 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4173 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4175 ret = send_cmd(sctx);
4183 static int send_hole(struct send_ctx *sctx, u64 end)
4185 struct fs_path *p = NULL;
4186 u64 offset = sctx->cur_inode_last_extent;
4190 p = fs_path_alloc();
4193 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4194 while (offset < end) {
4195 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4197 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4200 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4203 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4204 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4205 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4206 ret = send_cmd(sctx);
4216 static int send_write_or_clone(struct send_ctx *sctx,
4217 struct btrfs_path *path,
4218 struct btrfs_key *key,
4219 struct clone_root *clone_root)
4222 struct btrfs_file_extent_item *ei;
4223 u64 offset = key->offset;
4228 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4230 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4231 struct btrfs_file_extent_item);
4232 type = btrfs_file_extent_type(path->nodes[0], ei);
4233 if (type == BTRFS_FILE_EXTENT_INLINE) {
4234 len = btrfs_file_extent_inline_len(path->nodes[0],
4235 path->slots[0], ei);
4237 * it is possible the inline item won't cover the whole page,
4238 * but there may be items after this page. Make
4239 * sure to send the whole thing
4241 len = PAGE_CACHE_ALIGN(len);
4243 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4246 if (offset + len > sctx->cur_inode_size)
4247 len = sctx->cur_inode_size - offset;
4253 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4254 ret = send_clone(sctx, offset, len, clone_root);
4255 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4256 ret = send_update_extent(sctx, offset, len);
4260 if (l > BTRFS_SEND_READ_SIZE)
4261 l = BTRFS_SEND_READ_SIZE;
4262 ret = send_write(sctx, pos + offset, l);
4275 static int is_extent_unchanged(struct send_ctx *sctx,
4276 struct btrfs_path *left_path,
4277 struct btrfs_key *ekey)
4280 struct btrfs_key key;
4281 struct btrfs_path *path = NULL;
4282 struct extent_buffer *eb;
4284 struct btrfs_key found_key;
4285 struct btrfs_file_extent_item *ei;
4290 u64 left_offset_fixed;
4298 path = alloc_path_for_send();
4302 eb = left_path->nodes[0];
4303 slot = left_path->slots[0];
4304 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4305 left_type = btrfs_file_extent_type(eb, ei);
4307 if (left_type != BTRFS_FILE_EXTENT_REG) {
4311 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4312 left_len = btrfs_file_extent_num_bytes(eb, ei);
4313 left_offset = btrfs_file_extent_offset(eb, ei);
4314 left_gen = btrfs_file_extent_generation(eb, ei);
4317 * Following comments will refer to these graphics. L is the left
4318 * extents which we are checking at the moment. 1-8 are the right
4319 * extents that we iterate.
4322 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4325 * |--1--|-2b-|...(same as above)
4327 * Alternative situation. Happens on files where extents got split.
4329 * |-----------7-----------|-6-|
4331 * Alternative situation. Happens on files which got larger.
4334 * Nothing follows after 8.
4337 key.objectid = ekey->objectid;
4338 key.type = BTRFS_EXTENT_DATA_KEY;
4339 key.offset = ekey->offset;
4340 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4349 * Handle special case where the right side has no extents at all.
4351 eb = path->nodes[0];
4352 slot = path->slots[0];
4353 btrfs_item_key_to_cpu(eb, &found_key, slot);
4354 if (found_key.objectid != key.objectid ||
4355 found_key.type != key.type) {
4356 /* If we're a hole then just pretend nothing changed */
4357 ret = (left_disknr) ? 0 : 1;
4362 * We're now on 2a, 2b or 7.
4365 while (key.offset < ekey->offset + left_len) {
4366 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4367 right_type = btrfs_file_extent_type(eb, ei);
4368 if (right_type != BTRFS_FILE_EXTENT_REG) {
4373 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4374 right_len = btrfs_file_extent_num_bytes(eb, ei);
4375 right_offset = btrfs_file_extent_offset(eb, ei);
4376 right_gen = btrfs_file_extent_generation(eb, ei);
4379 * Are we at extent 8? If yes, we know the extent is changed.
4380 * This may only happen on the first iteration.
4382 if (found_key.offset + right_len <= ekey->offset) {
4383 /* If we're a hole just pretend nothing changed */
4384 ret = (left_disknr) ? 0 : 1;
4388 left_offset_fixed = left_offset;
4389 if (key.offset < ekey->offset) {
4390 /* Fix the right offset for 2a and 7. */
4391 right_offset += ekey->offset - key.offset;
4393 /* Fix the left offset for all behind 2a and 2b */
4394 left_offset_fixed += key.offset - ekey->offset;
4398 * Check if we have the same extent.
4400 if (left_disknr != right_disknr ||
4401 left_offset_fixed != right_offset ||
4402 left_gen != right_gen) {
4408 * Go to the next extent.
4410 ret = btrfs_next_item(sctx->parent_root, path);
4414 eb = path->nodes[0];
4415 slot = path->slots[0];
4416 btrfs_item_key_to_cpu(eb, &found_key, slot);
4418 if (ret || found_key.objectid != key.objectid ||
4419 found_key.type != key.type) {
4420 key.offset += right_len;
4423 if (found_key.offset != key.offset + right_len) {
4431 * We're now behind the left extent (treat as unchanged) or at the end
4432 * of the right side (treat as changed).
4434 if (key.offset >= ekey->offset + left_len)
4441 btrfs_free_path(path);
4445 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4447 struct btrfs_path *path;
4448 struct btrfs_root *root = sctx->send_root;
4449 struct btrfs_file_extent_item *fi;
4450 struct btrfs_key key;
4455 path = alloc_path_for_send();
4459 sctx->cur_inode_last_extent = 0;
4461 key.objectid = sctx->cur_ino;
4462 key.type = BTRFS_EXTENT_DATA_KEY;
4463 key.offset = offset;
4464 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4468 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4469 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4472 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4473 struct btrfs_file_extent_item);
4474 type = btrfs_file_extent_type(path->nodes[0], fi);
4475 if (type == BTRFS_FILE_EXTENT_INLINE) {
4476 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4477 path->slots[0], fi);
4478 extent_end = ALIGN(key.offset + size,
4479 sctx->send_root->sectorsize);
4481 extent_end = key.offset +
4482 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4484 sctx->cur_inode_last_extent = extent_end;
4486 btrfs_free_path(path);
4490 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4491 struct btrfs_key *key)
4493 struct btrfs_file_extent_item *fi;
4498 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4501 if (sctx->cur_inode_last_extent == (u64)-1) {
4502 ret = get_last_extent(sctx, key->offset - 1);
4507 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4508 struct btrfs_file_extent_item);
4509 type = btrfs_file_extent_type(path->nodes[0], fi);
4510 if (type == BTRFS_FILE_EXTENT_INLINE) {
4511 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4512 path->slots[0], fi);
4513 extent_end = ALIGN(key->offset + size,
4514 sctx->send_root->sectorsize);
4516 extent_end = key->offset +
4517 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4520 if (path->slots[0] == 0 &&
4521 sctx->cur_inode_last_extent < key->offset) {
4523 * We might have skipped entire leafs that contained only
4524 * file extent items for our current inode. These leafs have
4525 * a generation number smaller (older) than the one in the
4526 * current leaf and the leaf our last extent came from, and
4527 * are located between these 2 leafs.
4529 ret = get_last_extent(sctx, key->offset - 1);
4534 if (sctx->cur_inode_last_extent < key->offset)
4535 ret = send_hole(sctx, key->offset);
4536 sctx->cur_inode_last_extent = extent_end;
4540 static int process_extent(struct send_ctx *sctx,
4541 struct btrfs_path *path,
4542 struct btrfs_key *key)
4544 struct clone_root *found_clone = NULL;
4547 if (S_ISLNK(sctx->cur_inode_mode))
4550 if (sctx->parent_root && !sctx->cur_inode_new) {
4551 ret = is_extent_unchanged(sctx, path, key);
4559 struct btrfs_file_extent_item *ei;
4562 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4563 struct btrfs_file_extent_item);
4564 type = btrfs_file_extent_type(path->nodes[0], ei);
4565 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4566 type == BTRFS_FILE_EXTENT_REG) {
4568 * The send spec does not have a prealloc command yet,
4569 * so just leave a hole for prealloc'ed extents until
4570 * we have enough commands queued up to justify rev'ing
4573 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4578 /* Have a hole, just skip it. */
4579 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4586 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4587 sctx->cur_inode_size, &found_clone);
4588 if (ret != -ENOENT && ret < 0)
4591 ret = send_write_or_clone(sctx, path, key, found_clone);
4595 ret = maybe_send_hole(sctx, path, key);
4600 static int process_all_extents(struct send_ctx *sctx)
4603 struct btrfs_root *root;
4604 struct btrfs_path *path;
4605 struct btrfs_key key;
4606 struct btrfs_key found_key;
4607 struct extent_buffer *eb;
4610 root = sctx->send_root;
4611 path = alloc_path_for_send();
4615 key.objectid = sctx->cmp_key->objectid;
4616 key.type = BTRFS_EXTENT_DATA_KEY;
4618 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4623 eb = path->nodes[0];
4624 slot = path->slots[0];
4626 if (slot >= btrfs_header_nritems(eb)) {
4627 ret = btrfs_next_leaf(root, path);
4630 } else if (ret > 0) {
4637 btrfs_item_key_to_cpu(eb, &found_key, slot);
4639 if (found_key.objectid != key.objectid ||
4640 found_key.type != key.type) {
4645 ret = process_extent(sctx, path, &found_key);
4653 btrfs_free_path(path);
4657 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4659 int *refs_processed)
4663 if (sctx->cur_ino == 0)
4665 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4666 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4668 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4671 ret = process_recorded_refs(sctx, pending_move);
4675 *refs_processed = 1;
4680 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4691 int pending_move = 0;
4692 int refs_processed = 0;
4694 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4700 * We have processed the refs and thus need to advance send_progress.
4701 * Now, calls to get_cur_xxx will take the updated refs of the current
4702 * inode into account.
4704 * On the other hand, if our current inode is a directory and couldn't
4705 * be moved/renamed because its parent was renamed/moved too and it has
4706 * a higher inode number, we can only move/rename our current inode
4707 * after we moved/renamed its parent. Therefore in this case operate on
4708 * the old path (pre move/rename) of our current inode, and the
4709 * move/rename will be performed later.
4711 if (refs_processed && !pending_move)
4712 sctx->send_progress = sctx->cur_ino + 1;
4714 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4716 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4719 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4720 &left_mode, &left_uid, &left_gid, NULL);
4724 if (!sctx->parent_root || sctx->cur_inode_new) {
4726 if (!S_ISLNK(sctx->cur_inode_mode))
4729 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4730 NULL, NULL, &right_mode, &right_uid,
4735 if (left_uid != right_uid || left_gid != right_gid)
4737 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4741 if (S_ISREG(sctx->cur_inode_mode)) {
4742 if (need_send_hole(sctx)) {
4743 if (sctx->cur_inode_last_extent == (u64)-1) {
4744 ret = get_last_extent(sctx, (u64)-1);
4748 if (sctx->cur_inode_last_extent <
4749 sctx->cur_inode_size) {
4750 ret = send_hole(sctx, sctx->cur_inode_size);
4755 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4756 sctx->cur_inode_size);
4762 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4763 left_uid, left_gid);
4768 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4775 * If other directory inodes depended on our current directory
4776 * inode's move/rename, now do their move/rename operations.
4778 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4779 ret = apply_children_dir_moves(sctx);
4785 * Need to send that every time, no matter if it actually
4786 * changed between the two trees as we have done changes to
4789 sctx->send_progress = sctx->cur_ino + 1;
4790 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4798 static int changed_inode(struct send_ctx *sctx,
4799 enum btrfs_compare_tree_result result)
4802 struct btrfs_key *key = sctx->cmp_key;
4803 struct btrfs_inode_item *left_ii = NULL;
4804 struct btrfs_inode_item *right_ii = NULL;
4808 sctx->cur_ino = key->objectid;
4809 sctx->cur_inode_new_gen = 0;
4810 sctx->cur_inode_last_extent = (u64)-1;
4813 * Set send_progress to current inode. This will tell all get_cur_xxx
4814 * functions that the current inode's refs are not updated yet. Later,
4815 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4817 sctx->send_progress = sctx->cur_ino;
4819 if (result == BTRFS_COMPARE_TREE_NEW ||
4820 result == BTRFS_COMPARE_TREE_CHANGED) {
4821 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4822 sctx->left_path->slots[0],
4823 struct btrfs_inode_item);
4824 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4827 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4828 sctx->right_path->slots[0],
4829 struct btrfs_inode_item);
4830 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4833 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4834 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4835 sctx->right_path->slots[0],
4836 struct btrfs_inode_item);
4838 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4842 * The cur_ino = root dir case is special here. We can't treat
4843 * the inode as deleted+reused because it would generate a
4844 * stream that tries to delete/mkdir the root dir.
4846 if (left_gen != right_gen &&
4847 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4848 sctx->cur_inode_new_gen = 1;
4851 if (result == BTRFS_COMPARE_TREE_NEW) {
4852 sctx->cur_inode_gen = left_gen;
4853 sctx->cur_inode_new = 1;
4854 sctx->cur_inode_deleted = 0;
4855 sctx->cur_inode_size = btrfs_inode_size(
4856 sctx->left_path->nodes[0], left_ii);
4857 sctx->cur_inode_mode = btrfs_inode_mode(
4858 sctx->left_path->nodes[0], left_ii);
4859 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4860 ret = send_create_inode_if_needed(sctx);
4861 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4862 sctx->cur_inode_gen = right_gen;
4863 sctx->cur_inode_new = 0;
4864 sctx->cur_inode_deleted = 1;
4865 sctx->cur_inode_size = btrfs_inode_size(
4866 sctx->right_path->nodes[0], right_ii);
4867 sctx->cur_inode_mode = btrfs_inode_mode(
4868 sctx->right_path->nodes[0], right_ii);
4869 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4871 * We need to do some special handling in case the inode was
4872 * reported as changed with a changed generation number. This
4873 * means that the original inode was deleted and new inode
4874 * reused the same inum. So we have to treat the old inode as
4875 * deleted and the new one as new.
4877 if (sctx->cur_inode_new_gen) {
4879 * First, process the inode as if it was deleted.
4881 sctx->cur_inode_gen = right_gen;
4882 sctx->cur_inode_new = 0;
4883 sctx->cur_inode_deleted = 1;
4884 sctx->cur_inode_size = btrfs_inode_size(
4885 sctx->right_path->nodes[0], right_ii);
4886 sctx->cur_inode_mode = btrfs_inode_mode(
4887 sctx->right_path->nodes[0], right_ii);
4888 ret = process_all_refs(sctx,
4889 BTRFS_COMPARE_TREE_DELETED);
4894 * Now process the inode as if it was new.
4896 sctx->cur_inode_gen = left_gen;
4897 sctx->cur_inode_new = 1;
4898 sctx->cur_inode_deleted = 0;
4899 sctx->cur_inode_size = btrfs_inode_size(
4900 sctx->left_path->nodes[0], left_ii);
4901 sctx->cur_inode_mode = btrfs_inode_mode(
4902 sctx->left_path->nodes[0], left_ii);
4903 ret = send_create_inode_if_needed(sctx);
4907 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4911 * Advance send_progress now as we did not get into
4912 * process_recorded_refs_if_needed in the new_gen case.
4914 sctx->send_progress = sctx->cur_ino + 1;
4917 * Now process all extents and xattrs of the inode as if
4918 * they were all new.
4920 ret = process_all_extents(sctx);
4923 ret = process_all_new_xattrs(sctx);
4927 sctx->cur_inode_gen = left_gen;
4928 sctx->cur_inode_new = 0;
4929 sctx->cur_inode_new_gen = 0;
4930 sctx->cur_inode_deleted = 0;
4931 sctx->cur_inode_size = btrfs_inode_size(
4932 sctx->left_path->nodes[0], left_ii);
4933 sctx->cur_inode_mode = btrfs_inode_mode(
4934 sctx->left_path->nodes[0], left_ii);
4943 * We have to process new refs before deleted refs, but compare_trees gives us
4944 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4945 * first and later process them in process_recorded_refs.
4946 * For the cur_inode_new_gen case, we skip recording completely because
4947 * changed_inode did already initiate processing of refs. The reason for this is
4948 * that in this case, compare_tree actually compares the refs of 2 different
4949 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4950 * refs of the right tree as deleted and all refs of the left tree as new.
4952 static int changed_ref(struct send_ctx *sctx,
4953 enum btrfs_compare_tree_result result)
4957 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4959 if (!sctx->cur_inode_new_gen &&
4960 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4961 if (result == BTRFS_COMPARE_TREE_NEW)
4962 ret = record_new_ref(sctx);
4963 else if (result == BTRFS_COMPARE_TREE_DELETED)
4964 ret = record_deleted_ref(sctx);
4965 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4966 ret = record_changed_ref(sctx);
4973 * Process new/deleted/changed xattrs. We skip processing in the
4974 * cur_inode_new_gen case because changed_inode did already initiate processing
4975 * of xattrs. The reason is the same as in changed_ref
4977 static int changed_xattr(struct send_ctx *sctx,
4978 enum btrfs_compare_tree_result result)
4982 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4984 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4985 if (result == BTRFS_COMPARE_TREE_NEW)
4986 ret = process_new_xattr(sctx);
4987 else if (result == BTRFS_COMPARE_TREE_DELETED)
4988 ret = process_deleted_xattr(sctx);
4989 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4990 ret = process_changed_xattr(sctx);
4997 * Process new/deleted/changed extents. We skip processing in the
4998 * cur_inode_new_gen case because changed_inode did already initiate processing
4999 * of extents. The reason is the same as in changed_ref
5001 static int changed_extent(struct send_ctx *sctx,
5002 enum btrfs_compare_tree_result result)
5006 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5008 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5009 if (result != BTRFS_COMPARE_TREE_DELETED)
5010 ret = process_extent(sctx, sctx->left_path,
5017 static int dir_changed(struct send_ctx *sctx, u64 dir)
5019 u64 orig_gen, new_gen;
5022 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5027 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5032 return (orig_gen != new_gen) ? 1 : 0;
5035 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5036 struct btrfs_key *key)
5038 struct btrfs_inode_extref *extref;
5039 struct extent_buffer *leaf;
5040 u64 dirid = 0, last_dirid = 0;
5047 /* Easy case, just check this one dirid */
5048 if (key->type == BTRFS_INODE_REF_KEY) {
5049 dirid = key->offset;
5051 ret = dir_changed(sctx, dirid);
5055 leaf = path->nodes[0];
5056 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5057 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5058 while (cur_offset < item_size) {
5059 extref = (struct btrfs_inode_extref *)(ptr +
5061 dirid = btrfs_inode_extref_parent(leaf, extref);
5062 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5063 cur_offset += ref_name_len + sizeof(*extref);
5064 if (dirid == last_dirid)
5066 ret = dir_changed(sctx, dirid);
5076 * Updates compare related fields in sctx and simply forwards to the actual
5077 * changed_xxx functions.
5079 static int changed_cb(struct btrfs_root *left_root,
5080 struct btrfs_root *right_root,
5081 struct btrfs_path *left_path,
5082 struct btrfs_path *right_path,
5083 struct btrfs_key *key,
5084 enum btrfs_compare_tree_result result,
5088 struct send_ctx *sctx = ctx;
5090 if (result == BTRFS_COMPARE_TREE_SAME) {
5091 if (key->type == BTRFS_INODE_REF_KEY ||
5092 key->type == BTRFS_INODE_EXTREF_KEY) {
5093 ret = compare_refs(sctx, left_path, key);
5098 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5099 return maybe_send_hole(sctx, left_path, key);
5103 result = BTRFS_COMPARE_TREE_CHANGED;
5107 sctx->left_path = left_path;
5108 sctx->right_path = right_path;
5109 sctx->cmp_key = key;
5111 ret = finish_inode_if_needed(sctx, 0);
5115 /* Ignore non-FS objects */
5116 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5117 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5120 if (key->type == BTRFS_INODE_ITEM_KEY)
5121 ret = changed_inode(sctx, result);
5122 else if (key->type == BTRFS_INODE_REF_KEY ||
5123 key->type == BTRFS_INODE_EXTREF_KEY)
5124 ret = changed_ref(sctx, result);
5125 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5126 ret = changed_xattr(sctx, result);
5127 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5128 ret = changed_extent(sctx, result);
5134 static int full_send_tree(struct send_ctx *sctx)
5137 struct btrfs_trans_handle *trans = NULL;
5138 struct btrfs_root *send_root = sctx->send_root;
5139 struct btrfs_key key;
5140 struct btrfs_key found_key;
5141 struct btrfs_path *path;
5142 struct extent_buffer *eb;
5147 path = alloc_path_for_send();
5151 spin_lock(&send_root->root_item_lock);
5152 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5153 spin_unlock(&send_root->root_item_lock);
5155 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5156 key.type = BTRFS_INODE_ITEM_KEY;
5161 * We need to make sure the transaction does not get committed
5162 * while we do anything on commit roots. Join a transaction to prevent
5165 trans = btrfs_join_transaction(send_root);
5166 if (IS_ERR(trans)) {
5167 ret = PTR_ERR(trans);
5173 * Make sure the tree has not changed after re-joining. We detect this
5174 * by comparing start_ctransid and ctransid. They should always match.
5176 spin_lock(&send_root->root_item_lock);
5177 ctransid = btrfs_root_ctransid(&send_root->root_item);
5178 spin_unlock(&send_root->root_item_lock);
5180 if (ctransid != start_ctransid) {
5181 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
5182 "send was modified in between. This is "
5183 "probably a bug.\n");
5188 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5196 * When someone want to commit while we iterate, end the
5197 * joined transaction and rejoin.
5199 if (btrfs_should_end_transaction(trans, send_root)) {
5200 ret = btrfs_end_transaction(trans, send_root);
5204 btrfs_release_path(path);
5208 eb = path->nodes[0];
5209 slot = path->slots[0];
5210 btrfs_item_key_to_cpu(eb, &found_key, slot);
5212 ret = changed_cb(send_root, NULL, path, NULL,
5213 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5217 key.objectid = found_key.objectid;
5218 key.type = found_key.type;
5219 key.offset = found_key.offset + 1;
5221 ret = btrfs_next_item(send_root, path);
5231 ret = finish_inode_if_needed(sctx, 1);
5234 btrfs_free_path(path);
5237 ret = btrfs_end_transaction(trans, send_root);
5239 btrfs_end_transaction(trans, send_root);
5244 static int send_subvol(struct send_ctx *sctx)
5248 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5249 ret = send_header(sctx);
5254 ret = send_subvol_begin(sctx);
5258 if (sctx->parent_root) {
5259 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5263 ret = finish_inode_if_needed(sctx, 1);
5267 ret = full_send_tree(sctx);
5273 free_recorded_refs(sctx);
5277 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5279 spin_lock(&root->root_item_lock);
5280 root->send_in_progress--;
5282 * Not much left to do, we don't know why it's unbalanced and
5283 * can't blindly reset it to 0.
5285 if (root->send_in_progress < 0)
5286 btrfs_err(root->fs_info,
5287 "send_in_progres unbalanced %d root %llu\n",
5288 root->send_in_progress, root->root_key.objectid);
5289 spin_unlock(&root->root_item_lock);
5292 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5295 struct btrfs_root *send_root;
5296 struct btrfs_root *clone_root;
5297 struct btrfs_fs_info *fs_info;
5298 struct btrfs_ioctl_send_args *arg = NULL;
5299 struct btrfs_key key;
5300 struct send_ctx *sctx = NULL;
5302 u64 *clone_sources_tmp = NULL;
5303 int clone_sources_to_rollback = 0;
5304 int sort_clone_roots = 0;
5307 if (!capable(CAP_SYS_ADMIN))
5310 send_root = BTRFS_I(file_inode(mnt_file))->root;
5311 fs_info = send_root->fs_info;
5314 * The subvolume must remain read-only during send, protect against
5317 spin_lock(&send_root->root_item_lock);
5318 send_root->send_in_progress++;
5319 spin_unlock(&send_root->root_item_lock);
5322 * This is done when we lookup the root, it should already be complete
5323 * by the time we get here.
5325 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5328 * Userspace tools do the checks and warn the user if it's
5331 if (!btrfs_root_readonly(send_root)) {
5336 arg = memdup_user(arg_, sizeof(*arg));
5343 if (!access_ok(VERIFY_READ, arg->clone_sources,
5344 sizeof(*arg->clone_sources) *
5345 arg->clone_sources_count)) {
5350 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5355 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5361 INIT_LIST_HEAD(&sctx->new_refs);
5362 INIT_LIST_HEAD(&sctx->deleted_refs);
5363 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5364 INIT_LIST_HEAD(&sctx->name_cache_list);
5366 sctx->flags = arg->flags;
5368 sctx->send_filp = fget(arg->send_fd);
5369 if (!sctx->send_filp) {
5374 sctx->send_root = send_root;
5375 sctx->clone_roots_cnt = arg->clone_sources_count;
5377 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5378 sctx->send_buf = vmalloc(sctx->send_max_size);
5379 if (!sctx->send_buf) {
5384 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5385 if (!sctx->read_buf) {
5390 sctx->pending_dir_moves = RB_ROOT;
5391 sctx->waiting_dir_moves = RB_ROOT;
5393 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5394 (arg->clone_sources_count + 1));
5395 if (!sctx->clone_roots) {
5400 if (arg->clone_sources_count) {
5401 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5402 sizeof(*arg->clone_sources));
5403 if (!clone_sources_tmp) {
5408 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5409 arg->clone_sources_count *
5410 sizeof(*arg->clone_sources));
5416 for (i = 0; i < arg->clone_sources_count; i++) {
5417 key.objectid = clone_sources_tmp[i];
5418 key.type = BTRFS_ROOT_ITEM_KEY;
5419 key.offset = (u64)-1;
5421 index = srcu_read_lock(&fs_info->subvol_srcu);
5423 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5424 if (IS_ERR(clone_root)) {
5425 srcu_read_unlock(&fs_info->subvol_srcu, index);
5426 ret = PTR_ERR(clone_root);
5429 clone_sources_to_rollback = i + 1;
5430 spin_lock(&clone_root->root_item_lock);
5431 clone_root->send_in_progress++;
5432 if (!btrfs_root_readonly(clone_root)) {
5433 spin_unlock(&clone_root->root_item_lock);
5434 srcu_read_unlock(&fs_info->subvol_srcu, index);
5438 spin_unlock(&clone_root->root_item_lock);
5439 srcu_read_unlock(&fs_info->subvol_srcu, index);
5441 sctx->clone_roots[i].root = clone_root;
5443 vfree(clone_sources_tmp);
5444 clone_sources_tmp = NULL;
5447 if (arg->parent_root) {
5448 key.objectid = arg->parent_root;
5449 key.type = BTRFS_ROOT_ITEM_KEY;
5450 key.offset = (u64)-1;
5452 index = srcu_read_lock(&fs_info->subvol_srcu);
5454 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5455 if (IS_ERR(sctx->parent_root)) {
5456 srcu_read_unlock(&fs_info->subvol_srcu, index);
5457 ret = PTR_ERR(sctx->parent_root);
5461 spin_lock(&sctx->parent_root->root_item_lock);
5462 sctx->parent_root->send_in_progress++;
5463 if (!btrfs_root_readonly(sctx->parent_root)) {
5464 spin_unlock(&sctx->parent_root->root_item_lock);
5465 srcu_read_unlock(&fs_info->subvol_srcu, index);
5469 spin_unlock(&sctx->parent_root->root_item_lock);
5471 srcu_read_unlock(&fs_info->subvol_srcu, index);
5475 * Clones from send_root are allowed, but only if the clone source
5476 * is behind the current send position. This is checked while searching
5477 * for possible clone sources.
5479 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5481 /* We do a bsearch later */
5482 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5483 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5485 sort_clone_roots = 1;
5487 ret = send_subvol(sctx);
5491 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5492 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5495 ret = send_cmd(sctx);
5501 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5502 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5504 struct pending_dir_move *pm;
5506 n = rb_first(&sctx->pending_dir_moves);
5507 pm = rb_entry(n, struct pending_dir_move, node);
5508 while (!list_empty(&pm->list)) {
5509 struct pending_dir_move *pm2;
5511 pm2 = list_first_entry(&pm->list,
5512 struct pending_dir_move, list);
5513 free_pending_move(sctx, pm2);
5515 free_pending_move(sctx, pm);
5518 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5519 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5521 struct waiting_dir_move *dm;
5523 n = rb_first(&sctx->waiting_dir_moves);
5524 dm = rb_entry(n, struct waiting_dir_move, node);
5525 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5529 if (sort_clone_roots) {
5530 for (i = 0; i < sctx->clone_roots_cnt; i++)
5531 btrfs_root_dec_send_in_progress(
5532 sctx->clone_roots[i].root);
5534 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5535 btrfs_root_dec_send_in_progress(
5536 sctx->clone_roots[i].root);
5538 btrfs_root_dec_send_in_progress(send_root);
5540 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5541 btrfs_root_dec_send_in_progress(sctx->parent_root);
5544 vfree(clone_sources_tmp);
5547 if (sctx->send_filp)
5548 fput(sctx->send_filp);
5550 vfree(sctx->clone_roots);
5551 vfree(sctx->send_buf);
5552 vfree(sctx->read_buf);
5554 name_cache_free(sctx);