2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
38 static int g_verbose = 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
58 unsigned int reversed:1;
59 unsigned int virtual_mem:1;
65 #define FS_PATH_INLINE_SIZE \
66 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
69 /* reused for each extent */
71 struct btrfs_root *root;
78 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
79 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
82 struct file *send_filp;
88 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
89 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
91 struct btrfs_root *send_root;
92 struct btrfs_root *parent_root;
93 struct clone_root *clone_roots;
96 /* current state of the compare_tree call */
97 struct btrfs_path *left_path;
98 struct btrfs_path *right_path;
99 struct btrfs_key *cmp_key;
102 * infos of the currently processed inode. In case of deleted inodes,
103 * these are the values from the deleted inode.
108 int cur_inode_new_gen;
109 int cur_inode_deleted;
112 u64 cur_inode_last_extent;
116 struct list_head new_refs;
117 struct list_head deleted_refs;
119 struct radix_tree_root name_cache;
120 struct list_head name_cache_list;
126 * We process inodes by their increasing order, so if before an
127 * incremental send we reverse the parent/child relationship of
128 * directories such that a directory with a lower inode number was
129 * the parent of a directory with a higher inode number, and the one
130 * becoming the new parent got renamed too, we can't rename/move the
131 * directory with lower inode number when we finish processing it - we
132 * must process the directory with higher inode number first, then
133 * rename/move it and then rename/move the directory with lower inode
134 * number. Example follows.
136 * Tree state when the first send was performed:
148 * Tree state when the second (incremental) send is performed:
157 * The sequence of steps that lead to the second state was:
159 * mv /a/b/c/d /a/b/c2/d2
160 * mv /a/b/c /a/b/c2/d2/cc
162 * "c" has lower inode number, but we can't move it (2nd mv operation)
163 * before we move "d", which has higher inode number.
165 * So we just memorize which move/rename operations must be performed
166 * later when their respective parent is processed and moved/renamed.
169 /* Indexed by parent directory inode number. */
170 struct rb_root pending_dir_moves;
173 * Reverse index, indexed by the inode number of a directory that
174 * is waiting for the move/rename of its immediate parent before its
175 * own move/rename can be performed.
177 struct rb_root waiting_dir_moves;
180 struct pending_dir_move {
182 struct list_head list;
186 struct list_head update_refs;
189 struct waiting_dir_move {
194 struct name_cache_entry {
195 struct list_head list;
197 * radix_tree has only 32bit entries but we need to handle 64bit inums.
198 * We use the lower 32bit of the 64bit inum to store it in the tree. If
199 * more then one inum would fall into the same entry, we use radix_list
200 * to store the additional entries. radix_list is also used to store
201 * entries where two entries have the same inum but different
204 struct list_head radix_list;
210 int need_later_update;
215 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
217 static int need_send_hole(struct send_ctx *sctx)
219 return (sctx->parent_root && !sctx->cur_inode_new &&
220 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
221 S_ISREG(sctx->cur_inode_mode));
224 static void fs_path_reset(struct fs_path *p)
227 p->start = p->buf + p->buf_len - 1;
237 static struct fs_path *fs_path_alloc(void)
241 p = kmalloc(sizeof(*p), GFP_NOFS);
246 p->buf = p->inline_buf;
247 p->buf_len = FS_PATH_INLINE_SIZE;
252 static struct fs_path *fs_path_alloc_reversed(void)
264 static void fs_path_free(struct fs_path *p)
268 if (p->buf != p->inline_buf) {
277 static int fs_path_len(struct fs_path *p)
279 return p->end - p->start;
282 static int fs_path_ensure_buf(struct fs_path *p, int len)
290 if (p->buf_len >= len)
293 path_len = p->end - p->start;
294 old_buf_len = p->buf_len;
295 len = PAGE_ALIGN(len);
297 if (p->buf == p->inline_buf) {
298 tmp_buf = kmalloc(len, GFP_NOFS | __GFP_NOWARN);
300 tmp_buf = vmalloc(len);
305 memcpy(tmp_buf, p->buf, p->buf_len);
309 if (p->virtual_mem) {
310 tmp_buf = vmalloc(len);
313 memcpy(tmp_buf, p->buf, p->buf_len);
316 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
318 tmp_buf = vmalloc(len);
321 memcpy(tmp_buf, p->buf, p->buf_len);
330 tmp_buf = p->buf + old_buf_len - path_len - 1;
331 p->end = p->buf + p->buf_len - 1;
332 p->start = p->end - path_len;
333 memmove(p->start, tmp_buf, path_len + 1);
336 p->end = p->start + path_len;
341 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
346 new_len = p->end - p->start + name_len;
347 if (p->start != p->end)
349 ret = fs_path_ensure_buf(p, new_len);
354 if (p->start != p->end)
356 p->start -= name_len;
357 p->prepared = p->start;
359 if (p->start != p->end)
361 p->prepared = p->end;
370 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
374 ret = fs_path_prepare_for_add(p, name_len);
377 memcpy(p->prepared, name, name_len);
384 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
388 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
391 memcpy(p->prepared, p2->start, p2->end - p2->start);
398 static int fs_path_add_from_extent_buffer(struct fs_path *p,
399 struct extent_buffer *eb,
400 unsigned long off, int len)
404 ret = fs_path_prepare_for_add(p, len);
408 read_extent_buffer(eb, p->prepared, off, len);
415 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
419 p->reversed = from->reversed;
422 ret = fs_path_add_path(p, from);
428 static void fs_path_unreverse(struct fs_path *p)
437 len = p->end - p->start;
439 p->end = p->start + len;
440 memmove(p->start, tmp, len + 1);
444 static struct btrfs_path *alloc_path_for_send(void)
446 struct btrfs_path *path;
448 path = btrfs_alloc_path();
451 path->search_commit_root = 1;
452 path->skip_locking = 1;
456 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
466 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
467 /* TODO handle that correctly */
468 /*if (ret == -ERESTARTSYS) {
487 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
489 struct btrfs_tlv_header *hdr;
490 int total_len = sizeof(*hdr) + len;
491 int left = sctx->send_max_size - sctx->send_size;
493 if (unlikely(left < total_len))
496 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
497 hdr->tlv_type = cpu_to_le16(attr);
498 hdr->tlv_len = cpu_to_le16(len);
499 memcpy(hdr + 1, data, len);
500 sctx->send_size += total_len;
505 #define TLV_PUT_DEFINE_INT(bits) \
506 static int tlv_put_u##bits(struct send_ctx *sctx, \
507 u##bits attr, u##bits value) \
509 __le##bits __tmp = cpu_to_le##bits(value); \
510 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
513 TLV_PUT_DEFINE_INT(64)
515 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
516 const char *str, int len)
520 return tlv_put(sctx, attr, str, len);
523 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
526 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
529 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
530 struct extent_buffer *eb,
531 struct btrfs_timespec *ts)
533 struct btrfs_timespec bts;
534 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
535 return tlv_put(sctx, attr, &bts, sizeof(bts));
539 #define TLV_PUT(sctx, attrtype, attrlen, data) \
541 ret = tlv_put(sctx, attrtype, attrlen, data); \
543 goto tlv_put_failure; \
546 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
548 ret = tlv_put_u##bits(sctx, attrtype, value); \
550 goto tlv_put_failure; \
553 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
554 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
555 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
556 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
557 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
559 ret = tlv_put_string(sctx, attrtype, str, len); \
561 goto tlv_put_failure; \
563 #define TLV_PUT_PATH(sctx, attrtype, p) \
565 ret = tlv_put_string(sctx, attrtype, p->start, \
566 p->end - p->start); \
568 goto tlv_put_failure; \
570 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
572 ret = tlv_put_uuid(sctx, attrtype, uuid); \
574 goto tlv_put_failure; \
576 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
578 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
580 goto tlv_put_failure; \
583 static int send_header(struct send_ctx *sctx)
585 struct btrfs_stream_header hdr;
587 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
588 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
590 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
595 * For each command/item we want to send to userspace, we call this function.
597 static int begin_cmd(struct send_ctx *sctx, int cmd)
599 struct btrfs_cmd_header *hdr;
601 if (WARN_ON(!sctx->send_buf))
604 BUG_ON(sctx->send_size);
606 sctx->send_size += sizeof(*hdr);
607 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
608 hdr->cmd = cpu_to_le16(cmd);
613 static int send_cmd(struct send_ctx *sctx)
616 struct btrfs_cmd_header *hdr;
619 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
620 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
623 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
624 hdr->crc = cpu_to_le32(crc);
626 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
629 sctx->total_send_size += sctx->send_size;
630 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
637 * Sends a move instruction to user space
639 static int send_rename(struct send_ctx *sctx,
640 struct fs_path *from, struct fs_path *to)
644 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
646 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
650 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
651 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
653 ret = send_cmd(sctx);
661 * Sends a link instruction to user space
663 static int send_link(struct send_ctx *sctx,
664 struct fs_path *path, struct fs_path *lnk)
668 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
670 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
674 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
675 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
677 ret = send_cmd(sctx);
685 * Sends an unlink instruction to user space
687 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
691 verbose_printk("btrfs: send_unlink %s\n", path->start);
693 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
697 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
699 ret = send_cmd(sctx);
707 * Sends a rmdir instruction to user space
709 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
713 verbose_printk("btrfs: send_rmdir %s\n", path->start);
715 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
719 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
721 ret = send_cmd(sctx);
729 * Helper function to retrieve some fields from an inode item.
731 static int get_inode_info(struct btrfs_root *root,
732 u64 ino, u64 *size, u64 *gen,
733 u64 *mode, u64 *uid, u64 *gid,
737 struct btrfs_inode_item *ii;
738 struct btrfs_key key;
739 struct btrfs_path *path;
741 path = alloc_path_for_send();
746 key.type = BTRFS_INODE_ITEM_KEY;
748 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
756 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
757 struct btrfs_inode_item);
759 *size = btrfs_inode_size(path->nodes[0], ii);
761 *gen = btrfs_inode_generation(path->nodes[0], ii);
763 *mode = btrfs_inode_mode(path->nodes[0], ii);
765 *uid = btrfs_inode_uid(path->nodes[0], ii);
767 *gid = btrfs_inode_gid(path->nodes[0], ii);
769 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
772 btrfs_free_path(path);
776 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
781 * Helper function to iterate the entries in ONE btrfs_inode_ref or
782 * btrfs_inode_extref.
783 * The iterate callback may return a non zero value to stop iteration. This can
784 * be a negative value for error codes or 1 to simply stop it.
786 * path must point to the INODE_REF or INODE_EXTREF when called.
788 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
789 struct btrfs_key *found_key, int resolve,
790 iterate_inode_ref_t iterate, void *ctx)
792 struct extent_buffer *eb = path->nodes[0];
793 struct btrfs_item *item;
794 struct btrfs_inode_ref *iref;
795 struct btrfs_inode_extref *extref;
796 struct btrfs_path *tmp_path;
800 int slot = path->slots[0];
807 unsigned long name_off;
808 unsigned long elem_size;
811 p = fs_path_alloc_reversed();
815 tmp_path = alloc_path_for_send();
822 if (found_key->type == BTRFS_INODE_REF_KEY) {
823 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
824 struct btrfs_inode_ref);
825 item = btrfs_item_nr(slot);
826 total = btrfs_item_size(eb, item);
827 elem_size = sizeof(*iref);
829 ptr = btrfs_item_ptr_offset(eb, slot);
830 total = btrfs_item_size_nr(eb, slot);
831 elem_size = sizeof(*extref);
834 while (cur < total) {
837 if (found_key->type == BTRFS_INODE_REF_KEY) {
838 iref = (struct btrfs_inode_ref *)(ptr + cur);
839 name_len = btrfs_inode_ref_name_len(eb, iref);
840 name_off = (unsigned long)(iref + 1);
841 index = btrfs_inode_ref_index(eb, iref);
842 dir = found_key->offset;
844 extref = (struct btrfs_inode_extref *)(ptr + cur);
845 name_len = btrfs_inode_extref_name_len(eb, extref);
846 name_off = (unsigned long)&extref->name;
847 index = btrfs_inode_extref_index(eb, extref);
848 dir = btrfs_inode_extref_parent(eb, extref);
852 start = btrfs_ref_to_path(root, tmp_path, name_len,
856 ret = PTR_ERR(start);
859 if (start < p->buf) {
860 /* overflow , try again with larger buffer */
861 ret = fs_path_ensure_buf(p,
862 p->buf_len + p->buf - start);
865 start = btrfs_ref_to_path(root, tmp_path,
870 ret = PTR_ERR(start);
873 BUG_ON(start < p->buf);
877 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
883 cur += elem_size + name_len;
884 ret = iterate(num, dir, index, p, ctx);
891 btrfs_free_path(tmp_path);
896 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
897 const char *name, int name_len,
898 const char *data, int data_len,
902 * Helper function to iterate the entries in ONE btrfs_dir_item.
903 * The iterate callback may return a non zero value to stop iteration. This can
904 * be a negative value for error codes or 1 to simply stop it.
906 * path must point to the dir item when called.
908 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
909 struct btrfs_key *found_key,
910 iterate_dir_item_t iterate, void *ctx)
913 struct extent_buffer *eb;
914 struct btrfs_item *item;
915 struct btrfs_dir_item *di;
916 struct btrfs_key di_key;
931 buf = kmalloc(buf_len, GFP_NOFS);
938 slot = path->slots[0];
939 item = btrfs_item_nr(slot);
940 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
943 total = btrfs_item_size(eb, item);
946 while (cur < total) {
947 name_len = btrfs_dir_name_len(eb, di);
948 data_len = btrfs_dir_data_len(eb, di);
949 type = btrfs_dir_type(eb, di);
950 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
952 if (name_len + data_len > buf_len) {
953 buf_len = PAGE_ALIGN(name_len + data_len);
955 buf2 = vmalloc(buf_len);
962 buf2 = krealloc(buf, buf_len, GFP_NOFS);
964 buf2 = vmalloc(buf_len);
978 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
979 name_len + data_len);
981 len = sizeof(*di) + name_len + data_len;
982 di = (struct btrfs_dir_item *)((char *)di + len);
985 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
986 data_len, type, ctx);
1005 static int __copy_first_ref(int num, u64 dir, int index,
1006 struct fs_path *p, void *ctx)
1009 struct fs_path *pt = ctx;
1011 ret = fs_path_copy(pt, p);
1015 /* we want the first only */
1020 * Retrieve the first path of an inode. If an inode has more then one
1021 * ref/hardlink, this is ignored.
1023 static int get_inode_path(struct btrfs_root *root,
1024 u64 ino, struct fs_path *path)
1027 struct btrfs_key key, found_key;
1028 struct btrfs_path *p;
1030 p = alloc_path_for_send();
1034 fs_path_reset(path);
1037 key.type = BTRFS_INODE_REF_KEY;
1040 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1047 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1048 if (found_key.objectid != ino ||
1049 (found_key.type != BTRFS_INODE_REF_KEY &&
1050 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1055 ret = iterate_inode_ref(root, p, &found_key, 1,
1056 __copy_first_ref, path);
1066 struct backref_ctx {
1067 struct send_ctx *sctx;
1069 /* number of total found references */
1073 * used for clones found in send_root. clones found behind cur_objectid
1074 * and cur_offset are not considered as allowed clones.
1079 /* may be truncated in case it's the last extent in a file */
1082 /* Just to check for bugs in backref resolving */
1086 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1088 u64 root = (u64)(uintptr_t)key;
1089 struct clone_root *cr = (struct clone_root *)elt;
1091 if (root < cr->root->objectid)
1093 if (root > cr->root->objectid)
1098 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1100 struct clone_root *cr1 = (struct clone_root *)e1;
1101 struct clone_root *cr2 = (struct clone_root *)e2;
1103 if (cr1->root->objectid < cr2->root->objectid)
1105 if (cr1->root->objectid > cr2->root->objectid)
1111 * Called for every backref that is found for the current extent.
1112 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1114 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1116 struct backref_ctx *bctx = ctx_;
1117 struct clone_root *found;
1121 /* First check if the root is in the list of accepted clone sources */
1122 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1123 bctx->sctx->clone_roots_cnt,
1124 sizeof(struct clone_root),
1125 __clone_root_cmp_bsearch);
1129 if (found->root == bctx->sctx->send_root &&
1130 ino == bctx->cur_objectid &&
1131 offset == bctx->cur_offset) {
1132 bctx->found_itself = 1;
1136 * There are inodes that have extents that lie behind its i_size. Don't
1137 * accept clones from these extents.
1139 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1144 if (offset + bctx->extent_len > i_size)
1148 * Make sure we don't consider clones from send_root that are
1149 * behind the current inode/offset.
1151 if (found->root == bctx->sctx->send_root) {
1153 * TODO for the moment we don't accept clones from the inode
1154 * that is currently send. We may change this when
1155 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1158 if (ino >= bctx->cur_objectid)
1161 if (ino > bctx->cur_objectid)
1163 if (offset + bctx->extent_len > bctx->cur_offset)
1169 found->found_refs++;
1170 if (ino < found->ino) {
1172 found->offset = offset;
1173 } else if (found->ino == ino) {
1175 * same extent found more then once in the same file.
1177 if (found->offset > offset + bctx->extent_len)
1178 found->offset = offset;
1185 * Given an inode, offset and extent item, it finds a good clone for a clone
1186 * instruction. Returns -ENOENT when none could be found. The function makes
1187 * sure that the returned clone is usable at the point where sending is at the
1188 * moment. This means, that no clones are accepted which lie behind the current
1191 * path must point to the extent item when called.
1193 static int find_extent_clone(struct send_ctx *sctx,
1194 struct btrfs_path *path,
1195 u64 ino, u64 data_offset,
1197 struct clone_root **found)
1204 u64 extent_item_pos;
1206 struct btrfs_file_extent_item *fi;
1207 struct extent_buffer *eb = path->nodes[0];
1208 struct backref_ctx *backref_ctx = NULL;
1209 struct clone_root *cur_clone_root;
1210 struct btrfs_key found_key;
1211 struct btrfs_path *tmp_path;
1215 tmp_path = alloc_path_for_send();
1219 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1225 if (data_offset >= ino_size) {
1227 * There may be extents that lie behind the file's size.
1228 * I at least had this in combination with snapshotting while
1229 * writing large files.
1235 fi = btrfs_item_ptr(eb, path->slots[0],
1236 struct btrfs_file_extent_item);
1237 extent_type = btrfs_file_extent_type(eb, fi);
1238 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1242 compressed = btrfs_file_extent_compression(eb, fi);
1244 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1245 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1246 if (disk_byte == 0) {
1250 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1252 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1253 &found_key, &flags);
1254 btrfs_release_path(tmp_path);
1258 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1264 * Setup the clone roots.
1266 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1267 cur_clone_root = sctx->clone_roots + i;
1268 cur_clone_root->ino = (u64)-1;
1269 cur_clone_root->offset = 0;
1270 cur_clone_root->found_refs = 0;
1273 backref_ctx->sctx = sctx;
1274 backref_ctx->found = 0;
1275 backref_ctx->cur_objectid = ino;
1276 backref_ctx->cur_offset = data_offset;
1277 backref_ctx->found_itself = 0;
1278 backref_ctx->extent_len = num_bytes;
1281 * The last extent of a file may be too large due to page alignment.
1282 * We need to adjust extent_len in this case so that the checks in
1283 * __iterate_backrefs work.
1285 if (data_offset + num_bytes >= ino_size)
1286 backref_ctx->extent_len = ino_size - data_offset;
1289 * Now collect all backrefs.
1291 if (compressed == BTRFS_COMPRESS_NONE)
1292 extent_item_pos = logical - found_key.objectid;
1294 extent_item_pos = 0;
1296 extent_item_pos = logical - found_key.objectid;
1297 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1298 found_key.objectid, extent_item_pos, 1,
1299 __iterate_backrefs, backref_ctx);
1304 if (!backref_ctx->found_itself) {
1305 /* found a bug in backref code? */
1307 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1308 "send_root. inode=%llu, offset=%llu, "
1309 "disk_byte=%llu found extent=%llu\n",
1310 ino, data_offset, disk_byte, found_key.objectid);
1314 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1316 "num_bytes=%llu, logical=%llu\n",
1317 data_offset, ino, num_bytes, logical);
1319 if (!backref_ctx->found)
1320 verbose_printk("btrfs: no clones found\n");
1322 cur_clone_root = NULL;
1323 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1324 if (sctx->clone_roots[i].found_refs) {
1325 if (!cur_clone_root)
1326 cur_clone_root = sctx->clone_roots + i;
1327 else if (sctx->clone_roots[i].root == sctx->send_root)
1328 /* prefer clones from send_root over others */
1329 cur_clone_root = sctx->clone_roots + i;
1334 if (cur_clone_root) {
1335 if (compressed != BTRFS_COMPRESS_NONE) {
1337 * Offsets given by iterate_extent_inodes() are relative
1338 * to the start of the extent, we need to add logical
1339 * offset from the file extent item.
1340 * (See why at backref.c:check_extent_in_eb())
1342 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1345 *found = cur_clone_root;
1352 btrfs_free_path(tmp_path);
1357 static int read_symlink(struct btrfs_root *root,
1359 struct fs_path *dest)
1362 struct btrfs_path *path;
1363 struct btrfs_key key;
1364 struct btrfs_file_extent_item *ei;
1370 path = alloc_path_for_send();
1375 key.type = BTRFS_EXTENT_DATA_KEY;
1377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1382 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1383 struct btrfs_file_extent_item);
1384 type = btrfs_file_extent_type(path->nodes[0], ei);
1385 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1386 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1387 BUG_ON(compression);
1389 off = btrfs_file_extent_inline_start(ei);
1390 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1392 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1395 btrfs_free_path(path);
1400 * Helper function to generate a file name that is unique in the root of
1401 * send_root and parent_root. This is used to generate names for orphan inodes.
1403 static int gen_unique_name(struct send_ctx *sctx,
1405 struct fs_path *dest)
1408 struct btrfs_path *path;
1409 struct btrfs_dir_item *di;
1414 path = alloc_path_for_send();
1419 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1421 if (len >= sizeof(tmp)) {
1422 /* should really not happen */
1427 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1428 path, BTRFS_FIRST_FREE_OBJECTID,
1429 tmp, strlen(tmp), 0);
1430 btrfs_release_path(path);
1436 /* not unique, try again */
1441 if (!sctx->parent_root) {
1447 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1448 path, BTRFS_FIRST_FREE_OBJECTID,
1449 tmp, strlen(tmp), 0);
1450 btrfs_release_path(path);
1456 /* not unique, try again */
1464 ret = fs_path_add(dest, tmp, strlen(tmp));
1467 btrfs_free_path(path);
1472 inode_state_no_change,
1473 inode_state_will_create,
1474 inode_state_did_create,
1475 inode_state_will_delete,
1476 inode_state_did_delete,
1479 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1487 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1489 if (ret < 0 && ret != -ENOENT)
1493 if (!sctx->parent_root) {
1494 right_ret = -ENOENT;
1496 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1497 NULL, NULL, NULL, NULL);
1498 if (ret < 0 && ret != -ENOENT)
1503 if (!left_ret && !right_ret) {
1504 if (left_gen == gen && right_gen == gen) {
1505 ret = inode_state_no_change;
1506 } else if (left_gen == gen) {
1507 if (ino < sctx->send_progress)
1508 ret = inode_state_did_create;
1510 ret = inode_state_will_create;
1511 } else if (right_gen == gen) {
1512 if (ino < sctx->send_progress)
1513 ret = inode_state_did_delete;
1515 ret = inode_state_will_delete;
1519 } else if (!left_ret) {
1520 if (left_gen == gen) {
1521 if (ino < sctx->send_progress)
1522 ret = inode_state_did_create;
1524 ret = inode_state_will_create;
1528 } else if (!right_ret) {
1529 if (right_gen == gen) {
1530 if (ino < sctx->send_progress)
1531 ret = inode_state_did_delete;
1533 ret = inode_state_will_delete;
1545 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1549 ret = get_cur_inode_state(sctx, ino, gen);
1553 if (ret == inode_state_no_change ||
1554 ret == inode_state_did_create ||
1555 ret == inode_state_will_delete)
1565 * Helper function to lookup a dir item in a dir.
1567 static int lookup_dir_item_inode(struct btrfs_root *root,
1568 u64 dir, const char *name, int name_len,
1573 struct btrfs_dir_item *di;
1574 struct btrfs_key key;
1575 struct btrfs_path *path;
1577 path = alloc_path_for_send();
1581 di = btrfs_lookup_dir_item(NULL, root, path,
1582 dir, name, name_len, 0);
1591 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1592 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1596 *found_inode = key.objectid;
1597 *found_type = btrfs_dir_type(path->nodes[0], di);
1600 btrfs_free_path(path);
1605 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1606 * generation of the parent dir and the name of the dir entry.
1608 static int get_first_ref(struct btrfs_root *root, u64 ino,
1609 u64 *dir, u64 *dir_gen, struct fs_path *name)
1612 struct btrfs_key key;
1613 struct btrfs_key found_key;
1614 struct btrfs_path *path;
1618 path = alloc_path_for_send();
1623 key.type = BTRFS_INODE_REF_KEY;
1626 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1630 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1632 if (ret || found_key.objectid != ino ||
1633 (found_key.type != BTRFS_INODE_REF_KEY &&
1634 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1639 if (key.type == BTRFS_INODE_REF_KEY) {
1640 struct btrfs_inode_ref *iref;
1641 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1642 struct btrfs_inode_ref);
1643 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1644 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1645 (unsigned long)(iref + 1),
1647 parent_dir = found_key.offset;
1649 struct btrfs_inode_extref *extref;
1650 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1651 struct btrfs_inode_extref);
1652 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1653 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1654 (unsigned long)&extref->name, len);
1655 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1659 btrfs_release_path(path);
1661 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1669 btrfs_free_path(path);
1673 static int is_first_ref(struct btrfs_root *root,
1675 const char *name, int name_len)
1678 struct fs_path *tmp_name;
1682 tmp_name = fs_path_alloc();
1686 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1690 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1695 ret = !memcmp(tmp_name->start, name, name_len);
1698 fs_path_free(tmp_name);
1703 * Used by process_recorded_refs to determine if a new ref would overwrite an
1704 * already existing ref. In case it detects an overwrite, it returns the
1705 * inode/gen in who_ino/who_gen.
1706 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1707 * to make sure later references to the overwritten inode are possible.
1708 * Orphanizing is however only required for the first ref of an inode.
1709 * process_recorded_refs does an additional is_first_ref check to see if
1710 * orphanizing is really required.
1712 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1713 const char *name, int name_len,
1714 u64 *who_ino, u64 *who_gen)
1718 u64 other_inode = 0;
1721 if (!sctx->parent_root)
1724 ret = is_inode_existent(sctx, dir, dir_gen);
1729 * If we have a parent root we need to verify that the parent dir was
1730 * not delted and then re-created, if it was then we have no overwrite
1731 * and we can just unlink this entry.
1733 if (sctx->parent_root) {
1734 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1736 if (ret < 0 && ret != -ENOENT)
1746 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1747 &other_inode, &other_type);
1748 if (ret < 0 && ret != -ENOENT)
1756 * Check if the overwritten ref was already processed. If yes, the ref
1757 * was already unlinked/moved, so we can safely assume that we will not
1758 * overwrite anything at this point in time.
1760 if (other_inode > sctx->send_progress) {
1761 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1762 who_gen, NULL, NULL, NULL, NULL);
1767 *who_ino = other_inode;
1777 * Checks if the ref was overwritten by an already processed inode. This is
1778 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1779 * thus the orphan name needs be used.
1780 * process_recorded_refs also uses it to avoid unlinking of refs that were
1783 static int did_overwrite_ref(struct send_ctx *sctx,
1784 u64 dir, u64 dir_gen,
1785 u64 ino, u64 ino_gen,
1786 const char *name, int name_len)
1793 if (!sctx->parent_root)
1796 ret = is_inode_existent(sctx, dir, dir_gen);
1800 /* check if the ref was overwritten by another ref */
1801 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1802 &ow_inode, &other_type);
1803 if (ret < 0 && ret != -ENOENT)
1806 /* was never and will never be overwritten */
1811 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1816 if (ow_inode == ino && gen == ino_gen) {
1821 /* we know that it is or will be overwritten. check this now */
1822 if (ow_inode < sctx->send_progress)
1832 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1833 * that got overwritten. This is used by process_recorded_refs to determine
1834 * if it has to use the path as returned by get_cur_path or the orphan name.
1836 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1839 struct fs_path *name = NULL;
1843 if (!sctx->parent_root)
1846 name = fs_path_alloc();
1850 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1854 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1855 name->start, fs_path_len(name));
1863 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1864 * so we need to do some special handling in case we have clashes. This function
1865 * takes care of this with the help of name_cache_entry::radix_list.
1866 * In case of error, nce is kfreed.
1868 static int name_cache_insert(struct send_ctx *sctx,
1869 struct name_cache_entry *nce)
1872 struct list_head *nce_head;
1874 nce_head = radix_tree_lookup(&sctx->name_cache,
1875 (unsigned long)nce->ino);
1877 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1882 INIT_LIST_HEAD(nce_head);
1884 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1891 list_add_tail(&nce->radix_list, nce_head);
1892 list_add_tail(&nce->list, &sctx->name_cache_list);
1893 sctx->name_cache_size++;
1898 static void name_cache_delete(struct send_ctx *sctx,
1899 struct name_cache_entry *nce)
1901 struct list_head *nce_head;
1903 nce_head = radix_tree_lookup(&sctx->name_cache,
1904 (unsigned long)nce->ino);
1907 list_del(&nce->radix_list);
1908 list_del(&nce->list);
1909 sctx->name_cache_size--;
1911 if (list_empty(nce_head)) {
1912 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1917 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1920 struct list_head *nce_head;
1921 struct name_cache_entry *cur;
1923 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1927 list_for_each_entry(cur, nce_head, radix_list) {
1928 if (cur->ino == ino && cur->gen == gen)
1935 * Removes the entry from the list and adds it back to the end. This marks the
1936 * entry as recently used so that name_cache_clean_unused does not remove it.
1938 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1940 list_del(&nce->list);
1941 list_add_tail(&nce->list, &sctx->name_cache_list);
1945 * Remove some entries from the beginning of name_cache_list.
1947 static void name_cache_clean_unused(struct send_ctx *sctx)
1949 struct name_cache_entry *nce;
1951 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1954 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1955 nce = list_entry(sctx->name_cache_list.next,
1956 struct name_cache_entry, list);
1957 name_cache_delete(sctx, nce);
1962 static void name_cache_free(struct send_ctx *sctx)
1964 struct name_cache_entry *nce;
1966 while (!list_empty(&sctx->name_cache_list)) {
1967 nce = list_entry(sctx->name_cache_list.next,
1968 struct name_cache_entry, list);
1969 name_cache_delete(sctx, nce);
1975 * Used by get_cur_path for each ref up to the root.
1976 * Returns 0 if it succeeded.
1977 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1978 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1979 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1980 * Returns <0 in case of error.
1982 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1984 int skip_name_cache,
1987 struct fs_path *dest)
1991 struct btrfs_path *path = NULL;
1992 struct name_cache_entry *nce = NULL;
1994 if (skip_name_cache)
1997 * First check if we already did a call to this function with the same
1998 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1999 * return the cached result.
2001 nce = name_cache_search(sctx, ino, gen);
2003 if (ino < sctx->send_progress && nce->need_later_update) {
2004 name_cache_delete(sctx, nce);
2008 name_cache_used(sctx, nce);
2009 *parent_ino = nce->parent_ino;
2010 *parent_gen = nce->parent_gen;
2011 ret = fs_path_add(dest, nce->name, nce->name_len);
2019 path = alloc_path_for_send();
2024 * If the inode is not existent yet, add the orphan name and return 1.
2025 * This should only happen for the parent dir that we determine in
2028 ret = is_inode_existent(sctx, ino, gen);
2033 ret = gen_unique_name(sctx, ino, gen, dest);
2042 * Depending on whether the inode was already processed or not, use
2043 * send_root or parent_root for ref lookup.
2045 if (ino < sctx->send_progress && !skip_name_cache)
2046 ret = get_first_ref(sctx->send_root, ino,
2047 parent_ino, parent_gen, dest);
2049 ret = get_first_ref(sctx->parent_root, ino,
2050 parent_ino, parent_gen, dest);
2055 * Check if the ref was overwritten by an inode's ref that was processed
2056 * earlier. If yes, treat as orphan and return 1.
2058 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2059 dest->start, dest->end - dest->start);
2063 fs_path_reset(dest);
2064 ret = gen_unique_name(sctx, ino, gen, dest);
2069 if (skip_name_cache)
2074 * Store the result of the lookup in the name cache.
2076 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2084 nce->parent_ino = *parent_ino;
2085 nce->parent_gen = *parent_gen;
2086 nce->name_len = fs_path_len(dest);
2088 strcpy(nce->name, dest->start);
2090 if (ino < sctx->send_progress)
2091 nce->need_later_update = 0;
2093 nce->need_later_update = 1;
2095 nce_ret = name_cache_insert(sctx, nce);
2098 name_cache_clean_unused(sctx);
2101 btrfs_free_path(path);
2106 * Magic happens here. This function returns the first ref to an inode as it
2107 * would look like while receiving the stream at this point in time.
2108 * We walk the path up to the root. For every inode in between, we check if it
2109 * was already processed/sent. If yes, we continue with the parent as found
2110 * in send_root. If not, we continue with the parent as found in parent_root.
2111 * If we encounter an inode that was deleted at this point in time, we use the
2112 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2113 * that were not created yet and overwritten inodes/refs.
2115 * When do we have have orphan inodes:
2116 * 1. When an inode is freshly created and thus no valid refs are available yet
2117 * 2. When a directory lost all it's refs (deleted) but still has dir items
2118 * inside which were not processed yet (pending for move/delete). If anyone
2119 * tried to get the path to the dir items, it would get a path inside that
2121 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2122 * of an unprocessed inode. If in that case the first ref would be
2123 * overwritten, the overwritten inode gets "orphanized". Later when we
2124 * process this overwritten inode, it is restored at a new place by moving
2127 * sctx->send_progress tells this function at which point in time receiving
2130 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2131 struct fs_path *dest)
2134 struct fs_path *name = NULL;
2135 u64 parent_inode = 0;
2138 u64 start_ino = ino;
2139 u64 start_gen = gen;
2140 int skip_name_cache = 0;
2142 name = fs_path_alloc();
2148 if (is_waiting_for_move(sctx, ino))
2149 skip_name_cache = 1;
2153 fs_path_reset(dest);
2155 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2156 fs_path_reset(name);
2158 ret = __get_cur_name_and_parent(sctx, ino, gen, skip_name_cache,
2159 &parent_inode, &parent_gen, name);
2165 if (!skip_name_cache &&
2166 is_waiting_for_move(sctx, parent_inode)) {
2170 skip_name_cache = 1;
2174 ret = fs_path_add_path(dest, name);
2185 fs_path_unreverse(dest);
2190 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2192 static int send_subvol_begin(struct send_ctx *sctx)
2195 struct btrfs_root *send_root = sctx->send_root;
2196 struct btrfs_root *parent_root = sctx->parent_root;
2197 struct btrfs_path *path;
2198 struct btrfs_key key;
2199 struct btrfs_root_ref *ref;
2200 struct extent_buffer *leaf;
2204 path = btrfs_alloc_path();
2208 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2210 btrfs_free_path(path);
2214 key.objectid = send_root->objectid;
2215 key.type = BTRFS_ROOT_BACKREF_KEY;
2218 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2227 leaf = path->nodes[0];
2228 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2229 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2230 key.objectid != send_root->objectid) {
2234 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2235 namelen = btrfs_root_ref_name_len(leaf, ref);
2236 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2237 btrfs_release_path(path);
2240 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2244 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2249 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2250 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2251 sctx->send_root->root_item.uuid);
2252 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2253 le64_to_cpu(sctx->send_root->root_item.ctransid));
2255 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2256 sctx->parent_root->root_item.uuid);
2257 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2258 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2261 ret = send_cmd(sctx);
2265 btrfs_free_path(path);
2270 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2275 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2277 p = fs_path_alloc();
2281 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2285 ret = get_cur_path(sctx, ino, gen, p);
2288 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2289 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2291 ret = send_cmd(sctx);
2299 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2304 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2306 p = fs_path_alloc();
2310 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2314 ret = get_cur_path(sctx, ino, gen, p);
2317 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2318 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2320 ret = send_cmd(sctx);
2328 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2333 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2335 p = fs_path_alloc();
2339 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2343 ret = get_cur_path(sctx, ino, gen, p);
2346 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2347 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2348 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2350 ret = send_cmd(sctx);
2358 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2361 struct fs_path *p = NULL;
2362 struct btrfs_inode_item *ii;
2363 struct btrfs_path *path = NULL;
2364 struct extent_buffer *eb;
2365 struct btrfs_key key;
2368 verbose_printk("btrfs: send_utimes %llu\n", ino);
2370 p = fs_path_alloc();
2374 path = alloc_path_for_send();
2381 key.type = BTRFS_INODE_ITEM_KEY;
2383 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2387 eb = path->nodes[0];
2388 slot = path->slots[0];
2389 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2391 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2395 ret = get_cur_path(sctx, ino, gen, p);
2398 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2399 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2400 btrfs_inode_atime(ii));
2401 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2402 btrfs_inode_mtime(ii));
2403 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2404 btrfs_inode_ctime(ii));
2405 /* TODO Add otime support when the otime patches get into upstream */
2407 ret = send_cmd(sctx);
2412 btrfs_free_path(path);
2417 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2418 * a valid path yet because we did not process the refs yet. So, the inode
2419 * is created as orphan.
2421 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2430 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2432 p = fs_path_alloc();
2436 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2441 if (S_ISREG(mode)) {
2442 cmd = BTRFS_SEND_C_MKFILE;
2443 } else if (S_ISDIR(mode)) {
2444 cmd = BTRFS_SEND_C_MKDIR;
2445 } else if (S_ISLNK(mode)) {
2446 cmd = BTRFS_SEND_C_SYMLINK;
2447 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2448 cmd = BTRFS_SEND_C_MKNOD;
2449 } else if (S_ISFIFO(mode)) {
2450 cmd = BTRFS_SEND_C_MKFIFO;
2451 } else if (S_ISSOCK(mode)) {
2452 cmd = BTRFS_SEND_C_MKSOCK;
2454 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2455 (int)(mode & S_IFMT));
2460 ret = begin_cmd(sctx, cmd);
2464 ret = gen_unique_name(sctx, ino, gen, p);
2468 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2469 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2471 if (S_ISLNK(mode)) {
2473 ret = read_symlink(sctx->send_root, ino, p);
2476 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2477 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2478 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2479 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2480 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2483 ret = send_cmd(sctx);
2495 * We need some special handling for inodes that get processed before the parent
2496 * directory got created. See process_recorded_refs for details.
2497 * This function does the check if we already created the dir out of order.
2499 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2502 struct btrfs_path *path = NULL;
2503 struct btrfs_key key;
2504 struct btrfs_key found_key;
2505 struct btrfs_key di_key;
2506 struct extent_buffer *eb;
2507 struct btrfs_dir_item *di;
2510 path = alloc_path_for_send();
2517 key.type = BTRFS_DIR_INDEX_KEY;
2520 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2525 eb = path->nodes[0];
2526 slot = path->slots[0];
2527 btrfs_item_key_to_cpu(eb, &found_key, slot);
2529 if (ret || found_key.objectid != key.objectid ||
2530 found_key.type != key.type) {
2535 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2536 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2538 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2539 di_key.objectid < sctx->send_progress) {
2544 key.offset = found_key.offset + 1;
2545 btrfs_release_path(path);
2549 btrfs_free_path(path);
2554 * Only creates the inode if it is:
2555 * 1. Not a directory
2556 * 2. Or a directory which was not created already due to out of order
2557 * directories. See did_create_dir and process_recorded_refs for details.
2559 static int send_create_inode_if_needed(struct send_ctx *sctx)
2563 if (S_ISDIR(sctx->cur_inode_mode)) {
2564 ret = did_create_dir(sctx, sctx->cur_ino);
2573 ret = send_create_inode(sctx, sctx->cur_ino);
2581 struct recorded_ref {
2582 struct list_head list;
2585 struct fs_path *full_path;
2593 * We need to process new refs before deleted refs, but compare_tree gives us
2594 * everything mixed. So we first record all refs and later process them.
2595 * This function is a helper to record one ref.
2597 static int record_ref(struct list_head *head, u64 dir,
2598 u64 dir_gen, struct fs_path *path)
2600 struct recorded_ref *ref;
2602 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2607 ref->dir_gen = dir_gen;
2608 ref->full_path = path;
2610 ref->name = (char *)kbasename(ref->full_path->start);
2611 ref->name_len = ref->full_path->end - ref->name;
2612 ref->dir_path = ref->full_path->start;
2613 if (ref->name == ref->full_path->start)
2614 ref->dir_path_len = 0;
2616 ref->dir_path_len = ref->full_path->end -
2617 ref->full_path->start - 1 - ref->name_len;
2619 list_add_tail(&ref->list, head);
2623 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2625 struct recorded_ref *new;
2627 new = kmalloc(sizeof(*ref), GFP_NOFS);
2631 new->dir = ref->dir;
2632 new->dir_gen = ref->dir_gen;
2633 new->full_path = NULL;
2634 INIT_LIST_HEAD(&new->list);
2635 list_add_tail(&new->list, list);
2639 static void __free_recorded_refs(struct list_head *head)
2641 struct recorded_ref *cur;
2643 while (!list_empty(head)) {
2644 cur = list_entry(head->next, struct recorded_ref, list);
2645 fs_path_free(cur->full_path);
2646 list_del(&cur->list);
2651 static void free_recorded_refs(struct send_ctx *sctx)
2653 __free_recorded_refs(&sctx->new_refs);
2654 __free_recorded_refs(&sctx->deleted_refs);
2658 * Renames/moves a file/dir to its orphan name. Used when the first
2659 * ref of an unprocessed inode gets overwritten and for all non empty
2662 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2663 struct fs_path *path)
2666 struct fs_path *orphan;
2668 orphan = fs_path_alloc();
2672 ret = gen_unique_name(sctx, ino, gen, orphan);
2676 ret = send_rename(sctx, path, orphan);
2679 fs_path_free(orphan);
2684 * Returns 1 if a directory can be removed at this point in time.
2685 * We check this by iterating all dir items and checking if the inode behind
2686 * the dir item was already processed.
2688 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2691 struct btrfs_root *root = sctx->parent_root;
2692 struct btrfs_path *path;
2693 struct btrfs_key key;
2694 struct btrfs_key found_key;
2695 struct btrfs_key loc;
2696 struct btrfs_dir_item *di;
2699 * Don't try to rmdir the top/root subvolume dir.
2701 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2704 path = alloc_path_for_send();
2709 key.type = BTRFS_DIR_INDEX_KEY;
2713 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2717 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2720 if (ret || found_key.objectid != key.objectid ||
2721 found_key.type != key.type) {
2725 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2726 struct btrfs_dir_item);
2727 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2729 if (loc.objectid > send_progress) {
2734 btrfs_release_path(path);
2735 key.offset = found_key.offset + 1;
2741 btrfs_free_path(path);
2745 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2747 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2748 struct waiting_dir_move *entry;
2751 entry = rb_entry(n, struct waiting_dir_move, node);
2752 if (ino < entry->ino)
2754 else if (ino > entry->ino)
2762 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2764 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2765 struct rb_node *parent = NULL;
2766 struct waiting_dir_move *entry, *dm;
2768 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2775 entry = rb_entry(parent, struct waiting_dir_move, node);
2776 if (ino < entry->ino) {
2778 } else if (ino > entry->ino) {
2779 p = &(*p)->rb_right;
2786 rb_link_node(&dm->node, parent, p);
2787 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2791 static int del_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2793 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2794 struct waiting_dir_move *entry;
2797 entry = rb_entry(n, struct waiting_dir_move, node);
2798 if (ino < entry->ino) {
2800 } else if (ino > entry->ino) {
2803 rb_erase(&entry->node, &sctx->waiting_dir_moves);
2811 static int add_pending_dir_move(struct send_ctx *sctx, u64 parent_ino)
2813 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2814 struct rb_node *parent = NULL;
2815 struct pending_dir_move *entry, *pm;
2816 struct recorded_ref *cur;
2820 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2823 pm->parent_ino = parent_ino;
2824 pm->ino = sctx->cur_ino;
2825 pm->gen = sctx->cur_inode_gen;
2826 INIT_LIST_HEAD(&pm->list);
2827 INIT_LIST_HEAD(&pm->update_refs);
2828 RB_CLEAR_NODE(&pm->node);
2832 entry = rb_entry(parent, struct pending_dir_move, node);
2833 if (parent_ino < entry->parent_ino) {
2835 } else if (parent_ino > entry->parent_ino) {
2836 p = &(*p)->rb_right;
2843 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2844 ret = dup_ref(cur, &pm->update_refs);
2848 list_for_each_entry(cur, &sctx->new_refs, list) {
2849 ret = dup_ref(cur, &pm->update_refs);
2854 ret = add_waiting_dir_move(sctx, pm->ino);
2859 list_add_tail(&pm->list, &entry->list);
2861 rb_link_node(&pm->node, parent, p);
2862 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2867 __free_recorded_refs(&pm->update_refs);
2873 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
2876 struct rb_node *n = sctx->pending_dir_moves.rb_node;
2877 struct pending_dir_move *entry;
2880 entry = rb_entry(n, struct pending_dir_move, node);
2881 if (parent_ino < entry->parent_ino)
2883 else if (parent_ino > entry->parent_ino)
2891 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
2893 struct fs_path *from_path = NULL;
2894 struct fs_path *to_path = NULL;
2895 u64 orig_progress = sctx->send_progress;
2896 struct recorded_ref *cur;
2899 from_path = fs_path_alloc();
2903 sctx->send_progress = pm->ino;
2904 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
2908 to_path = fs_path_alloc();
2914 sctx->send_progress = sctx->cur_ino + 1;
2915 ret = del_waiting_dir_move(sctx, pm->ino);
2918 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
2922 ret = send_rename(sctx, from_path, to_path);
2926 ret = send_utimes(sctx, pm->ino, pm->gen);
2931 * After rename/move, need to update the utimes of both new parent(s)
2932 * and old parent(s).
2934 list_for_each_entry(cur, &pm->update_refs, list) {
2935 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
2941 fs_path_free(from_path);
2942 fs_path_free(to_path);
2943 sctx->send_progress = orig_progress;
2948 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
2950 if (!list_empty(&m->list))
2952 if (!RB_EMPTY_NODE(&m->node))
2953 rb_erase(&m->node, &sctx->pending_dir_moves);
2954 __free_recorded_refs(&m->update_refs);
2958 static void tail_append_pending_moves(struct pending_dir_move *moves,
2959 struct list_head *stack)
2961 if (list_empty(&moves->list)) {
2962 list_add_tail(&moves->list, stack);
2965 list_splice_init(&moves->list, &list);
2966 list_add_tail(&moves->list, stack);
2967 list_splice_tail(&list, stack);
2971 static int apply_children_dir_moves(struct send_ctx *sctx)
2973 struct pending_dir_move *pm;
2974 struct list_head stack;
2975 u64 parent_ino = sctx->cur_ino;
2978 pm = get_pending_dir_moves(sctx, parent_ino);
2982 INIT_LIST_HEAD(&stack);
2983 tail_append_pending_moves(pm, &stack);
2985 while (!list_empty(&stack)) {
2986 pm = list_first_entry(&stack, struct pending_dir_move, list);
2987 parent_ino = pm->ino;
2988 ret = apply_dir_move(sctx, pm);
2989 free_pending_move(sctx, pm);
2992 pm = get_pending_dir_moves(sctx, parent_ino);
2994 tail_append_pending_moves(pm, &stack);
2999 while (!list_empty(&stack)) {
3000 pm = list_first_entry(&stack, struct pending_dir_move, list);
3001 free_pending_move(sctx, pm);
3006 static int wait_for_parent_move(struct send_ctx *sctx,
3007 struct recorded_ref *parent_ref)
3010 u64 ino = parent_ref->dir;
3011 u64 parent_ino_before, parent_ino_after;
3012 u64 new_gen, old_gen;
3013 struct fs_path *path_before = NULL;
3014 struct fs_path *path_after = NULL;
3017 if (parent_ref->dir <= sctx->cur_ino)
3020 if (is_waiting_for_move(sctx, ino))
3023 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3024 NULL, NULL, NULL, NULL);
3030 ret = get_inode_info(sctx->send_root, ino, NULL, &new_gen,
3031 NULL, NULL, NULL, NULL);
3035 if (new_gen != old_gen)
3038 path_before = fs_path_alloc();
3042 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3044 if (ret == -ENOENT) {
3047 } else if (ret < 0) {
3051 path_after = fs_path_alloc();
3057 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3059 if (ret == -ENOENT) {
3062 } else if (ret < 0) {
3066 len1 = fs_path_len(path_before);
3067 len2 = fs_path_len(path_after);
3068 if ((parent_ino_before != parent_ino_after) && (len1 != len2 ||
3069 memcmp(path_before->start, path_after->start, len1))) {
3076 fs_path_free(path_before);
3077 fs_path_free(path_after);
3083 * This does all the move/link/unlink/rmdir magic.
3085 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3088 struct recorded_ref *cur;
3089 struct recorded_ref *cur2;
3090 struct list_head check_dirs;
3091 struct fs_path *valid_path = NULL;
3094 int did_overwrite = 0;
3097 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3100 * This should never happen as the root dir always has the same ref
3101 * which is always '..'
3103 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3104 INIT_LIST_HEAD(&check_dirs);
3106 valid_path = fs_path_alloc();
3113 * First, check if the first ref of the current inode was overwritten
3114 * before. If yes, we know that the current inode was already orphanized
3115 * and thus use the orphan name. If not, we can use get_cur_path to
3116 * get the path of the first ref as it would like while receiving at
3117 * this point in time.
3118 * New inodes are always orphan at the beginning, so force to use the
3119 * orphan name in this case.
3120 * The first ref is stored in valid_path and will be updated if it
3121 * gets moved around.
3123 if (!sctx->cur_inode_new) {
3124 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3125 sctx->cur_inode_gen);
3131 if (sctx->cur_inode_new || did_overwrite) {
3132 ret = gen_unique_name(sctx, sctx->cur_ino,
3133 sctx->cur_inode_gen, valid_path);
3138 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3144 list_for_each_entry(cur, &sctx->new_refs, list) {
3146 * We may have refs where the parent directory does not exist
3147 * yet. This happens if the parent directories inum is higher
3148 * the the current inum. To handle this case, we create the
3149 * parent directory out of order. But we need to check if this
3150 * did already happen before due to other refs in the same dir.
3152 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3155 if (ret == inode_state_will_create) {
3158 * First check if any of the current inodes refs did
3159 * already create the dir.
3161 list_for_each_entry(cur2, &sctx->new_refs, list) {
3164 if (cur2->dir == cur->dir) {
3171 * If that did not happen, check if a previous inode
3172 * did already create the dir.
3175 ret = did_create_dir(sctx, cur->dir);
3179 ret = send_create_inode(sctx, cur->dir);
3186 * Check if this new ref would overwrite the first ref of
3187 * another unprocessed inode. If yes, orphanize the
3188 * overwritten inode. If we find an overwritten ref that is
3189 * not the first ref, simply unlink it.
3191 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3192 cur->name, cur->name_len,
3193 &ow_inode, &ow_gen);
3197 ret = is_first_ref(sctx->parent_root,
3198 ow_inode, cur->dir, cur->name,
3203 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3208 ret = send_unlink(sctx, cur->full_path);
3215 * link/move the ref to the new place. If we have an orphan
3216 * inode, move it and update valid_path. If not, link or move
3217 * it depending on the inode mode.
3220 ret = send_rename(sctx, valid_path, cur->full_path);
3224 ret = fs_path_copy(valid_path, cur->full_path);
3228 if (S_ISDIR(sctx->cur_inode_mode)) {
3230 * Dirs can't be linked, so move it. For moved
3231 * dirs, we always have one new and one deleted
3232 * ref. The deleted ref is ignored later.
3234 if (wait_for_parent_move(sctx, cur)) {
3235 ret = add_pending_dir_move(sctx,
3239 ret = send_rename(sctx, valid_path,
3242 ret = fs_path_copy(valid_path,
3248 ret = send_link(sctx, cur->full_path,
3254 ret = dup_ref(cur, &check_dirs);
3259 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3261 * Check if we can already rmdir the directory. If not,
3262 * orphanize it. For every dir item inside that gets deleted
3263 * later, we do this check again and rmdir it then if possible.
3264 * See the use of check_dirs for more details.
3266 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
3270 ret = send_rmdir(sctx, valid_path);
3273 } else if (!is_orphan) {
3274 ret = orphanize_inode(sctx, sctx->cur_ino,
3275 sctx->cur_inode_gen, valid_path);
3281 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3282 ret = dup_ref(cur, &check_dirs);
3286 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3287 !list_empty(&sctx->deleted_refs)) {
3289 * We have a moved dir. Add the old parent to check_dirs
3291 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3293 ret = dup_ref(cur, &check_dirs);
3296 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3298 * We have a non dir inode. Go through all deleted refs and
3299 * unlink them if they were not already overwritten by other
3302 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3303 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3304 sctx->cur_ino, sctx->cur_inode_gen,
3305 cur->name, cur->name_len);
3309 ret = send_unlink(sctx, cur->full_path);
3313 ret = dup_ref(cur, &check_dirs);
3318 * If the inode is still orphan, unlink the orphan. This may
3319 * happen when a previous inode did overwrite the first ref
3320 * of this inode and no new refs were added for the current
3321 * inode. Unlinking does not mean that the inode is deleted in
3322 * all cases. There may still be links to this inode in other
3326 ret = send_unlink(sctx, valid_path);
3333 * We did collect all parent dirs where cur_inode was once located. We
3334 * now go through all these dirs and check if they are pending for
3335 * deletion and if it's finally possible to perform the rmdir now.
3336 * We also update the inode stats of the parent dirs here.
3338 list_for_each_entry(cur, &check_dirs, list) {
3340 * In case we had refs into dirs that were not processed yet,
3341 * we don't need to do the utime and rmdir logic for these dirs.
3342 * The dir will be processed later.
3344 if (cur->dir > sctx->cur_ino)
3347 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3351 if (ret == inode_state_did_create ||
3352 ret == inode_state_no_change) {
3353 /* TODO delayed utimes */
3354 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3357 } else if (ret == inode_state_did_delete) {
3358 ret = can_rmdir(sctx, cur->dir, sctx->cur_ino);
3362 ret = get_cur_path(sctx, cur->dir,
3363 cur->dir_gen, valid_path);
3366 ret = send_rmdir(sctx, valid_path);
3376 __free_recorded_refs(&check_dirs);
3377 free_recorded_refs(sctx);
3378 fs_path_free(valid_path);
3382 static int __record_new_ref(int num, u64 dir, int index,
3383 struct fs_path *name,
3387 struct send_ctx *sctx = ctx;
3391 p = fs_path_alloc();
3395 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3400 ret = get_cur_path(sctx, dir, gen, p);
3403 ret = fs_path_add_path(p, name);
3407 ret = record_ref(&sctx->new_refs, dir, gen, p);
3415 static int __record_deleted_ref(int num, u64 dir, int index,
3416 struct fs_path *name,
3420 struct send_ctx *sctx = ctx;
3424 p = fs_path_alloc();
3428 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3433 ret = get_cur_path(sctx, dir, gen, p);
3436 ret = fs_path_add_path(p, name);
3440 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3448 static int record_new_ref(struct send_ctx *sctx)
3452 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3453 sctx->cmp_key, 0, __record_new_ref, sctx);
3462 static int record_deleted_ref(struct send_ctx *sctx)
3466 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3467 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3476 struct find_ref_ctx {
3479 struct btrfs_root *root;
3480 struct fs_path *name;
3484 static int __find_iref(int num, u64 dir, int index,
3485 struct fs_path *name,
3488 struct find_ref_ctx *ctx = ctx_;
3492 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3493 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3495 * To avoid doing extra lookups we'll only do this if everything
3498 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3502 if (dir_gen != ctx->dir_gen)
3504 ctx->found_idx = num;
3510 static int find_iref(struct btrfs_root *root,
3511 struct btrfs_path *path,
3512 struct btrfs_key *key,
3513 u64 dir, u64 dir_gen, struct fs_path *name)
3516 struct find_ref_ctx ctx;
3520 ctx.dir_gen = dir_gen;
3524 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3528 if (ctx.found_idx == -1)
3531 return ctx.found_idx;
3534 static int __record_changed_new_ref(int num, u64 dir, int index,
3535 struct fs_path *name,
3540 struct send_ctx *sctx = ctx;
3542 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3547 ret = find_iref(sctx->parent_root, sctx->right_path,
3548 sctx->cmp_key, dir, dir_gen, name);
3550 ret = __record_new_ref(num, dir, index, name, sctx);
3557 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3558 struct fs_path *name,
3563 struct send_ctx *sctx = ctx;
3565 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3570 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3571 dir, dir_gen, name);
3573 ret = __record_deleted_ref(num, dir, index, name, sctx);
3580 static int record_changed_ref(struct send_ctx *sctx)
3584 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3585 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3588 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3589 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3599 * Record and process all refs at once. Needed when an inode changes the
3600 * generation number, which means that it was deleted and recreated.
3602 static int process_all_refs(struct send_ctx *sctx,
3603 enum btrfs_compare_tree_result cmd)
3606 struct btrfs_root *root;
3607 struct btrfs_path *path;
3608 struct btrfs_key key;
3609 struct btrfs_key found_key;
3610 struct extent_buffer *eb;
3612 iterate_inode_ref_t cb;
3613 int pending_move = 0;
3615 path = alloc_path_for_send();
3619 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3620 root = sctx->send_root;
3621 cb = __record_new_ref;
3622 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3623 root = sctx->parent_root;
3624 cb = __record_deleted_ref;
3629 key.objectid = sctx->cmp_key->objectid;
3630 key.type = BTRFS_INODE_REF_KEY;
3633 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3639 eb = path->nodes[0];
3640 slot = path->slots[0];
3641 btrfs_item_key_to_cpu(eb, &found_key, slot);
3643 if (found_key.objectid != key.objectid ||
3644 (found_key.type != BTRFS_INODE_REF_KEY &&
3645 found_key.type != BTRFS_INODE_EXTREF_KEY))
3648 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3649 btrfs_release_path(path);
3653 key.offset = found_key.offset + 1;
3655 btrfs_release_path(path);
3657 ret = process_recorded_refs(sctx, &pending_move);
3658 /* Only applicable to an incremental send. */
3659 ASSERT(pending_move == 0);
3662 btrfs_free_path(path);
3666 static int send_set_xattr(struct send_ctx *sctx,
3667 struct fs_path *path,
3668 const char *name, int name_len,
3669 const char *data, int data_len)
3673 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3677 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3678 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3679 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3681 ret = send_cmd(sctx);
3688 static int send_remove_xattr(struct send_ctx *sctx,
3689 struct fs_path *path,
3690 const char *name, int name_len)
3694 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3698 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3699 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3701 ret = send_cmd(sctx);
3708 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3709 const char *name, int name_len,
3710 const char *data, int data_len,
3714 struct send_ctx *sctx = ctx;
3716 posix_acl_xattr_header dummy_acl;
3718 p = fs_path_alloc();
3723 * This hack is needed because empty acl's are stored as zero byte
3724 * data in xattrs. Problem with that is, that receiving these zero byte
3725 * acl's will fail later. To fix this, we send a dummy acl list that
3726 * only contains the version number and no entries.
3728 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3729 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3730 if (data_len == 0) {
3731 dummy_acl.a_version =
3732 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3733 data = (char *)&dummy_acl;
3734 data_len = sizeof(dummy_acl);
3738 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3742 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3749 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3750 const char *name, int name_len,
3751 const char *data, int data_len,
3755 struct send_ctx *sctx = ctx;
3758 p = fs_path_alloc();
3762 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3766 ret = send_remove_xattr(sctx, p, name, name_len);
3773 static int process_new_xattr(struct send_ctx *sctx)
3777 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3778 sctx->cmp_key, __process_new_xattr, sctx);
3783 static int process_deleted_xattr(struct send_ctx *sctx)
3787 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3788 sctx->cmp_key, __process_deleted_xattr, sctx);
3793 struct find_xattr_ctx {
3801 static int __find_xattr(int num, struct btrfs_key *di_key,
3802 const char *name, int name_len,
3803 const char *data, int data_len,
3804 u8 type, void *vctx)
3806 struct find_xattr_ctx *ctx = vctx;
3808 if (name_len == ctx->name_len &&
3809 strncmp(name, ctx->name, name_len) == 0) {
3810 ctx->found_idx = num;
3811 ctx->found_data_len = data_len;
3812 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
3813 if (!ctx->found_data)
3820 static int find_xattr(struct btrfs_root *root,
3821 struct btrfs_path *path,
3822 struct btrfs_key *key,
3823 const char *name, int name_len,
3824 char **data, int *data_len)
3827 struct find_xattr_ctx ctx;
3830 ctx.name_len = name_len;
3832 ctx.found_data = NULL;
3833 ctx.found_data_len = 0;
3835 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
3839 if (ctx.found_idx == -1)
3842 *data = ctx.found_data;
3843 *data_len = ctx.found_data_len;
3845 kfree(ctx.found_data);
3847 return ctx.found_idx;
3851 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3852 const char *name, int name_len,
3853 const char *data, int data_len,
3857 struct send_ctx *sctx = ctx;
3858 char *found_data = NULL;
3859 int found_data_len = 0;
3861 ret = find_xattr(sctx->parent_root, sctx->right_path,
3862 sctx->cmp_key, name, name_len, &found_data,
3864 if (ret == -ENOENT) {
3865 ret = __process_new_xattr(num, di_key, name, name_len, data,
3866 data_len, type, ctx);
3867 } else if (ret >= 0) {
3868 if (data_len != found_data_len ||
3869 memcmp(data, found_data, data_len)) {
3870 ret = __process_new_xattr(num, di_key, name, name_len,
3871 data, data_len, type, ctx);
3881 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3882 const char *name, int name_len,
3883 const char *data, int data_len,
3887 struct send_ctx *sctx = ctx;
3889 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
3890 name, name_len, NULL, NULL);
3892 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3893 data_len, type, ctx);
3900 static int process_changed_xattr(struct send_ctx *sctx)
3904 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
3905 sctx->cmp_key, __process_changed_new_xattr, sctx);
3908 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
3909 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3915 static int process_all_new_xattrs(struct send_ctx *sctx)
3918 struct btrfs_root *root;
3919 struct btrfs_path *path;
3920 struct btrfs_key key;
3921 struct btrfs_key found_key;
3922 struct extent_buffer *eb;
3925 path = alloc_path_for_send();
3929 root = sctx->send_root;
3931 key.objectid = sctx->cmp_key->objectid;
3932 key.type = BTRFS_XATTR_ITEM_KEY;
3935 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3943 eb = path->nodes[0];
3944 slot = path->slots[0];
3945 btrfs_item_key_to_cpu(eb, &found_key, slot);
3947 if (found_key.objectid != key.objectid ||
3948 found_key.type != key.type) {
3953 ret = iterate_dir_item(root, path, &found_key,
3954 __process_new_xattr, sctx);
3958 btrfs_release_path(path);
3959 key.offset = found_key.offset + 1;
3963 btrfs_free_path(path);
3967 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
3969 struct btrfs_root *root = sctx->send_root;
3970 struct btrfs_fs_info *fs_info = root->fs_info;
3971 struct inode *inode;
3974 struct btrfs_key key;
3975 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
3977 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
3980 key.objectid = sctx->cur_ino;
3981 key.type = BTRFS_INODE_ITEM_KEY;
3984 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3986 return PTR_ERR(inode);
3988 if (offset + len > i_size_read(inode)) {
3989 if (offset > i_size_read(inode))
3992 len = offset - i_size_read(inode);
3997 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
3998 while (index <= last_index) {
3999 unsigned cur_len = min_t(unsigned, len,
4000 PAGE_CACHE_SIZE - pg_offset);
4001 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4007 if (!PageUptodate(page)) {
4008 btrfs_readpage(NULL, page);
4010 if (!PageUptodate(page)) {
4012 page_cache_release(page);
4019 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4022 page_cache_release(page);
4034 * Read some bytes from the current inode/file and send a write command to
4037 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4041 ssize_t num_read = 0;
4043 p = fs_path_alloc();
4047 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4049 num_read = fill_read_buf(sctx, offset, len);
4050 if (num_read <= 0) {
4056 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4060 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4064 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4065 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4066 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4068 ret = send_cmd(sctx);
4079 * Send a clone command to user space.
4081 static int send_clone(struct send_ctx *sctx,
4082 u64 offset, u32 len,
4083 struct clone_root *clone_root)
4089 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4090 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4091 clone_root->root->objectid, clone_root->ino,
4092 clone_root->offset);
4094 p = fs_path_alloc();
4098 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4102 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4106 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4107 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4108 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4110 if (clone_root->root == sctx->send_root) {
4111 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4112 &gen, NULL, NULL, NULL, NULL);
4115 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4117 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4122 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4123 clone_root->root->root_item.uuid);
4124 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4125 le64_to_cpu(clone_root->root->root_item.ctransid));
4126 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4127 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4128 clone_root->offset);
4130 ret = send_cmd(sctx);
4139 * Send an update extent command to user space.
4141 static int send_update_extent(struct send_ctx *sctx,
4142 u64 offset, u32 len)
4147 p = fs_path_alloc();
4151 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4155 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4159 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4160 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4161 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4163 ret = send_cmd(sctx);
4171 static int send_hole(struct send_ctx *sctx, u64 end)
4173 struct fs_path *p = NULL;
4174 u64 offset = sctx->cur_inode_last_extent;
4178 p = fs_path_alloc();
4181 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4182 while (offset < end) {
4183 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4185 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4188 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4191 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4192 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4193 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4194 ret = send_cmd(sctx);
4204 static int send_write_or_clone(struct send_ctx *sctx,
4205 struct btrfs_path *path,
4206 struct btrfs_key *key,
4207 struct clone_root *clone_root)
4210 struct btrfs_file_extent_item *ei;
4211 u64 offset = key->offset;
4216 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4218 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4219 struct btrfs_file_extent_item);
4220 type = btrfs_file_extent_type(path->nodes[0], ei);
4221 if (type == BTRFS_FILE_EXTENT_INLINE) {
4222 len = btrfs_file_extent_inline_len(path->nodes[0],
4223 path->slots[0], ei);
4225 * it is possible the inline item won't cover the whole page,
4226 * but there may be items after this page. Make
4227 * sure to send the whole thing
4229 len = PAGE_CACHE_ALIGN(len);
4231 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4234 if (offset + len > sctx->cur_inode_size)
4235 len = sctx->cur_inode_size - offset;
4241 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4242 ret = send_clone(sctx, offset, len, clone_root);
4243 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4244 ret = send_update_extent(sctx, offset, len);
4248 if (l > BTRFS_SEND_READ_SIZE)
4249 l = BTRFS_SEND_READ_SIZE;
4250 ret = send_write(sctx, pos + offset, l);
4263 static int is_extent_unchanged(struct send_ctx *sctx,
4264 struct btrfs_path *left_path,
4265 struct btrfs_key *ekey)
4268 struct btrfs_key key;
4269 struct btrfs_path *path = NULL;
4270 struct extent_buffer *eb;
4272 struct btrfs_key found_key;
4273 struct btrfs_file_extent_item *ei;
4278 u64 left_offset_fixed;
4286 path = alloc_path_for_send();
4290 eb = left_path->nodes[0];
4291 slot = left_path->slots[0];
4292 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4293 left_type = btrfs_file_extent_type(eb, ei);
4295 if (left_type != BTRFS_FILE_EXTENT_REG) {
4299 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4300 left_len = btrfs_file_extent_num_bytes(eb, ei);
4301 left_offset = btrfs_file_extent_offset(eb, ei);
4302 left_gen = btrfs_file_extent_generation(eb, ei);
4305 * Following comments will refer to these graphics. L is the left
4306 * extents which we are checking at the moment. 1-8 are the right
4307 * extents that we iterate.
4310 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4313 * |--1--|-2b-|...(same as above)
4315 * Alternative situation. Happens on files where extents got split.
4317 * |-----------7-----------|-6-|
4319 * Alternative situation. Happens on files which got larger.
4322 * Nothing follows after 8.
4325 key.objectid = ekey->objectid;
4326 key.type = BTRFS_EXTENT_DATA_KEY;
4327 key.offset = ekey->offset;
4328 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4337 * Handle special case where the right side has no extents at all.
4339 eb = path->nodes[0];
4340 slot = path->slots[0];
4341 btrfs_item_key_to_cpu(eb, &found_key, slot);
4342 if (found_key.objectid != key.objectid ||
4343 found_key.type != key.type) {
4344 /* If we're a hole then just pretend nothing changed */
4345 ret = (left_disknr) ? 0 : 1;
4350 * We're now on 2a, 2b or 7.
4353 while (key.offset < ekey->offset + left_len) {
4354 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4355 right_type = btrfs_file_extent_type(eb, ei);
4356 if (right_type != BTRFS_FILE_EXTENT_REG) {
4361 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4362 right_len = btrfs_file_extent_num_bytes(eb, ei);
4363 right_offset = btrfs_file_extent_offset(eb, ei);
4364 right_gen = btrfs_file_extent_generation(eb, ei);
4367 * Are we at extent 8? If yes, we know the extent is changed.
4368 * This may only happen on the first iteration.
4370 if (found_key.offset + right_len <= ekey->offset) {
4371 /* If we're a hole just pretend nothing changed */
4372 ret = (left_disknr) ? 0 : 1;
4376 left_offset_fixed = left_offset;
4377 if (key.offset < ekey->offset) {
4378 /* Fix the right offset for 2a and 7. */
4379 right_offset += ekey->offset - key.offset;
4381 /* Fix the left offset for all behind 2a and 2b */
4382 left_offset_fixed += key.offset - ekey->offset;
4386 * Check if we have the same extent.
4388 if (left_disknr != right_disknr ||
4389 left_offset_fixed != right_offset ||
4390 left_gen != right_gen) {
4396 * Go to the next extent.
4398 ret = btrfs_next_item(sctx->parent_root, path);
4402 eb = path->nodes[0];
4403 slot = path->slots[0];
4404 btrfs_item_key_to_cpu(eb, &found_key, slot);
4406 if (ret || found_key.objectid != key.objectid ||
4407 found_key.type != key.type) {
4408 key.offset += right_len;
4411 if (found_key.offset != key.offset + right_len) {
4419 * We're now behind the left extent (treat as unchanged) or at the end
4420 * of the right side (treat as changed).
4422 if (key.offset >= ekey->offset + left_len)
4429 btrfs_free_path(path);
4433 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4435 struct btrfs_path *path;
4436 struct btrfs_root *root = sctx->send_root;
4437 struct btrfs_file_extent_item *fi;
4438 struct btrfs_key key;
4443 path = alloc_path_for_send();
4447 sctx->cur_inode_last_extent = 0;
4449 key.objectid = sctx->cur_ino;
4450 key.type = BTRFS_EXTENT_DATA_KEY;
4451 key.offset = offset;
4452 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4456 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4457 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4460 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4461 struct btrfs_file_extent_item);
4462 type = btrfs_file_extent_type(path->nodes[0], fi);
4463 if (type == BTRFS_FILE_EXTENT_INLINE) {
4464 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4465 path->slots[0], fi);
4466 extent_end = ALIGN(key.offset + size,
4467 sctx->send_root->sectorsize);
4469 extent_end = key.offset +
4470 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4472 sctx->cur_inode_last_extent = extent_end;
4474 btrfs_free_path(path);
4478 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4479 struct btrfs_key *key)
4481 struct btrfs_file_extent_item *fi;
4486 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4489 if (sctx->cur_inode_last_extent == (u64)-1) {
4490 ret = get_last_extent(sctx, key->offset - 1);
4495 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4496 struct btrfs_file_extent_item);
4497 type = btrfs_file_extent_type(path->nodes[0], fi);
4498 if (type == BTRFS_FILE_EXTENT_INLINE) {
4499 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4500 path->slots[0], fi);
4501 extent_end = ALIGN(key->offset + size,
4502 sctx->send_root->sectorsize);
4504 extent_end = key->offset +
4505 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4508 if (path->slots[0] == 0 &&
4509 sctx->cur_inode_last_extent < key->offset) {
4511 * We might have skipped entire leafs that contained only
4512 * file extent items for our current inode. These leafs have
4513 * a generation number smaller (older) than the one in the
4514 * current leaf and the leaf our last extent came from, and
4515 * are located between these 2 leafs.
4517 ret = get_last_extent(sctx, key->offset - 1);
4522 if (sctx->cur_inode_last_extent < key->offset)
4523 ret = send_hole(sctx, key->offset);
4524 sctx->cur_inode_last_extent = extent_end;
4528 static int process_extent(struct send_ctx *sctx,
4529 struct btrfs_path *path,
4530 struct btrfs_key *key)
4532 struct clone_root *found_clone = NULL;
4535 if (S_ISLNK(sctx->cur_inode_mode))
4538 if (sctx->parent_root && !sctx->cur_inode_new) {
4539 ret = is_extent_unchanged(sctx, path, key);
4547 struct btrfs_file_extent_item *ei;
4550 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4551 struct btrfs_file_extent_item);
4552 type = btrfs_file_extent_type(path->nodes[0], ei);
4553 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4554 type == BTRFS_FILE_EXTENT_REG) {
4556 * The send spec does not have a prealloc command yet,
4557 * so just leave a hole for prealloc'ed extents until
4558 * we have enough commands queued up to justify rev'ing
4561 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4566 /* Have a hole, just skip it. */
4567 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4574 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4575 sctx->cur_inode_size, &found_clone);
4576 if (ret != -ENOENT && ret < 0)
4579 ret = send_write_or_clone(sctx, path, key, found_clone);
4583 ret = maybe_send_hole(sctx, path, key);
4588 static int process_all_extents(struct send_ctx *sctx)
4591 struct btrfs_root *root;
4592 struct btrfs_path *path;
4593 struct btrfs_key key;
4594 struct btrfs_key found_key;
4595 struct extent_buffer *eb;
4598 root = sctx->send_root;
4599 path = alloc_path_for_send();
4603 key.objectid = sctx->cmp_key->objectid;
4604 key.type = BTRFS_EXTENT_DATA_KEY;
4606 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4611 eb = path->nodes[0];
4612 slot = path->slots[0];
4614 if (slot >= btrfs_header_nritems(eb)) {
4615 ret = btrfs_next_leaf(root, path);
4618 } else if (ret > 0) {
4625 btrfs_item_key_to_cpu(eb, &found_key, slot);
4627 if (found_key.objectid != key.objectid ||
4628 found_key.type != key.type) {
4633 ret = process_extent(sctx, path, &found_key);
4641 btrfs_free_path(path);
4645 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4647 int *refs_processed)
4651 if (sctx->cur_ino == 0)
4653 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4654 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4656 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4659 ret = process_recorded_refs(sctx, pending_move);
4663 *refs_processed = 1;
4668 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4679 int pending_move = 0;
4680 int refs_processed = 0;
4682 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4688 * We have processed the refs and thus need to advance send_progress.
4689 * Now, calls to get_cur_xxx will take the updated refs of the current
4690 * inode into account.
4692 * On the other hand, if our current inode is a directory and couldn't
4693 * be moved/renamed because its parent was renamed/moved too and it has
4694 * a higher inode number, we can only move/rename our current inode
4695 * after we moved/renamed its parent. Therefore in this case operate on
4696 * the old path (pre move/rename) of our current inode, and the
4697 * move/rename will be performed later.
4699 if (refs_processed && !pending_move)
4700 sctx->send_progress = sctx->cur_ino + 1;
4702 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4704 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4707 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4708 &left_mode, &left_uid, &left_gid, NULL);
4712 if (!sctx->parent_root || sctx->cur_inode_new) {
4714 if (!S_ISLNK(sctx->cur_inode_mode))
4717 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4718 NULL, NULL, &right_mode, &right_uid,
4723 if (left_uid != right_uid || left_gid != right_gid)
4725 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4729 if (S_ISREG(sctx->cur_inode_mode)) {
4730 if (need_send_hole(sctx)) {
4731 if (sctx->cur_inode_last_extent == (u64)-1) {
4732 ret = get_last_extent(sctx, (u64)-1);
4736 if (sctx->cur_inode_last_extent <
4737 sctx->cur_inode_size) {
4738 ret = send_hole(sctx, sctx->cur_inode_size);
4743 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4744 sctx->cur_inode_size);
4750 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4751 left_uid, left_gid);
4756 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4763 * If other directory inodes depended on our current directory
4764 * inode's move/rename, now do their move/rename operations.
4766 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
4767 ret = apply_children_dir_moves(sctx);
4773 * Need to send that every time, no matter if it actually
4774 * changed between the two trees as we have done changes to
4777 sctx->send_progress = sctx->cur_ino + 1;
4778 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4786 static int changed_inode(struct send_ctx *sctx,
4787 enum btrfs_compare_tree_result result)
4790 struct btrfs_key *key = sctx->cmp_key;
4791 struct btrfs_inode_item *left_ii = NULL;
4792 struct btrfs_inode_item *right_ii = NULL;
4796 sctx->cur_ino = key->objectid;
4797 sctx->cur_inode_new_gen = 0;
4798 sctx->cur_inode_last_extent = (u64)-1;
4801 * Set send_progress to current inode. This will tell all get_cur_xxx
4802 * functions that the current inode's refs are not updated yet. Later,
4803 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4805 sctx->send_progress = sctx->cur_ino;
4807 if (result == BTRFS_COMPARE_TREE_NEW ||
4808 result == BTRFS_COMPARE_TREE_CHANGED) {
4809 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4810 sctx->left_path->slots[0],
4811 struct btrfs_inode_item);
4812 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4815 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4816 sctx->right_path->slots[0],
4817 struct btrfs_inode_item);
4818 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4821 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4822 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4823 sctx->right_path->slots[0],
4824 struct btrfs_inode_item);
4826 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4830 * The cur_ino = root dir case is special here. We can't treat
4831 * the inode as deleted+reused because it would generate a
4832 * stream that tries to delete/mkdir the root dir.
4834 if (left_gen != right_gen &&
4835 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4836 sctx->cur_inode_new_gen = 1;
4839 if (result == BTRFS_COMPARE_TREE_NEW) {
4840 sctx->cur_inode_gen = left_gen;
4841 sctx->cur_inode_new = 1;
4842 sctx->cur_inode_deleted = 0;
4843 sctx->cur_inode_size = btrfs_inode_size(
4844 sctx->left_path->nodes[0], left_ii);
4845 sctx->cur_inode_mode = btrfs_inode_mode(
4846 sctx->left_path->nodes[0], left_ii);
4847 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4848 ret = send_create_inode_if_needed(sctx);
4849 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4850 sctx->cur_inode_gen = right_gen;
4851 sctx->cur_inode_new = 0;
4852 sctx->cur_inode_deleted = 1;
4853 sctx->cur_inode_size = btrfs_inode_size(
4854 sctx->right_path->nodes[0], right_ii);
4855 sctx->cur_inode_mode = btrfs_inode_mode(
4856 sctx->right_path->nodes[0], right_ii);
4857 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4859 * We need to do some special handling in case the inode was
4860 * reported as changed with a changed generation number. This
4861 * means that the original inode was deleted and new inode
4862 * reused the same inum. So we have to treat the old inode as
4863 * deleted and the new one as new.
4865 if (sctx->cur_inode_new_gen) {
4867 * First, process the inode as if it was deleted.
4869 sctx->cur_inode_gen = right_gen;
4870 sctx->cur_inode_new = 0;
4871 sctx->cur_inode_deleted = 1;
4872 sctx->cur_inode_size = btrfs_inode_size(
4873 sctx->right_path->nodes[0], right_ii);
4874 sctx->cur_inode_mode = btrfs_inode_mode(
4875 sctx->right_path->nodes[0], right_ii);
4876 ret = process_all_refs(sctx,
4877 BTRFS_COMPARE_TREE_DELETED);
4882 * Now process the inode as if it was new.
4884 sctx->cur_inode_gen = left_gen;
4885 sctx->cur_inode_new = 1;
4886 sctx->cur_inode_deleted = 0;
4887 sctx->cur_inode_size = btrfs_inode_size(
4888 sctx->left_path->nodes[0], left_ii);
4889 sctx->cur_inode_mode = btrfs_inode_mode(
4890 sctx->left_path->nodes[0], left_ii);
4891 ret = send_create_inode_if_needed(sctx);
4895 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4899 * Advance send_progress now as we did not get into
4900 * process_recorded_refs_if_needed in the new_gen case.
4902 sctx->send_progress = sctx->cur_ino + 1;
4905 * Now process all extents and xattrs of the inode as if
4906 * they were all new.
4908 ret = process_all_extents(sctx);
4911 ret = process_all_new_xattrs(sctx);
4915 sctx->cur_inode_gen = left_gen;
4916 sctx->cur_inode_new = 0;
4917 sctx->cur_inode_new_gen = 0;
4918 sctx->cur_inode_deleted = 0;
4919 sctx->cur_inode_size = btrfs_inode_size(
4920 sctx->left_path->nodes[0], left_ii);
4921 sctx->cur_inode_mode = btrfs_inode_mode(
4922 sctx->left_path->nodes[0], left_ii);
4931 * We have to process new refs before deleted refs, but compare_trees gives us
4932 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4933 * first and later process them in process_recorded_refs.
4934 * For the cur_inode_new_gen case, we skip recording completely because
4935 * changed_inode did already initiate processing of refs. The reason for this is
4936 * that in this case, compare_tree actually compares the refs of 2 different
4937 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4938 * refs of the right tree as deleted and all refs of the left tree as new.
4940 static int changed_ref(struct send_ctx *sctx,
4941 enum btrfs_compare_tree_result result)
4945 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4947 if (!sctx->cur_inode_new_gen &&
4948 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4949 if (result == BTRFS_COMPARE_TREE_NEW)
4950 ret = record_new_ref(sctx);
4951 else if (result == BTRFS_COMPARE_TREE_DELETED)
4952 ret = record_deleted_ref(sctx);
4953 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4954 ret = record_changed_ref(sctx);
4961 * Process new/deleted/changed xattrs. We skip processing in the
4962 * cur_inode_new_gen case because changed_inode did already initiate processing
4963 * of xattrs. The reason is the same as in changed_ref
4965 static int changed_xattr(struct send_ctx *sctx,
4966 enum btrfs_compare_tree_result result)
4970 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4972 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4973 if (result == BTRFS_COMPARE_TREE_NEW)
4974 ret = process_new_xattr(sctx);
4975 else if (result == BTRFS_COMPARE_TREE_DELETED)
4976 ret = process_deleted_xattr(sctx);
4977 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4978 ret = process_changed_xattr(sctx);
4985 * Process new/deleted/changed extents. We skip processing in the
4986 * cur_inode_new_gen case because changed_inode did already initiate processing
4987 * of extents. The reason is the same as in changed_ref
4989 static int changed_extent(struct send_ctx *sctx,
4990 enum btrfs_compare_tree_result result)
4994 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4996 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4997 if (result != BTRFS_COMPARE_TREE_DELETED)
4998 ret = process_extent(sctx, sctx->left_path,
5005 static int dir_changed(struct send_ctx *sctx, u64 dir)
5007 u64 orig_gen, new_gen;
5010 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5015 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5020 return (orig_gen != new_gen) ? 1 : 0;
5023 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5024 struct btrfs_key *key)
5026 struct btrfs_inode_extref *extref;
5027 struct extent_buffer *leaf;
5028 u64 dirid = 0, last_dirid = 0;
5035 /* Easy case, just check this one dirid */
5036 if (key->type == BTRFS_INODE_REF_KEY) {
5037 dirid = key->offset;
5039 ret = dir_changed(sctx, dirid);
5043 leaf = path->nodes[0];
5044 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5045 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5046 while (cur_offset < item_size) {
5047 extref = (struct btrfs_inode_extref *)(ptr +
5049 dirid = btrfs_inode_extref_parent(leaf, extref);
5050 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5051 cur_offset += ref_name_len + sizeof(*extref);
5052 if (dirid == last_dirid)
5054 ret = dir_changed(sctx, dirid);
5064 * Updates compare related fields in sctx and simply forwards to the actual
5065 * changed_xxx functions.
5067 static int changed_cb(struct btrfs_root *left_root,
5068 struct btrfs_root *right_root,
5069 struct btrfs_path *left_path,
5070 struct btrfs_path *right_path,
5071 struct btrfs_key *key,
5072 enum btrfs_compare_tree_result result,
5076 struct send_ctx *sctx = ctx;
5078 if (result == BTRFS_COMPARE_TREE_SAME) {
5079 if (key->type == BTRFS_INODE_REF_KEY ||
5080 key->type == BTRFS_INODE_EXTREF_KEY) {
5081 ret = compare_refs(sctx, left_path, key);
5086 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5087 return maybe_send_hole(sctx, left_path, key);
5091 result = BTRFS_COMPARE_TREE_CHANGED;
5095 sctx->left_path = left_path;
5096 sctx->right_path = right_path;
5097 sctx->cmp_key = key;
5099 ret = finish_inode_if_needed(sctx, 0);
5103 /* Ignore non-FS objects */
5104 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5105 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5108 if (key->type == BTRFS_INODE_ITEM_KEY)
5109 ret = changed_inode(sctx, result);
5110 else if (key->type == BTRFS_INODE_REF_KEY ||
5111 key->type == BTRFS_INODE_EXTREF_KEY)
5112 ret = changed_ref(sctx, result);
5113 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5114 ret = changed_xattr(sctx, result);
5115 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5116 ret = changed_extent(sctx, result);
5122 static int full_send_tree(struct send_ctx *sctx)
5125 struct btrfs_root *send_root = sctx->send_root;
5126 struct btrfs_key key;
5127 struct btrfs_key found_key;
5128 struct btrfs_path *path;
5129 struct extent_buffer *eb;
5134 path = alloc_path_for_send();
5138 spin_lock(&send_root->root_item_lock);
5139 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
5140 spin_unlock(&send_root->root_item_lock);
5142 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5143 key.type = BTRFS_INODE_ITEM_KEY;
5147 * Make sure the tree has not changed after re-joining. We detect this
5148 * by comparing start_ctransid and ctransid. They should always match.
5150 spin_lock(&send_root->root_item_lock);
5151 ctransid = btrfs_root_ctransid(&send_root->root_item);
5152 spin_unlock(&send_root->root_item_lock);
5154 if (ctransid != start_ctransid) {
5155 WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
5156 "send was modified in between. This is "
5157 "probably a bug.\n");
5162 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5169 eb = path->nodes[0];
5170 slot = path->slots[0];
5171 btrfs_item_key_to_cpu(eb, &found_key, slot);
5173 ret = changed_cb(send_root, NULL, path, NULL,
5174 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5178 key.objectid = found_key.objectid;
5179 key.type = found_key.type;
5180 key.offset = found_key.offset + 1;
5182 ret = btrfs_next_item(send_root, path);
5192 ret = finish_inode_if_needed(sctx, 1);
5195 btrfs_free_path(path);
5199 static int send_subvol(struct send_ctx *sctx)
5203 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5204 ret = send_header(sctx);
5209 ret = send_subvol_begin(sctx);
5213 if (sctx->parent_root) {
5214 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5218 ret = finish_inode_if_needed(sctx, 1);
5222 ret = full_send_tree(sctx);
5228 free_recorded_refs(sctx);
5232 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5234 spin_lock(&root->root_item_lock);
5235 root->send_in_progress--;
5237 * Not much left to do, we don't know why it's unbalanced and
5238 * can't blindly reset it to 0.
5240 if (root->send_in_progress < 0)
5241 btrfs_err(root->fs_info,
5242 "send_in_progres unbalanced %d root %llu\n",
5243 root->send_in_progress, root->root_key.objectid);
5244 spin_unlock(&root->root_item_lock);
5247 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5250 struct btrfs_root *send_root;
5251 struct btrfs_root *clone_root;
5252 struct btrfs_fs_info *fs_info;
5253 struct btrfs_ioctl_send_args *arg = NULL;
5254 struct btrfs_key key;
5255 struct send_ctx *sctx = NULL;
5257 u64 *clone_sources_tmp = NULL;
5258 int clone_sources_to_rollback = 0;
5259 int sort_clone_roots = 0;
5262 if (!capable(CAP_SYS_ADMIN))
5265 send_root = BTRFS_I(file_inode(mnt_file))->root;
5266 fs_info = send_root->fs_info;
5269 * The subvolume must remain read-only during send, protect against
5272 spin_lock(&send_root->root_item_lock);
5273 send_root->send_in_progress++;
5274 spin_unlock(&send_root->root_item_lock);
5277 * This is done when we lookup the root, it should already be complete
5278 * by the time we get here.
5280 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5283 * Userspace tools do the checks and warn the user if it's
5286 if (!btrfs_root_readonly(send_root)) {
5291 arg = memdup_user(arg_, sizeof(*arg));
5298 if (!access_ok(VERIFY_READ, arg->clone_sources,
5299 sizeof(*arg->clone_sources) *
5300 arg->clone_sources_count)) {
5305 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5310 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5316 INIT_LIST_HEAD(&sctx->new_refs);
5317 INIT_LIST_HEAD(&sctx->deleted_refs);
5318 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5319 INIT_LIST_HEAD(&sctx->name_cache_list);
5321 sctx->flags = arg->flags;
5323 sctx->send_filp = fget(arg->send_fd);
5324 if (!sctx->send_filp) {
5329 sctx->send_root = send_root;
5330 sctx->clone_roots_cnt = arg->clone_sources_count;
5332 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5333 sctx->send_buf = vmalloc(sctx->send_max_size);
5334 if (!sctx->send_buf) {
5339 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5340 if (!sctx->read_buf) {
5345 sctx->pending_dir_moves = RB_ROOT;
5346 sctx->waiting_dir_moves = RB_ROOT;
5348 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5349 (arg->clone_sources_count + 1));
5350 if (!sctx->clone_roots) {
5355 if (arg->clone_sources_count) {
5356 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5357 sizeof(*arg->clone_sources));
5358 if (!clone_sources_tmp) {
5363 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5364 arg->clone_sources_count *
5365 sizeof(*arg->clone_sources));
5371 for (i = 0; i < arg->clone_sources_count; i++) {
5372 key.objectid = clone_sources_tmp[i];
5373 key.type = BTRFS_ROOT_ITEM_KEY;
5374 key.offset = (u64)-1;
5376 index = srcu_read_lock(&fs_info->subvol_srcu);
5378 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5379 if (IS_ERR(clone_root)) {
5380 srcu_read_unlock(&fs_info->subvol_srcu, index);
5381 ret = PTR_ERR(clone_root);
5384 clone_sources_to_rollback = i + 1;
5385 spin_lock(&clone_root->root_item_lock);
5386 clone_root->send_in_progress++;
5387 if (!btrfs_root_readonly(clone_root)) {
5388 spin_unlock(&clone_root->root_item_lock);
5389 srcu_read_unlock(&fs_info->subvol_srcu, index);
5393 spin_unlock(&clone_root->root_item_lock);
5394 srcu_read_unlock(&fs_info->subvol_srcu, index);
5396 sctx->clone_roots[i].root = clone_root;
5398 vfree(clone_sources_tmp);
5399 clone_sources_tmp = NULL;
5402 if (arg->parent_root) {
5403 key.objectid = arg->parent_root;
5404 key.type = BTRFS_ROOT_ITEM_KEY;
5405 key.offset = (u64)-1;
5407 index = srcu_read_lock(&fs_info->subvol_srcu);
5409 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5410 if (IS_ERR(sctx->parent_root)) {
5411 srcu_read_unlock(&fs_info->subvol_srcu, index);
5412 ret = PTR_ERR(sctx->parent_root);
5416 spin_lock(&sctx->parent_root->root_item_lock);
5417 sctx->parent_root->send_in_progress++;
5418 if (!btrfs_root_readonly(sctx->parent_root)) {
5419 spin_unlock(&sctx->parent_root->root_item_lock);
5420 srcu_read_unlock(&fs_info->subvol_srcu, index);
5424 spin_unlock(&sctx->parent_root->root_item_lock);
5426 srcu_read_unlock(&fs_info->subvol_srcu, index);
5430 * Clones from send_root are allowed, but only if the clone source
5431 * is behind the current send position. This is checked while searching
5432 * for possible clone sources.
5434 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5436 /* We do a bsearch later */
5437 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5438 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5440 sort_clone_roots = 1;
5442 ret = send_subvol(sctx);
5446 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5447 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5450 ret = send_cmd(sctx);
5456 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5457 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5459 struct pending_dir_move *pm;
5461 n = rb_first(&sctx->pending_dir_moves);
5462 pm = rb_entry(n, struct pending_dir_move, node);
5463 while (!list_empty(&pm->list)) {
5464 struct pending_dir_move *pm2;
5466 pm2 = list_first_entry(&pm->list,
5467 struct pending_dir_move, list);
5468 free_pending_move(sctx, pm2);
5470 free_pending_move(sctx, pm);
5473 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5474 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5476 struct waiting_dir_move *dm;
5478 n = rb_first(&sctx->waiting_dir_moves);
5479 dm = rb_entry(n, struct waiting_dir_move, node);
5480 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5484 if (sort_clone_roots) {
5485 for (i = 0; i < sctx->clone_roots_cnt; i++)
5486 btrfs_root_dec_send_in_progress(
5487 sctx->clone_roots[i].root);
5489 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5490 btrfs_root_dec_send_in_progress(
5491 sctx->clone_roots[i].root);
5493 btrfs_root_dec_send_in_progress(send_root);
5495 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5496 btrfs_root_dec_send_in_progress(sctx->parent_root);
5499 vfree(clone_sources_tmp);
5502 if (sctx->send_filp)
5503 fput(sctx->send_filp);
5505 vfree(sctx->clone_roots);
5506 vfree(sctx->send_buf);
5507 vfree(sctx->read_buf);
5509 name_cache_free(sctx);