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/crc32c.h>
28 #include <linux/vmalloc.h>
34 #include "btrfs_inode.h"
35 #include "transaction.h"
37 static int g_verbose = 0;
39 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
42 * A fs_path is a helper to dynamically build path names with unknown size.
43 * It reallocates the internal buffer on demand.
44 * It allows fast adding of path elements on the right side (normal path) and
45 * fast adding to the left side (reversed path). A reversed path can also be
46 * unreversed if needed.
64 #define FS_PATH_INLINE_SIZE \
65 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
68 /* reused for each extent */
70 struct btrfs_root *root;
77 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
78 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
81 struct file *send_filp;
87 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
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;
110 int cur_inode_first_ref_orphan;
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;
123 struct file *cur_inode_filp;
127 struct name_cache_entry {
128 struct list_head list;
129 struct list_head use_list;
135 int need_later_update;
140 static void fs_path_reset(struct fs_path *p)
143 p->start = p->buf + p->buf_len - 1;
153 static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
157 p = kmalloc(sizeof(*p), GFP_NOFS);
162 p->buf = p->inline_buf;
163 p->buf_len = FS_PATH_INLINE_SIZE;
168 static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
172 p = fs_path_alloc(sctx);
180 static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
184 if (p->buf != p->inline_buf) {
193 static int fs_path_len(struct fs_path *p)
195 return p->end - p->start;
198 static int fs_path_ensure_buf(struct fs_path *p, int len)
206 if (p->buf_len >= len)
209 path_len = p->end - p->start;
210 old_buf_len = p->buf_len;
211 len = PAGE_ALIGN(len);
213 if (p->buf == p->inline_buf) {
214 tmp_buf = kmalloc(len, GFP_NOFS);
216 tmp_buf = vmalloc(len);
221 memcpy(tmp_buf, p->buf, p->buf_len);
225 if (p->virtual_mem) {
226 tmp_buf = vmalloc(len);
229 memcpy(tmp_buf, p->buf, p->buf_len);
232 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
234 tmp_buf = vmalloc(len);
237 memcpy(tmp_buf, p->buf, p->buf_len);
246 tmp_buf = p->buf + old_buf_len - path_len - 1;
247 p->end = p->buf + p->buf_len - 1;
248 p->start = p->end - path_len;
249 memmove(p->start, tmp_buf, path_len + 1);
252 p->end = p->start + path_len;
257 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
262 new_len = p->end - p->start + name_len;
263 if (p->start != p->end)
265 ret = fs_path_ensure_buf(p, new_len);
270 if (p->start != p->end)
272 p->start -= name_len;
273 p->prepared = p->start;
275 if (p->start != p->end)
277 p->prepared = p->end;
286 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
290 ret = fs_path_prepare_for_add(p, name_len);
293 memcpy(p->prepared, name, name_len);
300 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
304 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
307 memcpy(p->prepared, p2->start, p2->end - p2->start);
314 static int fs_path_add_from_extent_buffer(struct fs_path *p,
315 struct extent_buffer *eb,
316 unsigned long off, int len)
320 ret = fs_path_prepare_for_add(p, len);
324 read_extent_buffer(eb, p->prepared, off, len);
331 static void fs_path_remove(struct fs_path *p)
334 while (p->start != p->end && *p->end != '/')
339 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
343 p->reversed = from->reversed;
346 ret = fs_path_add_path(p, from);
352 static void fs_path_unreverse(struct fs_path *p)
361 len = p->end - p->start;
363 p->end = p->start + len;
364 memmove(p->start, tmp, len + 1);
368 static struct btrfs_path *alloc_path_for_send(void)
370 struct btrfs_path *path;
372 path = btrfs_alloc_path();
375 path->search_commit_root = 1;
376 path->skip_locking = 1;
380 static int write_buf(struct send_ctx *sctx, const void *buf, u32 len)
390 ret = vfs_write(sctx->send_filp, (char *)buf + pos, len - pos,
392 /* TODO handle that correctly */
393 /*if (ret == -ERESTARTSYS) {
412 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
414 struct btrfs_tlv_header *hdr;
415 int total_len = sizeof(*hdr) + len;
416 int left = sctx->send_max_size - sctx->send_size;
418 if (unlikely(left < total_len))
421 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
422 hdr->tlv_type = cpu_to_le16(attr);
423 hdr->tlv_len = cpu_to_le16(len);
424 memcpy(hdr + 1, data, len);
425 sctx->send_size += total_len;
431 static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
433 return tlv_put(sctx, attr, &value, sizeof(value));
436 static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
438 __le16 tmp = cpu_to_le16(value);
439 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
442 static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
444 __le32 tmp = cpu_to_le32(value);
445 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
449 static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
451 __le64 tmp = cpu_to_le64(value);
452 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
455 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
456 const char *str, int len)
460 return tlv_put(sctx, attr, str, len);
463 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
466 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
470 static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
473 struct btrfs_timespec bts;
474 bts.sec = cpu_to_le64(ts->tv_sec);
475 bts.nsec = cpu_to_le32(ts->tv_nsec);
476 return tlv_put(sctx, attr, &bts, sizeof(bts));
480 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
481 struct extent_buffer *eb,
482 struct btrfs_timespec *ts)
484 struct btrfs_timespec bts;
485 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
486 return tlv_put(sctx, attr, &bts, sizeof(bts));
490 #define TLV_PUT(sctx, attrtype, attrlen, data) \
492 ret = tlv_put(sctx, attrtype, attrlen, data); \
494 goto tlv_put_failure; \
497 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
499 ret = tlv_put_u##bits(sctx, attrtype, value); \
501 goto tlv_put_failure; \
504 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
505 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
506 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
507 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
508 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
510 ret = tlv_put_string(sctx, attrtype, str, len); \
512 goto tlv_put_failure; \
514 #define TLV_PUT_PATH(sctx, attrtype, p) \
516 ret = tlv_put_string(sctx, attrtype, p->start, \
517 p->end - p->start); \
519 goto tlv_put_failure; \
521 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
523 ret = tlv_put_uuid(sctx, attrtype, uuid); \
525 goto tlv_put_failure; \
527 #define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
529 ret = tlv_put_timespec(sctx, attrtype, ts); \
531 goto tlv_put_failure; \
533 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
535 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
537 goto tlv_put_failure; \
540 static int send_header(struct send_ctx *sctx)
542 struct btrfs_stream_header hdr;
544 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
545 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
547 return write_buf(sctx, &hdr, sizeof(hdr));
551 * For each command/item we want to send to userspace, we call this function.
553 static int begin_cmd(struct send_ctx *sctx, int cmd)
555 struct btrfs_cmd_header *hdr;
557 if (!sctx->send_buf) {
562 BUG_ON(sctx->send_size);
564 sctx->send_size += sizeof(*hdr);
565 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
566 hdr->cmd = cpu_to_le16(cmd);
571 static int send_cmd(struct send_ctx *sctx)
574 struct btrfs_cmd_header *hdr;
577 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
578 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
581 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
582 hdr->crc = cpu_to_le32(crc);
584 ret = write_buf(sctx, sctx->send_buf, sctx->send_size);
586 sctx->total_send_size += sctx->send_size;
587 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
594 * Sends a move instruction to user space
596 static int send_rename(struct send_ctx *sctx,
597 struct fs_path *from, struct fs_path *to)
601 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
603 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
607 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
608 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
610 ret = send_cmd(sctx);
618 * Sends a link instruction to user space
620 static int send_link(struct send_ctx *sctx,
621 struct fs_path *path, struct fs_path *lnk)
625 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
627 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
631 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
632 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
634 ret = send_cmd(sctx);
642 * Sends an unlink instruction to user space
644 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
648 verbose_printk("btrfs: send_unlink %s\n", path->start);
650 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
654 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
656 ret = send_cmd(sctx);
664 * Sends a rmdir instruction to user space
666 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
670 verbose_printk("btrfs: send_rmdir %s\n", path->start);
672 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
676 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
678 ret = send_cmd(sctx);
686 * Helper function to retrieve some fields from an inode item.
688 static int get_inode_info(struct btrfs_root *root,
689 u64 ino, u64 *size, u64 *gen,
690 u64 *mode, u64 *uid, u64 *gid,
694 struct btrfs_inode_item *ii;
695 struct btrfs_key key;
696 struct btrfs_path *path;
698 path = alloc_path_for_send();
703 key.type = BTRFS_INODE_ITEM_KEY;
705 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
713 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
714 struct btrfs_inode_item);
716 *size = btrfs_inode_size(path->nodes[0], ii);
718 *gen = btrfs_inode_generation(path->nodes[0], ii);
720 *mode = btrfs_inode_mode(path->nodes[0], ii);
722 *uid = btrfs_inode_uid(path->nodes[0], ii);
724 *gid = btrfs_inode_gid(path->nodes[0], ii);
726 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
729 btrfs_free_path(path);
733 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
738 * Helper function to iterate the entries in ONE btrfs_inode_ref.
739 * The iterate callback may return a non zero value to stop iteration. This can
740 * be a negative value for error codes or 1 to simply stop it.
742 * path must point to the INODE_REF when called.
744 static int iterate_inode_ref(struct send_ctx *sctx,
745 struct btrfs_root *root, struct btrfs_path *path,
746 struct btrfs_key *found_key, int resolve,
747 iterate_inode_ref_t iterate, void *ctx)
749 struct extent_buffer *eb;
750 struct btrfs_item *item;
751 struct btrfs_inode_ref *iref;
752 struct btrfs_path *tmp_path;
764 p = fs_path_alloc_reversed(sctx);
768 tmp_path = alloc_path_for_send();
770 fs_path_free(sctx, p);
775 slot = path->slots[0];
776 item = btrfs_item_nr(eb, slot);
777 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
780 total = btrfs_item_size(eb, item);
783 while (cur < total) {
786 name_len = btrfs_inode_ref_name_len(eb, iref);
787 index = btrfs_inode_ref_index(eb, iref);
789 start = btrfs_iref_to_path(root, tmp_path, iref, eb,
790 found_key->offset, p->buf,
793 ret = PTR_ERR(start);
796 if (start < p->buf) {
797 /* overflow , try again with larger buffer */
798 ret = fs_path_ensure_buf(p,
799 p->buf_len + p->buf - start);
802 start = btrfs_iref_to_path(root, tmp_path, iref,
803 eb, found_key->offset, p->buf,
806 ret = PTR_ERR(start);
809 BUG_ON(start < p->buf);
813 ret = fs_path_add_from_extent_buffer(p, eb,
814 (unsigned long)(iref + 1), name_len);
820 len = sizeof(*iref) + name_len;
821 iref = (struct btrfs_inode_ref *)((char *)iref + len);
824 ret = iterate(num, found_key->offset, index, p, ctx);
832 btrfs_free_path(tmp_path);
833 fs_path_free(sctx, p);
837 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
838 const char *name, int name_len,
839 const char *data, int data_len,
843 * Helper function to iterate the entries in ONE btrfs_dir_item.
844 * The iterate callback may return a non zero value to stop iteration. This can
845 * be a negative value for error codes or 1 to simply stop it.
847 * path must point to the dir item when called.
849 static int iterate_dir_item(struct send_ctx *sctx,
850 struct btrfs_root *root, struct btrfs_path *path,
851 struct btrfs_key *found_key,
852 iterate_dir_item_t iterate, void *ctx)
855 struct extent_buffer *eb;
856 struct btrfs_item *item;
857 struct btrfs_dir_item *di;
858 struct btrfs_path *tmp_path = NULL;
859 struct btrfs_key di_key;
874 buf = kmalloc(buf_len, GFP_NOFS);
880 tmp_path = alloc_path_for_send();
887 slot = path->slots[0];
888 item = btrfs_item_nr(eb, slot);
889 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
892 total = btrfs_item_size(eb, item);
895 while (cur < total) {
896 name_len = btrfs_dir_name_len(eb, di);
897 data_len = btrfs_dir_data_len(eb, di);
898 type = btrfs_dir_type(eb, di);
899 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
901 if (name_len + data_len > buf_len) {
902 buf_len = PAGE_ALIGN(name_len + data_len);
904 buf2 = vmalloc(buf_len);
911 buf2 = krealloc(buf, buf_len, GFP_NOFS);
913 buf2 = vmalloc(buf_len);
927 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
928 name_len + data_len);
930 len = sizeof(*di) + name_len + data_len;
931 di = (struct btrfs_dir_item *)((char *)di + len);
934 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
935 data_len, type, ctx);
947 btrfs_free_path(tmp_path);
955 static int __copy_first_ref(int num, u64 dir, int index,
956 struct fs_path *p, void *ctx)
959 struct fs_path *pt = ctx;
961 ret = fs_path_copy(pt, p);
965 /* we want the first only */
970 * Retrieve the first path of an inode. If an inode has more then one
971 * ref/hardlink, this is ignored.
973 static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
974 u64 ino, struct fs_path *path)
977 struct btrfs_key key, found_key;
978 struct btrfs_path *p;
980 p = alloc_path_for_send();
987 key.type = BTRFS_INODE_REF_KEY;
990 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
997 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
998 if (found_key.objectid != ino ||
999 found_key.type != BTRFS_INODE_REF_KEY) {
1004 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1005 __copy_first_ref, path);
1015 struct backref_ctx {
1016 struct send_ctx *sctx;
1018 /* number of total found references */
1022 * used for clones found in send_root. clones found behind cur_objectid
1023 * and cur_offset are not considered as allowed clones.
1028 /* may be truncated in case it's the last extent in a file */
1031 /* Just to check for bugs in backref resolving */
1032 int found_in_send_root;
1035 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1037 u64 root = (u64)key;
1038 struct clone_root *cr = (struct clone_root *)elt;
1040 if (root < cr->root->objectid)
1042 if (root > cr->root->objectid)
1047 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1049 struct clone_root *cr1 = (struct clone_root *)e1;
1050 struct clone_root *cr2 = (struct clone_root *)e2;
1052 if (cr1->root->objectid < cr2->root->objectid)
1054 if (cr1->root->objectid > cr2->root->objectid)
1060 * Called for every backref that is found for the current extent.
1062 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1064 struct backref_ctx *bctx = ctx_;
1065 struct clone_root *found;
1069 /* First check if the root is in the list of accepted clone sources */
1070 found = bsearch((void *)root, bctx->sctx->clone_roots,
1071 bctx->sctx->clone_roots_cnt,
1072 sizeof(struct clone_root),
1073 __clone_root_cmp_bsearch);
1077 if (found->root == bctx->sctx->send_root &&
1078 ino == bctx->cur_objectid &&
1079 offset == bctx->cur_offset) {
1080 bctx->found_in_send_root = 1;
1084 * There are inodes that have extents that lie behind it's i_size. Don't
1085 * accept clones from these extents.
1087 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1092 if (offset + bctx->extent_len > i_size)
1096 * Make sure we don't consider clones from send_root that are
1097 * behind the current inode/offset.
1099 if (found->root == bctx->sctx->send_root) {
1101 * TODO for the moment we don't accept clones from the inode
1102 * that is currently send. We may change this when
1103 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1106 if (ino >= bctx->cur_objectid)
1108 /*if (ino > ctx->cur_objectid)
1110 if (offset + ctx->extent_len > ctx->cur_offset)
1114 found->found_refs++;
1116 found->offset = offset;
1121 found->found_refs++;
1122 if (ino < found->ino) {
1124 found->offset = offset;
1125 } else if (found->ino == ino) {
1127 * same extent found more then once in the same file.
1129 if (found->offset > offset + bctx->extent_len)
1130 found->offset = offset;
1137 * path must point to the extent item when called.
1139 static int find_extent_clone(struct send_ctx *sctx,
1140 struct btrfs_path *path,
1141 u64 ino, u64 data_offset,
1143 struct clone_root **found)
1149 u64 extent_item_pos;
1150 struct btrfs_file_extent_item *fi;
1151 struct extent_buffer *eb = path->nodes[0];
1152 struct backref_ctx backref_ctx;
1153 struct clone_root *cur_clone_root;
1154 struct btrfs_key found_key;
1155 struct btrfs_path *tmp_path;
1158 tmp_path = alloc_path_for_send();
1162 if (data_offset >= ino_size) {
1164 * There may be extents that lie behind the file's size.
1165 * I at least had this in combination with snapshotting while
1166 * writing large files.
1172 fi = btrfs_item_ptr(eb, path->slots[0],
1173 struct btrfs_file_extent_item);
1174 extent_type = btrfs_file_extent_type(eb, fi);
1175 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1180 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1181 logical = btrfs_file_extent_disk_bytenr(eb, fi);
1186 logical += btrfs_file_extent_offset(eb, fi);
1188 ret = extent_from_logical(sctx->send_root->fs_info,
1189 logical, tmp_path, &found_key);
1190 btrfs_release_path(tmp_path);
1194 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1200 * Setup the clone roots.
1202 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1203 cur_clone_root = sctx->clone_roots + i;
1204 cur_clone_root->ino = (u64)-1;
1205 cur_clone_root->offset = 0;
1206 cur_clone_root->found_refs = 0;
1209 backref_ctx.sctx = sctx;
1210 backref_ctx.found = 0;
1211 backref_ctx.cur_objectid = ino;
1212 backref_ctx.cur_offset = data_offset;
1213 backref_ctx.found_in_send_root = 0;
1214 backref_ctx.extent_len = num_bytes;
1217 * The last extent of a file may be too large due to page alignment.
1218 * We need to adjust extent_len in this case so that the checks in
1219 * __iterate_backrefs work.
1221 if (data_offset + num_bytes >= ino_size)
1222 backref_ctx.extent_len = ino_size - data_offset;
1225 * Now collect all backrefs.
1227 extent_item_pos = logical - found_key.objectid;
1228 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1229 found_key.objectid, extent_item_pos, 1,
1230 __iterate_backrefs, &backref_ctx);
1234 if (!backref_ctx.found_in_send_root) {
1235 /* found a bug in backref code? */
1237 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1238 "send_root. inode=%llu, offset=%llu, "
1240 ino, data_offset, logical);
1244 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1246 "num_bytes=%llu, logical=%llu\n",
1247 data_offset, ino, num_bytes, logical);
1249 if (!backref_ctx.found)
1250 verbose_printk("btrfs: no clones found\n");
1252 cur_clone_root = NULL;
1253 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1254 if (sctx->clone_roots[i].found_refs) {
1255 if (!cur_clone_root)
1256 cur_clone_root = sctx->clone_roots + i;
1257 else if (sctx->clone_roots[i].root == sctx->send_root)
1258 /* prefer clones from send_root over others */
1259 cur_clone_root = sctx->clone_roots + i;
1265 if (cur_clone_root) {
1266 *found = cur_clone_root;
1273 btrfs_free_path(tmp_path);
1277 static int read_symlink(struct send_ctx *sctx,
1278 struct btrfs_root *root,
1280 struct fs_path *dest)
1283 struct btrfs_path *path;
1284 struct btrfs_key key;
1285 struct btrfs_file_extent_item *ei;
1291 path = alloc_path_for_send();
1296 key.type = BTRFS_EXTENT_DATA_KEY;
1298 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1303 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1304 struct btrfs_file_extent_item);
1305 type = btrfs_file_extent_type(path->nodes[0], ei);
1306 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1307 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1308 BUG_ON(compression);
1310 off = btrfs_file_extent_inline_start(ei);
1311 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1313 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1318 btrfs_free_path(path);
1323 * Helper function to generate a file name that is unique in the root of
1324 * send_root and parent_root. This is used to generate names for orphan inodes.
1326 static int gen_unique_name(struct send_ctx *sctx,
1328 struct fs_path *dest)
1331 struct btrfs_path *path;
1332 struct btrfs_dir_item *di;
1337 path = alloc_path_for_send();
1342 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1344 if (len >= sizeof(tmp)) {
1345 /* should really not happen */
1350 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1351 path, BTRFS_FIRST_FREE_OBJECTID,
1352 tmp, strlen(tmp), 0);
1353 btrfs_release_path(path);
1359 /* not unique, try again */
1364 if (!sctx->parent_root) {
1370 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1371 path, BTRFS_FIRST_FREE_OBJECTID,
1372 tmp, strlen(tmp), 0);
1373 btrfs_release_path(path);
1379 /* not unique, try again */
1387 ret = fs_path_add(dest, tmp, strlen(tmp));
1390 btrfs_free_path(path);
1395 inode_state_no_change,
1396 inode_state_will_create,
1397 inode_state_did_create,
1398 inode_state_will_delete,
1399 inode_state_did_delete,
1402 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1410 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1412 if (ret < 0 && ret != -ENOENT)
1416 if (!sctx->parent_root) {
1417 right_ret = -ENOENT;
1419 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1420 NULL, NULL, NULL, NULL);
1421 if (ret < 0 && ret != -ENOENT)
1426 if (!left_ret && !right_ret) {
1427 if (left_gen == gen && right_gen == gen)
1428 ret = inode_state_no_change;
1429 else if (left_gen == gen) {
1430 if (ino < sctx->send_progress)
1431 ret = inode_state_did_create;
1433 ret = inode_state_will_create;
1434 } else if (right_gen == gen) {
1435 if (ino < sctx->send_progress)
1436 ret = inode_state_did_delete;
1438 ret = inode_state_will_delete;
1442 } else if (!left_ret) {
1443 if (left_gen == gen) {
1444 if (ino < sctx->send_progress)
1445 ret = inode_state_did_create;
1447 ret = inode_state_will_create;
1451 } else if (!right_ret) {
1452 if (right_gen == gen) {
1453 if (ino < sctx->send_progress)
1454 ret = inode_state_did_delete;
1456 ret = inode_state_will_delete;
1468 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1472 ret = get_cur_inode_state(sctx, ino, gen);
1476 if (ret == inode_state_no_change ||
1477 ret == inode_state_did_create ||
1478 ret == inode_state_will_delete)
1488 * Helper function to lookup a dir item in a dir.
1490 static int lookup_dir_item_inode(struct btrfs_root *root,
1491 u64 dir, const char *name, int name_len,
1496 struct btrfs_dir_item *di;
1497 struct btrfs_key key;
1498 struct btrfs_path *path;
1500 path = alloc_path_for_send();
1504 di = btrfs_lookup_dir_item(NULL, root, path,
1505 dir, name, name_len, 0);
1514 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1515 *found_inode = key.objectid;
1516 *found_type = btrfs_dir_type(path->nodes[0], di);
1519 btrfs_free_path(path);
1523 static int get_first_ref(struct send_ctx *sctx,
1524 struct btrfs_root *root, u64 ino,
1525 u64 *dir, u64 *dir_gen, struct fs_path *name)
1528 struct btrfs_key key;
1529 struct btrfs_key found_key;
1530 struct btrfs_path *path;
1531 struct btrfs_inode_ref *iref;
1534 path = alloc_path_for_send();
1539 key.type = BTRFS_INODE_REF_KEY;
1542 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1546 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1548 if (ret || found_key.objectid != key.objectid ||
1549 found_key.type != key.type) {
1554 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1555 struct btrfs_inode_ref);
1556 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1557 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1558 (unsigned long)(iref + 1), len);
1561 btrfs_release_path(path);
1563 ret = get_inode_info(root, found_key.offset, NULL, dir_gen, NULL, NULL,
1568 *dir = found_key.offset;
1571 btrfs_free_path(path);
1575 static int is_first_ref(struct send_ctx *sctx,
1576 struct btrfs_root *root,
1578 const char *name, int name_len)
1581 struct fs_path *tmp_name;
1585 tmp_name = fs_path_alloc(sctx);
1589 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1593 if (name_len != fs_path_len(tmp_name)) {
1598 ret = memcmp(tmp_name->start, name, name_len);
1605 fs_path_free(sctx, tmp_name);
1609 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1610 const char *name, int name_len,
1611 u64 *who_ino, u64 *who_gen)
1614 u64 other_inode = 0;
1617 if (!sctx->parent_root)
1620 ret = is_inode_existent(sctx, dir, dir_gen);
1624 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1625 &other_inode, &other_type);
1626 if (ret < 0 && ret != -ENOENT)
1633 if (other_inode > sctx->send_progress) {
1634 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1635 who_gen, NULL, NULL, NULL, NULL);
1640 *who_ino = other_inode;
1649 static int did_overwrite_ref(struct send_ctx *sctx,
1650 u64 dir, u64 dir_gen,
1651 u64 ino, u64 ino_gen,
1652 const char *name, int name_len)
1659 if (!sctx->parent_root)
1662 ret = is_inode_existent(sctx, dir, dir_gen);
1666 /* check if the ref was overwritten by another ref */
1667 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1668 &ow_inode, &other_type);
1669 if (ret < 0 && ret != -ENOENT)
1672 /* was never and will never be overwritten */
1677 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1682 if (ow_inode == ino && gen == ino_gen) {
1687 /* we know that it is or will be overwritten. check this now */
1688 if (ow_inode < sctx->send_progress)
1697 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1700 struct fs_path *name = NULL;
1704 if (!sctx->parent_root)
1707 name = fs_path_alloc(sctx);
1711 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1715 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1716 name->start, fs_path_len(name));
1721 fs_path_free(sctx, name);
1725 static int name_cache_insert(struct send_ctx *sctx,
1726 struct name_cache_entry *nce)
1729 struct name_cache_entry **ncea;
1731 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1740 ncea = kmalloc(sizeof(void *) * 2, GFP_NOFS);
1746 ret = radix_tree_insert(&sctx->name_cache, nce->ino, ncea);
1750 list_add_tail(&nce->list, &sctx->name_cache_list);
1751 sctx->name_cache_size++;
1756 static void name_cache_delete(struct send_ctx *sctx,
1757 struct name_cache_entry *nce)
1759 struct name_cache_entry **ncea;
1761 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1766 else if (ncea[1] == nce)
1771 if (!ncea[0] && !ncea[1]) {
1772 radix_tree_delete(&sctx->name_cache, nce->ino);
1776 list_del(&nce->list);
1778 sctx->name_cache_size--;
1781 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1784 struct name_cache_entry **ncea;
1786 ncea = radix_tree_lookup(&sctx->name_cache, ino);
1790 if (ncea[0] && ncea[0]->gen == gen)
1792 else if (ncea[1] && ncea[1]->gen == gen)
1797 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1799 list_del(&nce->list);
1800 list_add_tail(&nce->list, &sctx->name_cache_list);
1803 static void name_cache_clean_unused(struct send_ctx *sctx)
1805 struct name_cache_entry *nce;
1807 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1810 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1811 nce = list_entry(sctx->name_cache_list.next,
1812 struct name_cache_entry, list);
1813 name_cache_delete(sctx, nce);
1818 static void name_cache_free(struct send_ctx *sctx)
1820 struct name_cache_entry *nce;
1821 struct name_cache_entry *tmp;
1823 list_for_each_entry_safe(nce, tmp, &sctx->name_cache_list, list) {
1824 name_cache_delete(sctx, nce);
1828 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1832 struct fs_path *dest)
1836 struct btrfs_path *path = NULL;
1837 struct name_cache_entry *nce = NULL;
1839 nce = name_cache_search(sctx, ino, gen);
1841 if (ino < sctx->send_progress && nce->need_later_update) {
1842 name_cache_delete(sctx, nce);
1846 name_cache_used(sctx, nce);
1847 *parent_ino = nce->parent_ino;
1848 *parent_gen = nce->parent_gen;
1849 ret = fs_path_add(dest, nce->name, nce->name_len);
1857 path = alloc_path_for_send();
1861 ret = is_inode_existent(sctx, ino, gen);
1866 ret = gen_unique_name(sctx, ino, gen, dest);
1873 if (ino < sctx->send_progress)
1874 ret = get_first_ref(sctx, sctx->send_root, ino,
1875 parent_ino, parent_gen, dest);
1877 ret = get_first_ref(sctx, sctx->parent_root, ino,
1878 parent_ino, parent_gen, dest);
1882 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1883 dest->start, dest->end - dest->start);
1887 fs_path_reset(dest);
1888 ret = gen_unique_name(sctx, ino, gen, dest);
1895 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
1903 nce->parent_ino = *parent_ino;
1904 nce->parent_gen = *parent_gen;
1905 nce->name_len = fs_path_len(dest);
1907 strcpy(nce->name, dest->start);
1908 memset(&nce->use_list, 0, sizeof(nce->use_list));
1910 if (ino < sctx->send_progress)
1911 nce->need_later_update = 0;
1913 nce->need_later_update = 1;
1915 nce_ret = name_cache_insert(sctx, nce);
1918 name_cache_clean_unused(sctx);
1921 btrfs_free_path(path);
1926 * Magic happens here. This function returns the first ref to an inode as it
1927 * would look like while receiving the stream at this point in time.
1928 * We walk the path up to the root. For every inode in between, we check if it
1929 * was already processed/sent. If yes, we continue with the parent as found
1930 * in send_root. If not, we continue with the parent as found in parent_root.
1931 * If we encounter an inode that was deleted at this point in time, we use the
1932 * inodes "orphan" name instead of the real name and stop. Same with new inodes
1933 * that were not created yet and overwritten inodes/refs.
1935 * When do we have have orphan inodes:
1936 * 1. When an inode is freshly created and thus no valid refs are available yet
1937 * 2. When a directory lost all it's refs (deleted) but still has dir items
1938 * inside which were not processed yet (pending for move/delete). If anyone
1939 * tried to get the path to the dir items, it would get a path inside that
1941 * 3. When an inode is moved around or gets new links, it may overwrite the ref
1942 * of an unprocessed inode. If in that case the first ref would be
1943 * overwritten, the overwritten inode gets "orphanized". Later when we
1944 * process this overwritten inode, it is restored at a new place by moving
1947 * sctx->send_progress tells this function at which point in time receiving
1950 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
1951 struct fs_path *dest)
1954 struct fs_path *name = NULL;
1955 u64 parent_inode = 0;
1959 name = fs_path_alloc(sctx);
1966 fs_path_reset(dest);
1968 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
1969 fs_path_reset(name);
1971 ret = __get_cur_name_and_parent(sctx, ino, gen,
1972 &parent_inode, &parent_gen, name);
1978 ret = fs_path_add_path(dest, name);
1987 fs_path_free(sctx, name);
1989 fs_path_unreverse(dest);
1994 * Called for regular files when sending extents data. Opens a struct file
1995 * to read from the file.
1997 static int open_cur_inode_file(struct send_ctx *sctx)
2000 struct btrfs_key key;
2002 struct inode *inode;
2003 struct dentry *dentry;
2007 if (sctx->cur_inode_filp)
2010 key.objectid = sctx->cur_ino;
2011 key.type = BTRFS_INODE_ITEM_KEY;
2014 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2016 if (IS_ERR(inode)) {
2017 ret = PTR_ERR(inode);
2021 dentry = d_obtain_alias(inode);
2023 if (IS_ERR(dentry)) {
2024 ret = PTR_ERR(dentry);
2028 path.mnt = sctx->mnt;
2029 path.dentry = dentry;
2030 filp = dentry_open(&path, O_RDONLY | O_LARGEFILE, current_cred());
2034 ret = PTR_ERR(filp);
2037 sctx->cur_inode_filp = filp;
2041 * no xxxput required here as every vfs op
2042 * does it by itself on failure
2048 * Closes the struct file that was created in open_cur_inode_file
2050 static int close_cur_inode_file(struct send_ctx *sctx)
2054 if (!sctx->cur_inode_filp)
2057 ret = filp_close(sctx->cur_inode_filp, NULL);
2058 sctx->cur_inode_filp = NULL;
2065 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2067 static int send_subvol_begin(struct send_ctx *sctx)
2070 struct btrfs_root *send_root = sctx->send_root;
2071 struct btrfs_root *parent_root = sctx->parent_root;
2072 struct btrfs_path *path;
2073 struct btrfs_key key;
2074 struct btrfs_root_ref *ref;
2075 struct extent_buffer *leaf;
2079 path = alloc_path_for_send();
2083 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2085 btrfs_free_path(path);
2089 key.objectid = send_root->objectid;
2090 key.type = BTRFS_ROOT_BACKREF_KEY;
2093 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2102 leaf = path->nodes[0];
2103 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2104 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2105 key.objectid != send_root->objectid) {
2109 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2110 namelen = btrfs_root_ref_name_len(leaf, ref);
2111 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2112 btrfs_release_path(path);
2118 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2122 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2127 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2128 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2129 sctx->send_root->root_item.uuid);
2130 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2131 sctx->send_root->root_item.ctransid);
2133 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2134 sctx->parent_root->root_item.uuid);
2135 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2136 sctx->parent_root->root_item.ctransid);
2139 ret = send_cmd(sctx);
2143 btrfs_free_path(path);
2148 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2153 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2155 p = fs_path_alloc(sctx);
2159 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2163 ret = get_cur_path(sctx, ino, gen, p);
2166 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2167 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2169 ret = send_cmd(sctx);
2173 fs_path_free(sctx, p);
2177 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2182 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2184 p = fs_path_alloc(sctx);
2188 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2192 ret = get_cur_path(sctx, ino, gen, p);
2195 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2196 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2198 ret = send_cmd(sctx);
2202 fs_path_free(sctx, p);
2206 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2211 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2213 p = fs_path_alloc(sctx);
2217 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2221 ret = get_cur_path(sctx, ino, gen, p);
2224 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2225 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2226 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2228 ret = send_cmd(sctx);
2232 fs_path_free(sctx, p);
2236 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2239 struct fs_path *p = NULL;
2240 struct btrfs_inode_item *ii;
2241 struct btrfs_path *path = NULL;
2242 struct extent_buffer *eb;
2243 struct btrfs_key key;
2246 verbose_printk("btrfs: send_utimes %llu\n", ino);
2248 p = fs_path_alloc(sctx);
2252 path = alloc_path_for_send();
2259 key.type = BTRFS_INODE_ITEM_KEY;
2261 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2265 eb = path->nodes[0];
2266 slot = path->slots[0];
2267 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2269 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2273 ret = get_cur_path(sctx, ino, gen, p);
2276 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2277 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2278 btrfs_inode_atime(ii));
2279 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2280 btrfs_inode_mtime(ii));
2281 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2282 btrfs_inode_ctime(ii));
2285 ret = send_cmd(sctx);
2289 fs_path_free(sctx, p);
2290 btrfs_free_path(path);
2295 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2296 * a valid path yet because we did not process the refs yet. So, the inode
2297 * is created as orphan.
2299 static int send_create_inode(struct send_ctx *sctx, struct btrfs_path *path,
2300 struct btrfs_key *key)
2303 struct extent_buffer *eb = path->nodes[0];
2304 struct btrfs_inode_item *ii;
2306 int slot = path->slots[0];
2310 verbose_printk("btrfs: send_create_inode %llu\n", sctx->cur_ino);
2312 p = fs_path_alloc(sctx);
2316 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2317 mode = btrfs_inode_mode(eb, ii);
2320 cmd = BTRFS_SEND_C_MKFILE;
2321 else if (S_ISDIR(mode))
2322 cmd = BTRFS_SEND_C_MKDIR;
2323 else if (S_ISLNK(mode))
2324 cmd = BTRFS_SEND_C_SYMLINK;
2325 else if (S_ISCHR(mode) || S_ISBLK(mode))
2326 cmd = BTRFS_SEND_C_MKNOD;
2327 else if (S_ISFIFO(mode))
2328 cmd = BTRFS_SEND_C_MKFIFO;
2329 else if (S_ISSOCK(mode))
2330 cmd = BTRFS_SEND_C_MKSOCK;
2332 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2333 (int)(mode & S_IFMT));
2338 ret = begin_cmd(sctx, cmd);
2342 ret = gen_unique_name(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
2346 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2347 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, sctx->cur_ino);
2349 if (S_ISLNK(mode)) {
2351 ret = read_symlink(sctx, sctx->send_root, sctx->cur_ino, p);
2354 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2355 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2356 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2357 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, btrfs_inode_rdev(eb, ii));
2360 ret = send_cmd(sctx);
2367 fs_path_free(sctx, p);
2371 struct recorded_ref {
2372 struct list_head list;
2375 struct fs_path *full_path;
2383 * We need to process new refs before deleted refs, but compare_tree gives us
2384 * everything mixed. So we first record all refs and later process them.
2385 * This function is a helper to record one ref.
2387 static int record_ref(struct list_head *head, u64 dir,
2388 u64 dir_gen, struct fs_path *path)
2390 struct recorded_ref *ref;
2393 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2398 ref->dir_gen = dir_gen;
2399 ref->full_path = path;
2401 tmp = strrchr(ref->full_path->start, '/');
2403 ref->name_len = ref->full_path->end - ref->full_path->start;
2404 ref->name = ref->full_path->start;
2405 ref->dir_path_len = 0;
2406 ref->dir_path = ref->full_path->start;
2409 ref->name_len = ref->full_path->end - tmp;
2411 ref->dir_path = ref->full_path->start;
2412 ref->dir_path_len = ref->full_path->end -
2413 ref->full_path->start - 1 - ref->name_len;
2416 list_add_tail(&ref->list, head);
2420 static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
2422 struct recorded_ref *cur;
2423 struct recorded_ref *tmp;
2425 list_for_each_entry_safe(cur, tmp, head, list) {
2426 fs_path_free(sctx, cur->full_path);
2429 INIT_LIST_HEAD(head);
2432 static void free_recorded_refs(struct send_ctx *sctx)
2434 __free_recorded_refs(sctx, &sctx->new_refs);
2435 __free_recorded_refs(sctx, &sctx->deleted_refs);
2439 * Renames/moves a file/dir to it's orphan name. Used when the first
2440 * ref of an unprocessed inode gets overwritten and for all non empty
2443 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2444 struct fs_path *path)
2447 struct fs_path *orphan;
2449 orphan = fs_path_alloc(sctx);
2453 ret = gen_unique_name(sctx, ino, gen, orphan);
2457 ret = send_rename(sctx, path, orphan);
2460 fs_path_free(sctx, orphan);
2465 * Returns 1 if a directory can be removed at this point in time.
2466 * We check this by iterating all dir items and checking if the inode behind
2467 * the dir item was already processed.
2469 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2472 struct btrfs_root *root = sctx->parent_root;
2473 struct btrfs_path *path;
2474 struct btrfs_key key;
2475 struct btrfs_key found_key;
2476 struct btrfs_key loc;
2477 struct btrfs_dir_item *di;
2479 path = alloc_path_for_send();
2484 key.type = BTRFS_DIR_INDEX_KEY;
2488 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2495 if (ret || found_key.objectid != key.objectid ||
2496 found_key.type != key.type) {
2500 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2501 struct btrfs_dir_item);
2502 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2504 if (loc.objectid > send_progress) {
2509 btrfs_release_path(path);
2510 key.offset = found_key.offset + 1;
2516 btrfs_free_path(path);
2520 struct finish_unordered_dir_ctx {
2521 struct send_ctx *sctx;
2522 struct fs_path *cur_path;
2523 struct fs_path *dir_path;
2529 int __finish_unordered_dir(int num, struct btrfs_key *di_key,
2530 const char *name, int name_len,
2531 const char *data, int data_len,
2535 struct finish_unordered_dir_ctx *fctx = ctx;
2536 struct send_ctx *sctx = fctx->sctx;
2541 if (di_key->objectid >= fctx->dir_ino)
2544 fs_path_reset(fctx->cur_path);
2546 ret = get_inode_info(sctx->send_root, di_key->objectid,
2547 NULL, &di_gen, &di_mode, NULL, NULL, NULL);
2551 ret = is_first_ref(sctx, sctx->send_root, di_key->objectid,
2552 fctx->dir_ino, name, name_len);
2557 ret = gen_unique_name(sctx, di_key->objectid, di_gen,
2560 ret = get_cur_path(sctx, di_key->objectid, di_gen,
2566 ret = fs_path_add(fctx->dir_path, name, name_len);
2570 if (!fctx->delete_pass) {
2571 if (S_ISDIR(di_mode)) {
2572 ret = send_rename(sctx, fctx->cur_path,
2575 ret = send_link(sctx, fctx->dir_path,
2578 fctx->need_delete = 1;
2580 } else if (!S_ISDIR(di_mode)) {
2581 ret = send_unlink(sctx, fctx->cur_path);
2586 fs_path_remove(fctx->dir_path);
2593 * Go through all dir items and see if we find refs which could not be created
2594 * in the past because the dir did not exist at that time.
2596 static int finish_outoforder_dir(struct send_ctx *sctx, u64 dir, u64 dir_gen)
2599 struct btrfs_path *path = NULL;
2600 struct btrfs_key key;
2601 struct btrfs_key found_key;
2602 struct extent_buffer *eb;
2603 struct finish_unordered_dir_ctx fctx;
2606 path = alloc_path_for_send();
2612 memset(&fctx, 0, sizeof(fctx));
2614 fctx.cur_path = fs_path_alloc(sctx);
2615 fctx.dir_path = fs_path_alloc(sctx);
2616 if (!fctx.cur_path || !fctx.dir_path) {
2622 ret = get_cur_path(sctx, dir, dir_gen, fctx.dir_path);
2627 * We do two passes. The first links in the new refs and the second
2628 * deletes orphans if required. Deletion of orphans is not required for
2629 * directory inodes, as we always have only one ref and use rename
2630 * instead of link for those.
2635 key.type = BTRFS_DIR_ITEM_KEY;
2638 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2642 eb = path->nodes[0];
2643 slot = path->slots[0];
2644 btrfs_item_key_to_cpu(eb, &found_key, slot);
2646 if (found_key.objectid != key.objectid ||
2647 found_key.type != key.type) {
2648 btrfs_release_path(path);
2652 ret = iterate_dir_item(sctx, sctx->send_root, path,
2653 &found_key, __finish_unordered_dir,
2658 key.offset = found_key.offset + 1;
2659 btrfs_release_path(path);
2662 if (!fctx.delete_pass && fctx.need_delete) {
2663 fctx.delete_pass = 1;
2668 btrfs_free_path(path);
2669 fs_path_free(sctx, fctx.cur_path);
2670 fs_path_free(sctx, fctx.dir_path);
2675 * This does all the move/link/unlink/rmdir magic.
2677 static int process_recorded_refs(struct send_ctx *sctx)
2680 struct recorded_ref *cur;
2681 struct ulist *check_dirs = NULL;
2682 struct ulist_iterator uit;
2683 struct ulist_node *un;
2684 struct fs_path *valid_path = NULL;
2687 int did_overwrite = 0;
2690 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2692 valid_path = fs_path_alloc(sctx);
2698 check_dirs = ulist_alloc(GFP_NOFS);
2705 * First, check if the first ref of the current inode was overwritten
2706 * before. If yes, we know that the current inode was already orphanized
2707 * and thus use the orphan name. If not, we can use get_cur_path to
2708 * get the path of the first ref as it would like while receiving at
2709 * this point in time.
2710 * New inodes are always orphan at the beginning, so force to use the
2711 * orphan name in this case.
2712 * The first ref is stored in valid_path and will be updated if it
2713 * gets moved around.
2715 if (!sctx->cur_inode_new) {
2716 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2717 sctx->cur_inode_gen);
2723 if (sctx->cur_inode_new || did_overwrite) {
2724 ret = gen_unique_name(sctx, sctx->cur_ino,
2725 sctx->cur_inode_gen, valid_path);
2730 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2736 list_for_each_entry(cur, &sctx->new_refs, list) {
2738 * Check if this new ref would overwrite the first ref of
2739 * another unprocessed inode. If yes, orphanize the
2740 * overwritten inode. If we find an overwritten ref that is
2741 * not the first ref, simply unlink it.
2743 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2744 cur->name, cur->name_len,
2745 &ow_inode, &ow_gen);
2749 ret = is_first_ref(sctx, sctx->parent_root,
2750 ow_inode, cur->dir, cur->name,
2755 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2760 ret = send_unlink(sctx, cur->full_path);
2767 * link/move the ref to the new place. If we have an orphan
2768 * inode, move it and update valid_path. If not, link or move
2769 * it depending on the inode mode.
2771 if (is_orphan && !sctx->cur_inode_first_ref_orphan) {
2772 ret = send_rename(sctx, valid_path, cur->full_path);
2776 ret = fs_path_copy(valid_path, cur->full_path);
2780 if (S_ISDIR(sctx->cur_inode_mode)) {
2782 * Dirs can't be linked, so move it. For moved
2783 * dirs, we always have one new and one deleted
2784 * ref. The deleted ref is ignored later.
2786 ret = send_rename(sctx, valid_path,
2790 ret = fs_path_copy(valid_path, cur->full_path);
2794 ret = send_link(sctx, cur->full_path,
2800 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2806 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2808 * Check if we can already rmdir the directory. If not,
2809 * orphanize it. For every dir item inside that gets deleted
2810 * later, we do this check again and rmdir it then if possible.
2811 * See the use of check_dirs for more details.
2813 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2817 ret = send_rmdir(sctx, valid_path);
2820 } else if (!is_orphan) {
2821 ret = orphanize_inode(sctx, sctx->cur_ino,
2822 sctx->cur_inode_gen, valid_path);
2828 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2829 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2834 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2836 * We have a non dir inode. Go through all deleted refs and
2837 * unlink them if they were not already overwritten by other
2840 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2841 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2842 sctx->cur_ino, sctx->cur_inode_gen,
2843 cur->name, cur->name_len);
2848 * In case the inode was moved to a directory
2849 * that was not created yet (see
2850 * __record_new_ref), we can not unlink the ref
2851 * as it will be needed later when the parent
2852 * directory is created, so that we can move in
2853 * the inode to the new dir.
2856 sctx->cur_inode_first_ref_orphan) {
2857 ret = orphanize_inode(sctx,
2859 sctx->cur_inode_gen,
2863 ret = gen_unique_name(sctx,
2865 sctx->cur_inode_gen,
2872 ret = send_unlink(sctx, cur->full_path);
2877 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2884 * If the inode is still orphan, unlink the orphan. This may
2885 * happen when a previous inode did overwrite the first ref
2886 * of this inode and no new refs were added for the current
2888 * We can however not delete the orphan in case the inode relies
2889 * in a directory that was not created yet (see
2892 if (is_orphan && !sctx->cur_inode_first_ref_orphan) {
2893 ret = send_unlink(sctx, valid_path);
2900 * We did collect all parent dirs where cur_inode was once located. We
2901 * now go through all these dirs and check if they are pending for
2902 * deletion and if it's finally possible to perform the rmdir now.
2903 * We also update the inode stats of the parent dirs here.
2905 ULIST_ITER_INIT(&uit);
2906 while ((un = ulist_next(check_dirs, &uit))) {
2907 if (un->val > sctx->cur_ino)
2910 ret = get_cur_inode_state(sctx, un->val, un->aux);
2914 if (ret == inode_state_did_create ||
2915 ret == inode_state_no_change) {
2916 /* TODO delayed utimes */
2917 ret = send_utimes(sctx, un->val, un->aux);
2920 } else if (ret == inode_state_did_delete) {
2921 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
2925 ret = get_cur_path(sctx, un->val, un->aux,
2929 ret = send_rmdir(sctx, valid_path);
2937 * Current inode is now at it's new position, so we must increase
2940 sctx->send_progress = sctx->cur_ino + 1;
2943 * We may have a directory here that has pending refs which could not
2944 * be created before (because the dir did not exist before, see
2945 * __record_new_ref). finish_outoforder_dir will link/move the pending
2948 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_new) {
2949 ret = finish_outoforder_dir(sctx, sctx->cur_ino,
2950 sctx->cur_inode_gen);
2958 free_recorded_refs(sctx);
2959 ulist_free(check_dirs);
2960 fs_path_free(sctx, valid_path);
2964 static int __record_new_ref(int num, u64 dir, int index,
2965 struct fs_path *name,
2969 struct send_ctx *sctx = ctx;
2973 p = fs_path_alloc(sctx);
2977 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
2983 * The parent may be non-existent at this point in time. This happens
2984 * if the ino of the parent dir is higher then the current ino. In this
2985 * case, we can not process this ref until the parent dir is finally
2986 * created. If we reach the parent dir later, process_recorded_refs
2987 * will go through all dir items and process the refs that could not be
2988 * processed before. In case this is the first ref, we set
2989 * cur_inode_first_ref_orphan to 1 to inform process_recorded_refs to
2990 * keep an orphan of the inode so that it later can be used for
2993 ret = is_inode_existent(sctx, dir, gen);
2997 ret = is_first_ref(sctx, sctx->send_root, sctx->cur_ino, dir,
2998 name->start, fs_path_len(name));
3002 sctx->cur_inode_first_ref_orphan = 1;
3007 ret = get_cur_path(sctx, dir, gen, p);
3010 ret = fs_path_add_path(p, name);
3014 ret = record_ref(&sctx->new_refs, dir, gen, p);
3018 fs_path_free(sctx, p);
3022 static int __record_deleted_ref(int num, u64 dir, int index,
3023 struct fs_path *name,
3027 struct send_ctx *sctx = ctx;
3031 p = fs_path_alloc(sctx);
3035 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3040 ret = get_cur_path(sctx, dir, gen, p);
3043 ret = fs_path_add_path(p, name);
3047 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3051 fs_path_free(sctx, p);
3055 static int record_new_ref(struct send_ctx *sctx)
3059 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3060 sctx->cmp_key, 0, __record_new_ref, sctx);
3069 static int record_deleted_ref(struct send_ctx *sctx)
3073 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3074 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3083 struct find_ref_ctx {
3085 struct fs_path *name;
3089 static int __find_iref(int num, u64 dir, int index,
3090 struct fs_path *name,
3093 struct find_ref_ctx *ctx = ctx_;
3095 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3096 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3097 ctx->found_idx = num;
3103 static int find_iref(struct send_ctx *sctx,
3104 struct btrfs_root *root,
3105 struct btrfs_path *path,
3106 struct btrfs_key *key,
3107 u64 dir, struct fs_path *name)
3110 struct find_ref_ctx ctx;
3116 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3120 if (ctx.found_idx == -1)
3123 return ctx.found_idx;
3126 static int __record_changed_new_ref(int num, u64 dir, int index,
3127 struct fs_path *name,
3131 struct send_ctx *sctx = ctx;
3133 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3134 sctx->cmp_key, dir, name);
3136 ret = __record_new_ref(num, dir, index, name, sctx);
3143 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3144 struct fs_path *name,
3148 struct send_ctx *sctx = ctx;
3150 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3153 ret = __record_deleted_ref(num, dir, index, name, sctx);
3160 static int record_changed_ref(struct send_ctx *sctx)
3164 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3165 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3168 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3169 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3179 * Record and process all refs at once. Needed when an inode changes the
3180 * generation number, which means that it was deleted and recreated.
3182 static int process_all_refs(struct send_ctx *sctx,
3183 enum btrfs_compare_tree_result cmd)
3186 struct btrfs_root *root;
3187 struct btrfs_path *path;
3188 struct btrfs_key key;
3189 struct btrfs_key found_key;
3190 struct extent_buffer *eb;
3192 iterate_inode_ref_t cb;
3194 path = alloc_path_for_send();
3198 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3199 root = sctx->send_root;
3200 cb = __record_new_ref;
3201 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3202 root = sctx->parent_root;
3203 cb = __record_deleted_ref;
3208 key.objectid = sctx->cmp_key->objectid;
3209 key.type = BTRFS_INODE_REF_KEY;
3212 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3214 btrfs_release_path(path);
3218 btrfs_release_path(path);
3222 eb = path->nodes[0];
3223 slot = path->slots[0];
3224 btrfs_item_key_to_cpu(eb, &found_key, slot);
3226 if (found_key.objectid != key.objectid ||
3227 found_key.type != key.type) {
3228 btrfs_release_path(path);
3232 ret = iterate_inode_ref(sctx, sctx->parent_root, path,
3233 &found_key, 0, cb, sctx);
3234 btrfs_release_path(path);
3238 key.offset = found_key.offset + 1;
3241 ret = process_recorded_refs(sctx);
3244 btrfs_free_path(path);
3248 static int send_set_xattr(struct send_ctx *sctx,
3249 struct fs_path *path,
3250 const char *name, int name_len,
3251 const char *data, int data_len)
3255 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3259 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3260 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3261 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3263 ret = send_cmd(sctx);
3270 static int send_remove_xattr(struct send_ctx *sctx,
3271 struct fs_path *path,
3272 const char *name, int name_len)
3276 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3280 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3281 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3283 ret = send_cmd(sctx);
3290 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3291 const char *name, int name_len,
3292 const char *data, int data_len,
3296 struct send_ctx *sctx = ctx;
3298 posix_acl_xattr_header dummy_acl;
3300 p = fs_path_alloc(sctx);
3305 * This hack is needed because empty acl's are stored as zero byte
3306 * data in xattrs. Problem with that is, that receiving these zero byte
3307 * acl's will fail later. To fix this, we send a dummy acl list that
3308 * only contains the version number and no entries.
3310 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3311 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3312 if (data_len == 0) {
3313 dummy_acl.a_version =
3314 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3315 data = (char *)&dummy_acl;
3316 data_len = sizeof(dummy_acl);
3320 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3324 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3327 fs_path_free(sctx, p);
3331 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3332 const char *name, int name_len,
3333 const char *data, int data_len,
3337 struct send_ctx *sctx = ctx;
3340 p = fs_path_alloc(sctx);
3344 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3348 ret = send_remove_xattr(sctx, p, name, name_len);
3351 fs_path_free(sctx, p);
3355 static int process_new_xattr(struct send_ctx *sctx)
3359 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3360 sctx->cmp_key, __process_new_xattr, sctx);
3365 static int process_deleted_xattr(struct send_ctx *sctx)
3369 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3370 sctx->cmp_key, __process_deleted_xattr, sctx);
3375 struct find_xattr_ctx {
3383 static int __find_xattr(int num, struct btrfs_key *di_key,
3384 const char *name, int name_len,
3385 const char *data, int data_len,
3386 u8 type, void *vctx)
3388 struct find_xattr_ctx *ctx = vctx;
3390 if (name_len == ctx->name_len &&
3391 strncmp(name, ctx->name, name_len) == 0) {
3392 ctx->found_idx = num;
3393 ctx->found_data_len = data_len;
3394 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3395 if (!ctx->found_data)
3397 memcpy(ctx->found_data, data, data_len);
3403 static int find_xattr(struct send_ctx *sctx,
3404 struct btrfs_root *root,
3405 struct btrfs_path *path,
3406 struct btrfs_key *key,
3407 const char *name, int name_len,
3408 char **data, int *data_len)
3411 struct find_xattr_ctx ctx;
3414 ctx.name_len = name_len;
3416 ctx.found_data = NULL;
3417 ctx.found_data_len = 0;
3419 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3423 if (ctx.found_idx == -1)
3426 *data = ctx.found_data;
3427 *data_len = ctx.found_data_len;
3429 kfree(ctx.found_data);
3431 return ctx.found_idx;
3435 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3436 const char *name, int name_len,
3437 const char *data, int data_len,
3441 struct send_ctx *sctx = ctx;
3442 char *found_data = NULL;
3443 int found_data_len = 0;
3444 struct fs_path *p = NULL;
3446 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3447 sctx->cmp_key, name, name_len, &found_data,
3449 if (ret == -ENOENT) {
3450 ret = __process_new_xattr(num, di_key, name, name_len, data,
3451 data_len, type, ctx);
3452 } else if (ret >= 0) {
3453 if (data_len != found_data_len ||
3454 memcmp(data, found_data, data_len)) {
3455 ret = __process_new_xattr(num, di_key, name, name_len,
3456 data, data_len, type, ctx);
3463 fs_path_free(sctx, p);
3467 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3468 const char *name, int name_len,
3469 const char *data, int data_len,
3473 struct send_ctx *sctx = ctx;
3475 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3476 name, name_len, NULL, NULL);
3478 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3479 data_len, type, ctx);
3486 static int process_changed_xattr(struct send_ctx *sctx)
3490 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3491 sctx->cmp_key, __process_changed_new_xattr, sctx);
3494 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3495 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3501 static int process_all_new_xattrs(struct send_ctx *sctx)
3504 struct btrfs_root *root;
3505 struct btrfs_path *path;
3506 struct btrfs_key key;
3507 struct btrfs_key found_key;
3508 struct extent_buffer *eb;
3511 path = alloc_path_for_send();
3515 root = sctx->send_root;
3517 key.objectid = sctx->cmp_key->objectid;
3518 key.type = BTRFS_XATTR_ITEM_KEY;
3521 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3529 eb = path->nodes[0];
3530 slot = path->slots[0];
3531 btrfs_item_key_to_cpu(eb, &found_key, slot);
3533 if (found_key.objectid != key.objectid ||
3534 found_key.type != key.type) {
3539 ret = iterate_dir_item(sctx, root, path, &found_key,
3540 __process_new_xattr, sctx);
3544 btrfs_release_path(path);
3545 key.offset = found_key.offset + 1;
3549 btrfs_free_path(path);
3554 * Read some bytes from the current inode/file and send a write command to
3557 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3561 loff_t pos = offset;
3563 mm_segment_t old_fs;
3565 p = fs_path_alloc(sctx);
3570 * vfs normally only accepts user space buffers for security reasons.
3571 * we only read from the file and also only provide the read_buf buffer
3572 * to vfs. As this buffer does not come from a user space call, it's
3573 * ok to temporary allow kernel space buffers.
3578 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3580 ret = open_cur_inode_file(sctx);
3584 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3591 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3595 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3599 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3600 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3601 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, readed);
3603 ret = send_cmd(sctx);
3607 fs_path_free(sctx, p);
3615 * Send a clone command to user space.
3617 static int send_clone(struct send_ctx *sctx,
3618 u64 offset, u32 len,
3619 struct clone_root *clone_root)
3622 struct btrfs_root *clone_root2 = clone_root->root;
3626 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3627 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3628 clone_root->root->objectid, clone_root->ino,
3629 clone_root->offset);
3631 p = fs_path_alloc(sctx);
3635 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3639 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3643 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3644 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3645 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3647 if (clone_root2 == sctx->send_root) {
3648 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
3649 &gen, NULL, NULL, NULL, NULL);
3652 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3654 ret = get_inode_path(sctx, clone_root2, clone_root->ino, p);
3659 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
3660 clone_root2->root_item.uuid);
3661 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
3662 clone_root2->root_item.ctransid);
3663 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3664 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3665 clone_root->offset);
3667 ret = send_cmd(sctx);
3671 fs_path_free(sctx, p);
3675 static int send_write_or_clone(struct send_ctx *sctx,
3676 struct btrfs_path *path,
3677 struct btrfs_key *key,
3678 struct clone_root *clone_root)
3681 struct btrfs_file_extent_item *ei;
3682 u64 offset = key->offset;
3688 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3689 struct btrfs_file_extent_item);
3690 type = btrfs_file_extent_type(path->nodes[0], ei);
3691 if (type == BTRFS_FILE_EXTENT_INLINE)
3692 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3694 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3696 if (offset + len > sctx->cur_inode_size)
3697 len = sctx->cur_inode_size - offset;
3706 if (l > BTRFS_SEND_READ_SIZE)
3707 l = BTRFS_SEND_READ_SIZE;
3708 ret = send_write(sctx, pos + offset, l);
3717 ret = send_clone(sctx, offset, len, clone_root);
3724 static int is_extent_unchanged(struct send_ctx *sctx,
3725 struct btrfs_path *left_path,
3726 struct btrfs_key *ekey)
3729 struct btrfs_key key;
3730 struct btrfs_path *path = NULL;
3731 struct extent_buffer *eb;
3733 struct btrfs_key found_key;
3734 struct btrfs_file_extent_item *ei;
3739 u64 left_offset_fixed;
3745 path = alloc_path_for_send();
3749 eb = left_path->nodes[0];
3750 slot = left_path->slots[0];
3752 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3753 left_type = btrfs_file_extent_type(eb, ei);
3754 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3755 left_len = btrfs_file_extent_num_bytes(eb, ei);
3756 left_offset = btrfs_file_extent_offset(eb, ei);
3758 if (left_type != BTRFS_FILE_EXTENT_REG) {
3764 * Following comments will refer to these graphics. L is the left
3765 * extents which we are checking at the moment. 1-8 are the right
3766 * extents that we iterate.
3769 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3772 * |--1--|-2b-|...(same as above)
3774 * Alternative situation. Happens on files where extents got split.
3776 * |-----------7-----------|-6-|
3778 * Alternative situation. Happens on files which got larger.
3781 * Nothing follows after 8.
3784 key.objectid = ekey->objectid;
3785 key.type = BTRFS_EXTENT_DATA_KEY;
3786 key.offset = ekey->offset;
3787 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3796 * Handle special case where the right side has no extents at all.
3798 eb = path->nodes[0];
3799 slot = path->slots[0];
3800 btrfs_item_key_to_cpu(eb, &found_key, slot);
3801 if (found_key.objectid != key.objectid ||
3802 found_key.type != key.type) {
3808 * We're now on 2a, 2b or 7.
3811 while (key.offset < ekey->offset + left_len) {
3812 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3813 right_type = btrfs_file_extent_type(eb, ei);
3814 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3815 right_len = btrfs_file_extent_num_bytes(eb, ei);
3816 right_offset = btrfs_file_extent_offset(eb, ei);
3818 if (right_type != BTRFS_FILE_EXTENT_REG) {
3824 * Are we at extent 8? If yes, we know the extent is changed.
3825 * This may only happen on the first iteration.
3827 if (found_key.offset + right_len < ekey->offset) {
3832 left_offset_fixed = left_offset;
3833 if (key.offset < ekey->offset) {
3834 /* Fix the right offset for 2a and 7. */
3835 right_offset += ekey->offset - key.offset;
3837 /* Fix the left offset for all behind 2a and 2b */
3838 left_offset_fixed += key.offset - ekey->offset;
3842 * Check if we have the same extent.
3844 if (left_disknr + left_offset_fixed !=
3845 right_disknr + right_offset) {
3851 * Go to the next extent.
3853 ret = btrfs_next_item(sctx->parent_root, path);
3857 eb = path->nodes[0];
3858 slot = path->slots[0];
3859 btrfs_item_key_to_cpu(eb, &found_key, slot);
3861 if (ret || found_key.objectid != key.objectid ||
3862 found_key.type != key.type) {
3863 key.offset += right_len;
3866 if (found_key.offset != key.offset + right_len) {
3867 /* Should really not happen */
3876 * We're now behind the left extent (treat as unchanged) or at the end
3877 * of the right side (treat as changed).
3879 if (key.offset >= ekey->offset + left_len)
3886 btrfs_free_path(path);
3890 static int process_extent(struct send_ctx *sctx,
3891 struct btrfs_path *path,
3892 struct btrfs_key *key)
3895 struct clone_root *found_clone = NULL;
3897 if (S_ISLNK(sctx->cur_inode_mode))
3900 if (sctx->parent_root && !sctx->cur_inode_new) {
3901 ret = is_extent_unchanged(sctx, path, key);
3910 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3911 sctx->cur_inode_size, &found_clone);
3912 if (ret != -ENOENT && ret < 0)
3915 ret = send_write_or_clone(sctx, path, key, found_clone);
3921 static int process_all_extents(struct send_ctx *sctx)
3924 struct btrfs_root *root;
3925 struct btrfs_path *path;
3926 struct btrfs_key key;
3927 struct btrfs_key found_key;
3928 struct extent_buffer *eb;
3931 root = sctx->send_root;
3932 path = alloc_path_for_send();
3936 key.objectid = sctx->cmp_key->objectid;
3937 key.type = BTRFS_EXTENT_DATA_KEY;
3940 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3948 eb = path->nodes[0];
3949 slot = path->slots[0];
3950 btrfs_item_key_to_cpu(eb, &found_key, slot);
3952 if (found_key.objectid != key.objectid ||
3953 found_key.type != key.type) {
3958 ret = process_extent(sctx, path, &found_key);
3962 btrfs_release_path(path);
3963 key.offset = found_key.offset + 1;
3967 btrfs_free_path(path);
3971 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
3975 if (sctx->cur_ino == 0)
3977 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
3978 sctx->cmp_key->type <= BTRFS_INODE_REF_KEY)
3980 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
3983 ret = process_recorded_refs(sctx);
3989 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4001 ret = process_recorded_refs_if_needed(sctx, at_end);
4005 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4007 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4010 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4011 &left_mode, &left_uid, &left_gid, NULL);
4015 if (!S_ISLNK(sctx->cur_inode_mode)) {
4016 if (!sctx->parent_root || sctx->cur_inode_new) {
4020 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4021 NULL, NULL, &right_mode, &right_uid,
4026 if (left_uid != right_uid || left_gid != right_gid)
4028 if (left_mode != right_mode)
4033 if (S_ISREG(sctx->cur_inode_mode)) {
4034 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4035 sctx->cur_inode_size);
4041 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4042 left_uid, left_gid);
4047 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4054 * Need to send that every time, no matter if it actually changed
4055 * between the two trees as we have done changes to the inode before.
4057 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4065 static int changed_inode(struct send_ctx *sctx,
4066 enum btrfs_compare_tree_result result)
4069 struct btrfs_key *key = sctx->cmp_key;
4070 struct btrfs_inode_item *left_ii = NULL;
4071 struct btrfs_inode_item *right_ii = NULL;
4075 ret = close_cur_inode_file(sctx);
4079 sctx->cur_ino = key->objectid;
4080 sctx->cur_inode_new_gen = 0;
4081 sctx->cur_inode_first_ref_orphan = 0;
4082 sctx->send_progress = sctx->cur_ino;
4084 if (result == BTRFS_COMPARE_TREE_NEW ||
4085 result == BTRFS_COMPARE_TREE_CHANGED) {
4086 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4087 sctx->left_path->slots[0],
4088 struct btrfs_inode_item);
4089 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4092 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4093 sctx->right_path->slots[0],
4094 struct btrfs_inode_item);
4095 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4098 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4099 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4100 sctx->right_path->slots[0],
4101 struct btrfs_inode_item);
4103 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4105 if (left_gen != right_gen)
4106 sctx->cur_inode_new_gen = 1;
4109 if (result == BTRFS_COMPARE_TREE_NEW) {
4110 sctx->cur_inode_gen = left_gen;
4111 sctx->cur_inode_new = 1;
4112 sctx->cur_inode_deleted = 0;
4113 sctx->cur_inode_size = btrfs_inode_size(
4114 sctx->left_path->nodes[0], left_ii);
4115 sctx->cur_inode_mode = btrfs_inode_mode(
4116 sctx->left_path->nodes[0], left_ii);
4117 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4118 ret = send_create_inode(sctx, sctx->left_path,
4120 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4121 sctx->cur_inode_gen = right_gen;
4122 sctx->cur_inode_new = 0;
4123 sctx->cur_inode_deleted = 1;
4124 sctx->cur_inode_size = btrfs_inode_size(
4125 sctx->right_path->nodes[0], right_ii);
4126 sctx->cur_inode_mode = btrfs_inode_mode(
4127 sctx->right_path->nodes[0], right_ii);
4128 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4129 if (sctx->cur_inode_new_gen) {
4130 sctx->cur_inode_gen = right_gen;
4131 sctx->cur_inode_new = 0;
4132 sctx->cur_inode_deleted = 1;
4133 sctx->cur_inode_size = btrfs_inode_size(
4134 sctx->right_path->nodes[0], right_ii);
4135 sctx->cur_inode_mode = btrfs_inode_mode(
4136 sctx->right_path->nodes[0], right_ii);
4137 ret = process_all_refs(sctx,
4138 BTRFS_COMPARE_TREE_DELETED);
4142 sctx->cur_inode_gen = left_gen;
4143 sctx->cur_inode_new = 1;
4144 sctx->cur_inode_deleted = 0;
4145 sctx->cur_inode_size = btrfs_inode_size(
4146 sctx->left_path->nodes[0], left_ii);
4147 sctx->cur_inode_mode = btrfs_inode_mode(
4148 sctx->left_path->nodes[0], left_ii);
4149 ret = send_create_inode(sctx, sctx->left_path,
4154 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4157 ret = process_all_extents(sctx);
4160 ret = process_all_new_xattrs(sctx);
4164 sctx->cur_inode_gen = left_gen;
4165 sctx->cur_inode_new = 0;
4166 sctx->cur_inode_new_gen = 0;
4167 sctx->cur_inode_deleted = 0;
4168 sctx->cur_inode_size = btrfs_inode_size(
4169 sctx->left_path->nodes[0], left_ii);
4170 sctx->cur_inode_mode = btrfs_inode_mode(
4171 sctx->left_path->nodes[0], left_ii);
4179 static int changed_ref(struct send_ctx *sctx,
4180 enum btrfs_compare_tree_result result)
4184 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4186 if (!sctx->cur_inode_new_gen &&
4187 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4188 if (result == BTRFS_COMPARE_TREE_NEW)
4189 ret = record_new_ref(sctx);
4190 else if (result == BTRFS_COMPARE_TREE_DELETED)
4191 ret = record_deleted_ref(sctx);
4192 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4193 ret = record_changed_ref(sctx);
4199 static int changed_xattr(struct send_ctx *sctx,
4200 enum btrfs_compare_tree_result result)
4204 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4206 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4207 if (result == BTRFS_COMPARE_TREE_NEW)
4208 ret = process_new_xattr(sctx);
4209 else if (result == BTRFS_COMPARE_TREE_DELETED)
4210 ret = process_deleted_xattr(sctx);
4211 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4212 ret = process_changed_xattr(sctx);
4218 static int changed_extent(struct send_ctx *sctx,
4219 enum btrfs_compare_tree_result result)
4223 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4225 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4226 if (result != BTRFS_COMPARE_TREE_DELETED)
4227 ret = process_extent(sctx, sctx->left_path,
4235 static int changed_cb(struct btrfs_root *left_root,
4236 struct btrfs_root *right_root,
4237 struct btrfs_path *left_path,
4238 struct btrfs_path *right_path,
4239 struct btrfs_key *key,
4240 enum btrfs_compare_tree_result result,
4244 struct send_ctx *sctx = ctx;
4246 sctx->left_path = left_path;
4247 sctx->right_path = right_path;
4248 sctx->cmp_key = key;
4250 ret = finish_inode_if_needed(sctx, 0);
4254 if (key->type == BTRFS_INODE_ITEM_KEY)
4255 ret = changed_inode(sctx, result);
4256 else if (key->type == BTRFS_INODE_REF_KEY)
4257 ret = changed_ref(sctx, result);
4258 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4259 ret = changed_xattr(sctx, result);
4260 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4261 ret = changed_extent(sctx, result);
4267 static int full_send_tree(struct send_ctx *sctx)
4270 struct btrfs_trans_handle *trans = NULL;
4271 struct btrfs_root *send_root = sctx->send_root;
4272 struct btrfs_key key;
4273 struct btrfs_key found_key;
4274 struct btrfs_path *path;
4275 struct extent_buffer *eb;
4280 path = alloc_path_for_send();
4284 spin_lock(&send_root->root_times_lock);
4285 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4286 spin_unlock(&send_root->root_times_lock);
4288 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4289 key.type = BTRFS_INODE_ITEM_KEY;
4294 * We need to make sure the transaction does not get committed
4295 * while we do anything on commit roots. Join a transaction to prevent
4298 trans = btrfs_join_transaction(send_root);
4299 if (IS_ERR(trans)) {
4300 ret = PTR_ERR(trans);
4306 * Make sure the tree has not changed
4308 spin_lock(&send_root->root_times_lock);
4309 ctransid = btrfs_root_ctransid(&send_root->root_item);
4310 spin_unlock(&send_root->root_times_lock);
4312 if (ctransid != start_ctransid) {
4313 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4314 "send was modified in between. This is "
4315 "probably a bug.\n");
4320 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4328 * When someone want to commit while we iterate, end the
4329 * joined transaction and rejoin.
4331 if (btrfs_should_end_transaction(trans, send_root)) {
4332 ret = btrfs_end_transaction(trans, send_root);
4336 btrfs_release_path(path);
4340 eb = path->nodes[0];
4341 slot = path->slots[0];
4342 btrfs_item_key_to_cpu(eb, &found_key, slot);
4344 ret = changed_cb(send_root, NULL, path, NULL,
4345 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4349 key.objectid = found_key.objectid;
4350 key.type = found_key.type;
4351 key.offset = found_key.offset + 1;
4353 ret = btrfs_next_item(send_root, path);
4363 ret = finish_inode_if_needed(sctx, 1);
4366 btrfs_free_path(path);
4369 ret = btrfs_end_transaction(trans, send_root);
4371 btrfs_end_transaction(trans, send_root);
4376 static int send_subvol(struct send_ctx *sctx)
4380 ret = send_header(sctx);
4384 ret = send_subvol_begin(sctx);
4388 if (sctx->parent_root) {
4389 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4393 ret = finish_inode_if_needed(sctx, 1);
4397 ret = full_send_tree(sctx);
4404 ret = close_cur_inode_file(sctx);
4406 close_cur_inode_file(sctx);
4408 free_recorded_refs(sctx);
4412 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4415 struct btrfs_root *send_root;
4416 struct btrfs_root *clone_root;
4417 struct btrfs_fs_info *fs_info;
4418 struct btrfs_ioctl_send_args *arg = NULL;
4419 struct btrfs_key key;
4420 struct file *filp = NULL;
4421 struct send_ctx *sctx = NULL;
4423 u64 *clone_sources_tmp = NULL;
4425 if (!capable(CAP_SYS_ADMIN))
4428 send_root = BTRFS_I(fdentry(mnt_file)->d_inode)->root;
4429 fs_info = send_root->fs_info;
4431 arg = memdup_user(arg_, sizeof(*arg));
4438 if (!access_ok(VERIFY_READ, arg->clone_sources,
4439 sizeof(*arg->clone_sources *
4440 arg->clone_sources_count))) {
4445 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4451 INIT_LIST_HEAD(&sctx->new_refs);
4452 INIT_LIST_HEAD(&sctx->deleted_refs);
4453 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4454 INIT_LIST_HEAD(&sctx->name_cache_list);
4456 sctx->send_filp = fget(arg->send_fd);
4457 if (IS_ERR(sctx->send_filp)) {
4458 ret = PTR_ERR(sctx->send_filp);
4462 sctx->mnt = mnt_file->f_path.mnt;
4464 sctx->send_root = send_root;
4465 sctx->clone_roots_cnt = arg->clone_sources_count;
4467 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4468 sctx->send_buf = vmalloc(sctx->send_max_size);
4469 if (!sctx->send_buf) {
4474 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4475 if (!sctx->read_buf) {
4480 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4481 (arg->clone_sources_count + 1));
4482 if (!sctx->clone_roots) {
4487 if (arg->clone_sources_count) {
4488 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4489 sizeof(*arg->clone_sources));
4490 if (!clone_sources_tmp) {
4495 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4496 arg->clone_sources_count *
4497 sizeof(*arg->clone_sources));
4503 for (i = 0; i < arg->clone_sources_count; i++) {
4504 key.objectid = clone_sources_tmp[i];
4505 key.type = BTRFS_ROOT_ITEM_KEY;
4506 key.offset = (u64)-1;
4507 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4512 if (IS_ERR(clone_root)) {
4513 ret = PTR_ERR(clone_root);
4516 sctx->clone_roots[i].root = clone_root;
4518 vfree(clone_sources_tmp);
4519 clone_sources_tmp = NULL;
4522 if (arg->parent_root) {
4523 key.objectid = arg->parent_root;
4524 key.type = BTRFS_ROOT_ITEM_KEY;
4525 key.offset = (u64)-1;
4526 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4527 if (!sctx->parent_root) {
4534 * Clones from send_root are allowed, but only if the clone source
4535 * is behind the current send position. This is checked while searching
4536 * for possible clone sources.
4538 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4540 /* We do a bsearch later */
4541 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4542 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4545 ret = send_subvol(sctx);
4549 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4552 ret = send_cmd(sctx);
4560 vfree(clone_sources_tmp);
4563 if (sctx->send_filp)
4564 fput(sctx->send_filp);
4566 vfree(sctx->clone_roots);
4567 vfree(sctx->send_buf);
4568 vfree(sctx->read_buf);
4570 name_cache_free(sctx);