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;
115 struct list_head new_refs;
116 struct list_head deleted_refs;
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
122 struct file *cur_inode_filp;
126 struct name_cache_entry {
127 struct list_head list;
129 * radix_tree has only 32bit entries but we need to handle 64bit inums.
130 * We use the lower 32bit of the 64bit inum to store it in the tree. If
131 * more then one inum would fall into the same entry, we use radix_list
132 * to store the additional entries. radix_list is also used to store
133 * entries where two entries have the same inum but different
136 struct list_head radix_list;
142 int need_later_update;
147 static void fs_path_reset(struct fs_path *p)
150 p->start = p->buf + p->buf_len - 1;
160 static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
164 p = kmalloc(sizeof(*p), GFP_NOFS);
169 p->buf = p->inline_buf;
170 p->buf_len = FS_PATH_INLINE_SIZE;
175 static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
179 p = fs_path_alloc(sctx);
187 static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
191 if (p->buf != p->inline_buf) {
200 static int fs_path_len(struct fs_path *p)
202 return p->end - p->start;
205 static int fs_path_ensure_buf(struct fs_path *p, int len)
213 if (p->buf_len >= len)
216 path_len = p->end - p->start;
217 old_buf_len = p->buf_len;
218 len = PAGE_ALIGN(len);
220 if (p->buf == p->inline_buf) {
221 tmp_buf = kmalloc(len, GFP_NOFS);
223 tmp_buf = vmalloc(len);
228 memcpy(tmp_buf, p->buf, p->buf_len);
232 if (p->virtual_mem) {
233 tmp_buf = vmalloc(len);
236 memcpy(tmp_buf, p->buf, p->buf_len);
239 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
241 tmp_buf = vmalloc(len);
244 memcpy(tmp_buf, p->buf, p->buf_len);
253 tmp_buf = p->buf + old_buf_len - path_len - 1;
254 p->end = p->buf + p->buf_len - 1;
255 p->start = p->end - path_len;
256 memmove(p->start, tmp_buf, path_len + 1);
259 p->end = p->start + path_len;
264 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
269 new_len = p->end - p->start + name_len;
270 if (p->start != p->end)
272 ret = fs_path_ensure_buf(p, new_len);
277 if (p->start != p->end)
279 p->start -= name_len;
280 p->prepared = p->start;
282 if (p->start != p->end)
284 p->prepared = p->end;
293 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
297 ret = fs_path_prepare_for_add(p, name_len);
300 memcpy(p->prepared, name, name_len);
307 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
311 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
314 memcpy(p->prepared, p2->start, p2->end - p2->start);
321 static int fs_path_add_from_extent_buffer(struct fs_path *p,
322 struct extent_buffer *eb,
323 unsigned long off, int len)
327 ret = fs_path_prepare_for_add(p, len);
331 read_extent_buffer(eb, p->prepared, off, len);
339 static void fs_path_remove(struct fs_path *p)
342 while (p->start != p->end && *p->end != '/')
348 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
352 p->reversed = from->reversed;
355 ret = fs_path_add_path(p, from);
361 static void fs_path_unreverse(struct fs_path *p)
370 len = p->end - p->start;
372 p->end = p->start + len;
373 memmove(p->start, tmp, len + 1);
377 static struct btrfs_path *alloc_path_for_send(void)
379 struct btrfs_path *path;
381 path = btrfs_alloc_path();
384 path->search_commit_root = 1;
385 path->skip_locking = 1;
389 int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
399 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
400 /* TODO handle that correctly */
401 /*if (ret == -ERESTARTSYS) {
420 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
422 struct btrfs_tlv_header *hdr;
423 int total_len = sizeof(*hdr) + len;
424 int left = sctx->send_max_size - sctx->send_size;
426 if (unlikely(left < total_len))
429 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
430 hdr->tlv_type = cpu_to_le16(attr);
431 hdr->tlv_len = cpu_to_le16(len);
432 memcpy(hdr + 1, data, len);
433 sctx->send_size += total_len;
439 static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
441 return tlv_put(sctx, attr, &value, sizeof(value));
444 static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
446 __le16 tmp = cpu_to_le16(value);
447 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
450 static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
452 __le32 tmp = cpu_to_le32(value);
453 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
457 static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
459 __le64 tmp = cpu_to_le64(value);
460 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
463 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
464 const char *str, int len)
468 return tlv_put(sctx, attr, str, len);
471 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
474 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
478 static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
481 struct btrfs_timespec bts;
482 bts.sec = cpu_to_le64(ts->tv_sec);
483 bts.nsec = cpu_to_le32(ts->tv_nsec);
484 return tlv_put(sctx, attr, &bts, sizeof(bts));
488 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
489 struct extent_buffer *eb,
490 struct btrfs_timespec *ts)
492 struct btrfs_timespec bts;
493 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
494 return tlv_put(sctx, attr, &bts, sizeof(bts));
498 #define TLV_PUT(sctx, attrtype, attrlen, data) \
500 ret = tlv_put(sctx, attrtype, attrlen, data); \
502 goto tlv_put_failure; \
505 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
507 ret = tlv_put_u##bits(sctx, attrtype, value); \
509 goto tlv_put_failure; \
512 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
513 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
514 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
515 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
516 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
518 ret = tlv_put_string(sctx, attrtype, str, len); \
520 goto tlv_put_failure; \
522 #define TLV_PUT_PATH(sctx, attrtype, p) \
524 ret = tlv_put_string(sctx, attrtype, p->start, \
525 p->end - p->start); \
527 goto tlv_put_failure; \
529 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
531 ret = tlv_put_uuid(sctx, attrtype, uuid); \
533 goto tlv_put_failure; \
535 #define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
537 ret = tlv_put_timespec(sctx, attrtype, ts); \
539 goto tlv_put_failure; \
541 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
543 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
545 goto tlv_put_failure; \
548 static int send_header(struct send_ctx *sctx)
550 struct btrfs_stream_header hdr;
552 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
553 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
555 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
560 * For each command/item we want to send to userspace, we call this function.
562 static int begin_cmd(struct send_ctx *sctx, int cmd)
564 struct btrfs_cmd_header *hdr;
566 if (!sctx->send_buf) {
571 BUG_ON(sctx->send_size);
573 sctx->send_size += sizeof(*hdr);
574 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
575 hdr->cmd = cpu_to_le16(cmd);
580 static int send_cmd(struct send_ctx *sctx)
583 struct btrfs_cmd_header *hdr;
586 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
587 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
590 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
591 hdr->crc = cpu_to_le32(crc);
593 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
596 sctx->total_send_size += sctx->send_size;
597 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
604 * Sends a move instruction to user space
606 static int send_rename(struct send_ctx *sctx,
607 struct fs_path *from, struct fs_path *to)
611 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
613 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
617 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
618 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
620 ret = send_cmd(sctx);
628 * Sends a link instruction to user space
630 static int send_link(struct send_ctx *sctx,
631 struct fs_path *path, struct fs_path *lnk)
635 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
637 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
641 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
642 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
644 ret = send_cmd(sctx);
652 * Sends an unlink instruction to user space
654 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
658 verbose_printk("btrfs: send_unlink %s\n", path->start);
660 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
664 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
666 ret = send_cmd(sctx);
674 * Sends a rmdir instruction to user space
676 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
680 verbose_printk("btrfs: send_rmdir %s\n", path->start);
682 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
686 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
688 ret = send_cmd(sctx);
696 * Helper function to retrieve some fields from an inode item.
698 static int get_inode_info(struct btrfs_root *root,
699 u64 ino, u64 *size, u64 *gen,
700 u64 *mode, u64 *uid, u64 *gid,
704 struct btrfs_inode_item *ii;
705 struct btrfs_key key;
706 struct btrfs_path *path;
708 path = alloc_path_for_send();
713 key.type = BTRFS_INODE_ITEM_KEY;
715 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
723 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
724 struct btrfs_inode_item);
726 *size = btrfs_inode_size(path->nodes[0], ii);
728 *gen = btrfs_inode_generation(path->nodes[0], ii);
730 *mode = btrfs_inode_mode(path->nodes[0], ii);
732 *uid = btrfs_inode_uid(path->nodes[0], ii);
734 *gid = btrfs_inode_gid(path->nodes[0], ii);
736 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
739 btrfs_free_path(path);
743 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
748 * Helper function to iterate the entries in ONE btrfs_inode_ref.
749 * The iterate callback may return a non zero value to stop iteration. This can
750 * be a negative value for error codes or 1 to simply stop it.
752 * path must point to the INODE_REF when called.
754 static int iterate_inode_ref(struct send_ctx *sctx,
755 struct btrfs_root *root, struct btrfs_path *path,
756 struct btrfs_key *found_key, int resolve,
757 iterate_inode_ref_t iterate, void *ctx)
759 struct extent_buffer *eb;
760 struct btrfs_item *item;
761 struct btrfs_inode_ref *iref;
762 struct btrfs_path *tmp_path;
774 p = fs_path_alloc_reversed(sctx);
778 tmp_path = alloc_path_for_send();
780 fs_path_free(sctx, p);
785 slot = path->slots[0];
786 item = btrfs_item_nr(eb, slot);
787 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
790 total = btrfs_item_size(eb, item);
793 while (cur < total) {
796 name_len = btrfs_inode_ref_name_len(eb, iref);
797 index = btrfs_inode_ref_index(eb, iref);
799 start = btrfs_iref_to_path(root, tmp_path, iref, eb,
800 found_key->offset, p->buf,
803 ret = PTR_ERR(start);
806 if (start < p->buf) {
807 /* overflow , try again with larger buffer */
808 ret = fs_path_ensure_buf(p,
809 p->buf_len + p->buf - start);
812 start = btrfs_iref_to_path(root, tmp_path, iref,
813 eb, found_key->offset, p->buf,
816 ret = PTR_ERR(start);
819 BUG_ON(start < p->buf);
823 ret = fs_path_add_from_extent_buffer(p, eb,
824 (unsigned long)(iref + 1), name_len);
830 len = sizeof(*iref) + name_len;
831 iref = (struct btrfs_inode_ref *)((char *)iref + len);
834 ret = iterate(num, found_key->offset, index, p, ctx);
842 btrfs_free_path(tmp_path);
843 fs_path_free(sctx, p);
847 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
848 const char *name, int name_len,
849 const char *data, int data_len,
853 * Helper function to iterate the entries in ONE btrfs_dir_item.
854 * The iterate callback may return a non zero value to stop iteration. This can
855 * be a negative value for error codes or 1 to simply stop it.
857 * path must point to the dir item when called.
859 static int iterate_dir_item(struct send_ctx *sctx,
860 struct btrfs_root *root, struct btrfs_path *path,
861 struct btrfs_key *found_key,
862 iterate_dir_item_t iterate, void *ctx)
865 struct extent_buffer *eb;
866 struct btrfs_item *item;
867 struct btrfs_dir_item *di;
868 struct btrfs_key di_key;
883 buf = kmalloc(buf_len, GFP_NOFS);
890 slot = path->slots[0];
891 item = btrfs_item_nr(eb, slot);
892 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
895 total = btrfs_item_size(eb, item);
898 while (cur < total) {
899 name_len = btrfs_dir_name_len(eb, di);
900 data_len = btrfs_dir_data_len(eb, di);
901 type = btrfs_dir_type(eb, di);
902 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
904 if (name_len + data_len > buf_len) {
905 buf_len = PAGE_ALIGN(name_len + data_len);
907 buf2 = vmalloc(buf_len);
914 buf2 = krealloc(buf, buf_len, GFP_NOFS);
916 buf2 = vmalloc(buf_len);
930 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
931 name_len + data_len);
933 len = sizeof(*di) + name_len + data_len;
934 di = (struct btrfs_dir_item *)((char *)di + len);
937 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
938 data_len, type, ctx);
957 static int __copy_first_ref(int num, u64 dir, int index,
958 struct fs_path *p, void *ctx)
961 struct fs_path *pt = ctx;
963 ret = fs_path_copy(pt, p);
967 /* we want the first only */
972 * Retrieve the first path of an inode. If an inode has more then one
973 * ref/hardlink, this is ignored.
975 static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
976 u64 ino, struct fs_path *path)
979 struct btrfs_key key, found_key;
980 struct btrfs_path *p;
982 p = alloc_path_for_send();
989 key.type = BTRFS_INODE_REF_KEY;
992 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
999 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1000 if (found_key.objectid != ino ||
1001 found_key.type != BTRFS_INODE_REF_KEY) {
1006 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1007 __copy_first_ref, path);
1017 struct backref_ctx {
1018 struct send_ctx *sctx;
1020 /* number of total found references */
1024 * used for clones found in send_root. clones found behind cur_objectid
1025 * and cur_offset are not considered as allowed clones.
1030 /* may be truncated in case it's the last extent in a file */
1033 /* Just to check for bugs in backref resolving */
1037 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1039 u64 root = (u64)(uintptr_t)key;
1040 struct clone_root *cr = (struct clone_root *)elt;
1042 if (root < cr->root->objectid)
1044 if (root > cr->root->objectid)
1049 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1051 struct clone_root *cr1 = (struct clone_root *)e1;
1052 struct clone_root *cr2 = (struct clone_root *)e2;
1054 if (cr1->root->objectid < cr2->root->objectid)
1056 if (cr1->root->objectid > cr2->root->objectid)
1062 * Called for every backref that is found for the current extent.
1063 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1065 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1067 struct backref_ctx *bctx = ctx_;
1068 struct clone_root *found;
1072 /* First check if the root is in the list of accepted clone sources */
1073 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1074 bctx->sctx->clone_roots_cnt,
1075 sizeof(struct clone_root),
1076 __clone_root_cmp_bsearch);
1080 if (found->root == bctx->sctx->send_root &&
1081 ino == bctx->cur_objectid &&
1082 offset == bctx->cur_offset) {
1083 bctx->found_itself = 1;
1087 * There are inodes that have extents that lie behind its i_size. Don't
1088 * accept clones from these extents.
1090 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1095 if (offset + bctx->extent_len > i_size)
1099 * Make sure we don't consider clones from send_root that are
1100 * behind the current inode/offset.
1102 if (found->root == bctx->sctx->send_root) {
1104 * TODO for the moment we don't accept clones from the inode
1105 * that is currently send. We may change this when
1106 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1109 if (ino >= bctx->cur_objectid)
1112 if (ino > bctx->cur_objectid)
1114 if (offset + bctx->extent_len > bctx->cur_offset)
1120 found->found_refs++;
1121 if (ino < found->ino) {
1123 found->offset = offset;
1124 } else if (found->ino == ino) {
1126 * same extent found more then once in the same file.
1128 if (found->offset > offset + bctx->extent_len)
1129 found->offset = offset;
1136 * Given an inode, offset and extent item, it finds a good clone for a clone
1137 * instruction. Returns -ENOENT when none could be found. The function makes
1138 * sure that the returned clone is usable at the point where sending is at the
1139 * moment. This means, that no clones are accepted which lie behind the current
1142 * path must point to the extent item when called.
1144 static int find_extent_clone(struct send_ctx *sctx,
1145 struct btrfs_path *path,
1146 u64 ino, u64 data_offset,
1148 struct clone_root **found)
1155 u64 extent_item_pos;
1157 struct btrfs_file_extent_item *fi;
1158 struct extent_buffer *eb = path->nodes[0];
1159 struct backref_ctx *backref_ctx = NULL;
1160 struct clone_root *cur_clone_root;
1161 struct btrfs_key found_key;
1162 struct btrfs_path *tmp_path;
1166 tmp_path = alloc_path_for_send();
1170 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1176 if (data_offset >= ino_size) {
1178 * There may be extents that lie behind the file's size.
1179 * I at least had this in combination with snapshotting while
1180 * writing large files.
1186 fi = btrfs_item_ptr(eb, path->slots[0],
1187 struct btrfs_file_extent_item);
1188 extent_type = btrfs_file_extent_type(eb, fi);
1189 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1193 compressed = btrfs_file_extent_compression(eb, fi);
1195 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1196 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1197 if (disk_byte == 0) {
1201 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1203 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1204 &found_key, &flags);
1205 btrfs_release_path(tmp_path);
1209 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1215 * Setup the clone roots.
1217 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1218 cur_clone_root = sctx->clone_roots + i;
1219 cur_clone_root->ino = (u64)-1;
1220 cur_clone_root->offset = 0;
1221 cur_clone_root->found_refs = 0;
1224 backref_ctx->sctx = sctx;
1225 backref_ctx->found = 0;
1226 backref_ctx->cur_objectid = ino;
1227 backref_ctx->cur_offset = data_offset;
1228 backref_ctx->found_itself = 0;
1229 backref_ctx->extent_len = num_bytes;
1232 * The last extent of a file may be too large due to page alignment.
1233 * We need to adjust extent_len in this case so that the checks in
1234 * __iterate_backrefs work.
1236 if (data_offset + num_bytes >= ino_size)
1237 backref_ctx->extent_len = ino_size - data_offset;
1240 * Now collect all backrefs.
1242 if (compressed == BTRFS_COMPRESS_NONE)
1243 extent_item_pos = logical - found_key.objectid;
1245 extent_item_pos = 0;
1247 extent_item_pos = logical - found_key.objectid;
1248 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1249 found_key.objectid, extent_item_pos, 1,
1250 __iterate_backrefs, backref_ctx);
1255 if (!backref_ctx->found_itself) {
1256 /* found a bug in backref code? */
1258 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1259 "send_root. inode=%llu, offset=%llu, "
1260 "disk_byte=%llu found extent=%llu\n",
1261 ino, data_offset, disk_byte, found_key.objectid);
1265 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1267 "num_bytes=%llu, logical=%llu\n",
1268 data_offset, ino, num_bytes, logical);
1270 if (!backref_ctx->found)
1271 verbose_printk("btrfs: no clones found\n");
1273 cur_clone_root = NULL;
1274 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1275 if (sctx->clone_roots[i].found_refs) {
1276 if (!cur_clone_root)
1277 cur_clone_root = sctx->clone_roots + i;
1278 else if (sctx->clone_roots[i].root == sctx->send_root)
1279 /* prefer clones from send_root over others */
1280 cur_clone_root = sctx->clone_roots + i;
1285 if (cur_clone_root) {
1286 *found = cur_clone_root;
1293 btrfs_free_path(tmp_path);
1298 static int read_symlink(struct send_ctx *sctx,
1299 struct btrfs_root *root,
1301 struct fs_path *dest)
1304 struct btrfs_path *path;
1305 struct btrfs_key key;
1306 struct btrfs_file_extent_item *ei;
1312 path = alloc_path_for_send();
1317 key.type = BTRFS_EXTENT_DATA_KEY;
1319 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1324 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1325 struct btrfs_file_extent_item);
1326 type = btrfs_file_extent_type(path->nodes[0], ei);
1327 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1328 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1329 BUG_ON(compression);
1331 off = btrfs_file_extent_inline_start(ei);
1332 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1334 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1337 btrfs_free_path(path);
1342 * Helper function to generate a file name that is unique in the root of
1343 * send_root and parent_root. This is used to generate names for orphan inodes.
1345 static int gen_unique_name(struct send_ctx *sctx,
1347 struct fs_path *dest)
1350 struct btrfs_path *path;
1351 struct btrfs_dir_item *di;
1356 path = alloc_path_for_send();
1361 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1363 if (len >= sizeof(tmp)) {
1364 /* should really not happen */
1369 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1370 path, BTRFS_FIRST_FREE_OBJECTID,
1371 tmp, strlen(tmp), 0);
1372 btrfs_release_path(path);
1378 /* not unique, try again */
1383 if (!sctx->parent_root) {
1389 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1390 path, BTRFS_FIRST_FREE_OBJECTID,
1391 tmp, strlen(tmp), 0);
1392 btrfs_release_path(path);
1398 /* not unique, try again */
1406 ret = fs_path_add(dest, tmp, strlen(tmp));
1409 btrfs_free_path(path);
1414 inode_state_no_change,
1415 inode_state_will_create,
1416 inode_state_did_create,
1417 inode_state_will_delete,
1418 inode_state_did_delete,
1421 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1429 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1431 if (ret < 0 && ret != -ENOENT)
1435 if (!sctx->parent_root) {
1436 right_ret = -ENOENT;
1438 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1439 NULL, NULL, NULL, NULL);
1440 if (ret < 0 && ret != -ENOENT)
1445 if (!left_ret && !right_ret) {
1446 if (left_gen == gen && right_gen == gen) {
1447 ret = inode_state_no_change;
1448 } else if (left_gen == gen) {
1449 if (ino < sctx->send_progress)
1450 ret = inode_state_did_create;
1452 ret = inode_state_will_create;
1453 } else if (right_gen == gen) {
1454 if (ino < sctx->send_progress)
1455 ret = inode_state_did_delete;
1457 ret = inode_state_will_delete;
1461 } else if (!left_ret) {
1462 if (left_gen == gen) {
1463 if (ino < sctx->send_progress)
1464 ret = inode_state_did_create;
1466 ret = inode_state_will_create;
1470 } else if (!right_ret) {
1471 if (right_gen == gen) {
1472 if (ino < sctx->send_progress)
1473 ret = inode_state_did_delete;
1475 ret = inode_state_will_delete;
1487 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1491 ret = get_cur_inode_state(sctx, ino, gen);
1495 if (ret == inode_state_no_change ||
1496 ret == inode_state_did_create ||
1497 ret == inode_state_will_delete)
1507 * Helper function to lookup a dir item in a dir.
1509 static int lookup_dir_item_inode(struct btrfs_root *root,
1510 u64 dir, const char *name, int name_len,
1515 struct btrfs_dir_item *di;
1516 struct btrfs_key key;
1517 struct btrfs_path *path;
1519 path = alloc_path_for_send();
1523 di = btrfs_lookup_dir_item(NULL, root, path,
1524 dir, name, name_len, 0);
1533 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1534 *found_inode = key.objectid;
1535 *found_type = btrfs_dir_type(path->nodes[0], di);
1538 btrfs_free_path(path);
1543 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1544 * generation of the parent dir and the name of the dir entry.
1546 static int get_first_ref(struct send_ctx *sctx,
1547 struct btrfs_root *root, u64 ino,
1548 u64 *dir, u64 *dir_gen, struct fs_path *name)
1551 struct btrfs_key key;
1552 struct btrfs_key found_key;
1553 struct btrfs_path *path;
1554 struct btrfs_inode_ref *iref;
1557 path = alloc_path_for_send();
1562 key.type = BTRFS_INODE_REF_KEY;
1565 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1569 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1571 if (ret || found_key.objectid != key.objectid ||
1572 found_key.type != key.type) {
1577 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1578 struct btrfs_inode_ref);
1579 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1580 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1581 (unsigned long)(iref + 1), len);
1584 btrfs_release_path(path);
1586 ret = get_inode_info(root, found_key.offset, NULL, dir_gen, NULL, NULL,
1591 *dir = found_key.offset;
1594 btrfs_free_path(path);
1598 static int is_first_ref(struct send_ctx *sctx,
1599 struct btrfs_root *root,
1601 const char *name, int name_len)
1604 struct fs_path *tmp_name;
1608 tmp_name = fs_path_alloc(sctx);
1612 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1616 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1621 ret = !memcmp(tmp_name->start, name, name_len);
1624 fs_path_free(sctx, tmp_name);
1629 * Used by process_recorded_refs to determine if a new ref would overwrite an
1630 * already existing ref. In case it detects an overwrite, it returns the
1631 * inode/gen in who_ino/who_gen.
1632 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1633 * to make sure later references to the overwritten inode are possible.
1634 * Orphanizing is however only required for the first ref of an inode.
1635 * process_recorded_refs does an additional is_first_ref check to see if
1636 * orphanizing is really required.
1638 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1639 const char *name, int name_len,
1640 u64 *who_ino, u64 *who_gen)
1643 u64 other_inode = 0;
1646 if (!sctx->parent_root)
1649 ret = is_inode_existent(sctx, dir, dir_gen);
1653 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1654 &other_inode, &other_type);
1655 if (ret < 0 && ret != -ENOENT)
1663 * Check if the overwritten ref was already processed. If yes, the ref
1664 * was already unlinked/moved, so we can safely assume that we will not
1665 * overwrite anything at this point in time.
1667 if (other_inode > sctx->send_progress) {
1668 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1669 who_gen, NULL, NULL, NULL, NULL);
1674 *who_ino = other_inode;
1684 * Checks if the ref was overwritten by an already processed inode. This is
1685 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1686 * thus the orphan name needs be used.
1687 * process_recorded_refs also uses it to avoid unlinking of refs that were
1690 static int did_overwrite_ref(struct send_ctx *sctx,
1691 u64 dir, u64 dir_gen,
1692 u64 ino, u64 ino_gen,
1693 const char *name, int name_len)
1700 if (!sctx->parent_root)
1703 ret = is_inode_existent(sctx, dir, dir_gen);
1707 /* check if the ref was overwritten by another ref */
1708 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1709 &ow_inode, &other_type);
1710 if (ret < 0 && ret != -ENOENT)
1713 /* was never and will never be overwritten */
1718 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1723 if (ow_inode == ino && gen == ino_gen) {
1728 /* we know that it is or will be overwritten. check this now */
1729 if (ow_inode < sctx->send_progress)
1739 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1740 * that got overwritten. This is used by process_recorded_refs to determine
1741 * if it has to use the path as returned by get_cur_path or the orphan name.
1743 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1746 struct fs_path *name = NULL;
1750 if (!sctx->parent_root)
1753 name = fs_path_alloc(sctx);
1757 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1761 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1762 name->start, fs_path_len(name));
1765 fs_path_free(sctx, name);
1770 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1771 * so we need to do some special handling in case we have clashes. This function
1772 * takes care of this with the help of name_cache_entry::radix_list.
1773 * In case of error, nce is kfreed.
1775 static int name_cache_insert(struct send_ctx *sctx,
1776 struct name_cache_entry *nce)
1779 struct list_head *nce_head;
1781 nce_head = radix_tree_lookup(&sctx->name_cache,
1782 (unsigned long)nce->ino);
1784 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1787 INIT_LIST_HEAD(nce_head);
1789 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1796 list_add_tail(&nce->radix_list, nce_head);
1797 list_add_tail(&nce->list, &sctx->name_cache_list);
1798 sctx->name_cache_size++;
1803 static void name_cache_delete(struct send_ctx *sctx,
1804 struct name_cache_entry *nce)
1806 struct list_head *nce_head;
1808 nce_head = radix_tree_lookup(&sctx->name_cache,
1809 (unsigned long)nce->ino);
1812 list_del(&nce->radix_list);
1813 list_del(&nce->list);
1814 sctx->name_cache_size--;
1816 if (list_empty(nce_head)) {
1817 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1822 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1825 struct list_head *nce_head;
1826 struct name_cache_entry *cur;
1828 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1832 list_for_each_entry(cur, nce_head, radix_list) {
1833 if (cur->ino == ino && cur->gen == gen)
1840 * Removes the entry from the list and adds it back to the end. This marks the
1841 * entry as recently used so that name_cache_clean_unused does not remove it.
1843 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1845 list_del(&nce->list);
1846 list_add_tail(&nce->list, &sctx->name_cache_list);
1850 * Remove some entries from the beginning of name_cache_list.
1852 static void name_cache_clean_unused(struct send_ctx *sctx)
1854 struct name_cache_entry *nce;
1856 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1859 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1860 nce = list_entry(sctx->name_cache_list.next,
1861 struct name_cache_entry, list);
1862 name_cache_delete(sctx, nce);
1867 static void name_cache_free(struct send_ctx *sctx)
1869 struct name_cache_entry *nce;
1871 while (!list_empty(&sctx->name_cache_list)) {
1872 nce = list_entry(sctx->name_cache_list.next,
1873 struct name_cache_entry, list);
1874 name_cache_delete(sctx, nce);
1880 * Used by get_cur_path for each ref up to the root.
1881 * Returns 0 if it succeeded.
1882 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1883 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1884 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1885 * Returns <0 in case of error.
1887 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1891 struct fs_path *dest)
1895 struct btrfs_path *path = NULL;
1896 struct name_cache_entry *nce = NULL;
1899 * First check if we already did a call to this function with the same
1900 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1901 * return the cached result.
1903 nce = name_cache_search(sctx, ino, gen);
1905 if (ino < sctx->send_progress && nce->need_later_update) {
1906 name_cache_delete(sctx, nce);
1910 name_cache_used(sctx, nce);
1911 *parent_ino = nce->parent_ino;
1912 *parent_gen = nce->parent_gen;
1913 ret = fs_path_add(dest, nce->name, nce->name_len);
1921 path = alloc_path_for_send();
1926 * If the inode is not existent yet, add the orphan name and return 1.
1927 * This should only happen for the parent dir that we determine in
1930 ret = is_inode_existent(sctx, ino, gen);
1935 ret = gen_unique_name(sctx, ino, gen, dest);
1943 * Depending on whether the inode was already processed or not, use
1944 * send_root or parent_root for ref lookup.
1946 if (ino < sctx->send_progress)
1947 ret = get_first_ref(sctx, sctx->send_root, ino,
1948 parent_ino, parent_gen, dest);
1950 ret = get_first_ref(sctx, sctx->parent_root, ino,
1951 parent_ino, parent_gen, dest);
1956 * Check if the ref was overwritten by an inode's ref that was processed
1957 * earlier. If yes, treat as orphan and return 1.
1959 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1960 dest->start, dest->end - dest->start);
1964 fs_path_reset(dest);
1965 ret = gen_unique_name(sctx, ino, gen, dest);
1973 * Store the result of the lookup in the name cache.
1975 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
1983 nce->parent_ino = *parent_ino;
1984 nce->parent_gen = *parent_gen;
1985 nce->name_len = fs_path_len(dest);
1987 strcpy(nce->name, dest->start);
1989 if (ino < sctx->send_progress)
1990 nce->need_later_update = 0;
1992 nce->need_later_update = 1;
1994 nce_ret = name_cache_insert(sctx, nce);
1997 name_cache_clean_unused(sctx);
2000 btrfs_free_path(path);
2005 * Magic happens here. This function returns the first ref to an inode as it
2006 * would look like while receiving the stream at this point in time.
2007 * We walk the path up to the root. For every inode in between, we check if it
2008 * was already processed/sent. If yes, we continue with the parent as found
2009 * in send_root. If not, we continue with the parent as found in parent_root.
2010 * If we encounter an inode that was deleted at this point in time, we use the
2011 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2012 * that were not created yet and overwritten inodes/refs.
2014 * When do we have have orphan inodes:
2015 * 1. When an inode is freshly created and thus no valid refs are available yet
2016 * 2. When a directory lost all it's refs (deleted) but still has dir items
2017 * inside which were not processed yet (pending for move/delete). If anyone
2018 * tried to get the path to the dir items, it would get a path inside that
2020 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2021 * of an unprocessed inode. If in that case the first ref would be
2022 * overwritten, the overwritten inode gets "orphanized". Later when we
2023 * process this overwritten inode, it is restored at a new place by moving
2026 * sctx->send_progress tells this function at which point in time receiving
2029 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2030 struct fs_path *dest)
2033 struct fs_path *name = NULL;
2034 u64 parent_inode = 0;
2038 name = fs_path_alloc(sctx);
2045 fs_path_reset(dest);
2047 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2048 fs_path_reset(name);
2050 ret = __get_cur_name_and_parent(sctx, ino, gen,
2051 &parent_inode, &parent_gen, name);
2057 ret = fs_path_add_path(dest, name);
2066 fs_path_free(sctx, name);
2068 fs_path_unreverse(dest);
2073 * Called for regular files when sending extents data. Opens a struct file
2074 * to read from the file.
2076 static int open_cur_inode_file(struct send_ctx *sctx)
2079 struct btrfs_key key;
2081 struct inode *inode;
2082 struct dentry *dentry;
2086 if (sctx->cur_inode_filp)
2089 key.objectid = sctx->cur_ino;
2090 key.type = BTRFS_INODE_ITEM_KEY;
2093 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2095 if (IS_ERR(inode)) {
2096 ret = PTR_ERR(inode);
2100 dentry = d_obtain_alias(inode);
2102 if (IS_ERR(dentry)) {
2103 ret = PTR_ERR(dentry);
2107 path.mnt = sctx->mnt;
2108 path.dentry = dentry;
2109 filp = dentry_open(&path, O_RDONLY | O_LARGEFILE, current_cred());
2113 ret = PTR_ERR(filp);
2116 sctx->cur_inode_filp = filp;
2120 * no xxxput required here as every vfs op
2121 * does it by itself on failure
2127 * Closes the struct file that was created in open_cur_inode_file
2129 static int close_cur_inode_file(struct send_ctx *sctx)
2133 if (!sctx->cur_inode_filp)
2136 ret = filp_close(sctx->cur_inode_filp, NULL);
2137 sctx->cur_inode_filp = NULL;
2144 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2146 static int send_subvol_begin(struct send_ctx *sctx)
2149 struct btrfs_root *send_root = sctx->send_root;
2150 struct btrfs_root *parent_root = sctx->parent_root;
2151 struct btrfs_path *path;
2152 struct btrfs_key key;
2153 struct btrfs_root_ref *ref;
2154 struct extent_buffer *leaf;
2158 path = alloc_path_for_send();
2162 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2164 btrfs_free_path(path);
2168 key.objectid = send_root->objectid;
2169 key.type = BTRFS_ROOT_BACKREF_KEY;
2172 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2181 leaf = path->nodes[0];
2182 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2183 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2184 key.objectid != send_root->objectid) {
2188 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2189 namelen = btrfs_root_ref_name_len(leaf, ref);
2190 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2191 btrfs_release_path(path);
2194 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2198 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2203 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2204 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2205 sctx->send_root->root_item.uuid);
2206 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2207 sctx->send_root->root_item.ctransid);
2209 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2210 sctx->parent_root->root_item.uuid);
2211 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2212 sctx->parent_root->root_item.ctransid);
2215 ret = send_cmd(sctx);
2219 btrfs_free_path(path);
2224 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2229 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2231 p = fs_path_alloc(sctx);
2235 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2239 ret = get_cur_path(sctx, ino, gen, p);
2242 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2243 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2245 ret = send_cmd(sctx);
2249 fs_path_free(sctx, p);
2253 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2258 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2260 p = fs_path_alloc(sctx);
2264 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2268 ret = get_cur_path(sctx, ino, gen, p);
2271 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2272 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2274 ret = send_cmd(sctx);
2278 fs_path_free(sctx, p);
2282 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2287 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2289 p = fs_path_alloc(sctx);
2293 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2297 ret = get_cur_path(sctx, ino, gen, p);
2300 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2301 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2302 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2304 ret = send_cmd(sctx);
2308 fs_path_free(sctx, p);
2312 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2315 struct fs_path *p = NULL;
2316 struct btrfs_inode_item *ii;
2317 struct btrfs_path *path = NULL;
2318 struct extent_buffer *eb;
2319 struct btrfs_key key;
2322 verbose_printk("btrfs: send_utimes %llu\n", ino);
2324 p = fs_path_alloc(sctx);
2328 path = alloc_path_for_send();
2335 key.type = BTRFS_INODE_ITEM_KEY;
2337 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2341 eb = path->nodes[0];
2342 slot = path->slots[0];
2343 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2345 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2349 ret = get_cur_path(sctx, ino, gen, p);
2352 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2353 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2354 btrfs_inode_atime(ii));
2355 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2356 btrfs_inode_mtime(ii));
2357 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2358 btrfs_inode_ctime(ii));
2359 /* TODO Add otime support when the otime patches get into upstream */
2361 ret = send_cmd(sctx);
2365 fs_path_free(sctx, p);
2366 btrfs_free_path(path);
2371 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2372 * a valid path yet because we did not process the refs yet. So, the inode
2373 * is created as orphan.
2375 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2384 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2386 p = fs_path_alloc(sctx);
2390 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2395 if (S_ISREG(mode)) {
2396 cmd = BTRFS_SEND_C_MKFILE;
2397 } else if (S_ISDIR(mode)) {
2398 cmd = BTRFS_SEND_C_MKDIR;
2399 } else if (S_ISLNK(mode)) {
2400 cmd = BTRFS_SEND_C_SYMLINK;
2401 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2402 cmd = BTRFS_SEND_C_MKNOD;
2403 } else if (S_ISFIFO(mode)) {
2404 cmd = BTRFS_SEND_C_MKFIFO;
2405 } else if (S_ISSOCK(mode)) {
2406 cmd = BTRFS_SEND_C_MKSOCK;
2408 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2409 (int)(mode & S_IFMT));
2414 ret = begin_cmd(sctx, cmd);
2418 ret = gen_unique_name(sctx, ino, gen, p);
2422 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2423 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2425 if (S_ISLNK(mode)) {
2427 ret = read_symlink(sctx, sctx->send_root, ino, p);
2430 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2431 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2432 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2433 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, rdev);
2436 ret = send_cmd(sctx);
2443 fs_path_free(sctx, p);
2448 * We need some special handling for inodes that get processed before the parent
2449 * directory got created. See process_recorded_refs for details.
2450 * This function does the check if we already created the dir out of order.
2452 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2455 struct btrfs_path *path = NULL;
2456 struct btrfs_key key;
2457 struct btrfs_key found_key;
2458 struct btrfs_key di_key;
2459 struct extent_buffer *eb;
2460 struct btrfs_dir_item *di;
2463 path = alloc_path_for_send();
2470 key.type = BTRFS_DIR_INDEX_KEY;
2473 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2478 eb = path->nodes[0];
2479 slot = path->slots[0];
2480 btrfs_item_key_to_cpu(eb, &found_key, slot);
2482 if (ret || found_key.objectid != key.objectid ||
2483 found_key.type != key.type) {
2488 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2489 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2491 if (di_key.objectid < sctx->send_progress) {
2496 key.offset = found_key.offset + 1;
2497 btrfs_release_path(path);
2501 btrfs_free_path(path);
2506 * Only creates the inode if it is:
2507 * 1. Not a directory
2508 * 2. Or a directory which was not created already due to out of order
2509 * directories. See did_create_dir and process_recorded_refs for details.
2511 static int send_create_inode_if_needed(struct send_ctx *sctx)
2515 if (S_ISDIR(sctx->cur_inode_mode)) {
2516 ret = did_create_dir(sctx, sctx->cur_ino);
2525 ret = send_create_inode(sctx, sctx->cur_ino);
2533 struct recorded_ref {
2534 struct list_head list;
2537 struct fs_path *full_path;
2545 * We need to process new refs before deleted refs, but compare_tree gives us
2546 * everything mixed. So we first record all refs and later process them.
2547 * This function is a helper to record one ref.
2549 static int record_ref(struct list_head *head, u64 dir,
2550 u64 dir_gen, struct fs_path *path)
2552 struct recorded_ref *ref;
2555 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2560 ref->dir_gen = dir_gen;
2561 ref->full_path = path;
2563 tmp = strrchr(ref->full_path->start, '/');
2565 ref->name_len = ref->full_path->end - ref->full_path->start;
2566 ref->name = ref->full_path->start;
2567 ref->dir_path_len = 0;
2568 ref->dir_path = ref->full_path->start;
2571 ref->name_len = ref->full_path->end - tmp;
2573 ref->dir_path = ref->full_path->start;
2574 ref->dir_path_len = ref->full_path->end -
2575 ref->full_path->start - 1 - ref->name_len;
2578 list_add_tail(&ref->list, head);
2582 static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
2584 struct recorded_ref *cur;
2586 while (!list_empty(head)) {
2587 cur = list_entry(head->next, struct recorded_ref, list);
2588 fs_path_free(sctx, cur->full_path);
2589 list_del(&cur->list);
2594 static void free_recorded_refs(struct send_ctx *sctx)
2596 __free_recorded_refs(sctx, &sctx->new_refs);
2597 __free_recorded_refs(sctx, &sctx->deleted_refs);
2601 * Renames/moves a file/dir to its orphan name. Used when the first
2602 * ref of an unprocessed inode gets overwritten and for all non empty
2605 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2606 struct fs_path *path)
2609 struct fs_path *orphan;
2611 orphan = fs_path_alloc(sctx);
2615 ret = gen_unique_name(sctx, ino, gen, orphan);
2619 ret = send_rename(sctx, path, orphan);
2622 fs_path_free(sctx, orphan);
2627 * Returns 1 if a directory can be removed at this point in time.
2628 * We check this by iterating all dir items and checking if the inode behind
2629 * the dir item was already processed.
2631 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2634 struct btrfs_root *root = sctx->parent_root;
2635 struct btrfs_path *path;
2636 struct btrfs_key key;
2637 struct btrfs_key found_key;
2638 struct btrfs_key loc;
2639 struct btrfs_dir_item *di;
2642 * Don't try to rmdir the top/root subvolume dir.
2644 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2647 path = alloc_path_for_send();
2652 key.type = BTRFS_DIR_INDEX_KEY;
2656 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2660 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2663 if (ret || found_key.objectid != key.objectid ||
2664 found_key.type != key.type) {
2668 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2669 struct btrfs_dir_item);
2670 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2672 if (loc.objectid > send_progress) {
2677 btrfs_release_path(path);
2678 key.offset = found_key.offset + 1;
2684 btrfs_free_path(path);
2689 * This does all the move/link/unlink/rmdir magic.
2691 static int process_recorded_refs(struct send_ctx *sctx)
2694 struct recorded_ref *cur;
2695 struct recorded_ref *cur2;
2696 struct ulist *check_dirs = NULL;
2697 struct ulist_iterator uit;
2698 struct ulist_node *un;
2699 struct fs_path *valid_path = NULL;
2702 int did_overwrite = 0;
2705 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2708 * This should never happen as the root dir always has the same ref
2709 * which is always '..'
2711 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
2713 valid_path = fs_path_alloc(sctx);
2719 check_dirs = ulist_alloc(GFP_NOFS);
2726 * First, check if the first ref of the current inode was overwritten
2727 * before. If yes, we know that the current inode was already orphanized
2728 * and thus use the orphan name. If not, we can use get_cur_path to
2729 * get the path of the first ref as it would like while receiving at
2730 * this point in time.
2731 * New inodes are always orphan at the beginning, so force to use the
2732 * orphan name in this case.
2733 * The first ref is stored in valid_path and will be updated if it
2734 * gets moved around.
2736 if (!sctx->cur_inode_new) {
2737 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2738 sctx->cur_inode_gen);
2744 if (sctx->cur_inode_new || did_overwrite) {
2745 ret = gen_unique_name(sctx, sctx->cur_ino,
2746 sctx->cur_inode_gen, valid_path);
2751 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2757 list_for_each_entry(cur, &sctx->new_refs, list) {
2759 * We may have refs where the parent directory does not exist
2760 * yet. This happens if the parent directories inum is higher
2761 * the the current inum. To handle this case, we create the
2762 * parent directory out of order. But we need to check if this
2763 * did already happen before due to other refs in the same dir.
2765 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2768 if (ret == inode_state_will_create) {
2771 * First check if any of the current inodes refs did
2772 * already create the dir.
2774 list_for_each_entry(cur2, &sctx->new_refs, list) {
2777 if (cur2->dir == cur->dir) {
2784 * If that did not happen, check if a previous inode
2785 * did already create the dir.
2788 ret = did_create_dir(sctx, cur->dir);
2792 ret = send_create_inode(sctx, cur->dir);
2799 * Check if this new ref would overwrite the first ref of
2800 * another unprocessed inode. If yes, orphanize the
2801 * overwritten inode. If we find an overwritten ref that is
2802 * not the first ref, simply unlink it.
2804 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2805 cur->name, cur->name_len,
2806 &ow_inode, &ow_gen);
2810 ret = is_first_ref(sctx, sctx->parent_root,
2811 ow_inode, cur->dir, cur->name,
2816 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2821 ret = send_unlink(sctx, cur->full_path);
2828 * link/move the ref to the new place. If we have an orphan
2829 * inode, move it and update valid_path. If not, link or move
2830 * it depending on the inode mode.
2833 ret = send_rename(sctx, valid_path, cur->full_path);
2837 ret = fs_path_copy(valid_path, cur->full_path);
2841 if (S_ISDIR(sctx->cur_inode_mode)) {
2843 * Dirs can't be linked, so move it. For moved
2844 * dirs, we always have one new and one deleted
2845 * ref. The deleted ref is ignored later.
2847 ret = send_rename(sctx, valid_path,
2851 ret = fs_path_copy(valid_path, cur->full_path);
2855 ret = send_link(sctx, cur->full_path,
2861 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2867 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2869 * Check if we can already rmdir the directory. If not,
2870 * orphanize it. For every dir item inside that gets deleted
2871 * later, we do this check again and rmdir it then if possible.
2872 * See the use of check_dirs for more details.
2874 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2878 ret = send_rmdir(sctx, valid_path);
2881 } else if (!is_orphan) {
2882 ret = orphanize_inode(sctx, sctx->cur_ino,
2883 sctx->cur_inode_gen, valid_path);
2889 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2890 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2895 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2896 !list_empty(&sctx->deleted_refs)) {
2898 * We have a moved dir. Add the old parent to check_dirs
2900 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2902 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2906 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2908 * We have a non dir inode. Go through all deleted refs and
2909 * unlink them if they were not already overwritten by other
2912 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2913 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2914 sctx->cur_ino, sctx->cur_inode_gen,
2915 cur->name, cur->name_len);
2919 ret = send_unlink(sctx, cur->full_path);
2923 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2930 * If the inode is still orphan, unlink the orphan. This may
2931 * happen when a previous inode did overwrite the first ref
2932 * of this inode and no new refs were added for the current
2933 * inode. Unlinking does not mean that the inode is deleted in
2934 * all cases. There may still be links to this inode in other
2938 ret = send_unlink(sctx, valid_path);
2945 * We did collect all parent dirs where cur_inode was once located. We
2946 * now go through all these dirs and check if they are pending for
2947 * deletion and if it's finally possible to perform the rmdir now.
2948 * We also update the inode stats of the parent dirs here.
2950 ULIST_ITER_INIT(&uit);
2951 while ((un = ulist_next(check_dirs, &uit))) {
2953 * In case we had refs into dirs that were not processed yet,
2954 * we don't need to do the utime and rmdir logic for these dirs.
2955 * The dir will be processed later.
2957 if (un->val > sctx->cur_ino)
2960 ret = get_cur_inode_state(sctx, un->val, un->aux);
2964 if (ret == inode_state_did_create ||
2965 ret == inode_state_no_change) {
2966 /* TODO delayed utimes */
2967 ret = send_utimes(sctx, un->val, un->aux);
2970 } else if (ret == inode_state_did_delete) {
2971 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
2975 ret = get_cur_path(sctx, un->val, un->aux,
2979 ret = send_rmdir(sctx, valid_path);
2989 free_recorded_refs(sctx);
2990 ulist_free(check_dirs);
2991 fs_path_free(sctx, valid_path);
2995 static int __record_new_ref(int num, u64 dir, int index,
2996 struct fs_path *name,
3000 struct send_ctx *sctx = ctx;
3004 p = fs_path_alloc(sctx);
3008 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3013 ret = get_cur_path(sctx, dir, gen, p);
3016 ret = fs_path_add_path(p, name);
3020 ret = record_ref(&sctx->new_refs, dir, gen, p);
3024 fs_path_free(sctx, p);
3028 static int __record_deleted_ref(int num, u64 dir, int index,
3029 struct fs_path *name,
3033 struct send_ctx *sctx = ctx;
3037 p = fs_path_alloc(sctx);
3041 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3046 ret = get_cur_path(sctx, dir, gen, p);
3049 ret = fs_path_add_path(p, name);
3053 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3057 fs_path_free(sctx, p);
3061 static int record_new_ref(struct send_ctx *sctx)
3065 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3066 sctx->cmp_key, 0, __record_new_ref, sctx);
3075 static int record_deleted_ref(struct send_ctx *sctx)
3079 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3080 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3089 struct find_ref_ctx {
3091 struct fs_path *name;
3095 static int __find_iref(int num, u64 dir, int index,
3096 struct fs_path *name,
3099 struct find_ref_ctx *ctx = ctx_;
3101 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3102 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3103 ctx->found_idx = num;
3109 static int find_iref(struct send_ctx *sctx,
3110 struct btrfs_root *root,
3111 struct btrfs_path *path,
3112 struct btrfs_key *key,
3113 u64 dir, struct fs_path *name)
3116 struct find_ref_ctx ctx;
3122 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3126 if (ctx.found_idx == -1)
3129 return ctx.found_idx;
3132 static int __record_changed_new_ref(int num, u64 dir, int index,
3133 struct fs_path *name,
3137 struct send_ctx *sctx = ctx;
3139 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3140 sctx->cmp_key, dir, name);
3142 ret = __record_new_ref(num, dir, index, name, sctx);
3149 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3150 struct fs_path *name,
3154 struct send_ctx *sctx = ctx;
3156 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3159 ret = __record_deleted_ref(num, dir, index, name, sctx);
3166 static int record_changed_ref(struct send_ctx *sctx)
3170 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3171 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3174 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3175 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3185 * Record and process all refs at once. Needed when an inode changes the
3186 * generation number, which means that it was deleted and recreated.
3188 static int process_all_refs(struct send_ctx *sctx,
3189 enum btrfs_compare_tree_result cmd)
3192 struct btrfs_root *root;
3193 struct btrfs_path *path;
3194 struct btrfs_key key;
3195 struct btrfs_key found_key;
3196 struct extent_buffer *eb;
3198 iterate_inode_ref_t cb;
3200 path = alloc_path_for_send();
3204 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3205 root = sctx->send_root;
3206 cb = __record_new_ref;
3207 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3208 root = sctx->parent_root;
3209 cb = __record_deleted_ref;
3214 key.objectid = sctx->cmp_key->objectid;
3215 key.type = BTRFS_INODE_REF_KEY;
3218 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3224 eb = path->nodes[0];
3225 slot = path->slots[0];
3226 btrfs_item_key_to_cpu(eb, &found_key, slot);
3228 if (found_key.objectid != key.objectid ||
3229 found_key.type != key.type)
3232 ret = iterate_inode_ref(sctx, root, path, &found_key, 0, cb,
3234 btrfs_release_path(path);
3238 key.offset = found_key.offset + 1;
3240 btrfs_release_path(path);
3242 ret = process_recorded_refs(sctx);
3245 btrfs_free_path(path);
3249 static int send_set_xattr(struct send_ctx *sctx,
3250 struct fs_path *path,
3251 const char *name, int name_len,
3252 const char *data, int data_len)
3256 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3260 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3261 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3262 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3264 ret = send_cmd(sctx);
3271 static int send_remove_xattr(struct send_ctx *sctx,
3272 struct fs_path *path,
3273 const char *name, int name_len)
3277 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3281 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3282 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3284 ret = send_cmd(sctx);
3291 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3292 const char *name, int name_len,
3293 const char *data, int data_len,
3297 struct send_ctx *sctx = ctx;
3299 posix_acl_xattr_header dummy_acl;
3301 p = fs_path_alloc(sctx);
3306 * This hack is needed because empty acl's are stored as zero byte
3307 * data in xattrs. Problem with that is, that receiving these zero byte
3308 * acl's will fail later. To fix this, we send a dummy acl list that
3309 * only contains the version number and no entries.
3311 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3312 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3313 if (data_len == 0) {
3314 dummy_acl.a_version =
3315 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3316 data = (char *)&dummy_acl;
3317 data_len = sizeof(dummy_acl);
3321 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3325 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3328 fs_path_free(sctx, p);
3332 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3333 const char *name, int name_len,
3334 const char *data, int data_len,
3338 struct send_ctx *sctx = ctx;
3341 p = fs_path_alloc(sctx);
3345 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3349 ret = send_remove_xattr(sctx, p, name, name_len);
3352 fs_path_free(sctx, p);
3356 static int process_new_xattr(struct send_ctx *sctx)
3360 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3361 sctx->cmp_key, __process_new_xattr, sctx);
3366 static int process_deleted_xattr(struct send_ctx *sctx)
3370 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3371 sctx->cmp_key, __process_deleted_xattr, sctx);
3376 struct find_xattr_ctx {
3384 static int __find_xattr(int num, struct btrfs_key *di_key,
3385 const char *name, int name_len,
3386 const char *data, int data_len,
3387 u8 type, void *vctx)
3389 struct find_xattr_ctx *ctx = vctx;
3391 if (name_len == ctx->name_len &&
3392 strncmp(name, ctx->name, name_len) == 0) {
3393 ctx->found_idx = num;
3394 ctx->found_data_len = data_len;
3395 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3396 if (!ctx->found_data)
3398 memcpy(ctx->found_data, data, data_len);
3404 static int find_xattr(struct send_ctx *sctx,
3405 struct btrfs_root *root,
3406 struct btrfs_path *path,
3407 struct btrfs_key *key,
3408 const char *name, int name_len,
3409 char **data, int *data_len)
3412 struct find_xattr_ctx ctx;
3415 ctx.name_len = name_len;
3417 ctx.found_data = NULL;
3418 ctx.found_data_len = 0;
3420 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3424 if (ctx.found_idx == -1)
3427 *data = ctx.found_data;
3428 *data_len = ctx.found_data_len;
3430 kfree(ctx.found_data);
3432 return ctx.found_idx;
3436 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3437 const char *name, int name_len,
3438 const char *data, int data_len,
3442 struct send_ctx *sctx = ctx;
3443 char *found_data = NULL;
3444 int found_data_len = 0;
3445 struct fs_path *p = NULL;
3447 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3448 sctx->cmp_key, name, name_len, &found_data,
3450 if (ret == -ENOENT) {
3451 ret = __process_new_xattr(num, di_key, name, name_len, data,
3452 data_len, type, ctx);
3453 } else if (ret >= 0) {
3454 if (data_len != found_data_len ||
3455 memcmp(data, found_data, data_len)) {
3456 ret = __process_new_xattr(num, di_key, name, name_len,
3457 data, data_len, type, ctx);
3464 fs_path_free(sctx, p);
3468 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3469 const char *name, int name_len,
3470 const char *data, int data_len,
3474 struct send_ctx *sctx = ctx;
3476 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3477 name, name_len, NULL, NULL);
3479 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3480 data_len, type, ctx);
3487 static int process_changed_xattr(struct send_ctx *sctx)
3491 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3492 sctx->cmp_key, __process_changed_new_xattr, sctx);
3495 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3496 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3502 static int process_all_new_xattrs(struct send_ctx *sctx)
3505 struct btrfs_root *root;
3506 struct btrfs_path *path;
3507 struct btrfs_key key;
3508 struct btrfs_key found_key;
3509 struct extent_buffer *eb;
3512 path = alloc_path_for_send();
3516 root = sctx->send_root;
3518 key.objectid = sctx->cmp_key->objectid;
3519 key.type = BTRFS_XATTR_ITEM_KEY;
3522 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3530 eb = path->nodes[0];
3531 slot = path->slots[0];
3532 btrfs_item_key_to_cpu(eb, &found_key, slot);
3534 if (found_key.objectid != key.objectid ||
3535 found_key.type != key.type) {
3540 ret = iterate_dir_item(sctx, root, path, &found_key,
3541 __process_new_xattr, sctx);
3545 btrfs_release_path(path);
3546 key.offset = found_key.offset + 1;
3550 btrfs_free_path(path);
3555 * Read some bytes from the current inode/file and send a write command to
3558 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3562 loff_t pos = offset;
3564 mm_segment_t old_fs;
3566 p = fs_path_alloc(sctx);
3571 * vfs normally only accepts user space buffers for security reasons.
3572 * we only read from the file and also only provide the read_buf buffer
3573 * to vfs. As this buffer does not come from a user space call, it's
3574 * ok to temporary allow kernel space buffers.
3579 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3581 ret = open_cur_inode_file(sctx);
3585 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3592 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3596 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3600 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3601 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3602 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
3604 ret = send_cmd(sctx);
3608 fs_path_free(sctx, p);
3616 * Send a clone command to user space.
3618 static int send_clone(struct send_ctx *sctx,
3619 u64 offset, u32 len,
3620 struct clone_root *clone_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_root->root == 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_root->root,
3655 clone_root->ino, p);
3660 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
3661 clone_root->root->root_item.uuid);
3662 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
3663 clone_root->root->root_item.ctransid);
3664 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3665 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3666 clone_root->offset);
3668 ret = send_cmd(sctx);
3672 fs_path_free(sctx, p);
3676 static int send_write_or_clone(struct send_ctx *sctx,
3677 struct btrfs_path *path,
3678 struct btrfs_key *key,
3679 struct clone_root *clone_root)
3682 struct btrfs_file_extent_item *ei;
3683 u64 offset = key->offset;
3689 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3690 struct btrfs_file_extent_item);
3691 type = btrfs_file_extent_type(path->nodes[0], ei);
3692 if (type == BTRFS_FILE_EXTENT_INLINE) {
3693 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3695 * it is possible the inline item won't cover the whole page,
3696 * but there may be items after this page. Make
3697 * sure to send the whole thing
3699 len = PAGE_CACHE_ALIGN(len);
3701 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3704 if (offset + len > sctx->cur_inode_size)
3705 len = sctx->cur_inode_size - offset;
3714 if (l > BTRFS_SEND_READ_SIZE)
3715 l = BTRFS_SEND_READ_SIZE;
3716 ret = send_write(sctx, pos + offset, l);
3725 ret = send_clone(sctx, offset, len, clone_root);
3732 static int is_extent_unchanged(struct send_ctx *sctx,
3733 struct btrfs_path *left_path,
3734 struct btrfs_key *ekey)
3737 struct btrfs_key key;
3738 struct btrfs_path *path = NULL;
3739 struct extent_buffer *eb;
3741 struct btrfs_key found_key;
3742 struct btrfs_file_extent_item *ei;
3747 u64 left_offset_fixed;
3755 path = alloc_path_for_send();
3759 eb = left_path->nodes[0];
3760 slot = left_path->slots[0];
3761 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3762 left_type = btrfs_file_extent_type(eb, ei);
3764 if (left_type != BTRFS_FILE_EXTENT_REG) {
3768 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3769 left_len = btrfs_file_extent_num_bytes(eb, ei);
3770 left_offset = btrfs_file_extent_offset(eb, ei);
3771 left_gen = btrfs_file_extent_generation(eb, ei);
3774 * Following comments will refer to these graphics. L is the left
3775 * extents which we are checking at the moment. 1-8 are the right
3776 * extents that we iterate.
3779 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3782 * |--1--|-2b-|...(same as above)
3784 * Alternative situation. Happens on files where extents got split.
3786 * |-----------7-----------|-6-|
3788 * Alternative situation. Happens on files which got larger.
3791 * Nothing follows after 8.
3794 key.objectid = ekey->objectid;
3795 key.type = BTRFS_EXTENT_DATA_KEY;
3796 key.offset = ekey->offset;
3797 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3806 * Handle special case where the right side has no extents at all.
3808 eb = path->nodes[0];
3809 slot = path->slots[0];
3810 btrfs_item_key_to_cpu(eb, &found_key, slot);
3811 if (found_key.objectid != key.objectid ||
3812 found_key.type != key.type) {
3818 * We're now on 2a, 2b or 7.
3821 while (key.offset < ekey->offset + left_len) {
3822 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3823 right_type = btrfs_file_extent_type(eb, ei);
3824 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3825 right_len = btrfs_file_extent_num_bytes(eb, ei);
3826 right_offset = btrfs_file_extent_offset(eb, ei);
3827 right_gen = btrfs_file_extent_generation(eb, ei);
3829 if (right_type != BTRFS_FILE_EXTENT_REG) {
3835 * Are we at extent 8? If yes, we know the extent is changed.
3836 * This may only happen on the first iteration.
3838 if (found_key.offset + right_len <= ekey->offset) {
3843 left_offset_fixed = left_offset;
3844 if (key.offset < ekey->offset) {
3845 /* Fix the right offset for 2a and 7. */
3846 right_offset += ekey->offset - key.offset;
3848 /* Fix the left offset for all behind 2a and 2b */
3849 left_offset_fixed += key.offset - ekey->offset;
3853 * Check if we have the same extent.
3855 if (left_disknr != right_disknr ||
3856 left_offset_fixed != right_offset ||
3857 left_gen != right_gen) {
3863 * Go to the next extent.
3865 ret = btrfs_next_item(sctx->parent_root, path);
3869 eb = path->nodes[0];
3870 slot = path->slots[0];
3871 btrfs_item_key_to_cpu(eb, &found_key, slot);
3873 if (ret || found_key.objectid != key.objectid ||
3874 found_key.type != key.type) {
3875 key.offset += right_len;
3878 if (found_key.offset != key.offset + right_len) {
3879 /* Should really not happen */
3888 * We're now behind the left extent (treat as unchanged) or at the end
3889 * of the right side (treat as changed).
3891 if (key.offset >= ekey->offset + left_len)
3898 btrfs_free_path(path);
3902 static int process_extent(struct send_ctx *sctx,
3903 struct btrfs_path *path,
3904 struct btrfs_key *key)
3907 struct clone_root *found_clone = NULL;
3909 if (S_ISLNK(sctx->cur_inode_mode))
3912 if (sctx->parent_root && !sctx->cur_inode_new) {
3913 ret = is_extent_unchanged(sctx, path, key);
3922 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3923 sctx->cur_inode_size, &found_clone);
3924 if (ret != -ENOENT && ret < 0)
3927 ret = send_write_or_clone(sctx, path, key, found_clone);
3933 static int process_all_extents(struct send_ctx *sctx)
3936 struct btrfs_root *root;
3937 struct btrfs_path *path;
3938 struct btrfs_key key;
3939 struct btrfs_key found_key;
3940 struct extent_buffer *eb;
3943 root = sctx->send_root;
3944 path = alloc_path_for_send();
3948 key.objectid = sctx->cmp_key->objectid;
3949 key.type = BTRFS_EXTENT_DATA_KEY;
3952 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3960 eb = path->nodes[0];
3961 slot = path->slots[0];
3962 btrfs_item_key_to_cpu(eb, &found_key, slot);
3964 if (found_key.objectid != key.objectid ||
3965 found_key.type != key.type) {
3970 ret = process_extent(sctx, path, &found_key);
3974 btrfs_release_path(path);
3975 key.offset = found_key.offset + 1;
3979 btrfs_free_path(path);
3983 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
3987 if (sctx->cur_ino == 0)
3989 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
3990 sctx->cmp_key->type <= BTRFS_INODE_REF_KEY)
3992 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
3995 ret = process_recorded_refs(sctx);
4000 * We have processed the refs and thus need to advance send_progress.
4001 * Now, calls to get_cur_xxx will take the updated refs of the current
4002 * inode into account.
4004 sctx->send_progress = sctx->cur_ino + 1;
4010 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4022 ret = process_recorded_refs_if_needed(sctx, at_end);
4026 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4028 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4031 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4032 &left_mode, &left_uid, &left_gid, NULL);
4036 if (!S_ISLNK(sctx->cur_inode_mode)) {
4037 if (!sctx->parent_root || sctx->cur_inode_new) {
4041 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4042 NULL, NULL, &right_mode, &right_uid,
4047 if (left_uid != right_uid || left_gid != right_gid)
4049 if (left_mode != right_mode)
4054 if (S_ISREG(sctx->cur_inode_mode)) {
4055 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4056 sctx->cur_inode_size);
4062 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4063 left_uid, left_gid);
4068 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4075 * Need to send that every time, no matter if it actually changed
4076 * between the two trees as we have done changes to the inode before.
4078 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4086 static int changed_inode(struct send_ctx *sctx,
4087 enum btrfs_compare_tree_result result)
4090 struct btrfs_key *key = sctx->cmp_key;
4091 struct btrfs_inode_item *left_ii = NULL;
4092 struct btrfs_inode_item *right_ii = NULL;
4096 ret = close_cur_inode_file(sctx);
4100 sctx->cur_ino = key->objectid;
4101 sctx->cur_inode_new_gen = 0;
4104 * Set send_progress to current inode. This will tell all get_cur_xxx
4105 * functions that the current inode's refs are not updated yet. Later,
4106 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4108 sctx->send_progress = sctx->cur_ino;
4110 if (result == BTRFS_COMPARE_TREE_NEW ||
4111 result == BTRFS_COMPARE_TREE_CHANGED) {
4112 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4113 sctx->left_path->slots[0],
4114 struct btrfs_inode_item);
4115 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4118 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4119 sctx->right_path->slots[0],
4120 struct btrfs_inode_item);
4121 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4124 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4125 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4126 sctx->right_path->slots[0],
4127 struct btrfs_inode_item);
4129 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4133 * The cur_ino = root dir case is special here. We can't treat
4134 * the inode as deleted+reused because it would generate a
4135 * stream that tries to delete/mkdir the root dir.
4137 if (left_gen != right_gen &&
4138 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4139 sctx->cur_inode_new_gen = 1;
4142 if (result == BTRFS_COMPARE_TREE_NEW) {
4143 sctx->cur_inode_gen = left_gen;
4144 sctx->cur_inode_new = 1;
4145 sctx->cur_inode_deleted = 0;
4146 sctx->cur_inode_size = btrfs_inode_size(
4147 sctx->left_path->nodes[0], left_ii);
4148 sctx->cur_inode_mode = btrfs_inode_mode(
4149 sctx->left_path->nodes[0], left_ii);
4150 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4151 ret = send_create_inode_if_needed(sctx);
4152 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4153 sctx->cur_inode_gen = right_gen;
4154 sctx->cur_inode_new = 0;
4155 sctx->cur_inode_deleted = 1;
4156 sctx->cur_inode_size = btrfs_inode_size(
4157 sctx->right_path->nodes[0], right_ii);
4158 sctx->cur_inode_mode = btrfs_inode_mode(
4159 sctx->right_path->nodes[0], right_ii);
4160 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4162 * We need to do some special handling in case the inode was
4163 * reported as changed with a changed generation number. This
4164 * means that the original inode was deleted and new inode
4165 * reused the same inum. So we have to treat the old inode as
4166 * deleted and the new one as new.
4168 if (sctx->cur_inode_new_gen) {
4170 * First, process the inode as if it was deleted.
4172 sctx->cur_inode_gen = right_gen;
4173 sctx->cur_inode_new = 0;
4174 sctx->cur_inode_deleted = 1;
4175 sctx->cur_inode_size = btrfs_inode_size(
4176 sctx->right_path->nodes[0], right_ii);
4177 sctx->cur_inode_mode = btrfs_inode_mode(
4178 sctx->right_path->nodes[0], right_ii);
4179 ret = process_all_refs(sctx,
4180 BTRFS_COMPARE_TREE_DELETED);
4185 * Now process the inode as if it was new.
4187 sctx->cur_inode_gen = left_gen;
4188 sctx->cur_inode_new = 1;
4189 sctx->cur_inode_deleted = 0;
4190 sctx->cur_inode_size = btrfs_inode_size(
4191 sctx->left_path->nodes[0], left_ii);
4192 sctx->cur_inode_mode = btrfs_inode_mode(
4193 sctx->left_path->nodes[0], left_ii);
4194 ret = send_create_inode_if_needed(sctx);
4198 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4202 * Advance send_progress now as we did not get into
4203 * process_recorded_refs_if_needed in the new_gen case.
4205 sctx->send_progress = sctx->cur_ino + 1;
4208 * Now process all extents and xattrs of the inode as if
4209 * they were all new.
4211 ret = process_all_extents(sctx);
4214 ret = process_all_new_xattrs(sctx);
4218 sctx->cur_inode_gen = left_gen;
4219 sctx->cur_inode_new = 0;
4220 sctx->cur_inode_new_gen = 0;
4221 sctx->cur_inode_deleted = 0;
4222 sctx->cur_inode_size = btrfs_inode_size(
4223 sctx->left_path->nodes[0], left_ii);
4224 sctx->cur_inode_mode = btrfs_inode_mode(
4225 sctx->left_path->nodes[0], left_ii);
4234 * We have to process new refs before deleted refs, but compare_trees gives us
4235 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4236 * first and later process them in process_recorded_refs.
4237 * For the cur_inode_new_gen case, we skip recording completely because
4238 * changed_inode did already initiate processing of refs. The reason for this is
4239 * that in this case, compare_tree actually compares the refs of 2 different
4240 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4241 * refs of the right tree as deleted and all refs of the left tree as new.
4243 static int changed_ref(struct send_ctx *sctx,
4244 enum btrfs_compare_tree_result result)
4248 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4250 if (!sctx->cur_inode_new_gen &&
4251 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4252 if (result == BTRFS_COMPARE_TREE_NEW)
4253 ret = record_new_ref(sctx);
4254 else if (result == BTRFS_COMPARE_TREE_DELETED)
4255 ret = record_deleted_ref(sctx);
4256 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4257 ret = record_changed_ref(sctx);
4264 * Process new/deleted/changed xattrs. We skip processing in the
4265 * cur_inode_new_gen case because changed_inode did already initiate processing
4266 * of xattrs. The reason is the same as in changed_ref
4268 static int changed_xattr(struct send_ctx *sctx,
4269 enum btrfs_compare_tree_result result)
4273 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4275 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4276 if (result == BTRFS_COMPARE_TREE_NEW)
4277 ret = process_new_xattr(sctx);
4278 else if (result == BTRFS_COMPARE_TREE_DELETED)
4279 ret = process_deleted_xattr(sctx);
4280 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4281 ret = process_changed_xattr(sctx);
4288 * Process new/deleted/changed extents. We skip processing in the
4289 * cur_inode_new_gen case because changed_inode did already initiate processing
4290 * of extents. The reason is the same as in changed_ref
4292 static int changed_extent(struct send_ctx *sctx,
4293 enum btrfs_compare_tree_result result)
4297 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4299 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4300 if (result != BTRFS_COMPARE_TREE_DELETED)
4301 ret = process_extent(sctx, sctx->left_path,
4309 * Updates compare related fields in sctx and simply forwards to the actual
4310 * changed_xxx functions.
4312 static int changed_cb(struct btrfs_root *left_root,
4313 struct btrfs_root *right_root,
4314 struct btrfs_path *left_path,
4315 struct btrfs_path *right_path,
4316 struct btrfs_key *key,
4317 enum btrfs_compare_tree_result result,
4321 struct send_ctx *sctx = ctx;
4323 sctx->left_path = left_path;
4324 sctx->right_path = right_path;
4325 sctx->cmp_key = key;
4327 ret = finish_inode_if_needed(sctx, 0);
4331 /* Ignore non-FS objects */
4332 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
4333 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
4336 if (key->type == BTRFS_INODE_ITEM_KEY)
4337 ret = changed_inode(sctx, result);
4338 else if (key->type == BTRFS_INODE_REF_KEY)
4339 ret = changed_ref(sctx, result);
4340 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4341 ret = changed_xattr(sctx, result);
4342 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4343 ret = changed_extent(sctx, result);
4349 static int full_send_tree(struct send_ctx *sctx)
4352 struct btrfs_trans_handle *trans = NULL;
4353 struct btrfs_root *send_root = sctx->send_root;
4354 struct btrfs_key key;
4355 struct btrfs_key found_key;
4356 struct btrfs_path *path;
4357 struct extent_buffer *eb;
4362 path = alloc_path_for_send();
4366 spin_lock(&send_root->root_times_lock);
4367 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4368 spin_unlock(&send_root->root_times_lock);
4370 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4371 key.type = BTRFS_INODE_ITEM_KEY;
4376 * We need to make sure the transaction does not get committed
4377 * while we do anything on commit roots. Join a transaction to prevent
4380 trans = btrfs_join_transaction(send_root);
4381 if (IS_ERR(trans)) {
4382 ret = PTR_ERR(trans);
4388 * Make sure the tree has not changed after re-joining. We detect this
4389 * by comparing start_ctransid and ctransid. They should always match.
4391 spin_lock(&send_root->root_times_lock);
4392 ctransid = btrfs_root_ctransid(&send_root->root_item);
4393 spin_unlock(&send_root->root_times_lock);
4395 if (ctransid != start_ctransid) {
4396 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4397 "send was modified in between. This is "
4398 "probably a bug.\n");
4403 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4411 * When someone want to commit while we iterate, end the
4412 * joined transaction and rejoin.
4414 if (btrfs_should_end_transaction(trans, send_root)) {
4415 ret = btrfs_end_transaction(trans, send_root);
4419 btrfs_release_path(path);
4423 eb = path->nodes[0];
4424 slot = path->slots[0];
4425 btrfs_item_key_to_cpu(eb, &found_key, slot);
4427 ret = changed_cb(send_root, NULL, path, NULL,
4428 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4432 key.objectid = found_key.objectid;
4433 key.type = found_key.type;
4434 key.offset = found_key.offset + 1;
4436 ret = btrfs_next_item(send_root, path);
4446 ret = finish_inode_if_needed(sctx, 1);
4449 btrfs_free_path(path);
4452 ret = btrfs_end_transaction(trans, send_root);
4454 btrfs_end_transaction(trans, send_root);
4459 static int send_subvol(struct send_ctx *sctx)
4463 ret = send_header(sctx);
4467 ret = send_subvol_begin(sctx);
4471 if (sctx->parent_root) {
4472 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4476 ret = finish_inode_if_needed(sctx, 1);
4480 ret = full_send_tree(sctx);
4487 ret = close_cur_inode_file(sctx);
4489 close_cur_inode_file(sctx);
4491 free_recorded_refs(sctx);
4495 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4498 struct btrfs_root *send_root;
4499 struct btrfs_root *clone_root;
4500 struct btrfs_fs_info *fs_info;
4501 struct btrfs_ioctl_send_args *arg = NULL;
4502 struct btrfs_key key;
4503 struct file *filp = NULL;
4504 struct send_ctx *sctx = NULL;
4506 u64 *clone_sources_tmp = NULL;
4508 if (!capable(CAP_SYS_ADMIN))
4511 send_root = BTRFS_I(fdentry(mnt_file)->d_inode)->root;
4512 fs_info = send_root->fs_info;
4514 arg = memdup_user(arg_, sizeof(*arg));
4521 if (!access_ok(VERIFY_READ, arg->clone_sources,
4522 sizeof(*arg->clone_sources *
4523 arg->clone_sources_count))) {
4528 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4534 INIT_LIST_HEAD(&sctx->new_refs);
4535 INIT_LIST_HEAD(&sctx->deleted_refs);
4536 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4537 INIT_LIST_HEAD(&sctx->name_cache_list);
4539 sctx->send_filp = fget(arg->send_fd);
4540 if (IS_ERR(sctx->send_filp)) {
4541 ret = PTR_ERR(sctx->send_filp);
4545 sctx->mnt = mnt_file->f_path.mnt;
4547 sctx->send_root = send_root;
4548 sctx->clone_roots_cnt = arg->clone_sources_count;
4550 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4551 sctx->send_buf = vmalloc(sctx->send_max_size);
4552 if (!sctx->send_buf) {
4557 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4558 if (!sctx->read_buf) {
4563 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4564 (arg->clone_sources_count + 1));
4565 if (!sctx->clone_roots) {
4570 if (arg->clone_sources_count) {
4571 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4572 sizeof(*arg->clone_sources));
4573 if (!clone_sources_tmp) {
4578 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4579 arg->clone_sources_count *
4580 sizeof(*arg->clone_sources));
4586 for (i = 0; i < arg->clone_sources_count; i++) {
4587 key.objectid = clone_sources_tmp[i];
4588 key.type = BTRFS_ROOT_ITEM_KEY;
4589 key.offset = (u64)-1;
4590 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4595 if (IS_ERR(clone_root)) {
4596 ret = PTR_ERR(clone_root);
4599 sctx->clone_roots[i].root = clone_root;
4601 vfree(clone_sources_tmp);
4602 clone_sources_tmp = NULL;
4605 if (arg->parent_root) {
4606 key.objectid = arg->parent_root;
4607 key.type = BTRFS_ROOT_ITEM_KEY;
4608 key.offset = (u64)-1;
4609 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4610 if (!sctx->parent_root) {
4617 * Clones from send_root are allowed, but only if the clone source
4618 * is behind the current send position. This is checked while searching
4619 * for possible clone sources.
4621 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4623 /* We do a bsearch later */
4624 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4625 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4628 ret = send_subvol(sctx);
4632 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4635 ret = send_cmd(sctx);
4643 vfree(clone_sources_tmp);
4646 if (sctx->send_filp)
4647 fput(sctx->send_filp);
4649 vfree(sctx->clone_roots);
4650 vfree(sctx->send_buf);
4651 vfree(sctx->read_buf);
4653 name_cache_free(sctx);