2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/vmalloc.h>
28 #include <linux/string.h>
35 #include "btrfs_inode.h"
36 #include "transaction.h"
38 static int g_verbose = 0;
40 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
43 * A fs_path is a helper to dynamically build path names with unknown size.
44 * It reallocates the internal buffer on demand.
45 * It allows fast adding of path elements on the right side (normal path) and
46 * fast adding to the left side (reversed path). A reversed path can also be
47 * unreversed if needed.
56 unsigned short buf_len:15;
57 unsigned short reversed:1;
61 * Average path length does not exceed 200 bytes, we'll have
62 * better packing in the slab and higher chance to satisfy
63 * a allocation later during send.
68 #define FS_PATH_INLINE_SIZE \
69 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
72 /* reused for each extent */
74 struct btrfs_root *root;
81 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
82 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
85 struct file *send_filp;
91 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
92 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
94 struct btrfs_root *send_root;
95 struct btrfs_root *parent_root;
96 struct clone_root *clone_roots;
99 /* current state of the compare_tree call */
100 struct btrfs_path *left_path;
101 struct btrfs_path *right_path;
102 struct btrfs_key *cmp_key;
105 * infos of the currently processed inode. In case of deleted inodes,
106 * these are the values from the deleted inode.
111 int cur_inode_new_gen;
112 int cur_inode_deleted;
116 u64 cur_inode_last_extent;
120 struct list_head new_refs;
121 struct list_head deleted_refs;
123 struct radix_tree_root name_cache;
124 struct list_head name_cache_list;
127 struct file_ra_state ra;
132 * We process inodes by their increasing order, so if before an
133 * incremental send we reverse the parent/child relationship of
134 * directories such that a directory with a lower inode number was
135 * the parent of a directory with a higher inode number, and the one
136 * becoming the new parent got renamed too, we can't rename/move the
137 * directory with lower inode number when we finish processing it - we
138 * must process the directory with higher inode number first, then
139 * rename/move it and then rename/move the directory with lower inode
140 * number. Example follows.
142 * Tree state when the first send was performed:
154 * Tree state when the second (incremental) send is performed:
163 * The sequence of steps that lead to the second state was:
165 * mv /a/b/c/d /a/b/c2/d2
166 * mv /a/b/c /a/b/c2/d2/cc
168 * "c" has lower inode number, but we can't move it (2nd mv operation)
169 * before we move "d", which has higher inode number.
171 * So we just memorize which move/rename operations must be performed
172 * later when their respective parent is processed and moved/renamed.
175 /* Indexed by parent directory inode number. */
176 struct rb_root pending_dir_moves;
179 * Reverse index, indexed by the inode number of a directory that
180 * is waiting for the move/rename of its immediate parent before its
181 * own move/rename can be performed.
183 struct rb_root waiting_dir_moves;
186 * A directory that is going to be rm'ed might have a child directory
187 * which is in the pending directory moves index above. In this case,
188 * the directory can only be removed after the move/rename of its child
189 * is performed. Example:
209 * Sequence of steps that lead to the send snapshot:
210 * rm -f /a/b/c/foo.txt
212 * mv /a/b/c/x /a/b/YY
215 * When the child is processed, its move/rename is delayed until its
216 * parent is processed (as explained above), but all other operations
217 * like update utimes, chown, chgrp, etc, are performed and the paths
218 * that it uses for those operations must use the orphanized name of
219 * its parent (the directory we're going to rm later), so we need to
220 * memorize that name.
222 * Indexed by the inode number of the directory to be deleted.
224 struct rb_root orphan_dirs;
227 struct pending_dir_move {
229 struct list_head list;
233 struct list_head update_refs;
236 struct waiting_dir_move {
240 * There might be some directory that could not be removed because it
241 * was waiting for this directory inode to be moved first. Therefore
242 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
247 struct orphan_dir_info {
253 struct name_cache_entry {
254 struct list_head list;
256 * radix_tree has only 32bit entries but we need to handle 64bit inums.
257 * We use the lower 32bit of the 64bit inum to store it in the tree. If
258 * more then one inum would fall into the same entry, we use radix_list
259 * to store the additional entries. radix_list is also used to store
260 * entries where two entries have the same inum but different
263 struct list_head radix_list;
269 int need_later_update;
274 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
276 static struct waiting_dir_move *
277 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
279 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
281 static int need_send_hole(struct send_ctx *sctx)
283 return (sctx->parent_root && !sctx->cur_inode_new &&
284 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
285 S_ISREG(sctx->cur_inode_mode));
288 static void fs_path_reset(struct fs_path *p)
291 p->start = p->buf + p->buf_len - 1;
301 static struct fs_path *fs_path_alloc(void)
305 p = kmalloc(sizeof(*p), GFP_NOFS);
309 p->buf = p->inline_buf;
310 p->buf_len = FS_PATH_INLINE_SIZE;
315 static struct fs_path *fs_path_alloc_reversed(void)
327 static void fs_path_free(struct fs_path *p)
331 if (p->buf != p->inline_buf)
336 static int fs_path_len(struct fs_path *p)
338 return p->end - p->start;
341 static int fs_path_ensure_buf(struct fs_path *p, int len)
349 if (p->buf_len >= len)
352 path_len = p->end - p->start;
353 old_buf_len = p->buf_len;
356 * First time the inline_buf does not suffice
358 if (p->buf == p->inline_buf)
359 tmp_buf = kmalloc(len, GFP_NOFS);
361 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
366 * The real size of the buffer is bigger, this will let the fast path
367 * happen most of the time
369 p->buf_len = ksize(p->buf);
372 tmp_buf = p->buf + old_buf_len - path_len - 1;
373 p->end = p->buf + p->buf_len - 1;
374 p->start = p->end - path_len;
375 memmove(p->start, tmp_buf, path_len + 1);
378 p->end = p->start + path_len;
383 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
389 new_len = p->end - p->start + name_len;
390 if (p->start != p->end)
392 ret = fs_path_ensure_buf(p, new_len);
397 if (p->start != p->end)
399 p->start -= name_len;
400 *prepared = p->start;
402 if (p->start != p->end)
413 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
418 ret = fs_path_prepare_for_add(p, name_len, &prepared);
421 memcpy(prepared, name, name_len);
427 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
432 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
435 memcpy(prepared, p2->start, p2->end - p2->start);
441 static int fs_path_add_from_extent_buffer(struct fs_path *p,
442 struct extent_buffer *eb,
443 unsigned long off, int len)
448 ret = fs_path_prepare_for_add(p, len, &prepared);
452 read_extent_buffer(eb, prepared, off, len);
458 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
462 p->reversed = from->reversed;
465 ret = fs_path_add_path(p, from);
471 static void fs_path_unreverse(struct fs_path *p)
480 len = p->end - p->start;
482 p->end = p->start + len;
483 memmove(p->start, tmp, len + 1);
487 static struct btrfs_path *alloc_path_for_send(void)
489 struct btrfs_path *path;
491 path = btrfs_alloc_path();
494 path->search_commit_root = 1;
495 path->skip_locking = 1;
496 path->need_commit_sem = 1;
500 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
510 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
511 /* TODO handle that correctly */
512 /*if (ret == -ERESTARTSYS) {
531 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
533 struct btrfs_tlv_header *hdr;
534 int total_len = sizeof(*hdr) + len;
535 int left = sctx->send_max_size - sctx->send_size;
537 if (unlikely(left < total_len))
540 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
541 hdr->tlv_type = cpu_to_le16(attr);
542 hdr->tlv_len = cpu_to_le16(len);
543 memcpy(hdr + 1, data, len);
544 sctx->send_size += total_len;
549 #define TLV_PUT_DEFINE_INT(bits) \
550 static int tlv_put_u##bits(struct send_ctx *sctx, \
551 u##bits attr, u##bits value) \
553 __le##bits __tmp = cpu_to_le##bits(value); \
554 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
557 TLV_PUT_DEFINE_INT(64)
559 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
560 const char *str, int len)
564 return tlv_put(sctx, attr, str, len);
567 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
570 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
573 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
574 struct extent_buffer *eb,
575 struct btrfs_timespec *ts)
577 struct btrfs_timespec bts;
578 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
579 return tlv_put(sctx, attr, &bts, sizeof(bts));
583 #define TLV_PUT(sctx, attrtype, attrlen, data) \
585 ret = tlv_put(sctx, attrtype, attrlen, data); \
587 goto tlv_put_failure; \
590 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
592 ret = tlv_put_u##bits(sctx, attrtype, value); \
594 goto tlv_put_failure; \
597 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
598 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
599 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
600 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
601 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
603 ret = tlv_put_string(sctx, attrtype, str, len); \
605 goto tlv_put_failure; \
607 #define TLV_PUT_PATH(sctx, attrtype, p) \
609 ret = tlv_put_string(sctx, attrtype, p->start, \
610 p->end - p->start); \
612 goto tlv_put_failure; \
614 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
616 ret = tlv_put_uuid(sctx, attrtype, uuid); \
618 goto tlv_put_failure; \
620 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
622 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
624 goto tlv_put_failure; \
627 static int send_header(struct send_ctx *sctx)
629 struct btrfs_stream_header hdr;
631 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
632 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
634 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
639 * For each command/item we want to send to userspace, we call this function.
641 static int begin_cmd(struct send_ctx *sctx, int cmd)
643 struct btrfs_cmd_header *hdr;
645 if (WARN_ON(!sctx->send_buf))
648 BUG_ON(sctx->send_size);
650 sctx->send_size += sizeof(*hdr);
651 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
652 hdr->cmd = cpu_to_le16(cmd);
657 static int send_cmd(struct send_ctx *sctx)
660 struct btrfs_cmd_header *hdr;
663 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
664 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
667 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
668 hdr->crc = cpu_to_le32(crc);
670 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
673 sctx->total_send_size += sctx->send_size;
674 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
681 * Sends a move instruction to user space
683 static int send_rename(struct send_ctx *sctx,
684 struct fs_path *from, struct fs_path *to)
688 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
690 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
694 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
695 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
697 ret = send_cmd(sctx);
705 * Sends a link instruction to user space
707 static int send_link(struct send_ctx *sctx,
708 struct fs_path *path, struct fs_path *lnk)
712 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
714 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
718 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
719 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
721 ret = send_cmd(sctx);
729 * Sends an unlink instruction to user space
731 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
735 verbose_printk("btrfs: send_unlink %s\n", path->start);
737 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
741 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
743 ret = send_cmd(sctx);
751 * Sends a rmdir instruction to user space
753 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
757 verbose_printk("btrfs: send_rmdir %s\n", path->start);
759 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
763 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
765 ret = send_cmd(sctx);
773 * Helper function to retrieve some fields from an inode item.
775 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
776 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
780 struct btrfs_inode_item *ii;
781 struct btrfs_key key;
784 key.type = BTRFS_INODE_ITEM_KEY;
786 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
793 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
794 struct btrfs_inode_item);
796 *size = btrfs_inode_size(path->nodes[0], ii);
798 *gen = btrfs_inode_generation(path->nodes[0], ii);
800 *mode = btrfs_inode_mode(path->nodes[0], ii);
802 *uid = btrfs_inode_uid(path->nodes[0], ii);
804 *gid = btrfs_inode_gid(path->nodes[0], ii);
806 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
811 static int get_inode_info(struct btrfs_root *root,
812 u64 ino, u64 *size, u64 *gen,
813 u64 *mode, u64 *uid, u64 *gid,
816 struct btrfs_path *path;
819 path = alloc_path_for_send();
822 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
824 btrfs_free_path(path);
828 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
833 * Helper function to iterate the entries in ONE btrfs_inode_ref or
834 * btrfs_inode_extref.
835 * The iterate callback may return a non zero value to stop iteration. This can
836 * be a negative value for error codes or 1 to simply stop it.
838 * path must point to the INODE_REF or INODE_EXTREF when called.
840 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
841 struct btrfs_key *found_key, int resolve,
842 iterate_inode_ref_t iterate, void *ctx)
844 struct extent_buffer *eb = path->nodes[0];
845 struct btrfs_item *item;
846 struct btrfs_inode_ref *iref;
847 struct btrfs_inode_extref *extref;
848 struct btrfs_path *tmp_path;
852 int slot = path->slots[0];
859 unsigned long name_off;
860 unsigned long elem_size;
863 p = fs_path_alloc_reversed();
867 tmp_path = alloc_path_for_send();
874 if (found_key->type == BTRFS_INODE_REF_KEY) {
875 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
876 struct btrfs_inode_ref);
877 item = btrfs_item_nr(slot);
878 total = btrfs_item_size(eb, item);
879 elem_size = sizeof(*iref);
881 ptr = btrfs_item_ptr_offset(eb, slot);
882 total = btrfs_item_size_nr(eb, slot);
883 elem_size = sizeof(*extref);
886 while (cur < total) {
889 if (found_key->type == BTRFS_INODE_REF_KEY) {
890 iref = (struct btrfs_inode_ref *)(ptr + cur);
891 name_len = btrfs_inode_ref_name_len(eb, iref);
892 name_off = (unsigned long)(iref + 1);
893 index = btrfs_inode_ref_index(eb, iref);
894 dir = found_key->offset;
896 extref = (struct btrfs_inode_extref *)(ptr + cur);
897 name_len = btrfs_inode_extref_name_len(eb, extref);
898 name_off = (unsigned long)&extref->name;
899 index = btrfs_inode_extref_index(eb, extref);
900 dir = btrfs_inode_extref_parent(eb, extref);
904 start = btrfs_ref_to_path(root, tmp_path, name_len,
908 ret = PTR_ERR(start);
911 if (start < p->buf) {
912 /* overflow , try again with larger buffer */
913 ret = fs_path_ensure_buf(p,
914 p->buf_len + p->buf - start);
917 start = btrfs_ref_to_path(root, tmp_path,
922 ret = PTR_ERR(start);
925 BUG_ON(start < p->buf);
929 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
935 cur += elem_size + name_len;
936 ret = iterate(num, dir, index, p, ctx);
943 btrfs_free_path(tmp_path);
948 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
949 const char *name, int name_len,
950 const char *data, int data_len,
954 * Helper function to iterate the entries in ONE btrfs_dir_item.
955 * The iterate callback may return a non zero value to stop iteration. This can
956 * be a negative value for error codes or 1 to simply stop it.
958 * path must point to the dir item when called.
960 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
961 struct btrfs_key *found_key,
962 iterate_dir_item_t iterate, void *ctx)
965 struct extent_buffer *eb;
966 struct btrfs_item *item;
967 struct btrfs_dir_item *di;
968 struct btrfs_key di_key;
970 const int buf_len = PATH_MAX;
980 buf = kmalloc(buf_len, GFP_NOFS);
987 slot = path->slots[0];
988 item = btrfs_item_nr(slot);
989 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
992 total = btrfs_item_size(eb, item);
995 while (cur < total) {
996 name_len = btrfs_dir_name_len(eb, di);
997 data_len = btrfs_dir_data_len(eb, di);
998 type = btrfs_dir_type(eb, di);
999 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1004 if (name_len + data_len > buf_len) {
1005 ret = -ENAMETOOLONG;
1009 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1010 name_len + data_len);
1012 len = sizeof(*di) + name_len + data_len;
1013 di = (struct btrfs_dir_item *)((char *)di + len);
1016 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1017 data_len, type, ctx);
1033 static int __copy_first_ref(int num, u64 dir, int index,
1034 struct fs_path *p, void *ctx)
1037 struct fs_path *pt = ctx;
1039 ret = fs_path_copy(pt, p);
1043 /* we want the first only */
1048 * Retrieve the first path of an inode. If an inode has more then one
1049 * ref/hardlink, this is ignored.
1051 static int get_inode_path(struct btrfs_root *root,
1052 u64 ino, struct fs_path *path)
1055 struct btrfs_key key, found_key;
1056 struct btrfs_path *p;
1058 p = alloc_path_for_send();
1062 fs_path_reset(path);
1065 key.type = BTRFS_INODE_REF_KEY;
1068 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1075 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1076 if (found_key.objectid != ino ||
1077 (found_key.type != BTRFS_INODE_REF_KEY &&
1078 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1083 ret = iterate_inode_ref(root, p, &found_key, 1,
1084 __copy_first_ref, path);
1094 struct backref_ctx {
1095 struct send_ctx *sctx;
1097 struct btrfs_path *path;
1098 /* number of total found references */
1102 * used for clones found in send_root. clones found behind cur_objectid
1103 * and cur_offset are not considered as allowed clones.
1108 /* may be truncated in case it's the last extent in a file */
1111 /* Just to check for bugs in backref resolving */
1115 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1117 u64 root = (u64)(uintptr_t)key;
1118 struct clone_root *cr = (struct clone_root *)elt;
1120 if (root < cr->root->objectid)
1122 if (root > cr->root->objectid)
1127 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1129 struct clone_root *cr1 = (struct clone_root *)e1;
1130 struct clone_root *cr2 = (struct clone_root *)e2;
1132 if (cr1->root->objectid < cr2->root->objectid)
1134 if (cr1->root->objectid > cr2->root->objectid)
1140 * Called for every backref that is found for the current extent.
1141 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1143 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1145 struct backref_ctx *bctx = ctx_;
1146 struct clone_root *found;
1150 /* First check if the root is in the list of accepted clone sources */
1151 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1152 bctx->sctx->clone_roots_cnt,
1153 sizeof(struct clone_root),
1154 __clone_root_cmp_bsearch);
1158 if (found->root == bctx->sctx->send_root &&
1159 ino == bctx->cur_objectid &&
1160 offset == bctx->cur_offset) {
1161 bctx->found_itself = 1;
1165 * There are inodes that have extents that lie behind its i_size. Don't
1166 * accept clones from these extents.
1168 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1170 btrfs_release_path(bctx->path);
1174 if (offset + bctx->extent_len > i_size)
1178 * Make sure we don't consider clones from send_root that are
1179 * behind the current inode/offset.
1181 if (found->root == bctx->sctx->send_root) {
1183 * TODO for the moment we don't accept clones from the inode
1184 * that is currently send. We may change this when
1185 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1188 if (ino >= bctx->cur_objectid)
1191 if (ino > bctx->cur_objectid)
1193 if (offset + bctx->extent_len > bctx->cur_offset)
1199 found->found_refs++;
1200 if (ino < found->ino) {
1202 found->offset = offset;
1203 } else if (found->ino == ino) {
1205 * same extent found more then once in the same file.
1207 if (found->offset > offset + bctx->extent_len)
1208 found->offset = offset;
1215 * Given an inode, offset and extent item, it finds a good clone for a clone
1216 * instruction. Returns -ENOENT when none could be found. The function makes
1217 * sure that the returned clone is usable at the point where sending is at the
1218 * moment. This means, that no clones are accepted which lie behind the current
1221 * path must point to the extent item when called.
1223 static int find_extent_clone(struct send_ctx *sctx,
1224 struct btrfs_path *path,
1225 u64 ino, u64 data_offset,
1227 struct clone_root **found)
1234 u64 extent_item_pos;
1236 struct btrfs_file_extent_item *fi;
1237 struct extent_buffer *eb = path->nodes[0];
1238 struct backref_ctx *backref_ctx = NULL;
1239 struct clone_root *cur_clone_root;
1240 struct btrfs_key found_key;
1241 struct btrfs_path *tmp_path;
1245 tmp_path = alloc_path_for_send();
1249 /* We only use this path under the commit sem */
1250 tmp_path->need_commit_sem = 0;
1252 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1258 backref_ctx->path = tmp_path;
1260 if (data_offset >= ino_size) {
1262 * There may be extents that lie behind the file's size.
1263 * I at least had this in combination with snapshotting while
1264 * writing large files.
1270 fi = btrfs_item_ptr(eb, path->slots[0],
1271 struct btrfs_file_extent_item);
1272 extent_type = btrfs_file_extent_type(eb, fi);
1273 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1277 compressed = btrfs_file_extent_compression(eb, fi);
1279 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1280 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1281 if (disk_byte == 0) {
1285 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1287 down_read(&sctx->send_root->fs_info->commit_root_sem);
1288 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1289 &found_key, &flags);
1290 up_read(&sctx->send_root->fs_info->commit_root_sem);
1291 btrfs_release_path(tmp_path);
1295 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1301 * Setup the clone roots.
1303 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1304 cur_clone_root = sctx->clone_roots + i;
1305 cur_clone_root->ino = (u64)-1;
1306 cur_clone_root->offset = 0;
1307 cur_clone_root->found_refs = 0;
1310 backref_ctx->sctx = sctx;
1311 backref_ctx->found = 0;
1312 backref_ctx->cur_objectid = ino;
1313 backref_ctx->cur_offset = data_offset;
1314 backref_ctx->found_itself = 0;
1315 backref_ctx->extent_len = num_bytes;
1318 * The last extent of a file may be too large due to page alignment.
1319 * We need to adjust extent_len in this case so that the checks in
1320 * __iterate_backrefs work.
1322 if (data_offset + num_bytes >= ino_size)
1323 backref_ctx->extent_len = ino_size - data_offset;
1326 * Now collect all backrefs.
1328 if (compressed == BTRFS_COMPRESS_NONE)
1329 extent_item_pos = logical - found_key.objectid;
1331 extent_item_pos = 0;
1332 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1333 found_key.objectid, extent_item_pos, 1,
1334 __iterate_backrefs, backref_ctx);
1339 if (!backref_ctx->found_itself) {
1340 /* found a bug in backref code? */
1342 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1343 "send_root. inode=%llu, offset=%llu, "
1344 "disk_byte=%llu found extent=%llu\n",
1345 ino, data_offset, disk_byte, found_key.objectid);
1349 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1351 "num_bytes=%llu, logical=%llu\n",
1352 data_offset, ino, num_bytes, logical);
1354 if (!backref_ctx->found)
1355 verbose_printk("btrfs: no clones found\n");
1357 cur_clone_root = NULL;
1358 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1359 if (sctx->clone_roots[i].found_refs) {
1360 if (!cur_clone_root)
1361 cur_clone_root = sctx->clone_roots + i;
1362 else if (sctx->clone_roots[i].root == sctx->send_root)
1363 /* prefer clones from send_root over others */
1364 cur_clone_root = sctx->clone_roots + i;
1369 if (cur_clone_root) {
1370 if (compressed != BTRFS_COMPRESS_NONE) {
1372 * Offsets given by iterate_extent_inodes() are relative
1373 * to the start of the extent, we need to add logical
1374 * offset from the file extent item.
1375 * (See why at backref.c:check_extent_in_eb())
1377 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1380 *found = cur_clone_root;
1387 btrfs_free_path(tmp_path);
1392 static int read_symlink(struct btrfs_root *root,
1394 struct fs_path *dest)
1397 struct btrfs_path *path;
1398 struct btrfs_key key;
1399 struct btrfs_file_extent_item *ei;
1405 path = alloc_path_for_send();
1410 key.type = BTRFS_EXTENT_DATA_KEY;
1412 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1417 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1418 struct btrfs_file_extent_item);
1419 type = btrfs_file_extent_type(path->nodes[0], ei);
1420 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1421 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1422 BUG_ON(compression);
1424 off = btrfs_file_extent_inline_start(ei);
1425 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1427 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1430 btrfs_free_path(path);
1435 * Helper function to generate a file name that is unique in the root of
1436 * send_root and parent_root. This is used to generate names for orphan inodes.
1438 static int gen_unique_name(struct send_ctx *sctx,
1440 struct fs_path *dest)
1443 struct btrfs_path *path;
1444 struct btrfs_dir_item *di;
1449 path = alloc_path_for_send();
1454 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1456 ASSERT(len < sizeof(tmp));
1458 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1459 path, BTRFS_FIRST_FREE_OBJECTID,
1460 tmp, strlen(tmp), 0);
1461 btrfs_release_path(path);
1467 /* not unique, try again */
1472 if (!sctx->parent_root) {
1478 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1479 path, BTRFS_FIRST_FREE_OBJECTID,
1480 tmp, strlen(tmp), 0);
1481 btrfs_release_path(path);
1487 /* not unique, try again */
1495 ret = fs_path_add(dest, tmp, strlen(tmp));
1498 btrfs_free_path(path);
1503 inode_state_no_change,
1504 inode_state_will_create,
1505 inode_state_did_create,
1506 inode_state_will_delete,
1507 inode_state_did_delete,
1510 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1518 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1520 if (ret < 0 && ret != -ENOENT)
1524 if (!sctx->parent_root) {
1525 right_ret = -ENOENT;
1527 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1528 NULL, NULL, NULL, NULL);
1529 if (ret < 0 && ret != -ENOENT)
1534 if (!left_ret && !right_ret) {
1535 if (left_gen == gen && right_gen == gen) {
1536 ret = inode_state_no_change;
1537 } else if (left_gen == gen) {
1538 if (ino < sctx->send_progress)
1539 ret = inode_state_did_create;
1541 ret = inode_state_will_create;
1542 } else if (right_gen == gen) {
1543 if (ino < sctx->send_progress)
1544 ret = inode_state_did_delete;
1546 ret = inode_state_will_delete;
1550 } else if (!left_ret) {
1551 if (left_gen == gen) {
1552 if (ino < sctx->send_progress)
1553 ret = inode_state_did_create;
1555 ret = inode_state_will_create;
1559 } else if (!right_ret) {
1560 if (right_gen == gen) {
1561 if (ino < sctx->send_progress)
1562 ret = inode_state_did_delete;
1564 ret = inode_state_will_delete;
1576 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1580 ret = get_cur_inode_state(sctx, ino, gen);
1584 if (ret == inode_state_no_change ||
1585 ret == inode_state_did_create ||
1586 ret == inode_state_will_delete)
1596 * Helper function to lookup a dir item in a dir.
1598 static int lookup_dir_item_inode(struct btrfs_root *root,
1599 u64 dir, const char *name, int name_len,
1604 struct btrfs_dir_item *di;
1605 struct btrfs_key key;
1606 struct btrfs_path *path;
1608 path = alloc_path_for_send();
1612 di = btrfs_lookup_dir_item(NULL, root, path,
1613 dir, name, name_len, 0);
1622 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1623 *found_inode = key.objectid;
1624 *found_type = btrfs_dir_type(path->nodes[0], di);
1627 btrfs_free_path(path);
1632 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1633 * generation of the parent dir and the name of the dir entry.
1635 static int get_first_ref(struct btrfs_root *root, u64 ino,
1636 u64 *dir, u64 *dir_gen, struct fs_path *name)
1639 struct btrfs_key key;
1640 struct btrfs_key found_key;
1641 struct btrfs_path *path;
1645 path = alloc_path_for_send();
1650 key.type = BTRFS_INODE_REF_KEY;
1653 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1657 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1659 if (ret || found_key.objectid != ino ||
1660 (found_key.type != BTRFS_INODE_REF_KEY &&
1661 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1666 if (key.type == BTRFS_INODE_REF_KEY) {
1667 struct btrfs_inode_ref *iref;
1668 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1669 struct btrfs_inode_ref);
1670 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1671 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1672 (unsigned long)(iref + 1),
1674 parent_dir = found_key.offset;
1676 struct btrfs_inode_extref *extref;
1677 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1678 struct btrfs_inode_extref);
1679 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1680 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1681 (unsigned long)&extref->name, len);
1682 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1686 btrfs_release_path(path);
1688 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1696 btrfs_free_path(path);
1700 static int is_first_ref(struct btrfs_root *root,
1702 const char *name, int name_len)
1705 struct fs_path *tmp_name;
1709 tmp_name = fs_path_alloc();
1713 ret = get_first_ref(root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1717 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1722 ret = !memcmp(tmp_name->start, name, name_len);
1725 fs_path_free(tmp_name);
1730 * Used by process_recorded_refs to determine if a new ref would overwrite an
1731 * already existing ref. In case it detects an overwrite, it returns the
1732 * inode/gen in who_ino/who_gen.
1733 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1734 * to make sure later references to the overwritten inode are possible.
1735 * Orphanizing is however only required for the first ref of an inode.
1736 * process_recorded_refs does an additional is_first_ref check to see if
1737 * orphanizing is really required.
1739 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1740 const char *name, int name_len,
1741 u64 *who_ino, u64 *who_gen)
1745 u64 other_inode = 0;
1748 if (!sctx->parent_root)
1751 ret = is_inode_existent(sctx, dir, dir_gen);
1756 * If we have a parent root we need to verify that the parent dir was
1757 * not delted and then re-created, if it was then we have no overwrite
1758 * and we can just unlink this entry.
1760 if (sctx->parent_root) {
1761 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1763 if (ret < 0 && ret != -ENOENT)
1773 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1774 &other_inode, &other_type);
1775 if (ret < 0 && ret != -ENOENT)
1783 * Check if the overwritten ref was already processed. If yes, the ref
1784 * was already unlinked/moved, so we can safely assume that we will not
1785 * overwrite anything at this point in time.
1787 if (other_inode > sctx->send_progress) {
1788 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1789 who_gen, NULL, NULL, NULL, NULL);
1794 *who_ino = other_inode;
1804 * Checks if the ref was overwritten by an already processed inode. This is
1805 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1806 * thus the orphan name needs be used.
1807 * process_recorded_refs also uses it to avoid unlinking of refs that were
1810 static int did_overwrite_ref(struct send_ctx *sctx,
1811 u64 dir, u64 dir_gen,
1812 u64 ino, u64 ino_gen,
1813 const char *name, int name_len)
1820 if (!sctx->parent_root)
1823 ret = is_inode_existent(sctx, dir, dir_gen);
1827 /* check if the ref was overwritten by another ref */
1828 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1829 &ow_inode, &other_type);
1830 if (ret < 0 && ret != -ENOENT)
1833 /* was never and will never be overwritten */
1838 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1843 if (ow_inode == ino && gen == ino_gen) {
1848 /* we know that it is or will be overwritten. check this now */
1849 if (ow_inode < sctx->send_progress)
1859 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1860 * that got overwritten. This is used by process_recorded_refs to determine
1861 * if it has to use the path as returned by get_cur_path or the orphan name.
1863 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1866 struct fs_path *name = NULL;
1870 if (!sctx->parent_root)
1873 name = fs_path_alloc();
1877 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1881 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1882 name->start, fs_path_len(name));
1890 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1891 * so we need to do some special handling in case we have clashes. This function
1892 * takes care of this with the help of name_cache_entry::radix_list.
1893 * In case of error, nce is kfreed.
1895 static int name_cache_insert(struct send_ctx *sctx,
1896 struct name_cache_entry *nce)
1899 struct list_head *nce_head;
1901 nce_head = radix_tree_lookup(&sctx->name_cache,
1902 (unsigned long)nce->ino);
1904 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1909 INIT_LIST_HEAD(nce_head);
1911 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1918 list_add_tail(&nce->radix_list, nce_head);
1919 list_add_tail(&nce->list, &sctx->name_cache_list);
1920 sctx->name_cache_size++;
1925 static void name_cache_delete(struct send_ctx *sctx,
1926 struct name_cache_entry *nce)
1928 struct list_head *nce_head;
1930 nce_head = radix_tree_lookup(&sctx->name_cache,
1931 (unsigned long)nce->ino);
1933 btrfs_err(sctx->send_root->fs_info,
1934 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1935 nce->ino, sctx->name_cache_size);
1938 list_del(&nce->radix_list);
1939 list_del(&nce->list);
1940 sctx->name_cache_size--;
1943 * We may not get to the final release of nce_head if the lookup fails
1945 if (nce_head && list_empty(nce_head)) {
1946 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1951 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1954 struct list_head *nce_head;
1955 struct name_cache_entry *cur;
1957 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1961 list_for_each_entry(cur, nce_head, radix_list) {
1962 if (cur->ino == ino && cur->gen == gen)
1969 * Removes the entry from the list and adds it back to the end. This marks the
1970 * entry as recently used so that name_cache_clean_unused does not remove it.
1972 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1974 list_del(&nce->list);
1975 list_add_tail(&nce->list, &sctx->name_cache_list);
1979 * Remove some entries from the beginning of name_cache_list.
1981 static void name_cache_clean_unused(struct send_ctx *sctx)
1983 struct name_cache_entry *nce;
1985 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1988 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1989 nce = list_entry(sctx->name_cache_list.next,
1990 struct name_cache_entry, list);
1991 name_cache_delete(sctx, nce);
1996 static void name_cache_free(struct send_ctx *sctx)
1998 struct name_cache_entry *nce;
2000 while (!list_empty(&sctx->name_cache_list)) {
2001 nce = list_entry(sctx->name_cache_list.next,
2002 struct name_cache_entry, list);
2003 name_cache_delete(sctx, nce);
2009 * Used by get_cur_path for each ref up to the root.
2010 * Returns 0 if it succeeded.
2011 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2012 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2013 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2014 * Returns <0 in case of error.
2016 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2020 struct fs_path *dest)
2024 struct btrfs_path *path = NULL;
2025 struct name_cache_entry *nce = NULL;
2028 * First check if we already did a call to this function with the same
2029 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2030 * return the cached result.
2032 nce = name_cache_search(sctx, ino, gen);
2034 if (ino < sctx->send_progress && nce->need_later_update) {
2035 name_cache_delete(sctx, nce);
2039 name_cache_used(sctx, nce);
2040 *parent_ino = nce->parent_ino;
2041 *parent_gen = nce->parent_gen;
2042 ret = fs_path_add(dest, nce->name, nce->name_len);
2050 path = alloc_path_for_send();
2055 * If the inode is not existent yet, add the orphan name and return 1.
2056 * This should only happen for the parent dir that we determine in
2059 ret = is_inode_existent(sctx, ino, gen);
2064 ret = gen_unique_name(sctx, ino, gen, dest);
2072 * Depending on whether the inode was already processed or not, use
2073 * send_root or parent_root for ref lookup.
2075 if (ino < sctx->send_progress)
2076 ret = get_first_ref(sctx->send_root, ino,
2077 parent_ino, parent_gen, dest);
2079 ret = get_first_ref(sctx->parent_root, ino,
2080 parent_ino, parent_gen, dest);
2085 * Check if the ref was overwritten by an inode's ref that was processed
2086 * earlier. If yes, treat as orphan and return 1.
2088 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2089 dest->start, dest->end - dest->start);
2093 fs_path_reset(dest);
2094 ret = gen_unique_name(sctx, ino, gen, dest);
2102 * Store the result of the lookup in the name cache.
2104 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2112 nce->parent_ino = *parent_ino;
2113 nce->parent_gen = *parent_gen;
2114 nce->name_len = fs_path_len(dest);
2116 strcpy(nce->name, dest->start);
2118 if (ino < sctx->send_progress)
2119 nce->need_later_update = 0;
2121 nce->need_later_update = 1;
2123 nce_ret = name_cache_insert(sctx, nce);
2126 name_cache_clean_unused(sctx);
2129 btrfs_free_path(path);
2134 * Magic happens here. This function returns the first ref to an inode as it
2135 * would look like while receiving the stream at this point in time.
2136 * We walk the path up to the root. For every inode in between, we check if it
2137 * was already processed/sent. If yes, we continue with the parent as found
2138 * in send_root. If not, we continue with the parent as found in parent_root.
2139 * If we encounter an inode that was deleted at this point in time, we use the
2140 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2141 * that were not created yet and overwritten inodes/refs.
2143 * When do we have have orphan inodes:
2144 * 1. When an inode is freshly created and thus no valid refs are available yet
2145 * 2. When a directory lost all it's refs (deleted) but still has dir items
2146 * inside which were not processed yet (pending for move/delete). If anyone
2147 * tried to get the path to the dir items, it would get a path inside that
2149 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2150 * of an unprocessed inode. If in that case the first ref would be
2151 * overwritten, the overwritten inode gets "orphanized". Later when we
2152 * process this overwritten inode, it is restored at a new place by moving
2155 * sctx->send_progress tells this function at which point in time receiving
2158 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2159 struct fs_path *dest)
2162 struct fs_path *name = NULL;
2163 u64 parent_inode = 0;
2167 name = fs_path_alloc();
2174 fs_path_reset(dest);
2176 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2177 fs_path_reset(name);
2179 if (is_waiting_for_rm(sctx, ino)) {
2180 ret = gen_unique_name(sctx, ino, gen, name);
2183 ret = fs_path_add_path(dest, name);
2187 if (is_waiting_for_move(sctx, ino)) {
2188 ret = get_first_ref(sctx->parent_root, ino,
2189 &parent_inode, &parent_gen, name);
2191 ret = __get_cur_name_and_parent(sctx, ino, gen,
2201 ret = fs_path_add_path(dest, name);
2212 fs_path_unreverse(dest);
2217 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2219 static int send_subvol_begin(struct send_ctx *sctx)
2222 struct btrfs_root *send_root = sctx->send_root;
2223 struct btrfs_root *parent_root = sctx->parent_root;
2224 struct btrfs_path *path;
2225 struct btrfs_key key;
2226 struct btrfs_root_ref *ref;
2227 struct extent_buffer *leaf;
2231 path = btrfs_alloc_path();
2235 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2237 btrfs_free_path(path);
2241 key.objectid = send_root->objectid;
2242 key.type = BTRFS_ROOT_BACKREF_KEY;
2245 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2254 leaf = path->nodes[0];
2255 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2256 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2257 key.objectid != send_root->objectid) {
2261 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2262 namelen = btrfs_root_ref_name_len(leaf, ref);
2263 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2264 btrfs_release_path(path);
2267 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2271 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2276 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2277 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2278 sctx->send_root->root_item.uuid);
2279 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2280 le64_to_cpu(sctx->send_root->root_item.ctransid));
2282 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2283 sctx->parent_root->root_item.uuid);
2284 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2285 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2288 ret = send_cmd(sctx);
2292 btrfs_free_path(path);
2297 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2302 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2304 p = fs_path_alloc();
2308 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2312 ret = get_cur_path(sctx, ino, gen, p);
2315 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2316 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2318 ret = send_cmd(sctx);
2326 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2331 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2333 p = fs_path_alloc();
2337 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2341 ret = get_cur_path(sctx, ino, gen, p);
2344 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2345 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2347 ret = send_cmd(sctx);
2355 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2360 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2362 p = fs_path_alloc();
2366 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2370 ret = get_cur_path(sctx, ino, gen, p);
2373 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2374 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2375 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2377 ret = send_cmd(sctx);
2385 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2388 struct fs_path *p = NULL;
2389 struct btrfs_inode_item *ii;
2390 struct btrfs_path *path = NULL;
2391 struct extent_buffer *eb;
2392 struct btrfs_key key;
2395 verbose_printk("btrfs: send_utimes %llu\n", ino);
2397 p = fs_path_alloc();
2401 path = alloc_path_for_send();
2408 key.type = BTRFS_INODE_ITEM_KEY;
2410 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2414 eb = path->nodes[0];
2415 slot = path->slots[0];
2416 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2418 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2422 ret = get_cur_path(sctx, ino, gen, p);
2425 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2426 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2427 btrfs_inode_atime(ii));
2428 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2429 btrfs_inode_mtime(ii));
2430 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2431 btrfs_inode_ctime(ii));
2432 /* TODO Add otime support when the otime patches get into upstream */
2434 ret = send_cmd(sctx);
2439 btrfs_free_path(path);
2444 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2445 * a valid path yet because we did not process the refs yet. So, the inode
2446 * is created as orphan.
2448 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2457 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2459 p = fs_path_alloc();
2463 if (ino != sctx->cur_ino) {
2464 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2469 gen = sctx->cur_inode_gen;
2470 mode = sctx->cur_inode_mode;
2471 rdev = sctx->cur_inode_rdev;
2474 if (S_ISREG(mode)) {
2475 cmd = BTRFS_SEND_C_MKFILE;
2476 } else if (S_ISDIR(mode)) {
2477 cmd = BTRFS_SEND_C_MKDIR;
2478 } else if (S_ISLNK(mode)) {
2479 cmd = BTRFS_SEND_C_SYMLINK;
2480 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2481 cmd = BTRFS_SEND_C_MKNOD;
2482 } else if (S_ISFIFO(mode)) {
2483 cmd = BTRFS_SEND_C_MKFIFO;
2484 } else if (S_ISSOCK(mode)) {
2485 cmd = BTRFS_SEND_C_MKSOCK;
2487 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2488 (int)(mode & S_IFMT));
2493 ret = begin_cmd(sctx, cmd);
2497 ret = gen_unique_name(sctx, ino, gen, p);
2501 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2502 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2504 if (S_ISLNK(mode)) {
2506 ret = read_symlink(sctx->send_root, ino, p);
2509 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2510 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2511 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2512 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2513 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2516 ret = send_cmd(sctx);
2528 * We need some special handling for inodes that get processed before the parent
2529 * directory got created. See process_recorded_refs for details.
2530 * This function does the check if we already created the dir out of order.
2532 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2535 struct btrfs_path *path = NULL;
2536 struct btrfs_key key;
2537 struct btrfs_key found_key;
2538 struct btrfs_key di_key;
2539 struct extent_buffer *eb;
2540 struct btrfs_dir_item *di;
2543 path = alloc_path_for_send();
2550 key.type = BTRFS_DIR_INDEX_KEY;
2552 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2557 eb = path->nodes[0];
2558 slot = path->slots[0];
2559 if (slot >= btrfs_header_nritems(eb)) {
2560 ret = btrfs_next_leaf(sctx->send_root, path);
2563 } else if (ret > 0) {
2570 btrfs_item_key_to_cpu(eb, &found_key, slot);
2571 if (found_key.objectid != key.objectid ||
2572 found_key.type != key.type) {
2577 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2578 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2580 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2581 di_key.objectid < sctx->send_progress) {
2590 btrfs_free_path(path);
2595 * Only creates the inode if it is:
2596 * 1. Not a directory
2597 * 2. Or a directory which was not created already due to out of order
2598 * directories. See did_create_dir and process_recorded_refs for details.
2600 static int send_create_inode_if_needed(struct send_ctx *sctx)
2604 if (S_ISDIR(sctx->cur_inode_mode)) {
2605 ret = did_create_dir(sctx, sctx->cur_ino);
2614 ret = send_create_inode(sctx, sctx->cur_ino);
2622 struct recorded_ref {
2623 struct list_head list;
2626 struct fs_path *full_path;
2634 * We need to process new refs before deleted refs, but compare_tree gives us
2635 * everything mixed. So we first record all refs and later process them.
2636 * This function is a helper to record one ref.
2638 static int __record_ref(struct list_head *head, u64 dir,
2639 u64 dir_gen, struct fs_path *path)
2641 struct recorded_ref *ref;
2643 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2648 ref->dir_gen = dir_gen;
2649 ref->full_path = path;
2651 ref->name = (char *)kbasename(ref->full_path->start);
2652 ref->name_len = ref->full_path->end - ref->name;
2653 ref->dir_path = ref->full_path->start;
2654 if (ref->name == ref->full_path->start)
2655 ref->dir_path_len = 0;
2657 ref->dir_path_len = ref->full_path->end -
2658 ref->full_path->start - 1 - ref->name_len;
2660 list_add_tail(&ref->list, head);
2664 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2666 struct recorded_ref *new;
2668 new = kmalloc(sizeof(*ref), GFP_NOFS);
2672 new->dir = ref->dir;
2673 new->dir_gen = ref->dir_gen;
2674 new->full_path = NULL;
2675 INIT_LIST_HEAD(&new->list);
2676 list_add_tail(&new->list, list);
2680 static void __free_recorded_refs(struct list_head *head)
2682 struct recorded_ref *cur;
2684 while (!list_empty(head)) {
2685 cur = list_entry(head->next, struct recorded_ref, list);
2686 fs_path_free(cur->full_path);
2687 list_del(&cur->list);
2692 static void free_recorded_refs(struct send_ctx *sctx)
2694 __free_recorded_refs(&sctx->new_refs);
2695 __free_recorded_refs(&sctx->deleted_refs);
2699 * Renames/moves a file/dir to its orphan name. Used when the first
2700 * ref of an unprocessed inode gets overwritten and for all non empty
2703 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2704 struct fs_path *path)
2707 struct fs_path *orphan;
2709 orphan = fs_path_alloc();
2713 ret = gen_unique_name(sctx, ino, gen, orphan);
2717 ret = send_rename(sctx, path, orphan);
2720 fs_path_free(orphan);
2724 static struct orphan_dir_info *
2725 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2727 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2728 struct rb_node *parent = NULL;
2729 struct orphan_dir_info *entry, *odi;
2731 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2733 return ERR_PTR(-ENOMEM);
2739 entry = rb_entry(parent, struct orphan_dir_info, node);
2740 if (dir_ino < entry->ino) {
2742 } else if (dir_ino > entry->ino) {
2743 p = &(*p)->rb_right;
2750 rb_link_node(&odi->node, parent, p);
2751 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2755 static struct orphan_dir_info *
2756 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2758 struct rb_node *n = sctx->orphan_dirs.rb_node;
2759 struct orphan_dir_info *entry;
2762 entry = rb_entry(n, struct orphan_dir_info, node);
2763 if (dir_ino < entry->ino)
2765 else if (dir_ino > entry->ino)
2773 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2775 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2780 static void free_orphan_dir_info(struct send_ctx *sctx,
2781 struct orphan_dir_info *odi)
2785 rb_erase(&odi->node, &sctx->orphan_dirs);
2790 * Returns 1 if a directory can be removed at this point in time.
2791 * We check this by iterating all dir items and checking if the inode behind
2792 * the dir item was already processed.
2794 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2798 struct btrfs_root *root = sctx->parent_root;
2799 struct btrfs_path *path;
2800 struct btrfs_key key;
2801 struct btrfs_key found_key;
2802 struct btrfs_key loc;
2803 struct btrfs_dir_item *di;
2806 * Don't try to rmdir the top/root subvolume dir.
2808 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2811 path = alloc_path_for_send();
2816 key.type = BTRFS_DIR_INDEX_KEY;
2818 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2823 struct waiting_dir_move *dm;
2825 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2826 ret = btrfs_next_leaf(root, path);
2833 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2835 if (found_key.objectid != key.objectid ||
2836 found_key.type != key.type)
2839 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2840 struct btrfs_dir_item);
2841 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2843 dm = get_waiting_dir_move(sctx, loc.objectid);
2845 struct orphan_dir_info *odi;
2847 odi = add_orphan_dir_info(sctx, dir);
2853 dm->rmdir_ino = dir;
2858 if (loc.objectid > send_progress) {
2869 btrfs_free_path(path);
2873 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2875 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2877 return entry != NULL;
2880 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2882 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2883 struct rb_node *parent = NULL;
2884 struct waiting_dir_move *entry, *dm;
2886 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2894 entry = rb_entry(parent, struct waiting_dir_move, node);
2895 if (ino < entry->ino) {
2897 } else if (ino > entry->ino) {
2898 p = &(*p)->rb_right;
2905 rb_link_node(&dm->node, parent, p);
2906 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2910 static struct waiting_dir_move *
2911 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2913 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2914 struct waiting_dir_move *entry;
2917 entry = rb_entry(n, struct waiting_dir_move, node);
2918 if (ino < entry->ino)
2920 else if (ino > entry->ino)
2928 static void free_waiting_dir_move(struct send_ctx *sctx,
2929 struct waiting_dir_move *dm)
2933 rb_erase(&dm->node, &sctx->waiting_dir_moves);
2937 static int add_pending_dir_move(struct send_ctx *sctx,
2942 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
2943 struct rb_node *parent = NULL;
2944 struct pending_dir_move *entry = NULL, *pm;
2945 struct recorded_ref *cur;
2949 pm = kmalloc(sizeof(*pm), GFP_NOFS);
2952 pm->parent_ino = parent_ino;
2955 INIT_LIST_HEAD(&pm->list);
2956 INIT_LIST_HEAD(&pm->update_refs);
2957 RB_CLEAR_NODE(&pm->node);
2961 entry = rb_entry(parent, struct pending_dir_move, node);
2962 if (parent_ino < entry->parent_ino) {
2964 } else if (parent_ino > entry->parent_ino) {
2965 p = &(*p)->rb_right;
2972 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2973 ret = dup_ref(cur, &pm->update_refs);
2977 list_for_each_entry(cur, &sctx->new_refs, list) {
2978 ret = dup_ref(cur, &pm->update_refs);
2983 ret = add_waiting_dir_move(sctx, pm->ino);
2988 list_add_tail(&pm->list, &entry->list);
2990 rb_link_node(&pm->node, parent, p);
2991 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
2996 __free_recorded_refs(&pm->update_refs);
3002 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3005 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3006 struct pending_dir_move *entry;
3009 entry = rb_entry(n, struct pending_dir_move, node);
3010 if (parent_ino < entry->parent_ino)
3012 else if (parent_ino > entry->parent_ino)
3020 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3022 struct fs_path *from_path = NULL;
3023 struct fs_path *to_path = NULL;
3024 struct fs_path *name = NULL;
3025 u64 orig_progress = sctx->send_progress;
3026 struct recorded_ref *cur;
3027 u64 parent_ino, parent_gen;
3028 struct waiting_dir_move *dm = NULL;
3032 name = fs_path_alloc();
3033 from_path = fs_path_alloc();
3034 if (!name || !from_path) {
3039 dm = get_waiting_dir_move(sctx, pm->ino);
3041 rmdir_ino = dm->rmdir_ino;
3042 free_waiting_dir_move(sctx, dm);
3044 ret = get_first_ref(sctx->parent_root, pm->ino,
3045 &parent_ino, &parent_gen, name);
3049 if (parent_ino == sctx->cur_ino) {
3050 /* child only renamed, not moved */
3051 ASSERT(parent_gen == sctx->cur_inode_gen);
3052 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3056 ret = fs_path_add_path(from_path, name);
3060 /* child moved and maybe renamed too */
3061 sctx->send_progress = pm->ino;
3062 ret = get_cur_path(sctx, pm->ino, pm->gen, from_path);
3070 to_path = fs_path_alloc();
3076 sctx->send_progress = sctx->cur_ino + 1;
3077 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3081 ret = send_rename(sctx, from_path, to_path);
3086 struct orphan_dir_info *odi;
3088 odi = get_orphan_dir_info(sctx, rmdir_ino);
3090 /* already deleted */
3093 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3099 name = fs_path_alloc();
3104 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3107 ret = send_rmdir(sctx, name);
3110 free_orphan_dir_info(sctx, odi);
3114 ret = send_utimes(sctx, pm->ino, pm->gen);
3119 * After rename/move, need to update the utimes of both new parent(s)
3120 * and old parent(s).
3122 list_for_each_entry(cur, &pm->update_refs, list) {
3123 if (cur->dir == rmdir_ino)
3125 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3132 fs_path_free(from_path);
3133 fs_path_free(to_path);
3134 sctx->send_progress = orig_progress;
3139 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3141 if (!list_empty(&m->list))
3143 if (!RB_EMPTY_NODE(&m->node))
3144 rb_erase(&m->node, &sctx->pending_dir_moves);
3145 __free_recorded_refs(&m->update_refs);
3149 static void tail_append_pending_moves(struct pending_dir_move *moves,
3150 struct list_head *stack)
3152 if (list_empty(&moves->list)) {
3153 list_add_tail(&moves->list, stack);
3156 list_splice_init(&moves->list, &list);
3157 list_add_tail(&moves->list, stack);
3158 list_splice_tail(&list, stack);
3162 static int apply_children_dir_moves(struct send_ctx *sctx)
3164 struct pending_dir_move *pm;
3165 struct list_head stack;
3166 u64 parent_ino = sctx->cur_ino;
3169 pm = get_pending_dir_moves(sctx, parent_ino);
3173 INIT_LIST_HEAD(&stack);
3174 tail_append_pending_moves(pm, &stack);
3176 while (!list_empty(&stack)) {
3177 pm = list_first_entry(&stack, struct pending_dir_move, list);
3178 parent_ino = pm->ino;
3179 ret = apply_dir_move(sctx, pm);
3180 free_pending_move(sctx, pm);
3183 pm = get_pending_dir_moves(sctx, parent_ino);
3185 tail_append_pending_moves(pm, &stack);
3190 while (!list_empty(&stack)) {
3191 pm = list_first_entry(&stack, struct pending_dir_move, list);
3192 free_pending_move(sctx, pm);
3197 static int wait_for_parent_move(struct send_ctx *sctx,
3198 struct recorded_ref *parent_ref)
3201 u64 ino = parent_ref->dir;
3202 u64 parent_ino_before, parent_ino_after;
3204 struct fs_path *path_before = NULL;
3205 struct fs_path *path_after = NULL;
3207 int register_upper_dirs;
3210 if (is_waiting_for_move(sctx, ino))
3213 if (parent_ref->dir <= sctx->cur_ino)
3216 ret = get_inode_info(sctx->parent_root, ino, NULL, &old_gen,
3217 NULL, NULL, NULL, NULL);
3223 if (parent_ref->dir_gen != old_gen)
3226 path_before = fs_path_alloc();
3230 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3232 if (ret == -ENOENT) {
3235 } else if (ret < 0) {
3239 path_after = fs_path_alloc();
3245 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3247 if (ret == -ENOENT) {
3250 } else if (ret < 0) {
3254 len1 = fs_path_len(path_before);
3255 len2 = fs_path_len(path_after);
3256 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3257 memcmp(path_before->start, path_after->start, len1)) {
3264 * Ok, our new most direct ancestor has a higher inode number but
3265 * wasn't moved/renamed. So maybe some of the new ancestors higher in
3266 * the hierarchy have an higher inode number too *and* were renamed
3267 * or moved - in this case we need to wait for the ancestor's rename
3268 * or move operation before we can do the move/rename for the current
3271 register_upper_dirs = 0;
3272 ino = parent_ino_after;
3274 while ((ret == 0 || register_upper_dirs) && ino > sctx->cur_ino) {
3277 fs_path_reset(path_before);
3278 fs_path_reset(path_after);
3280 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3281 &parent_gen, path_after);
3284 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3286 if (ret == -ENOENT) {
3289 } else if (ret < 0) {
3293 len1 = fs_path_len(path_before);
3294 len2 = fs_path_len(path_after);
3295 if (parent_ino_before != parent_ino_after || len1 != len2 ||
3296 memcmp(path_before->start, path_after->start, len1)) {
3298 if (register_upper_dirs) {
3301 register_upper_dirs = 1;
3302 ino = parent_ref->dir;
3303 gen = parent_ref->dir_gen;
3306 } else if (register_upper_dirs) {
3307 ret = add_pending_dir_move(sctx, ino, gen,
3309 if (ret < 0 && ret != -EEXIST)
3313 ino = parent_ino_after;
3318 fs_path_free(path_before);
3319 fs_path_free(path_after);
3325 * This does all the move/link/unlink/rmdir magic.
3327 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3330 struct recorded_ref *cur;
3331 struct recorded_ref *cur2;
3332 struct list_head check_dirs;
3333 struct fs_path *valid_path = NULL;
3336 int did_overwrite = 0;
3338 u64 last_dir_ino_rm = 0;
3340 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3343 * This should never happen as the root dir always has the same ref
3344 * which is always '..'
3346 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3347 INIT_LIST_HEAD(&check_dirs);
3349 valid_path = fs_path_alloc();
3356 * First, check if the first ref of the current inode was overwritten
3357 * before. If yes, we know that the current inode was already orphanized
3358 * and thus use the orphan name. If not, we can use get_cur_path to
3359 * get the path of the first ref as it would like while receiving at
3360 * this point in time.
3361 * New inodes are always orphan at the beginning, so force to use the
3362 * orphan name in this case.
3363 * The first ref is stored in valid_path and will be updated if it
3364 * gets moved around.
3366 if (!sctx->cur_inode_new) {
3367 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3368 sctx->cur_inode_gen);
3374 if (sctx->cur_inode_new || did_overwrite) {
3375 ret = gen_unique_name(sctx, sctx->cur_ino,
3376 sctx->cur_inode_gen, valid_path);
3381 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3387 list_for_each_entry(cur, &sctx->new_refs, list) {
3389 * We may have refs where the parent directory does not exist
3390 * yet. This happens if the parent directories inum is higher
3391 * the the current inum. To handle this case, we create the
3392 * parent directory out of order. But we need to check if this
3393 * did already happen before due to other refs in the same dir.
3395 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3398 if (ret == inode_state_will_create) {
3401 * First check if any of the current inodes refs did
3402 * already create the dir.
3404 list_for_each_entry(cur2, &sctx->new_refs, list) {
3407 if (cur2->dir == cur->dir) {
3414 * If that did not happen, check if a previous inode
3415 * did already create the dir.
3418 ret = did_create_dir(sctx, cur->dir);
3422 ret = send_create_inode(sctx, cur->dir);
3429 * Check if this new ref would overwrite the first ref of
3430 * another unprocessed inode. If yes, orphanize the
3431 * overwritten inode. If we find an overwritten ref that is
3432 * not the first ref, simply unlink it.
3434 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3435 cur->name, cur->name_len,
3436 &ow_inode, &ow_gen);
3440 ret = is_first_ref(sctx->parent_root,
3441 ow_inode, cur->dir, cur->name,
3446 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3451 ret = send_unlink(sctx, cur->full_path);
3458 * link/move the ref to the new place. If we have an orphan
3459 * inode, move it and update valid_path. If not, link or move
3460 * it depending on the inode mode.
3463 ret = send_rename(sctx, valid_path, cur->full_path);
3467 ret = fs_path_copy(valid_path, cur->full_path);
3471 if (S_ISDIR(sctx->cur_inode_mode)) {
3473 * Dirs can't be linked, so move it. For moved
3474 * dirs, we always have one new and one deleted
3475 * ref. The deleted ref is ignored later.
3477 ret = wait_for_parent_move(sctx, cur);
3481 ret = add_pending_dir_move(sctx,
3483 sctx->cur_inode_gen,
3487 ret = send_rename(sctx, valid_path,
3490 ret = fs_path_copy(valid_path,
3496 ret = send_link(sctx, cur->full_path,
3502 ret = dup_ref(cur, &check_dirs);
3507 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3509 * Check if we can already rmdir the directory. If not,
3510 * orphanize it. For every dir item inside that gets deleted
3511 * later, we do this check again and rmdir it then if possible.
3512 * See the use of check_dirs for more details.
3514 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3519 ret = send_rmdir(sctx, valid_path);
3522 } else if (!is_orphan) {
3523 ret = orphanize_inode(sctx, sctx->cur_ino,
3524 sctx->cur_inode_gen, valid_path);
3530 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3531 ret = dup_ref(cur, &check_dirs);
3535 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3536 !list_empty(&sctx->deleted_refs)) {
3538 * We have a moved dir. Add the old parent to check_dirs
3540 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3542 ret = dup_ref(cur, &check_dirs);
3545 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3547 * We have a non dir inode. Go through all deleted refs and
3548 * unlink them if they were not already overwritten by other
3551 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3552 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3553 sctx->cur_ino, sctx->cur_inode_gen,
3554 cur->name, cur->name_len);
3558 ret = send_unlink(sctx, cur->full_path);
3562 ret = dup_ref(cur, &check_dirs);
3567 * If the inode is still orphan, unlink the orphan. This may
3568 * happen when a previous inode did overwrite the first ref
3569 * of this inode and no new refs were added for the current
3570 * inode. Unlinking does not mean that the inode is deleted in
3571 * all cases. There may still be links to this inode in other
3575 ret = send_unlink(sctx, valid_path);
3582 * We did collect all parent dirs where cur_inode was once located. We
3583 * now go through all these dirs and check if they are pending for
3584 * deletion and if it's finally possible to perform the rmdir now.
3585 * We also update the inode stats of the parent dirs here.
3587 list_for_each_entry(cur, &check_dirs, list) {
3589 * In case we had refs into dirs that were not processed yet,
3590 * we don't need to do the utime and rmdir logic for these dirs.
3591 * The dir will be processed later.
3593 if (cur->dir > sctx->cur_ino)
3596 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3600 if (ret == inode_state_did_create ||
3601 ret == inode_state_no_change) {
3602 /* TODO delayed utimes */
3603 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3606 } else if (ret == inode_state_did_delete &&
3607 cur->dir != last_dir_ino_rm) {
3608 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3613 ret = get_cur_path(sctx, cur->dir,
3614 cur->dir_gen, valid_path);
3617 ret = send_rmdir(sctx, valid_path);
3620 last_dir_ino_rm = cur->dir;
3628 __free_recorded_refs(&check_dirs);
3629 free_recorded_refs(sctx);
3630 fs_path_free(valid_path);
3634 static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3635 struct fs_path *name, void *ctx, struct list_head *refs)
3638 struct send_ctx *sctx = ctx;
3642 p = fs_path_alloc();
3646 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
3651 ret = get_cur_path(sctx, dir, gen, p);
3654 ret = fs_path_add_path(p, name);
3658 ret = __record_ref(refs, dir, gen, p);
3666 static int __record_new_ref(int num, u64 dir, int index,
3667 struct fs_path *name,
3670 struct send_ctx *sctx = ctx;
3671 return record_ref(sctx->send_root, num, dir, index, name,
3672 ctx, &sctx->new_refs);
3676 static int __record_deleted_ref(int num, u64 dir, int index,
3677 struct fs_path *name,
3680 struct send_ctx *sctx = ctx;
3681 return record_ref(sctx->parent_root, num, dir, index, name,
3682 ctx, &sctx->deleted_refs);
3685 static int record_new_ref(struct send_ctx *sctx)
3689 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3690 sctx->cmp_key, 0, __record_new_ref, sctx);
3699 static int record_deleted_ref(struct send_ctx *sctx)
3703 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3704 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3713 struct find_ref_ctx {
3716 struct btrfs_root *root;
3717 struct fs_path *name;
3721 static int __find_iref(int num, u64 dir, int index,
3722 struct fs_path *name,
3725 struct find_ref_ctx *ctx = ctx_;
3729 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3730 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3732 * To avoid doing extra lookups we'll only do this if everything
3735 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3739 if (dir_gen != ctx->dir_gen)
3741 ctx->found_idx = num;
3747 static int find_iref(struct btrfs_root *root,
3748 struct btrfs_path *path,
3749 struct btrfs_key *key,
3750 u64 dir, u64 dir_gen, struct fs_path *name)
3753 struct find_ref_ctx ctx;
3757 ctx.dir_gen = dir_gen;
3761 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3765 if (ctx.found_idx == -1)
3768 return ctx.found_idx;
3771 static int __record_changed_new_ref(int num, u64 dir, int index,
3772 struct fs_path *name,
3777 struct send_ctx *sctx = ctx;
3779 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3784 ret = find_iref(sctx->parent_root, sctx->right_path,
3785 sctx->cmp_key, dir, dir_gen, name);
3787 ret = __record_new_ref(num, dir, index, name, sctx);
3794 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3795 struct fs_path *name,
3800 struct send_ctx *sctx = ctx;
3802 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
3807 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
3808 dir, dir_gen, name);
3810 ret = __record_deleted_ref(num, dir, index, name, sctx);
3817 static int record_changed_ref(struct send_ctx *sctx)
3821 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3822 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3825 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3826 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3836 * Record and process all refs at once. Needed when an inode changes the
3837 * generation number, which means that it was deleted and recreated.
3839 static int process_all_refs(struct send_ctx *sctx,
3840 enum btrfs_compare_tree_result cmd)
3843 struct btrfs_root *root;
3844 struct btrfs_path *path;
3845 struct btrfs_key key;
3846 struct btrfs_key found_key;
3847 struct extent_buffer *eb;
3849 iterate_inode_ref_t cb;
3850 int pending_move = 0;
3852 path = alloc_path_for_send();
3856 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3857 root = sctx->send_root;
3858 cb = __record_new_ref;
3859 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3860 root = sctx->parent_root;
3861 cb = __record_deleted_ref;
3863 btrfs_err(sctx->send_root->fs_info,
3864 "Wrong command %d in process_all_refs", cmd);
3869 key.objectid = sctx->cmp_key->objectid;
3870 key.type = BTRFS_INODE_REF_KEY;
3872 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3877 eb = path->nodes[0];
3878 slot = path->slots[0];
3879 if (slot >= btrfs_header_nritems(eb)) {
3880 ret = btrfs_next_leaf(root, path);
3888 btrfs_item_key_to_cpu(eb, &found_key, slot);
3890 if (found_key.objectid != key.objectid ||
3891 (found_key.type != BTRFS_INODE_REF_KEY &&
3892 found_key.type != BTRFS_INODE_EXTREF_KEY))
3895 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
3901 btrfs_release_path(path);
3903 ret = process_recorded_refs(sctx, &pending_move);
3904 /* Only applicable to an incremental send. */
3905 ASSERT(pending_move == 0);
3908 btrfs_free_path(path);
3912 static int send_set_xattr(struct send_ctx *sctx,
3913 struct fs_path *path,
3914 const char *name, int name_len,
3915 const char *data, int data_len)
3919 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3923 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3924 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3925 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3927 ret = send_cmd(sctx);
3934 static int send_remove_xattr(struct send_ctx *sctx,
3935 struct fs_path *path,
3936 const char *name, int name_len)
3940 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3944 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3945 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3947 ret = send_cmd(sctx);
3954 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3955 const char *name, int name_len,
3956 const char *data, int data_len,
3960 struct send_ctx *sctx = ctx;
3962 posix_acl_xattr_header dummy_acl;
3964 p = fs_path_alloc();
3969 * This hack is needed because empty acl's are stored as zero byte
3970 * data in xattrs. Problem with that is, that receiving these zero byte
3971 * acl's will fail later. To fix this, we send a dummy acl list that
3972 * only contains the version number and no entries.
3974 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3975 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3976 if (data_len == 0) {
3977 dummy_acl.a_version =
3978 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3979 data = (char *)&dummy_acl;
3980 data_len = sizeof(dummy_acl);
3984 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3988 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3995 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3996 const char *name, int name_len,
3997 const char *data, int data_len,
4001 struct send_ctx *sctx = ctx;
4004 p = fs_path_alloc();
4008 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4012 ret = send_remove_xattr(sctx, p, name, name_len);
4019 static int process_new_xattr(struct send_ctx *sctx)
4023 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4024 sctx->cmp_key, __process_new_xattr, sctx);
4029 static int process_deleted_xattr(struct send_ctx *sctx)
4033 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4034 sctx->cmp_key, __process_deleted_xattr, sctx);
4039 struct find_xattr_ctx {
4047 static int __find_xattr(int num, struct btrfs_key *di_key,
4048 const char *name, int name_len,
4049 const char *data, int data_len,
4050 u8 type, void *vctx)
4052 struct find_xattr_ctx *ctx = vctx;
4054 if (name_len == ctx->name_len &&
4055 strncmp(name, ctx->name, name_len) == 0) {
4056 ctx->found_idx = num;
4057 ctx->found_data_len = data_len;
4058 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
4059 if (!ctx->found_data)
4066 static int find_xattr(struct btrfs_root *root,
4067 struct btrfs_path *path,
4068 struct btrfs_key *key,
4069 const char *name, int name_len,
4070 char **data, int *data_len)
4073 struct find_xattr_ctx ctx;
4076 ctx.name_len = name_len;
4078 ctx.found_data = NULL;
4079 ctx.found_data_len = 0;
4081 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4085 if (ctx.found_idx == -1)
4088 *data = ctx.found_data;
4089 *data_len = ctx.found_data_len;
4091 kfree(ctx.found_data);
4093 return ctx.found_idx;
4097 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4098 const char *name, int name_len,
4099 const char *data, int data_len,
4103 struct send_ctx *sctx = ctx;
4104 char *found_data = NULL;
4105 int found_data_len = 0;
4107 ret = find_xattr(sctx->parent_root, sctx->right_path,
4108 sctx->cmp_key, name, name_len, &found_data,
4110 if (ret == -ENOENT) {
4111 ret = __process_new_xattr(num, di_key, name, name_len, data,
4112 data_len, type, ctx);
4113 } else if (ret >= 0) {
4114 if (data_len != found_data_len ||
4115 memcmp(data, found_data, data_len)) {
4116 ret = __process_new_xattr(num, di_key, name, name_len,
4117 data, data_len, type, ctx);
4127 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4128 const char *name, int name_len,
4129 const char *data, int data_len,
4133 struct send_ctx *sctx = ctx;
4135 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4136 name, name_len, NULL, NULL);
4138 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4139 data_len, type, ctx);
4146 static int process_changed_xattr(struct send_ctx *sctx)
4150 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4151 sctx->cmp_key, __process_changed_new_xattr, sctx);
4154 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4155 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4161 static int process_all_new_xattrs(struct send_ctx *sctx)
4164 struct btrfs_root *root;
4165 struct btrfs_path *path;
4166 struct btrfs_key key;
4167 struct btrfs_key found_key;
4168 struct extent_buffer *eb;
4171 path = alloc_path_for_send();
4175 root = sctx->send_root;
4177 key.objectid = sctx->cmp_key->objectid;
4178 key.type = BTRFS_XATTR_ITEM_KEY;
4180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4185 eb = path->nodes[0];
4186 slot = path->slots[0];
4187 if (slot >= btrfs_header_nritems(eb)) {
4188 ret = btrfs_next_leaf(root, path);
4191 } else if (ret > 0) {
4198 btrfs_item_key_to_cpu(eb, &found_key, slot);
4199 if (found_key.objectid != key.objectid ||
4200 found_key.type != key.type) {
4205 ret = iterate_dir_item(root, path, &found_key,
4206 __process_new_xattr, sctx);
4214 btrfs_free_path(path);
4218 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4220 struct btrfs_root *root = sctx->send_root;
4221 struct btrfs_fs_info *fs_info = root->fs_info;
4222 struct inode *inode;
4225 struct btrfs_key key;
4226 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4228 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4231 key.objectid = sctx->cur_ino;
4232 key.type = BTRFS_INODE_ITEM_KEY;
4235 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4237 return PTR_ERR(inode);
4239 if (offset + len > i_size_read(inode)) {
4240 if (offset > i_size_read(inode))
4243 len = offset - i_size_read(inode);
4248 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4250 /* initial readahead */
4251 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4252 file_ra_state_init(&sctx->ra, inode->i_mapping);
4253 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4254 last_index - index + 1);
4256 while (index <= last_index) {
4257 unsigned cur_len = min_t(unsigned, len,
4258 PAGE_CACHE_SIZE - pg_offset);
4259 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4265 if (!PageUptodate(page)) {
4266 btrfs_readpage(NULL, page);
4268 if (!PageUptodate(page)) {
4270 page_cache_release(page);
4277 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4280 page_cache_release(page);
4292 * Read some bytes from the current inode/file and send a write command to
4295 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4299 ssize_t num_read = 0;
4301 p = fs_path_alloc();
4305 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4307 num_read = fill_read_buf(sctx, offset, len);
4308 if (num_read <= 0) {
4314 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4318 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4322 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4323 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4324 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4326 ret = send_cmd(sctx);
4337 * Send a clone command to user space.
4339 static int send_clone(struct send_ctx *sctx,
4340 u64 offset, u32 len,
4341 struct clone_root *clone_root)
4347 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4348 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4349 clone_root->root->objectid, clone_root->ino,
4350 clone_root->offset);
4352 p = fs_path_alloc();
4356 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4360 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4364 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4365 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4366 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4368 if (clone_root->root == sctx->send_root) {
4369 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4370 &gen, NULL, NULL, NULL, NULL);
4373 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4375 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4380 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4381 clone_root->root->root_item.uuid);
4382 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4383 le64_to_cpu(clone_root->root->root_item.ctransid));
4384 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4385 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4386 clone_root->offset);
4388 ret = send_cmd(sctx);
4397 * Send an update extent command to user space.
4399 static int send_update_extent(struct send_ctx *sctx,
4400 u64 offset, u32 len)
4405 p = fs_path_alloc();
4409 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4413 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4417 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4418 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4419 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4421 ret = send_cmd(sctx);
4429 static int send_hole(struct send_ctx *sctx, u64 end)
4431 struct fs_path *p = NULL;
4432 u64 offset = sctx->cur_inode_last_extent;
4436 p = fs_path_alloc();
4439 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4441 goto tlv_put_failure;
4442 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4443 while (offset < end) {
4444 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4446 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4449 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4450 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4451 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4452 ret = send_cmd(sctx);
4462 static int send_write_or_clone(struct send_ctx *sctx,
4463 struct btrfs_path *path,
4464 struct btrfs_key *key,
4465 struct clone_root *clone_root)
4468 struct btrfs_file_extent_item *ei;
4469 u64 offset = key->offset;
4474 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4476 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4477 struct btrfs_file_extent_item);
4478 type = btrfs_file_extent_type(path->nodes[0], ei);
4479 if (type == BTRFS_FILE_EXTENT_INLINE) {
4480 len = btrfs_file_extent_inline_len(path->nodes[0],
4481 path->slots[0], ei);
4483 * it is possible the inline item won't cover the whole page,
4484 * but there may be items after this page. Make
4485 * sure to send the whole thing
4487 len = PAGE_CACHE_ALIGN(len);
4489 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4492 if (offset + len > sctx->cur_inode_size)
4493 len = sctx->cur_inode_size - offset;
4499 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4500 ret = send_clone(sctx, offset, len, clone_root);
4501 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4502 ret = send_update_extent(sctx, offset, len);
4506 if (l > BTRFS_SEND_READ_SIZE)
4507 l = BTRFS_SEND_READ_SIZE;
4508 ret = send_write(sctx, pos + offset, l);
4521 static int is_extent_unchanged(struct send_ctx *sctx,
4522 struct btrfs_path *left_path,
4523 struct btrfs_key *ekey)
4526 struct btrfs_key key;
4527 struct btrfs_path *path = NULL;
4528 struct extent_buffer *eb;
4530 struct btrfs_key found_key;
4531 struct btrfs_file_extent_item *ei;
4536 u64 left_offset_fixed;
4544 path = alloc_path_for_send();
4548 eb = left_path->nodes[0];
4549 slot = left_path->slots[0];
4550 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4551 left_type = btrfs_file_extent_type(eb, ei);
4553 if (left_type != BTRFS_FILE_EXTENT_REG) {
4557 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4558 left_len = btrfs_file_extent_num_bytes(eb, ei);
4559 left_offset = btrfs_file_extent_offset(eb, ei);
4560 left_gen = btrfs_file_extent_generation(eb, ei);
4563 * Following comments will refer to these graphics. L is the left
4564 * extents which we are checking at the moment. 1-8 are the right
4565 * extents that we iterate.
4568 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4571 * |--1--|-2b-|...(same as above)
4573 * Alternative situation. Happens on files where extents got split.
4575 * |-----------7-----------|-6-|
4577 * Alternative situation. Happens on files which got larger.
4580 * Nothing follows after 8.
4583 key.objectid = ekey->objectid;
4584 key.type = BTRFS_EXTENT_DATA_KEY;
4585 key.offset = ekey->offset;
4586 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4595 * Handle special case where the right side has no extents at all.
4597 eb = path->nodes[0];
4598 slot = path->slots[0];
4599 btrfs_item_key_to_cpu(eb, &found_key, slot);
4600 if (found_key.objectid != key.objectid ||
4601 found_key.type != key.type) {
4602 /* If we're a hole then just pretend nothing changed */
4603 ret = (left_disknr) ? 0 : 1;
4608 * We're now on 2a, 2b or 7.
4611 while (key.offset < ekey->offset + left_len) {
4612 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4613 right_type = btrfs_file_extent_type(eb, ei);
4614 if (right_type != BTRFS_FILE_EXTENT_REG) {
4619 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4620 right_len = btrfs_file_extent_num_bytes(eb, ei);
4621 right_offset = btrfs_file_extent_offset(eb, ei);
4622 right_gen = btrfs_file_extent_generation(eb, ei);
4625 * Are we at extent 8? If yes, we know the extent is changed.
4626 * This may only happen on the first iteration.
4628 if (found_key.offset + right_len <= ekey->offset) {
4629 /* If we're a hole just pretend nothing changed */
4630 ret = (left_disknr) ? 0 : 1;
4634 left_offset_fixed = left_offset;
4635 if (key.offset < ekey->offset) {
4636 /* Fix the right offset for 2a and 7. */
4637 right_offset += ekey->offset - key.offset;
4639 /* Fix the left offset for all behind 2a and 2b */
4640 left_offset_fixed += key.offset - ekey->offset;
4644 * Check if we have the same extent.
4646 if (left_disknr != right_disknr ||
4647 left_offset_fixed != right_offset ||
4648 left_gen != right_gen) {
4654 * Go to the next extent.
4656 ret = btrfs_next_item(sctx->parent_root, path);
4660 eb = path->nodes[0];
4661 slot = path->slots[0];
4662 btrfs_item_key_to_cpu(eb, &found_key, slot);
4664 if (ret || found_key.objectid != key.objectid ||
4665 found_key.type != key.type) {
4666 key.offset += right_len;
4669 if (found_key.offset != key.offset + right_len) {
4677 * We're now behind the left extent (treat as unchanged) or at the end
4678 * of the right side (treat as changed).
4680 if (key.offset >= ekey->offset + left_len)
4687 btrfs_free_path(path);
4691 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4693 struct btrfs_path *path;
4694 struct btrfs_root *root = sctx->send_root;
4695 struct btrfs_file_extent_item *fi;
4696 struct btrfs_key key;
4701 path = alloc_path_for_send();
4705 sctx->cur_inode_last_extent = 0;
4707 key.objectid = sctx->cur_ino;
4708 key.type = BTRFS_EXTENT_DATA_KEY;
4709 key.offset = offset;
4710 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4714 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4715 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4718 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4719 struct btrfs_file_extent_item);
4720 type = btrfs_file_extent_type(path->nodes[0], fi);
4721 if (type == BTRFS_FILE_EXTENT_INLINE) {
4722 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4723 path->slots[0], fi);
4724 extent_end = ALIGN(key.offset + size,
4725 sctx->send_root->sectorsize);
4727 extent_end = key.offset +
4728 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4730 sctx->cur_inode_last_extent = extent_end;
4732 btrfs_free_path(path);
4736 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4737 struct btrfs_key *key)
4739 struct btrfs_file_extent_item *fi;
4744 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4747 if (sctx->cur_inode_last_extent == (u64)-1) {
4748 ret = get_last_extent(sctx, key->offset - 1);
4753 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4754 struct btrfs_file_extent_item);
4755 type = btrfs_file_extent_type(path->nodes[0], fi);
4756 if (type == BTRFS_FILE_EXTENT_INLINE) {
4757 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4758 path->slots[0], fi);
4759 extent_end = ALIGN(key->offset + size,
4760 sctx->send_root->sectorsize);
4762 extent_end = key->offset +
4763 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4766 if (path->slots[0] == 0 &&
4767 sctx->cur_inode_last_extent < key->offset) {
4769 * We might have skipped entire leafs that contained only
4770 * file extent items for our current inode. These leafs have
4771 * a generation number smaller (older) than the one in the
4772 * current leaf and the leaf our last extent came from, and
4773 * are located between these 2 leafs.
4775 ret = get_last_extent(sctx, key->offset - 1);
4780 if (sctx->cur_inode_last_extent < key->offset)
4781 ret = send_hole(sctx, key->offset);
4782 sctx->cur_inode_last_extent = extent_end;
4786 static int process_extent(struct send_ctx *sctx,
4787 struct btrfs_path *path,
4788 struct btrfs_key *key)
4790 struct clone_root *found_clone = NULL;
4793 if (S_ISLNK(sctx->cur_inode_mode))
4796 if (sctx->parent_root && !sctx->cur_inode_new) {
4797 ret = is_extent_unchanged(sctx, path, key);
4805 struct btrfs_file_extent_item *ei;
4808 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4809 struct btrfs_file_extent_item);
4810 type = btrfs_file_extent_type(path->nodes[0], ei);
4811 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
4812 type == BTRFS_FILE_EXTENT_REG) {
4814 * The send spec does not have a prealloc command yet,
4815 * so just leave a hole for prealloc'ed extents until
4816 * we have enough commands queued up to justify rev'ing
4819 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
4824 /* Have a hole, just skip it. */
4825 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
4832 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
4833 sctx->cur_inode_size, &found_clone);
4834 if (ret != -ENOENT && ret < 0)
4837 ret = send_write_or_clone(sctx, path, key, found_clone);
4841 ret = maybe_send_hole(sctx, path, key);
4846 static int process_all_extents(struct send_ctx *sctx)
4849 struct btrfs_root *root;
4850 struct btrfs_path *path;
4851 struct btrfs_key key;
4852 struct btrfs_key found_key;
4853 struct extent_buffer *eb;
4856 root = sctx->send_root;
4857 path = alloc_path_for_send();
4861 key.objectid = sctx->cmp_key->objectid;
4862 key.type = BTRFS_EXTENT_DATA_KEY;
4864 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4869 eb = path->nodes[0];
4870 slot = path->slots[0];
4872 if (slot >= btrfs_header_nritems(eb)) {
4873 ret = btrfs_next_leaf(root, path);
4876 } else if (ret > 0) {
4883 btrfs_item_key_to_cpu(eb, &found_key, slot);
4885 if (found_key.objectid != key.objectid ||
4886 found_key.type != key.type) {
4891 ret = process_extent(sctx, path, &found_key);
4899 btrfs_free_path(path);
4903 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
4905 int *refs_processed)
4909 if (sctx->cur_ino == 0)
4911 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4912 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4914 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4917 ret = process_recorded_refs(sctx, pending_move);
4921 *refs_processed = 1;
4926 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4937 int pending_move = 0;
4938 int refs_processed = 0;
4940 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
4946 * We have processed the refs and thus need to advance send_progress.
4947 * Now, calls to get_cur_xxx will take the updated refs of the current
4948 * inode into account.
4950 * On the other hand, if our current inode is a directory and couldn't
4951 * be moved/renamed because its parent was renamed/moved too and it has
4952 * a higher inode number, we can only move/rename our current inode
4953 * after we moved/renamed its parent. Therefore in this case operate on
4954 * the old path (pre move/rename) of our current inode, and the
4955 * move/rename will be performed later.
4957 if (refs_processed && !pending_move)
4958 sctx->send_progress = sctx->cur_ino + 1;
4960 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4962 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4965 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4966 &left_mode, &left_uid, &left_gid, NULL);
4970 if (!sctx->parent_root || sctx->cur_inode_new) {
4972 if (!S_ISLNK(sctx->cur_inode_mode))
4975 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4976 NULL, NULL, &right_mode, &right_uid,
4981 if (left_uid != right_uid || left_gid != right_gid)
4983 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4987 if (S_ISREG(sctx->cur_inode_mode)) {
4988 if (need_send_hole(sctx)) {
4989 if (sctx->cur_inode_last_extent == (u64)-1 ||
4990 sctx->cur_inode_last_extent <
4991 sctx->cur_inode_size) {
4992 ret = get_last_extent(sctx, (u64)-1);
4996 if (sctx->cur_inode_last_extent <
4997 sctx->cur_inode_size) {
4998 ret = send_hole(sctx, sctx->cur_inode_size);
5003 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5004 sctx->cur_inode_size);
5010 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5011 left_uid, left_gid);
5016 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5023 * If other directory inodes depended on our current directory
5024 * inode's move/rename, now do their move/rename operations.
5026 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5027 ret = apply_children_dir_moves(sctx);
5031 * Need to send that every time, no matter if it actually
5032 * changed between the two trees as we have done changes to
5033 * the inode before. If our inode is a directory and it's
5034 * waiting to be moved/renamed, we will send its utimes when
5035 * it's moved/renamed, therefore we don't need to do it here.
5037 sctx->send_progress = sctx->cur_ino + 1;
5038 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5047 static int changed_inode(struct send_ctx *sctx,
5048 enum btrfs_compare_tree_result result)
5051 struct btrfs_key *key = sctx->cmp_key;
5052 struct btrfs_inode_item *left_ii = NULL;
5053 struct btrfs_inode_item *right_ii = NULL;
5057 sctx->cur_ino = key->objectid;
5058 sctx->cur_inode_new_gen = 0;
5059 sctx->cur_inode_last_extent = (u64)-1;
5062 * Set send_progress to current inode. This will tell all get_cur_xxx
5063 * functions that the current inode's refs are not updated yet. Later,
5064 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5066 sctx->send_progress = sctx->cur_ino;
5068 if (result == BTRFS_COMPARE_TREE_NEW ||
5069 result == BTRFS_COMPARE_TREE_CHANGED) {
5070 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5071 sctx->left_path->slots[0],
5072 struct btrfs_inode_item);
5073 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5076 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5077 sctx->right_path->slots[0],
5078 struct btrfs_inode_item);
5079 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5082 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5083 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5084 sctx->right_path->slots[0],
5085 struct btrfs_inode_item);
5087 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5091 * The cur_ino = root dir case is special here. We can't treat
5092 * the inode as deleted+reused because it would generate a
5093 * stream that tries to delete/mkdir the root dir.
5095 if (left_gen != right_gen &&
5096 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5097 sctx->cur_inode_new_gen = 1;
5100 if (result == BTRFS_COMPARE_TREE_NEW) {
5101 sctx->cur_inode_gen = left_gen;
5102 sctx->cur_inode_new = 1;
5103 sctx->cur_inode_deleted = 0;
5104 sctx->cur_inode_size = btrfs_inode_size(
5105 sctx->left_path->nodes[0], left_ii);
5106 sctx->cur_inode_mode = btrfs_inode_mode(
5107 sctx->left_path->nodes[0], left_ii);
5108 sctx->cur_inode_rdev = btrfs_inode_rdev(
5109 sctx->left_path->nodes[0], left_ii);
5110 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5111 ret = send_create_inode_if_needed(sctx);
5112 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5113 sctx->cur_inode_gen = right_gen;
5114 sctx->cur_inode_new = 0;
5115 sctx->cur_inode_deleted = 1;
5116 sctx->cur_inode_size = btrfs_inode_size(
5117 sctx->right_path->nodes[0], right_ii);
5118 sctx->cur_inode_mode = btrfs_inode_mode(
5119 sctx->right_path->nodes[0], right_ii);
5120 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5122 * We need to do some special handling in case the inode was
5123 * reported as changed with a changed generation number. This
5124 * means that the original inode was deleted and new inode
5125 * reused the same inum. So we have to treat the old inode as
5126 * deleted and the new one as new.
5128 if (sctx->cur_inode_new_gen) {
5130 * First, process the inode as if it was deleted.
5132 sctx->cur_inode_gen = right_gen;
5133 sctx->cur_inode_new = 0;
5134 sctx->cur_inode_deleted = 1;
5135 sctx->cur_inode_size = btrfs_inode_size(
5136 sctx->right_path->nodes[0], right_ii);
5137 sctx->cur_inode_mode = btrfs_inode_mode(
5138 sctx->right_path->nodes[0], right_ii);
5139 ret = process_all_refs(sctx,
5140 BTRFS_COMPARE_TREE_DELETED);
5145 * Now process the inode as if it was new.
5147 sctx->cur_inode_gen = left_gen;
5148 sctx->cur_inode_new = 1;
5149 sctx->cur_inode_deleted = 0;
5150 sctx->cur_inode_size = btrfs_inode_size(
5151 sctx->left_path->nodes[0], left_ii);
5152 sctx->cur_inode_mode = btrfs_inode_mode(
5153 sctx->left_path->nodes[0], left_ii);
5154 sctx->cur_inode_rdev = btrfs_inode_rdev(
5155 sctx->left_path->nodes[0], left_ii);
5156 ret = send_create_inode_if_needed(sctx);
5160 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5164 * Advance send_progress now as we did not get into
5165 * process_recorded_refs_if_needed in the new_gen case.
5167 sctx->send_progress = sctx->cur_ino + 1;
5170 * Now process all extents and xattrs of the inode as if
5171 * they were all new.
5173 ret = process_all_extents(sctx);
5176 ret = process_all_new_xattrs(sctx);
5180 sctx->cur_inode_gen = left_gen;
5181 sctx->cur_inode_new = 0;
5182 sctx->cur_inode_new_gen = 0;
5183 sctx->cur_inode_deleted = 0;
5184 sctx->cur_inode_size = btrfs_inode_size(
5185 sctx->left_path->nodes[0], left_ii);
5186 sctx->cur_inode_mode = btrfs_inode_mode(
5187 sctx->left_path->nodes[0], left_ii);
5196 * We have to process new refs before deleted refs, but compare_trees gives us
5197 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5198 * first and later process them in process_recorded_refs.
5199 * For the cur_inode_new_gen case, we skip recording completely because
5200 * changed_inode did already initiate processing of refs. The reason for this is
5201 * that in this case, compare_tree actually compares the refs of 2 different
5202 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5203 * refs of the right tree as deleted and all refs of the left tree as new.
5205 static int changed_ref(struct send_ctx *sctx,
5206 enum btrfs_compare_tree_result result)
5210 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5212 if (!sctx->cur_inode_new_gen &&
5213 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5214 if (result == BTRFS_COMPARE_TREE_NEW)
5215 ret = record_new_ref(sctx);
5216 else if (result == BTRFS_COMPARE_TREE_DELETED)
5217 ret = record_deleted_ref(sctx);
5218 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5219 ret = record_changed_ref(sctx);
5226 * Process new/deleted/changed xattrs. We skip processing in the
5227 * cur_inode_new_gen case because changed_inode did already initiate processing
5228 * of xattrs. The reason is the same as in changed_ref
5230 static int changed_xattr(struct send_ctx *sctx,
5231 enum btrfs_compare_tree_result result)
5235 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5237 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5238 if (result == BTRFS_COMPARE_TREE_NEW)
5239 ret = process_new_xattr(sctx);
5240 else if (result == BTRFS_COMPARE_TREE_DELETED)
5241 ret = process_deleted_xattr(sctx);
5242 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5243 ret = process_changed_xattr(sctx);
5250 * Process new/deleted/changed extents. We skip processing in the
5251 * cur_inode_new_gen case because changed_inode did already initiate processing
5252 * of extents. The reason is the same as in changed_ref
5254 static int changed_extent(struct send_ctx *sctx,
5255 enum btrfs_compare_tree_result result)
5259 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5261 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5262 if (result != BTRFS_COMPARE_TREE_DELETED)
5263 ret = process_extent(sctx, sctx->left_path,
5270 static int dir_changed(struct send_ctx *sctx, u64 dir)
5272 u64 orig_gen, new_gen;
5275 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5280 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5285 return (orig_gen != new_gen) ? 1 : 0;
5288 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5289 struct btrfs_key *key)
5291 struct btrfs_inode_extref *extref;
5292 struct extent_buffer *leaf;
5293 u64 dirid = 0, last_dirid = 0;
5300 /* Easy case, just check this one dirid */
5301 if (key->type == BTRFS_INODE_REF_KEY) {
5302 dirid = key->offset;
5304 ret = dir_changed(sctx, dirid);
5308 leaf = path->nodes[0];
5309 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5310 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5311 while (cur_offset < item_size) {
5312 extref = (struct btrfs_inode_extref *)(ptr +
5314 dirid = btrfs_inode_extref_parent(leaf, extref);
5315 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5316 cur_offset += ref_name_len + sizeof(*extref);
5317 if (dirid == last_dirid)
5319 ret = dir_changed(sctx, dirid);
5329 * Updates compare related fields in sctx and simply forwards to the actual
5330 * changed_xxx functions.
5332 static int changed_cb(struct btrfs_root *left_root,
5333 struct btrfs_root *right_root,
5334 struct btrfs_path *left_path,
5335 struct btrfs_path *right_path,
5336 struct btrfs_key *key,
5337 enum btrfs_compare_tree_result result,
5341 struct send_ctx *sctx = ctx;
5343 if (result == BTRFS_COMPARE_TREE_SAME) {
5344 if (key->type == BTRFS_INODE_REF_KEY ||
5345 key->type == BTRFS_INODE_EXTREF_KEY) {
5346 ret = compare_refs(sctx, left_path, key);
5351 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5352 return maybe_send_hole(sctx, left_path, key);
5356 result = BTRFS_COMPARE_TREE_CHANGED;
5360 sctx->left_path = left_path;
5361 sctx->right_path = right_path;
5362 sctx->cmp_key = key;
5364 ret = finish_inode_if_needed(sctx, 0);
5368 /* Ignore non-FS objects */
5369 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5370 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5373 if (key->type == BTRFS_INODE_ITEM_KEY)
5374 ret = changed_inode(sctx, result);
5375 else if (key->type == BTRFS_INODE_REF_KEY ||
5376 key->type == BTRFS_INODE_EXTREF_KEY)
5377 ret = changed_ref(sctx, result);
5378 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5379 ret = changed_xattr(sctx, result);
5380 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5381 ret = changed_extent(sctx, result);
5387 static int full_send_tree(struct send_ctx *sctx)
5390 struct btrfs_root *send_root = sctx->send_root;
5391 struct btrfs_key key;
5392 struct btrfs_key found_key;
5393 struct btrfs_path *path;
5394 struct extent_buffer *eb;
5397 path = alloc_path_for_send();
5401 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5402 key.type = BTRFS_INODE_ITEM_KEY;
5405 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5412 eb = path->nodes[0];
5413 slot = path->slots[0];
5414 btrfs_item_key_to_cpu(eb, &found_key, slot);
5416 ret = changed_cb(send_root, NULL, path, NULL,
5417 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5421 key.objectid = found_key.objectid;
5422 key.type = found_key.type;
5423 key.offset = found_key.offset + 1;
5425 ret = btrfs_next_item(send_root, path);
5435 ret = finish_inode_if_needed(sctx, 1);
5438 btrfs_free_path(path);
5442 static int send_subvol(struct send_ctx *sctx)
5446 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5447 ret = send_header(sctx);
5452 ret = send_subvol_begin(sctx);
5456 if (sctx->parent_root) {
5457 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5461 ret = finish_inode_if_needed(sctx, 1);
5465 ret = full_send_tree(sctx);
5471 free_recorded_refs(sctx);
5475 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5477 spin_lock(&root->root_item_lock);
5478 root->send_in_progress--;
5480 * Not much left to do, we don't know why it's unbalanced and
5481 * can't blindly reset it to 0.
5483 if (root->send_in_progress < 0)
5484 btrfs_err(root->fs_info,
5485 "send_in_progres unbalanced %d root %llu\n",
5486 root->send_in_progress, root->root_key.objectid);
5487 spin_unlock(&root->root_item_lock);
5490 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5493 struct btrfs_root *send_root;
5494 struct btrfs_root *clone_root;
5495 struct btrfs_fs_info *fs_info;
5496 struct btrfs_ioctl_send_args *arg = NULL;
5497 struct btrfs_key key;
5498 struct send_ctx *sctx = NULL;
5500 u64 *clone_sources_tmp = NULL;
5501 int clone_sources_to_rollback = 0;
5502 int sort_clone_roots = 0;
5505 if (!capable(CAP_SYS_ADMIN))
5508 send_root = BTRFS_I(file_inode(mnt_file))->root;
5509 fs_info = send_root->fs_info;
5512 * The subvolume must remain read-only during send, protect against
5515 spin_lock(&send_root->root_item_lock);
5516 send_root->send_in_progress++;
5517 spin_unlock(&send_root->root_item_lock);
5520 * This is done when we lookup the root, it should already be complete
5521 * by the time we get here.
5523 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5526 * Userspace tools do the checks and warn the user if it's
5529 if (!btrfs_root_readonly(send_root)) {
5534 arg = memdup_user(arg_, sizeof(*arg));
5541 if (!access_ok(VERIFY_READ, arg->clone_sources,
5542 sizeof(*arg->clone_sources) *
5543 arg->clone_sources_count)) {
5548 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5553 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5559 INIT_LIST_HEAD(&sctx->new_refs);
5560 INIT_LIST_HEAD(&sctx->deleted_refs);
5561 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5562 INIT_LIST_HEAD(&sctx->name_cache_list);
5564 sctx->flags = arg->flags;
5566 sctx->send_filp = fget(arg->send_fd);
5567 if (!sctx->send_filp) {
5572 sctx->send_root = send_root;
5573 sctx->clone_roots_cnt = arg->clone_sources_count;
5575 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5576 sctx->send_buf = vmalloc(sctx->send_max_size);
5577 if (!sctx->send_buf) {
5582 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5583 if (!sctx->read_buf) {
5588 sctx->pending_dir_moves = RB_ROOT;
5589 sctx->waiting_dir_moves = RB_ROOT;
5590 sctx->orphan_dirs = RB_ROOT;
5592 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5593 (arg->clone_sources_count + 1));
5594 if (!sctx->clone_roots) {
5599 if (arg->clone_sources_count) {
5600 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5601 sizeof(*arg->clone_sources));
5602 if (!clone_sources_tmp) {
5607 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5608 arg->clone_sources_count *
5609 sizeof(*arg->clone_sources));
5615 for (i = 0; i < arg->clone_sources_count; i++) {
5616 key.objectid = clone_sources_tmp[i];
5617 key.type = BTRFS_ROOT_ITEM_KEY;
5618 key.offset = (u64)-1;
5620 index = srcu_read_lock(&fs_info->subvol_srcu);
5622 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5623 if (IS_ERR(clone_root)) {
5624 srcu_read_unlock(&fs_info->subvol_srcu, index);
5625 ret = PTR_ERR(clone_root);
5628 clone_sources_to_rollback = i + 1;
5629 spin_lock(&clone_root->root_item_lock);
5630 clone_root->send_in_progress++;
5631 if (!btrfs_root_readonly(clone_root)) {
5632 spin_unlock(&clone_root->root_item_lock);
5633 srcu_read_unlock(&fs_info->subvol_srcu, index);
5637 spin_unlock(&clone_root->root_item_lock);
5638 srcu_read_unlock(&fs_info->subvol_srcu, index);
5640 sctx->clone_roots[i].root = clone_root;
5642 vfree(clone_sources_tmp);
5643 clone_sources_tmp = NULL;
5646 if (arg->parent_root) {
5647 key.objectid = arg->parent_root;
5648 key.type = BTRFS_ROOT_ITEM_KEY;
5649 key.offset = (u64)-1;
5651 index = srcu_read_lock(&fs_info->subvol_srcu);
5653 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5654 if (IS_ERR(sctx->parent_root)) {
5655 srcu_read_unlock(&fs_info->subvol_srcu, index);
5656 ret = PTR_ERR(sctx->parent_root);
5660 spin_lock(&sctx->parent_root->root_item_lock);
5661 sctx->parent_root->send_in_progress++;
5662 if (!btrfs_root_readonly(sctx->parent_root)) {
5663 spin_unlock(&sctx->parent_root->root_item_lock);
5664 srcu_read_unlock(&fs_info->subvol_srcu, index);
5668 spin_unlock(&sctx->parent_root->root_item_lock);
5670 srcu_read_unlock(&fs_info->subvol_srcu, index);
5674 * Clones from send_root are allowed, but only if the clone source
5675 * is behind the current send position. This is checked while searching
5676 * for possible clone sources.
5678 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5680 /* We do a bsearch later */
5681 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5682 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5684 sort_clone_roots = 1;
5686 current->journal_info = (void *)BTRFS_SEND_TRANS_STUB;
5687 ret = send_subvol(sctx);
5688 current->journal_info = NULL;
5692 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5693 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5696 ret = send_cmd(sctx);
5702 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5703 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5705 struct pending_dir_move *pm;
5707 n = rb_first(&sctx->pending_dir_moves);
5708 pm = rb_entry(n, struct pending_dir_move, node);
5709 while (!list_empty(&pm->list)) {
5710 struct pending_dir_move *pm2;
5712 pm2 = list_first_entry(&pm->list,
5713 struct pending_dir_move, list);
5714 free_pending_move(sctx, pm2);
5716 free_pending_move(sctx, pm);
5719 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5720 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5722 struct waiting_dir_move *dm;
5724 n = rb_first(&sctx->waiting_dir_moves);
5725 dm = rb_entry(n, struct waiting_dir_move, node);
5726 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5730 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
5731 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
5733 struct orphan_dir_info *odi;
5735 n = rb_first(&sctx->orphan_dirs);
5736 odi = rb_entry(n, struct orphan_dir_info, node);
5737 free_orphan_dir_info(sctx, odi);
5740 if (sort_clone_roots) {
5741 for (i = 0; i < sctx->clone_roots_cnt; i++)
5742 btrfs_root_dec_send_in_progress(
5743 sctx->clone_roots[i].root);
5745 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5746 btrfs_root_dec_send_in_progress(
5747 sctx->clone_roots[i].root);
5749 btrfs_root_dec_send_in_progress(send_root);
5751 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
5752 btrfs_root_dec_send_in_progress(sctx->parent_root);
5755 vfree(clone_sources_tmp);
5758 if (sctx->send_filp)
5759 fput(sctx->send_filp);
5761 vfree(sctx->clone_roots);
5762 vfree(sctx->send_buf);
5763 vfree(sctx->read_buf);
5765 name_cache_free(sctx);