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;
234 struct list_head update_refs;
237 struct waiting_dir_move {
241 * There might be some directory that could not be removed because it
242 * was waiting for this directory inode to be moved first. Therefore
243 * after this directory is moved, we can try to rmdir the ino rmdir_ino.
249 struct orphan_dir_info {
255 struct name_cache_entry {
256 struct list_head list;
258 * radix_tree has only 32bit entries but we need to handle 64bit inums.
259 * We use the lower 32bit of the 64bit inum to store it in the tree. If
260 * more then one inum would fall into the same entry, we use radix_list
261 * to store the additional entries. radix_list is also used to store
262 * entries where two entries have the same inum but different
265 struct list_head radix_list;
271 int need_later_update;
276 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino);
278 static struct waiting_dir_move *
279 get_waiting_dir_move(struct send_ctx *sctx, u64 ino);
281 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino);
283 static int need_send_hole(struct send_ctx *sctx)
285 return (sctx->parent_root && !sctx->cur_inode_new &&
286 !sctx->cur_inode_new_gen && !sctx->cur_inode_deleted &&
287 S_ISREG(sctx->cur_inode_mode));
290 static void fs_path_reset(struct fs_path *p)
293 p->start = p->buf + p->buf_len - 1;
303 static struct fs_path *fs_path_alloc(void)
307 p = kmalloc(sizeof(*p), GFP_NOFS);
311 p->buf = p->inline_buf;
312 p->buf_len = FS_PATH_INLINE_SIZE;
317 static struct fs_path *fs_path_alloc_reversed(void)
329 static void fs_path_free(struct fs_path *p)
333 if (p->buf != p->inline_buf)
338 static int fs_path_len(struct fs_path *p)
340 return p->end - p->start;
343 static int fs_path_ensure_buf(struct fs_path *p, int len)
351 if (p->buf_len >= len)
354 if (len > PATH_MAX) {
359 path_len = p->end - p->start;
360 old_buf_len = p->buf_len;
363 * First time the inline_buf does not suffice
365 if (p->buf == p->inline_buf) {
366 tmp_buf = kmalloc(len, GFP_NOFS);
368 memcpy(tmp_buf, p->buf, old_buf_len);
370 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
376 * The real size of the buffer is bigger, this will let the fast path
377 * happen most of the time
379 p->buf_len = ksize(p->buf);
382 tmp_buf = p->buf + old_buf_len - path_len - 1;
383 p->end = p->buf + p->buf_len - 1;
384 p->start = p->end - path_len;
385 memmove(p->start, tmp_buf, path_len + 1);
388 p->end = p->start + path_len;
393 static int fs_path_prepare_for_add(struct fs_path *p, int name_len,
399 new_len = p->end - p->start + name_len;
400 if (p->start != p->end)
402 ret = fs_path_ensure_buf(p, new_len);
407 if (p->start != p->end)
409 p->start -= name_len;
410 *prepared = p->start;
412 if (p->start != p->end)
423 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
428 ret = fs_path_prepare_for_add(p, name_len, &prepared);
431 memcpy(prepared, name, name_len);
437 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
442 ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared);
445 memcpy(prepared, p2->start, p2->end - p2->start);
451 static int fs_path_add_from_extent_buffer(struct fs_path *p,
452 struct extent_buffer *eb,
453 unsigned long off, int len)
458 ret = fs_path_prepare_for_add(p, len, &prepared);
462 read_extent_buffer(eb, prepared, off, len);
468 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
472 p->reversed = from->reversed;
475 ret = fs_path_add_path(p, from);
481 static void fs_path_unreverse(struct fs_path *p)
490 len = p->end - p->start;
492 p->end = p->start + len;
493 memmove(p->start, tmp, len + 1);
497 static struct btrfs_path *alloc_path_for_send(void)
499 struct btrfs_path *path;
501 path = btrfs_alloc_path();
504 path->search_commit_root = 1;
505 path->skip_locking = 1;
506 path->need_commit_sem = 1;
510 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
520 ret = vfs_write(filp, (__force const char __user *)buf + pos,
522 /* TODO handle that correctly */
523 /*if (ret == -ERESTARTSYS) {
542 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
544 struct btrfs_tlv_header *hdr;
545 int total_len = sizeof(*hdr) + len;
546 int left = sctx->send_max_size - sctx->send_size;
548 if (unlikely(left < total_len))
551 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
552 hdr->tlv_type = cpu_to_le16(attr);
553 hdr->tlv_len = cpu_to_le16(len);
554 memcpy(hdr + 1, data, len);
555 sctx->send_size += total_len;
560 #define TLV_PUT_DEFINE_INT(bits) \
561 static int tlv_put_u##bits(struct send_ctx *sctx, \
562 u##bits attr, u##bits value) \
564 __le##bits __tmp = cpu_to_le##bits(value); \
565 return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \
568 TLV_PUT_DEFINE_INT(64)
570 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
571 const char *str, int len)
575 return tlv_put(sctx, attr, str, len);
578 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
581 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
584 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
585 struct extent_buffer *eb,
586 struct btrfs_timespec *ts)
588 struct btrfs_timespec bts;
589 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
590 return tlv_put(sctx, attr, &bts, sizeof(bts));
594 #define TLV_PUT(sctx, attrtype, attrlen, data) \
596 ret = tlv_put(sctx, attrtype, attrlen, data); \
598 goto tlv_put_failure; \
601 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
603 ret = tlv_put_u##bits(sctx, attrtype, value); \
605 goto tlv_put_failure; \
608 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
609 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
610 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
611 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
612 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
614 ret = tlv_put_string(sctx, attrtype, str, len); \
616 goto tlv_put_failure; \
618 #define TLV_PUT_PATH(sctx, attrtype, p) \
620 ret = tlv_put_string(sctx, attrtype, p->start, \
621 p->end - p->start); \
623 goto tlv_put_failure; \
625 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
627 ret = tlv_put_uuid(sctx, attrtype, uuid); \
629 goto tlv_put_failure; \
631 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
633 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
635 goto tlv_put_failure; \
638 static int send_header(struct send_ctx *sctx)
640 struct btrfs_stream_header hdr;
642 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
643 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
645 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
650 * For each command/item we want to send to userspace, we call this function.
652 static int begin_cmd(struct send_ctx *sctx, int cmd)
654 struct btrfs_cmd_header *hdr;
656 if (WARN_ON(!sctx->send_buf))
659 BUG_ON(sctx->send_size);
661 sctx->send_size += sizeof(*hdr);
662 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
663 hdr->cmd = cpu_to_le16(cmd);
668 static int send_cmd(struct send_ctx *sctx)
671 struct btrfs_cmd_header *hdr;
674 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
675 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
678 crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
679 hdr->crc = cpu_to_le32(crc);
681 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
684 sctx->total_send_size += sctx->send_size;
685 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
692 * Sends a move instruction to user space
694 static int send_rename(struct send_ctx *sctx,
695 struct fs_path *from, struct fs_path *to)
699 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
701 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
705 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
706 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
708 ret = send_cmd(sctx);
716 * Sends a link instruction to user space
718 static int send_link(struct send_ctx *sctx,
719 struct fs_path *path, struct fs_path *lnk)
723 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
725 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
729 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
730 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
732 ret = send_cmd(sctx);
740 * Sends an unlink instruction to user space
742 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
746 verbose_printk("btrfs: send_unlink %s\n", path->start);
748 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
752 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
754 ret = send_cmd(sctx);
762 * Sends a rmdir instruction to user space
764 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
768 verbose_printk("btrfs: send_rmdir %s\n", path->start);
770 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
774 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
776 ret = send_cmd(sctx);
784 * Helper function to retrieve some fields from an inode item.
786 static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path,
787 u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid,
791 struct btrfs_inode_item *ii;
792 struct btrfs_key key;
795 key.type = BTRFS_INODE_ITEM_KEY;
797 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
804 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
805 struct btrfs_inode_item);
807 *size = btrfs_inode_size(path->nodes[0], ii);
809 *gen = btrfs_inode_generation(path->nodes[0], ii);
811 *mode = btrfs_inode_mode(path->nodes[0], ii);
813 *uid = btrfs_inode_uid(path->nodes[0], ii);
815 *gid = btrfs_inode_gid(path->nodes[0], ii);
817 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
822 static int get_inode_info(struct btrfs_root *root,
823 u64 ino, u64 *size, u64 *gen,
824 u64 *mode, u64 *uid, u64 *gid,
827 struct btrfs_path *path;
830 path = alloc_path_for_send();
833 ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid,
835 btrfs_free_path(path);
839 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
844 * Helper function to iterate the entries in ONE btrfs_inode_ref or
845 * btrfs_inode_extref.
846 * The iterate callback may return a non zero value to stop iteration. This can
847 * be a negative value for error codes or 1 to simply stop it.
849 * path must point to the INODE_REF or INODE_EXTREF when called.
851 static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path,
852 struct btrfs_key *found_key, int resolve,
853 iterate_inode_ref_t iterate, void *ctx)
855 struct extent_buffer *eb = path->nodes[0];
856 struct btrfs_item *item;
857 struct btrfs_inode_ref *iref;
858 struct btrfs_inode_extref *extref;
859 struct btrfs_path *tmp_path;
863 int slot = path->slots[0];
870 unsigned long name_off;
871 unsigned long elem_size;
874 p = fs_path_alloc_reversed();
878 tmp_path = alloc_path_for_send();
885 if (found_key->type == BTRFS_INODE_REF_KEY) {
886 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
887 struct btrfs_inode_ref);
888 item = btrfs_item_nr(slot);
889 total = btrfs_item_size(eb, item);
890 elem_size = sizeof(*iref);
892 ptr = btrfs_item_ptr_offset(eb, slot);
893 total = btrfs_item_size_nr(eb, slot);
894 elem_size = sizeof(*extref);
897 while (cur < total) {
900 if (found_key->type == BTRFS_INODE_REF_KEY) {
901 iref = (struct btrfs_inode_ref *)(ptr + cur);
902 name_len = btrfs_inode_ref_name_len(eb, iref);
903 name_off = (unsigned long)(iref + 1);
904 index = btrfs_inode_ref_index(eb, iref);
905 dir = found_key->offset;
907 extref = (struct btrfs_inode_extref *)(ptr + cur);
908 name_len = btrfs_inode_extref_name_len(eb, extref);
909 name_off = (unsigned long)&extref->name;
910 index = btrfs_inode_extref_index(eb, extref);
911 dir = btrfs_inode_extref_parent(eb, extref);
915 start = btrfs_ref_to_path(root, tmp_path, name_len,
919 ret = PTR_ERR(start);
922 if (start < p->buf) {
923 /* overflow , try again with larger buffer */
924 ret = fs_path_ensure_buf(p,
925 p->buf_len + p->buf - start);
928 start = btrfs_ref_to_path(root, tmp_path,
933 ret = PTR_ERR(start);
936 BUG_ON(start < p->buf);
940 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
946 cur += elem_size + name_len;
947 ret = iterate(num, dir, index, p, ctx);
954 btrfs_free_path(tmp_path);
959 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
960 const char *name, int name_len,
961 const char *data, int data_len,
965 * Helper function to iterate the entries in ONE btrfs_dir_item.
966 * The iterate callback may return a non zero value to stop iteration. This can
967 * be a negative value for error codes or 1 to simply stop it.
969 * path must point to the dir item when called.
971 static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path,
972 struct btrfs_key *found_key,
973 iterate_dir_item_t iterate, void *ctx)
976 struct extent_buffer *eb;
977 struct btrfs_item *item;
978 struct btrfs_dir_item *di;
979 struct btrfs_key di_key;
992 * Start with a small buffer (1 page). If later we end up needing more
993 * space, which can happen for xattrs on a fs with a leaf size greater
994 * then the page size, attempt to increase the buffer. Typically xattr
998 buf = kmalloc(buf_len, GFP_NOFS);
1004 eb = path->nodes[0];
1005 slot = path->slots[0];
1006 item = btrfs_item_nr(slot);
1007 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1010 total = btrfs_item_size(eb, item);
1013 while (cur < total) {
1014 name_len = btrfs_dir_name_len(eb, di);
1015 data_len = btrfs_dir_data_len(eb, di);
1016 type = btrfs_dir_type(eb, di);
1017 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1019 if (type == BTRFS_FT_XATTR) {
1020 if (name_len > XATTR_NAME_MAX) {
1021 ret = -ENAMETOOLONG;
1024 if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(root)) {
1032 if (name_len + data_len > PATH_MAX) {
1033 ret = -ENAMETOOLONG;
1038 if (name_len + data_len > buf_len) {
1039 buf_len = name_len + data_len;
1040 if (is_vmalloc_addr(buf)) {
1044 char *tmp = krealloc(buf, buf_len,
1045 GFP_NOFS | __GFP_NOWARN);
1052 buf = vmalloc(buf_len);
1060 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
1061 name_len + data_len);
1063 len = sizeof(*di) + name_len + data_len;
1064 di = (struct btrfs_dir_item *)((char *)di + len);
1067 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
1068 data_len, type, ctx);
1084 static int __copy_first_ref(int num, u64 dir, int index,
1085 struct fs_path *p, void *ctx)
1088 struct fs_path *pt = ctx;
1090 ret = fs_path_copy(pt, p);
1094 /* we want the first only */
1099 * Retrieve the first path of an inode. If an inode has more then one
1100 * ref/hardlink, this is ignored.
1102 static int get_inode_path(struct btrfs_root *root,
1103 u64 ino, struct fs_path *path)
1106 struct btrfs_key key, found_key;
1107 struct btrfs_path *p;
1109 p = alloc_path_for_send();
1113 fs_path_reset(path);
1116 key.type = BTRFS_INODE_REF_KEY;
1119 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1126 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1127 if (found_key.objectid != ino ||
1128 (found_key.type != BTRFS_INODE_REF_KEY &&
1129 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1134 ret = iterate_inode_ref(root, p, &found_key, 1,
1135 __copy_first_ref, path);
1145 struct backref_ctx {
1146 struct send_ctx *sctx;
1148 struct btrfs_path *path;
1149 /* number of total found references */
1153 * used for clones found in send_root. clones found behind cur_objectid
1154 * and cur_offset are not considered as allowed clones.
1159 /* may be truncated in case it's the last extent in a file */
1162 /* data offset in the file extent item */
1165 /* Just to check for bugs in backref resolving */
1169 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1171 u64 root = (u64)(uintptr_t)key;
1172 struct clone_root *cr = (struct clone_root *)elt;
1174 if (root < cr->root->objectid)
1176 if (root > cr->root->objectid)
1181 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1183 struct clone_root *cr1 = (struct clone_root *)e1;
1184 struct clone_root *cr2 = (struct clone_root *)e2;
1186 if (cr1->root->objectid < cr2->root->objectid)
1188 if (cr1->root->objectid > cr2->root->objectid)
1194 * Called for every backref that is found for the current extent.
1195 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1197 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1199 struct backref_ctx *bctx = ctx_;
1200 struct clone_root *found;
1204 /* First check if the root is in the list of accepted clone sources */
1205 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1206 bctx->sctx->clone_roots_cnt,
1207 sizeof(struct clone_root),
1208 __clone_root_cmp_bsearch);
1212 if (found->root == bctx->sctx->send_root &&
1213 ino == bctx->cur_objectid &&
1214 offset == bctx->cur_offset) {
1215 bctx->found_itself = 1;
1219 * There are inodes that have extents that lie behind its i_size. Don't
1220 * accept clones from these extents.
1222 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1224 btrfs_release_path(bctx->path);
1228 if (offset + bctx->data_offset + bctx->extent_len > i_size)
1232 * Make sure we don't consider clones from send_root that are
1233 * behind the current inode/offset.
1235 if (found->root == bctx->sctx->send_root) {
1237 * TODO for the moment we don't accept clones from the inode
1238 * that is currently send. We may change this when
1239 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1242 if (ino >= bctx->cur_objectid)
1245 if (ino > bctx->cur_objectid)
1247 if (offset + bctx->extent_len > bctx->cur_offset)
1253 found->found_refs++;
1254 if (ino < found->ino) {
1256 found->offset = offset;
1257 } else if (found->ino == ino) {
1259 * same extent found more then once in the same file.
1261 if (found->offset > offset + bctx->extent_len)
1262 found->offset = offset;
1269 * Given an inode, offset and extent item, it finds a good clone for a clone
1270 * instruction. Returns -ENOENT when none could be found. The function makes
1271 * sure that the returned clone is usable at the point where sending is at the
1272 * moment. This means, that no clones are accepted which lie behind the current
1275 * path must point to the extent item when called.
1277 static int find_extent_clone(struct send_ctx *sctx,
1278 struct btrfs_path *path,
1279 u64 ino, u64 data_offset,
1281 struct clone_root **found)
1288 u64 extent_item_pos;
1290 struct btrfs_file_extent_item *fi;
1291 struct extent_buffer *eb = path->nodes[0];
1292 struct backref_ctx *backref_ctx = NULL;
1293 struct clone_root *cur_clone_root;
1294 struct btrfs_key found_key;
1295 struct btrfs_path *tmp_path;
1299 tmp_path = alloc_path_for_send();
1303 /* We only use this path under the commit sem */
1304 tmp_path->need_commit_sem = 0;
1306 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1312 backref_ctx->path = tmp_path;
1314 if (data_offset >= ino_size) {
1316 * There may be extents that lie behind the file's size.
1317 * I at least had this in combination with snapshotting while
1318 * writing large files.
1324 fi = btrfs_item_ptr(eb, path->slots[0],
1325 struct btrfs_file_extent_item);
1326 extent_type = btrfs_file_extent_type(eb, fi);
1327 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1331 compressed = btrfs_file_extent_compression(eb, fi);
1333 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1334 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1335 if (disk_byte == 0) {
1339 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1341 down_read(&sctx->send_root->fs_info->commit_root_sem);
1342 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1343 &found_key, &flags);
1344 up_read(&sctx->send_root->fs_info->commit_root_sem);
1345 btrfs_release_path(tmp_path);
1349 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1355 * Setup the clone roots.
1357 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1358 cur_clone_root = sctx->clone_roots + i;
1359 cur_clone_root->ino = (u64)-1;
1360 cur_clone_root->offset = 0;
1361 cur_clone_root->found_refs = 0;
1364 backref_ctx->sctx = sctx;
1365 backref_ctx->found = 0;
1366 backref_ctx->cur_objectid = ino;
1367 backref_ctx->cur_offset = data_offset;
1368 backref_ctx->found_itself = 0;
1369 backref_ctx->extent_len = num_bytes;
1371 * For non-compressed extents iterate_extent_inodes() gives us extent
1372 * offsets that already take into account the data offset, but not for
1373 * compressed extents, since the offset is logical and not relative to
1374 * the physical extent locations. We must take this into account to
1375 * avoid sending clone offsets that go beyond the source file's size,
1376 * which would result in the clone ioctl failing with -EINVAL on the
1379 if (compressed == BTRFS_COMPRESS_NONE)
1380 backref_ctx->data_offset = 0;
1382 backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi);
1385 * The last extent of a file may be too large due to page alignment.
1386 * We need to adjust extent_len in this case so that the checks in
1387 * __iterate_backrefs work.
1389 if (data_offset + num_bytes >= ino_size)
1390 backref_ctx->extent_len = ino_size - data_offset;
1393 * Now collect all backrefs.
1395 if (compressed == BTRFS_COMPRESS_NONE)
1396 extent_item_pos = logical - found_key.objectid;
1398 extent_item_pos = 0;
1399 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1400 found_key.objectid, extent_item_pos, 1,
1401 __iterate_backrefs, backref_ctx);
1406 if (!backref_ctx->found_itself) {
1407 /* found a bug in backref code? */
1409 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1410 "send_root. inode=%llu, offset=%llu, "
1411 "disk_byte=%llu found extent=%llu",
1412 ino, data_offset, disk_byte, found_key.objectid);
1416 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1418 "num_bytes=%llu, logical=%llu\n",
1419 data_offset, ino, num_bytes, logical);
1421 if (!backref_ctx->found)
1422 verbose_printk("btrfs: no clones found\n");
1424 cur_clone_root = NULL;
1425 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1426 if (sctx->clone_roots[i].found_refs) {
1427 if (!cur_clone_root)
1428 cur_clone_root = sctx->clone_roots + i;
1429 else if (sctx->clone_roots[i].root == sctx->send_root)
1430 /* prefer clones from send_root over others */
1431 cur_clone_root = sctx->clone_roots + i;
1436 if (cur_clone_root) {
1437 if (compressed != BTRFS_COMPRESS_NONE) {
1439 * Offsets given by iterate_extent_inodes() are relative
1440 * to the start of the extent, we need to add logical
1441 * offset from the file extent item.
1442 * (See why at backref.c:check_extent_in_eb())
1444 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1447 *found = cur_clone_root;
1454 btrfs_free_path(tmp_path);
1459 static int read_symlink(struct btrfs_root *root,
1461 struct fs_path *dest)
1464 struct btrfs_path *path;
1465 struct btrfs_key key;
1466 struct btrfs_file_extent_item *ei;
1472 path = alloc_path_for_send();
1477 key.type = BTRFS_EXTENT_DATA_KEY;
1479 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1484 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1485 struct btrfs_file_extent_item);
1486 type = btrfs_file_extent_type(path->nodes[0], ei);
1487 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1488 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1489 BUG_ON(compression);
1491 off = btrfs_file_extent_inline_start(ei);
1492 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1494 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1497 btrfs_free_path(path);
1502 * Helper function to generate a file name that is unique in the root of
1503 * send_root and parent_root. This is used to generate names for orphan inodes.
1505 static int gen_unique_name(struct send_ctx *sctx,
1507 struct fs_path *dest)
1510 struct btrfs_path *path;
1511 struct btrfs_dir_item *di;
1516 path = alloc_path_for_send();
1521 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1523 ASSERT(len < sizeof(tmp));
1525 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1526 path, BTRFS_FIRST_FREE_OBJECTID,
1527 tmp, strlen(tmp), 0);
1528 btrfs_release_path(path);
1534 /* not unique, try again */
1539 if (!sctx->parent_root) {
1545 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1546 path, BTRFS_FIRST_FREE_OBJECTID,
1547 tmp, strlen(tmp), 0);
1548 btrfs_release_path(path);
1554 /* not unique, try again */
1562 ret = fs_path_add(dest, tmp, strlen(tmp));
1565 btrfs_free_path(path);
1570 inode_state_no_change,
1571 inode_state_will_create,
1572 inode_state_did_create,
1573 inode_state_will_delete,
1574 inode_state_did_delete,
1577 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1585 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1587 if (ret < 0 && ret != -ENOENT)
1591 if (!sctx->parent_root) {
1592 right_ret = -ENOENT;
1594 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1595 NULL, NULL, NULL, NULL);
1596 if (ret < 0 && ret != -ENOENT)
1601 if (!left_ret && !right_ret) {
1602 if (left_gen == gen && right_gen == gen) {
1603 ret = inode_state_no_change;
1604 } else if (left_gen == gen) {
1605 if (ino < sctx->send_progress)
1606 ret = inode_state_did_create;
1608 ret = inode_state_will_create;
1609 } else if (right_gen == gen) {
1610 if (ino < sctx->send_progress)
1611 ret = inode_state_did_delete;
1613 ret = inode_state_will_delete;
1617 } else if (!left_ret) {
1618 if (left_gen == gen) {
1619 if (ino < sctx->send_progress)
1620 ret = inode_state_did_create;
1622 ret = inode_state_will_create;
1626 } else if (!right_ret) {
1627 if (right_gen == gen) {
1628 if (ino < sctx->send_progress)
1629 ret = inode_state_did_delete;
1631 ret = inode_state_will_delete;
1643 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1647 ret = get_cur_inode_state(sctx, ino, gen);
1651 if (ret == inode_state_no_change ||
1652 ret == inode_state_did_create ||
1653 ret == inode_state_will_delete)
1663 * Helper function to lookup a dir item in a dir.
1665 static int lookup_dir_item_inode(struct btrfs_root *root,
1666 u64 dir, const char *name, int name_len,
1671 struct btrfs_dir_item *di;
1672 struct btrfs_key key;
1673 struct btrfs_path *path;
1675 path = alloc_path_for_send();
1679 di = btrfs_lookup_dir_item(NULL, root, path,
1680 dir, name, name_len, 0);
1689 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1690 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1694 *found_inode = key.objectid;
1695 *found_type = btrfs_dir_type(path->nodes[0], di);
1698 btrfs_free_path(path);
1703 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1704 * generation of the parent dir and the name of the dir entry.
1706 static int get_first_ref(struct btrfs_root *root, u64 ino,
1707 u64 *dir, u64 *dir_gen, struct fs_path *name)
1710 struct btrfs_key key;
1711 struct btrfs_key found_key;
1712 struct btrfs_path *path;
1716 path = alloc_path_for_send();
1721 key.type = BTRFS_INODE_REF_KEY;
1724 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1728 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1730 if (ret || found_key.objectid != ino ||
1731 (found_key.type != BTRFS_INODE_REF_KEY &&
1732 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1737 if (found_key.type == BTRFS_INODE_REF_KEY) {
1738 struct btrfs_inode_ref *iref;
1739 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1740 struct btrfs_inode_ref);
1741 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1742 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1743 (unsigned long)(iref + 1),
1745 parent_dir = found_key.offset;
1747 struct btrfs_inode_extref *extref;
1748 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1749 struct btrfs_inode_extref);
1750 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1751 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1752 (unsigned long)&extref->name, len);
1753 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1757 btrfs_release_path(path);
1760 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1769 btrfs_free_path(path);
1773 static int is_first_ref(struct btrfs_root *root,
1775 const char *name, int name_len)
1778 struct fs_path *tmp_name;
1781 tmp_name = fs_path_alloc();
1785 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1789 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1794 ret = !memcmp(tmp_name->start, name, name_len);
1797 fs_path_free(tmp_name);
1802 * Used by process_recorded_refs to determine if a new ref would overwrite an
1803 * already existing ref. In case it detects an overwrite, it returns the
1804 * inode/gen in who_ino/who_gen.
1805 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1806 * to make sure later references to the overwritten inode are possible.
1807 * Orphanizing is however only required for the first ref of an inode.
1808 * process_recorded_refs does an additional is_first_ref check to see if
1809 * orphanizing is really required.
1811 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1812 const char *name, int name_len,
1813 u64 *who_ino, u64 *who_gen)
1817 u64 other_inode = 0;
1820 if (!sctx->parent_root)
1823 ret = is_inode_existent(sctx, dir, dir_gen);
1828 * If we have a parent root we need to verify that the parent dir was
1829 * not delted and then re-created, if it was then we have no overwrite
1830 * and we can just unlink this entry.
1832 if (sctx->parent_root) {
1833 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1835 if (ret < 0 && ret != -ENOENT)
1845 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1846 &other_inode, &other_type);
1847 if (ret < 0 && ret != -ENOENT)
1855 * Check if the overwritten ref was already processed. If yes, the ref
1856 * was already unlinked/moved, so we can safely assume that we will not
1857 * overwrite anything at this point in time.
1859 if (other_inode > sctx->send_progress) {
1860 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1861 who_gen, NULL, NULL, NULL, NULL);
1866 *who_ino = other_inode;
1876 * Checks if the ref was overwritten by an already processed inode. This is
1877 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1878 * thus the orphan name needs be used.
1879 * process_recorded_refs also uses it to avoid unlinking of refs that were
1882 static int did_overwrite_ref(struct send_ctx *sctx,
1883 u64 dir, u64 dir_gen,
1884 u64 ino, u64 ino_gen,
1885 const char *name, int name_len)
1892 if (!sctx->parent_root)
1895 ret = is_inode_existent(sctx, dir, dir_gen);
1899 /* check if the ref was overwritten by another ref */
1900 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1901 &ow_inode, &other_type);
1902 if (ret < 0 && ret != -ENOENT)
1905 /* was never and will never be overwritten */
1910 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1915 if (ow_inode == ino && gen == ino_gen) {
1921 * We know that it is or will be overwritten. Check this now.
1922 * The current inode being processed might have been the one that caused
1923 * inode 'ino' to be orphanized, therefore check if ow_inode matches
1924 * the current inode being processed.
1926 if ((ow_inode < sctx->send_progress) ||
1927 (ino != sctx->cur_ino && ow_inode == sctx->cur_ino &&
1928 gen == sctx->cur_inode_gen))
1938 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1939 * that got overwritten. This is used by process_recorded_refs to determine
1940 * if it has to use the path as returned by get_cur_path or the orphan name.
1942 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1945 struct fs_path *name = NULL;
1949 if (!sctx->parent_root)
1952 name = fs_path_alloc();
1956 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1960 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1961 name->start, fs_path_len(name));
1969 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1970 * so we need to do some special handling in case we have clashes. This function
1971 * takes care of this with the help of name_cache_entry::radix_list.
1972 * In case of error, nce is kfreed.
1974 static int name_cache_insert(struct send_ctx *sctx,
1975 struct name_cache_entry *nce)
1978 struct list_head *nce_head;
1980 nce_head = radix_tree_lookup(&sctx->name_cache,
1981 (unsigned long)nce->ino);
1983 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1988 INIT_LIST_HEAD(nce_head);
1990 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1997 list_add_tail(&nce->radix_list, nce_head);
1998 list_add_tail(&nce->list, &sctx->name_cache_list);
1999 sctx->name_cache_size++;
2004 static void name_cache_delete(struct send_ctx *sctx,
2005 struct name_cache_entry *nce)
2007 struct list_head *nce_head;
2009 nce_head = radix_tree_lookup(&sctx->name_cache,
2010 (unsigned long)nce->ino);
2012 btrfs_err(sctx->send_root->fs_info,
2013 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
2014 nce->ino, sctx->name_cache_size);
2017 list_del(&nce->radix_list);
2018 list_del(&nce->list);
2019 sctx->name_cache_size--;
2022 * We may not get to the final release of nce_head if the lookup fails
2024 if (nce_head && list_empty(nce_head)) {
2025 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2030 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2033 struct list_head *nce_head;
2034 struct name_cache_entry *cur;
2036 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2040 list_for_each_entry(cur, nce_head, radix_list) {
2041 if (cur->ino == ino && cur->gen == gen)
2048 * Removes the entry from the list and adds it back to the end. This marks the
2049 * entry as recently used so that name_cache_clean_unused does not remove it.
2051 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2053 list_del(&nce->list);
2054 list_add_tail(&nce->list, &sctx->name_cache_list);
2058 * Remove some entries from the beginning of name_cache_list.
2060 static void name_cache_clean_unused(struct send_ctx *sctx)
2062 struct name_cache_entry *nce;
2064 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2067 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2068 nce = list_entry(sctx->name_cache_list.next,
2069 struct name_cache_entry, list);
2070 name_cache_delete(sctx, nce);
2075 static void name_cache_free(struct send_ctx *sctx)
2077 struct name_cache_entry *nce;
2079 while (!list_empty(&sctx->name_cache_list)) {
2080 nce = list_entry(sctx->name_cache_list.next,
2081 struct name_cache_entry, list);
2082 name_cache_delete(sctx, nce);
2088 * Used by get_cur_path for each ref up to the root.
2089 * Returns 0 if it succeeded.
2090 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2091 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2092 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2093 * Returns <0 in case of error.
2095 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2099 struct fs_path *dest)
2103 struct name_cache_entry *nce = NULL;
2106 * First check if we already did a call to this function with the same
2107 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2108 * return the cached result.
2110 nce = name_cache_search(sctx, ino, gen);
2112 if (ino < sctx->send_progress && nce->need_later_update) {
2113 name_cache_delete(sctx, nce);
2117 name_cache_used(sctx, nce);
2118 *parent_ino = nce->parent_ino;
2119 *parent_gen = nce->parent_gen;
2120 ret = fs_path_add(dest, nce->name, nce->name_len);
2129 * If the inode is not existent yet, add the orphan name and return 1.
2130 * This should only happen for the parent dir that we determine in
2133 ret = is_inode_existent(sctx, ino, gen);
2138 ret = gen_unique_name(sctx, ino, gen, dest);
2146 * Depending on whether the inode was already processed or not, use
2147 * send_root or parent_root for ref lookup.
2149 if (ino < sctx->send_progress)
2150 ret = get_first_ref(sctx->send_root, ino,
2151 parent_ino, parent_gen, dest);
2153 ret = get_first_ref(sctx->parent_root, ino,
2154 parent_ino, parent_gen, dest);
2159 * Check if the ref was overwritten by an inode's ref that was processed
2160 * earlier. If yes, treat as orphan and return 1.
2162 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2163 dest->start, dest->end - dest->start);
2167 fs_path_reset(dest);
2168 ret = gen_unique_name(sctx, ino, gen, dest);
2176 * Store the result of the lookup in the name cache.
2178 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2186 nce->parent_ino = *parent_ino;
2187 nce->parent_gen = *parent_gen;
2188 nce->name_len = fs_path_len(dest);
2190 strcpy(nce->name, dest->start);
2192 if (ino < sctx->send_progress)
2193 nce->need_later_update = 0;
2195 nce->need_later_update = 1;
2197 nce_ret = name_cache_insert(sctx, nce);
2200 name_cache_clean_unused(sctx);
2207 * Magic happens here. This function returns the first ref to an inode as it
2208 * would look like while receiving the stream at this point in time.
2209 * We walk the path up to the root. For every inode in between, we check if it
2210 * was already processed/sent. If yes, we continue with the parent as found
2211 * in send_root. If not, we continue with the parent as found in parent_root.
2212 * If we encounter an inode that was deleted at this point in time, we use the
2213 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2214 * that were not created yet and overwritten inodes/refs.
2216 * When do we have have orphan inodes:
2217 * 1. When an inode is freshly created and thus no valid refs are available yet
2218 * 2. When a directory lost all it's refs (deleted) but still has dir items
2219 * inside which were not processed yet (pending for move/delete). If anyone
2220 * tried to get the path to the dir items, it would get a path inside that
2222 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2223 * of an unprocessed inode. If in that case the first ref would be
2224 * overwritten, the overwritten inode gets "orphanized". Later when we
2225 * process this overwritten inode, it is restored at a new place by moving
2228 * sctx->send_progress tells this function at which point in time receiving
2231 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2232 struct fs_path *dest)
2235 struct fs_path *name = NULL;
2236 u64 parent_inode = 0;
2240 name = fs_path_alloc();
2247 fs_path_reset(dest);
2249 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2250 struct waiting_dir_move *wdm;
2252 fs_path_reset(name);
2254 if (is_waiting_for_rm(sctx, ino)) {
2255 ret = gen_unique_name(sctx, ino, gen, name);
2258 ret = fs_path_add_path(dest, name);
2262 wdm = get_waiting_dir_move(sctx, ino);
2263 if (wdm && wdm->orphanized) {
2264 ret = gen_unique_name(sctx, ino, gen, name);
2267 ret = get_first_ref(sctx->parent_root, ino,
2268 &parent_inode, &parent_gen, name);
2270 ret = __get_cur_name_and_parent(sctx, ino, gen,
2280 ret = fs_path_add_path(dest, name);
2291 fs_path_unreverse(dest);
2296 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2298 static int send_subvol_begin(struct send_ctx *sctx)
2301 struct btrfs_root *send_root = sctx->send_root;
2302 struct btrfs_root *parent_root = sctx->parent_root;
2303 struct btrfs_path *path;
2304 struct btrfs_key key;
2305 struct btrfs_root_ref *ref;
2306 struct extent_buffer *leaf;
2310 path = btrfs_alloc_path();
2314 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2316 btrfs_free_path(path);
2320 key.objectid = send_root->objectid;
2321 key.type = BTRFS_ROOT_BACKREF_KEY;
2324 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2333 leaf = path->nodes[0];
2334 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2335 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2336 key.objectid != send_root->objectid) {
2340 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2341 namelen = btrfs_root_ref_name_len(leaf, ref);
2342 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2343 btrfs_release_path(path);
2346 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2350 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2355 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2356 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2357 sctx->send_root->root_item.uuid);
2358 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2359 le64_to_cpu(sctx->send_root->root_item.ctransid));
2361 if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid))
2362 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2363 parent_root->root_item.received_uuid);
2365 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2366 parent_root->root_item.uuid);
2367 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2368 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2371 ret = send_cmd(sctx);
2375 btrfs_free_path(path);
2380 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2385 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2387 p = fs_path_alloc();
2391 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2395 ret = get_cur_path(sctx, ino, gen, p);
2398 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2399 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2401 ret = send_cmd(sctx);
2409 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2414 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2416 p = fs_path_alloc();
2420 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2424 ret = get_cur_path(sctx, ino, gen, p);
2427 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2428 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2430 ret = send_cmd(sctx);
2438 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2443 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2445 p = fs_path_alloc();
2449 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2453 ret = get_cur_path(sctx, ino, gen, p);
2456 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2457 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2458 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2460 ret = send_cmd(sctx);
2468 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2471 struct fs_path *p = NULL;
2472 struct btrfs_inode_item *ii;
2473 struct btrfs_path *path = NULL;
2474 struct extent_buffer *eb;
2475 struct btrfs_key key;
2478 verbose_printk("btrfs: send_utimes %llu\n", ino);
2480 p = fs_path_alloc();
2484 path = alloc_path_for_send();
2491 key.type = BTRFS_INODE_ITEM_KEY;
2493 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2497 eb = path->nodes[0];
2498 slot = path->slots[0];
2499 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2501 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2505 ret = get_cur_path(sctx, ino, gen, p);
2508 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2509 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2510 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2511 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2512 /* TODO Add otime support when the otime patches get into upstream */
2514 ret = send_cmd(sctx);
2519 btrfs_free_path(path);
2524 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2525 * a valid path yet because we did not process the refs yet. So, the inode
2526 * is created as orphan.
2528 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2537 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2539 p = fs_path_alloc();
2543 if (ino != sctx->cur_ino) {
2544 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2549 gen = sctx->cur_inode_gen;
2550 mode = sctx->cur_inode_mode;
2551 rdev = sctx->cur_inode_rdev;
2554 if (S_ISREG(mode)) {
2555 cmd = BTRFS_SEND_C_MKFILE;
2556 } else if (S_ISDIR(mode)) {
2557 cmd = BTRFS_SEND_C_MKDIR;
2558 } else if (S_ISLNK(mode)) {
2559 cmd = BTRFS_SEND_C_SYMLINK;
2560 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2561 cmd = BTRFS_SEND_C_MKNOD;
2562 } else if (S_ISFIFO(mode)) {
2563 cmd = BTRFS_SEND_C_MKFIFO;
2564 } else if (S_ISSOCK(mode)) {
2565 cmd = BTRFS_SEND_C_MKSOCK;
2567 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2568 (int)(mode & S_IFMT));
2573 ret = begin_cmd(sctx, cmd);
2577 ret = gen_unique_name(sctx, ino, gen, p);
2581 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2582 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2584 if (S_ISLNK(mode)) {
2586 ret = read_symlink(sctx->send_root, ino, p);
2589 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2590 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2591 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2592 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2593 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2596 ret = send_cmd(sctx);
2608 * We need some special handling for inodes that get processed before the parent
2609 * directory got created. See process_recorded_refs for details.
2610 * This function does the check if we already created the dir out of order.
2612 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2615 struct btrfs_path *path = NULL;
2616 struct btrfs_key key;
2617 struct btrfs_key found_key;
2618 struct btrfs_key di_key;
2619 struct extent_buffer *eb;
2620 struct btrfs_dir_item *di;
2623 path = alloc_path_for_send();
2630 key.type = BTRFS_DIR_INDEX_KEY;
2632 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2637 eb = path->nodes[0];
2638 slot = path->slots[0];
2639 if (slot >= btrfs_header_nritems(eb)) {
2640 ret = btrfs_next_leaf(sctx->send_root, path);
2643 } else if (ret > 0) {
2650 btrfs_item_key_to_cpu(eb, &found_key, slot);
2651 if (found_key.objectid != key.objectid ||
2652 found_key.type != key.type) {
2657 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2658 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2660 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2661 di_key.objectid < sctx->send_progress) {
2670 btrfs_free_path(path);
2675 * Only creates the inode if it is:
2676 * 1. Not a directory
2677 * 2. Or a directory which was not created already due to out of order
2678 * directories. See did_create_dir and process_recorded_refs for details.
2680 static int send_create_inode_if_needed(struct send_ctx *sctx)
2684 if (S_ISDIR(sctx->cur_inode_mode)) {
2685 ret = did_create_dir(sctx, sctx->cur_ino);
2694 ret = send_create_inode(sctx, sctx->cur_ino);
2702 struct recorded_ref {
2703 struct list_head list;
2706 struct fs_path *full_path;
2714 * We need to process new refs before deleted refs, but compare_tree gives us
2715 * everything mixed. So we first record all refs and later process them.
2716 * This function is a helper to record one ref.
2718 static int __record_ref(struct list_head *head, u64 dir,
2719 u64 dir_gen, struct fs_path *path)
2721 struct recorded_ref *ref;
2723 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2728 ref->dir_gen = dir_gen;
2729 ref->full_path = path;
2731 ref->name = (char *)kbasename(ref->full_path->start);
2732 ref->name_len = ref->full_path->end - ref->name;
2733 ref->dir_path = ref->full_path->start;
2734 if (ref->name == ref->full_path->start)
2735 ref->dir_path_len = 0;
2737 ref->dir_path_len = ref->full_path->end -
2738 ref->full_path->start - 1 - ref->name_len;
2740 list_add_tail(&ref->list, head);
2744 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2746 struct recorded_ref *new;
2748 new = kmalloc(sizeof(*ref), GFP_NOFS);
2752 new->dir = ref->dir;
2753 new->dir_gen = ref->dir_gen;
2754 new->full_path = NULL;
2755 INIT_LIST_HEAD(&new->list);
2756 list_add_tail(&new->list, list);
2760 static void __free_recorded_refs(struct list_head *head)
2762 struct recorded_ref *cur;
2764 while (!list_empty(head)) {
2765 cur = list_entry(head->next, struct recorded_ref, list);
2766 fs_path_free(cur->full_path);
2767 list_del(&cur->list);
2772 static void free_recorded_refs(struct send_ctx *sctx)
2774 __free_recorded_refs(&sctx->new_refs);
2775 __free_recorded_refs(&sctx->deleted_refs);
2779 * Renames/moves a file/dir to its orphan name. Used when the first
2780 * ref of an unprocessed inode gets overwritten and for all non empty
2783 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2784 struct fs_path *path)
2787 struct fs_path *orphan;
2789 orphan = fs_path_alloc();
2793 ret = gen_unique_name(sctx, ino, gen, orphan);
2797 ret = send_rename(sctx, path, orphan);
2800 fs_path_free(orphan);
2804 static struct orphan_dir_info *
2805 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2807 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2808 struct rb_node *parent = NULL;
2809 struct orphan_dir_info *entry, *odi;
2811 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2813 return ERR_PTR(-ENOMEM);
2819 entry = rb_entry(parent, struct orphan_dir_info, node);
2820 if (dir_ino < entry->ino) {
2822 } else if (dir_ino > entry->ino) {
2823 p = &(*p)->rb_right;
2830 rb_link_node(&odi->node, parent, p);
2831 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2835 static struct orphan_dir_info *
2836 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2838 struct rb_node *n = sctx->orphan_dirs.rb_node;
2839 struct orphan_dir_info *entry;
2842 entry = rb_entry(n, struct orphan_dir_info, node);
2843 if (dir_ino < entry->ino)
2845 else if (dir_ino > entry->ino)
2853 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2855 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2860 static void free_orphan_dir_info(struct send_ctx *sctx,
2861 struct orphan_dir_info *odi)
2865 rb_erase(&odi->node, &sctx->orphan_dirs);
2870 * Returns 1 if a directory can be removed at this point in time.
2871 * We check this by iterating all dir items and checking if the inode behind
2872 * the dir item was already processed.
2874 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2878 struct btrfs_root *root = sctx->parent_root;
2879 struct btrfs_path *path;
2880 struct btrfs_key key;
2881 struct btrfs_key found_key;
2882 struct btrfs_key loc;
2883 struct btrfs_dir_item *di;
2886 * Don't try to rmdir the top/root subvolume dir.
2888 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2891 path = alloc_path_for_send();
2896 key.type = BTRFS_DIR_INDEX_KEY;
2898 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2903 struct waiting_dir_move *dm;
2905 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2906 ret = btrfs_next_leaf(root, path);
2913 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2915 if (found_key.objectid != key.objectid ||
2916 found_key.type != key.type)
2919 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2920 struct btrfs_dir_item);
2921 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2923 dm = get_waiting_dir_move(sctx, loc.objectid);
2925 struct orphan_dir_info *odi;
2927 odi = add_orphan_dir_info(sctx, dir);
2933 dm->rmdir_ino = dir;
2938 if (loc.objectid > send_progress) {
2949 btrfs_free_path(path);
2953 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2955 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2957 return entry != NULL;
2960 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
2962 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2963 struct rb_node *parent = NULL;
2964 struct waiting_dir_move *entry, *dm;
2966 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2971 dm->orphanized = orphanized;
2975 entry = rb_entry(parent, struct waiting_dir_move, node);
2976 if (ino < entry->ino) {
2978 } else if (ino > entry->ino) {
2979 p = &(*p)->rb_right;
2986 rb_link_node(&dm->node, parent, p);
2987 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2991 static struct waiting_dir_move *
2992 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2994 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2995 struct waiting_dir_move *entry;
2998 entry = rb_entry(n, struct waiting_dir_move, node);
2999 if (ino < entry->ino)
3001 else if (ino > entry->ino)
3009 static void free_waiting_dir_move(struct send_ctx *sctx,
3010 struct waiting_dir_move *dm)
3014 rb_erase(&dm->node, &sctx->waiting_dir_moves);
3018 static int add_pending_dir_move(struct send_ctx *sctx,
3022 struct list_head *new_refs,
3023 struct list_head *deleted_refs,
3024 const bool is_orphan)
3026 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3027 struct rb_node *parent = NULL;
3028 struct pending_dir_move *entry = NULL, *pm;
3029 struct recorded_ref *cur;
3033 pm = kmalloc(sizeof(*pm), GFP_NOFS);
3036 pm->parent_ino = parent_ino;
3039 pm->is_orphan = is_orphan;
3040 INIT_LIST_HEAD(&pm->list);
3041 INIT_LIST_HEAD(&pm->update_refs);
3042 RB_CLEAR_NODE(&pm->node);
3046 entry = rb_entry(parent, struct pending_dir_move, node);
3047 if (parent_ino < entry->parent_ino) {
3049 } else if (parent_ino > entry->parent_ino) {
3050 p = &(*p)->rb_right;
3057 list_for_each_entry(cur, deleted_refs, list) {
3058 ret = dup_ref(cur, &pm->update_refs);
3062 list_for_each_entry(cur, new_refs, list) {
3063 ret = dup_ref(cur, &pm->update_refs);
3068 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3073 list_add_tail(&pm->list, &entry->list);
3075 rb_link_node(&pm->node, parent, p);
3076 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3081 __free_recorded_refs(&pm->update_refs);
3087 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3090 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3091 struct pending_dir_move *entry;
3094 entry = rb_entry(n, struct pending_dir_move, node);
3095 if (parent_ino < entry->parent_ino)
3097 else if (parent_ino > entry->parent_ino)
3105 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3107 struct fs_path *from_path = NULL;
3108 struct fs_path *to_path = NULL;
3109 struct fs_path *name = NULL;
3110 u64 orig_progress = sctx->send_progress;
3111 struct recorded_ref *cur;
3112 u64 parent_ino, parent_gen;
3113 struct waiting_dir_move *dm = NULL;
3117 name = fs_path_alloc();
3118 from_path = fs_path_alloc();
3119 if (!name || !from_path) {
3124 dm = get_waiting_dir_move(sctx, pm->ino);
3126 rmdir_ino = dm->rmdir_ino;
3127 free_waiting_dir_move(sctx, dm);
3129 if (pm->is_orphan) {
3130 ret = gen_unique_name(sctx, pm->ino,
3131 pm->gen, from_path);
3133 ret = get_first_ref(sctx->parent_root, pm->ino,
3134 &parent_ino, &parent_gen, name);
3137 ret = get_cur_path(sctx, parent_ino, parent_gen,
3141 ret = fs_path_add_path(from_path, name);
3146 sctx->send_progress = sctx->cur_ino + 1;
3147 fs_path_reset(name);
3150 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3154 ret = send_rename(sctx, from_path, to_path);
3159 struct orphan_dir_info *odi;
3161 odi = get_orphan_dir_info(sctx, rmdir_ino);
3163 /* already deleted */
3166 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3172 name = fs_path_alloc();
3177 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3180 ret = send_rmdir(sctx, name);
3183 free_orphan_dir_info(sctx, odi);
3187 ret = send_utimes(sctx, pm->ino, pm->gen);
3192 * After rename/move, need to update the utimes of both new parent(s)
3193 * and old parent(s).
3195 list_for_each_entry(cur, &pm->update_refs, list) {
3196 if (cur->dir == rmdir_ino)
3198 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3205 fs_path_free(from_path);
3206 fs_path_free(to_path);
3207 sctx->send_progress = orig_progress;
3212 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3214 if (!list_empty(&m->list))
3216 if (!RB_EMPTY_NODE(&m->node))
3217 rb_erase(&m->node, &sctx->pending_dir_moves);
3218 __free_recorded_refs(&m->update_refs);
3222 static void tail_append_pending_moves(struct pending_dir_move *moves,
3223 struct list_head *stack)
3225 if (list_empty(&moves->list)) {
3226 list_add_tail(&moves->list, stack);
3229 list_splice_init(&moves->list, &list);
3230 list_add_tail(&moves->list, stack);
3231 list_splice_tail(&list, stack);
3235 static int apply_children_dir_moves(struct send_ctx *sctx)
3237 struct pending_dir_move *pm;
3238 struct list_head stack;
3239 u64 parent_ino = sctx->cur_ino;
3242 pm = get_pending_dir_moves(sctx, parent_ino);
3246 INIT_LIST_HEAD(&stack);
3247 tail_append_pending_moves(pm, &stack);
3249 while (!list_empty(&stack)) {
3250 pm = list_first_entry(&stack, struct pending_dir_move, list);
3251 parent_ino = pm->ino;
3252 ret = apply_dir_move(sctx, pm);
3253 free_pending_move(sctx, pm);
3256 pm = get_pending_dir_moves(sctx, parent_ino);
3258 tail_append_pending_moves(pm, &stack);
3263 while (!list_empty(&stack)) {
3264 pm = list_first_entry(&stack, struct pending_dir_move, list);
3265 free_pending_move(sctx, pm);
3271 * We might need to delay a directory rename even when no ancestor directory
3272 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3273 * renamed. This happens when we rename a directory to the old name (the name
3274 * in the parent root) of some other unrelated directory that got its rename
3275 * delayed due to some ancestor with higher number that got renamed.
3281 * |---- a/ (ino 257)
3282 * | |---- file (ino 260)
3284 * |---- b/ (ino 258)
3285 * |---- c/ (ino 259)
3289 * |---- a/ (ino 258)
3290 * |---- x/ (ino 259)
3291 * |---- y/ (ino 257)
3292 * |----- file (ino 260)
3294 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3295 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3296 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3299 * 1 - rename 259 from 'c' to 'x'
3300 * 2 - rename 257 from 'a' to 'x/y'
3301 * 3 - rename 258 from 'b' to 'a'
3303 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3304 * be done right away and < 0 on error.
3306 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3307 struct recorded_ref *parent_ref,
3308 const bool is_orphan)
3310 struct btrfs_path *path;
3311 struct btrfs_key key;
3312 struct btrfs_key di_key;
3313 struct btrfs_dir_item *di;
3318 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3321 path = alloc_path_for_send();
3325 key.objectid = parent_ref->dir;
3326 key.type = BTRFS_DIR_ITEM_KEY;
3327 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3329 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3332 } else if (ret > 0) {
3337 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3338 parent_ref->name, parent_ref->name_len);
3344 * di_key.objectid has the number of the inode that has a dentry in the
3345 * parent directory with the same name that sctx->cur_ino is being
3346 * renamed to. We need to check if that inode is in the send root as
3347 * well and if it is currently marked as an inode with a pending rename,
3348 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3349 * that it happens after that other inode is renamed.
3351 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3352 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3357 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3358 &left_gen, NULL, NULL, NULL, NULL);
3361 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3362 &right_gen, NULL, NULL, NULL, NULL);
3369 /* Different inode, no need to delay the rename of sctx->cur_ino */
3370 if (right_gen != left_gen) {
3375 if (is_waiting_for_move(sctx, di_key.objectid)) {
3376 ret = add_pending_dir_move(sctx,
3378 sctx->cur_inode_gen,
3381 &sctx->deleted_refs,
3387 btrfs_free_path(path);
3392 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3393 * Return 1 if true, 0 if false and < 0 on error.
3395 static int is_ancestor(struct btrfs_root *root,
3399 struct fs_path *fs_path)
3403 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3408 fs_path_reset(fs_path);
3409 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3411 if (ret == -ENOENT && ino == ino2)
3416 return parent_gen == ino1_gen ? 1 : 0;
3422 static int wait_for_parent_move(struct send_ctx *sctx,
3423 struct recorded_ref *parent_ref,
3424 const bool is_orphan)
3427 u64 ino = parent_ref->dir;
3428 u64 parent_ino_before, parent_ino_after;
3429 struct fs_path *path_before = NULL;
3430 struct fs_path *path_after = NULL;
3433 path_after = fs_path_alloc();
3434 path_before = fs_path_alloc();
3435 if (!path_after || !path_before) {
3441 * Our current directory inode may not yet be renamed/moved because some
3442 * ancestor (immediate or not) has to be renamed/moved first. So find if
3443 * such ancestor exists and make sure our own rename/move happens after
3444 * that ancestor is processed to avoid path build infinite loops (done
3445 * at get_cur_path()).
3447 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3448 if (is_waiting_for_move(sctx, ino)) {
3450 * If the current inode is an ancestor of ino in the
3451 * parent root, we need to delay the rename of the
3452 * current inode, otherwise don't delayed the rename
3453 * because we can end up with a circular dependency
3454 * of renames, resulting in some directories never
3455 * getting the respective rename operations issued in
3456 * the send stream or getting into infinite path build
3459 ret = is_ancestor(sctx->parent_root,
3460 sctx->cur_ino, sctx->cur_inode_gen,
3465 fs_path_reset(path_before);
3466 fs_path_reset(path_after);
3468 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3472 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3474 if (ret < 0 && ret != -ENOENT) {
3476 } else if (ret == -ENOENT) {
3481 len1 = fs_path_len(path_before);
3482 len2 = fs_path_len(path_after);
3483 if (ino > sctx->cur_ino &&
3484 (parent_ino_before != parent_ino_after || len1 != len2 ||
3485 memcmp(path_before->start, path_after->start, len1))) {
3489 ino = parent_ino_after;
3493 fs_path_free(path_before);
3494 fs_path_free(path_after);
3497 ret = add_pending_dir_move(sctx,
3499 sctx->cur_inode_gen,
3502 &sctx->deleted_refs,
3512 * This does all the move/link/unlink/rmdir magic.
3514 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3517 struct recorded_ref *cur;
3518 struct recorded_ref *cur2;
3519 struct list_head check_dirs;
3520 struct fs_path *valid_path = NULL;
3523 int did_overwrite = 0;
3525 u64 last_dir_ino_rm = 0;
3526 bool can_rename = true;
3528 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3531 * This should never happen as the root dir always has the same ref
3532 * which is always '..'
3534 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3535 INIT_LIST_HEAD(&check_dirs);
3537 valid_path = fs_path_alloc();
3544 * First, check if the first ref of the current inode was overwritten
3545 * before. If yes, we know that the current inode was already orphanized
3546 * and thus use the orphan name. If not, we can use get_cur_path to
3547 * get the path of the first ref as it would like while receiving at
3548 * this point in time.
3549 * New inodes are always orphan at the beginning, so force to use the
3550 * orphan name in this case.
3551 * The first ref is stored in valid_path and will be updated if it
3552 * gets moved around.
3554 if (!sctx->cur_inode_new) {
3555 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3556 sctx->cur_inode_gen);
3562 if (sctx->cur_inode_new || did_overwrite) {
3563 ret = gen_unique_name(sctx, sctx->cur_ino,
3564 sctx->cur_inode_gen, valid_path);
3569 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3575 list_for_each_entry(cur, &sctx->new_refs, list) {
3577 * We may have refs where the parent directory does not exist
3578 * yet. This happens if the parent directories inum is higher
3579 * the the current inum. To handle this case, we create the
3580 * parent directory out of order. But we need to check if this
3581 * did already happen before due to other refs in the same dir.
3583 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3586 if (ret == inode_state_will_create) {
3589 * First check if any of the current inodes refs did
3590 * already create the dir.
3592 list_for_each_entry(cur2, &sctx->new_refs, list) {
3595 if (cur2->dir == cur->dir) {
3602 * If that did not happen, check if a previous inode
3603 * did already create the dir.
3606 ret = did_create_dir(sctx, cur->dir);
3610 ret = send_create_inode(sctx, cur->dir);
3617 * Check if this new ref would overwrite the first ref of
3618 * another unprocessed inode. If yes, orphanize the
3619 * overwritten inode. If we find an overwritten ref that is
3620 * not the first ref, simply unlink it.
3622 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3623 cur->name, cur->name_len,
3624 &ow_inode, &ow_gen);
3628 ret = is_first_ref(sctx->parent_root,
3629 ow_inode, cur->dir, cur->name,
3634 struct name_cache_entry *nce;
3636 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3641 * Make sure we clear our orphanized inode's
3642 * name from the name cache. This is because the
3643 * inode ow_inode might be an ancestor of some
3644 * other inode that will be orphanized as well
3645 * later and has an inode number greater than
3646 * sctx->send_progress. We need to prevent
3647 * future name lookups from using the old name
3648 * and get instead the orphan name.
3650 nce = name_cache_search(sctx, ow_inode, ow_gen);
3652 name_cache_delete(sctx, nce);
3656 ret = send_unlink(sctx, cur->full_path);
3662 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3663 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3672 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3674 ret = wait_for_parent_move(sctx, cur, is_orphan);
3684 * link/move the ref to the new place. If we have an orphan
3685 * inode, move it and update valid_path. If not, link or move
3686 * it depending on the inode mode.
3688 if (is_orphan && can_rename) {
3689 ret = send_rename(sctx, valid_path, cur->full_path);
3693 ret = fs_path_copy(valid_path, cur->full_path);
3696 } else if (can_rename) {
3697 if (S_ISDIR(sctx->cur_inode_mode)) {
3699 * Dirs can't be linked, so move it. For moved
3700 * dirs, we always have one new and one deleted
3701 * ref. The deleted ref is ignored later.
3703 ret = send_rename(sctx, valid_path,
3706 ret = fs_path_copy(valid_path,
3711 ret = send_link(sctx, cur->full_path,
3717 ret = dup_ref(cur, &check_dirs);
3722 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3724 * Check if we can already rmdir the directory. If not,
3725 * orphanize it. For every dir item inside that gets deleted
3726 * later, we do this check again and rmdir it then if possible.
3727 * See the use of check_dirs for more details.
3729 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3734 ret = send_rmdir(sctx, valid_path);
3737 } else if (!is_orphan) {
3738 ret = orphanize_inode(sctx, sctx->cur_ino,
3739 sctx->cur_inode_gen, valid_path);
3745 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3746 ret = dup_ref(cur, &check_dirs);
3750 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3751 !list_empty(&sctx->deleted_refs)) {
3753 * We have a moved dir. Add the old parent to check_dirs
3755 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3757 ret = dup_ref(cur, &check_dirs);
3760 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3762 * We have a non dir inode. Go through all deleted refs and
3763 * unlink them if they were not already overwritten by other
3766 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3767 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3768 sctx->cur_ino, sctx->cur_inode_gen,
3769 cur->name, cur->name_len);
3773 ret = send_unlink(sctx, cur->full_path);
3777 ret = dup_ref(cur, &check_dirs);
3782 * If the inode is still orphan, unlink the orphan. This may
3783 * happen when a previous inode did overwrite the first ref
3784 * of this inode and no new refs were added for the current
3785 * inode. Unlinking does not mean that the inode is deleted in
3786 * all cases. There may still be links to this inode in other
3790 ret = send_unlink(sctx, valid_path);
3797 * We did collect all parent dirs where cur_inode was once located. We
3798 * now go through all these dirs and check if they are pending for
3799 * deletion and if it's finally possible to perform the rmdir now.
3800 * We also update the inode stats of the parent dirs here.
3802 list_for_each_entry(cur, &check_dirs, list) {
3804 * In case we had refs into dirs that were not processed yet,
3805 * we don't need to do the utime and rmdir logic for these dirs.
3806 * The dir will be processed later.
3808 if (cur->dir > sctx->cur_ino)
3811 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3815 if (ret == inode_state_did_create ||
3816 ret == inode_state_no_change) {
3817 /* TODO delayed utimes */
3818 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3821 } else if (ret == inode_state_did_delete &&
3822 cur->dir != last_dir_ino_rm) {
3823 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3828 ret = get_cur_path(sctx, cur->dir,
3829 cur->dir_gen, valid_path);
3832 ret = send_rmdir(sctx, valid_path);
3835 last_dir_ino_rm = cur->dir;
3843 __free_recorded_refs(&check_dirs);
3844 free_recorded_refs(sctx);
3845 fs_path_free(valid_path);
3849 static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3850 struct fs_path *name, void *ctx, struct list_head *refs)
3853 struct send_ctx *sctx = ctx;
3857 p = fs_path_alloc();
3861 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
3866 ret = get_cur_path(sctx, dir, gen, p);
3869 ret = fs_path_add_path(p, name);
3873 ret = __record_ref(refs, dir, gen, p);
3881 static int __record_new_ref(int num, u64 dir, int index,
3882 struct fs_path *name,
3885 struct send_ctx *sctx = ctx;
3886 return record_ref(sctx->send_root, num, dir, index, name,
3887 ctx, &sctx->new_refs);
3891 static int __record_deleted_ref(int num, u64 dir, int index,
3892 struct fs_path *name,
3895 struct send_ctx *sctx = ctx;
3896 return record_ref(sctx->parent_root, num, dir, index, name,
3897 ctx, &sctx->deleted_refs);
3900 static int record_new_ref(struct send_ctx *sctx)
3904 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3905 sctx->cmp_key, 0, __record_new_ref, sctx);
3914 static int record_deleted_ref(struct send_ctx *sctx)
3918 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3919 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3928 struct find_ref_ctx {
3931 struct btrfs_root *root;
3932 struct fs_path *name;
3936 static int __find_iref(int num, u64 dir, int index,
3937 struct fs_path *name,
3940 struct find_ref_ctx *ctx = ctx_;
3944 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3945 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3947 * To avoid doing extra lookups we'll only do this if everything
3950 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3954 if (dir_gen != ctx->dir_gen)
3956 ctx->found_idx = num;
3962 static int find_iref(struct btrfs_root *root,
3963 struct btrfs_path *path,
3964 struct btrfs_key *key,
3965 u64 dir, u64 dir_gen, struct fs_path *name)
3968 struct find_ref_ctx ctx;
3972 ctx.dir_gen = dir_gen;
3976 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3980 if (ctx.found_idx == -1)
3983 return ctx.found_idx;
3986 static int __record_changed_new_ref(int num, u64 dir, int index,
3987 struct fs_path *name,
3992 struct send_ctx *sctx = ctx;
3994 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3999 ret = find_iref(sctx->parent_root, sctx->right_path,
4000 sctx->cmp_key, dir, dir_gen, name);
4002 ret = __record_new_ref(num, dir, index, name, sctx);
4009 static int __record_changed_deleted_ref(int num, u64 dir, int index,
4010 struct fs_path *name,
4015 struct send_ctx *sctx = ctx;
4017 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4022 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4023 dir, dir_gen, name);
4025 ret = __record_deleted_ref(num, dir, index, name, sctx);
4032 static int record_changed_ref(struct send_ctx *sctx)
4036 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4037 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4040 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4041 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4051 * Record and process all refs at once. Needed when an inode changes the
4052 * generation number, which means that it was deleted and recreated.
4054 static int process_all_refs(struct send_ctx *sctx,
4055 enum btrfs_compare_tree_result cmd)
4058 struct btrfs_root *root;
4059 struct btrfs_path *path;
4060 struct btrfs_key key;
4061 struct btrfs_key found_key;
4062 struct extent_buffer *eb;
4064 iterate_inode_ref_t cb;
4065 int pending_move = 0;
4067 path = alloc_path_for_send();
4071 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4072 root = sctx->send_root;
4073 cb = __record_new_ref;
4074 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4075 root = sctx->parent_root;
4076 cb = __record_deleted_ref;
4078 btrfs_err(sctx->send_root->fs_info,
4079 "Wrong command %d in process_all_refs", cmd);
4084 key.objectid = sctx->cmp_key->objectid;
4085 key.type = BTRFS_INODE_REF_KEY;
4087 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4092 eb = path->nodes[0];
4093 slot = path->slots[0];
4094 if (slot >= btrfs_header_nritems(eb)) {
4095 ret = btrfs_next_leaf(root, path);
4103 btrfs_item_key_to_cpu(eb, &found_key, slot);
4105 if (found_key.objectid != key.objectid ||
4106 (found_key.type != BTRFS_INODE_REF_KEY &&
4107 found_key.type != BTRFS_INODE_EXTREF_KEY))
4110 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4116 btrfs_release_path(path);
4118 ret = process_recorded_refs(sctx, &pending_move);
4119 /* Only applicable to an incremental send. */
4120 ASSERT(pending_move == 0);
4123 btrfs_free_path(path);
4127 static int send_set_xattr(struct send_ctx *sctx,
4128 struct fs_path *path,
4129 const char *name, int name_len,
4130 const char *data, int data_len)
4134 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4138 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4139 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4140 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4142 ret = send_cmd(sctx);
4149 static int send_remove_xattr(struct send_ctx *sctx,
4150 struct fs_path *path,
4151 const char *name, int name_len)
4155 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4159 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4160 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4162 ret = send_cmd(sctx);
4169 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4170 const char *name, int name_len,
4171 const char *data, int data_len,
4175 struct send_ctx *sctx = ctx;
4177 posix_acl_xattr_header dummy_acl;
4179 p = fs_path_alloc();
4184 * This hack is needed because empty acl's are stored as zero byte
4185 * data in xattrs. Problem with that is, that receiving these zero byte
4186 * acl's will fail later. To fix this, we send a dummy acl list that
4187 * only contains the version number and no entries.
4189 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4190 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4191 if (data_len == 0) {
4192 dummy_acl.a_version =
4193 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4194 data = (char *)&dummy_acl;
4195 data_len = sizeof(dummy_acl);
4199 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4203 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4210 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4211 const char *name, int name_len,
4212 const char *data, int data_len,
4216 struct send_ctx *sctx = ctx;
4219 p = fs_path_alloc();
4223 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4227 ret = send_remove_xattr(sctx, p, name, name_len);
4234 static int process_new_xattr(struct send_ctx *sctx)
4238 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4239 sctx->cmp_key, __process_new_xattr, sctx);
4244 static int process_deleted_xattr(struct send_ctx *sctx)
4248 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4249 sctx->cmp_key, __process_deleted_xattr, sctx);
4254 struct find_xattr_ctx {
4262 static int __find_xattr(int num, struct btrfs_key *di_key,
4263 const char *name, int name_len,
4264 const char *data, int data_len,
4265 u8 type, void *vctx)
4267 struct find_xattr_ctx *ctx = vctx;
4269 if (name_len == ctx->name_len &&
4270 strncmp(name, ctx->name, name_len) == 0) {
4271 ctx->found_idx = num;
4272 ctx->found_data_len = data_len;
4273 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
4274 if (!ctx->found_data)
4281 static int find_xattr(struct btrfs_root *root,
4282 struct btrfs_path *path,
4283 struct btrfs_key *key,
4284 const char *name, int name_len,
4285 char **data, int *data_len)
4288 struct find_xattr_ctx ctx;
4291 ctx.name_len = name_len;
4293 ctx.found_data = NULL;
4294 ctx.found_data_len = 0;
4296 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4300 if (ctx.found_idx == -1)
4303 *data = ctx.found_data;
4304 *data_len = ctx.found_data_len;
4306 kfree(ctx.found_data);
4308 return ctx.found_idx;
4312 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4313 const char *name, int name_len,
4314 const char *data, int data_len,
4318 struct send_ctx *sctx = ctx;
4319 char *found_data = NULL;
4320 int found_data_len = 0;
4322 ret = find_xattr(sctx->parent_root, sctx->right_path,
4323 sctx->cmp_key, name, name_len, &found_data,
4325 if (ret == -ENOENT) {
4326 ret = __process_new_xattr(num, di_key, name, name_len, data,
4327 data_len, type, ctx);
4328 } else if (ret >= 0) {
4329 if (data_len != found_data_len ||
4330 memcmp(data, found_data, data_len)) {
4331 ret = __process_new_xattr(num, di_key, name, name_len,
4332 data, data_len, type, ctx);
4342 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4343 const char *name, int name_len,
4344 const char *data, int data_len,
4348 struct send_ctx *sctx = ctx;
4350 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4351 name, name_len, NULL, NULL);
4353 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4354 data_len, type, ctx);
4361 static int process_changed_xattr(struct send_ctx *sctx)
4365 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4366 sctx->cmp_key, __process_changed_new_xattr, sctx);
4369 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4370 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4376 static int process_all_new_xattrs(struct send_ctx *sctx)
4379 struct btrfs_root *root;
4380 struct btrfs_path *path;
4381 struct btrfs_key key;
4382 struct btrfs_key found_key;
4383 struct extent_buffer *eb;
4386 path = alloc_path_for_send();
4390 root = sctx->send_root;
4392 key.objectid = sctx->cmp_key->objectid;
4393 key.type = BTRFS_XATTR_ITEM_KEY;
4395 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4400 eb = path->nodes[0];
4401 slot = path->slots[0];
4402 if (slot >= btrfs_header_nritems(eb)) {
4403 ret = btrfs_next_leaf(root, path);
4406 } else if (ret > 0) {
4413 btrfs_item_key_to_cpu(eb, &found_key, slot);
4414 if (found_key.objectid != key.objectid ||
4415 found_key.type != key.type) {
4420 ret = iterate_dir_item(root, path, &found_key,
4421 __process_new_xattr, sctx);
4429 btrfs_free_path(path);
4433 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4435 struct btrfs_root *root = sctx->send_root;
4436 struct btrfs_fs_info *fs_info = root->fs_info;
4437 struct inode *inode;
4440 struct btrfs_key key;
4441 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4443 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4446 key.objectid = sctx->cur_ino;
4447 key.type = BTRFS_INODE_ITEM_KEY;
4450 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4452 return PTR_ERR(inode);
4454 if (offset + len > i_size_read(inode)) {
4455 if (offset > i_size_read(inode))
4458 len = offset - i_size_read(inode);
4463 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4465 /* initial readahead */
4466 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4467 file_ra_state_init(&sctx->ra, inode->i_mapping);
4468 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4469 last_index - index + 1);
4471 while (index <= last_index) {
4472 unsigned cur_len = min_t(unsigned, len,
4473 PAGE_CACHE_SIZE - pg_offset);
4474 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4480 if (!PageUptodate(page)) {
4481 btrfs_readpage(NULL, page);
4483 if (!PageUptodate(page)) {
4485 page_cache_release(page);
4492 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4495 page_cache_release(page);
4507 * Read some bytes from the current inode/file and send a write command to
4510 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4514 ssize_t num_read = 0;
4516 p = fs_path_alloc();
4520 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4522 num_read = fill_read_buf(sctx, offset, len);
4523 if (num_read <= 0) {
4529 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4533 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4537 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4538 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4539 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4541 ret = send_cmd(sctx);
4552 * Send a clone command to user space.
4554 static int send_clone(struct send_ctx *sctx,
4555 u64 offset, u32 len,
4556 struct clone_root *clone_root)
4562 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4563 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4564 clone_root->root->objectid, clone_root->ino,
4565 clone_root->offset);
4567 p = fs_path_alloc();
4571 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4575 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4579 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4580 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4581 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4583 if (clone_root->root == sctx->send_root) {
4584 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4585 &gen, NULL, NULL, NULL, NULL);
4588 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4590 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4596 * If the parent we're using has a received_uuid set then use that as
4597 * our clone source as that is what we will look for when doing a
4600 * This covers the case that we create a snapshot off of a received
4601 * subvolume and then use that as the parent and try to receive on a
4604 if (!btrfs_is_empty_uuid(clone_root->root->root_item.received_uuid))
4605 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4606 clone_root->root->root_item.received_uuid);
4608 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4609 clone_root->root->root_item.uuid);
4610 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4611 le64_to_cpu(clone_root->root->root_item.ctransid));
4612 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4613 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4614 clone_root->offset);
4616 ret = send_cmd(sctx);
4625 * Send an update extent command to user space.
4627 static int send_update_extent(struct send_ctx *sctx,
4628 u64 offset, u32 len)
4633 p = fs_path_alloc();
4637 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4641 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4645 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4646 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4647 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4649 ret = send_cmd(sctx);
4657 static int send_hole(struct send_ctx *sctx, u64 end)
4659 struct fs_path *p = NULL;
4660 u64 offset = sctx->cur_inode_last_extent;
4664 p = fs_path_alloc();
4667 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4669 goto tlv_put_failure;
4670 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4671 while (offset < end) {
4672 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4674 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4677 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4678 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4679 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4680 ret = send_cmd(sctx);
4690 static int send_write_or_clone(struct send_ctx *sctx,
4691 struct btrfs_path *path,
4692 struct btrfs_key *key,
4693 struct clone_root *clone_root)
4696 struct btrfs_file_extent_item *ei;
4697 u64 offset = key->offset;
4702 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4704 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4705 struct btrfs_file_extent_item);
4706 type = btrfs_file_extent_type(path->nodes[0], ei);
4707 if (type == BTRFS_FILE_EXTENT_INLINE) {
4708 len = btrfs_file_extent_inline_len(path->nodes[0],
4709 path->slots[0], ei);
4711 * it is possible the inline item won't cover the whole page,
4712 * but there may be items after this page. Make
4713 * sure to send the whole thing
4715 len = PAGE_CACHE_ALIGN(len);
4717 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4720 if (offset + len > sctx->cur_inode_size)
4721 len = sctx->cur_inode_size - offset;
4727 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4728 ret = send_clone(sctx, offset, len, clone_root);
4729 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4730 ret = send_update_extent(sctx, offset, len);
4734 if (l > BTRFS_SEND_READ_SIZE)
4735 l = BTRFS_SEND_READ_SIZE;
4736 ret = send_write(sctx, pos + offset, l);
4749 static int is_extent_unchanged(struct send_ctx *sctx,
4750 struct btrfs_path *left_path,
4751 struct btrfs_key *ekey)
4754 struct btrfs_key key;
4755 struct btrfs_path *path = NULL;
4756 struct extent_buffer *eb;
4758 struct btrfs_key found_key;
4759 struct btrfs_file_extent_item *ei;
4764 u64 left_offset_fixed;
4772 path = alloc_path_for_send();
4776 eb = left_path->nodes[0];
4777 slot = left_path->slots[0];
4778 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4779 left_type = btrfs_file_extent_type(eb, ei);
4781 if (left_type != BTRFS_FILE_EXTENT_REG) {
4785 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4786 left_len = btrfs_file_extent_num_bytes(eb, ei);
4787 left_offset = btrfs_file_extent_offset(eb, ei);
4788 left_gen = btrfs_file_extent_generation(eb, ei);
4791 * Following comments will refer to these graphics. L is the left
4792 * extents which we are checking at the moment. 1-8 are the right
4793 * extents that we iterate.
4796 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4799 * |--1--|-2b-|...(same as above)
4801 * Alternative situation. Happens on files where extents got split.
4803 * |-----------7-----------|-6-|
4805 * Alternative situation. Happens on files which got larger.
4808 * Nothing follows after 8.
4811 key.objectid = ekey->objectid;
4812 key.type = BTRFS_EXTENT_DATA_KEY;
4813 key.offset = ekey->offset;
4814 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4823 * Handle special case where the right side has no extents at all.
4825 eb = path->nodes[0];
4826 slot = path->slots[0];
4827 btrfs_item_key_to_cpu(eb, &found_key, slot);
4828 if (found_key.objectid != key.objectid ||
4829 found_key.type != key.type) {
4830 /* If we're a hole then just pretend nothing changed */
4831 ret = (left_disknr) ? 0 : 1;
4836 * We're now on 2a, 2b or 7.
4839 while (key.offset < ekey->offset + left_len) {
4840 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4841 right_type = btrfs_file_extent_type(eb, ei);
4842 if (right_type != BTRFS_FILE_EXTENT_REG) {
4847 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4848 right_len = btrfs_file_extent_num_bytes(eb, ei);
4849 right_offset = btrfs_file_extent_offset(eb, ei);
4850 right_gen = btrfs_file_extent_generation(eb, ei);
4853 * Are we at extent 8? If yes, we know the extent is changed.
4854 * This may only happen on the first iteration.
4856 if (found_key.offset + right_len <= ekey->offset) {
4857 /* If we're a hole just pretend nothing changed */
4858 ret = (left_disknr) ? 0 : 1;
4862 left_offset_fixed = left_offset;
4863 if (key.offset < ekey->offset) {
4864 /* Fix the right offset for 2a and 7. */
4865 right_offset += ekey->offset - key.offset;
4867 /* Fix the left offset for all behind 2a and 2b */
4868 left_offset_fixed += key.offset - ekey->offset;
4872 * Check if we have the same extent.
4874 if (left_disknr != right_disknr ||
4875 left_offset_fixed != right_offset ||
4876 left_gen != right_gen) {
4882 * Go to the next extent.
4884 ret = btrfs_next_item(sctx->parent_root, path);
4888 eb = path->nodes[0];
4889 slot = path->slots[0];
4890 btrfs_item_key_to_cpu(eb, &found_key, slot);
4892 if (ret || found_key.objectid != key.objectid ||
4893 found_key.type != key.type) {
4894 key.offset += right_len;
4897 if (found_key.offset != key.offset + right_len) {
4905 * We're now behind the left extent (treat as unchanged) or at the end
4906 * of the right side (treat as changed).
4908 if (key.offset >= ekey->offset + left_len)
4915 btrfs_free_path(path);
4919 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4921 struct btrfs_path *path;
4922 struct btrfs_root *root = sctx->send_root;
4923 struct btrfs_file_extent_item *fi;
4924 struct btrfs_key key;
4929 path = alloc_path_for_send();
4933 sctx->cur_inode_last_extent = 0;
4935 key.objectid = sctx->cur_ino;
4936 key.type = BTRFS_EXTENT_DATA_KEY;
4937 key.offset = offset;
4938 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4942 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4943 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4946 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4947 struct btrfs_file_extent_item);
4948 type = btrfs_file_extent_type(path->nodes[0], fi);
4949 if (type == BTRFS_FILE_EXTENT_INLINE) {
4950 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4951 path->slots[0], fi);
4952 extent_end = ALIGN(key.offset + size,
4953 sctx->send_root->sectorsize);
4955 extent_end = key.offset +
4956 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4958 sctx->cur_inode_last_extent = extent_end;
4960 btrfs_free_path(path);
4964 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4965 struct btrfs_key *key)
4967 struct btrfs_file_extent_item *fi;
4972 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4975 if (sctx->cur_inode_last_extent == (u64)-1) {
4976 ret = get_last_extent(sctx, key->offset - 1);
4981 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4982 struct btrfs_file_extent_item);
4983 type = btrfs_file_extent_type(path->nodes[0], fi);
4984 if (type == BTRFS_FILE_EXTENT_INLINE) {
4985 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4986 path->slots[0], fi);
4987 extent_end = ALIGN(key->offset + size,
4988 sctx->send_root->sectorsize);
4990 extent_end = key->offset +
4991 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4994 if (path->slots[0] == 0 &&
4995 sctx->cur_inode_last_extent < key->offset) {
4997 * We might have skipped entire leafs that contained only
4998 * file extent items for our current inode. These leafs have
4999 * a generation number smaller (older) than the one in the
5000 * current leaf and the leaf our last extent came from, and
5001 * are located between these 2 leafs.
5003 ret = get_last_extent(sctx, key->offset - 1);
5008 if (sctx->cur_inode_last_extent < key->offset)
5009 ret = send_hole(sctx, key->offset);
5010 sctx->cur_inode_last_extent = extent_end;
5014 static int process_extent(struct send_ctx *sctx,
5015 struct btrfs_path *path,
5016 struct btrfs_key *key)
5018 struct clone_root *found_clone = NULL;
5021 if (S_ISLNK(sctx->cur_inode_mode))
5024 if (sctx->parent_root && !sctx->cur_inode_new) {
5025 ret = is_extent_unchanged(sctx, path, key);
5033 struct btrfs_file_extent_item *ei;
5036 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5037 struct btrfs_file_extent_item);
5038 type = btrfs_file_extent_type(path->nodes[0], ei);
5039 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5040 type == BTRFS_FILE_EXTENT_REG) {
5042 * The send spec does not have a prealloc command yet,
5043 * so just leave a hole for prealloc'ed extents until
5044 * we have enough commands queued up to justify rev'ing
5047 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5052 /* Have a hole, just skip it. */
5053 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5060 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5061 sctx->cur_inode_size, &found_clone);
5062 if (ret != -ENOENT && ret < 0)
5065 ret = send_write_or_clone(sctx, path, key, found_clone);
5069 ret = maybe_send_hole(sctx, path, key);
5074 static int process_all_extents(struct send_ctx *sctx)
5077 struct btrfs_root *root;
5078 struct btrfs_path *path;
5079 struct btrfs_key key;
5080 struct btrfs_key found_key;
5081 struct extent_buffer *eb;
5084 root = sctx->send_root;
5085 path = alloc_path_for_send();
5089 key.objectid = sctx->cmp_key->objectid;
5090 key.type = BTRFS_EXTENT_DATA_KEY;
5092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5097 eb = path->nodes[0];
5098 slot = path->slots[0];
5100 if (slot >= btrfs_header_nritems(eb)) {
5101 ret = btrfs_next_leaf(root, path);
5104 } else if (ret > 0) {
5111 btrfs_item_key_to_cpu(eb, &found_key, slot);
5113 if (found_key.objectid != key.objectid ||
5114 found_key.type != key.type) {
5119 ret = process_extent(sctx, path, &found_key);
5127 btrfs_free_path(path);
5131 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5133 int *refs_processed)
5137 if (sctx->cur_ino == 0)
5139 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5140 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5142 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5145 ret = process_recorded_refs(sctx, pending_move);
5149 *refs_processed = 1;
5154 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5165 int pending_move = 0;
5166 int refs_processed = 0;
5168 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5174 * We have processed the refs and thus need to advance send_progress.
5175 * Now, calls to get_cur_xxx will take the updated refs of the current
5176 * inode into account.
5178 * On the other hand, if our current inode is a directory and couldn't
5179 * be moved/renamed because its parent was renamed/moved too and it has
5180 * a higher inode number, we can only move/rename our current inode
5181 * after we moved/renamed its parent. Therefore in this case operate on
5182 * the old path (pre move/rename) of our current inode, and the
5183 * move/rename will be performed later.
5185 if (refs_processed && !pending_move)
5186 sctx->send_progress = sctx->cur_ino + 1;
5188 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5190 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5193 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5194 &left_mode, &left_uid, &left_gid, NULL);
5198 if (!sctx->parent_root || sctx->cur_inode_new) {
5200 if (!S_ISLNK(sctx->cur_inode_mode))
5203 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5204 NULL, NULL, &right_mode, &right_uid,
5209 if (left_uid != right_uid || left_gid != right_gid)
5211 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5215 if (S_ISREG(sctx->cur_inode_mode)) {
5216 if (need_send_hole(sctx)) {
5217 if (sctx->cur_inode_last_extent == (u64)-1 ||
5218 sctx->cur_inode_last_extent <
5219 sctx->cur_inode_size) {
5220 ret = get_last_extent(sctx, (u64)-1);
5224 if (sctx->cur_inode_last_extent <
5225 sctx->cur_inode_size) {
5226 ret = send_hole(sctx, sctx->cur_inode_size);
5231 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5232 sctx->cur_inode_size);
5238 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5239 left_uid, left_gid);
5244 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5251 * If other directory inodes depended on our current directory
5252 * inode's move/rename, now do their move/rename operations.
5254 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5255 ret = apply_children_dir_moves(sctx);
5259 * Need to send that every time, no matter if it actually
5260 * changed between the two trees as we have done changes to
5261 * the inode before. If our inode is a directory and it's
5262 * waiting to be moved/renamed, we will send its utimes when
5263 * it's moved/renamed, therefore we don't need to do it here.
5265 sctx->send_progress = sctx->cur_ino + 1;
5266 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5275 static int changed_inode(struct send_ctx *sctx,
5276 enum btrfs_compare_tree_result result)
5279 struct btrfs_key *key = sctx->cmp_key;
5280 struct btrfs_inode_item *left_ii = NULL;
5281 struct btrfs_inode_item *right_ii = NULL;
5285 sctx->cur_ino = key->objectid;
5286 sctx->cur_inode_new_gen = 0;
5287 sctx->cur_inode_last_extent = (u64)-1;
5290 * Set send_progress to current inode. This will tell all get_cur_xxx
5291 * functions that the current inode's refs are not updated yet. Later,
5292 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5294 sctx->send_progress = sctx->cur_ino;
5296 if (result == BTRFS_COMPARE_TREE_NEW ||
5297 result == BTRFS_COMPARE_TREE_CHANGED) {
5298 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5299 sctx->left_path->slots[0],
5300 struct btrfs_inode_item);
5301 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5304 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5305 sctx->right_path->slots[0],
5306 struct btrfs_inode_item);
5307 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5310 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5311 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5312 sctx->right_path->slots[0],
5313 struct btrfs_inode_item);
5315 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5319 * The cur_ino = root dir case is special here. We can't treat
5320 * the inode as deleted+reused because it would generate a
5321 * stream that tries to delete/mkdir the root dir.
5323 if (left_gen != right_gen &&
5324 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5325 sctx->cur_inode_new_gen = 1;
5328 if (result == BTRFS_COMPARE_TREE_NEW) {
5329 sctx->cur_inode_gen = left_gen;
5330 sctx->cur_inode_new = 1;
5331 sctx->cur_inode_deleted = 0;
5332 sctx->cur_inode_size = btrfs_inode_size(
5333 sctx->left_path->nodes[0], left_ii);
5334 sctx->cur_inode_mode = btrfs_inode_mode(
5335 sctx->left_path->nodes[0], left_ii);
5336 sctx->cur_inode_rdev = btrfs_inode_rdev(
5337 sctx->left_path->nodes[0], left_ii);
5338 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5339 ret = send_create_inode_if_needed(sctx);
5340 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5341 sctx->cur_inode_gen = right_gen;
5342 sctx->cur_inode_new = 0;
5343 sctx->cur_inode_deleted = 1;
5344 sctx->cur_inode_size = btrfs_inode_size(
5345 sctx->right_path->nodes[0], right_ii);
5346 sctx->cur_inode_mode = btrfs_inode_mode(
5347 sctx->right_path->nodes[0], right_ii);
5348 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5350 * We need to do some special handling in case the inode was
5351 * reported as changed with a changed generation number. This
5352 * means that the original inode was deleted and new inode
5353 * reused the same inum. So we have to treat the old inode as
5354 * deleted and the new one as new.
5356 if (sctx->cur_inode_new_gen) {
5358 * First, process the inode as if it was deleted.
5360 sctx->cur_inode_gen = right_gen;
5361 sctx->cur_inode_new = 0;
5362 sctx->cur_inode_deleted = 1;
5363 sctx->cur_inode_size = btrfs_inode_size(
5364 sctx->right_path->nodes[0], right_ii);
5365 sctx->cur_inode_mode = btrfs_inode_mode(
5366 sctx->right_path->nodes[0], right_ii);
5367 ret = process_all_refs(sctx,
5368 BTRFS_COMPARE_TREE_DELETED);
5373 * Now process the inode as if it was new.
5375 sctx->cur_inode_gen = left_gen;
5376 sctx->cur_inode_new = 1;
5377 sctx->cur_inode_deleted = 0;
5378 sctx->cur_inode_size = btrfs_inode_size(
5379 sctx->left_path->nodes[0], left_ii);
5380 sctx->cur_inode_mode = btrfs_inode_mode(
5381 sctx->left_path->nodes[0], left_ii);
5382 sctx->cur_inode_rdev = btrfs_inode_rdev(
5383 sctx->left_path->nodes[0], left_ii);
5384 ret = send_create_inode_if_needed(sctx);
5388 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5392 * Advance send_progress now as we did not get into
5393 * process_recorded_refs_if_needed in the new_gen case.
5395 sctx->send_progress = sctx->cur_ino + 1;
5398 * Now process all extents and xattrs of the inode as if
5399 * they were all new.
5401 ret = process_all_extents(sctx);
5404 ret = process_all_new_xattrs(sctx);
5408 sctx->cur_inode_gen = left_gen;
5409 sctx->cur_inode_new = 0;
5410 sctx->cur_inode_new_gen = 0;
5411 sctx->cur_inode_deleted = 0;
5412 sctx->cur_inode_size = btrfs_inode_size(
5413 sctx->left_path->nodes[0], left_ii);
5414 sctx->cur_inode_mode = btrfs_inode_mode(
5415 sctx->left_path->nodes[0], left_ii);
5424 * We have to process new refs before deleted refs, but compare_trees gives us
5425 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5426 * first and later process them in process_recorded_refs.
5427 * For the cur_inode_new_gen case, we skip recording completely because
5428 * changed_inode did already initiate processing of refs. The reason for this is
5429 * that in this case, compare_tree actually compares the refs of 2 different
5430 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5431 * refs of the right tree as deleted and all refs of the left tree as new.
5433 static int changed_ref(struct send_ctx *sctx,
5434 enum btrfs_compare_tree_result result)
5438 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5440 if (!sctx->cur_inode_new_gen &&
5441 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5442 if (result == BTRFS_COMPARE_TREE_NEW)
5443 ret = record_new_ref(sctx);
5444 else if (result == BTRFS_COMPARE_TREE_DELETED)
5445 ret = record_deleted_ref(sctx);
5446 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5447 ret = record_changed_ref(sctx);
5454 * Process new/deleted/changed xattrs. We skip processing in the
5455 * cur_inode_new_gen case because changed_inode did already initiate processing
5456 * of xattrs. The reason is the same as in changed_ref
5458 static int changed_xattr(struct send_ctx *sctx,
5459 enum btrfs_compare_tree_result result)
5463 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5465 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5466 if (result == BTRFS_COMPARE_TREE_NEW)
5467 ret = process_new_xattr(sctx);
5468 else if (result == BTRFS_COMPARE_TREE_DELETED)
5469 ret = process_deleted_xattr(sctx);
5470 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5471 ret = process_changed_xattr(sctx);
5478 * Process new/deleted/changed extents. We skip processing in the
5479 * cur_inode_new_gen case because changed_inode did already initiate processing
5480 * of extents. The reason is the same as in changed_ref
5482 static int changed_extent(struct send_ctx *sctx,
5483 enum btrfs_compare_tree_result result)
5487 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5489 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5490 if (result != BTRFS_COMPARE_TREE_DELETED)
5491 ret = process_extent(sctx, sctx->left_path,
5498 static int dir_changed(struct send_ctx *sctx, u64 dir)
5500 u64 orig_gen, new_gen;
5503 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5508 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5513 return (orig_gen != new_gen) ? 1 : 0;
5516 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5517 struct btrfs_key *key)
5519 struct btrfs_inode_extref *extref;
5520 struct extent_buffer *leaf;
5521 u64 dirid = 0, last_dirid = 0;
5528 /* Easy case, just check this one dirid */
5529 if (key->type == BTRFS_INODE_REF_KEY) {
5530 dirid = key->offset;
5532 ret = dir_changed(sctx, dirid);
5536 leaf = path->nodes[0];
5537 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5538 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5539 while (cur_offset < item_size) {
5540 extref = (struct btrfs_inode_extref *)(ptr +
5542 dirid = btrfs_inode_extref_parent(leaf, extref);
5543 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5544 cur_offset += ref_name_len + sizeof(*extref);
5545 if (dirid == last_dirid)
5547 ret = dir_changed(sctx, dirid);
5557 * Updates compare related fields in sctx and simply forwards to the actual
5558 * changed_xxx functions.
5560 static int changed_cb(struct btrfs_root *left_root,
5561 struct btrfs_root *right_root,
5562 struct btrfs_path *left_path,
5563 struct btrfs_path *right_path,
5564 struct btrfs_key *key,
5565 enum btrfs_compare_tree_result result,
5569 struct send_ctx *sctx = ctx;
5571 if (result == BTRFS_COMPARE_TREE_SAME) {
5572 if (key->type == BTRFS_INODE_REF_KEY ||
5573 key->type == BTRFS_INODE_EXTREF_KEY) {
5574 ret = compare_refs(sctx, left_path, key);
5579 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5580 return maybe_send_hole(sctx, left_path, key);
5584 result = BTRFS_COMPARE_TREE_CHANGED;
5588 sctx->left_path = left_path;
5589 sctx->right_path = right_path;
5590 sctx->cmp_key = key;
5592 ret = finish_inode_if_needed(sctx, 0);
5596 /* Ignore non-FS objects */
5597 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5598 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5601 if (key->type == BTRFS_INODE_ITEM_KEY)
5602 ret = changed_inode(sctx, result);
5603 else if (key->type == BTRFS_INODE_REF_KEY ||
5604 key->type == BTRFS_INODE_EXTREF_KEY)
5605 ret = changed_ref(sctx, result);
5606 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5607 ret = changed_xattr(sctx, result);
5608 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5609 ret = changed_extent(sctx, result);
5615 static int full_send_tree(struct send_ctx *sctx)
5618 struct btrfs_root *send_root = sctx->send_root;
5619 struct btrfs_key key;
5620 struct btrfs_key found_key;
5621 struct btrfs_path *path;
5622 struct extent_buffer *eb;
5625 path = alloc_path_for_send();
5629 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5630 key.type = BTRFS_INODE_ITEM_KEY;
5633 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5640 eb = path->nodes[0];
5641 slot = path->slots[0];
5642 btrfs_item_key_to_cpu(eb, &found_key, slot);
5644 ret = changed_cb(send_root, NULL, path, NULL,
5645 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5649 key.objectid = found_key.objectid;
5650 key.type = found_key.type;
5651 key.offset = found_key.offset + 1;
5653 ret = btrfs_next_item(send_root, path);
5663 ret = finish_inode_if_needed(sctx, 1);
5666 btrfs_free_path(path);
5670 static int send_subvol(struct send_ctx *sctx)
5674 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5675 ret = send_header(sctx);
5680 ret = send_subvol_begin(sctx);
5684 if (sctx->parent_root) {
5685 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5689 ret = finish_inode_if_needed(sctx, 1);
5693 ret = full_send_tree(sctx);
5699 free_recorded_refs(sctx);
5704 * If orphan cleanup did remove any orphans from a root, it means the tree
5705 * was modified and therefore the commit root is not the same as the current
5706 * root anymore. This is a problem, because send uses the commit root and
5707 * therefore can see inode items that don't exist in the current root anymore,
5708 * and for example make calls to btrfs_iget, which will do tree lookups based
5709 * on the current root and not on the commit root. Those lookups will fail,
5710 * returning a -ESTALE error, and making send fail with that error. So make
5711 * sure a send does not see any orphans we have just removed, and that it will
5712 * see the same inodes regardless of whether a transaction commit happened
5713 * before it started (meaning that the commit root will be the same as the
5714 * current root) or not.
5716 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5719 struct btrfs_trans_handle *trans = NULL;
5722 if (sctx->parent_root &&
5723 sctx->parent_root->node != sctx->parent_root->commit_root)
5726 for (i = 0; i < sctx->clone_roots_cnt; i++)
5727 if (sctx->clone_roots[i].root->node !=
5728 sctx->clone_roots[i].root->commit_root)
5732 return btrfs_end_transaction(trans, sctx->send_root);
5737 /* Use any root, all fs roots will get their commit roots updated. */
5739 trans = btrfs_join_transaction(sctx->send_root);
5741 return PTR_ERR(trans);
5745 return btrfs_commit_transaction(trans, sctx->send_root);
5748 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5750 spin_lock(&root->root_item_lock);
5751 root->send_in_progress--;
5753 * Not much left to do, we don't know why it's unbalanced and
5754 * can't blindly reset it to 0.
5756 if (root->send_in_progress < 0)
5757 btrfs_err(root->fs_info,
5758 "send_in_progres unbalanced %d root %llu",
5759 root->send_in_progress, root->root_key.objectid);
5760 spin_unlock(&root->root_item_lock);
5763 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5766 struct btrfs_root *send_root;
5767 struct btrfs_root *clone_root;
5768 struct btrfs_fs_info *fs_info;
5769 struct btrfs_ioctl_send_args *arg = NULL;
5770 struct btrfs_key key;
5771 struct send_ctx *sctx = NULL;
5773 u64 *clone_sources_tmp = NULL;
5774 int clone_sources_to_rollback = 0;
5775 int sort_clone_roots = 0;
5778 if (!capable(CAP_SYS_ADMIN))
5781 send_root = BTRFS_I(file_inode(mnt_file))->root;
5782 fs_info = send_root->fs_info;
5785 * The subvolume must remain read-only during send, protect against
5786 * making it RW. This also protects against deletion.
5788 spin_lock(&send_root->root_item_lock);
5789 send_root->send_in_progress++;
5790 spin_unlock(&send_root->root_item_lock);
5793 * This is done when we lookup the root, it should already be complete
5794 * by the time we get here.
5796 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5799 * Userspace tools do the checks and warn the user if it's
5802 if (!btrfs_root_readonly(send_root)) {
5807 arg = memdup_user(arg_, sizeof(*arg));
5814 if (!access_ok(VERIFY_READ, arg->clone_sources,
5815 sizeof(*arg->clone_sources) *
5816 arg->clone_sources_count)) {
5821 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5826 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5832 INIT_LIST_HEAD(&sctx->new_refs);
5833 INIT_LIST_HEAD(&sctx->deleted_refs);
5834 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5835 INIT_LIST_HEAD(&sctx->name_cache_list);
5837 sctx->flags = arg->flags;
5839 sctx->send_filp = fget(arg->send_fd);
5840 if (!sctx->send_filp) {
5845 sctx->send_root = send_root;
5847 * Unlikely but possible, if the subvolume is marked for deletion but
5848 * is slow to remove the directory entry, send can still be started
5850 if (btrfs_root_dead(sctx->send_root)) {
5855 sctx->clone_roots_cnt = arg->clone_sources_count;
5857 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5858 sctx->send_buf = vmalloc(sctx->send_max_size);
5859 if (!sctx->send_buf) {
5864 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5865 if (!sctx->read_buf) {
5870 sctx->pending_dir_moves = RB_ROOT;
5871 sctx->waiting_dir_moves = RB_ROOT;
5872 sctx->orphan_dirs = RB_ROOT;
5874 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5875 (arg->clone_sources_count + 1));
5876 if (!sctx->clone_roots) {
5881 if (arg->clone_sources_count) {
5882 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5883 sizeof(*arg->clone_sources));
5884 if (!clone_sources_tmp) {
5889 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5890 arg->clone_sources_count *
5891 sizeof(*arg->clone_sources));
5897 for (i = 0; i < arg->clone_sources_count; i++) {
5898 key.objectid = clone_sources_tmp[i];
5899 key.type = BTRFS_ROOT_ITEM_KEY;
5900 key.offset = (u64)-1;
5902 index = srcu_read_lock(&fs_info->subvol_srcu);
5904 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5905 if (IS_ERR(clone_root)) {
5906 srcu_read_unlock(&fs_info->subvol_srcu, index);
5907 ret = PTR_ERR(clone_root);
5910 spin_lock(&clone_root->root_item_lock);
5911 if (!btrfs_root_readonly(clone_root) ||
5912 btrfs_root_dead(clone_root)) {
5913 spin_unlock(&clone_root->root_item_lock);
5914 srcu_read_unlock(&fs_info->subvol_srcu, index);
5918 clone_root->send_in_progress++;
5919 spin_unlock(&clone_root->root_item_lock);
5920 srcu_read_unlock(&fs_info->subvol_srcu, index);
5922 sctx->clone_roots[i].root = clone_root;
5923 clone_sources_to_rollback = i + 1;
5925 vfree(clone_sources_tmp);
5926 clone_sources_tmp = NULL;
5929 if (arg->parent_root) {
5930 key.objectid = arg->parent_root;
5931 key.type = BTRFS_ROOT_ITEM_KEY;
5932 key.offset = (u64)-1;
5934 index = srcu_read_lock(&fs_info->subvol_srcu);
5936 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5937 if (IS_ERR(sctx->parent_root)) {
5938 srcu_read_unlock(&fs_info->subvol_srcu, index);
5939 ret = PTR_ERR(sctx->parent_root);
5943 spin_lock(&sctx->parent_root->root_item_lock);
5944 sctx->parent_root->send_in_progress++;
5945 if (!btrfs_root_readonly(sctx->parent_root) ||
5946 btrfs_root_dead(sctx->parent_root)) {
5947 spin_unlock(&sctx->parent_root->root_item_lock);
5948 srcu_read_unlock(&fs_info->subvol_srcu, index);
5952 spin_unlock(&sctx->parent_root->root_item_lock);
5954 srcu_read_unlock(&fs_info->subvol_srcu, index);
5958 * Clones from send_root are allowed, but only if the clone source
5959 * is behind the current send position. This is checked while searching
5960 * for possible clone sources.
5962 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5964 /* We do a bsearch later */
5965 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5966 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5968 sort_clone_roots = 1;
5970 ret = ensure_commit_roots_uptodate(sctx);
5974 current->journal_info = BTRFS_SEND_TRANS_STUB;
5975 ret = send_subvol(sctx);
5976 current->journal_info = NULL;
5980 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5981 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5984 ret = send_cmd(sctx);
5990 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5991 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5993 struct pending_dir_move *pm;
5995 n = rb_first(&sctx->pending_dir_moves);
5996 pm = rb_entry(n, struct pending_dir_move, node);
5997 while (!list_empty(&pm->list)) {
5998 struct pending_dir_move *pm2;
6000 pm2 = list_first_entry(&pm->list,
6001 struct pending_dir_move, list);
6002 free_pending_move(sctx, pm2);
6004 free_pending_move(sctx, pm);
6007 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
6008 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
6010 struct waiting_dir_move *dm;
6012 n = rb_first(&sctx->waiting_dir_moves);
6013 dm = rb_entry(n, struct waiting_dir_move, node);
6014 rb_erase(&dm->node, &sctx->waiting_dir_moves);
6018 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
6019 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
6021 struct orphan_dir_info *odi;
6023 n = rb_first(&sctx->orphan_dirs);
6024 odi = rb_entry(n, struct orphan_dir_info, node);
6025 free_orphan_dir_info(sctx, odi);
6028 if (sort_clone_roots) {
6029 for (i = 0; i < sctx->clone_roots_cnt; i++)
6030 btrfs_root_dec_send_in_progress(
6031 sctx->clone_roots[i].root);
6033 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
6034 btrfs_root_dec_send_in_progress(
6035 sctx->clone_roots[i].root);
6037 btrfs_root_dec_send_in_progress(send_root);
6039 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6040 btrfs_root_dec_send_in_progress(sctx->parent_root);
6043 vfree(clone_sources_tmp);
6046 if (sctx->send_filp)
6047 fput(sctx->send_filp);
6049 vfree(sctx->clone_roots);
6050 vfree(sctx->send_buf);
6051 vfree(sctx->read_buf);
6053 name_cache_free(sctx);