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 /* Just to check for bugs in backref resolving */
1166 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1168 u64 root = (u64)(uintptr_t)key;
1169 struct clone_root *cr = (struct clone_root *)elt;
1171 if (root < cr->root->objectid)
1173 if (root > cr->root->objectid)
1178 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1180 struct clone_root *cr1 = (struct clone_root *)e1;
1181 struct clone_root *cr2 = (struct clone_root *)e2;
1183 if (cr1->root->objectid < cr2->root->objectid)
1185 if (cr1->root->objectid > cr2->root->objectid)
1191 * Called for every backref that is found for the current extent.
1192 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1194 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1196 struct backref_ctx *bctx = ctx_;
1197 struct clone_root *found;
1201 /* First check if the root is in the list of accepted clone sources */
1202 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1203 bctx->sctx->clone_roots_cnt,
1204 sizeof(struct clone_root),
1205 __clone_root_cmp_bsearch);
1209 if (found->root == bctx->sctx->send_root &&
1210 ino == bctx->cur_objectid &&
1211 offset == bctx->cur_offset) {
1212 bctx->found_itself = 1;
1216 * There are inodes that have extents that lie behind its i_size. Don't
1217 * accept clones from these extents.
1219 ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1221 btrfs_release_path(bctx->path);
1225 if (offset + bctx->extent_len > i_size)
1229 * Make sure we don't consider clones from send_root that are
1230 * behind the current inode/offset.
1232 if (found->root == bctx->sctx->send_root) {
1234 * TODO for the moment we don't accept clones from the inode
1235 * that is currently send. We may change this when
1236 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1239 if (ino >= bctx->cur_objectid)
1242 if (ino > bctx->cur_objectid)
1244 if (offset + bctx->extent_len > bctx->cur_offset)
1250 found->found_refs++;
1251 if (ino < found->ino) {
1253 found->offset = offset;
1254 } else if (found->ino == ino) {
1256 * same extent found more then once in the same file.
1258 if (found->offset > offset + bctx->extent_len)
1259 found->offset = offset;
1266 * Given an inode, offset and extent item, it finds a good clone for a clone
1267 * instruction. Returns -ENOENT when none could be found. The function makes
1268 * sure that the returned clone is usable at the point where sending is at the
1269 * moment. This means, that no clones are accepted which lie behind the current
1272 * path must point to the extent item when called.
1274 static int find_extent_clone(struct send_ctx *sctx,
1275 struct btrfs_path *path,
1276 u64 ino, u64 data_offset,
1278 struct clone_root **found)
1285 u64 extent_item_pos;
1287 struct btrfs_file_extent_item *fi;
1288 struct extent_buffer *eb = path->nodes[0];
1289 struct backref_ctx *backref_ctx = NULL;
1290 struct clone_root *cur_clone_root;
1291 struct btrfs_key found_key;
1292 struct btrfs_path *tmp_path;
1296 tmp_path = alloc_path_for_send();
1300 /* We only use this path under the commit sem */
1301 tmp_path->need_commit_sem = 0;
1303 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1309 backref_ctx->path = tmp_path;
1311 if (data_offset >= ino_size) {
1313 * There may be extents that lie behind the file's size.
1314 * I at least had this in combination with snapshotting while
1315 * writing large files.
1321 fi = btrfs_item_ptr(eb, path->slots[0],
1322 struct btrfs_file_extent_item);
1323 extent_type = btrfs_file_extent_type(eb, fi);
1324 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1328 compressed = btrfs_file_extent_compression(eb, fi);
1330 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1331 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1332 if (disk_byte == 0) {
1336 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1338 down_read(&sctx->send_root->fs_info->commit_root_sem);
1339 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1340 &found_key, &flags);
1341 up_read(&sctx->send_root->fs_info->commit_root_sem);
1342 btrfs_release_path(tmp_path);
1346 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1352 * Setup the clone roots.
1354 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1355 cur_clone_root = sctx->clone_roots + i;
1356 cur_clone_root->ino = (u64)-1;
1357 cur_clone_root->offset = 0;
1358 cur_clone_root->found_refs = 0;
1361 backref_ctx->sctx = sctx;
1362 backref_ctx->found = 0;
1363 backref_ctx->cur_objectid = ino;
1364 backref_ctx->cur_offset = data_offset;
1365 backref_ctx->found_itself = 0;
1366 backref_ctx->extent_len = num_bytes;
1369 * The last extent of a file may be too large due to page alignment.
1370 * We need to adjust extent_len in this case so that the checks in
1371 * __iterate_backrefs work.
1373 if (data_offset + num_bytes >= ino_size)
1374 backref_ctx->extent_len = ino_size - data_offset;
1377 * Now collect all backrefs.
1379 if (compressed == BTRFS_COMPRESS_NONE)
1380 extent_item_pos = logical - found_key.objectid;
1382 extent_item_pos = 0;
1383 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1384 found_key.objectid, extent_item_pos, 1,
1385 __iterate_backrefs, backref_ctx);
1390 if (!backref_ctx->found_itself) {
1391 /* found a bug in backref code? */
1393 btrfs_err(sctx->send_root->fs_info, "did not find backref in "
1394 "send_root. inode=%llu, offset=%llu, "
1395 "disk_byte=%llu found extent=%llu",
1396 ino, data_offset, disk_byte, found_key.objectid);
1400 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1402 "num_bytes=%llu, logical=%llu\n",
1403 data_offset, ino, num_bytes, logical);
1405 if (!backref_ctx->found)
1406 verbose_printk("btrfs: no clones found\n");
1408 cur_clone_root = NULL;
1409 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1410 if (sctx->clone_roots[i].found_refs) {
1411 if (!cur_clone_root)
1412 cur_clone_root = sctx->clone_roots + i;
1413 else if (sctx->clone_roots[i].root == sctx->send_root)
1414 /* prefer clones from send_root over others */
1415 cur_clone_root = sctx->clone_roots + i;
1420 if (cur_clone_root) {
1421 if (compressed != BTRFS_COMPRESS_NONE) {
1423 * Offsets given by iterate_extent_inodes() are relative
1424 * to the start of the extent, we need to add logical
1425 * offset from the file extent item.
1426 * (See why at backref.c:check_extent_in_eb())
1428 cur_clone_root->offset += btrfs_file_extent_offset(eb,
1431 *found = cur_clone_root;
1438 btrfs_free_path(tmp_path);
1443 static int read_symlink(struct btrfs_root *root,
1445 struct fs_path *dest)
1448 struct btrfs_path *path;
1449 struct btrfs_key key;
1450 struct btrfs_file_extent_item *ei;
1456 path = alloc_path_for_send();
1461 key.type = BTRFS_EXTENT_DATA_KEY;
1463 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1468 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1469 struct btrfs_file_extent_item);
1470 type = btrfs_file_extent_type(path->nodes[0], ei);
1471 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1472 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1473 BUG_ON(compression);
1475 off = btrfs_file_extent_inline_start(ei);
1476 len = btrfs_file_extent_inline_len(path->nodes[0], path->slots[0], ei);
1478 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1481 btrfs_free_path(path);
1486 * Helper function to generate a file name that is unique in the root of
1487 * send_root and parent_root. This is used to generate names for orphan inodes.
1489 static int gen_unique_name(struct send_ctx *sctx,
1491 struct fs_path *dest)
1494 struct btrfs_path *path;
1495 struct btrfs_dir_item *di;
1500 path = alloc_path_for_send();
1505 len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu",
1507 ASSERT(len < sizeof(tmp));
1509 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1510 path, BTRFS_FIRST_FREE_OBJECTID,
1511 tmp, strlen(tmp), 0);
1512 btrfs_release_path(path);
1518 /* not unique, try again */
1523 if (!sctx->parent_root) {
1529 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1530 path, BTRFS_FIRST_FREE_OBJECTID,
1531 tmp, strlen(tmp), 0);
1532 btrfs_release_path(path);
1538 /* not unique, try again */
1546 ret = fs_path_add(dest, tmp, strlen(tmp));
1549 btrfs_free_path(path);
1554 inode_state_no_change,
1555 inode_state_will_create,
1556 inode_state_did_create,
1557 inode_state_will_delete,
1558 inode_state_did_delete,
1561 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1569 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1571 if (ret < 0 && ret != -ENOENT)
1575 if (!sctx->parent_root) {
1576 right_ret = -ENOENT;
1578 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1579 NULL, NULL, NULL, NULL);
1580 if (ret < 0 && ret != -ENOENT)
1585 if (!left_ret && !right_ret) {
1586 if (left_gen == gen && right_gen == gen) {
1587 ret = inode_state_no_change;
1588 } else if (left_gen == gen) {
1589 if (ino < sctx->send_progress)
1590 ret = inode_state_did_create;
1592 ret = inode_state_will_create;
1593 } else if (right_gen == gen) {
1594 if (ino < sctx->send_progress)
1595 ret = inode_state_did_delete;
1597 ret = inode_state_will_delete;
1601 } else if (!left_ret) {
1602 if (left_gen == gen) {
1603 if (ino < sctx->send_progress)
1604 ret = inode_state_did_create;
1606 ret = inode_state_will_create;
1610 } else if (!right_ret) {
1611 if (right_gen == gen) {
1612 if (ino < sctx->send_progress)
1613 ret = inode_state_did_delete;
1615 ret = inode_state_will_delete;
1627 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1631 ret = get_cur_inode_state(sctx, ino, gen);
1635 if (ret == inode_state_no_change ||
1636 ret == inode_state_did_create ||
1637 ret == inode_state_will_delete)
1647 * Helper function to lookup a dir item in a dir.
1649 static int lookup_dir_item_inode(struct btrfs_root *root,
1650 u64 dir, const char *name, int name_len,
1655 struct btrfs_dir_item *di;
1656 struct btrfs_key key;
1657 struct btrfs_path *path;
1659 path = alloc_path_for_send();
1663 di = btrfs_lookup_dir_item(NULL, root, path,
1664 dir, name, name_len, 0);
1673 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1674 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1678 *found_inode = key.objectid;
1679 *found_type = btrfs_dir_type(path->nodes[0], di);
1682 btrfs_free_path(path);
1687 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1688 * generation of the parent dir and the name of the dir entry.
1690 static int get_first_ref(struct btrfs_root *root, u64 ino,
1691 u64 *dir, u64 *dir_gen, struct fs_path *name)
1694 struct btrfs_key key;
1695 struct btrfs_key found_key;
1696 struct btrfs_path *path;
1700 path = alloc_path_for_send();
1705 key.type = BTRFS_INODE_REF_KEY;
1708 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1712 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1714 if (ret || found_key.objectid != ino ||
1715 (found_key.type != BTRFS_INODE_REF_KEY &&
1716 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1721 if (found_key.type == BTRFS_INODE_REF_KEY) {
1722 struct btrfs_inode_ref *iref;
1723 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1724 struct btrfs_inode_ref);
1725 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1726 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1727 (unsigned long)(iref + 1),
1729 parent_dir = found_key.offset;
1731 struct btrfs_inode_extref *extref;
1732 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1733 struct btrfs_inode_extref);
1734 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1735 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1736 (unsigned long)&extref->name, len);
1737 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1741 btrfs_release_path(path);
1744 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL,
1753 btrfs_free_path(path);
1757 static int is_first_ref(struct btrfs_root *root,
1759 const char *name, int name_len)
1762 struct fs_path *tmp_name;
1765 tmp_name = fs_path_alloc();
1769 ret = get_first_ref(root, ino, &tmp_dir, NULL, tmp_name);
1773 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1778 ret = !memcmp(tmp_name->start, name, name_len);
1781 fs_path_free(tmp_name);
1786 * Used by process_recorded_refs to determine if a new ref would overwrite an
1787 * already existing ref. In case it detects an overwrite, it returns the
1788 * inode/gen in who_ino/who_gen.
1789 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1790 * to make sure later references to the overwritten inode are possible.
1791 * Orphanizing is however only required for the first ref of an inode.
1792 * process_recorded_refs does an additional is_first_ref check to see if
1793 * orphanizing is really required.
1795 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1796 const char *name, int name_len,
1797 u64 *who_ino, u64 *who_gen)
1801 u64 other_inode = 0;
1804 if (!sctx->parent_root)
1807 ret = is_inode_existent(sctx, dir, dir_gen);
1812 * If we have a parent root we need to verify that the parent dir was
1813 * not delted and then re-created, if it was then we have no overwrite
1814 * and we can just unlink this entry.
1816 if (sctx->parent_root) {
1817 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL,
1819 if (ret < 0 && ret != -ENOENT)
1829 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1830 &other_inode, &other_type);
1831 if (ret < 0 && ret != -ENOENT)
1839 * Check if the overwritten ref was already processed. If yes, the ref
1840 * was already unlinked/moved, so we can safely assume that we will not
1841 * overwrite anything at this point in time.
1843 if (other_inode > sctx->send_progress) {
1844 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1845 who_gen, NULL, NULL, NULL, NULL);
1850 *who_ino = other_inode;
1860 * Checks if the ref was overwritten by an already processed inode. This is
1861 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1862 * thus the orphan name needs be used.
1863 * process_recorded_refs also uses it to avoid unlinking of refs that were
1866 static int did_overwrite_ref(struct send_ctx *sctx,
1867 u64 dir, u64 dir_gen,
1868 u64 ino, u64 ino_gen,
1869 const char *name, int name_len)
1876 if (!sctx->parent_root)
1879 ret = is_inode_existent(sctx, dir, dir_gen);
1883 /* check if the ref was overwritten by another ref */
1884 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1885 &ow_inode, &other_type);
1886 if (ret < 0 && ret != -ENOENT)
1889 /* was never and will never be overwritten */
1894 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1899 if (ow_inode == ino && gen == ino_gen) {
1905 * We know that it is or will be overwritten. Check this now.
1906 * The current inode being processed might have been the one that caused
1907 * inode 'ino' to be orphanized, therefore ow_inode can actually be the
1908 * same as sctx->send_progress.
1910 if (ow_inode <= sctx->send_progress)
1920 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1921 * that got overwritten. This is used by process_recorded_refs to determine
1922 * if it has to use the path as returned by get_cur_path or the orphan name.
1924 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1927 struct fs_path *name = NULL;
1931 if (!sctx->parent_root)
1934 name = fs_path_alloc();
1938 ret = get_first_ref(sctx->parent_root, ino, &dir, &dir_gen, name);
1942 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1943 name->start, fs_path_len(name));
1951 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1952 * so we need to do some special handling in case we have clashes. This function
1953 * takes care of this with the help of name_cache_entry::radix_list.
1954 * In case of error, nce is kfreed.
1956 static int name_cache_insert(struct send_ctx *sctx,
1957 struct name_cache_entry *nce)
1960 struct list_head *nce_head;
1962 nce_head = radix_tree_lookup(&sctx->name_cache,
1963 (unsigned long)nce->ino);
1965 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1970 INIT_LIST_HEAD(nce_head);
1972 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1979 list_add_tail(&nce->radix_list, nce_head);
1980 list_add_tail(&nce->list, &sctx->name_cache_list);
1981 sctx->name_cache_size++;
1986 static void name_cache_delete(struct send_ctx *sctx,
1987 struct name_cache_entry *nce)
1989 struct list_head *nce_head;
1991 nce_head = radix_tree_lookup(&sctx->name_cache,
1992 (unsigned long)nce->ino);
1994 btrfs_err(sctx->send_root->fs_info,
1995 "name_cache_delete lookup failed ino %llu cache size %d, leaking memory",
1996 nce->ino, sctx->name_cache_size);
1999 list_del(&nce->radix_list);
2000 list_del(&nce->list);
2001 sctx->name_cache_size--;
2004 * We may not get to the final release of nce_head if the lookup fails
2006 if (nce_head && list_empty(nce_head)) {
2007 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
2012 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
2015 struct list_head *nce_head;
2016 struct name_cache_entry *cur;
2018 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
2022 list_for_each_entry(cur, nce_head, radix_list) {
2023 if (cur->ino == ino && cur->gen == gen)
2030 * Removes the entry from the list and adds it back to the end. This marks the
2031 * entry as recently used so that name_cache_clean_unused does not remove it.
2033 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
2035 list_del(&nce->list);
2036 list_add_tail(&nce->list, &sctx->name_cache_list);
2040 * Remove some entries from the beginning of name_cache_list.
2042 static void name_cache_clean_unused(struct send_ctx *sctx)
2044 struct name_cache_entry *nce;
2046 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
2049 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
2050 nce = list_entry(sctx->name_cache_list.next,
2051 struct name_cache_entry, list);
2052 name_cache_delete(sctx, nce);
2057 static void name_cache_free(struct send_ctx *sctx)
2059 struct name_cache_entry *nce;
2061 while (!list_empty(&sctx->name_cache_list)) {
2062 nce = list_entry(sctx->name_cache_list.next,
2063 struct name_cache_entry, list);
2064 name_cache_delete(sctx, nce);
2070 * Used by get_cur_path for each ref up to the root.
2071 * Returns 0 if it succeeded.
2072 * Returns 1 if the inode is not existent or got overwritten. In that case, the
2073 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
2074 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
2075 * Returns <0 in case of error.
2077 static int __get_cur_name_and_parent(struct send_ctx *sctx,
2081 struct fs_path *dest)
2085 struct name_cache_entry *nce = NULL;
2088 * First check if we already did a call to this function with the same
2089 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
2090 * return the cached result.
2092 nce = name_cache_search(sctx, ino, gen);
2094 if (ino < sctx->send_progress && nce->need_later_update) {
2095 name_cache_delete(sctx, nce);
2099 name_cache_used(sctx, nce);
2100 *parent_ino = nce->parent_ino;
2101 *parent_gen = nce->parent_gen;
2102 ret = fs_path_add(dest, nce->name, nce->name_len);
2111 * If the inode is not existent yet, add the orphan name and return 1.
2112 * This should only happen for the parent dir that we determine in
2115 ret = is_inode_existent(sctx, ino, gen);
2120 ret = gen_unique_name(sctx, ino, gen, dest);
2128 * Depending on whether the inode was already processed or not, use
2129 * send_root or parent_root for ref lookup.
2131 if (ino < sctx->send_progress)
2132 ret = get_first_ref(sctx->send_root, ino,
2133 parent_ino, parent_gen, dest);
2135 ret = get_first_ref(sctx->parent_root, ino,
2136 parent_ino, parent_gen, dest);
2141 * Check if the ref was overwritten by an inode's ref that was processed
2142 * earlier. If yes, treat as orphan and return 1.
2144 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
2145 dest->start, dest->end - dest->start);
2149 fs_path_reset(dest);
2150 ret = gen_unique_name(sctx, ino, gen, dest);
2158 * Store the result of the lookup in the name cache.
2160 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2168 nce->parent_ino = *parent_ino;
2169 nce->parent_gen = *parent_gen;
2170 nce->name_len = fs_path_len(dest);
2172 strcpy(nce->name, dest->start);
2174 if (ino < sctx->send_progress)
2175 nce->need_later_update = 0;
2177 nce->need_later_update = 1;
2179 nce_ret = name_cache_insert(sctx, nce);
2182 name_cache_clean_unused(sctx);
2189 * Magic happens here. This function returns the first ref to an inode as it
2190 * would look like while receiving the stream at this point in time.
2191 * We walk the path up to the root. For every inode in between, we check if it
2192 * was already processed/sent. If yes, we continue with the parent as found
2193 * in send_root. If not, we continue with the parent as found in parent_root.
2194 * If we encounter an inode that was deleted at this point in time, we use the
2195 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2196 * that were not created yet and overwritten inodes/refs.
2198 * When do we have have orphan inodes:
2199 * 1. When an inode is freshly created and thus no valid refs are available yet
2200 * 2. When a directory lost all it's refs (deleted) but still has dir items
2201 * inside which were not processed yet (pending for move/delete). If anyone
2202 * tried to get the path to the dir items, it would get a path inside that
2204 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2205 * of an unprocessed inode. If in that case the first ref would be
2206 * overwritten, the overwritten inode gets "orphanized". Later when we
2207 * process this overwritten inode, it is restored at a new place by moving
2210 * sctx->send_progress tells this function at which point in time receiving
2213 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2214 struct fs_path *dest)
2217 struct fs_path *name = NULL;
2218 u64 parent_inode = 0;
2222 name = fs_path_alloc();
2229 fs_path_reset(dest);
2231 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2232 struct waiting_dir_move *wdm;
2234 fs_path_reset(name);
2236 if (is_waiting_for_rm(sctx, ino)) {
2237 ret = gen_unique_name(sctx, ino, gen, name);
2240 ret = fs_path_add_path(dest, name);
2244 wdm = get_waiting_dir_move(sctx, ino);
2245 if (wdm && wdm->orphanized) {
2246 ret = gen_unique_name(sctx, ino, gen, name);
2249 ret = get_first_ref(sctx->parent_root, ino,
2250 &parent_inode, &parent_gen, name);
2252 ret = __get_cur_name_and_parent(sctx, ino, gen,
2262 ret = fs_path_add_path(dest, name);
2273 fs_path_unreverse(dest);
2278 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2280 static int send_subvol_begin(struct send_ctx *sctx)
2283 struct btrfs_root *send_root = sctx->send_root;
2284 struct btrfs_root *parent_root = sctx->parent_root;
2285 struct btrfs_path *path;
2286 struct btrfs_key key;
2287 struct btrfs_root_ref *ref;
2288 struct extent_buffer *leaf;
2292 path = btrfs_alloc_path();
2296 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2298 btrfs_free_path(path);
2302 key.objectid = send_root->objectid;
2303 key.type = BTRFS_ROOT_BACKREF_KEY;
2306 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2315 leaf = path->nodes[0];
2316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2317 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2318 key.objectid != send_root->objectid) {
2322 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2323 namelen = btrfs_root_ref_name_len(leaf, ref);
2324 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2325 btrfs_release_path(path);
2328 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2332 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2337 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2338 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2339 sctx->send_root->root_item.uuid);
2340 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2341 le64_to_cpu(sctx->send_root->root_item.ctransid));
2343 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2344 sctx->parent_root->root_item.uuid);
2345 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2346 le64_to_cpu(sctx->parent_root->root_item.ctransid));
2349 ret = send_cmd(sctx);
2353 btrfs_free_path(path);
2358 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2363 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2365 p = fs_path_alloc();
2369 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2373 ret = get_cur_path(sctx, ino, gen, p);
2376 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2377 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2379 ret = send_cmd(sctx);
2387 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2392 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2394 p = fs_path_alloc();
2398 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2402 ret = get_cur_path(sctx, ino, gen, p);
2405 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2406 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2408 ret = send_cmd(sctx);
2416 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2421 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2423 p = fs_path_alloc();
2427 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2431 ret = get_cur_path(sctx, ino, gen, p);
2434 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2435 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2436 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2438 ret = send_cmd(sctx);
2446 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2449 struct fs_path *p = NULL;
2450 struct btrfs_inode_item *ii;
2451 struct btrfs_path *path = NULL;
2452 struct extent_buffer *eb;
2453 struct btrfs_key key;
2456 verbose_printk("btrfs: send_utimes %llu\n", ino);
2458 p = fs_path_alloc();
2462 path = alloc_path_for_send();
2469 key.type = BTRFS_INODE_ITEM_KEY;
2471 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2475 eb = path->nodes[0];
2476 slot = path->slots[0];
2477 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2479 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2483 ret = get_cur_path(sctx, ino, gen, p);
2486 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2487 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime);
2488 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime);
2489 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime);
2490 /* TODO Add otime support when the otime patches get into upstream */
2492 ret = send_cmd(sctx);
2497 btrfs_free_path(path);
2502 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2503 * a valid path yet because we did not process the refs yet. So, the inode
2504 * is created as orphan.
2506 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2515 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2517 p = fs_path_alloc();
2521 if (ino != sctx->cur_ino) {
2522 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode,
2527 gen = sctx->cur_inode_gen;
2528 mode = sctx->cur_inode_mode;
2529 rdev = sctx->cur_inode_rdev;
2532 if (S_ISREG(mode)) {
2533 cmd = BTRFS_SEND_C_MKFILE;
2534 } else if (S_ISDIR(mode)) {
2535 cmd = BTRFS_SEND_C_MKDIR;
2536 } else if (S_ISLNK(mode)) {
2537 cmd = BTRFS_SEND_C_SYMLINK;
2538 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2539 cmd = BTRFS_SEND_C_MKNOD;
2540 } else if (S_ISFIFO(mode)) {
2541 cmd = BTRFS_SEND_C_MKFIFO;
2542 } else if (S_ISSOCK(mode)) {
2543 cmd = BTRFS_SEND_C_MKSOCK;
2545 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2546 (int)(mode & S_IFMT));
2551 ret = begin_cmd(sctx, cmd);
2555 ret = gen_unique_name(sctx, ino, gen, p);
2559 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2560 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2562 if (S_ISLNK(mode)) {
2564 ret = read_symlink(sctx->send_root, ino, p);
2567 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2568 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2569 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2570 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2571 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2574 ret = send_cmd(sctx);
2586 * We need some special handling for inodes that get processed before the parent
2587 * directory got created. See process_recorded_refs for details.
2588 * This function does the check if we already created the dir out of order.
2590 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2593 struct btrfs_path *path = NULL;
2594 struct btrfs_key key;
2595 struct btrfs_key found_key;
2596 struct btrfs_key di_key;
2597 struct extent_buffer *eb;
2598 struct btrfs_dir_item *di;
2601 path = alloc_path_for_send();
2608 key.type = BTRFS_DIR_INDEX_KEY;
2610 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2615 eb = path->nodes[0];
2616 slot = path->slots[0];
2617 if (slot >= btrfs_header_nritems(eb)) {
2618 ret = btrfs_next_leaf(sctx->send_root, path);
2621 } else if (ret > 0) {
2628 btrfs_item_key_to_cpu(eb, &found_key, slot);
2629 if (found_key.objectid != key.objectid ||
2630 found_key.type != key.type) {
2635 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2636 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2638 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2639 di_key.objectid < sctx->send_progress) {
2648 btrfs_free_path(path);
2653 * Only creates the inode if it is:
2654 * 1. Not a directory
2655 * 2. Or a directory which was not created already due to out of order
2656 * directories. See did_create_dir and process_recorded_refs for details.
2658 static int send_create_inode_if_needed(struct send_ctx *sctx)
2662 if (S_ISDIR(sctx->cur_inode_mode)) {
2663 ret = did_create_dir(sctx, sctx->cur_ino);
2672 ret = send_create_inode(sctx, sctx->cur_ino);
2680 struct recorded_ref {
2681 struct list_head list;
2684 struct fs_path *full_path;
2692 * We need to process new refs before deleted refs, but compare_tree gives us
2693 * everything mixed. So we first record all refs and later process them.
2694 * This function is a helper to record one ref.
2696 static int __record_ref(struct list_head *head, u64 dir,
2697 u64 dir_gen, struct fs_path *path)
2699 struct recorded_ref *ref;
2701 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2706 ref->dir_gen = dir_gen;
2707 ref->full_path = path;
2709 ref->name = (char *)kbasename(ref->full_path->start);
2710 ref->name_len = ref->full_path->end - ref->name;
2711 ref->dir_path = ref->full_path->start;
2712 if (ref->name == ref->full_path->start)
2713 ref->dir_path_len = 0;
2715 ref->dir_path_len = ref->full_path->end -
2716 ref->full_path->start - 1 - ref->name_len;
2718 list_add_tail(&ref->list, head);
2722 static int dup_ref(struct recorded_ref *ref, struct list_head *list)
2724 struct recorded_ref *new;
2726 new = kmalloc(sizeof(*ref), GFP_NOFS);
2730 new->dir = ref->dir;
2731 new->dir_gen = ref->dir_gen;
2732 new->full_path = NULL;
2733 INIT_LIST_HEAD(&new->list);
2734 list_add_tail(&new->list, list);
2738 static void __free_recorded_refs(struct list_head *head)
2740 struct recorded_ref *cur;
2742 while (!list_empty(head)) {
2743 cur = list_entry(head->next, struct recorded_ref, list);
2744 fs_path_free(cur->full_path);
2745 list_del(&cur->list);
2750 static void free_recorded_refs(struct send_ctx *sctx)
2752 __free_recorded_refs(&sctx->new_refs);
2753 __free_recorded_refs(&sctx->deleted_refs);
2757 * Renames/moves a file/dir to its orphan name. Used when the first
2758 * ref of an unprocessed inode gets overwritten and for all non empty
2761 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2762 struct fs_path *path)
2765 struct fs_path *orphan;
2767 orphan = fs_path_alloc();
2771 ret = gen_unique_name(sctx, ino, gen, orphan);
2775 ret = send_rename(sctx, path, orphan);
2778 fs_path_free(orphan);
2782 static struct orphan_dir_info *
2783 add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2785 struct rb_node **p = &sctx->orphan_dirs.rb_node;
2786 struct rb_node *parent = NULL;
2787 struct orphan_dir_info *entry, *odi;
2789 odi = kmalloc(sizeof(*odi), GFP_NOFS);
2791 return ERR_PTR(-ENOMEM);
2797 entry = rb_entry(parent, struct orphan_dir_info, node);
2798 if (dir_ino < entry->ino) {
2800 } else if (dir_ino > entry->ino) {
2801 p = &(*p)->rb_right;
2808 rb_link_node(&odi->node, parent, p);
2809 rb_insert_color(&odi->node, &sctx->orphan_dirs);
2813 static struct orphan_dir_info *
2814 get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino)
2816 struct rb_node *n = sctx->orphan_dirs.rb_node;
2817 struct orphan_dir_info *entry;
2820 entry = rb_entry(n, struct orphan_dir_info, node);
2821 if (dir_ino < entry->ino)
2823 else if (dir_ino > entry->ino)
2831 static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino)
2833 struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino);
2838 static void free_orphan_dir_info(struct send_ctx *sctx,
2839 struct orphan_dir_info *odi)
2843 rb_erase(&odi->node, &sctx->orphan_dirs);
2848 * Returns 1 if a directory can be removed at this point in time.
2849 * We check this by iterating all dir items and checking if the inode behind
2850 * the dir item was already processed.
2852 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen,
2856 struct btrfs_root *root = sctx->parent_root;
2857 struct btrfs_path *path;
2858 struct btrfs_key key;
2859 struct btrfs_key found_key;
2860 struct btrfs_key loc;
2861 struct btrfs_dir_item *di;
2864 * Don't try to rmdir the top/root subvolume dir.
2866 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2869 path = alloc_path_for_send();
2874 key.type = BTRFS_DIR_INDEX_KEY;
2876 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2881 struct waiting_dir_move *dm;
2883 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2884 ret = btrfs_next_leaf(root, path);
2891 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2893 if (found_key.objectid != key.objectid ||
2894 found_key.type != key.type)
2897 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2898 struct btrfs_dir_item);
2899 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2901 dm = get_waiting_dir_move(sctx, loc.objectid);
2903 struct orphan_dir_info *odi;
2905 odi = add_orphan_dir_info(sctx, dir);
2911 dm->rmdir_ino = dir;
2916 if (loc.objectid > send_progress) {
2927 btrfs_free_path(path);
2931 static int is_waiting_for_move(struct send_ctx *sctx, u64 ino)
2933 struct waiting_dir_move *entry = get_waiting_dir_move(sctx, ino);
2935 return entry != NULL;
2938 static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized)
2940 struct rb_node **p = &sctx->waiting_dir_moves.rb_node;
2941 struct rb_node *parent = NULL;
2942 struct waiting_dir_move *entry, *dm;
2944 dm = kmalloc(sizeof(*dm), GFP_NOFS);
2949 dm->orphanized = orphanized;
2953 entry = rb_entry(parent, struct waiting_dir_move, node);
2954 if (ino < entry->ino) {
2956 } else if (ino > entry->ino) {
2957 p = &(*p)->rb_right;
2964 rb_link_node(&dm->node, parent, p);
2965 rb_insert_color(&dm->node, &sctx->waiting_dir_moves);
2969 static struct waiting_dir_move *
2970 get_waiting_dir_move(struct send_ctx *sctx, u64 ino)
2972 struct rb_node *n = sctx->waiting_dir_moves.rb_node;
2973 struct waiting_dir_move *entry;
2976 entry = rb_entry(n, struct waiting_dir_move, node);
2977 if (ino < entry->ino)
2979 else if (ino > entry->ino)
2987 static void free_waiting_dir_move(struct send_ctx *sctx,
2988 struct waiting_dir_move *dm)
2992 rb_erase(&dm->node, &sctx->waiting_dir_moves);
2996 static int add_pending_dir_move(struct send_ctx *sctx,
3000 struct list_head *new_refs,
3001 struct list_head *deleted_refs,
3002 const bool is_orphan)
3004 struct rb_node **p = &sctx->pending_dir_moves.rb_node;
3005 struct rb_node *parent = NULL;
3006 struct pending_dir_move *entry = NULL, *pm;
3007 struct recorded_ref *cur;
3011 pm = kmalloc(sizeof(*pm), GFP_NOFS);
3014 pm->parent_ino = parent_ino;
3017 pm->is_orphan = is_orphan;
3018 INIT_LIST_HEAD(&pm->list);
3019 INIT_LIST_HEAD(&pm->update_refs);
3020 RB_CLEAR_NODE(&pm->node);
3024 entry = rb_entry(parent, struct pending_dir_move, node);
3025 if (parent_ino < entry->parent_ino) {
3027 } else if (parent_ino > entry->parent_ino) {
3028 p = &(*p)->rb_right;
3035 list_for_each_entry(cur, deleted_refs, list) {
3036 ret = dup_ref(cur, &pm->update_refs);
3040 list_for_each_entry(cur, new_refs, list) {
3041 ret = dup_ref(cur, &pm->update_refs);
3046 ret = add_waiting_dir_move(sctx, pm->ino, is_orphan);
3051 list_add_tail(&pm->list, &entry->list);
3053 rb_link_node(&pm->node, parent, p);
3054 rb_insert_color(&pm->node, &sctx->pending_dir_moves);
3059 __free_recorded_refs(&pm->update_refs);
3065 static struct pending_dir_move *get_pending_dir_moves(struct send_ctx *sctx,
3068 struct rb_node *n = sctx->pending_dir_moves.rb_node;
3069 struct pending_dir_move *entry;
3072 entry = rb_entry(n, struct pending_dir_move, node);
3073 if (parent_ino < entry->parent_ino)
3075 else if (parent_ino > entry->parent_ino)
3083 static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm)
3085 struct fs_path *from_path = NULL;
3086 struct fs_path *to_path = NULL;
3087 struct fs_path *name = NULL;
3088 u64 orig_progress = sctx->send_progress;
3089 struct recorded_ref *cur;
3090 u64 parent_ino, parent_gen;
3091 struct waiting_dir_move *dm = NULL;
3095 name = fs_path_alloc();
3096 from_path = fs_path_alloc();
3097 if (!name || !from_path) {
3102 dm = get_waiting_dir_move(sctx, pm->ino);
3104 rmdir_ino = dm->rmdir_ino;
3105 free_waiting_dir_move(sctx, dm);
3107 if (pm->is_orphan) {
3108 ret = gen_unique_name(sctx, pm->ino,
3109 pm->gen, from_path);
3111 ret = get_first_ref(sctx->parent_root, pm->ino,
3112 &parent_ino, &parent_gen, name);
3115 ret = get_cur_path(sctx, parent_ino, parent_gen,
3119 ret = fs_path_add_path(from_path, name);
3124 sctx->send_progress = sctx->cur_ino + 1;
3125 fs_path_reset(name);
3128 ret = get_cur_path(sctx, pm->ino, pm->gen, to_path);
3132 ret = send_rename(sctx, from_path, to_path);
3137 struct orphan_dir_info *odi;
3139 odi = get_orphan_dir_info(sctx, rmdir_ino);
3141 /* already deleted */
3144 ret = can_rmdir(sctx, rmdir_ino, odi->gen, sctx->cur_ino + 1);
3150 name = fs_path_alloc();
3155 ret = get_cur_path(sctx, rmdir_ino, odi->gen, name);
3158 ret = send_rmdir(sctx, name);
3161 free_orphan_dir_info(sctx, odi);
3165 ret = send_utimes(sctx, pm->ino, pm->gen);
3170 * After rename/move, need to update the utimes of both new parent(s)
3171 * and old parent(s).
3173 list_for_each_entry(cur, &pm->update_refs, list) {
3174 if (cur->dir == rmdir_ino)
3176 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3183 fs_path_free(from_path);
3184 fs_path_free(to_path);
3185 sctx->send_progress = orig_progress;
3190 static void free_pending_move(struct send_ctx *sctx, struct pending_dir_move *m)
3192 if (!list_empty(&m->list))
3194 if (!RB_EMPTY_NODE(&m->node))
3195 rb_erase(&m->node, &sctx->pending_dir_moves);
3196 __free_recorded_refs(&m->update_refs);
3200 static void tail_append_pending_moves(struct pending_dir_move *moves,
3201 struct list_head *stack)
3203 if (list_empty(&moves->list)) {
3204 list_add_tail(&moves->list, stack);
3207 list_splice_init(&moves->list, &list);
3208 list_add_tail(&moves->list, stack);
3209 list_splice_tail(&list, stack);
3213 static int apply_children_dir_moves(struct send_ctx *sctx)
3215 struct pending_dir_move *pm;
3216 struct list_head stack;
3217 u64 parent_ino = sctx->cur_ino;
3220 pm = get_pending_dir_moves(sctx, parent_ino);
3224 INIT_LIST_HEAD(&stack);
3225 tail_append_pending_moves(pm, &stack);
3227 while (!list_empty(&stack)) {
3228 pm = list_first_entry(&stack, struct pending_dir_move, list);
3229 parent_ino = pm->ino;
3230 ret = apply_dir_move(sctx, pm);
3231 free_pending_move(sctx, pm);
3234 pm = get_pending_dir_moves(sctx, parent_ino);
3236 tail_append_pending_moves(pm, &stack);
3241 while (!list_empty(&stack)) {
3242 pm = list_first_entry(&stack, struct pending_dir_move, list);
3243 free_pending_move(sctx, pm);
3249 * We might need to delay a directory rename even when no ancestor directory
3250 * (in the send root) with a higher inode number than ours (sctx->cur_ino) was
3251 * renamed. This happens when we rename a directory to the old name (the name
3252 * in the parent root) of some other unrelated directory that got its rename
3253 * delayed due to some ancestor with higher number that got renamed.
3259 * |---- a/ (ino 257)
3260 * | |---- file (ino 260)
3262 * |---- b/ (ino 258)
3263 * |---- c/ (ino 259)
3267 * |---- a/ (ino 258)
3268 * |---- x/ (ino 259)
3269 * |---- y/ (ino 257)
3270 * |----- file (ino 260)
3272 * Here we can not rename 258 from 'b' to 'a' without the rename of inode 257
3273 * from 'a' to 'x/y' happening first, which in turn depends on the rename of
3274 * inode 259 from 'c' to 'x'. So the order of rename commands the send stream
3277 * 1 - rename 259 from 'c' to 'x'
3278 * 2 - rename 257 from 'a' to 'x/y'
3279 * 3 - rename 258 from 'b' to 'a'
3281 * Returns 1 if the rename of sctx->cur_ino needs to be delayed, 0 if it can
3282 * be done right away and < 0 on error.
3284 static int wait_for_dest_dir_move(struct send_ctx *sctx,
3285 struct recorded_ref *parent_ref,
3286 const bool is_orphan)
3288 struct btrfs_path *path;
3289 struct btrfs_key key;
3290 struct btrfs_key di_key;
3291 struct btrfs_dir_item *di;
3296 if (RB_EMPTY_ROOT(&sctx->waiting_dir_moves))
3299 path = alloc_path_for_send();
3303 key.objectid = parent_ref->dir;
3304 key.type = BTRFS_DIR_ITEM_KEY;
3305 key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len);
3307 ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0);
3310 } else if (ret > 0) {
3315 di = btrfs_match_dir_item_name(sctx->parent_root, path,
3316 parent_ref->name, parent_ref->name_len);
3322 * di_key.objectid has the number of the inode that has a dentry in the
3323 * parent directory with the same name that sctx->cur_ino is being
3324 * renamed to. We need to check if that inode is in the send root as
3325 * well and if it is currently marked as an inode with a pending rename,
3326 * if it is, we need to delay the rename of sctx->cur_ino as well, so
3327 * that it happens after that other inode is renamed.
3329 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key);
3330 if (di_key.type != BTRFS_INODE_ITEM_KEY) {
3335 ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL,
3336 &left_gen, NULL, NULL, NULL, NULL);
3339 ret = get_inode_info(sctx->send_root, di_key.objectid, NULL,
3340 &right_gen, NULL, NULL, NULL, NULL);
3347 /* Different inode, no need to delay the rename of sctx->cur_ino */
3348 if (right_gen != left_gen) {
3353 if (is_waiting_for_move(sctx, di_key.objectid)) {
3354 ret = add_pending_dir_move(sctx,
3356 sctx->cur_inode_gen,
3359 &sctx->deleted_refs,
3365 btrfs_free_path(path);
3370 * Check if ino ino1 is an ancestor of inode ino2 in the given root.
3371 * Return 1 if true, 0 if false and < 0 on error.
3373 static int is_ancestor(struct btrfs_root *root,
3377 struct fs_path *fs_path)
3381 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3386 fs_path_reset(fs_path);
3387 ret = get_first_ref(root, ino, &parent, &parent_gen, fs_path);
3389 if (ret == -ENOENT && ino == ino2)
3394 return parent_gen == ino1_gen ? 1 : 0;
3400 static int wait_for_parent_move(struct send_ctx *sctx,
3401 struct recorded_ref *parent_ref,
3402 const bool is_orphan)
3405 u64 ino = parent_ref->dir;
3406 u64 parent_ino_before, parent_ino_after;
3407 struct fs_path *path_before = NULL;
3408 struct fs_path *path_after = NULL;
3411 path_after = fs_path_alloc();
3412 path_before = fs_path_alloc();
3413 if (!path_after || !path_before) {
3419 * Our current directory inode may not yet be renamed/moved because some
3420 * ancestor (immediate or not) has to be renamed/moved first. So find if
3421 * such ancestor exists and make sure our own rename/move happens after
3422 * that ancestor is processed to avoid path build infinite loops (done
3423 * at get_cur_path()).
3425 while (ino > BTRFS_FIRST_FREE_OBJECTID) {
3426 if (is_waiting_for_move(sctx, ino)) {
3428 * If the current inode is an ancestor of ino in the
3429 * parent root, we need to delay the rename of the
3430 * current inode, otherwise don't delayed the rename
3431 * because we can end up with a circular dependency
3432 * of renames, resulting in some directories never
3433 * getting the respective rename operations issued in
3434 * the send stream or getting into infinite path build
3437 ret = is_ancestor(sctx->parent_root,
3438 sctx->cur_ino, sctx->cur_inode_gen,
3443 fs_path_reset(path_before);
3444 fs_path_reset(path_after);
3446 ret = get_first_ref(sctx->send_root, ino, &parent_ino_after,
3450 ret = get_first_ref(sctx->parent_root, ino, &parent_ino_before,
3452 if (ret < 0 && ret != -ENOENT) {
3454 } else if (ret == -ENOENT) {
3459 len1 = fs_path_len(path_before);
3460 len2 = fs_path_len(path_after);
3461 if (ino > sctx->cur_ino &&
3462 (parent_ino_before != parent_ino_after || len1 != len2 ||
3463 memcmp(path_before->start, path_after->start, len1))) {
3467 ino = parent_ino_after;
3471 fs_path_free(path_before);
3472 fs_path_free(path_after);
3475 ret = add_pending_dir_move(sctx,
3477 sctx->cur_inode_gen,
3480 &sctx->deleted_refs,
3490 * This does all the move/link/unlink/rmdir magic.
3492 static int process_recorded_refs(struct send_ctx *sctx, int *pending_move)
3495 struct recorded_ref *cur;
3496 struct recorded_ref *cur2;
3497 struct list_head check_dirs;
3498 struct fs_path *valid_path = NULL;
3501 int did_overwrite = 0;
3503 u64 last_dir_ino_rm = 0;
3504 bool can_rename = true;
3506 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
3509 * This should never happen as the root dir always has the same ref
3510 * which is always '..'
3512 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
3513 INIT_LIST_HEAD(&check_dirs);
3515 valid_path = fs_path_alloc();
3522 * First, check if the first ref of the current inode was overwritten
3523 * before. If yes, we know that the current inode was already orphanized
3524 * and thus use the orphan name. If not, we can use get_cur_path to
3525 * get the path of the first ref as it would like while receiving at
3526 * this point in time.
3527 * New inodes are always orphan at the beginning, so force to use the
3528 * orphan name in this case.
3529 * The first ref is stored in valid_path and will be updated if it
3530 * gets moved around.
3532 if (!sctx->cur_inode_new) {
3533 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
3534 sctx->cur_inode_gen);
3540 if (sctx->cur_inode_new || did_overwrite) {
3541 ret = gen_unique_name(sctx, sctx->cur_ino,
3542 sctx->cur_inode_gen, valid_path);
3547 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3553 list_for_each_entry(cur, &sctx->new_refs, list) {
3555 * We may have refs where the parent directory does not exist
3556 * yet. This happens if the parent directories inum is higher
3557 * the the current inum. To handle this case, we create the
3558 * parent directory out of order. But we need to check if this
3559 * did already happen before due to other refs in the same dir.
3561 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3564 if (ret == inode_state_will_create) {
3567 * First check if any of the current inodes refs did
3568 * already create the dir.
3570 list_for_each_entry(cur2, &sctx->new_refs, list) {
3573 if (cur2->dir == cur->dir) {
3580 * If that did not happen, check if a previous inode
3581 * did already create the dir.
3584 ret = did_create_dir(sctx, cur->dir);
3588 ret = send_create_inode(sctx, cur->dir);
3595 * Check if this new ref would overwrite the first ref of
3596 * another unprocessed inode. If yes, orphanize the
3597 * overwritten inode. If we find an overwritten ref that is
3598 * not the first ref, simply unlink it.
3600 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3601 cur->name, cur->name_len,
3602 &ow_inode, &ow_gen);
3606 ret = is_first_ref(sctx->parent_root,
3607 ow_inode, cur->dir, cur->name,
3612 struct name_cache_entry *nce;
3614 ret = orphanize_inode(sctx, ow_inode, ow_gen,
3619 * Make sure we clear our orphanized inode's
3620 * name from the name cache. This is because the
3621 * inode ow_inode might be an ancestor of some
3622 * other inode that will be orphanized as well
3623 * later and has an inode number greater than
3624 * sctx->send_progress. We need to prevent
3625 * future name lookups from using the old name
3626 * and get instead the orphan name.
3628 nce = name_cache_search(sctx, ow_inode, ow_gen);
3630 name_cache_delete(sctx, nce);
3634 ret = send_unlink(sctx, cur->full_path);
3640 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) {
3641 ret = wait_for_dest_dir_move(sctx, cur, is_orphan);
3650 if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root &&
3652 ret = wait_for_parent_move(sctx, cur, is_orphan);
3662 * link/move the ref to the new place. If we have an orphan
3663 * inode, move it and update valid_path. If not, link or move
3664 * it depending on the inode mode.
3666 if (is_orphan && can_rename) {
3667 ret = send_rename(sctx, valid_path, cur->full_path);
3671 ret = fs_path_copy(valid_path, cur->full_path);
3674 } else if (can_rename) {
3675 if (S_ISDIR(sctx->cur_inode_mode)) {
3677 * Dirs can't be linked, so move it. For moved
3678 * dirs, we always have one new and one deleted
3679 * ref. The deleted ref is ignored later.
3681 ret = send_rename(sctx, valid_path,
3684 ret = fs_path_copy(valid_path,
3689 ret = send_link(sctx, cur->full_path,
3695 ret = dup_ref(cur, &check_dirs);
3700 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
3702 * Check if we can already rmdir the directory. If not,
3703 * orphanize it. For every dir item inside that gets deleted
3704 * later, we do this check again and rmdir it then if possible.
3705 * See the use of check_dirs for more details.
3707 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen,
3712 ret = send_rmdir(sctx, valid_path);
3715 } else if (!is_orphan) {
3716 ret = orphanize_inode(sctx, sctx->cur_ino,
3717 sctx->cur_inode_gen, valid_path);
3723 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3724 ret = dup_ref(cur, &check_dirs);
3728 } else if (S_ISDIR(sctx->cur_inode_mode) &&
3729 !list_empty(&sctx->deleted_refs)) {
3731 * We have a moved dir. Add the old parent to check_dirs
3733 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
3735 ret = dup_ref(cur, &check_dirs);
3738 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
3740 * We have a non dir inode. Go through all deleted refs and
3741 * unlink them if they were not already overwritten by other
3744 list_for_each_entry(cur, &sctx->deleted_refs, list) {
3745 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
3746 sctx->cur_ino, sctx->cur_inode_gen,
3747 cur->name, cur->name_len);
3751 ret = send_unlink(sctx, cur->full_path);
3755 ret = dup_ref(cur, &check_dirs);
3760 * If the inode is still orphan, unlink the orphan. This may
3761 * happen when a previous inode did overwrite the first ref
3762 * of this inode and no new refs were added for the current
3763 * inode. Unlinking does not mean that the inode is deleted in
3764 * all cases. There may still be links to this inode in other
3768 ret = send_unlink(sctx, valid_path);
3775 * We did collect all parent dirs where cur_inode was once located. We
3776 * now go through all these dirs and check if they are pending for
3777 * deletion and if it's finally possible to perform the rmdir now.
3778 * We also update the inode stats of the parent dirs here.
3780 list_for_each_entry(cur, &check_dirs, list) {
3782 * In case we had refs into dirs that were not processed yet,
3783 * we don't need to do the utime and rmdir logic for these dirs.
3784 * The dir will be processed later.
3786 if (cur->dir > sctx->cur_ino)
3789 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3793 if (ret == inode_state_did_create ||
3794 ret == inode_state_no_change) {
3795 /* TODO delayed utimes */
3796 ret = send_utimes(sctx, cur->dir, cur->dir_gen);
3799 } else if (ret == inode_state_did_delete &&
3800 cur->dir != last_dir_ino_rm) {
3801 ret = can_rmdir(sctx, cur->dir, cur->dir_gen,
3806 ret = get_cur_path(sctx, cur->dir,
3807 cur->dir_gen, valid_path);
3810 ret = send_rmdir(sctx, valid_path);
3813 last_dir_ino_rm = cur->dir;
3821 __free_recorded_refs(&check_dirs);
3822 free_recorded_refs(sctx);
3823 fs_path_free(valid_path);
3827 static int record_ref(struct btrfs_root *root, int num, u64 dir, int index,
3828 struct fs_path *name, void *ctx, struct list_head *refs)
3831 struct send_ctx *sctx = ctx;
3835 p = fs_path_alloc();
3839 ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL,
3844 ret = get_cur_path(sctx, dir, gen, p);
3847 ret = fs_path_add_path(p, name);
3851 ret = __record_ref(refs, dir, gen, p);
3859 static int __record_new_ref(int num, u64 dir, int index,
3860 struct fs_path *name,
3863 struct send_ctx *sctx = ctx;
3864 return record_ref(sctx->send_root, num, dir, index, name,
3865 ctx, &sctx->new_refs);
3869 static int __record_deleted_ref(int num, u64 dir, int index,
3870 struct fs_path *name,
3873 struct send_ctx *sctx = ctx;
3874 return record_ref(sctx->parent_root, num, dir, index, name,
3875 ctx, &sctx->deleted_refs);
3878 static int record_new_ref(struct send_ctx *sctx)
3882 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
3883 sctx->cmp_key, 0, __record_new_ref, sctx);
3892 static int record_deleted_ref(struct send_ctx *sctx)
3896 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
3897 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3906 struct find_ref_ctx {
3909 struct btrfs_root *root;
3910 struct fs_path *name;
3914 static int __find_iref(int num, u64 dir, int index,
3915 struct fs_path *name,
3918 struct find_ref_ctx *ctx = ctx_;
3922 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3923 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3925 * To avoid doing extra lookups we'll only do this if everything
3928 ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL,
3932 if (dir_gen != ctx->dir_gen)
3934 ctx->found_idx = num;
3940 static int find_iref(struct btrfs_root *root,
3941 struct btrfs_path *path,
3942 struct btrfs_key *key,
3943 u64 dir, u64 dir_gen, struct fs_path *name)
3946 struct find_ref_ctx ctx;
3950 ctx.dir_gen = dir_gen;
3954 ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx);
3958 if (ctx.found_idx == -1)
3961 return ctx.found_idx;
3964 static int __record_changed_new_ref(int num, u64 dir, int index,
3965 struct fs_path *name,
3970 struct send_ctx *sctx = ctx;
3972 ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL,
3977 ret = find_iref(sctx->parent_root, sctx->right_path,
3978 sctx->cmp_key, dir, dir_gen, name);
3980 ret = __record_new_ref(num, dir, index, name, sctx);
3987 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3988 struct fs_path *name,
3993 struct send_ctx *sctx = ctx;
3995 ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL,
4000 ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key,
4001 dir, dir_gen, name);
4003 ret = __record_deleted_ref(num, dir, index, name, sctx);
4010 static int record_changed_ref(struct send_ctx *sctx)
4014 ret = iterate_inode_ref(sctx->send_root, sctx->left_path,
4015 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
4018 ret = iterate_inode_ref(sctx->parent_root, sctx->right_path,
4019 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
4029 * Record and process all refs at once. Needed when an inode changes the
4030 * generation number, which means that it was deleted and recreated.
4032 static int process_all_refs(struct send_ctx *sctx,
4033 enum btrfs_compare_tree_result cmd)
4036 struct btrfs_root *root;
4037 struct btrfs_path *path;
4038 struct btrfs_key key;
4039 struct btrfs_key found_key;
4040 struct extent_buffer *eb;
4042 iterate_inode_ref_t cb;
4043 int pending_move = 0;
4045 path = alloc_path_for_send();
4049 if (cmd == BTRFS_COMPARE_TREE_NEW) {
4050 root = sctx->send_root;
4051 cb = __record_new_ref;
4052 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
4053 root = sctx->parent_root;
4054 cb = __record_deleted_ref;
4056 btrfs_err(sctx->send_root->fs_info,
4057 "Wrong command %d in process_all_refs", cmd);
4062 key.objectid = sctx->cmp_key->objectid;
4063 key.type = BTRFS_INODE_REF_KEY;
4065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4070 eb = path->nodes[0];
4071 slot = path->slots[0];
4072 if (slot >= btrfs_header_nritems(eb)) {
4073 ret = btrfs_next_leaf(root, path);
4081 btrfs_item_key_to_cpu(eb, &found_key, slot);
4083 if (found_key.objectid != key.objectid ||
4084 (found_key.type != BTRFS_INODE_REF_KEY &&
4085 found_key.type != BTRFS_INODE_EXTREF_KEY))
4088 ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx);
4094 btrfs_release_path(path);
4096 ret = process_recorded_refs(sctx, &pending_move);
4097 /* Only applicable to an incremental send. */
4098 ASSERT(pending_move == 0);
4101 btrfs_free_path(path);
4105 static int send_set_xattr(struct send_ctx *sctx,
4106 struct fs_path *path,
4107 const char *name, int name_len,
4108 const char *data, int data_len)
4112 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
4116 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4117 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4118 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
4120 ret = send_cmd(sctx);
4127 static int send_remove_xattr(struct send_ctx *sctx,
4128 struct fs_path *path,
4129 const char *name, int name_len)
4133 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
4137 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
4138 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
4140 ret = send_cmd(sctx);
4147 static int __process_new_xattr(int num, struct btrfs_key *di_key,
4148 const char *name, int name_len,
4149 const char *data, int data_len,
4153 struct send_ctx *sctx = ctx;
4155 posix_acl_xattr_header dummy_acl;
4157 p = fs_path_alloc();
4162 * This hack is needed because empty acl's are stored as zero byte
4163 * data in xattrs. Problem with that is, that receiving these zero byte
4164 * acl's will fail later. To fix this, we send a dummy acl list that
4165 * only contains the version number and no entries.
4167 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
4168 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
4169 if (data_len == 0) {
4170 dummy_acl.a_version =
4171 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
4172 data = (char *)&dummy_acl;
4173 data_len = sizeof(dummy_acl);
4177 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4181 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
4188 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
4189 const char *name, int name_len,
4190 const char *data, int data_len,
4194 struct send_ctx *sctx = ctx;
4197 p = fs_path_alloc();
4201 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4205 ret = send_remove_xattr(sctx, p, name, name_len);
4212 static int process_new_xattr(struct send_ctx *sctx)
4216 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4217 sctx->cmp_key, __process_new_xattr, sctx);
4222 static int process_deleted_xattr(struct send_ctx *sctx)
4226 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4227 sctx->cmp_key, __process_deleted_xattr, sctx);
4232 struct find_xattr_ctx {
4240 static int __find_xattr(int num, struct btrfs_key *di_key,
4241 const char *name, int name_len,
4242 const char *data, int data_len,
4243 u8 type, void *vctx)
4245 struct find_xattr_ctx *ctx = vctx;
4247 if (name_len == ctx->name_len &&
4248 strncmp(name, ctx->name, name_len) == 0) {
4249 ctx->found_idx = num;
4250 ctx->found_data_len = data_len;
4251 ctx->found_data = kmemdup(data, data_len, GFP_NOFS);
4252 if (!ctx->found_data)
4259 static int find_xattr(struct btrfs_root *root,
4260 struct btrfs_path *path,
4261 struct btrfs_key *key,
4262 const char *name, int name_len,
4263 char **data, int *data_len)
4266 struct find_xattr_ctx ctx;
4269 ctx.name_len = name_len;
4271 ctx.found_data = NULL;
4272 ctx.found_data_len = 0;
4274 ret = iterate_dir_item(root, path, key, __find_xattr, &ctx);
4278 if (ctx.found_idx == -1)
4281 *data = ctx.found_data;
4282 *data_len = ctx.found_data_len;
4284 kfree(ctx.found_data);
4286 return ctx.found_idx;
4290 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
4291 const char *name, int name_len,
4292 const char *data, int data_len,
4296 struct send_ctx *sctx = ctx;
4297 char *found_data = NULL;
4298 int found_data_len = 0;
4300 ret = find_xattr(sctx->parent_root, sctx->right_path,
4301 sctx->cmp_key, name, name_len, &found_data,
4303 if (ret == -ENOENT) {
4304 ret = __process_new_xattr(num, di_key, name, name_len, data,
4305 data_len, type, ctx);
4306 } else if (ret >= 0) {
4307 if (data_len != found_data_len ||
4308 memcmp(data, found_data, data_len)) {
4309 ret = __process_new_xattr(num, di_key, name, name_len,
4310 data, data_len, type, ctx);
4320 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
4321 const char *name, int name_len,
4322 const char *data, int data_len,
4326 struct send_ctx *sctx = ctx;
4328 ret = find_xattr(sctx->send_root, sctx->left_path, sctx->cmp_key,
4329 name, name_len, NULL, NULL);
4331 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
4332 data_len, type, ctx);
4339 static int process_changed_xattr(struct send_ctx *sctx)
4343 ret = iterate_dir_item(sctx->send_root, sctx->left_path,
4344 sctx->cmp_key, __process_changed_new_xattr, sctx);
4347 ret = iterate_dir_item(sctx->parent_root, sctx->right_path,
4348 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
4354 static int process_all_new_xattrs(struct send_ctx *sctx)
4357 struct btrfs_root *root;
4358 struct btrfs_path *path;
4359 struct btrfs_key key;
4360 struct btrfs_key found_key;
4361 struct extent_buffer *eb;
4364 path = alloc_path_for_send();
4368 root = sctx->send_root;
4370 key.objectid = sctx->cmp_key->objectid;
4371 key.type = BTRFS_XATTR_ITEM_KEY;
4373 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4378 eb = path->nodes[0];
4379 slot = path->slots[0];
4380 if (slot >= btrfs_header_nritems(eb)) {
4381 ret = btrfs_next_leaf(root, path);
4384 } else if (ret > 0) {
4391 btrfs_item_key_to_cpu(eb, &found_key, slot);
4392 if (found_key.objectid != key.objectid ||
4393 found_key.type != key.type) {
4398 ret = iterate_dir_item(root, path, &found_key,
4399 __process_new_xattr, sctx);
4407 btrfs_free_path(path);
4411 static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
4413 struct btrfs_root *root = sctx->send_root;
4414 struct btrfs_fs_info *fs_info = root->fs_info;
4415 struct inode *inode;
4418 struct btrfs_key key;
4419 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
4421 unsigned pg_offset = offset & ~PAGE_CACHE_MASK;
4424 key.objectid = sctx->cur_ino;
4425 key.type = BTRFS_INODE_ITEM_KEY;
4428 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4430 return PTR_ERR(inode);
4432 if (offset + len > i_size_read(inode)) {
4433 if (offset > i_size_read(inode))
4436 len = offset - i_size_read(inode);
4441 last_index = (offset + len - 1) >> PAGE_CACHE_SHIFT;
4443 /* initial readahead */
4444 memset(&sctx->ra, 0, sizeof(struct file_ra_state));
4445 file_ra_state_init(&sctx->ra, inode->i_mapping);
4446 btrfs_force_ra(inode->i_mapping, &sctx->ra, NULL, index,
4447 last_index - index + 1);
4449 while (index <= last_index) {
4450 unsigned cur_len = min_t(unsigned, len,
4451 PAGE_CACHE_SIZE - pg_offset);
4452 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
4458 if (!PageUptodate(page)) {
4459 btrfs_readpage(NULL, page);
4461 if (!PageUptodate(page)) {
4463 page_cache_release(page);
4470 memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
4473 page_cache_release(page);
4485 * Read some bytes from the current inode/file and send a write command to
4488 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
4492 ssize_t num_read = 0;
4494 p = fs_path_alloc();
4498 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
4500 num_read = fill_read_buf(sctx, offset, len);
4501 if (num_read <= 0) {
4507 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4511 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4515 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4516 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4517 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
4519 ret = send_cmd(sctx);
4530 * Send a clone command to user space.
4532 static int send_clone(struct send_ctx *sctx,
4533 u64 offset, u32 len,
4534 struct clone_root *clone_root)
4540 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
4541 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
4542 clone_root->root->objectid, clone_root->ino,
4543 clone_root->offset);
4545 p = fs_path_alloc();
4549 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
4553 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4557 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4558 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
4559 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4561 if (clone_root->root == sctx->send_root) {
4562 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
4563 &gen, NULL, NULL, NULL, NULL);
4566 ret = get_cur_path(sctx, clone_root->ino, gen, p);
4568 ret = get_inode_path(clone_root->root, clone_root->ino, p);
4573 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
4574 clone_root->root->root_item.uuid);
4575 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
4576 le64_to_cpu(clone_root->root->root_item.ctransid));
4577 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
4578 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
4579 clone_root->offset);
4581 ret = send_cmd(sctx);
4590 * Send an update extent command to user space.
4592 static int send_update_extent(struct send_ctx *sctx,
4593 u64 offset, u32 len)
4598 p = fs_path_alloc();
4602 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
4606 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4610 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4611 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4612 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
4614 ret = send_cmd(sctx);
4622 static int send_hole(struct send_ctx *sctx, u64 end)
4624 struct fs_path *p = NULL;
4625 u64 offset = sctx->cur_inode_last_extent;
4629 p = fs_path_alloc();
4632 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
4634 goto tlv_put_failure;
4635 memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
4636 while (offset < end) {
4637 len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
4639 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
4642 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
4643 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
4644 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
4645 ret = send_cmd(sctx);
4655 static int send_write_or_clone(struct send_ctx *sctx,
4656 struct btrfs_path *path,
4657 struct btrfs_key *key,
4658 struct clone_root *clone_root)
4661 struct btrfs_file_extent_item *ei;
4662 u64 offset = key->offset;
4667 u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
4669 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4670 struct btrfs_file_extent_item);
4671 type = btrfs_file_extent_type(path->nodes[0], ei);
4672 if (type == BTRFS_FILE_EXTENT_INLINE) {
4673 len = btrfs_file_extent_inline_len(path->nodes[0],
4674 path->slots[0], ei);
4676 * it is possible the inline item won't cover the whole page,
4677 * but there may be items after this page. Make
4678 * sure to send the whole thing
4680 len = PAGE_CACHE_ALIGN(len);
4682 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
4685 if (offset + len > sctx->cur_inode_size)
4686 len = sctx->cur_inode_size - offset;
4692 if (clone_root && IS_ALIGNED(offset + len, bs)) {
4693 ret = send_clone(sctx, offset, len, clone_root);
4694 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
4695 ret = send_update_extent(sctx, offset, len);
4699 if (l > BTRFS_SEND_READ_SIZE)
4700 l = BTRFS_SEND_READ_SIZE;
4701 ret = send_write(sctx, pos + offset, l);
4714 static int is_extent_unchanged(struct send_ctx *sctx,
4715 struct btrfs_path *left_path,
4716 struct btrfs_key *ekey)
4719 struct btrfs_key key;
4720 struct btrfs_path *path = NULL;
4721 struct extent_buffer *eb;
4723 struct btrfs_key found_key;
4724 struct btrfs_file_extent_item *ei;
4729 u64 left_offset_fixed;
4737 path = alloc_path_for_send();
4741 eb = left_path->nodes[0];
4742 slot = left_path->slots[0];
4743 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4744 left_type = btrfs_file_extent_type(eb, ei);
4746 if (left_type != BTRFS_FILE_EXTENT_REG) {
4750 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4751 left_len = btrfs_file_extent_num_bytes(eb, ei);
4752 left_offset = btrfs_file_extent_offset(eb, ei);
4753 left_gen = btrfs_file_extent_generation(eb, ei);
4756 * Following comments will refer to these graphics. L is the left
4757 * extents which we are checking at the moment. 1-8 are the right
4758 * extents that we iterate.
4761 * |-1-|-2a-|-3-|-4-|-5-|-6-|
4764 * |--1--|-2b-|...(same as above)
4766 * Alternative situation. Happens on files where extents got split.
4768 * |-----------7-----------|-6-|
4770 * Alternative situation. Happens on files which got larger.
4773 * Nothing follows after 8.
4776 key.objectid = ekey->objectid;
4777 key.type = BTRFS_EXTENT_DATA_KEY;
4778 key.offset = ekey->offset;
4779 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
4788 * Handle special case where the right side has no extents at all.
4790 eb = path->nodes[0];
4791 slot = path->slots[0];
4792 btrfs_item_key_to_cpu(eb, &found_key, slot);
4793 if (found_key.objectid != key.objectid ||
4794 found_key.type != key.type) {
4795 /* If we're a hole then just pretend nothing changed */
4796 ret = (left_disknr) ? 0 : 1;
4801 * We're now on 2a, 2b or 7.
4804 while (key.offset < ekey->offset + left_len) {
4805 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
4806 right_type = btrfs_file_extent_type(eb, ei);
4807 if (right_type != BTRFS_FILE_EXTENT_REG) {
4812 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
4813 right_len = btrfs_file_extent_num_bytes(eb, ei);
4814 right_offset = btrfs_file_extent_offset(eb, ei);
4815 right_gen = btrfs_file_extent_generation(eb, ei);
4818 * Are we at extent 8? If yes, we know the extent is changed.
4819 * This may only happen on the first iteration.
4821 if (found_key.offset + right_len <= ekey->offset) {
4822 /* If we're a hole just pretend nothing changed */
4823 ret = (left_disknr) ? 0 : 1;
4827 left_offset_fixed = left_offset;
4828 if (key.offset < ekey->offset) {
4829 /* Fix the right offset for 2a and 7. */
4830 right_offset += ekey->offset - key.offset;
4832 /* Fix the left offset for all behind 2a and 2b */
4833 left_offset_fixed += key.offset - ekey->offset;
4837 * Check if we have the same extent.
4839 if (left_disknr != right_disknr ||
4840 left_offset_fixed != right_offset ||
4841 left_gen != right_gen) {
4847 * Go to the next extent.
4849 ret = btrfs_next_item(sctx->parent_root, path);
4853 eb = path->nodes[0];
4854 slot = path->slots[0];
4855 btrfs_item_key_to_cpu(eb, &found_key, slot);
4857 if (ret || found_key.objectid != key.objectid ||
4858 found_key.type != key.type) {
4859 key.offset += right_len;
4862 if (found_key.offset != key.offset + right_len) {
4870 * We're now behind the left extent (treat as unchanged) or at the end
4871 * of the right side (treat as changed).
4873 if (key.offset >= ekey->offset + left_len)
4880 btrfs_free_path(path);
4884 static int get_last_extent(struct send_ctx *sctx, u64 offset)
4886 struct btrfs_path *path;
4887 struct btrfs_root *root = sctx->send_root;
4888 struct btrfs_file_extent_item *fi;
4889 struct btrfs_key key;
4894 path = alloc_path_for_send();
4898 sctx->cur_inode_last_extent = 0;
4900 key.objectid = sctx->cur_ino;
4901 key.type = BTRFS_EXTENT_DATA_KEY;
4902 key.offset = offset;
4903 ret = btrfs_search_slot_for_read(root, &key, path, 0, 1);
4907 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4908 if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY)
4911 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
4912 struct btrfs_file_extent_item);
4913 type = btrfs_file_extent_type(path->nodes[0], fi);
4914 if (type == BTRFS_FILE_EXTENT_INLINE) {
4915 u64 size = btrfs_file_extent_inline_len(path->nodes[0],
4916 path->slots[0], fi);
4917 extent_end = ALIGN(key.offset + size,
4918 sctx->send_root->sectorsize);
4920 extent_end = key.offset +
4921 btrfs_file_extent_num_bytes(path->nodes[0], fi);
4923 sctx->cur_inode_last_extent = extent_end;
4925 btrfs_free_path(path);
4929 static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path,
4930 struct btrfs_key *key)
4932 struct btrfs_file_extent_item *fi;
4937 if (sctx->cur_ino != key->objectid || !need_send_hole(sctx))
4940 if (sctx->cur_inode_last_extent == (u64)-1) {
4941 ret = get_last_extent(sctx, key->offset - 1);
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);
4959 if (path->slots[0] == 0 &&
4960 sctx->cur_inode_last_extent < key->offset) {
4962 * We might have skipped entire leafs that contained only
4963 * file extent items for our current inode. These leafs have
4964 * a generation number smaller (older) than the one in the
4965 * current leaf and the leaf our last extent came from, and
4966 * are located between these 2 leafs.
4968 ret = get_last_extent(sctx, key->offset - 1);
4973 if (sctx->cur_inode_last_extent < key->offset)
4974 ret = send_hole(sctx, key->offset);
4975 sctx->cur_inode_last_extent = extent_end;
4979 static int process_extent(struct send_ctx *sctx,
4980 struct btrfs_path *path,
4981 struct btrfs_key *key)
4983 struct clone_root *found_clone = NULL;
4986 if (S_ISLNK(sctx->cur_inode_mode))
4989 if (sctx->parent_root && !sctx->cur_inode_new) {
4990 ret = is_extent_unchanged(sctx, path, key);
4998 struct btrfs_file_extent_item *ei;
5001 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5002 struct btrfs_file_extent_item);
5003 type = btrfs_file_extent_type(path->nodes[0], ei);
5004 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
5005 type == BTRFS_FILE_EXTENT_REG) {
5007 * The send spec does not have a prealloc command yet,
5008 * so just leave a hole for prealloc'ed extents until
5009 * we have enough commands queued up to justify rev'ing
5012 if (type == BTRFS_FILE_EXTENT_PREALLOC) {
5017 /* Have a hole, just skip it. */
5018 if (btrfs_file_extent_disk_bytenr(path->nodes[0], ei) == 0) {
5025 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
5026 sctx->cur_inode_size, &found_clone);
5027 if (ret != -ENOENT && ret < 0)
5030 ret = send_write_or_clone(sctx, path, key, found_clone);
5034 ret = maybe_send_hole(sctx, path, key);
5039 static int process_all_extents(struct send_ctx *sctx)
5042 struct btrfs_root *root;
5043 struct btrfs_path *path;
5044 struct btrfs_key key;
5045 struct btrfs_key found_key;
5046 struct extent_buffer *eb;
5049 root = sctx->send_root;
5050 path = alloc_path_for_send();
5054 key.objectid = sctx->cmp_key->objectid;
5055 key.type = BTRFS_EXTENT_DATA_KEY;
5057 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5062 eb = path->nodes[0];
5063 slot = path->slots[0];
5065 if (slot >= btrfs_header_nritems(eb)) {
5066 ret = btrfs_next_leaf(root, path);
5069 } else if (ret > 0) {
5076 btrfs_item_key_to_cpu(eb, &found_key, slot);
5078 if (found_key.objectid != key.objectid ||
5079 found_key.type != key.type) {
5084 ret = process_extent(sctx, path, &found_key);
5092 btrfs_free_path(path);
5096 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end,
5098 int *refs_processed)
5102 if (sctx->cur_ino == 0)
5104 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
5105 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
5107 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
5110 ret = process_recorded_refs(sctx, pending_move);
5114 *refs_processed = 1;
5119 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
5130 int pending_move = 0;
5131 int refs_processed = 0;
5133 ret = process_recorded_refs_if_needed(sctx, at_end, &pending_move,
5139 * We have processed the refs and thus need to advance send_progress.
5140 * Now, calls to get_cur_xxx will take the updated refs of the current
5141 * inode into account.
5143 * On the other hand, if our current inode is a directory and couldn't
5144 * be moved/renamed because its parent was renamed/moved too and it has
5145 * a higher inode number, we can only move/rename our current inode
5146 * after we moved/renamed its parent. Therefore in this case operate on
5147 * the old path (pre move/rename) of our current inode, and the
5148 * move/rename will be performed later.
5150 if (refs_processed && !pending_move)
5151 sctx->send_progress = sctx->cur_ino + 1;
5153 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
5155 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
5158 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
5159 &left_mode, &left_uid, &left_gid, NULL);
5163 if (!sctx->parent_root || sctx->cur_inode_new) {
5165 if (!S_ISLNK(sctx->cur_inode_mode))
5168 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
5169 NULL, NULL, &right_mode, &right_uid,
5174 if (left_uid != right_uid || left_gid != right_gid)
5176 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
5180 if (S_ISREG(sctx->cur_inode_mode)) {
5181 if (need_send_hole(sctx)) {
5182 if (sctx->cur_inode_last_extent == (u64)-1 ||
5183 sctx->cur_inode_last_extent <
5184 sctx->cur_inode_size) {
5185 ret = get_last_extent(sctx, (u64)-1);
5189 if (sctx->cur_inode_last_extent <
5190 sctx->cur_inode_size) {
5191 ret = send_hole(sctx, sctx->cur_inode_size);
5196 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5197 sctx->cur_inode_size);
5203 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5204 left_uid, left_gid);
5209 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
5216 * If other directory inodes depended on our current directory
5217 * inode's move/rename, now do their move/rename operations.
5219 if (!is_waiting_for_move(sctx, sctx->cur_ino)) {
5220 ret = apply_children_dir_moves(sctx);
5224 * Need to send that every time, no matter if it actually
5225 * changed between the two trees as we have done changes to
5226 * the inode before. If our inode is a directory and it's
5227 * waiting to be moved/renamed, we will send its utimes when
5228 * it's moved/renamed, therefore we don't need to do it here.
5230 sctx->send_progress = sctx->cur_ino + 1;
5231 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
5240 static int changed_inode(struct send_ctx *sctx,
5241 enum btrfs_compare_tree_result result)
5244 struct btrfs_key *key = sctx->cmp_key;
5245 struct btrfs_inode_item *left_ii = NULL;
5246 struct btrfs_inode_item *right_ii = NULL;
5250 sctx->cur_ino = key->objectid;
5251 sctx->cur_inode_new_gen = 0;
5252 sctx->cur_inode_last_extent = (u64)-1;
5255 * Set send_progress to current inode. This will tell all get_cur_xxx
5256 * functions that the current inode's refs are not updated yet. Later,
5257 * when process_recorded_refs is finished, it is set to cur_ino + 1.
5259 sctx->send_progress = sctx->cur_ino;
5261 if (result == BTRFS_COMPARE_TREE_NEW ||
5262 result == BTRFS_COMPARE_TREE_CHANGED) {
5263 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
5264 sctx->left_path->slots[0],
5265 struct btrfs_inode_item);
5266 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
5269 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5270 sctx->right_path->slots[0],
5271 struct btrfs_inode_item);
5272 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5275 if (result == BTRFS_COMPARE_TREE_CHANGED) {
5276 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
5277 sctx->right_path->slots[0],
5278 struct btrfs_inode_item);
5280 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
5284 * The cur_ino = root dir case is special here. We can't treat
5285 * the inode as deleted+reused because it would generate a
5286 * stream that tries to delete/mkdir the root dir.
5288 if (left_gen != right_gen &&
5289 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5290 sctx->cur_inode_new_gen = 1;
5293 if (result == BTRFS_COMPARE_TREE_NEW) {
5294 sctx->cur_inode_gen = left_gen;
5295 sctx->cur_inode_new = 1;
5296 sctx->cur_inode_deleted = 0;
5297 sctx->cur_inode_size = btrfs_inode_size(
5298 sctx->left_path->nodes[0], left_ii);
5299 sctx->cur_inode_mode = btrfs_inode_mode(
5300 sctx->left_path->nodes[0], left_ii);
5301 sctx->cur_inode_rdev = btrfs_inode_rdev(
5302 sctx->left_path->nodes[0], left_ii);
5303 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
5304 ret = send_create_inode_if_needed(sctx);
5305 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
5306 sctx->cur_inode_gen = right_gen;
5307 sctx->cur_inode_new = 0;
5308 sctx->cur_inode_deleted = 1;
5309 sctx->cur_inode_size = btrfs_inode_size(
5310 sctx->right_path->nodes[0], right_ii);
5311 sctx->cur_inode_mode = btrfs_inode_mode(
5312 sctx->right_path->nodes[0], right_ii);
5313 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
5315 * We need to do some special handling in case the inode was
5316 * reported as changed with a changed generation number. This
5317 * means that the original inode was deleted and new inode
5318 * reused the same inum. So we have to treat the old inode as
5319 * deleted and the new one as new.
5321 if (sctx->cur_inode_new_gen) {
5323 * First, process the inode as if it was deleted.
5325 sctx->cur_inode_gen = right_gen;
5326 sctx->cur_inode_new = 0;
5327 sctx->cur_inode_deleted = 1;
5328 sctx->cur_inode_size = btrfs_inode_size(
5329 sctx->right_path->nodes[0], right_ii);
5330 sctx->cur_inode_mode = btrfs_inode_mode(
5331 sctx->right_path->nodes[0], right_ii);
5332 ret = process_all_refs(sctx,
5333 BTRFS_COMPARE_TREE_DELETED);
5338 * Now process the inode as if it was new.
5340 sctx->cur_inode_gen = left_gen;
5341 sctx->cur_inode_new = 1;
5342 sctx->cur_inode_deleted = 0;
5343 sctx->cur_inode_size = btrfs_inode_size(
5344 sctx->left_path->nodes[0], left_ii);
5345 sctx->cur_inode_mode = btrfs_inode_mode(
5346 sctx->left_path->nodes[0], left_ii);
5347 sctx->cur_inode_rdev = btrfs_inode_rdev(
5348 sctx->left_path->nodes[0], left_ii);
5349 ret = send_create_inode_if_needed(sctx);
5353 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
5357 * Advance send_progress now as we did not get into
5358 * process_recorded_refs_if_needed in the new_gen case.
5360 sctx->send_progress = sctx->cur_ino + 1;
5363 * Now process all extents and xattrs of the inode as if
5364 * they were all new.
5366 ret = process_all_extents(sctx);
5369 ret = process_all_new_xattrs(sctx);
5373 sctx->cur_inode_gen = left_gen;
5374 sctx->cur_inode_new = 0;
5375 sctx->cur_inode_new_gen = 0;
5376 sctx->cur_inode_deleted = 0;
5377 sctx->cur_inode_size = btrfs_inode_size(
5378 sctx->left_path->nodes[0], left_ii);
5379 sctx->cur_inode_mode = btrfs_inode_mode(
5380 sctx->left_path->nodes[0], left_ii);
5389 * We have to process new refs before deleted refs, but compare_trees gives us
5390 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
5391 * first and later process them in process_recorded_refs.
5392 * For the cur_inode_new_gen case, we skip recording completely because
5393 * changed_inode did already initiate processing of refs. The reason for this is
5394 * that in this case, compare_tree actually compares the refs of 2 different
5395 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
5396 * refs of the right tree as deleted and all refs of the left tree as new.
5398 static int changed_ref(struct send_ctx *sctx,
5399 enum btrfs_compare_tree_result result)
5403 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5405 if (!sctx->cur_inode_new_gen &&
5406 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
5407 if (result == BTRFS_COMPARE_TREE_NEW)
5408 ret = record_new_ref(sctx);
5409 else if (result == BTRFS_COMPARE_TREE_DELETED)
5410 ret = record_deleted_ref(sctx);
5411 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5412 ret = record_changed_ref(sctx);
5419 * Process new/deleted/changed xattrs. We skip processing in the
5420 * cur_inode_new_gen case because changed_inode did already initiate processing
5421 * of xattrs. The reason is the same as in changed_ref
5423 static int changed_xattr(struct send_ctx *sctx,
5424 enum btrfs_compare_tree_result result)
5428 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5430 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5431 if (result == BTRFS_COMPARE_TREE_NEW)
5432 ret = process_new_xattr(sctx);
5433 else if (result == BTRFS_COMPARE_TREE_DELETED)
5434 ret = process_deleted_xattr(sctx);
5435 else if (result == BTRFS_COMPARE_TREE_CHANGED)
5436 ret = process_changed_xattr(sctx);
5443 * Process new/deleted/changed extents. We skip processing in the
5444 * cur_inode_new_gen case because changed_inode did already initiate processing
5445 * of extents. The reason is the same as in changed_ref
5447 static int changed_extent(struct send_ctx *sctx,
5448 enum btrfs_compare_tree_result result)
5452 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
5454 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
5455 if (result != BTRFS_COMPARE_TREE_DELETED)
5456 ret = process_extent(sctx, sctx->left_path,
5463 static int dir_changed(struct send_ctx *sctx, u64 dir)
5465 u64 orig_gen, new_gen;
5468 ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL,
5473 ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL,
5478 return (orig_gen != new_gen) ? 1 : 0;
5481 static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path,
5482 struct btrfs_key *key)
5484 struct btrfs_inode_extref *extref;
5485 struct extent_buffer *leaf;
5486 u64 dirid = 0, last_dirid = 0;
5493 /* Easy case, just check this one dirid */
5494 if (key->type == BTRFS_INODE_REF_KEY) {
5495 dirid = key->offset;
5497 ret = dir_changed(sctx, dirid);
5501 leaf = path->nodes[0];
5502 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
5503 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
5504 while (cur_offset < item_size) {
5505 extref = (struct btrfs_inode_extref *)(ptr +
5507 dirid = btrfs_inode_extref_parent(leaf, extref);
5508 ref_name_len = btrfs_inode_extref_name_len(leaf, extref);
5509 cur_offset += ref_name_len + sizeof(*extref);
5510 if (dirid == last_dirid)
5512 ret = dir_changed(sctx, dirid);
5522 * Updates compare related fields in sctx and simply forwards to the actual
5523 * changed_xxx functions.
5525 static int changed_cb(struct btrfs_root *left_root,
5526 struct btrfs_root *right_root,
5527 struct btrfs_path *left_path,
5528 struct btrfs_path *right_path,
5529 struct btrfs_key *key,
5530 enum btrfs_compare_tree_result result,
5534 struct send_ctx *sctx = ctx;
5536 if (result == BTRFS_COMPARE_TREE_SAME) {
5537 if (key->type == BTRFS_INODE_REF_KEY ||
5538 key->type == BTRFS_INODE_EXTREF_KEY) {
5539 ret = compare_refs(sctx, left_path, key);
5544 } else if (key->type == BTRFS_EXTENT_DATA_KEY) {
5545 return maybe_send_hole(sctx, left_path, key);
5549 result = BTRFS_COMPARE_TREE_CHANGED;
5553 sctx->left_path = left_path;
5554 sctx->right_path = right_path;
5555 sctx->cmp_key = key;
5557 ret = finish_inode_if_needed(sctx, 0);
5561 /* Ignore non-FS objects */
5562 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
5563 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
5566 if (key->type == BTRFS_INODE_ITEM_KEY)
5567 ret = changed_inode(sctx, result);
5568 else if (key->type == BTRFS_INODE_REF_KEY ||
5569 key->type == BTRFS_INODE_EXTREF_KEY)
5570 ret = changed_ref(sctx, result);
5571 else if (key->type == BTRFS_XATTR_ITEM_KEY)
5572 ret = changed_xattr(sctx, result);
5573 else if (key->type == BTRFS_EXTENT_DATA_KEY)
5574 ret = changed_extent(sctx, result);
5580 static int full_send_tree(struct send_ctx *sctx)
5583 struct btrfs_root *send_root = sctx->send_root;
5584 struct btrfs_key key;
5585 struct btrfs_key found_key;
5586 struct btrfs_path *path;
5587 struct extent_buffer *eb;
5590 path = alloc_path_for_send();
5594 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
5595 key.type = BTRFS_INODE_ITEM_KEY;
5598 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
5605 eb = path->nodes[0];
5606 slot = path->slots[0];
5607 btrfs_item_key_to_cpu(eb, &found_key, slot);
5609 ret = changed_cb(send_root, NULL, path, NULL,
5610 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
5614 key.objectid = found_key.objectid;
5615 key.type = found_key.type;
5616 key.offset = found_key.offset + 1;
5618 ret = btrfs_next_item(send_root, path);
5628 ret = finish_inode_if_needed(sctx, 1);
5631 btrfs_free_path(path);
5635 static int send_subvol(struct send_ctx *sctx)
5639 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
5640 ret = send_header(sctx);
5645 ret = send_subvol_begin(sctx);
5649 if (sctx->parent_root) {
5650 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
5654 ret = finish_inode_if_needed(sctx, 1);
5658 ret = full_send_tree(sctx);
5664 free_recorded_refs(sctx);
5669 * If orphan cleanup did remove any orphans from a root, it means the tree
5670 * was modified and therefore the commit root is not the same as the current
5671 * root anymore. This is a problem, because send uses the commit root and
5672 * therefore can see inode items that don't exist in the current root anymore,
5673 * and for example make calls to btrfs_iget, which will do tree lookups based
5674 * on the current root and not on the commit root. Those lookups will fail,
5675 * returning a -ESTALE error, and making send fail with that error. So make
5676 * sure a send does not see any orphans we have just removed, and that it will
5677 * see the same inodes regardless of whether a transaction commit happened
5678 * before it started (meaning that the commit root will be the same as the
5679 * current root) or not.
5681 static int ensure_commit_roots_uptodate(struct send_ctx *sctx)
5684 struct btrfs_trans_handle *trans = NULL;
5687 if (sctx->parent_root &&
5688 sctx->parent_root->node != sctx->parent_root->commit_root)
5691 for (i = 0; i < sctx->clone_roots_cnt; i++)
5692 if (sctx->clone_roots[i].root->node !=
5693 sctx->clone_roots[i].root->commit_root)
5697 return btrfs_end_transaction(trans, sctx->send_root);
5702 /* Use any root, all fs roots will get their commit roots updated. */
5704 trans = btrfs_join_transaction(sctx->send_root);
5706 return PTR_ERR(trans);
5710 return btrfs_commit_transaction(trans, sctx->send_root);
5713 static void btrfs_root_dec_send_in_progress(struct btrfs_root* root)
5715 spin_lock(&root->root_item_lock);
5716 root->send_in_progress--;
5718 * Not much left to do, we don't know why it's unbalanced and
5719 * can't blindly reset it to 0.
5721 if (root->send_in_progress < 0)
5722 btrfs_err(root->fs_info,
5723 "send_in_progres unbalanced %d root %llu",
5724 root->send_in_progress, root->root_key.objectid);
5725 spin_unlock(&root->root_item_lock);
5728 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
5731 struct btrfs_root *send_root;
5732 struct btrfs_root *clone_root;
5733 struct btrfs_fs_info *fs_info;
5734 struct btrfs_ioctl_send_args *arg = NULL;
5735 struct btrfs_key key;
5736 struct send_ctx *sctx = NULL;
5738 u64 *clone_sources_tmp = NULL;
5739 int clone_sources_to_rollback = 0;
5740 int sort_clone_roots = 0;
5743 if (!capable(CAP_SYS_ADMIN))
5746 send_root = BTRFS_I(file_inode(mnt_file))->root;
5747 fs_info = send_root->fs_info;
5750 * The subvolume must remain read-only during send, protect against
5751 * making it RW. This also protects against deletion.
5753 spin_lock(&send_root->root_item_lock);
5754 send_root->send_in_progress++;
5755 spin_unlock(&send_root->root_item_lock);
5758 * This is done when we lookup the root, it should already be complete
5759 * by the time we get here.
5761 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
5764 * Userspace tools do the checks and warn the user if it's
5767 if (!btrfs_root_readonly(send_root)) {
5772 arg = memdup_user(arg_, sizeof(*arg));
5779 if (!access_ok(VERIFY_READ, arg->clone_sources,
5780 sizeof(*arg->clone_sources) *
5781 arg->clone_sources_count)) {
5786 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
5791 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
5797 INIT_LIST_HEAD(&sctx->new_refs);
5798 INIT_LIST_HEAD(&sctx->deleted_refs);
5799 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
5800 INIT_LIST_HEAD(&sctx->name_cache_list);
5802 sctx->flags = arg->flags;
5804 sctx->send_filp = fget(arg->send_fd);
5805 if (!sctx->send_filp) {
5810 sctx->send_root = send_root;
5812 * Unlikely but possible, if the subvolume is marked for deletion but
5813 * is slow to remove the directory entry, send can still be started
5815 if (btrfs_root_dead(sctx->send_root)) {
5820 sctx->clone_roots_cnt = arg->clone_sources_count;
5822 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
5823 sctx->send_buf = vmalloc(sctx->send_max_size);
5824 if (!sctx->send_buf) {
5829 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
5830 if (!sctx->read_buf) {
5835 sctx->pending_dir_moves = RB_ROOT;
5836 sctx->waiting_dir_moves = RB_ROOT;
5837 sctx->orphan_dirs = RB_ROOT;
5839 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
5840 (arg->clone_sources_count + 1));
5841 if (!sctx->clone_roots) {
5846 if (arg->clone_sources_count) {
5847 clone_sources_tmp = vmalloc(arg->clone_sources_count *
5848 sizeof(*arg->clone_sources));
5849 if (!clone_sources_tmp) {
5854 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
5855 arg->clone_sources_count *
5856 sizeof(*arg->clone_sources));
5862 for (i = 0; i < arg->clone_sources_count; i++) {
5863 key.objectid = clone_sources_tmp[i];
5864 key.type = BTRFS_ROOT_ITEM_KEY;
5865 key.offset = (u64)-1;
5867 index = srcu_read_lock(&fs_info->subvol_srcu);
5869 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
5870 if (IS_ERR(clone_root)) {
5871 srcu_read_unlock(&fs_info->subvol_srcu, index);
5872 ret = PTR_ERR(clone_root);
5875 spin_lock(&clone_root->root_item_lock);
5876 if (!btrfs_root_readonly(clone_root) ||
5877 btrfs_root_dead(clone_root)) {
5878 spin_unlock(&clone_root->root_item_lock);
5879 srcu_read_unlock(&fs_info->subvol_srcu, index);
5883 clone_root->send_in_progress++;
5884 spin_unlock(&clone_root->root_item_lock);
5885 srcu_read_unlock(&fs_info->subvol_srcu, index);
5887 sctx->clone_roots[i].root = clone_root;
5888 clone_sources_to_rollback = i + 1;
5890 vfree(clone_sources_tmp);
5891 clone_sources_tmp = NULL;
5894 if (arg->parent_root) {
5895 key.objectid = arg->parent_root;
5896 key.type = BTRFS_ROOT_ITEM_KEY;
5897 key.offset = (u64)-1;
5899 index = srcu_read_lock(&fs_info->subvol_srcu);
5901 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
5902 if (IS_ERR(sctx->parent_root)) {
5903 srcu_read_unlock(&fs_info->subvol_srcu, index);
5904 ret = PTR_ERR(sctx->parent_root);
5908 spin_lock(&sctx->parent_root->root_item_lock);
5909 sctx->parent_root->send_in_progress++;
5910 if (!btrfs_root_readonly(sctx->parent_root) ||
5911 btrfs_root_dead(sctx->parent_root)) {
5912 spin_unlock(&sctx->parent_root->root_item_lock);
5913 srcu_read_unlock(&fs_info->subvol_srcu, index);
5917 spin_unlock(&sctx->parent_root->root_item_lock);
5919 srcu_read_unlock(&fs_info->subvol_srcu, index);
5923 * Clones from send_root are allowed, but only if the clone source
5924 * is behind the current send position. This is checked while searching
5925 * for possible clone sources.
5927 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
5929 /* We do a bsearch later */
5930 sort(sctx->clone_roots, sctx->clone_roots_cnt,
5931 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
5933 sort_clone_roots = 1;
5935 ret = ensure_commit_roots_uptodate(sctx);
5939 current->journal_info = BTRFS_SEND_TRANS_STUB;
5940 ret = send_subvol(sctx);
5941 current->journal_info = NULL;
5945 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
5946 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
5949 ret = send_cmd(sctx);
5955 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->pending_dir_moves));
5956 while (sctx && !RB_EMPTY_ROOT(&sctx->pending_dir_moves)) {
5958 struct pending_dir_move *pm;
5960 n = rb_first(&sctx->pending_dir_moves);
5961 pm = rb_entry(n, struct pending_dir_move, node);
5962 while (!list_empty(&pm->list)) {
5963 struct pending_dir_move *pm2;
5965 pm2 = list_first_entry(&pm->list,
5966 struct pending_dir_move, list);
5967 free_pending_move(sctx, pm2);
5969 free_pending_move(sctx, pm);
5972 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves));
5973 while (sctx && !RB_EMPTY_ROOT(&sctx->waiting_dir_moves)) {
5975 struct waiting_dir_move *dm;
5977 n = rb_first(&sctx->waiting_dir_moves);
5978 dm = rb_entry(n, struct waiting_dir_move, node);
5979 rb_erase(&dm->node, &sctx->waiting_dir_moves);
5983 WARN_ON(sctx && !ret && !RB_EMPTY_ROOT(&sctx->orphan_dirs));
5984 while (sctx && !RB_EMPTY_ROOT(&sctx->orphan_dirs)) {
5986 struct orphan_dir_info *odi;
5988 n = rb_first(&sctx->orphan_dirs);
5989 odi = rb_entry(n, struct orphan_dir_info, node);
5990 free_orphan_dir_info(sctx, odi);
5993 if (sort_clone_roots) {
5994 for (i = 0; i < sctx->clone_roots_cnt; i++)
5995 btrfs_root_dec_send_in_progress(
5996 sctx->clone_roots[i].root);
5998 for (i = 0; sctx && i < clone_sources_to_rollback; i++)
5999 btrfs_root_dec_send_in_progress(
6000 sctx->clone_roots[i].root);
6002 btrfs_root_dec_send_in_progress(send_root);
6004 if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
6005 btrfs_root_dec_send_in_progress(sctx->parent_root);
6008 vfree(clone_sources_tmp);
6011 if (sctx->send_filp)
6012 fput(sctx->send_filp);
6014 vfree(sctx->clone_roots);
6015 vfree(sctx->send_buf);
6016 vfree(sctx->read_buf);
6018 name_cache_free(sctx);