2 * Copyright (C) 2012 Alexander Block. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bsearch.h>
21 #include <linux/file.h>
22 #include <linux/sort.h>
23 #include <linux/mount.h>
24 #include <linux/xattr.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/radix-tree.h>
27 #include <linux/crc32c.h>
28 #include <linux/vmalloc.h>
34 #include "btrfs_inode.h"
35 #include "transaction.h"
37 static int g_verbose = 0;
39 #define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
42 * A fs_path is a helper to dynamically build path names with unknown size.
43 * It reallocates the internal buffer on demand.
44 * It allows fast adding of path elements on the right side (normal path) and
45 * fast adding to the left side (reversed path). A reversed path can also be
46 * unreversed if needed.
64 #define FS_PATH_INLINE_SIZE \
65 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
68 /* reused for each extent */
70 struct btrfs_root *root;
77 #define SEND_CTX_MAX_NAME_CACHE_SIZE 128
78 #define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
81 struct file *send_filp;
87 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
88 u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */
92 struct btrfs_root *send_root;
93 struct btrfs_root *parent_root;
94 struct clone_root *clone_roots;
97 /* current state of the compare_tree call */
98 struct btrfs_path *left_path;
99 struct btrfs_path *right_path;
100 struct btrfs_key *cmp_key;
103 * infos of the currently processed inode. In case of deleted inodes,
104 * these are the values from the deleted inode.
109 int cur_inode_new_gen;
110 int cur_inode_deleted;
116 struct list_head new_refs;
117 struct list_head deleted_refs;
119 struct radix_tree_root name_cache;
120 struct list_head name_cache_list;
123 struct file *cur_inode_filp;
127 struct name_cache_entry {
128 struct list_head list;
130 * radix_tree has only 32bit entries but we need to handle 64bit inums.
131 * We use the lower 32bit of the 64bit inum to store it in the tree. If
132 * more then one inum would fall into the same entry, we use radix_list
133 * to store the additional entries. radix_list is also used to store
134 * entries where two entries have the same inum but different
137 struct list_head radix_list;
143 int need_later_update;
148 static void fs_path_reset(struct fs_path *p)
151 p->start = p->buf + p->buf_len - 1;
161 static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
165 p = kmalloc(sizeof(*p), GFP_NOFS);
170 p->buf = p->inline_buf;
171 p->buf_len = FS_PATH_INLINE_SIZE;
176 static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
180 p = fs_path_alloc(sctx);
188 static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
192 if (p->buf != p->inline_buf) {
201 static int fs_path_len(struct fs_path *p)
203 return p->end - p->start;
206 static int fs_path_ensure_buf(struct fs_path *p, int len)
214 if (p->buf_len >= len)
217 path_len = p->end - p->start;
218 old_buf_len = p->buf_len;
219 len = PAGE_ALIGN(len);
221 if (p->buf == p->inline_buf) {
222 tmp_buf = kmalloc(len, GFP_NOFS);
224 tmp_buf = vmalloc(len);
229 memcpy(tmp_buf, p->buf, p->buf_len);
233 if (p->virtual_mem) {
234 tmp_buf = vmalloc(len);
237 memcpy(tmp_buf, p->buf, p->buf_len);
240 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
242 tmp_buf = vmalloc(len);
245 memcpy(tmp_buf, p->buf, p->buf_len);
254 tmp_buf = p->buf + old_buf_len - path_len - 1;
255 p->end = p->buf + p->buf_len - 1;
256 p->start = p->end - path_len;
257 memmove(p->start, tmp_buf, path_len + 1);
260 p->end = p->start + path_len;
265 static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
270 new_len = p->end - p->start + name_len;
271 if (p->start != p->end)
273 ret = fs_path_ensure_buf(p, new_len);
278 if (p->start != p->end)
280 p->start -= name_len;
281 p->prepared = p->start;
283 if (p->start != p->end)
285 p->prepared = p->end;
294 static int fs_path_add(struct fs_path *p, const char *name, int name_len)
298 ret = fs_path_prepare_for_add(p, name_len);
301 memcpy(p->prepared, name, name_len);
308 static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
312 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
315 memcpy(p->prepared, p2->start, p2->end - p2->start);
322 static int fs_path_add_from_extent_buffer(struct fs_path *p,
323 struct extent_buffer *eb,
324 unsigned long off, int len)
328 ret = fs_path_prepare_for_add(p, len);
332 read_extent_buffer(eb, p->prepared, off, len);
340 static void fs_path_remove(struct fs_path *p)
343 while (p->start != p->end && *p->end != '/')
349 static int fs_path_copy(struct fs_path *p, struct fs_path *from)
353 p->reversed = from->reversed;
356 ret = fs_path_add_path(p, from);
362 static void fs_path_unreverse(struct fs_path *p)
371 len = p->end - p->start;
373 p->end = p->start + len;
374 memmove(p->start, tmp, len + 1);
378 static struct btrfs_path *alloc_path_for_send(void)
380 struct btrfs_path *path;
382 path = btrfs_alloc_path();
385 path->search_commit_root = 1;
386 path->skip_locking = 1;
390 static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off)
400 ret = vfs_write(filp, (char *)buf + pos, len - pos, off);
401 /* TODO handle that correctly */
402 /*if (ret == -ERESTARTSYS) {
421 static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
423 struct btrfs_tlv_header *hdr;
424 int total_len = sizeof(*hdr) + len;
425 int left = sctx->send_max_size - sctx->send_size;
427 if (unlikely(left < total_len))
430 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
431 hdr->tlv_type = cpu_to_le16(attr);
432 hdr->tlv_len = cpu_to_le16(len);
433 memcpy(hdr + 1, data, len);
434 sctx->send_size += total_len;
440 static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
442 return tlv_put(sctx, attr, &value, sizeof(value));
445 static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
447 __le16 tmp = cpu_to_le16(value);
448 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
451 static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
453 __le32 tmp = cpu_to_le32(value);
454 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
458 static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
460 __le64 tmp = cpu_to_le64(value);
461 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
464 static int tlv_put_string(struct send_ctx *sctx, u16 attr,
465 const char *str, int len)
469 return tlv_put(sctx, attr, str, len);
472 static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
475 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
479 static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
482 struct btrfs_timespec bts;
483 bts.sec = cpu_to_le64(ts->tv_sec);
484 bts.nsec = cpu_to_le32(ts->tv_nsec);
485 return tlv_put(sctx, attr, &bts, sizeof(bts));
489 static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
490 struct extent_buffer *eb,
491 struct btrfs_timespec *ts)
493 struct btrfs_timespec bts;
494 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
495 return tlv_put(sctx, attr, &bts, sizeof(bts));
499 #define TLV_PUT(sctx, attrtype, attrlen, data) \
501 ret = tlv_put(sctx, attrtype, attrlen, data); \
503 goto tlv_put_failure; \
506 #define TLV_PUT_INT(sctx, attrtype, bits, value) \
508 ret = tlv_put_u##bits(sctx, attrtype, value); \
510 goto tlv_put_failure; \
513 #define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
514 #define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
515 #define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
516 #define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
517 #define TLV_PUT_STRING(sctx, attrtype, str, len) \
519 ret = tlv_put_string(sctx, attrtype, str, len); \
521 goto tlv_put_failure; \
523 #define TLV_PUT_PATH(sctx, attrtype, p) \
525 ret = tlv_put_string(sctx, attrtype, p->start, \
526 p->end - p->start); \
528 goto tlv_put_failure; \
530 #define TLV_PUT_UUID(sctx, attrtype, uuid) \
532 ret = tlv_put_uuid(sctx, attrtype, uuid); \
534 goto tlv_put_failure; \
536 #define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
538 ret = tlv_put_timespec(sctx, attrtype, ts); \
540 goto tlv_put_failure; \
542 #define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
544 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
546 goto tlv_put_failure; \
549 static int send_header(struct send_ctx *sctx)
551 struct btrfs_stream_header hdr;
553 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
554 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
556 return write_buf(sctx->send_filp, &hdr, sizeof(hdr),
561 * For each command/item we want to send to userspace, we call this function.
563 static int begin_cmd(struct send_ctx *sctx, int cmd)
565 struct btrfs_cmd_header *hdr;
567 if (!sctx->send_buf) {
572 BUG_ON(sctx->send_size);
574 sctx->send_size += sizeof(*hdr);
575 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
576 hdr->cmd = cpu_to_le16(cmd);
581 static int send_cmd(struct send_ctx *sctx)
584 struct btrfs_cmd_header *hdr;
587 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
588 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
591 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
592 hdr->crc = cpu_to_le32(crc);
594 ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
597 sctx->total_send_size += sctx->send_size;
598 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
605 * Sends a move instruction to user space
607 static int send_rename(struct send_ctx *sctx,
608 struct fs_path *from, struct fs_path *to)
612 verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
614 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
618 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
619 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
621 ret = send_cmd(sctx);
629 * Sends a link instruction to user space
631 static int send_link(struct send_ctx *sctx,
632 struct fs_path *path, struct fs_path *lnk)
636 verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
638 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
642 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
643 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
645 ret = send_cmd(sctx);
653 * Sends an unlink instruction to user space
655 static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
659 verbose_printk("btrfs: send_unlink %s\n", path->start);
661 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
665 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
667 ret = send_cmd(sctx);
675 * Sends a rmdir instruction to user space
677 static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
681 verbose_printk("btrfs: send_rmdir %s\n", path->start);
683 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
687 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
689 ret = send_cmd(sctx);
697 * Helper function to retrieve some fields from an inode item.
699 static int get_inode_info(struct btrfs_root *root,
700 u64 ino, u64 *size, u64 *gen,
701 u64 *mode, u64 *uid, u64 *gid,
705 struct btrfs_inode_item *ii;
706 struct btrfs_key key;
707 struct btrfs_path *path;
709 path = alloc_path_for_send();
714 key.type = BTRFS_INODE_ITEM_KEY;
716 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
724 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
725 struct btrfs_inode_item);
727 *size = btrfs_inode_size(path->nodes[0], ii);
729 *gen = btrfs_inode_generation(path->nodes[0], ii);
731 *mode = btrfs_inode_mode(path->nodes[0], ii);
733 *uid = btrfs_inode_uid(path->nodes[0], ii);
735 *gid = btrfs_inode_gid(path->nodes[0], ii);
737 *rdev = btrfs_inode_rdev(path->nodes[0], ii);
740 btrfs_free_path(path);
744 typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
749 * Helper function to iterate the entries in ONE btrfs_inode_ref or
750 * btrfs_inode_extref.
751 * The iterate callback may return a non zero value to stop iteration. This can
752 * be a negative value for error codes or 1 to simply stop it.
754 * path must point to the INODE_REF or INODE_EXTREF when called.
756 static int iterate_inode_ref(struct send_ctx *sctx,
757 struct btrfs_root *root, struct btrfs_path *path,
758 struct btrfs_key *found_key, int resolve,
759 iterate_inode_ref_t iterate, void *ctx)
761 struct extent_buffer *eb = path->nodes[0];
762 struct btrfs_item *item;
763 struct btrfs_inode_ref *iref;
764 struct btrfs_inode_extref *extref;
765 struct btrfs_path *tmp_path;
769 int slot = path->slots[0];
776 unsigned long name_off;
777 unsigned long elem_size;
780 p = fs_path_alloc_reversed(sctx);
784 tmp_path = alloc_path_for_send();
786 fs_path_free(sctx, p);
791 if (found_key->type == BTRFS_INODE_REF_KEY) {
792 ptr = (unsigned long)btrfs_item_ptr(eb, slot,
793 struct btrfs_inode_ref);
794 item = btrfs_item_nr(eb, slot);
795 total = btrfs_item_size(eb, item);
796 elem_size = sizeof(*iref);
798 ptr = btrfs_item_ptr_offset(eb, slot);
799 total = btrfs_item_size_nr(eb, slot);
800 elem_size = sizeof(*extref);
803 while (cur < total) {
806 if (found_key->type == BTRFS_INODE_REF_KEY) {
807 iref = (struct btrfs_inode_ref *)(ptr + cur);
808 name_len = btrfs_inode_ref_name_len(eb, iref);
809 name_off = (unsigned long)(iref + 1);
810 index = btrfs_inode_ref_index(eb, iref);
811 dir = found_key->offset;
813 extref = (struct btrfs_inode_extref *)(ptr + cur);
814 name_len = btrfs_inode_extref_name_len(eb, extref);
815 name_off = (unsigned long)&extref->name;
816 index = btrfs_inode_extref_index(eb, extref);
817 dir = btrfs_inode_extref_parent(eb, extref);
821 start = btrfs_ref_to_path(root, tmp_path, name_len,
825 ret = PTR_ERR(start);
828 if (start < p->buf) {
829 /* overflow , try again with larger buffer */
830 ret = fs_path_ensure_buf(p,
831 p->buf_len + p->buf - start);
834 start = btrfs_ref_to_path(root, tmp_path,
839 ret = PTR_ERR(start);
842 BUG_ON(start < p->buf);
846 ret = fs_path_add_from_extent_buffer(p, eb, name_off,
852 cur += elem_size + name_len;
853 ret = iterate(num, dir, index, p, ctx);
860 btrfs_free_path(tmp_path);
861 fs_path_free(sctx, p);
865 typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
866 const char *name, int name_len,
867 const char *data, int data_len,
871 * Helper function to iterate the entries in ONE btrfs_dir_item.
872 * The iterate callback may return a non zero value to stop iteration. This can
873 * be a negative value for error codes or 1 to simply stop it.
875 * path must point to the dir item when called.
877 static int iterate_dir_item(struct send_ctx *sctx,
878 struct btrfs_root *root, struct btrfs_path *path,
879 struct btrfs_key *found_key,
880 iterate_dir_item_t iterate, void *ctx)
883 struct extent_buffer *eb;
884 struct btrfs_item *item;
885 struct btrfs_dir_item *di;
886 struct btrfs_key di_key;
901 buf = kmalloc(buf_len, GFP_NOFS);
908 slot = path->slots[0];
909 item = btrfs_item_nr(eb, slot);
910 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
913 total = btrfs_item_size(eb, item);
916 while (cur < total) {
917 name_len = btrfs_dir_name_len(eb, di);
918 data_len = btrfs_dir_data_len(eb, di);
919 type = btrfs_dir_type(eb, di);
920 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
922 if (name_len + data_len > buf_len) {
923 buf_len = PAGE_ALIGN(name_len + data_len);
925 buf2 = vmalloc(buf_len);
932 buf2 = krealloc(buf, buf_len, GFP_NOFS);
934 buf2 = vmalloc(buf_len);
948 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
949 name_len + data_len);
951 len = sizeof(*di) + name_len + data_len;
952 di = (struct btrfs_dir_item *)((char *)di + len);
955 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
956 data_len, type, ctx);
975 static int __copy_first_ref(int num, u64 dir, int index,
976 struct fs_path *p, void *ctx)
979 struct fs_path *pt = ctx;
981 ret = fs_path_copy(pt, p);
985 /* we want the first only */
990 * Retrieve the first path of an inode. If an inode has more then one
991 * ref/hardlink, this is ignored.
993 static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
994 u64 ino, struct fs_path *path)
997 struct btrfs_key key, found_key;
998 struct btrfs_path *p;
1000 p = alloc_path_for_send();
1004 fs_path_reset(path);
1007 key.type = BTRFS_INODE_REF_KEY;
1010 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
1017 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
1018 if (found_key.objectid != ino ||
1019 (found_key.type != BTRFS_INODE_REF_KEY &&
1020 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1025 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1026 __copy_first_ref, path);
1036 struct backref_ctx {
1037 struct send_ctx *sctx;
1039 /* number of total found references */
1043 * used for clones found in send_root. clones found behind cur_objectid
1044 * and cur_offset are not considered as allowed clones.
1049 /* may be truncated in case it's the last extent in a file */
1052 /* Just to check for bugs in backref resolving */
1056 static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1058 u64 root = (u64)(uintptr_t)key;
1059 struct clone_root *cr = (struct clone_root *)elt;
1061 if (root < cr->root->objectid)
1063 if (root > cr->root->objectid)
1068 static int __clone_root_cmp_sort(const void *e1, const void *e2)
1070 struct clone_root *cr1 = (struct clone_root *)e1;
1071 struct clone_root *cr2 = (struct clone_root *)e2;
1073 if (cr1->root->objectid < cr2->root->objectid)
1075 if (cr1->root->objectid > cr2->root->objectid)
1081 * Called for every backref that is found for the current extent.
1082 * Results are collected in sctx->clone_roots->ino/offset/found_refs
1084 static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1086 struct backref_ctx *bctx = ctx_;
1087 struct clone_root *found;
1091 /* First check if the root is in the list of accepted clone sources */
1092 found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
1093 bctx->sctx->clone_roots_cnt,
1094 sizeof(struct clone_root),
1095 __clone_root_cmp_bsearch);
1099 if (found->root == bctx->sctx->send_root &&
1100 ino == bctx->cur_objectid &&
1101 offset == bctx->cur_offset) {
1102 bctx->found_itself = 1;
1106 * There are inodes that have extents that lie behind its i_size. Don't
1107 * accept clones from these extents.
1109 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL,
1114 if (offset + bctx->extent_len > i_size)
1118 * Make sure we don't consider clones from send_root that are
1119 * behind the current inode/offset.
1121 if (found->root == bctx->sctx->send_root) {
1123 * TODO for the moment we don't accept clones from the inode
1124 * that is currently send. We may change this when
1125 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1128 if (ino >= bctx->cur_objectid)
1131 if (ino > bctx->cur_objectid)
1133 if (offset + bctx->extent_len > bctx->cur_offset)
1139 found->found_refs++;
1140 if (ino < found->ino) {
1142 found->offset = offset;
1143 } else if (found->ino == ino) {
1145 * same extent found more then once in the same file.
1147 if (found->offset > offset + bctx->extent_len)
1148 found->offset = offset;
1155 * Given an inode, offset and extent item, it finds a good clone for a clone
1156 * instruction. Returns -ENOENT when none could be found. The function makes
1157 * sure that the returned clone is usable at the point where sending is at the
1158 * moment. This means, that no clones are accepted which lie behind the current
1161 * path must point to the extent item when called.
1163 static int find_extent_clone(struct send_ctx *sctx,
1164 struct btrfs_path *path,
1165 u64 ino, u64 data_offset,
1167 struct clone_root **found)
1174 u64 extent_item_pos;
1176 struct btrfs_file_extent_item *fi;
1177 struct extent_buffer *eb = path->nodes[0];
1178 struct backref_ctx *backref_ctx = NULL;
1179 struct clone_root *cur_clone_root;
1180 struct btrfs_key found_key;
1181 struct btrfs_path *tmp_path;
1185 tmp_path = alloc_path_for_send();
1189 backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_NOFS);
1195 if (data_offset >= ino_size) {
1197 * There may be extents that lie behind the file's size.
1198 * I at least had this in combination with snapshotting while
1199 * writing large files.
1205 fi = btrfs_item_ptr(eb, path->slots[0],
1206 struct btrfs_file_extent_item);
1207 extent_type = btrfs_file_extent_type(eb, fi);
1208 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1212 compressed = btrfs_file_extent_compression(eb, fi);
1214 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1215 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
1216 if (disk_byte == 0) {
1220 logical = disk_byte + btrfs_file_extent_offset(eb, fi);
1222 ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
1223 &found_key, &flags);
1224 btrfs_release_path(tmp_path);
1228 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1234 * Setup the clone roots.
1236 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1237 cur_clone_root = sctx->clone_roots + i;
1238 cur_clone_root->ino = (u64)-1;
1239 cur_clone_root->offset = 0;
1240 cur_clone_root->found_refs = 0;
1243 backref_ctx->sctx = sctx;
1244 backref_ctx->found = 0;
1245 backref_ctx->cur_objectid = ino;
1246 backref_ctx->cur_offset = data_offset;
1247 backref_ctx->found_itself = 0;
1248 backref_ctx->extent_len = num_bytes;
1251 * The last extent of a file may be too large due to page alignment.
1252 * We need to adjust extent_len in this case so that the checks in
1253 * __iterate_backrefs work.
1255 if (data_offset + num_bytes >= ino_size)
1256 backref_ctx->extent_len = ino_size - data_offset;
1259 * Now collect all backrefs.
1261 if (compressed == BTRFS_COMPRESS_NONE)
1262 extent_item_pos = logical - found_key.objectid;
1264 extent_item_pos = 0;
1266 extent_item_pos = logical - found_key.objectid;
1267 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1268 found_key.objectid, extent_item_pos, 1,
1269 __iterate_backrefs, backref_ctx);
1274 if (!backref_ctx->found_itself) {
1275 /* found a bug in backref code? */
1277 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1278 "send_root. inode=%llu, offset=%llu, "
1279 "disk_byte=%llu found extent=%llu\n",
1280 ino, data_offset, disk_byte, found_key.objectid);
1284 verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1286 "num_bytes=%llu, logical=%llu\n",
1287 data_offset, ino, num_bytes, logical);
1289 if (!backref_ctx->found)
1290 verbose_printk("btrfs: no clones found\n");
1292 cur_clone_root = NULL;
1293 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1294 if (sctx->clone_roots[i].found_refs) {
1295 if (!cur_clone_root)
1296 cur_clone_root = sctx->clone_roots + i;
1297 else if (sctx->clone_roots[i].root == sctx->send_root)
1298 /* prefer clones from send_root over others */
1299 cur_clone_root = sctx->clone_roots + i;
1304 if (cur_clone_root) {
1305 *found = cur_clone_root;
1312 btrfs_free_path(tmp_path);
1317 static int read_symlink(struct send_ctx *sctx,
1318 struct btrfs_root *root,
1320 struct fs_path *dest)
1323 struct btrfs_path *path;
1324 struct btrfs_key key;
1325 struct btrfs_file_extent_item *ei;
1331 path = alloc_path_for_send();
1336 key.type = BTRFS_EXTENT_DATA_KEY;
1338 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1343 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1344 struct btrfs_file_extent_item);
1345 type = btrfs_file_extent_type(path->nodes[0], ei);
1346 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1347 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1348 BUG_ON(compression);
1350 off = btrfs_file_extent_inline_start(ei);
1351 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1353 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1356 btrfs_free_path(path);
1361 * Helper function to generate a file name that is unique in the root of
1362 * send_root and parent_root. This is used to generate names for orphan inodes.
1364 static int gen_unique_name(struct send_ctx *sctx,
1366 struct fs_path *dest)
1369 struct btrfs_path *path;
1370 struct btrfs_dir_item *di;
1375 path = alloc_path_for_send();
1380 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1382 if (len >= sizeof(tmp)) {
1383 /* should really not happen */
1388 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1389 path, BTRFS_FIRST_FREE_OBJECTID,
1390 tmp, strlen(tmp), 0);
1391 btrfs_release_path(path);
1397 /* not unique, try again */
1402 if (!sctx->parent_root) {
1408 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1409 path, BTRFS_FIRST_FREE_OBJECTID,
1410 tmp, strlen(tmp), 0);
1411 btrfs_release_path(path);
1417 /* not unique, try again */
1425 ret = fs_path_add(dest, tmp, strlen(tmp));
1428 btrfs_free_path(path);
1433 inode_state_no_change,
1434 inode_state_will_create,
1435 inode_state_did_create,
1436 inode_state_will_delete,
1437 inode_state_did_delete,
1440 static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1448 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1450 if (ret < 0 && ret != -ENOENT)
1454 if (!sctx->parent_root) {
1455 right_ret = -ENOENT;
1457 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1458 NULL, NULL, NULL, NULL);
1459 if (ret < 0 && ret != -ENOENT)
1464 if (!left_ret && !right_ret) {
1465 if (left_gen == gen && right_gen == gen) {
1466 ret = inode_state_no_change;
1467 } else if (left_gen == gen) {
1468 if (ino < sctx->send_progress)
1469 ret = inode_state_did_create;
1471 ret = inode_state_will_create;
1472 } else if (right_gen == gen) {
1473 if (ino < sctx->send_progress)
1474 ret = inode_state_did_delete;
1476 ret = inode_state_will_delete;
1480 } else if (!left_ret) {
1481 if (left_gen == gen) {
1482 if (ino < sctx->send_progress)
1483 ret = inode_state_did_create;
1485 ret = inode_state_will_create;
1489 } else if (!right_ret) {
1490 if (right_gen == gen) {
1491 if (ino < sctx->send_progress)
1492 ret = inode_state_did_delete;
1494 ret = inode_state_will_delete;
1506 static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1510 ret = get_cur_inode_state(sctx, ino, gen);
1514 if (ret == inode_state_no_change ||
1515 ret == inode_state_did_create ||
1516 ret == inode_state_will_delete)
1526 * Helper function to lookup a dir item in a dir.
1528 static int lookup_dir_item_inode(struct btrfs_root *root,
1529 u64 dir, const char *name, int name_len,
1534 struct btrfs_dir_item *di;
1535 struct btrfs_key key;
1536 struct btrfs_path *path;
1538 path = alloc_path_for_send();
1542 di = btrfs_lookup_dir_item(NULL, root, path,
1543 dir, name, name_len, 0);
1552 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1553 if (key.type == BTRFS_ROOT_ITEM_KEY) {
1557 *found_inode = key.objectid;
1558 *found_type = btrfs_dir_type(path->nodes[0], di);
1561 btrfs_free_path(path);
1566 * Looks up the first btrfs_inode_ref of a given ino. It returns the parent dir,
1567 * generation of the parent dir and the name of the dir entry.
1569 static int get_first_ref(struct send_ctx *sctx,
1570 struct btrfs_root *root, u64 ino,
1571 u64 *dir, u64 *dir_gen, struct fs_path *name)
1574 struct btrfs_key key;
1575 struct btrfs_key found_key;
1576 struct btrfs_path *path;
1580 path = alloc_path_for_send();
1585 key.type = BTRFS_INODE_REF_KEY;
1588 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1592 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1594 if (ret || found_key.objectid != ino ||
1595 (found_key.type != BTRFS_INODE_REF_KEY &&
1596 found_key.type != BTRFS_INODE_EXTREF_KEY)) {
1601 if (key.type == BTRFS_INODE_REF_KEY) {
1602 struct btrfs_inode_ref *iref;
1603 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1604 struct btrfs_inode_ref);
1605 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1606 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1607 (unsigned long)(iref + 1),
1609 parent_dir = found_key.offset;
1611 struct btrfs_inode_extref *extref;
1612 extref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1613 struct btrfs_inode_extref);
1614 len = btrfs_inode_extref_name_len(path->nodes[0], extref);
1615 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1616 (unsigned long)&extref->name, len);
1617 parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref);
1621 btrfs_release_path(path);
1623 ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, NULL,
1631 btrfs_free_path(path);
1635 static int is_first_ref(struct send_ctx *sctx,
1636 struct btrfs_root *root,
1638 const char *name, int name_len)
1641 struct fs_path *tmp_name;
1645 tmp_name = fs_path_alloc(sctx);
1649 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1653 if (dir != tmp_dir || name_len != fs_path_len(tmp_name)) {
1658 ret = !memcmp(tmp_name->start, name, name_len);
1661 fs_path_free(sctx, tmp_name);
1666 * Used by process_recorded_refs to determine if a new ref would overwrite an
1667 * already existing ref. In case it detects an overwrite, it returns the
1668 * inode/gen in who_ino/who_gen.
1669 * When an overwrite is detected, process_recorded_refs does proper orphanizing
1670 * to make sure later references to the overwritten inode are possible.
1671 * Orphanizing is however only required for the first ref of an inode.
1672 * process_recorded_refs does an additional is_first_ref check to see if
1673 * orphanizing is really required.
1675 static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1676 const char *name, int name_len,
1677 u64 *who_ino, u64 *who_gen)
1680 u64 other_inode = 0;
1683 if (!sctx->parent_root)
1686 ret = is_inode_existent(sctx, dir, dir_gen);
1690 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1691 &other_inode, &other_type);
1692 if (ret < 0 && ret != -ENOENT)
1700 * Check if the overwritten ref was already processed. If yes, the ref
1701 * was already unlinked/moved, so we can safely assume that we will not
1702 * overwrite anything at this point in time.
1704 if (other_inode > sctx->send_progress) {
1705 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1706 who_gen, NULL, NULL, NULL, NULL);
1711 *who_ino = other_inode;
1721 * Checks if the ref was overwritten by an already processed inode. This is
1722 * used by __get_cur_name_and_parent to find out if the ref was orphanized and
1723 * thus the orphan name needs be used.
1724 * process_recorded_refs also uses it to avoid unlinking of refs that were
1727 static int did_overwrite_ref(struct send_ctx *sctx,
1728 u64 dir, u64 dir_gen,
1729 u64 ino, u64 ino_gen,
1730 const char *name, int name_len)
1737 if (!sctx->parent_root)
1740 ret = is_inode_existent(sctx, dir, dir_gen);
1744 /* check if the ref was overwritten by another ref */
1745 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1746 &ow_inode, &other_type);
1747 if (ret < 0 && ret != -ENOENT)
1750 /* was never and will never be overwritten */
1755 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1760 if (ow_inode == ino && gen == ino_gen) {
1765 /* we know that it is or will be overwritten. check this now */
1766 if (ow_inode < sctx->send_progress)
1776 * Same as did_overwrite_ref, but also checks if it is the first ref of an inode
1777 * that got overwritten. This is used by process_recorded_refs to determine
1778 * if it has to use the path as returned by get_cur_path or the orphan name.
1780 static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1783 struct fs_path *name = NULL;
1787 if (!sctx->parent_root)
1790 name = fs_path_alloc(sctx);
1794 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1798 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1799 name->start, fs_path_len(name));
1802 fs_path_free(sctx, name);
1807 * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit,
1808 * so we need to do some special handling in case we have clashes. This function
1809 * takes care of this with the help of name_cache_entry::radix_list.
1810 * In case of error, nce is kfreed.
1812 static int name_cache_insert(struct send_ctx *sctx,
1813 struct name_cache_entry *nce)
1816 struct list_head *nce_head;
1818 nce_head = radix_tree_lookup(&sctx->name_cache,
1819 (unsigned long)nce->ino);
1821 nce_head = kmalloc(sizeof(*nce_head), GFP_NOFS);
1826 INIT_LIST_HEAD(nce_head);
1828 ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head);
1835 list_add_tail(&nce->radix_list, nce_head);
1836 list_add_tail(&nce->list, &sctx->name_cache_list);
1837 sctx->name_cache_size++;
1842 static void name_cache_delete(struct send_ctx *sctx,
1843 struct name_cache_entry *nce)
1845 struct list_head *nce_head;
1847 nce_head = radix_tree_lookup(&sctx->name_cache,
1848 (unsigned long)nce->ino);
1851 list_del(&nce->radix_list);
1852 list_del(&nce->list);
1853 sctx->name_cache_size--;
1855 if (list_empty(nce_head)) {
1856 radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino);
1861 static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1864 struct list_head *nce_head;
1865 struct name_cache_entry *cur;
1867 nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino);
1871 list_for_each_entry(cur, nce_head, radix_list) {
1872 if (cur->ino == ino && cur->gen == gen)
1879 * Removes the entry from the list and adds it back to the end. This marks the
1880 * entry as recently used so that name_cache_clean_unused does not remove it.
1882 static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1884 list_del(&nce->list);
1885 list_add_tail(&nce->list, &sctx->name_cache_list);
1889 * Remove some entries from the beginning of name_cache_list.
1891 static void name_cache_clean_unused(struct send_ctx *sctx)
1893 struct name_cache_entry *nce;
1895 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1898 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1899 nce = list_entry(sctx->name_cache_list.next,
1900 struct name_cache_entry, list);
1901 name_cache_delete(sctx, nce);
1906 static void name_cache_free(struct send_ctx *sctx)
1908 struct name_cache_entry *nce;
1910 while (!list_empty(&sctx->name_cache_list)) {
1911 nce = list_entry(sctx->name_cache_list.next,
1912 struct name_cache_entry, list);
1913 name_cache_delete(sctx, nce);
1919 * Used by get_cur_path for each ref up to the root.
1920 * Returns 0 if it succeeded.
1921 * Returns 1 if the inode is not existent or got overwritten. In that case, the
1922 * name is an orphan name. This instructs get_cur_path to stop iterating. If 1
1923 * is returned, parent_ino/parent_gen are not guaranteed to be valid.
1924 * Returns <0 in case of error.
1926 static int __get_cur_name_and_parent(struct send_ctx *sctx,
1930 struct fs_path *dest)
1934 struct btrfs_path *path = NULL;
1935 struct name_cache_entry *nce = NULL;
1938 * First check if we already did a call to this function with the same
1939 * ino/gen. If yes, check if the cache entry is still up-to-date. If yes
1940 * return the cached result.
1942 nce = name_cache_search(sctx, ino, gen);
1944 if (ino < sctx->send_progress && nce->need_later_update) {
1945 name_cache_delete(sctx, nce);
1949 name_cache_used(sctx, nce);
1950 *parent_ino = nce->parent_ino;
1951 *parent_gen = nce->parent_gen;
1952 ret = fs_path_add(dest, nce->name, nce->name_len);
1960 path = alloc_path_for_send();
1965 * If the inode is not existent yet, add the orphan name and return 1.
1966 * This should only happen for the parent dir that we determine in
1969 ret = is_inode_existent(sctx, ino, gen);
1974 ret = gen_unique_name(sctx, ino, gen, dest);
1982 * Depending on whether the inode was already processed or not, use
1983 * send_root or parent_root for ref lookup.
1985 if (ino < sctx->send_progress)
1986 ret = get_first_ref(sctx, sctx->send_root, ino,
1987 parent_ino, parent_gen, dest);
1989 ret = get_first_ref(sctx, sctx->parent_root, ino,
1990 parent_ino, parent_gen, dest);
1995 * Check if the ref was overwritten by an inode's ref that was processed
1996 * earlier. If yes, treat as orphan and return 1.
1998 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1999 dest->start, dest->end - dest->start);
2003 fs_path_reset(dest);
2004 ret = gen_unique_name(sctx, ino, gen, dest);
2012 * Store the result of the lookup in the name cache.
2014 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
2022 nce->parent_ino = *parent_ino;
2023 nce->parent_gen = *parent_gen;
2024 nce->name_len = fs_path_len(dest);
2026 strcpy(nce->name, dest->start);
2028 if (ino < sctx->send_progress)
2029 nce->need_later_update = 0;
2031 nce->need_later_update = 1;
2033 nce_ret = name_cache_insert(sctx, nce);
2036 name_cache_clean_unused(sctx);
2039 btrfs_free_path(path);
2044 * Magic happens here. This function returns the first ref to an inode as it
2045 * would look like while receiving the stream at this point in time.
2046 * We walk the path up to the root. For every inode in between, we check if it
2047 * was already processed/sent. If yes, we continue with the parent as found
2048 * in send_root. If not, we continue with the parent as found in parent_root.
2049 * If we encounter an inode that was deleted at this point in time, we use the
2050 * inodes "orphan" name instead of the real name and stop. Same with new inodes
2051 * that were not created yet and overwritten inodes/refs.
2053 * When do we have have orphan inodes:
2054 * 1. When an inode is freshly created and thus no valid refs are available yet
2055 * 2. When a directory lost all it's refs (deleted) but still has dir items
2056 * inside which were not processed yet (pending for move/delete). If anyone
2057 * tried to get the path to the dir items, it would get a path inside that
2059 * 3. When an inode is moved around or gets new links, it may overwrite the ref
2060 * of an unprocessed inode. If in that case the first ref would be
2061 * overwritten, the overwritten inode gets "orphanized". Later when we
2062 * process this overwritten inode, it is restored at a new place by moving
2065 * sctx->send_progress tells this function at which point in time receiving
2068 static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
2069 struct fs_path *dest)
2072 struct fs_path *name = NULL;
2073 u64 parent_inode = 0;
2077 name = fs_path_alloc(sctx);
2084 fs_path_reset(dest);
2086 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
2087 fs_path_reset(name);
2089 ret = __get_cur_name_and_parent(sctx, ino, gen,
2090 &parent_inode, &parent_gen, name);
2096 ret = fs_path_add_path(dest, name);
2105 fs_path_free(sctx, name);
2107 fs_path_unreverse(dest);
2112 * Called for regular files when sending extents data. Opens a struct file
2113 * to read from the file.
2115 static int open_cur_inode_file(struct send_ctx *sctx)
2118 struct btrfs_key key;
2120 struct inode *inode;
2121 struct dentry *dentry;
2125 if (sctx->cur_inode_filp)
2128 key.objectid = sctx->cur_ino;
2129 key.type = BTRFS_INODE_ITEM_KEY;
2132 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2134 if (IS_ERR(inode)) {
2135 ret = PTR_ERR(inode);
2139 dentry = d_obtain_alias(inode);
2141 if (IS_ERR(dentry)) {
2142 ret = PTR_ERR(dentry);
2146 path.mnt = sctx->mnt;
2147 path.dentry = dentry;
2148 filp = dentry_open(&path, O_RDONLY | O_LARGEFILE, current_cred());
2152 ret = PTR_ERR(filp);
2155 sctx->cur_inode_filp = filp;
2159 * no xxxput required here as every vfs op
2160 * does it by itself on failure
2166 * Closes the struct file that was created in open_cur_inode_file
2168 static int close_cur_inode_file(struct send_ctx *sctx)
2172 if (!sctx->cur_inode_filp)
2175 ret = filp_close(sctx->cur_inode_filp, NULL);
2176 sctx->cur_inode_filp = NULL;
2183 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2185 static int send_subvol_begin(struct send_ctx *sctx)
2188 struct btrfs_root *send_root = sctx->send_root;
2189 struct btrfs_root *parent_root = sctx->parent_root;
2190 struct btrfs_path *path;
2191 struct btrfs_key key;
2192 struct btrfs_root_ref *ref;
2193 struct extent_buffer *leaf;
2197 path = alloc_path_for_send();
2201 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2203 btrfs_free_path(path);
2207 key.objectid = send_root->objectid;
2208 key.type = BTRFS_ROOT_BACKREF_KEY;
2211 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2220 leaf = path->nodes[0];
2221 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2222 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2223 key.objectid != send_root->objectid) {
2227 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2228 namelen = btrfs_root_ref_name_len(leaf, ref);
2229 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2230 btrfs_release_path(path);
2233 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2237 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2242 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2243 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2244 sctx->send_root->root_item.uuid);
2245 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2246 sctx->send_root->root_item.ctransid);
2248 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2249 sctx->parent_root->root_item.uuid);
2250 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2251 sctx->parent_root->root_item.ctransid);
2254 ret = send_cmd(sctx);
2258 btrfs_free_path(path);
2263 static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2268 verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2270 p = fs_path_alloc(sctx);
2274 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2278 ret = get_cur_path(sctx, ino, gen, p);
2281 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2282 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2284 ret = send_cmd(sctx);
2288 fs_path_free(sctx, p);
2292 static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2297 verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2299 p = fs_path_alloc(sctx);
2303 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2307 ret = get_cur_path(sctx, ino, gen, p);
2310 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2311 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2313 ret = send_cmd(sctx);
2317 fs_path_free(sctx, p);
2321 static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2326 verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2328 p = fs_path_alloc(sctx);
2332 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2336 ret = get_cur_path(sctx, ino, gen, p);
2339 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2340 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2341 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2343 ret = send_cmd(sctx);
2347 fs_path_free(sctx, p);
2351 static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2354 struct fs_path *p = NULL;
2355 struct btrfs_inode_item *ii;
2356 struct btrfs_path *path = NULL;
2357 struct extent_buffer *eb;
2358 struct btrfs_key key;
2361 verbose_printk("btrfs: send_utimes %llu\n", ino);
2363 p = fs_path_alloc(sctx);
2367 path = alloc_path_for_send();
2374 key.type = BTRFS_INODE_ITEM_KEY;
2376 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2380 eb = path->nodes[0];
2381 slot = path->slots[0];
2382 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2384 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2388 ret = get_cur_path(sctx, ino, gen, p);
2391 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2392 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2393 btrfs_inode_atime(ii));
2394 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2395 btrfs_inode_mtime(ii));
2396 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2397 btrfs_inode_ctime(ii));
2398 /* TODO Add otime support when the otime patches get into upstream */
2400 ret = send_cmd(sctx);
2404 fs_path_free(sctx, p);
2405 btrfs_free_path(path);
2410 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2411 * a valid path yet because we did not process the refs yet. So, the inode
2412 * is created as orphan.
2414 static int send_create_inode(struct send_ctx *sctx, u64 ino)
2423 verbose_printk("btrfs: send_create_inode %llu\n", ino);
2425 p = fs_path_alloc(sctx);
2429 ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, NULL,
2434 if (S_ISREG(mode)) {
2435 cmd = BTRFS_SEND_C_MKFILE;
2436 } else if (S_ISDIR(mode)) {
2437 cmd = BTRFS_SEND_C_MKDIR;
2438 } else if (S_ISLNK(mode)) {
2439 cmd = BTRFS_SEND_C_SYMLINK;
2440 } else if (S_ISCHR(mode) || S_ISBLK(mode)) {
2441 cmd = BTRFS_SEND_C_MKNOD;
2442 } else if (S_ISFIFO(mode)) {
2443 cmd = BTRFS_SEND_C_MKFIFO;
2444 } else if (S_ISSOCK(mode)) {
2445 cmd = BTRFS_SEND_C_MKSOCK;
2447 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2448 (int)(mode & S_IFMT));
2453 ret = begin_cmd(sctx, cmd);
2457 ret = gen_unique_name(sctx, ino, gen, p);
2461 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2462 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, ino);
2464 if (S_ISLNK(mode)) {
2466 ret = read_symlink(sctx, sctx->send_root, ino, p);
2469 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2470 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2471 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2472 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, new_encode_dev(rdev));
2473 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode);
2476 ret = send_cmd(sctx);
2483 fs_path_free(sctx, p);
2488 * We need some special handling for inodes that get processed before the parent
2489 * directory got created. See process_recorded_refs for details.
2490 * This function does the check if we already created the dir out of order.
2492 static int did_create_dir(struct send_ctx *sctx, u64 dir)
2495 struct btrfs_path *path = NULL;
2496 struct btrfs_key key;
2497 struct btrfs_key found_key;
2498 struct btrfs_key di_key;
2499 struct extent_buffer *eb;
2500 struct btrfs_dir_item *di;
2503 path = alloc_path_for_send();
2510 key.type = BTRFS_DIR_INDEX_KEY;
2513 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2518 eb = path->nodes[0];
2519 slot = path->slots[0];
2520 btrfs_item_key_to_cpu(eb, &found_key, slot);
2522 if (ret || found_key.objectid != key.objectid ||
2523 found_key.type != key.type) {
2528 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2529 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2531 if (di_key.type != BTRFS_ROOT_ITEM_KEY &&
2532 di_key.objectid < sctx->send_progress) {
2537 key.offset = found_key.offset + 1;
2538 btrfs_release_path(path);
2542 btrfs_free_path(path);
2547 * Only creates the inode if it is:
2548 * 1. Not a directory
2549 * 2. Or a directory which was not created already due to out of order
2550 * directories. See did_create_dir and process_recorded_refs for details.
2552 static int send_create_inode_if_needed(struct send_ctx *sctx)
2556 if (S_ISDIR(sctx->cur_inode_mode)) {
2557 ret = did_create_dir(sctx, sctx->cur_ino);
2566 ret = send_create_inode(sctx, sctx->cur_ino);
2574 struct recorded_ref {
2575 struct list_head list;
2578 struct fs_path *full_path;
2586 * We need to process new refs before deleted refs, but compare_tree gives us
2587 * everything mixed. So we first record all refs and later process them.
2588 * This function is a helper to record one ref.
2590 static int record_ref(struct list_head *head, u64 dir,
2591 u64 dir_gen, struct fs_path *path)
2593 struct recorded_ref *ref;
2596 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2601 ref->dir_gen = dir_gen;
2602 ref->full_path = path;
2604 tmp = strrchr(ref->full_path->start, '/');
2606 ref->name_len = ref->full_path->end - ref->full_path->start;
2607 ref->name = ref->full_path->start;
2608 ref->dir_path_len = 0;
2609 ref->dir_path = ref->full_path->start;
2612 ref->name_len = ref->full_path->end - tmp;
2614 ref->dir_path = ref->full_path->start;
2615 ref->dir_path_len = ref->full_path->end -
2616 ref->full_path->start - 1 - ref->name_len;
2619 list_add_tail(&ref->list, head);
2623 static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
2625 struct recorded_ref *cur;
2627 while (!list_empty(head)) {
2628 cur = list_entry(head->next, struct recorded_ref, list);
2629 fs_path_free(sctx, cur->full_path);
2630 list_del(&cur->list);
2635 static void free_recorded_refs(struct send_ctx *sctx)
2637 __free_recorded_refs(sctx, &sctx->new_refs);
2638 __free_recorded_refs(sctx, &sctx->deleted_refs);
2642 * Renames/moves a file/dir to its orphan name. Used when the first
2643 * ref of an unprocessed inode gets overwritten and for all non empty
2646 static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2647 struct fs_path *path)
2650 struct fs_path *orphan;
2652 orphan = fs_path_alloc(sctx);
2656 ret = gen_unique_name(sctx, ino, gen, orphan);
2660 ret = send_rename(sctx, path, orphan);
2663 fs_path_free(sctx, orphan);
2668 * Returns 1 if a directory can be removed at this point in time.
2669 * We check this by iterating all dir items and checking if the inode behind
2670 * the dir item was already processed.
2672 static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2675 struct btrfs_root *root = sctx->parent_root;
2676 struct btrfs_path *path;
2677 struct btrfs_key key;
2678 struct btrfs_key found_key;
2679 struct btrfs_key loc;
2680 struct btrfs_dir_item *di;
2683 * Don't try to rmdir the top/root subvolume dir.
2685 if (dir == BTRFS_FIRST_FREE_OBJECTID)
2688 path = alloc_path_for_send();
2693 key.type = BTRFS_DIR_INDEX_KEY;
2697 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2701 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2704 if (ret || found_key.objectid != key.objectid ||
2705 found_key.type != key.type) {
2709 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2710 struct btrfs_dir_item);
2711 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2713 if (loc.objectid > send_progress) {
2718 btrfs_release_path(path);
2719 key.offset = found_key.offset + 1;
2725 btrfs_free_path(path);
2730 * This does all the move/link/unlink/rmdir magic.
2732 static int process_recorded_refs(struct send_ctx *sctx)
2735 struct recorded_ref *cur;
2736 struct recorded_ref *cur2;
2737 struct ulist *check_dirs = NULL;
2738 struct ulist_iterator uit;
2739 struct ulist_node *un;
2740 struct fs_path *valid_path = NULL;
2743 int did_overwrite = 0;
2746 verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2749 * This should never happen as the root dir always has the same ref
2750 * which is always '..'
2752 BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID);
2754 valid_path = fs_path_alloc(sctx);
2760 check_dirs = ulist_alloc(GFP_NOFS);
2767 * First, check if the first ref of the current inode was overwritten
2768 * before. If yes, we know that the current inode was already orphanized
2769 * and thus use the orphan name. If not, we can use get_cur_path to
2770 * get the path of the first ref as it would like while receiving at
2771 * this point in time.
2772 * New inodes are always orphan at the beginning, so force to use the
2773 * orphan name in this case.
2774 * The first ref is stored in valid_path and will be updated if it
2775 * gets moved around.
2777 if (!sctx->cur_inode_new) {
2778 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2779 sctx->cur_inode_gen);
2785 if (sctx->cur_inode_new || did_overwrite) {
2786 ret = gen_unique_name(sctx, sctx->cur_ino,
2787 sctx->cur_inode_gen, valid_path);
2792 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2798 list_for_each_entry(cur, &sctx->new_refs, list) {
2800 * We may have refs where the parent directory does not exist
2801 * yet. This happens if the parent directories inum is higher
2802 * the the current inum. To handle this case, we create the
2803 * parent directory out of order. But we need to check if this
2804 * did already happen before due to other refs in the same dir.
2806 ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
2809 if (ret == inode_state_will_create) {
2812 * First check if any of the current inodes refs did
2813 * already create the dir.
2815 list_for_each_entry(cur2, &sctx->new_refs, list) {
2818 if (cur2->dir == cur->dir) {
2825 * If that did not happen, check if a previous inode
2826 * did already create the dir.
2829 ret = did_create_dir(sctx, cur->dir);
2833 ret = send_create_inode(sctx, cur->dir);
2840 * Check if this new ref would overwrite the first ref of
2841 * another unprocessed inode. If yes, orphanize the
2842 * overwritten inode. If we find an overwritten ref that is
2843 * not the first ref, simply unlink it.
2845 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2846 cur->name, cur->name_len,
2847 &ow_inode, &ow_gen);
2851 ret = is_first_ref(sctx, sctx->parent_root,
2852 ow_inode, cur->dir, cur->name,
2857 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2862 ret = send_unlink(sctx, cur->full_path);
2869 * link/move the ref to the new place. If we have an orphan
2870 * inode, move it and update valid_path. If not, link or move
2871 * it depending on the inode mode.
2874 ret = send_rename(sctx, valid_path, cur->full_path);
2878 ret = fs_path_copy(valid_path, cur->full_path);
2882 if (S_ISDIR(sctx->cur_inode_mode)) {
2884 * Dirs can't be linked, so move it. For moved
2885 * dirs, we always have one new and one deleted
2886 * ref. The deleted ref is ignored later.
2888 ret = send_rename(sctx, valid_path,
2892 ret = fs_path_copy(valid_path, cur->full_path);
2896 ret = send_link(sctx, cur->full_path,
2902 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2908 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2910 * Check if we can already rmdir the directory. If not,
2911 * orphanize it. For every dir item inside that gets deleted
2912 * later, we do this check again and rmdir it then if possible.
2913 * See the use of check_dirs for more details.
2915 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2919 ret = send_rmdir(sctx, valid_path);
2922 } else if (!is_orphan) {
2923 ret = orphanize_inode(sctx, sctx->cur_ino,
2924 sctx->cur_inode_gen, valid_path);
2930 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2931 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2936 } else if (S_ISDIR(sctx->cur_inode_mode) &&
2937 !list_empty(&sctx->deleted_refs)) {
2939 * We have a moved dir. Add the old parent to check_dirs
2941 cur = list_entry(sctx->deleted_refs.next, struct recorded_ref,
2943 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2947 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2949 * We have a non dir inode. Go through all deleted refs and
2950 * unlink them if they were not already overwritten by other
2953 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2954 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2955 sctx->cur_ino, sctx->cur_inode_gen,
2956 cur->name, cur->name_len);
2960 ret = send_unlink(sctx, cur->full_path);
2964 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2971 * If the inode is still orphan, unlink the orphan. This may
2972 * happen when a previous inode did overwrite the first ref
2973 * of this inode and no new refs were added for the current
2974 * inode. Unlinking does not mean that the inode is deleted in
2975 * all cases. There may still be links to this inode in other
2979 ret = send_unlink(sctx, valid_path);
2986 * We did collect all parent dirs where cur_inode was once located. We
2987 * now go through all these dirs and check if they are pending for
2988 * deletion and if it's finally possible to perform the rmdir now.
2989 * We also update the inode stats of the parent dirs here.
2991 ULIST_ITER_INIT(&uit);
2992 while ((un = ulist_next(check_dirs, &uit))) {
2994 * In case we had refs into dirs that were not processed yet,
2995 * we don't need to do the utime and rmdir logic for these dirs.
2996 * The dir will be processed later.
2998 if (un->val > sctx->cur_ino)
3001 ret = get_cur_inode_state(sctx, un->val, un->aux);
3005 if (ret == inode_state_did_create ||
3006 ret == inode_state_no_change) {
3007 /* TODO delayed utimes */
3008 ret = send_utimes(sctx, un->val, un->aux);
3011 } else if (ret == inode_state_did_delete) {
3012 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
3016 ret = get_cur_path(sctx, un->val, un->aux,
3020 ret = send_rmdir(sctx, valid_path);
3030 free_recorded_refs(sctx);
3031 ulist_free(check_dirs);
3032 fs_path_free(sctx, valid_path);
3036 static int __record_new_ref(int num, u64 dir, int index,
3037 struct fs_path *name,
3041 struct send_ctx *sctx = ctx;
3045 p = fs_path_alloc(sctx);
3049 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
3054 ret = get_cur_path(sctx, dir, gen, p);
3057 ret = fs_path_add_path(p, name);
3061 ret = record_ref(&sctx->new_refs, dir, gen, p);
3065 fs_path_free(sctx, p);
3069 static int __record_deleted_ref(int num, u64 dir, int index,
3070 struct fs_path *name,
3074 struct send_ctx *sctx = ctx;
3078 p = fs_path_alloc(sctx);
3082 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3087 ret = get_cur_path(sctx, dir, gen, p);
3090 ret = fs_path_add_path(p, name);
3094 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3098 fs_path_free(sctx, p);
3102 static int record_new_ref(struct send_ctx *sctx)
3106 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3107 sctx->cmp_key, 0, __record_new_ref, sctx);
3116 static int record_deleted_ref(struct send_ctx *sctx)
3120 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3121 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3130 struct find_ref_ctx {
3132 struct fs_path *name;
3136 static int __find_iref(int num, u64 dir, int index,
3137 struct fs_path *name,
3140 struct find_ref_ctx *ctx = ctx_;
3142 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3143 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3144 ctx->found_idx = num;
3150 static int find_iref(struct send_ctx *sctx,
3151 struct btrfs_root *root,
3152 struct btrfs_path *path,
3153 struct btrfs_key *key,
3154 u64 dir, struct fs_path *name)
3157 struct find_ref_ctx ctx;
3163 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3167 if (ctx.found_idx == -1)
3170 return ctx.found_idx;
3173 static int __record_changed_new_ref(int num, u64 dir, int index,
3174 struct fs_path *name,
3178 struct send_ctx *sctx = ctx;
3180 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3181 sctx->cmp_key, dir, name);
3183 ret = __record_new_ref(num, dir, index, name, sctx);
3190 static int __record_changed_deleted_ref(int num, u64 dir, int index,
3191 struct fs_path *name,
3195 struct send_ctx *sctx = ctx;
3197 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3200 ret = __record_deleted_ref(num, dir, index, name, sctx);
3207 static int record_changed_ref(struct send_ctx *sctx)
3211 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3212 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3215 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3216 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3226 * Record and process all refs at once. Needed when an inode changes the
3227 * generation number, which means that it was deleted and recreated.
3229 static int process_all_refs(struct send_ctx *sctx,
3230 enum btrfs_compare_tree_result cmd)
3233 struct btrfs_root *root;
3234 struct btrfs_path *path;
3235 struct btrfs_key key;
3236 struct btrfs_key found_key;
3237 struct extent_buffer *eb;
3239 iterate_inode_ref_t cb;
3241 path = alloc_path_for_send();
3245 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3246 root = sctx->send_root;
3247 cb = __record_new_ref;
3248 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3249 root = sctx->parent_root;
3250 cb = __record_deleted_ref;
3255 key.objectid = sctx->cmp_key->objectid;
3256 key.type = BTRFS_INODE_REF_KEY;
3259 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3265 eb = path->nodes[0];
3266 slot = path->slots[0];
3267 btrfs_item_key_to_cpu(eb, &found_key, slot);
3269 if (found_key.objectid != key.objectid ||
3270 (found_key.type != BTRFS_INODE_REF_KEY &&
3271 found_key.type != BTRFS_INODE_EXTREF_KEY))
3274 ret = iterate_inode_ref(sctx, root, path, &found_key, 0, cb,
3276 btrfs_release_path(path);
3280 key.offset = found_key.offset + 1;
3282 btrfs_release_path(path);
3284 ret = process_recorded_refs(sctx);
3287 btrfs_free_path(path);
3291 static int send_set_xattr(struct send_ctx *sctx,
3292 struct fs_path *path,
3293 const char *name, int name_len,
3294 const char *data, int data_len)
3298 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3302 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3303 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3304 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3306 ret = send_cmd(sctx);
3313 static int send_remove_xattr(struct send_ctx *sctx,
3314 struct fs_path *path,
3315 const char *name, int name_len)
3319 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3323 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3324 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3326 ret = send_cmd(sctx);
3333 static int __process_new_xattr(int num, struct btrfs_key *di_key,
3334 const char *name, int name_len,
3335 const char *data, int data_len,
3339 struct send_ctx *sctx = ctx;
3341 posix_acl_xattr_header dummy_acl;
3343 p = fs_path_alloc(sctx);
3348 * This hack is needed because empty acl's are stored as zero byte
3349 * data in xattrs. Problem with that is, that receiving these zero byte
3350 * acl's will fail later. To fix this, we send a dummy acl list that
3351 * only contains the version number and no entries.
3353 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3354 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3355 if (data_len == 0) {
3356 dummy_acl.a_version =
3357 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3358 data = (char *)&dummy_acl;
3359 data_len = sizeof(dummy_acl);
3363 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3367 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3370 fs_path_free(sctx, p);
3374 static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3375 const char *name, int name_len,
3376 const char *data, int data_len,
3380 struct send_ctx *sctx = ctx;
3383 p = fs_path_alloc(sctx);
3387 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3391 ret = send_remove_xattr(sctx, p, name, name_len);
3394 fs_path_free(sctx, p);
3398 static int process_new_xattr(struct send_ctx *sctx)
3402 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3403 sctx->cmp_key, __process_new_xattr, sctx);
3408 static int process_deleted_xattr(struct send_ctx *sctx)
3412 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3413 sctx->cmp_key, __process_deleted_xattr, sctx);
3418 struct find_xattr_ctx {
3426 static int __find_xattr(int num, struct btrfs_key *di_key,
3427 const char *name, int name_len,
3428 const char *data, int data_len,
3429 u8 type, void *vctx)
3431 struct find_xattr_ctx *ctx = vctx;
3433 if (name_len == ctx->name_len &&
3434 strncmp(name, ctx->name, name_len) == 0) {
3435 ctx->found_idx = num;
3436 ctx->found_data_len = data_len;
3437 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3438 if (!ctx->found_data)
3440 memcpy(ctx->found_data, data, data_len);
3446 static int find_xattr(struct send_ctx *sctx,
3447 struct btrfs_root *root,
3448 struct btrfs_path *path,
3449 struct btrfs_key *key,
3450 const char *name, int name_len,
3451 char **data, int *data_len)
3454 struct find_xattr_ctx ctx;
3457 ctx.name_len = name_len;
3459 ctx.found_data = NULL;
3460 ctx.found_data_len = 0;
3462 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3466 if (ctx.found_idx == -1)
3469 *data = ctx.found_data;
3470 *data_len = ctx.found_data_len;
3472 kfree(ctx.found_data);
3474 return ctx.found_idx;
3478 static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3479 const char *name, int name_len,
3480 const char *data, int data_len,
3484 struct send_ctx *sctx = ctx;
3485 char *found_data = NULL;
3486 int found_data_len = 0;
3488 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3489 sctx->cmp_key, name, name_len, &found_data,
3491 if (ret == -ENOENT) {
3492 ret = __process_new_xattr(num, di_key, name, name_len, data,
3493 data_len, type, ctx);
3494 } else if (ret >= 0) {
3495 if (data_len != found_data_len ||
3496 memcmp(data, found_data, data_len)) {
3497 ret = __process_new_xattr(num, di_key, name, name_len,
3498 data, data_len, type, ctx);
3508 static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3509 const char *name, int name_len,
3510 const char *data, int data_len,
3514 struct send_ctx *sctx = ctx;
3516 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3517 name, name_len, NULL, NULL);
3519 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3520 data_len, type, ctx);
3527 static int process_changed_xattr(struct send_ctx *sctx)
3531 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3532 sctx->cmp_key, __process_changed_new_xattr, sctx);
3535 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3536 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3542 static int process_all_new_xattrs(struct send_ctx *sctx)
3545 struct btrfs_root *root;
3546 struct btrfs_path *path;
3547 struct btrfs_key key;
3548 struct btrfs_key found_key;
3549 struct extent_buffer *eb;
3552 path = alloc_path_for_send();
3556 root = sctx->send_root;
3558 key.objectid = sctx->cmp_key->objectid;
3559 key.type = BTRFS_XATTR_ITEM_KEY;
3562 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3570 eb = path->nodes[0];
3571 slot = path->slots[0];
3572 btrfs_item_key_to_cpu(eb, &found_key, slot);
3574 if (found_key.objectid != key.objectid ||
3575 found_key.type != key.type) {
3580 ret = iterate_dir_item(sctx, root, path, &found_key,
3581 __process_new_xattr, sctx);
3585 btrfs_release_path(path);
3586 key.offset = found_key.offset + 1;
3590 btrfs_free_path(path);
3595 * Read some bytes from the current inode/file and send a write command to
3598 static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3602 loff_t pos = offset;
3604 mm_segment_t old_fs;
3606 p = fs_path_alloc(sctx);
3611 * vfs normally only accepts user space buffers for security reasons.
3612 * we only read from the file and also only provide the read_buf buffer
3613 * to vfs. As this buffer does not come from a user space call, it's
3614 * ok to temporary allow kernel space buffers.
3619 verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3621 ret = open_cur_inode_file(sctx);
3625 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3632 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3636 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3640 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3641 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3642 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
3644 ret = send_cmd(sctx);
3648 fs_path_free(sctx, p);
3656 * Send a clone command to user space.
3658 static int send_clone(struct send_ctx *sctx,
3659 u64 offset, u32 len,
3660 struct clone_root *clone_root)
3666 verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3667 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3668 clone_root->root->objectid, clone_root->ino,
3669 clone_root->offset);
3671 p = fs_path_alloc(sctx);
3675 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3679 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3683 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3684 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3685 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3687 if (clone_root->root == sctx->send_root) {
3688 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
3689 &gen, NULL, NULL, NULL, NULL);
3692 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3694 ret = get_inode_path(sctx, clone_root->root,
3695 clone_root->ino, p);
3700 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
3701 clone_root->root->root_item.uuid);
3702 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
3703 clone_root->root->root_item.ctransid);
3704 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3705 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3706 clone_root->offset);
3708 ret = send_cmd(sctx);
3712 fs_path_free(sctx, p);
3717 * Send an update extent command to user space.
3719 static int send_update_extent(struct send_ctx *sctx,
3720 u64 offset, u32 len)
3725 p = fs_path_alloc(sctx);
3729 ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT);
3733 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3737 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3738 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3739 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len);
3741 ret = send_cmd(sctx);
3745 fs_path_free(sctx, p);
3749 static int send_write_or_clone(struct send_ctx *sctx,
3750 struct btrfs_path *path,
3751 struct btrfs_key *key,
3752 struct clone_root *clone_root)
3755 struct btrfs_file_extent_item *ei;
3756 u64 offset = key->offset;
3762 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3763 struct btrfs_file_extent_item);
3764 type = btrfs_file_extent_type(path->nodes[0], ei);
3765 if (type == BTRFS_FILE_EXTENT_INLINE) {
3766 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3768 * it is possible the inline item won't cover the whole page,
3769 * but there may be items after this page. Make
3770 * sure to send the whole thing
3772 len = PAGE_CACHE_ALIGN(len);
3774 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3777 if (offset + len > sctx->cur_inode_size)
3778 len = sctx->cur_inode_size - offset;
3785 ret = send_clone(sctx, offset, len, clone_root);
3786 } else if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) {
3787 ret = send_update_extent(sctx, offset, len);
3791 if (l > BTRFS_SEND_READ_SIZE)
3792 l = BTRFS_SEND_READ_SIZE;
3793 ret = send_write(sctx, pos + offset, l);
3806 static int is_extent_unchanged(struct send_ctx *sctx,
3807 struct btrfs_path *left_path,
3808 struct btrfs_key *ekey)
3811 struct btrfs_key key;
3812 struct btrfs_path *path = NULL;
3813 struct extent_buffer *eb;
3815 struct btrfs_key found_key;
3816 struct btrfs_file_extent_item *ei;
3821 u64 left_offset_fixed;
3829 path = alloc_path_for_send();
3833 eb = left_path->nodes[0];
3834 slot = left_path->slots[0];
3835 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3836 left_type = btrfs_file_extent_type(eb, ei);
3838 if (left_type != BTRFS_FILE_EXTENT_REG) {
3842 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3843 left_len = btrfs_file_extent_num_bytes(eb, ei);
3844 left_offset = btrfs_file_extent_offset(eb, ei);
3845 left_gen = btrfs_file_extent_generation(eb, ei);
3848 * Following comments will refer to these graphics. L is the left
3849 * extents which we are checking at the moment. 1-8 are the right
3850 * extents that we iterate.
3853 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3856 * |--1--|-2b-|...(same as above)
3858 * Alternative situation. Happens on files where extents got split.
3860 * |-----------7-----------|-6-|
3862 * Alternative situation. Happens on files which got larger.
3865 * Nothing follows after 8.
3868 key.objectid = ekey->objectid;
3869 key.type = BTRFS_EXTENT_DATA_KEY;
3870 key.offset = ekey->offset;
3871 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3880 * Handle special case where the right side has no extents at all.
3882 eb = path->nodes[0];
3883 slot = path->slots[0];
3884 btrfs_item_key_to_cpu(eb, &found_key, slot);
3885 if (found_key.objectid != key.objectid ||
3886 found_key.type != key.type) {
3892 * We're now on 2a, 2b or 7.
3895 while (key.offset < ekey->offset + left_len) {
3896 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3897 right_type = btrfs_file_extent_type(eb, ei);
3898 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3899 right_len = btrfs_file_extent_num_bytes(eb, ei);
3900 right_offset = btrfs_file_extent_offset(eb, ei);
3901 right_gen = btrfs_file_extent_generation(eb, ei);
3903 if (right_type != BTRFS_FILE_EXTENT_REG) {
3909 * Are we at extent 8? If yes, we know the extent is changed.
3910 * This may only happen on the first iteration.
3912 if (found_key.offset + right_len <= ekey->offset) {
3917 left_offset_fixed = left_offset;
3918 if (key.offset < ekey->offset) {
3919 /* Fix the right offset for 2a and 7. */
3920 right_offset += ekey->offset - key.offset;
3922 /* Fix the left offset for all behind 2a and 2b */
3923 left_offset_fixed += key.offset - ekey->offset;
3927 * Check if we have the same extent.
3929 if (left_disknr != right_disknr ||
3930 left_offset_fixed != right_offset ||
3931 left_gen != right_gen) {
3937 * Go to the next extent.
3939 ret = btrfs_next_item(sctx->parent_root, path);
3943 eb = path->nodes[0];
3944 slot = path->slots[0];
3945 btrfs_item_key_to_cpu(eb, &found_key, slot);
3947 if (ret || found_key.objectid != key.objectid ||
3948 found_key.type != key.type) {
3949 key.offset += right_len;
3952 if (found_key.offset != key.offset + right_len) {
3960 * We're now behind the left extent (treat as unchanged) or at the end
3961 * of the right side (treat as changed).
3963 if (key.offset >= ekey->offset + left_len)
3970 btrfs_free_path(path);
3974 static int process_extent(struct send_ctx *sctx,
3975 struct btrfs_path *path,
3976 struct btrfs_key *key)
3979 struct clone_root *found_clone = NULL;
3981 if (S_ISLNK(sctx->cur_inode_mode))
3984 if (sctx->parent_root && !sctx->cur_inode_new) {
3985 ret = is_extent_unchanged(sctx, path, key);
3994 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3995 sctx->cur_inode_size, &found_clone);
3996 if (ret != -ENOENT && ret < 0)
3999 ret = send_write_or_clone(sctx, path, key, found_clone);
4005 static int process_all_extents(struct send_ctx *sctx)
4008 struct btrfs_root *root;
4009 struct btrfs_path *path;
4010 struct btrfs_key key;
4011 struct btrfs_key found_key;
4012 struct extent_buffer *eb;
4015 root = sctx->send_root;
4016 path = alloc_path_for_send();
4020 key.objectid = sctx->cmp_key->objectid;
4021 key.type = BTRFS_EXTENT_DATA_KEY;
4024 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
4032 eb = path->nodes[0];
4033 slot = path->slots[0];
4034 btrfs_item_key_to_cpu(eb, &found_key, slot);
4036 if (found_key.objectid != key.objectid ||
4037 found_key.type != key.type) {
4042 ret = process_extent(sctx, path, &found_key);
4046 btrfs_release_path(path);
4047 key.offset = found_key.offset + 1;
4051 btrfs_free_path(path);
4055 static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
4059 if (sctx->cur_ino == 0)
4061 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
4062 sctx->cmp_key->type <= BTRFS_INODE_EXTREF_KEY)
4064 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
4067 ret = process_recorded_refs(sctx);
4072 * We have processed the refs and thus need to advance send_progress.
4073 * Now, calls to get_cur_xxx will take the updated refs of the current
4074 * inode into account.
4076 sctx->send_progress = sctx->cur_ino + 1;
4082 static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
4094 ret = process_recorded_refs_if_needed(sctx, at_end);
4098 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4100 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4103 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4104 &left_mode, &left_uid, &left_gid, NULL);
4108 if (!sctx->parent_root || sctx->cur_inode_new) {
4110 if (!S_ISLNK(sctx->cur_inode_mode))
4113 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4114 NULL, NULL, &right_mode, &right_uid,
4119 if (left_uid != right_uid || left_gid != right_gid)
4121 if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode)
4125 if (S_ISREG(sctx->cur_inode_mode)) {
4126 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4127 sctx->cur_inode_size);
4133 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4134 left_uid, left_gid);
4139 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4146 * Need to send that every time, no matter if it actually changed
4147 * between the two trees as we have done changes to the inode before.
4149 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4157 static int changed_inode(struct send_ctx *sctx,
4158 enum btrfs_compare_tree_result result)
4161 struct btrfs_key *key = sctx->cmp_key;
4162 struct btrfs_inode_item *left_ii = NULL;
4163 struct btrfs_inode_item *right_ii = NULL;
4167 ret = close_cur_inode_file(sctx);
4171 sctx->cur_ino = key->objectid;
4172 sctx->cur_inode_new_gen = 0;
4175 * Set send_progress to current inode. This will tell all get_cur_xxx
4176 * functions that the current inode's refs are not updated yet. Later,
4177 * when process_recorded_refs is finished, it is set to cur_ino + 1.
4179 sctx->send_progress = sctx->cur_ino;
4181 if (result == BTRFS_COMPARE_TREE_NEW ||
4182 result == BTRFS_COMPARE_TREE_CHANGED) {
4183 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4184 sctx->left_path->slots[0],
4185 struct btrfs_inode_item);
4186 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4189 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4190 sctx->right_path->slots[0],
4191 struct btrfs_inode_item);
4192 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4195 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4196 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4197 sctx->right_path->slots[0],
4198 struct btrfs_inode_item);
4200 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4204 * The cur_ino = root dir case is special here. We can't treat
4205 * the inode as deleted+reused because it would generate a
4206 * stream that tries to delete/mkdir the root dir.
4208 if (left_gen != right_gen &&
4209 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4210 sctx->cur_inode_new_gen = 1;
4213 if (result == BTRFS_COMPARE_TREE_NEW) {
4214 sctx->cur_inode_gen = left_gen;
4215 sctx->cur_inode_new = 1;
4216 sctx->cur_inode_deleted = 0;
4217 sctx->cur_inode_size = btrfs_inode_size(
4218 sctx->left_path->nodes[0], left_ii);
4219 sctx->cur_inode_mode = btrfs_inode_mode(
4220 sctx->left_path->nodes[0], left_ii);
4221 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4222 ret = send_create_inode_if_needed(sctx);
4223 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4224 sctx->cur_inode_gen = right_gen;
4225 sctx->cur_inode_new = 0;
4226 sctx->cur_inode_deleted = 1;
4227 sctx->cur_inode_size = btrfs_inode_size(
4228 sctx->right_path->nodes[0], right_ii);
4229 sctx->cur_inode_mode = btrfs_inode_mode(
4230 sctx->right_path->nodes[0], right_ii);
4231 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4233 * We need to do some special handling in case the inode was
4234 * reported as changed with a changed generation number. This
4235 * means that the original inode was deleted and new inode
4236 * reused the same inum. So we have to treat the old inode as
4237 * deleted and the new one as new.
4239 if (sctx->cur_inode_new_gen) {
4241 * First, process the inode as if it was deleted.
4243 sctx->cur_inode_gen = right_gen;
4244 sctx->cur_inode_new = 0;
4245 sctx->cur_inode_deleted = 1;
4246 sctx->cur_inode_size = btrfs_inode_size(
4247 sctx->right_path->nodes[0], right_ii);
4248 sctx->cur_inode_mode = btrfs_inode_mode(
4249 sctx->right_path->nodes[0], right_ii);
4250 ret = process_all_refs(sctx,
4251 BTRFS_COMPARE_TREE_DELETED);
4256 * Now process the inode as if it was new.
4258 sctx->cur_inode_gen = left_gen;
4259 sctx->cur_inode_new = 1;
4260 sctx->cur_inode_deleted = 0;
4261 sctx->cur_inode_size = btrfs_inode_size(
4262 sctx->left_path->nodes[0], left_ii);
4263 sctx->cur_inode_mode = btrfs_inode_mode(
4264 sctx->left_path->nodes[0], left_ii);
4265 ret = send_create_inode_if_needed(sctx);
4269 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4273 * Advance send_progress now as we did not get into
4274 * process_recorded_refs_if_needed in the new_gen case.
4276 sctx->send_progress = sctx->cur_ino + 1;
4279 * Now process all extents and xattrs of the inode as if
4280 * they were all new.
4282 ret = process_all_extents(sctx);
4285 ret = process_all_new_xattrs(sctx);
4289 sctx->cur_inode_gen = left_gen;
4290 sctx->cur_inode_new = 0;
4291 sctx->cur_inode_new_gen = 0;
4292 sctx->cur_inode_deleted = 0;
4293 sctx->cur_inode_size = btrfs_inode_size(
4294 sctx->left_path->nodes[0], left_ii);
4295 sctx->cur_inode_mode = btrfs_inode_mode(
4296 sctx->left_path->nodes[0], left_ii);
4305 * We have to process new refs before deleted refs, but compare_trees gives us
4306 * the new and deleted refs mixed. To fix this, we record the new/deleted refs
4307 * first and later process them in process_recorded_refs.
4308 * For the cur_inode_new_gen case, we skip recording completely because
4309 * changed_inode did already initiate processing of refs. The reason for this is
4310 * that in this case, compare_tree actually compares the refs of 2 different
4311 * inodes. To fix this, process_all_refs is used in changed_inode to handle all
4312 * refs of the right tree as deleted and all refs of the left tree as new.
4314 static int changed_ref(struct send_ctx *sctx,
4315 enum btrfs_compare_tree_result result)
4319 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4321 if (!sctx->cur_inode_new_gen &&
4322 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4323 if (result == BTRFS_COMPARE_TREE_NEW)
4324 ret = record_new_ref(sctx);
4325 else if (result == BTRFS_COMPARE_TREE_DELETED)
4326 ret = record_deleted_ref(sctx);
4327 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4328 ret = record_changed_ref(sctx);
4335 * Process new/deleted/changed xattrs. We skip processing in the
4336 * cur_inode_new_gen case because changed_inode did already initiate processing
4337 * of xattrs. The reason is the same as in changed_ref
4339 static int changed_xattr(struct send_ctx *sctx,
4340 enum btrfs_compare_tree_result result)
4344 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4346 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4347 if (result == BTRFS_COMPARE_TREE_NEW)
4348 ret = process_new_xattr(sctx);
4349 else if (result == BTRFS_COMPARE_TREE_DELETED)
4350 ret = process_deleted_xattr(sctx);
4351 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4352 ret = process_changed_xattr(sctx);
4359 * Process new/deleted/changed extents. We skip processing in the
4360 * cur_inode_new_gen case because changed_inode did already initiate processing
4361 * of extents. The reason is the same as in changed_ref
4363 static int changed_extent(struct send_ctx *sctx,
4364 enum btrfs_compare_tree_result result)
4368 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4370 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4371 if (result != BTRFS_COMPARE_TREE_DELETED)
4372 ret = process_extent(sctx, sctx->left_path,
4380 * Updates compare related fields in sctx and simply forwards to the actual
4381 * changed_xxx functions.
4383 static int changed_cb(struct btrfs_root *left_root,
4384 struct btrfs_root *right_root,
4385 struct btrfs_path *left_path,
4386 struct btrfs_path *right_path,
4387 struct btrfs_key *key,
4388 enum btrfs_compare_tree_result result,
4392 struct send_ctx *sctx = ctx;
4394 sctx->left_path = left_path;
4395 sctx->right_path = right_path;
4396 sctx->cmp_key = key;
4398 ret = finish_inode_if_needed(sctx, 0);
4402 /* Ignore non-FS objects */
4403 if (key->objectid == BTRFS_FREE_INO_OBJECTID ||
4404 key->objectid == BTRFS_FREE_SPACE_OBJECTID)
4407 if (key->type == BTRFS_INODE_ITEM_KEY)
4408 ret = changed_inode(sctx, result);
4409 else if (key->type == BTRFS_INODE_REF_KEY ||
4410 key->type == BTRFS_INODE_EXTREF_KEY)
4411 ret = changed_ref(sctx, result);
4412 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4413 ret = changed_xattr(sctx, result);
4414 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4415 ret = changed_extent(sctx, result);
4421 static int full_send_tree(struct send_ctx *sctx)
4424 struct btrfs_trans_handle *trans = NULL;
4425 struct btrfs_root *send_root = sctx->send_root;
4426 struct btrfs_key key;
4427 struct btrfs_key found_key;
4428 struct btrfs_path *path;
4429 struct extent_buffer *eb;
4434 path = alloc_path_for_send();
4438 spin_lock(&send_root->root_item_lock);
4439 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4440 spin_unlock(&send_root->root_item_lock);
4442 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4443 key.type = BTRFS_INODE_ITEM_KEY;
4448 * We need to make sure the transaction does not get committed
4449 * while we do anything on commit roots. Join a transaction to prevent
4452 trans = btrfs_join_transaction(send_root);
4453 if (IS_ERR(trans)) {
4454 ret = PTR_ERR(trans);
4460 * Make sure the tree has not changed after re-joining. We detect this
4461 * by comparing start_ctransid and ctransid. They should always match.
4463 spin_lock(&send_root->root_item_lock);
4464 ctransid = btrfs_root_ctransid(&send_root->root_item);
4465 spin_unlock(&send_root->root_item_lock);
4467 if (ctransid != start_ctransid) {
4468 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4469 "send was modified in between. This is "
4470 "probably a bug.\n");
4475 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4483 * When someone want to commit while we iterate, end the
4484 * joined transaction and rejoin.
4486 if (btrfs_should_end_transaction(trans, send_root)) {
4487 ret = btrfs_end_transaction(trans, send_root);
4491 btrfs_release_path(path);
4495 eb = path->nodes[0];
4496 slot = path->slots[0];
4497 btrfs_item_key_to_cpu(eb, &found_key, slot);
4499 ret = changed_cb(send_root, NULL, path, NULL,
4500 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4504 key.objectid = found_key.objectid;
4505 key.type = found_key.type;
4506 key.offset = found_key.offset + 1;
4508 ret = btrfs_next_item(send_root, path);
4518 ret = finish_inode_if_needed(sctx, 1);
4521 btrfs_free_path(path);
4524 ret = btrfs_end_transaction(trans, send_root);
4526 btrfs_end_transaction(trans, send_root);
4531 static int send_subvol(struct send_ctx *sctx)
4535 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_STREAM_HEADER)) {
4536 ret = send_header(sctx);
4541 ret = send_subvol_begin(sctx);
4545 if (sctx->parent_root) {
4546 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4550 ret = finish_inode_if_needed(sctx, 1);
4554 ret = full_send_tree(sctx);
4561 ret = close_cur_inode_file(sctx);
4563 close_cur_inode_file(sctx);
4565 free_recorded_refs(sctx);
4569 long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4572 struct btrfs_root *send_root;
4573 struct btrfs_root *clone_root;
4574 struct btrfs_fs_info *fs_info;
4575 struct btrfs_ioctl_send_args *arg = NULL;
4576 struct btrfs_key key;
4577 struct send_ctx *sctx = NULL;
4579 u64 *clone_sources_tmp = NULL;
4581 if (!capable(CAP_SYS_ADMIN))
4584 send_root = BTRFS_I(file_inode(mnt_file))->root;
4585 fs_info = send_root->fs_info;
4588 * This is done when we lookup the root, it should already be complete
4589 * by the time we get here.
4591 WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE);
4594 * If we just created this root we need to make sure that the orphan
4595 * cleanup has been done and committed since we search the commit root,
4596 * so check its commit root transid with our otransid and if they match
4597 * commit the transaction to make sure everything is updated.
4599 down_read(&send_root->fs_info->extent_commit_sem);
4600 if (btrfs_header_generation(send_root->commit_root) ==
4601 btrfs_root_otransid(&send_root->root_item)) {
4602 struct btrfs_trans_handle *trans;
4604 up_read(&send_root->fs_info->extent_commit_sem);
4606 trans = btrfs_attach_transaction_barrier(send_root);
4607 if (IS_ERR(trans)) {
4608 if (PTR_ERR(trans) != -ENOENT) {
4609 ret = PTR_ERR(trans);
4612 /* ENOENT means theres no transaction */
4614 ret = btrfs_commit_transaction(trans, send_root);
4619 up_read(&send_root->fs_info->extent_commit_sem);
4622 arg = memdup_user(arg_, sizeof(*arg));
4629 if (!access_ok(VERIFY_READ, arg->clone_sources,
4630 sizeof(*arg->clone_sources) *
4631 arg->clone_sources_count)) {
4636 if (arg->flags & ~BTRFS_SEND_FLAG_MASK) {
4641 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4647 INIT_LIST_HEAD(&sctx->new_refs);
4648 INIT_LIST_HEAD(&sctx->deleted_refs);
4649 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4650 INIT_LIST_HEAD(&sctx->name_cache_list);
4652 sctx->flags = arg->flags;
4654 sctx->send_filp = fget(arg->send_fd);
4655 if (!sctx->send_filp) {
4660 sctx->mnt = mnt_file->f_path.mnt;
4662 sctx->send_root = send_root;
4663 sctx->clone_roots_cnt = arg->clone_sources_count;
4665 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4666 sctx->send_buf = vmalloc(sctx->send_max_size);
4667 if (!sctx->send_buf) {
4672 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4673 if (!sctx->read_buf) {
4678 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4679 (arg->clone_sources_count + 1));
4680 if (!sctx->clone_roots) {
4685 if (arg->clone_sources_count) {
4686 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4687 sizeof(*arg->clone_sources));
4688 if (!clone_sources_tmp) {
4693 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4694 arg->clone_sources_count *
4695 sizeof(*arg->clone_sources));
4701 for (i = 0; i < arg->clone_sources_count; i++) {
4702 key.objectid = clone_sources_tmp[i];
4703 key.type = BTRFS_ROOT_ITEM_KEY;
4704 key.offset = (u64)-1;
4705 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4710 if (IS_ERR(clone_root)) {
4711 ret = PTR_ERR(clone_root);
4714 sctx->clone_roots[i].root = clone_root;
4716 vfree(clone_sources_tmp);
4717 clone_sources_tmp = NULL;
4720 if (arg->parent_root) {
4721 key.objectid = arg->parent_root;
4722 key.type = BTRFS_ROOT_ITEM_KEY;
4723 key.offset = (u64)-1;
4724 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4725 if (!sctx->parent_root) {
4732 * Clones from send_root are allowed, but only if the clone source
4733 * is behind the current send position. This is checked while searching
4734 * for possible clone sources.
4736 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4738 /* We do a bsearch later */
4739 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4740 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4743 ret = send_subvol(sctx);
4747 if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) {
4748 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4751 ret = send_cmd(sctx);
4758 vfree(clone_sources_tmp);
4761 if (sctx->send_filp)
4762 fput(sctx->send_filp);
4764 vfree(sctx->clone_roots);
4765 vfree(sctx->send_buf);
4766 vfree(sctx->read_buf);
4768 name_cache_free(sctx);