aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorChris Mason <chris.mason@fusionio.com>2012-07-25 19:17:39 -0400
committerChris Mason <chris.mason@fusionio.com>2012-07-25 19:19:10 -0400
commit113c1cb530e10bcada93d88ffaa6b521aae2d251 (patch)
treee41776281314eb8fdc6b07e1533491e5f9ec21af
parentcd1cfc49153ba2bef247e500d8bd4d135193ece9 (diff)
parent31db9f7c23fbf7e95026143f79645de6507b583b (diff)
Merge branch 'send-v2' of git://github.com/ablock84/linux-btrfs into for-linus
This is the kernel portion of btrfs send/receive Conflicts: fs/btrfs/Makefile fs/btrfs/backref.h fs/btrfs/ctree.c fs/btrfs/ioctl.c fs/btrfs/ioctl.h Signed-off-by: Chris Mason <chris.mason@fusionio.com>
Diffstat
-rw-r--r--fs/btrfs/Makefile2
-rw-r--r--fs/btrfs/backref.c10
-rw-r--r--fs/btrfs/backref.h4
-rw-r--r--fs/btrfs/check-integrity.c7
-rw-r--r--fs/btrfs/ctree.c459
-rw-r--r--fs/btrfs/ctree.h62
-rw-r--r--fs/btrfs/disk-io.c8
-rw-r--r--fs/btrfs/inode.c4
-rw-r--r--fs/btrfs/ioctl.c114
-rw-r--r--fs/btrfs/ioctl.h26
-rw-r--r--fs/btrfs/root-tree.c107
-rw-r--r--fs/btrfs/send.c4570
-rw-r--r--fs/btrfs/send.h133
-rw-r--r--fs/btrfs/transaction.c17
14 files changed, 5483 insertions, 40 deletions
diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 0bc4d3a10a5f..d7fcdba141a2 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -8,7 +8,7 @@ btrfs-y += super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
8 extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \ 8 extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
9 export.o tree-log.o free-space-cache.o zlib.o lzo.o \ 9 export.o tree-log.o free-space-cache.o zlib.o lzo.o \
10 compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \ 10 compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \
11 reada.o backref.o ulist.o qgroup.o 11 reada.o backref.o ulist.o qgroup.o send.o
12 12
13btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o 13btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o
14btrfs-$(CONFIG_BTRFS_FS_CHECK_INTEGRITY) += check-integrity.o 14btrfs-$(CONFIG_BTRFS_FS_CHECK_INTEGRITY) += check-integrity.o
diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index 7d80ddd8f544..a256f3b2a845 100644
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -1122,10 +1122,10 @@ static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1122 * required for the path to fit into the buffer. in that case, the returned 1122 * required for the path to fit into the buffer. in that case, the returned
1123 * value will be smaller than dest. callers must check this! 1123 * value will be smaller than dest. callers must check this!
1124 */ 1124 */
1125static char *iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, 1125char *btrfs_iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1126 struct btrfs_inode_ref *iref, 1126 struct btrfs_inode_ref *iref,
1127 struct extent_buffer *eb_in, u64 parent, 1127 struct extent_buffer *eb_in, u64 parent,
1128 char *dest, u32 size) 1128 char *dest, u32 size)
1129{ 1129{
1130 u32 len; 1130 u32 len;
1131 int slot; 1131 int slot;
@@ -1531,7 +1531,7 @@ static int inode_to_path(u64 inum, struct btrfs_inode_ref *iref,
1531 ipath->fspath->bytes_left - s_ptr : 0; 1531 ipath->fspath->bytes_left - s_ptr : 0;
1532 1532
1533 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; 1533 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1534 fspath = iref_to_path(ipath->fs_root, ipath->btrfs_path, iref, eb, 1534 fspath = btrfs_iref_to_path(ipath->fs_root, ipath->btrfs_path, iref, eb,
1535 inum, fspath_min, bytes_left); 1535 inum, fspath_min, bytes_left);
1536 if (IS_ERR(fspath)) 1536 if (IS_ERR(fspath))
1537 return PTR_ERR(fspath); 1537 return PTR_ERR(fspath);
diff --git a/fs/btrfs/backref.h b/fs/btrfs/backref.h
index 3a1ad3e2dcb0..032f4dc7eab8 100644
--- a/fs/btrfs/backref.h
+++ b/fs/btrfs/backref.h
@@ -21,6 +21,7 @@
21 21
22#include "ioctl.h" 22#include "ioctl.h"
23#include "ulist.h" 23#include "ulist.h"
24#include "extent_io.h"
24 25
25#define BTRFS_BACKREF_SEARCH_COMMIT_ROOT ((struct btrfs_trans_handle *)0) 26#define BTRFS_BACKREF_SEARCH_COMMIT_ROOT ((struct btrfs_trans_handle *)0)
26 27
@@ -59,6 +60,9 @@ int paths_from_inode(u64 inum, struct inode_fs_paths *ipath);
59int btrfs_find_all_roots(struct btrfs_trans_handle *trans, 60int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
60 struct btrfs_fs_info *fs_info, u64 bytenr, 61 struct btrfs_fs_info *fs_info, u64 bytenr,
61 u64 time_seq, struct ulist **roots); 62 u64 time_seq, struct ulist **roots);
63char *btrfs_iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
64 struct btrfs_inode_ref *iref, struct extent_buffer *eb,
65 u64 parent, char *dest, u32 size);
62 66
63struct btrfs_data_container *init_data_container(u32 total_bytes); 67struct btrfs_data_container *init_data_container(u32 total_bytes);
64struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, 68struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
diff --git a/fs/btrfs/check-integrity.c b/fs/btrfs/check-integrity.c
index da6e9364a5e3..9197e2e33407 100644
--- a/fs/btrfs/check-integrity.c
+++ b/fs/btrfs/check-integrity.c
@@ -1032,6 +1032,7 @@ continue_with_current_leaf_stack_frame:
1032 struct btrfs_disk_key *disk_key; 1032 struct btrfs_disk_key *disk_key;
1033 u8 type; 1033 u8 type;
1034 u32 item_offset; 1034 u32 item_offset;
1035 u32 item_size;
1035 1036
1036 if (disk_item_offset + sizeof(struct btrfs_item) > 1037 if (disk_item_offset + sizeof(struct btrfs_item) >
1037 sf->block_ctx->len) { 1038 sf->block_ctx->len) {
@@ -1047,6 +1048,7 @@ leaf_item_out_of_bounce_error:
1047 disk_item_offset, 1048 disk_item_offset,
1048 sizeof(struct btrfs_item)); 1049 sizeof(struct btrfs_item));
1049 item_offset = le32_to_cpu(disk_item.offset); 1050 item_offset = le32_to_cpu(disk_item.offset);
1051 item_size = le32_to_cpu(disk_item.size);
1050 disk_key = &disk_item.key; 1052 disk_key = &disk_item.key;
1051 type = disk_key->type; 1053 type = disk_key->type;
1052 1054
@@ -1057,14 +1059,13 @@ leaf_item_out_of_bounce_error:
1057 1059
1058 root_item_offset = item_offset + 1060 root_item_offset = item_offset +
1059 offsetof(struct btrfs_leaf, items); 1061 offsetof(struct btrfs_leaf, items);
1060 if (root_item_offset + 1062 if (root_item_offset + item_size >
1061 sizeof(struct btrfs_root_item) >
1062 sf->block_ctx->len) 1063 sf->block_ctx->len)
1063 goto leaf_item_out_of_bounce_error; 1064 goto leaf_item_out_of_bounce_error;
1064 btrfsic_read_from_block_data( 1065 btrfsic_read_from_block_data(
1065 sf->block_ctx, &root_item, 1066 sf->block_ctx, &root_item,
1066 root_item_offset, 1067 root_item_offset,
1067 sizeof(struct btrfs_root_item)); 1068 item_size);
1068 next_bytenr = le64_to_cpu(root_item.bytenr); 1069 next_bytenr = le64_to_cpu(root_item.bytenr);
1069 1070
1070 sf->error = 1071 sf->error =
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index fb21431fe4e0..9d7621f271ff 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -2837,23 +2837,25 @@ again:
2837 goto again; 2837 goto again;
2838 } 2838 }
2839 } else { 2839 } else {
2840 if (p->slots[0] >= btrfs_header_nritems(leaf)) { 2840 if (p->slots[0] == 0) {
2841 /* we're sitting on an invalid slot */ 2841 ret = btrfs_prev_leaf(root, p);
2842 if (p->slots[0] == 0) { 2842 if (ret < 0)
2843 ret = btrfs_prev_leaf(root, p); 2843 return ret;
2844 if (ret <= 0) 2844 if (!ret) {
2845 return ret; 2845 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2846 if (!return_any) 2846 return 0;
2847 return 1;
2848 /*
2849 * no lower item found, return the next
2850 * higher instead
2851 */
2852 return_any = 0;
2853 find_higher = 1;
2854 btrfs_release_path(p);
2855 goto again;
2856 } 2847 }
2848 if (!return_any)
2849 return 1;
2850 /*
2851 * no lower item found, return the next
2852 * higher instead
2853 */
2854 return_any = 0;
2855 find_higher = 1;
2856 btrfs_release_path(p);
2857 goto again;
2858 } else {
2857 --p->slots[0]; 2859 --p->slots[0];
2858 } 2860 }
2859 } 2861 }
@@ -5070,6 +5072,431 @@ out:
5070 return ret; 5072 return ret;
5071} 5073}
5072 5074
5075static void tree_move_down(struct btrfs_root *root,
5076 struct btrfs_path *path,
5077 int *level, int root_level)
5078{
5079 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5080 path->slots[*level]);
5081 path->slots[*level - 1] = 0;
5082 (*level)--;
5083}
5084
5085static int tree_move_next_or_upnext(struct btrfs_root *root,
5086 struct btrfs_path *path,
5087 int *level, int root_level)
5088{
5089 int ret = 0;
5090 int nritems;
5091 nritems = btrfs_header_nritems(path->nodes[*level]);
5092
5093 path->slots[*level]++;
5094
5095 while (path->slots[*level] == nritems) {
5096 if (*level == root_level)
5097 return -1;
5098
5099 /* move upnext */
5100 path->slots[*level] = 0;
5101 free_extent_buffer(path->nodes[*level]);
5102 path->nodes[*level] = NULL;
5103 (*level)++;
5104 path->slots[*level]++;
5105
5106 nritems = btrfs_header_nritems(path->nodes[*level]);
5107 ret = 1;
5108 }
5109 return ret;
5110}
5111
5112/*
5113 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5114 * or down.
5115 */
5116static int tree_advance(struct btrfs_root *root,
5117 struct btrfs_path *path,
5118 int *level, int root_level,
5119 int allow_down,
5120 struct btrfs_key *key)
5121{
5122 int ret;
5123
5124 if (*level == 0 || !allow_down) {
5125 ret = tree_move_next_or_upnext(root, path, level, root_level);
5126 } else {
5127 tree_move_down(root, path, level, root_level);
5128 ret = 0;
5129 }
5130 if (ret >= 0) {
5131 if (*level == 0)
5132 btrfs_item_key_to_cpu(path->nodes[*level], key,
5133 path->slots[*level]);
5134 else
5135 btrfs_node_key_to_cpu(path->nodes[*level], key,
5136 path->slots[*level]);
5137 }
5138 return ret;
5139}
5140
5141static int tree_compare_item(struct btrfs_root *left_root,
5142 struct btrfs_path *left_path,
5143 struct btrfs_path *right_path,
5144 char *tmp_buf)
5145{
5146 int cmp;
5147 int len1, len2;
5148 unsigned long off1, off2;
5149
5150 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5151 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5152 if (len1 != len2)
5153 return 1;
5154
5155 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5156 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5157 right_path->slots[0]);
5158
5159 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5160
5161 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5162 if (cmp)
5163 return 1;
5164 return 0;
5165}
5166
5167#define ADVANCE 1
5168#define ADVANCE_ONLY_NEXT -1
5169
5170/*
5171 * This function compares two trees and calls the provided callback for
5172 * every changed/new/deleted item it finds.
5173 * If shared tree blocks are encountered, whole subtrees are skipped, making
5174 * the compare pretty fast on snapshotted subvolumes.
5175 *
5176 * This currently works on commit roots only. As commit roots are read only,
5177 * we don't do any locking. The commit roots are protected with transactions.
5178 * Transactions are ended and rejoined when a commit is tried in between.
5179 *
5180 * This function checks for modifications done to the trees while comparing.
5181 * If it detects a change, it aborts immediately.
5182 */
5183int btrfs_compare_trees(struct btrfs_root *left_root,
5184 struct btrfs_root *right_root,
5185 btrfs_changed_cb_t changed_cb, void *ctx)
5186{
5187 int ret;
5188 int cmp;
5189 struct btrfs_trans_handle *trans = NULL;
5190 struct btrfs_path *left_path = NULL;
5191 struct btrfs_path *right_path = NULL;
5192 struct btrfs_key left_key;
5193 struct btrfs_key right_key;
5194 char *tmp_buf = NULL;
5195 int left_root_level;
5196 int right_root_level;
5197 int left_level;
5198 int right_level;
5199 int left_end_reached;
5200 int right_end_reached;
5201 int advance_left;
5202 int advance_right;
5203 u64 left_blockptr;
5204 u64 right_blockptr;
5205 u64 left_start_ctransid;
5206 u64 right_start_ctransid;
5207 u64 ctransid;
5208
5209 left_path = btrfs_alloc_path();
5210 if (!left_path) {
5211 ret = -ENOMEM;
5212 goto out;
5213 }
5214 right_path = btrfs_alloc_path();
5215 if (!right_path) {
5216 ret = -ENOMEM;
5217 goto out;
5218 }
5219
5220 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5221 if (!tmp_buf) {
5222 ret = -ENOMEM;
5223 goto out;
5224 }
5225
5226 left_path->search_commit_root = 1;
5227 left_path->skip_locking = 1;
5228 right_path->search_commit_root = 1;
5229 right_path->skip_locking = 1;
5230
5231 spin_lock(&left_root->root_times_lock);
5232 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5233 spin_unlock(&left_root->root_times_lock);
5234
5235 spin_lock(&right_root->root_times_lock);
5236 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5237 spin_unlock(&right_root->root_times_lock);
5238
5239 trans = btrfs_join_transaction(left_root);
5240 if (IS_ERR(trans)) {
5241 ret = PTR_ERR(trans);
5242 trans = NULL;
5243 goto out;
5244 }
5245
5246 /*
5247 * Strategy: Go to the first items of both trees. Then do
5248 *
5249 * If both trees are at level 0
5250 * Compare keys of current items
5251 * If left < right treat left item as new, advance left tree
5252 * and repeat
5253 * If left > right treat right item as deleted, advance right tree
5254 * and repeat
5255 * If left == right do deep compare of items, treat as changed if
5256 * needed, advance both trees and repeat
5257 * If both trees are at the same level but not at level 0
5258 * Compare keys of current nodes/leafs
5259 * If left < right advance left tree and repeat
5260 * If left > right advance right tree and repeat
5261 * If left == right compare blockptrs of the next nodes/leafs
5262 * If they match advance both trees but stay at the same level
5263 * and repeat
5264 * If they don't match advance both trees while allowing to go
5265 * deeper and repeat
5266 * If tree levels are different
5267 * Advance the tree that needs it and repeat
5268 *
5269 * Advancing a tree means:
5270 * If we are at level 0, try to go to the next slot. If that's not
5271 * possible, go one level up and repeat. Stop when we found a level
5272 * where we could go to the next slot. We may at this point be on a
5273 * node or a leaf.
5274 *
5275 * If we are not at level 0 and not on shared tree blocks, go one
5276 * level deeper.
5277 *
5278 * If we are not at level 0 and on shared tree blocks, go one slot to
5279 * the right if possible or go up and right.
5280 */
5281
5282 left_level = btrfs_header_level(left_root->commit_root);
5283 left_root_level = left_level;
5284 left_path->nodes[left_level] = left_root->commit_root;
5285 extent_buffer_get(left_path->nodes[left_level]);
5286
5287 right_level = btrfs_header_level(right_root->commit_root);
5288 right_root_level = right_level;
5289 right_path->nodes[right_level] = right_root->commit_root;
5290 extent_buffer_get(right_path->nodes[right_level]);
5291
5292 if (left_level == 0)
5293 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5294 &left_key, left_path->slots[left_level]);
5295 else
5296 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5297 &left_key, left_path->slots[left_level]);
5298 if (right_level == 0)
5299 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5300 &right_key, right_path->slots[right_level]);
5301 else
5302 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5303 &right_key, right_path->slots[right_level]);
5304
5305 left_end_reached = right_end_reached = 0;
5306 advance_left = advance_right = 0;
5307
5308 while (1) {
5309 /*
5310 * We need to make sure the transaction does not get committed
5311 * while we do anything on commit roots. This means, we need to
5312 * join and leave transactions for every item that we process.
5313 */
5314 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5315 btrfs_release_path(left_path);
5316 btrfs_release_path(right_path);
5317
5318 ret = btrfs_end_transaction(trans, left_root);
5319 trans = NULL;
5320 if (ret < 0)
5321 goto out;
5322 }
5323 /* now rejoin the transaction */
5324 if (!trans) {
5325 trans = btrfs_join_transaction(left_root);
5326 if (IS_ERR(trans)) {
5327 ret = PTR_ERR(trans);
5328 trans = NULL;
5329 goto out;
5330 }
5331
5332 spin_lock(&left_root->root_times_lock);
5333 ctransid = btrfs_root_ctransid(&left_root->root_item);
5334 spin_unlock(&left_root->root_times_lock);
5335 if (ctransid != left_start_ctransid)
5336 left_start_ctransid = 0;
5337
5338 spin_lock(&right_root->root_times_lock);
5339 ctransid = btrfs_root_ctransid(&right_root->root_item);
5340 spin_unlock(&right_root->root_times_lock);
5341 if (ctransid != right_start_ctransid)
5342 right_start_ctransid = 0;
5343
5344 if (!left_start_ctransid || !right_start_ctransid) {
5345 WARN(1, KERN_WARNING
5346 "btrfs: btrfs_compare_tree detected "
5347 "a change in one of the trees while "
5348 "iterating. This is probably a "
5349 "bug.\n");
5350 ret = -EIO;
5351 goto out;
5352 }
5353
5354 /*
5355 * the commit root may have changed, so start again
5356 * where we stopped
5357 */
5358 left_path->lowest_level = left_level;
5359 right_path->lowest_level = right_level;
5360 ret = btrfs_search_slot(NULL, left_root,
5361 &left_key, left_path, 0, 0);
5362 if (ret < 0)
5363 goto out;
5364 ret = btrfs_search_slot(NULL, right_root,
5365 &right_key, right_path, 0, 0);
5366 if (ret < 0)
5367 goto out;
5368 }
5369
5370 if (advance_left && !left_end_reached) {
5371 ret = tree_advance(left_root, left_path, &left_level,
5372 left_root_level,
5373 advance_left != ADVANCE_ONLY_NEXT,
5374 &left_key);
5375 if (ret < 0)
5376 left_end_reached = ADVANCE;
5377 advance_left = 0;
5378 }
5379 if (advance_right && !right_end_reached) {
5380 ret = tree_advance(right_root, right_path, &right_level,
5381 right_root_level,
5382 advance_right != ADVANCE_ONLY_NEXT,
5383 &right_key);
5384 if (ret < 0)
5385 right_end_reached = ADVANCE;
5386 advance_right = 0;
5387 }
5388
5389 if (left_end_reached && right_end_reached) {
5390 ret = 0;
5391 goto out;
5392 } else if (left_end_reached) {
5393 if (right_level == 0) {
5394 ret = changed_cb(left_root, right_root,
5395 left_path, right_path,
5396 &right_key,
5397 BTRFS_COMPARE_TREE_DELETED,
5398 ctx);
5399 if (ret < 0)
5400 goto out;
5401 }
5402 advance_right = ADVANCE;
5403 continue;
5404 } else if (right_end_reached) {
5405 if (left_level == 0) {
5406 ret = changed_cb(left_root, right_root,
5407 left_path, right_path,
5408 &left_key,
5409 BTRFS_COMPARE_TREE_NEW,
5410 ctx);
5411 if (ret < 0)
5412 goto out;
5413 }
5414 advance_left = ADVANCE;
5415 continue;
5416 }
5417
5418 if (left_level == 0 && right_level == 0) {
5419 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5420 if (cmp < 0) {
5421 ret = changed_cb(left_root, right_root,
5422 left_path, right_path,
5423 &left_key,
5424 BTRFS_COMPARE_TREE_NEW,
5425 ctx);
5426 if (ret < 0)
5427 goto out;
5428 advance_left = ADVANCE;
5429 } else if (cmp > 0) {
5430 ret = changed_cb(left_root, right_root,
5431 left_path, right_path,
5432 &right_key,
5433 BTRFS_COMPARE_TREE_DELETED,
5434 ctx);
5435 if (ret < 0)
5436 goto out;
5437 advance_right = ADVANCE;
5438 } else {
5439 ret = tree_compare_item(left_root, left_path,
5440 right_path, tmp_buf);
5441 if (ret) {
5442 ret = changed_cb(left_root, right_root,
5443 left_path, right_path,
5444 &left_key,
5445 BTRFS_COMPARE_TREE_CHANGED,
5446 ctx);
5447 if (ret < 0)
5448 goto out;
5449 }
5450 advance_left = ADVANCE;
5451 advance_right = ADVANCE;
5452 }
5453 } else if (left_level == right_level) {
5454 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5455 if (cmp < 0) {
5456 advance_left = ADVANCE;
5457 } else if (cmp > 0) {
5458 advance_right = ADVANCE;
5459 } else {
5460 left_blockptr = btrfs_node_blockptr(
5461 left_path->nodes[left_level],
5462 left_path->slots[left_level]);
5463 right_blockptr = btrfs_node_blockptr(
5464 right_path->nodes[right_level],
5465 right_path->slots[right_level]);
5466 if (left_blockptr == right_blockptr) {
5467 /*
5468 * As we're on a shared block, don't
5469 * allow to go deeper.
5470 */
5471 advance_left = ADVANCE_ONLY_NEXT;
5472 advance_right = ADVANCE_ONLY_NEXT;
5473 } else {
5474 advance_left = ADVANCE;
5475 advance_right = ADVANCE;
5476 }
5477 }
5478 } else if (left_level < right_level) {
5479 advance_right = ADVANCE;
5480 } else {
5481 advance_left = ADVANCE;
5482 }
5483 }
5484
5485out:
5486 btrfs_free_path(left_path);
5487 btrfs_free_path(right_path);
5488 kfree(tmp_buf);
5489
5490 if (trans) {
5491 if (!ret)
5492 ret = btrfs_end_transaction(trans, left_root);
5493 else
5494 btrfs_end_transaction(trans, left_root);
5495 }
5496
5497 return ret;
5498}
5499
5073/* 5500/*
5074 * this is similar to btrfs_next_leaf, but does not try to preserve 5501 * this is similar to btrfs_next_leaf, but does not try to preserve
5075 * and fixup the path. It looks for and returns the next key in the 5502 * and fixup the path. It looks for and returns the next key in the
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 0f369da5cd97..adb1cd7ceb9b 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -712,6 +712,36 @@ struct btrfs_root_item {
712 struct btrfs_disk_key drop_progress; 712 struct btrfs_disk_key drop_progress;
713 u8 drop_level; 713 u8 drop_level;
714 u8 level; 714 u8 level;
715
716 /*
717 * The following fields appear after subvol_uuids+subvol_times
718 * were introduced.
719 */
720
721 /*
722 * This generation number is used to test if the new fields are valid
723 * and up to date while reading the root item. Everytime the root item
724 * is written out, the "generation" field is copied into this field. If
725 * anyone ever mounted the fs with an older kernel, we will have
726 * mismatching generation values here and thus must invalidate the
727 * new fields. See btrfs_update_root and btrfs_find_last_root for
728 * details.
729 * the offset of generation_v2 is also used as the start for the memset
730 * when invalidating the fields.
731 */
732 __le64 generation_v2;
733 u8 uuid[BTRFS_UUID_SIZE];
734 u8 parent_uuid[BTRFS_UUID_SIZE];
735 u8 received_uuid[BTRFS_UUID_SIZE];
736 __le64 ctransid; /* updated when an inode changes */
737 __le64 otransid; /* trans when created */
738 __le64 stransid; /* trans when sent. non-zero for received subvol */
739 __le64 rtransid; /* trans when received. non-zero for received subvol */
740 struct btrfs_timespec ctime;
741 struct btrfs_timespec otime;
742 struct btrfs_timespec stime;
743 struct btrfs_timespec rtime;
744 __le64 reserved[8]; /* for future */
715} __attribute__ ((__packed__)); 745} __attribute__ ((__packed__));
716 746
717/* 747/*
@@ -1520,6 +1550,8 @@ struct btrfs_root {
1520 dev_t anon_dev; 1550 dev_t anon_dev;
1521 1551
1522 int force_cow; 1552 int force_cow;
1553
1554 spinlock_t root_times_lock;
1523}; 1555};
1524 1556
1525struct btrfs_ioctl_defrag_range_args { 1557struct btrfs_ioctl_defrag_range_args {
@@ -2358,6 +2390,16 @@ BTRFS_SETGET_STACK_FUNCS(root_used, struct btrfs_root_item, bytes_used, 64);
2358BTRFS_SETGET_STACK_FUNCS(root_limit, struct btrfs_root_item, byte_limit, 64); 2390BTRFS_SETGET_STACK_FUNCS(root_limit, struct btrfs_root_item, byte_limit, 64);
2359BTRFS_SETGET_STACK_FUNCS(root_last_snapshot, struct btrfs_root_item, 2391BTRFS_SETGET_STACK_FUNCS(root_last_snapshot, struct btrfs_root_item,
2360 last_snapshot, 64); 2392 last_snapshot, 64);
2393BTRFS_SETGET_STACK_FUNCS(root_generation_v2, struct btrfs_root_item,
2394 generation_v2, 64);
2395BTRFS_SETGET_STACK_FUNCS(root_ctransid, struct btrfs_root_item,
2396 ctransid, 64);
2397BTRFS_SETGET_STACK_FUNCS(root_otransid, struct btrfs_root_item,
2398 otransid, 64);
2399BTRFS_SETGET_STACK_FUNCS(root_stransid, struct btrfs_root_item,
2400 stransid, 64);
2401BTRFS_SETGET_STACK_FUNCS(root_rtransid, struct btrfs_root_item,
2402 rtransid, 64);
2361 2403
2362static inline bool btrfs_root_readonly(struct btrfs_root *root) 2404static inline bool btrfs_root_readonly(struct btrfs_root *root)
2363{ 2405{
@@ -2893,6 +2935,21 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
2893 struct btrfs_key *max_key, 2935 struct btrfs_key *max_key,
2894 struct btrfs_path *path, int cache_only, 2936 struct btrfs_path *path, int cache_only,
2895 u64 min_trans); 2937 u64 min_trans);
2938enum btrfs_compare_tree_result {
2939 BTRFS_COMPARE_TREE_NEW,
2940 BTRFS_COMPARE_TREE_DELETED,
2941 BTRFS_COMPARE_TREE_CHANGED,
2942};
2943typedef int (*btrfs_changed_cb_t)(struct btrfs_root *left_root,
2944 struct btrfs_root *right_root,
2945 struct btrfs_path *left_path,
2946 struct btrfs_path *right_path,
2947 struct btrfs_key *key,
2948 enum btrfs_compare_tree_result result,
2949 void *ctx);
2950int btrfs_compare_trees(struct btrfs_root *left_root,
2951 struct btrfs_root *right_root,
2952 btrfs_changed_cb_t cb, void *ctx);
2896int btrfs_cow_block(struct btrfs_trans_handle *trans, 2953int btrfs_cow_block(struct btrfs_trans_handle *trans,
2897 struct btrfs_root *root, struct extent_buffer *buf, 2954 struct btrfs_root *root, struct extent_buffer *buf,
2898 struct extent_buffer *parent, int parent_slot, 2955 struct extent_buffer *parent, int parent_slot,
@@ -3046,6 +3103,9 @@ int __must_check btrfs_update_root(struct btrfs_trans_handle *trans,
3046 struct btrfs_root *root, 3103 struct btrfs_root *root,
3047 struct btrfs_key *key, 3104 struct btrfs_key *key,
3048 struct btrfs_root_item *item); 3105 struct btrfs_root_item *item);
3106void btrfs_read_root_item(struct btrfs_root *root,
3107 struct extent_buffer *eb, int slot,
3108 struct btrfs_root_item *item);
3049int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct 3109int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
3050 btrfs_root_item *item, struct btrfs_key *key); 3110 btrfs_root_item *item, struct btrfs_key *key);
3051int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid); 3111int btrfs_find_dead_roots(struct btrfs_root *root, u64 objectid);
@@ -3053,6 +3113,8 @@ int btrfs_find_orphan_roots(struct btrfs_root *tree_root);
3053void btrfs_set_root_node(struct btrfs_root_item *item, 3113void btrfs_set_root_node(struct btrfs_root_item *item,
3054 struct extent_buffer *node); 3114 struct extent_buffer *node);
3055void btrfs_check_and_init_root_item(struct btrfs_root_item *item); 3115void btrfs_check_and_init_root_item(struct btrfs_root_item *item);
3116void btrfs_update_root_times(struct btrfs_trans_handle *trans,
3117 struct btrfs_root *root);
3056 3118
3057/* dir-item.c */ 3119/* dir-item.c */
3058int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, 3120int btrfs_insert_dir_item(struct btrfs_trans_handle *trans,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 05f4fb6e0607..502b20c56e84 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1182,6 +1182,8 @@ static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1182 root->defrag_running = 0; 1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid; 1183 root->root_key.objectid = objectid;
1184 root->anon_dev = 0; 1184 root->anon_dev = 0;
1185
1186 spin_lock_init(&root->root_times_lock);
1185} 1187}
1186 1188
1187static int __must_check find_and_setup_root(struct btrfs_root *tree_root, 1189static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
@@ -1402,6 +1404,7 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1402 u64 generation; 1404 u64 generation;
1403 u32 blocksize; 1405 u32 blocksize;
1404 int ret = 0; 1406 int ret = 0;
1407 int slot;
1405 1408
1406 root = btrfs_alloc_root(fs_info); 1409 root = btrfs_alloc_root(fs_info);
1407 if (!root) 1410 if (!root)
@@ -1428,9 +1431,8 @@ struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1428 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 1431 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1429 if (ret == 0) { 1432 if (ret == 0) {
1430 l = path->nodes[0]; 1433 l = path->nodes[0];
1431 read_extent_buffer(l, &root->root_item, 1434 slot = path->slots[0];
1432 btrfs_item_ptr_offset(l, path->slots[0]), 1435 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1433 sizeof(root->root_item));
1434 memcpy(&root->root_key, location, sizeof(*location)); 1436 memcpy(&root->root_key, location, sizeof(*location));
1435 } 1437 }
1436 btrfs_free_path(path); 1438 btrfs_free_path(path);
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 144f4642b2a9..dac1fc21d809 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2734,6 +2734,8 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2734 */ 2734 */
2735 if (!btrfs_is_free_space_inode(inode) 2735 if (!btrfs_is_free_space_inode(inode)
2736 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 2736 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2737 btrfs_update_root_times(trans, root);
2738
2737 ret = btrfs_delayed_update_inode(trans, root, inode); 2739 ret = btrfs_delayed_update_inode(trans, root, inode);
2738 if (!ret) 2740 if (!ret)
2739 btrfs_set_inode_last_trans(trans, inode); 2741 btrfs_set_inode_last_trans(trans, inode);
@@ -4728,6 +4730,8 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4728 trace_btrfs_inode_new(inode); 4730 trace_btrfs_inode_new(inode);
4729 btrfs_set_inode_last_trans(trans, inode); 4731 btrfs_set_inode_last_trans(trans, inode);
4730 4732
4733 btrfs_update_root_times(trans, root);
4734
4731 return inode; 4735 return inode;
4732fail: 4736fail:
4733 if (dir) 4737 if (dir)
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 3f3cbe928a1a..43f0012016e3 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -41,6 +41,7 @@
41#include <linux/vmalloc.h> 41#include <linux/vmalloc.h>
42#include <linux/slab.h> 42#include <linux/slab.h>
43#include <linux/blkdev.h> 43#include <linux/blkdev.h>
44#include <linux/uuid.h>
44#include "compat.h" 45#include "compat.h"
45#include "ctree.h" 46#include "ctree.h"
46#include "disk-io.h" 47#include "disk-io.h"
@@ -53,6 +54,7 @@
53#include "inode-map.h" 54#include "inode-map.h"
54#include "backref.h" 55#include "backref.h"
55#include "rcu-string.h" 56#include "rcu-string.h"
57#include "send.h"
56 58
57/* Mask out flags that are inappropriate for the given type of inode. */ 59/* Mask out flags that are inappropriate for the given type of inode. */
58static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags) 60static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags)
@@ -347,11 +349,13 @@ static noinline int create_subvol(struct btrfs_root *root,
347 struct btrfs_root *new_root; 349 struct btrfs_root *new_root;
348 struct dentry *parent = dentry->d_parent; 350 struct dentry *parent = dentry->d_parent;
349 struct inode *dir; 351 struct inode *dir;
352 struct timespec cur_time = CURRENT_TIME;
350 int ret; 353 int ret;
351 int err; 354 int err;
352 u64 objectid; 355 u64 objectid;
353 u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 356 u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
354 u64 index = 0; 357 u64 index = 0;
358 uuid_le new_uuid;
355 359
356 ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid); 360 ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid);
357 if (ret) 361 if (ret)
@@ -395,8 +399,9 @@ static noinline int create_subvol(struct btrfs_root *root,
395 BTRFS_UUID_SIZE); 399 BTRFS_UUID_SIZE);
396 btrfs_mark_buffer_dirty(leaf); 400 btrfs_mark_buffer_dirty(leaf);
397 401
402 memset(&root_item, 0, sizeof(root_item));
403
398 inode_item = &root_item.inode; 404 inode_item = &root_item.inode;
399 memset(inode_item, 0, sizeof(*inode_item));
400 inode_item->generation = cpu_to_le64(1); 405 inode_item->generation = cpu_to_le64(1);
401 inode_item->size = cpu_to_le64(3); 406 inode_item->size = cpu_to_le64(3);
402 inode_item->nlink = cpu_to_le32(1); 407 inode_item->nlink = cpu_to_le32(1);
@@ -414,8 +419,15 @@ static noinline int create_subvol(struct btrfs_root *root,
414 btrfs_set_root_used(&root_item, leaf->len); 419 btrfs_set_root_used(&root_item, leaf->len);
415 btrfs_set_root_last_snapshot(&root_item, 0); 420 btrfs_set_root_last_snapshot(&root_item, 0);
416 421
417 memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress)); 422 btrfs_set_root_generation_v2(&root_item,
418 root_item.drop_level = 0; 423 btrfs_root_generation(&root_item));
424 uuid_le_gen(&new_uuid);
425 memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE);
426 root_item.otime.sec = cpu_to_le64(cur_time.tv_sec);
427 root_item.otime.nsec = cpu_to_le64(cur_time.tv_nsec);
428 root_item.ctime = root_item.otime;
429 btrfs_set_root_ctransid(&root_item, trans->transid);
430 btrfs_set_root_otransid(&root_item, trans->transid);
419 431
420 btrfs_tree_unlock(leaf); 432 btrfs_tree_unlock(leaf);
421 free_extent_buffer(leaf); 433 free_extent_buffer(leaf);
@@ -2370,6 +2382,10 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2370 goto out_drop_write; 2382 goto out_drop_write;
2371 } 2383 }
2372 2384
2385 ret = -EXDEV;
2386 if (src_file->f_path.mnt != file->f_path.mnt)
2387 goto out_fput;
2388
2373 src = src_file->f_dentry->d_inode; 2389 src = src_file->f_dentry->d_inode;
2374 2390
2375 ret = -EINVAL; 2391 ret = -EINVAL;
@@ -2390,7 +2406,7 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2390 goto out_fput; 2406 goto out_fput;
2391 2407
2392 ret = -EXDEV; 2408 ret = -EXDEV;
2393 if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root) 2409 if (src->i_sb != inode->i_sb)
2394 goto out_fput; 2410 goto out_fput;
2395 2411
2396 ret = -ENOMEM; 2412 ret = -ENOMEM;
@@ -2464,13 +2480,14 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
2464 * note the key will change type as we walk through the 2480 * note the key will change type as we walk through the
2465 * tree. 2481 * tree.
2466 */ 2482 */
2467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2483 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
2484 0, 0);
2468 if (ret < 0) 2485 if (ret < 0)
2469 goto out; 2486 goto out;
2470 2487
2471 nritems = btrfs_header_nritems(path->nodes[0]); 2488 nritems = btrfs_header_nritems(path->nodes[0]);
2472 if (path->slots[0] >= nritems) { 2489 if (path->slots[0] >= nritems) {
2473 ret = btrfs_next_leaf(root, path); 2490 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
2474 if (ret < 0) 2491 if (ret < 0)
2475 goto out; 2492 goto out;
2476 if (ret > 0) 2493 if (ret > 0)
@@ -3589,6 +3606,87 @@ out:
3589 return ret; 3606 return ret;
3590} 3607}
3591 3608
3609static long btrfs_ioctl_set_received_subvol(struct file *file,
3610 void __user *arg)
3611{
3612 struct btrfs_ioctl_received_subvol_args *sa = NULL;
3613 struct inode *inode = fdentry(file)->d_inode;
3614 struct btrfs_root *root = BTRFS_I(inode)->root;
3615 struct btrfs_root_item *root_item = &root->root_item;
3616 struct btrfs_trans_handle *trans;
3617 struct timespec ct = CURRENT_TIME;
3618 int ret = 0;
3619
3620 ret = mnt_want_write_file(file);
3621 if (ret < 0)
3622 return ret;
3623
3624 down_write(&root->fs_info->subvol_sem);
3625
3626 if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) {
3627 ret = -EINVAL;
3628 goto out;
3629 }
3630
3631 if (btrfs_root_readonly(root)) {
3632 ret = -EROFS;
3633 goto out;
3634 }
3635
3636 if (!inode_owner_or_capable(inode)) {
3637 ret = -EACCES;
3638 goto out;
3639 }
3640
3641 sa = memdup_user(arg, sizeof(*sa));
3642 if (IS_ERR(sa)) {
3643 ret = PTR_ERR(sa);
3644 sa = NULL;
3645 goto out;
3646 }
3647
3648 trans = btrfs_start_transaction(root, 1);
3649 if (IS_ERR(trans)) {
3650 ret = PTR_ERR(trans);
3651 trans = NULL;
3652 goto out;
3653 }
3654
3655 sa->rtransid = trans->transid;
3656 sa->rtime.sec = ct.tv_sec;
3657 sa->rtime.nsec = ct.tv_nsec;
3658
3659 memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
3660 btrfs_set_root_stransid(root_item, sa->stransid);
3661 btrfs_set_root_rtransid(root_item, sa->rtransid);
3662 root_item->stime.sec = cpu_to_le64(sa->stime.sec);
3663 root_item->stime.nsec = cpu_to_le32(sa->stime.nsec);
3664 root_item->rtime.sec = cpu_to_le64(sa->rtime.sec);
3665 root_item->rtime.nsec = cpu_to_le32(sa->rtime.nsec);
3666
3667 ret = btrfs_update_root(trans, root->fs_info->tree_root,
3668 &root->root_key, &root->root_item);
3669 if (ret < 0) {
3670 btrfs_end_transaction(trans, root);
3671 trans = NULL;
3672 goto out;
3673 } else {
3674 ret = btrfs_commit_transaction(trans, root);
3675 if (ret < 0)
3676 goto out;
3677 }
3678
3679 ret = copy_to_user(arg, sa, sizeof(*sa));
3680 if (ret)
3681 ret = -EFAULT;
3682
3683out:
3684 kfree(sa);
3685 up_write(&root->fs_info->subvol_sem);
3686 mnt_drop_write_file(file);
3687 return ret;
3688}
3689
3592long btrfs_ioctl(struct file *file, unsigned int 3690long btrfs_ioctl(struct file *file, unsigned int
3593 cmd, unsigned long arg) 3691 cmd, unsigned long arg)
3594{ 3692{
@@ -3673,6 +3771,10 @@ long btrfs_ioctl(struct file *file, unsigned int
3673 return btrfs_ioctl_balance_ctl(root, arg); 3771 return btrfs_ioctl_balance_ctl(root, arg);
3674 case BTRFS_IOC_BALANCE_PROGRESS: 3772 case BTRFS_IOC_BALANCE_PROGRESS:
3675 return btrfs_ioctl_balance_progress(root, argp); 3773 return btrfs_ioctl_balance_progress(root, argp);
3774 case BTRFS_IOC_SET_RECEIVED_SUBVOL:
3775 return btrfs_ioctl_set_received_subvol(file, argp);
3776 case BTRFS_IOC_SEND:
3777 return btrfs_ioctl_send(file, argp);
3676 case BTRFS_IOC_GET_DEV_STATS: 3778 case BTRFS_IOC_GET_DEV_STATS:
3677 return btrfs_ioctl_get_dev_stats(root, argp); 3779 return btrfs_ioctl_get_dev_stats(root, argp);
3678 case BTRFS_IOC_QUOTA_CTL: 3780 case BTRFS_IOC_QUOTA_CTL:
diff --git a/fs/btrfs/ioctl.h b/fs/btrfs/ioctl.h
index 3f9701d571ea..731e2875ab93 100644
--- a/fs/btrfs/ioctl.h
+++ b/fs/btrfs/ioctl.h
@@ -348,6 +348,29 @@ struct btrfs_ioctl_qgroup_create_args {
348 __u64 create; 348 __u64 create;
349 __u64 qgroupid; 349 __u64 qgroupid;
350}; 350};
351struct btrfs_ioctl_timespec {
352 __u64 sec;
353 __u32 nsec;
354};
355
356struct btrfs_ioctl_received_subvol_args {
357 char uuid[BTRFS_UUID_SIZE]; /* in */
358 __u64 stransid; /* in */
359 __u64 rtransid; /* out */
360 struct btrfs_ioctl_timespec stime; /* in */
361 struct btrfs_ioctl_timespec rtime; /* out */
362 __u64 flags; /* in */
363 __u64 reserved[16]; /* in */
364};
365
366struct btrfs_ioctl_send_args {
367 __s64 send_fd; /* in */
368 __u64 clone_sources_count; /* in */
369 __u64 __user *clone_sources; /* in */
370 __u64 parent_root; /* in */
371 __u64 flags; /* in */
372 __u64 reserved[4]; /* in */
373};
351 374
352#define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \ 375#define BTRFS_IOC_SNAP_CREATE _IOW(BTRFS_IOCTL_MAGIC, 1, \
353 struct btrfs_ioctl_vol_args) 376 struct btrfs_ioctl_vol_args)
@@ -415,6 +438,9 @@ struct btrfs_ioctl_qgroup_create_args {
415 struct btrfs_ioctl_ino_path_args) 438 struct btrfs_ioctl_ino_path_args)
416#define BTRFS_IOC_LOGICAL_INO _IOWR(BTRFS_IOCTL_MAGIC, 36, \ 439#define BTRFS_IOC_LOGICAL_INO _IOWR(BTRFS_IOCTL_MAGIC, 36, \
417 struct btrfs_ioctl_ino_path_args) 440 struct btrfs_ioctl_ino_path_args)
441#define BTRFS_IOC_SET_RECEIVED_SUBVOL _IOWR(BTRFS_IOCTL_MAGIC, 37, \
442 struct btrfs_ioctl_received_subvol_args)
443#define BTRFS_IOC_SEND _IOW(BTRFS_IOCTL_MAGIC, 38, struct btrfs_ioctl_send_args)
418#define BTRFS_IOC_DEVICES_READY _IOR(BTRFS_IOCTL_MAGIC, 39, \ 444#define BTRFS_IOC_DEVICES_READY _IOR(BTRFS_IOCTL_MAGIC, 39, \
419 struct btrfs_ioctl_vol_args) 445 struct btrfs_ioctl_vol_args)
420#define BTRFS_IOC_QUOTA_CTL _IOWR(BTRFS_IOCTL_MAGIC, 40, \ 446#define BTRFS_IOC_QUOTA_CTL _IOWR(BTRFS_IOCTL_MAGIC, 40, \
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 24fb8ce4e071..6bb465cca20f 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -16,12 +16,55 @@
16 * Boston, MA 021110-1307, USA. 16 * Boston, MA 021110-1307, USA.
17 */ 17 */
18 18
19#include <linux/uuid.h>
19#include "ctree.h" 20#include "ctree.h"
20#include "transaction.h" 21#include "transaction.h"
21#include "disk-io.h" 22#include "disk-io.h"
22#include "print-tree.h" 23#include "print-tree.h"
23 24
24/* 25/*
26 * Read a root item from the tree. In case we detect a root item smaller then
27 * sizeof(root_item), we know it's an old version of the root structure and
28 * initialize all new fields to zero. The same happens if we detect mismatching
29 * generation numbers as then we know the root was once mounted with an older
30 * kernel that was not aware of the root item structure change.
31 */
32void btrfs_read_root_item(struct btrfs_root *root,
33 struct extent_buffer *eb, int slot,
34 struct btrfs_root_item *item)
35{
36 uuid_le uuid;
37 int len;
38 int need_reset = 0;
39
40 len = btrfs_item_size_nr(eb, slot);
41 read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
42 min_t(int, len, (int)sizeof(*item)));
43 if (len < sizeof(*item))
44 need_reset = 1;
45 if (!need_reset && btrfs_root_generation(item)
46 != btrfs_root_generation_v2(item)) {
47 if (btrfs_root_generation_v2(item) != 0) {
48 printk(KERN_WARNING "btrfs: mismatching "
49 "generation and generation_v2 "
50 "found in root item. This root "
51 "was probably mounted with an "
52 "older kernel. Resetting all "
53 "new fields.\n");
54 }
55 need_reset = 1;
56 }
57 if (need_reset) {
58 memset(&item->generation_v2, 0,
59 sizeof(*item) - offsetof(struct btrfs_root_item,
60 generation_v2));
61
62 uuid_le_gen(&uuid);
63 memcpy(item->uuid, uuid.b, BTRFS_UUID_SIZE);
64 }
65}
66
67/*
25 * lookup the root with the highest offset for a given objectid. The key we do 68 * lookup the root with the highest offset for a given objectid. The key we do
26 * find is copied into 'key'. If we find something return 0, otherwise 1, < 0 69 * find is copied into 'key'. If we find something return 0, otherwise 1, < 0
27 * on error. 70 * on error.
@@ -61,10 +104,10 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
61 goto out; 104 goto out;
62 } 105 }
63 if (item) 106 if (item)
64 read_extent_buffer(l, item, btrfs_item_ptr_offset(l, slot), 107 btrfs_read_root_item(root, l, slot, item);
65 sizeof(*item));
66 if (key) 108 if (key)
67 memcpy(key, &found_key, sizeof(found_key)); 109 memcpy(key, &found_key, sizeof(found_key));
110
68 ret = 0; 111 ret = 0;
69out: 112out:
70 btrfs_free_path(path); 113 btrfs_free_path(path);
@@ -91,16 +134,15 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
91 int ret; 134 int ret;
92 int slot; 135 int slot;
93 unsigned long ptr; 136 unsigned long ptr;
137 int old_len;
94 138
95 path = btrfs_alloc_path(); 139 path = btrfs_alloc_path();
96 if (!path) 140 if (!path)
97 return -ENOMEM; 141 return -ENOMEM;
98 142
99 ret = btrfs_search_slot(trans, root, key, path, 0, 1); 143 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
100 if (ret < 0) { 144 if (ret < 0)
101 btrfs_abort_transaction(trans, root, ret); 145 goto out_abort;
102 goto out;
103 }
104 146
105 if (ret != 0) { 147 if (ret != 0) {
106 btrfs_print_leaf(root, path->nodes[0]); 148 btrfs_print_leaf(root, path->nodes[0]);
@@ -113,16 +155,56 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
113 l = path->nodes[0]; 155 l = path->nodes[0];
114 slot = path->slots[0]; 156 slot = path->slots[0];
115 ptr = btrfs_item_ptr_offset(l, slot); 157 ptr = btrfs_item_ptr_offset(l, slot);
158 old_len = btrfs_item_size_nr(l, slot);
159
160 /*
161 * If this is the first time we update the root item which originated
162 * from an older kernel, we need to enlarge the item size to make room
163 * for the added fields.
164 */
165 if (old_len < sizeof(*item)) {
166 btrfs_release_path(path);
167 ret = btrfs_search_slot(trans, root, key, path,
168 -1, 1);
169 if (ret < 0)
170 goto out_abort;
171 ret = btrfs_del_item(trans, root, path);
172 if (ret < 0)
173 goto out_abort;
174 btrfs_release_path(path);
175 ret = btrfs_insert_empty_item(trans, root, path,
176 key, sizeof(*item));
177 if (ret < 0)
178 goto out_abort;
179 l = path->nodes[0];
180 slot = path->slots[0];
181 ptr = btrfs_item_ptr_offset(l, slot);
182 }
183
184 /*
185 * Update generation_v2 so at the next mount we know the new root
186 * fields are valid.
187 */
188 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
189
116 write_extent_buffer(l, item, ptr, sizeof(*item)); 190 write_extent_buffer(l, item, ptr, sizeof(*item));
117 btrfs_mark_buffer_dirty(path->nodes[0]); 191 btrfs_mark_buffer_dirty(path->nodes[0]);
118out: 192out:
119 btrfs_free_path(path); 193 btrfs_free_path(path);
120 return ret; 194 return ret;
195
196out_abort:
197 btrfs_abort_transaction(trans, root, ret);
198 goto out;
121} 199}
122 200
123int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root, 201int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
124 struct btrfs_key *key, struct btrfs_root_item *item) 202 struct btrfs_key *key, struct btrfs_root_item *item)
125{ 203{
204 /*
205 * Make sure generation v1 and v2 match. See update_root for details.
206 */
207 btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
126 return btrfs_insert_item(trans, root, key, item, sizeof(*item)); 208 return btrfs_insert_item(trans, root, key, item, sizeof(*item));
127} 209}
128 210
@@ -454,3 +536,16 @@ void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
454 root_item->byte_limit = 0; 536 root_item->byte_limit = 0;
455 } 537 }
456} 538}
539
540void btrfs_update_root_times(struct btrfs_trans_handle *trans,
541 struct btrfs_root *root)
542{
543 struct btrfs_root_item *item = &root->root_item;
544 struct timespec ct = CURRENT_TIME;
545
546 spin_lock(&root->root_times_lock);
547 item->ctransid = trans->transid;
548 item->ctime.sec = cpu_to_le64(ct.tv_sec);
549 item->ctime.nsec = cpu_to_le64(ct.tv_nsec);
550 spin_unlock(&root->root_times_lock);
551}
diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
new file mode 100644
index 000000000000..5394cb75012a
--- /dev/null
+++ b/fs/btrfs/send.c
@@ -0,0 +1,4570 @@
1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19#include <linux/bsearch.h>
20#include <linux/fs.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
29#include "send.h"
30#include "backref.h"
31#include "locking.h"
32#include "disk-io.h"
33#include "btrfs_inode.h"
34#include "transaction.h"
35
36static int g_verbose = 0;
37
38#define verbose_printk(...) if (g_verbose) printk(__VA_ARGS__)
39
40/*
41 * A fs_path is a helper to dynamically build path names with unknown size.
42 * It reallocates the internal buffer on demand.
43 * It allows fast adding of path elements on the right side (normal path) and
44 * fast adding to the left side (reversed path). A reversed path can also be
45 * unreversed if needed.
46 */
47struct fs_path {
48 union {
49 struct {
50 char *start;
51 char *end;
52 char *prepared;
53
54 char *buf;
55 int buf_len;
56 int reversed:1;
57 int virtual_mem:1;
58 char inline_buf[];
59 };
60 char pad[PAGE_SIZE];
61 };
62};
63#define FS_PATH_INLINE_SIZE \
64 (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf))
65
66
67/* reused for each extent */
68struct clone_root {
69 struct btrfs_root *root;
70 u64 ino;
71 u64 offset;
72
73 u64 found_refs;
74};
75
76#define SEND_CTX_MAX_NAME_CACHE_SIZE 128
77#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2)
78
79struct send_ctx {
80 struct file *send_filp;
81 loff_t send_off;
82 char *send_buf;
83 u32 send_size;
84 u32 send_max_size;
85 u64 total_send_size;
86 u64 cmd_send_size[BTRFS_SEND_C_MAX + 1];
87
88 struct vfsmount *mnt;
89
90 struct btrfs_root *send_root;
91 struct btrfs_root *parent_root;
92 struct clone_root *clone_roots;
93 int clone_roots_cnt;
94
95 /* current state of the compare_tree call */
96 struct btrfs_path *left_path;
97 struct btrfs_path *right_path;
98 struct btrfs_key *cmp_key;
99
100 /*
101 * infos of the currently processed inode. In case of deleted inodes,
102 * these are the values from the deleted inode.
103 */
104 u64 cur_ino;
105 u64 cur_inode_gen;
106 int cur_inode_new;
107 int cur_inode_new_gen;
108 int cur_inode_deleted;
109 int cur_inode_first_ref_orphan;
110 u64 cur_inode_size;
111 u64 cur_inode_mode;
112
113 u64 send_progress;
114
115 struct list_head new_refs;
116 struct list_head deleted_refs;
117
118 struct radix_tree_root name_cache;
119 struct list_head name_cache_list;
120 int name_cache_size;
121
122 struct file *cur_inode_filp;
123 char *read_buf;
124};
125
126struct name_cache_entry {
127 struct list_head list;
128 struct list_head use_list;
129 u64 ino;
130 u64 gen;
131 u64 parent_ino;
132 u64 parent_gen;
133 int ret;
134 int need_later_update;
135 int name_len;
136 char name[];
137};
138
139static void fs_path_reset(struct fs_path *p)
140{
141 if (p->reversed) {
142 p->start = p->buf + p->buf_len - 1;
143 p->end = p->start;
144 *p->start = 0;
145 } else {
146 p->start = p->buf;
147 p->end = p->start;
148 *p->start = 0;
149 }
150}
151
152static struct fs_path *fs_path_alloc(struct send_ctx *sctx)
153{
154 struct fs_path *p;
155
156 p = kmalloc(sizeof(*p), GFP_NOFS);
157 if (!p)
158 return NULL;
159 p->reversed = 0;
160 p->virtual_mem = 0;
161 p->buf = p->inline_buf;
162 p->buf_len = FS_PATH_INLINE_SIZE;
163 fs_path_reset(p);
164 return p;
165}
166
167static struct fs_path *fs_path_alloc_reversed(struct send_ctx *sctx)
168{
169 struct fs_path *p;
170
171 p = fs_path_alloc(sctx);
172 if (!p)
173 return NULL;
174 p->reversed = 1;
175 fs_path_reset(p);
176 return p;
177}
178
179static void fs_path_free(struct send_ctx *sctx, struct fs_path *p)
180{
181 if (!p)
182 return;
183 if (p->buf != p->inline_buf) {
184 if (p->virtual_mem)
185 vfree(p->buf);
186 else
187 kfree(p->buf);
188 }
189 kfree(p);
190}
191
192static int fs_path_len(struct fs_path *p)
193{
194 return p->end - p->start;
195}
196
197static int fs_path_ensure_buf(struct fs_path *p, int len)
198{
199 char *tmp_buf;
200 int path_len;
201 int old_buf_len;
202
203 len++;
204
205 if (p->buf_len >= len)
206 return 0;
207
208 path_len = p->end - p->start;
209 old_buf_len = p->buf_len;
210 len = PAGE_ALIGN(len);
211
212 if (p->buf == p->inline_buf) {
213 tmp_buf = kmalloc(len, GFP_NOFS);
214 if (!tmp_buf) {
215 tmp_buf = vmalloc(len);
216 if (!tmp_buf)
217 return -ENOMEM;
218 p->virtual_mem = 1;
219 }
220 memcpy(tmp_buf, p->buf, p->buf_len);
221 p->buf = tmp_buf;
222 p->buf_len = len;
223 } else {
224 if (p->virtual_mem) {
225 tmp_buf = vmalloc(len);
226 if (!tmp_buf)
227 return -ENOMEM;
228 memcpy(tmp_buf, p->buf, p->buf_len);
229 vfree(p->buf);
230 } else {
231 tmp_buf = krealloc(p->buf, len, GFP_NOFS);
232 if (!tmp_buf) {
233 tmp_buf = vmalloc(len);
234 if (!tmp_buf)
235 return -ENOMEM;
236 memcpy(tmp_buf, p->buf, p->buf_len);
237 kfree(p->buf);
238 p->virtual_mem = 1;
239 }
240 }
241 p->buf = tmp_buf;
242 p->buf_len = len;
243 }
244 if (p->reversed) {
245 tmp_buf = p->buf + old_buf_len - path_len - 1;
246 p->end = p->buf + p->buf_len - 1;
247 p->start = p->end - path_len;
248 memmove(p->start, tmp_buf, path_len + 1);
249 } else {
250 p->start = p->buf;
251 p->end = p->start + path_len;
252 }
253 return 0;
254}
255
256static int fs_path_prepare_for_add(struct fs_path *p, int name_len)
257{
258 int ret;
259 int new_len;
260
261 new_len = p->end - p->start + name_len;
262 if (p->start != p->end)
263 new_len++;
264 ret = fs_path_ensure_buf(p, new_len);
265 if (ret < 0)
266 goto out;
267
268 if (p->reversed) {
269 if (p->start != p->end)
270 *--p->start = '/';
271 p->start -= name_len;
272 p->prepared = p->start;
273 } else {
274 if (p->start != p->end)
275 *p->end++ = '/';
276 p->prepared = p->end;
277 p->end += name_len;
278 *p->end = 0;
279 }
280
281out:
282 return ret;
283}
284
285static int fs_path_add(struct fs_path *p, const char *name, int name_len)
286{
287 int ret;
288
289 ret = fs_path_prepare_for_add(p, name_len);
290 if (ret < 0)
291 goto out;
292 memcpy(p->prepared, name, name_len);
293 p->prepared = NULL;
294
295out:
296 return ret;
297}
298
299static int fs_path_add_path(struct fs_path *p, struct fs_path *p2)
300{
301 int ret;
302
303 ret = fs_path_prepare_for_add(p, p2->end - p2->start);
304 if (ret < 0)
305 goto out;
306 memcpy(p->prepared, p2->start, p2->end - p2->start);
307 p->prepared = NULL;
308
309out:
310 return ret;
311}
312
313static int fs_path_add_from_extent_buffer(struct fs_path *p,
314 struct extent_buffer *eb,
315 unsigned long off, int len)
316{
317 int ret;
318
319 ret = fs_path_prepare_for_add(p, len);
320 if (ret < 0)
321 goto out;
322
323 read_extent_buffer(eb, p->prepared, off, len);
324 p->prepared = NULL;
325
326out:
327 return ret;
328}
329
330static void fs_path_remove(struct fs_path *p)
331{
332 BUG_ON(p->reversed);
333 while (p->start != p->end && *p->end != '/')
334 p->end--;
335 *p->end = 0;
336}
337
338static int fs_path_copy(struct fs_path *p, struct fs_path *from)
339{
340 int ret;
341
342 p->reversed = from->reversed;
343 fs_path_reset(p);
344
345 ret = fs_path_add_path(p, from);
346
347 return ret;
348}
349
350
351static void fs_path_unreverse(struct fs_path *p)
352{
353 char *tmp;
354 int len;
355
356 if (!p->reversed)
357 return;
358
359 tmp = p->start;
360 len = p->end - p->start;
361 p->start = p->buf;
362 p->end = p->start + len;
363 memmove(p->start, tmp, len + 1);
364 p->reversed = 0;
365}
366
367static struct btrfs_path *alloc_path_for_send(void)
368{
369 struct btrfs_path *path;
370
371 path = btrfs_alloc_path();
372 if (!path)
373 return NULL;
374 path->search_commit_root = 1;
375 path->skip_locking = 1;
376 return path;
377}
378
379static int write_buf(struct send_ctx *sctx, const void *buf, u32 len)
380{
381 int ret;
382 mm_segment_t old_fs;
383 u32 pos = 0;
384
385 old_fs = get_fs();
386 set_fs(KERNEL_DS);
387
388 while (pos < len) {
389 ret = vfs_write(sctx->send_filp, (char *)buf + pos, len - pos,
390 &sctx->send_off);
391 /* TODO handle that correctly */
392 /*if (ret == -ERESTARTSYS) {
393 continue;
394 }*/
395 if (ret < 0)
396 goto out;
397 if (ret == 0) {
398 ret = -EIO;
399 goto out;
400 }
401 pos += ret;
402 }
403
404 ret = 0;
405
406out:
407 set_fs(old_fs);
408 return ret;
409}
410
411static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len)
412{
413 struct btrfs_tlv_header *hdr;
414 int total_len = sizeof(*hdr) + len;
415 int left = sctx->send_max_size - sctx->send_size;
416
417 if (unlikely(left < total_len))
418 return -EOVERFLOW;
419
420 hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
421 hdr->tlv_type = cpu_to_le16(attr);
422 hdr->tlv_len = cpu_to_le16(len);
423 memcpy(hdr + 1, data, len);
424 sctx->send_size += total_len;
425
426 return 0;
427}
428
429#if 0
430static int tlv_put_u8(struct send_ctx *sctx, u16 attr, u8 value)
431{
432 return tlv_put(sctx, attr, &value, sizeof(value));
433}
434
435static int tlv_put_u16(struct send_ctx *sctx, u16 attr, u16 value)
436{
437 __le16 tmp = cpu_to_le16(value);
438 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
439}
440
441static int tlv_put_u32(struct send_ctx *sctx, u16 attr, u32 value)
442{
443 __le32 tmp = cpu_to_le32(value);
444 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
445}
446#endif
447
448static int tlv_put_u64(struct send_ctx *sctx, u16 attr, u64 value)
449{
450 __le64 tmp = cpu_to_le64(value);
451 return tlv_put(sctx, attr, &tmp, sizeof(tmp));
452}
453
454static int tlv_put_string(struct send_ctx *sctx, u16 attr,
455 const char *str, int len)
456{
457 if (len == -1)
458 len = strlen(str);
459 return tlv_put(sctx, attr, str, len);
460}
461
462static int tlv_put_uuid(struct send_ctx *sctx, u16 attr,
463 const u8 *uuid)
464{
465 return tlv_put(sctx, attr, uuid, BTRFS_UUID_SIZE);
466}
467
468#if 0
469static int tlv_put_timespec(struct send_ctx *sctx, u16 attr,
470 struct timespec *ts)
471{
472 struct btrfs_timespec bts;
473 bts.sec = cpu_to_le64(ts->tv_sec);
474 bts.nsec = cpu_to_le32(ts->tv_nsec);
475 return tlv_put(sctx, attr, &bts, sizeof(bts));
476}
477#endif
478
479static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr,
480 struct extent_buffer *eb,
481 struct btrfs_timespec *ts)
482{
483 struct btrfs_timespec bts;
484 read_extent_buffer(eb, &bts, (unsigned long)ts, sizeof(bts));
485 return tlv_put(sctx, attr, &bts, sizeof(bts));
486}
487
488
489#define TLV_PUT(sctx, attrtype, attrlen, data) \
490 do { \
491 ret = tlv_put(sctx, attrtype, attrlen, data); \
492 if (ret < 0) \
493 goto tlv_put_failure; \
494 } while (0)
495
496#define TLV_PUT_INT(sctx, attrtype, bits, value) \
497 do { \
498 ret = tlv_put_u##bits(sctx, attrtype, value); \
499 if (ret < 0) \
500 goto tlv_put_failure; \
501 } while (0)
502
503#define TLV_PUT_U8(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 8, data)
504#define TLV_PUT_U16(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 16, data)
505#define TLV_PUT_U32(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 32, data)
506#define TLV_PUT_U64(sctx, attrtype, data) TLV_PUT_INT(sctx, attrtype, 64, data)
507#define TLV_PUT_STRING(sctx, attrtype, str, len) \
508 do { \
509 ret = tlv_put_string(sctx, attrtype, str, len); \
510 if (ret < 0) \
511 goto tlv_put_failure; \
512 } while (0)
513#define TLV_PUT_PATH(sctx, attrtype, p) \
514 do { \
515 ret = tlv_put_string(sctx, attrtype, p->start, \
516 p->end - p->start); \
517 if (ret < 0) \
518 goto tlv_put_failure; \
519 } while(0)
520#define TLV_PUT_UUID(sctx, attrtype, uuid) \
521 do { \
522 ret = tlv_put_uuid(sctx, attrtype, uuid); \
523 if (ret < 0) \
524 goto tlv_put_failure; \
525 } while (0)
526#define TLV_PUT_TIMESPEC(sctx, attrtype, ts) \
527 do { \
528 ret = tlv_put_timespec(sctx, attrtype, ts); \
529 if (ret < 0) \
530 goto tlv_put_failure; \
531 } while (0)
532#define TLV_PUT_BTRFS_TIMESPEC(sctx, attrtype, eb, ts) \
533 do { \
534 ret = tlv_put_btrfs_timespec(sctx, attrtype, eb, ts); \
535 if (ret < 0) \
536 goto tlv_put_failure; \
537 } while (0)
538
539static int send_header(struct send_ctx *sctx)
540{
541 struct btrfs_stream_header hdr;
542
543 strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC);
544 hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION);
545
546 return write_buf(sctx, &hdr, sizeof(hdr));
547}
548
549/*
550 * For each command/item we want to send to userspace, we call this function.
551 */
552static int begin_cmd(struct send_ctx *sctx, int cmd)
553{
554 struct btrfs_cmd_header *hdr;
555
556 if (!sctx->send_buf) {
557 WARN_ON(1);
558 return -EINVAL;
559 }
560
561 BUG_ON(sctx->send_size);
562
563 sctx->send_size += sizeof(*hdr);
564 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
565 hdr->cmd = cpu_to_le16(cmd);
566
567 return 0;
568}
569
570static int send_cmd(struct send_ctx *sctx)
571{
572 int ret;
573 struct btrfs_cmd_header *hdr;
574 u32 crc;
575
576 hdr = (struct btrfs_cmd_header *)sctx->send_buf;
577 hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
578 hdr->crc = 0;
579
580 crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
581 hdr->crc = cpu_to_le32(crc);
582
583 ret = write_buf(sctx, sctx->send_buf, sctx->send_size);
584
585 sctx->total_send_size += sctx->send_size;
586 sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
587 sctx->send_size = 0;
588
589 return ret;
590}
591
592/*
593 * Sends a move instruction to user space
594 */
595static int send_rename(struct send_ctx *sctx,
596 struct fs_path *from, struct fs_path *to)
597{
598 int ret;
599
600verbose_printk("btrfs: send_rename %s -> %s\n", from->start, to->start);
601
602 ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME);
603 if (ret < 0)
604 goto out;
605
606 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from);
607 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to);
608
609 ret = send_cmd(sctx);
610
611tlv_put_failure:
612out:
613 return ret;
614}
615
616/*
617 * Sends a link instruction to user space
618 */
619static int send_link(struct send_ctx *sctx,
620 struct fs_path *path, struct fs_path *lnk)
621{
622 int ret;
623
624verbose_printk("btrfs: send_link %s -> %s\n", path->start, lnk->start);
625
626 ret = begin_cmd(sctx, BTRFS_SEND_C_LINK);
627 if (ret < 0)
628 goto out;
629
630 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
631 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk);
632
633 ret = send_cmd(sctx);
634
635tlv_put_failure:
636out:
637 return ret;
638}
639
640/*
641 * Sends an unlink instruction to user space
642 */
643static int send_unlink(struct send_ctx *sctx, struct fs_path *path)
644{
645 int ret;
646
647verbose_printk("btrfs: send_unlink %s\n", path->start);
648
649 ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK);
650 if (ret < 0)
651 goto out;
652
653 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
654
655 ret = send_cmd(sctx);
656
657tlv_put_failure:
658out:
659 return ret;
660}
661
662/*
663 * Sends a rmdir instruction to user space
664 */
665static int send_rmdir(struct send_ctx *sctx, struct fs_path *path)
666{
667 int ret;
668
669verbose_printk("btrfs: send_rmdir %s\n", path->start);
670
671 ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR);
672 if (ret < 0)
673 goto out;
674
675 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
676
677 ret = send_cmd(sctx);
678
679tlv_put_failure:
680out:
681 return ret;
682}
683
684/*
685 * Helper function to retrieve some fields from an inode item.
686 */
687static int get_inode_info(struct btrfs_root *root,
688 u64 ino, u64 *size, u64 *gen,
689 u64 *mode, u64 *uid, u64 *gid)
690{
691 int ret;
692 struct btrfs_inode_item *ii;
693 struct btrfs_key key;
694 struct btrfs_path *path;
695
696 path = alloc_path_for_send();
697 if (!path)
698 return -ENOMEM;
699
700 key.objectid = ino;
701 key.type = BTRFS_INODE_ITEM_KEY;
702 key.offset = 0;
703 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
704 if (ret < 0)
705 goto out;
706 if (ret) {
707 ret = -ENOENT;
708 goto out;
709 }
710
711 ii = btrfs_item_ptr(path->nodes[0], path->slots[0],
712 struct btrfs_inode_item);
713 if (size)
714 *size = btrfs_inode_size(path->nodes[0], ii);
715 if (gen)
716 *gen = btrfs_inode_generation(path->nodes[0], ii);
717 if (mode)
718 *mode = btrfs_inode_mode(path->nodes[0], ii);
719 if (uid)
720 *uid = btrfs_inode_uid(path->nodes[0], ii);
721 if (gid)
722 *gid = btrfs_inode_gid(path->nodes[0], ii);
723
724out:
725 btrfs_free_path(path);
726 return ret;
727}
728
729typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index,
730 struct fs_path *p,
731 void *ctx);
732
733/*
734 * Helper function to iterate the entries in ONE btrfs_inode_ref.
735 * The iterate callback may return a non zero value to stop iteration. This can
736 * be a negative value for error codes or 1 to simply stop it.
737 *
738 * path must point to the INODE_REF when called.
739 */
740static int iterate_inode_ref(struct send_ctx *sctx,
741 struct btrfs_root *root, struct btrfs_path *path,
742 struct btrfs_key *found_key, int resolve,
743 iterate_inode_ref_t iterate, void *ctx)
744{
745 struct extent_buffer *eb;
746 struct btrfs_item *item;
747 struct btrfs_inode_ref *iref;
748 struct btrfs_path *tmp_path;
749 struct fs_path *p;
750 u32 cur;
751 u32 len;
752 u32 total;
753 int slot;
754 u32 name_len;
755 char *start;
756 int ret = 0;
757 int num;
758 int index;
759
760 p = fs_path_alloc_reversed(sctx);
761 if (!p)
762 return -ENOMEM;
763
764 tmp_path = alloc_path_for_send();
765 if (!tmp_path) {
766 fs_path_free(sctx, p);
767 return -ENOMEM;
768 }
769
770 eb = path->nodes[0];
771 slot = path->slots[0];
772 item = btrfs_item_nr(eb, slot);
773 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
774 cur = 0;
775 len = 0;
776 total = btrfs_item_size(eb, item);
777
778 num = 0;
779 while (cur < total) {
780 fs_path_reset(p);
781
782 name_len = btrfs_inode_ref_name_len(eb, iref);
783 index = btrfs_inode_ref_index(eb, iref);
784 if (resolve) {
785 start = btrfs_iref_to_path(root, tmp_path, iref, eb,
786 found_key->offset, p->buf,
787 p->buf_len);
788 if (IS_ERR(start)) {
789 ret = PTR_ERR(start);
790 goto out;
791 }
792 if (start < p->buf) {
793 /* overflow , try again with larger buffer */
794 ret = fs_path_ensure_buf(p,
795 p->buf_len + p->buf - start);
796 if (ret < 0)
797 goto out;
798 start = btrfs_iref_to_path(root, tmp_path, iref,
799 eb, found_key->offset, p->buf,
800 p->buf_len);
801 if (IS_ERR(start)) {
802 ret = PTR_ERR(start);
803 goto out;
804 }
805 BUG_ON(start < p->buf);
806 }
807 p->start = start;
808 } else {
809 ret = fs_path_add_from_extent_buffer(p, eb,
810 (unsigned long)(iref + 1), name_len);
811 if (ret < 0)
812 goto out;
813 }
814
815
816 len = sizeof(*iref) + name_len;
817 iref = (struct btrfs_inode_ref *)((char *)iref + len);
818 cur += len;
819
820 ret = iterate(num, found_key->offset, index, p, ctx);
821 if (ret)
822 goto out;
823
824 num++;
825 }
826
827out:
828 btrfs_free_path(tmp_path);
829 fs_path_free(sctx, p);
830 return ret;
831}
832
833typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key,
834 const char *name, int name_len,
835 const char *data, int data_len,
836 u8 type, void *ctx);
837
838/*
839 * Helper function to iterate the entries in ONE btrfs_dir_item.
840 * The iterate callback may return a non zero value to stop iteration. This can
841 * be a negative value for error codes or 1 to simply stop it.
842 *
843 * path must point to the dir item when called.
844 */
845static int iterate_dir_item(struct send_ctx *sctx,
846 struct btrfs_root *root, struct btrfs_path *path,
847 struct btrfs_key *found_key,
848 iterate_dir_item_t iterate, void *ctx)
849{
850 int ret = 0;
851 struct extent_buffer *eb;
852 struct btrfs_item *item;
853 struct btrfs_dir_item *di;
854 struct btrfs_path *tmp_path = NULL;
855 struct btrfs_key di_key;
856 char *buf = NULL;
857 char *buf2 = NULL;
858 int buf_len;
859 int buf_virtual = 0;
860 u32 name_len;
861 u32 data_len;
862 u32 cur;
863 u32 len;
864 u32 total;
865 int slot;
866 int num;
867 u8 type;
868
869 buf_len = PAGE_SIZE;
870 buf = kmalloc(buf_len, GFP_NOFS);
871 if (!buf) {
872 ret = -ENOMEM;
873 goto out;
874 }
875
876 tmp_path = alloc_path_for_send();
877 if (!tmp_path) {
878 ret = -ENOMEM;
879 goto out;
880 }
881
882 eb = path->nodes[0];
883 slot = path->slots[0];
884 item = btrfs_item_nr(eb, slot);
885 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
886 cur = 0;
887 len = 0;
888 total = btrfs_item_size(eb, item);
889
890 num = 0;
891 while (cur < total) {
892 name_len = btrfs_dir_name_len(eb, di);
893 data_len = btrfs_dir_data_len(eb, di);
894 type = btrfs_dir_type(eb, di);
895 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
896
897 if (name_len + data_len > buf_len) {
898 buf_len = PAGE_ALIGN(name_len + data_len);
899 if (buf_virtual) {
900 buf2 = vmalloc(buf_len);
901 if (!buf2) {
902 ret = -ENOMEM;
903 goto out;
904 }
905 vfree(buf);
906 } else {
907 buf2 = krealloc(buf, buf_len, GFP_NOFS);
908 if (!buf2) {
909 buf2 = vmalloc(buf_len);
910 if (!buf2) {
911 ret = -ENOMEM;
912 goto out;
913 }
914 kfree(buf);
915 buf_virtual = 1;
916 }
917 }
918
919 buf = buf2;
920 buf2 = NULL;
921 }
922
923 read_extent_buffer(eb, buf, (unsigned long)(di + 1),
924 name_len + data_len);
925
926 len = sizeof(*di) + name_len + data_len;
927 di = (struct btrfs_dir_item *)((char *)di + len);
928 cur += len;
929
930 ret = iterate(num, &di_key, buf, name_len, buf + name_len,
931 data_len, type, ctx);
932 if (ret < 0)
933 goto out;
934 if (ret) {
935 ret = 0;
936 goto out;
937 }
938
939 num++;
940 }
941
942out:
943 btrfs_free_path(tmp_path);
944 if (buf_virtual)
945 vfree(buf);
946 else
947 kfree(buf);
948 return ret;
949}
950
951static int __copy_first_ref(int num, u64 dir, int index,
952 struct fs_path *p, void *ctx)
953{
954 int ret;
955 struct fs_path *pt = ctx;
956
957 ret = fs_path_copy(pt, p);
958 if (ret < 0)
959 return ret;
960
961 /* we want the first only */
962 return 1;
963}
964
965/*
966 * Retrieve the first path of an inode. If an inode has more then one
967 * ref/hardlink, this is ignored.
968 */
969static int get_inode_path(struct send_ctx *sctx, struct btrfs_root *root,
970 u64 ino, struct fs_path *path)
971{
972 int ret;
973 struct btrfs_key key, found_key;
974 struct btrfs_path *p;
975
976 p = alloc_path_for_send();
977 if (!p)
978 return -ENOMEM;
979
980 fs_path_reset(path);
981
982 key.objectid = ino;
983 key.type = BTRFS_INODE_REF_KEY;
984 key.offset = 0;
985
986 ret = btrfs_search_slot_for_read(root, &key, p, 1, 0);
987 if (ret < 0)
988 goto out;
989 if (ret) {
990 ret = 1;
991 goto out;
992 }
993 btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]);
994 if (found_key.objectid != ino ||
995 found_key.type != BTRFS_INODE_REF_KEY) {
996 ret = -ENOENT;
997 goto out;
998 }
999
1000 ret = iterate_inode_ref(sctx, root, p, &found_key, 1,
1001 __copy_first_ref, path);
1002 if (ret < 0)
1003 goto out;
1004 ret = 0;
1005
1006out:
1007 btrfs_free_path(p);
1008 return ret;
1009}
1010
1011struct backref_ctx {
1012 struct send_ctx *sctx;
1013
1014 /* number of total found references */
1015 u64 found;
1016
1017 /*
1018 * used for clones found in send_root. clones found behind cur_objectid
1019 * and cur_offset are not considered as allowed clones.
1020 */
1021 u64 cur_objectid;
1022 u64 cur_offset;
1023
1024 /* may be truncated in case it's the last extent in a file */
1025 u64 extent_len;
1026
1027 /* Just to check for bugs in backref resolving */
1028 int found_in_send_root;
1029};
1030
1031static int __clone_root_cmp_bsearch(const void *key, const void *elt)
1032{
1033 u64 root = (u64)key;
1034 struct clone_root *cr = (struct clone_root *)elt;
1035
1036 if (root < cr->root->objectid)
1037 return -1;
1038 if (root > cr->root->objectid)
1039 return 1;
1040 return 0;
1041}
1042
1043static int __clone_root_cmp_sort(const void *e1, const void *e2)
1044{
1045 struct clone_root *cr1 = (struct clone_root *)e1;
1046 struct clone_root *cr2 = (struct clone_root *)e2;
1047
1048 if (cr1->root->objectid < cr2->root->objectid)
1049 return -1;
1050 if (cr1->root->objectid > cr2->root->objectid)
1051 return 1;
1052 return 0;
1053}
1054
1055/*
1056 * Called for every backref that is found for the current extent.
1057 */
1058static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_)
1059{
1060 struct backref_ctx *bctx = ctx_;
1061 struct clone_root *found;
1062 int ret;
1063 u64 i_size;
1064
1065 /* First check if the root is in the list of accepted clone sources */
1066 found = bsearch((void *)root, bctx->sctx->clone_roots,
1067 bctx->sctx->clone_roots_cnt,
1068 sizeof(struct clone_root),
1069 __clone_root_cmp_bsearch);
1070 if (!found)
1071 return 0;
1072
1073 if (found->root == bctx->sctx->send_root &&
1074 ino == bctx->cur_objectid &&
1075 offset == bctx->cur_offset) {
1076 bctx->found_in_send_root = 1;
1077 }
1078
1079 /*
1080 * There are inodes that have extents that lie behind it's i_size. Don't
1081 * accept clones from these extents.
1082 */
1083 ret = get_inode_info(found->root, ino, &i_size, NULL, NULL, NULL, NULL);
1084 if (ret < 0)
1085 return ret;
1086
1087 if (offset + bctx->extent_len > i_size)
1088 return 0;
1089
1090 /*
1091 * Make sure we don't consider clones from send_root that are
1092 * behind the current inode/offset.
1093 */
1094 if (found->root == bctx->sctx->send_root) {
1095 /*
1096 * TODO for the moment we don't accept clones from the inode
1097 * that is currently send. We may change this when
1098 * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1099 * file.
1100 */
1101 if (ino >= bctx->cur_objectid)
1102 return 0;
1103 /*if (ino > ctx->cur_objectid)
1104 return 0;
1105 if (offset + ctx->extent_len > ctx->cur_offset)
1106 return 0;*/
1107
1108 bctx->found++;
1109 found->found_refs++;
1110 found->ino = ino;
1111 found->offset = offset;
1112 return 0;
1113 }
1114
1115 bctx->found++;
1116 found->found_refs++;
1117 if (ino < found->ino) {
1118 found->ino = ino;
1119 found->offset = offset;
1120 } else if (found->ino == ino) {
1121 /*
1122 * same extent found more then once in the same file.
1123 */
1124 if (found->offset > offset + bctx->extent_len)
1125 found->offset = offset;
1126 }
1127
1128 return 0;
1129}
1130
1131/*
1132 * path must point to the extent item when called.
1133 */
1134static int find_extent_clone(struct send_ctx *sctx,
1135 struct btrfs_path *path,
1136 u64 ino, u64 data_offset,
1137 u64 ino_size,
1138 struct clone_root **found)
1139{
1140 int ret;
1141 int extent_type;
1142 u64 logical;
1143 u64 num_bytes;
1144 u64 extent_item_pos;
1145 struct btrfs_file_extent_item *fi;
1146 struct extent_buffer *eb = path->nodes[0];
1147 struct backref_ctx backref_ctx;
1148 struct clone_root *cur_clone_root;
1149 struct btrfs_key found_key;
1150 struct btrfs_path *tmp_path;
1151 u32 i;
1152
1153 tmp_path = alloc_path_for_send();
1154 if (!tmp_path)
1155 return -ENOMEM;
1156
1157 if (data_offset >= ino_size) {
1158 /*
1159 * There may be extents that lie behind the file's size.
1160 * I at least had this in combination with snapshotting while
1161 * writing large files.
1162 */
1163 ret = 0;
1164 goto out;
1165 }
1166
1167 fi = btrfs_item_ptr(eb, path->slots[0],
1168 struct btrfs_file_extent_item);
1169 extent_type = btrfs_file_extent_type(eb, fi);
1170 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1171 ret = -ENOENT;
1172 goto out;
1173 }
1174
1175 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1176 logical = btrfs_file_extent_disk_bytenr(eb, fi);
1177 if (logical == 0) {
1178 ret = -ENOENT;
1179 goto out;
1180 }
1181 logical += btrfs_file_extent_offset(eb, fi);
1182
1183 ret = extent_from_logical(sctx->send_root->fs_info,
1184 logical, tmp_path, &found_key);
1185 btrfs_release_path(tmp_path);
1186
1187 if (ret < 0)
1188 goto out;
1189 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1190 ret = -EIO;
1191 goto out;
1192 }
1193
1194 /*
1195 * Setup the clone roots.
1196 */
1197 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1198 cur_clone_root = sctx->clone_roots + i;
1199 cur_clone_root->ino = (u64)-1;
1200 cur_clone_root->offset = 0;
1201 cur_clone_root->found_refs = 0;
1202 }
1203
1204 backref_ctx.sctx = sctx;
1205 backref_ctx.found = 0;
1206 backref_ctx.cur_objectid = ino;
1207 backref_ctx.cur_offset = data_offset;
1208 backref_ctx.found_in_send_root = 0;
1209 backref_ctx.extent_len = num_bytes;
1210
1211 /*
1212 * The last extent of a file may be too large due to page alignment.
1213 * We need to adjust extent_len in this case so that the checks in
1214 * __iterate_backrefs work.
1215 */
1216 if (data_offset + num_bytes >= ino_size)
1217 backref_ctx.extent_len = ino_size - data_offset;
1218
1219 /*
1220 * Now collect all backrefs.
1221 */
1222 extent_item_pos = logical - found_key.objectid;
1223 ret = iterate_extent_inodes(sctx->send_root->fs_info,
1224 found_key.objectid, extent_item_pos, 1,
1225 __iterate_backrefs, &backref_ctx);
1226 if (ret < 0)
1227 goto out;
1228
1229 if (!backref_ctx.found_in_send_root) {
1230 /* found a bug in backref code? */
1231 ret = -EIO;
1232 printk(KERN_ERR "btrfs: ERROR did not find backref in "
1233 "send_root. inode=%llu, offset=%llu, "
1234 "logical=%llu\n",
1235 ino, data_offset, logical);
1236 goto out;
1237 }
1238
1239verbose_printk(KERN_DEBUG "btrfs: find_extent_clone: data_offset=%llu, "
1240 "ino=%llu, "
1241 "num_bytes=%llu, logical=%llu\n",
1242 data_offset, ino, num_bytes, logical);
1243
1244 if (!backref_ctx.found)
1245 verbose_printk("btrfs: no clones found\n");
1246
1247 cur_clone_root = NULL;
1248 for (i = 0; i < sctx->clone_roots_cnt; i++) {
1249 if (sctx->clone_roots[i].found_refs) {
1250 if (!cur_clone_root)
1251 cur_clone_root = sctx->clone_roots + i;
1252 else if (sctx->clone_roots[i].root == sctx->send_root)
1253 /* prefer clones from send_root over others */
1254 cur_clone_root = sctx->clone_roots + i;
1255 break;
1256 }
1257
1258 }
1259
1260 if (cur_clone_root) {
1261 *found = cur_clone_root;
1262 ret = 0;
1263 } else {
1264 ret = -ENOENT;
1265 }
1266
1267out:
1268 btrfs_free_path(tmp_path);
1269 return ret;
1270}
1271
1272static int read_symlink(struct send_ctx *sctx,
1273 struct btrfs_root *root,
1274 u64 ino,
1275 struct fs_path *dest)
1276{
1277 int ret;
1278 struct btrfs_path *path;
1279 struct btrfs_key key;
1280 struct btrfs_file_extent_item *ei;
1281 u8 type;
1282 u8 compression;
1283 unsigned long off;
1284 int len;
1285
1286 path = alloc_path_for_send();
1287 if (!path)
1288 return -ENOMEM;
1289
1290 key.objectid = ino;
1291 key.type = BTRFS_EXTENT_DATA_KEY;
1292 key.offset = 0;
1293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1294 if (ret < 0)
1295 goto out;
1296 BUG_ON(ret);
1297
1298 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1299 struct btrfs_file_extent_item);
1300 type = btrfs_file_extent_type(path->nodes[0], ei);
1301 compression = btrfs_file_extent_compression(path->nodes[0], ei);
1302 BUG_ON(type != BTRFS_FILE_EXTENT_INLINE);
1303 BUG_ON(compression);
1304
1305 off = btrfs_file_extent_inline_start(ei);
1306 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
1307
1308 ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len);
1309 if (ret < 0)
1310 goto out;
1311
1312out:
1313 btrfs_free_path(path);
1314 return ret;
1315}
1316
1317/*
1318 * Helper function to generate a file name that is unique in the root of
1319 * send_root and parent_root. This is used to generate names for orphan inodes.
1320 */
1321static int gen_unique_name(struct send_ctx *sctx,
1322 u64 ino, u64 gen,
1323 struct fs_path *dest)
1324{
1325 int ret = 0;
1326 struct btrfs_path *path;
1327 struct btrfs_dir_item *di;
1328 char tmp[64];
1329 int len;
1330 u64 idx = 0;
1331
1332 path = alloc_path_for_send();
1333 if (!path)
1334 return -ENOMEM;
1335
1336 while (1) {
1337 len = snprintf(tmp, sizeof(tmp) - 1, "o%llu-%llu-%llu",
1338 ino, gen, idx);
1339 if (len >= sizeof(tmp)) {
1340 /* should really not happen */
1341 ret = -EOVERFLOW;
1342 goto out;
1343 }
1344
1345 di = btrfs_lookup_dir_item(NULL, sctx->send_root,
1346 path, BTRFS_FIRST_FREE_OBJECTID,
1347 tmp, strlen(tmp), 0);
1348 btrfs_release_path(path);
1349 if (IS_ERR(di)) {
1350 ret = PTR_ERR(di);
1351 goto out;
1352 }
1353 if (di) {
1354 /* not unique, try again */
1355 idx++;
1356 continue;
1357 }
1358
1359 if (!sctx->parent_root) {
1360 /* unique */
1361 ret = 0;
1362 break;
1363 }
1364
1365 di = btrfs_lookup_dir_item(NULL, sctx->parent_root,
1366 path, BTRFS_FIRST_FREE_OBJECTID,
1367 tmp, strlen(tmp), 0);
1368 btrfs_release_path(path);
1369 if (IS_ERR(di)) {
1370 ret = PTR_ERR(di);
1371 goto out;
1372 }
1373 if (di) {
1374 /* not unique, try again */
1375 idx++;
1376 continue;
1377 }
1378 /* unique */
1379 break;
1380 }
1381
1382 ret = fs_path_add(dest, tmp, strlen(tmp));
1383
1384out:
1385 btrfs_free_path(path);
1386 return ret;
1387}
1388
1389enum inode_state {
1390 inode_state_no_change,
1391 inode_state_will_create,
1392 inode_state_did_create,
1393 inode_state_will_delete,
1394 inode_state_did_delete,
1395};
1396
1397static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen)
1398{
1399 int ret;
1400 int left_ret;
1401 int right_ret;
1402 u64 left_gen;
1403 u64 right_gen;
1404
1405 ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL,
1406 NULL);
1407 if (ret < 0 && ret != -ENOENT)
1408 goto out;
1409 left_ret = ret;
1410
1411 if (!sctx->parent_root) {
1412 right_ret = -ENOENT;
1413 } else {
1414 ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen,
1415 NULL, NULL, NULL);
1416 if (ret < 0 && ret != -ENOENT)
1417 goto out;
1418 right_ret = ret;
1419 }
1420
1421 if (!left_ret && !right_ret) {
1422 if (left_gen == gen && right_gen == gen)
1423 ret = inode_state_no_change;
1424 else if (left_gen == gen) {
1425 if (ino < sctx->send_progress)
1426 ret = inode_state_did_create;
1427 else
1428 ret = inode_state_will_create;
1429 } else if (right_gen == gen) {
1430 if (ino < sctx->send_progress)
1431 ret = inode_state_did_delete;
1432 else
1433 ret = inode_state_will_delete;
1434 } else {
1435 ret = -ENOENT;
1436 }
1437 } else if (!left_ret) {
1438 if (left_gen == gen) {
1439 if (ino < sctx->send_progress)
1440 ret = inode_state_did_create;
1441 else
1442 ret = inode_state_will_create;
1443 } else {
1444 ret = -ENOENT;
1445 }
1446 } else if (!right_ret) {
1447 if (right_gen == gen) {
1448 if (ino < sctx->send_progress)
1449 ret = inode_state_did_delete;
1450 else
1451 ret = inode_state_will_delete;
1452 } else {
1453 ret = -ENOENT;
1454 }
1455 } else {
1456 ret = -ENOENT;
1457 }
1458
1459out:
1460 return ret;
1461}
1462
1463static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen)
1464{
1465 int ret;
1466
1467 ret = get_cur_inode_state(sctx, ino, gen);
1468 if (ret < 0)
1469 goto out;
1470
1471 if (ret == inode_state_no_change ||
1472 ret == inode_state_did_create ||
1473 ret == inode_state_will_delete)
1474 ret = 1;
1475 else
1476 ret = 0;
1477
1478out:
1479 return ret;
1480}
1481
1482/*
1483 * Helper function to lookup a dir item in a dir.
1484 */
1485static int lookup_dir_item_inode(struct btrfs_root *root,
1486 u64 dir, const char *name, int name_len,
1487 u64 *found_inode,
1488 u8 *found_type)
1489{
1490 int ret = 0;
1491 struct btrfs_dir_item *di;
1492 struct btrfs_key key;
1493 struct btrfs_path *path;
1494
1495 path = alloc_path_for_send();
1496 if (!path)
1497 return -ENOMEM;
1498
1499 di = btrfs_lookup_dir_item(NULL, root, path,
1500 dir, name, name_len, 0);
1501 if (!di) {
1502 ret = -ENOENT;
1503 goto out;
1504 }
1505 if (IS_ERR(di)) {
1506 ret = PTR_ERR(di);
1507 goto out;
1508 }
1509 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
1510 *found_inode = key.objectid;
1511 *found_type = btrfs_dir_type(path->nodes[0], di);
1512
1513out:
1514 btrfs_free_path(path);
1515 return ret;
1516}
1517
1518static int get_first_ref(struct send_ctx *sctx,
1519 struct btrfs_root *root, u64 ino,
1520 u64 *dir, u64 *dir_gen, struct fs_path *name)
1521{
1522 int ret;
1523 struct btrfs_key key;
1524 struct btrfs_key found_key;
1525 struct btrfs_path *path;
1526 struct btrfs_inode_ref *iref;
1527 int len;
1528
1529 path = alloc_path_for_send();
1530 if (!path)
1531 return -ENOMEM;
1532
1533 key.objectid = ino;
1534 key.type = BTRFS_INODE_REF_KEY;
1535 key.offset = 0;
1536
1537 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
1538 if (ret < 0)
1539 goto out;
1540 if (!ret)
1541 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1542 path->slots[0]);
1543 if (ret || found_key.objectid != key.objectid ||
1544 found_key.type != key.type) {
1545 ret = -ENOENT;
1546 goto out;
1547 }
1548
1549 iref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1550 struct btrfs_inode_ref);
1551 len = btrfs_inode_ref_name_len(path->nodes[0], iref);
1552 ret = fs_path_add_from_extent_buffer(name, path->nodes[0],
1553 (unsigned long)(iref + 1), len);
1554 if (ret < 0)
1555 goto out;
1556 btrfs_release_path(path);
1557
1558 ret = get_inode_info(root, found_key.offset, NULL, dir_gen, NULL, NULL,
1559 NULL);
1560 if (ret < 0)
1561 goto out;
1562
1563 *dir = found_key.offset;
1564
1565out:
1566 btrfs_free_path(path);
1567 return ret;
1568}
1569
1570static int is_first_ref(struct send_ctx *sctx,
1571 struct btrfs_root *root,
1572 u64 ino, u64 dir,
1573 const char *name, int name_len)
1574{
1575 int ret;
1576 struct fs_path *tmp_name;
1577 u64 tmp_dir;
1578 u64 tmp_dir_gen;
1579
1580 tmp_name = fs_path_alloc(sctx);
1581 if (!tmp_name)
1582 return -ENOMEM;
1583
1584 ret = get_first_ref(sctx, root, ino, &tmp_dir, &tmp_dir_gen, tmp_name);
1585 if (ret < 0)
1586 goto out;
1587
1588 if (name_len != fs_path_len(tmp_name)) {
1589 ret = 0;
1590 goto out;
1591 }
1592
1593 ret = memcmp(tmp_name->start, name, name_len);
1594 if (ret)
1595 ret = 0;
1596 else
1597 ret = 1;
1598
1599out:
1600 fs_path_free(sctx, tmp_name);
1601 return ret;
1602}
1603
1604static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen,
1605 const char *name, int name_len,
1606 u64 *who_ino, u64 *who_gen)
1607{
1608 int ret = 0;
1609 u64 other_inode = 0;
1610 u8 other_type = 0;
1611
1612 if (!sctx->parent_root)
1613 goto out;
1614
1615 ret = is_inode_existent(sctx, dir, dir_gen);
1616 if (ret <= 0)
1617 goto out;
1618
1619 ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len,
1620 &other_inode, &other_type);
1621 if (ret < 0 && ret != -ENOENT)
1622 goto out;
1623 if (ret) {
1624 ret = 0;
1625 goto out;
1626 }
1627
1628 if (other_inode > sctx->send_progress) {
1629 ret = get_inode_info(sctx->parent_root, other_inode, NULL,
1630 who_gen, NULL, NULL, NULL);
1631 if (ret < 0)
1632 goto out;
1633
1634 ret = 1;
1635 *who_ino = other_inode;
1636 } else {
1637 ret = 0;
1638 }
1639
1640out:
1641 return ret;
1642}
1643
1644static int did_overwrite_ref(struct send_ctx *sctx,
1645 u64 dir, u64 dir_gen,
1646 u64 ino, u64 ino_gen,
1647 const char *name, int name_len)
1648{
1649 int ret = 0;
1650 u64 gen;
1651 u64 ow_inode;
1652 u8 other_type;
1653
1654 if (!sctx->parent_root)
1655 goto out;
1656
1657 ret = is_inode_existent(sctx, dir, dir_gen);
1658 if (ret <= 0)
1659 goto out;
1660
1661 /* check if the ref was overwritten by another ref */
1662 ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len,
1663 &ow_inode, &other_type);
1664 if (ret < 0 && ret != -ENOENT)
1665 goto out;
1666 if (ret) {
1667 /* was never and will never be overwritten */
1668 ret = 0;
1669 goto out;
1670 }
1671
1672 ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL,
1673 NULL);
1674 if (ret < 0)
1675 goto out;
1676
1677 if (ow_inode == ino && gen == ino_gen) {
1678 ret = 0;
1679 goto out;
1680 }
1681
1682 /* we know that it is or will be overwritten. check this now */
1683 if (ow_inode < sctx->send_progress)
1684 ret = 1;
1685 else
1686 ret = 0;
1687
1688out:
1689 return ret;
1690}
1691
1692static int did_overwrite_first_ref(struct send_ctx *sctx, u64 ino, u64 gen)
1693{
1694 int ret = 0;
1695 struct fs_path *name = NULL;
1696 u64 dir;
1697 u64 dir_gen;
1698
1699 if (!sctx->parent_root)
1700 goto out;
1701
1702 name = fs_path_alloc(sctx);
1703 if (!name)
1704 return -ENOMEM;
1705
1706 ret = get_first_ref(sctx, sctx->parent_root, ino, &dir, &dir_gen, name);
1707 if (ret < 0)
1708 goto out;
1709
1710 ret = did_overwrite_ref(sctx, dir, dir_gen, ino, gen,
1711 name->start, fs_path_len(name));
1712 if (ret < 0)
1713 goto out;
1714
1715out:
1716 fs_path_free(sctx, name);
1717 return ret;
1718}
1719
1720static int name_cache_insert(struct send_ctx *sctx,
1721 struct name_cache_entry *nce)
1722{
1723 int ret = 0;
1724 struct name_cache_entry **ncea;
1725
1726 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1727 if (ncea) {
1728 if (!ncea[0])
1729 ncea[0] = nce;
1730 else if (!ncea[1])
1731 ncea[1] = nce;
1732 else
1733 BUG();
1734 } else {
1735 ncea = kmalloc(sizeof(void *) * 2, GFP_NOFS);
1736 if (!ncea)
1737 return -ENOMEM;
1738
1739 ncea[0] = nce;
1740 ncea[1] = NULL;
1741 ret = radix_tree_insert(&sctx->name_cache, nce->ino, ncea);
1742 if (ret < 0)
1743 return ret;
1744 }
1745 list_add_tail(&nce->list, &sctx->name_cache_list);
1746 sctx->name_cache_size++;
1747
1748 return ret;
1749}
1750
1751static void name_cache_delete(struct send_ctx *sctx,
1752 struct name_cache_entry *nce)
1753{
1754 struct name_cache_entry **ncea;
1755
1756 ncea = radix_tree_lookup(&sctx->name_cache, nce->ino);
1757 BUG_ON(!ncea);
1758
1759 if (ncea[0] == nce)
1760 ncea[0] = NULL;
1761 else if (ncea[1] == nce)
1762 ncea[1] = NULL;
1763 else
1764 BUG();
1765
1766 if (!ncea[0] && !ncea[1]) {
1767 radix_tree_delete(&sctx->name_cache, nce->ino);
1768 kfree(ncea);
1769 }
1770
1771 list_del(&nce->list);
1772
1773 sctx->name_cache_size--;
1774}
1775
1776static struct name_cache_entry *name_cache_search(struct send_ctx *sctx,
1777 u64 ino, u64 gen)
1778{
1779 struct name_cache_entry **ncea;
1780
1781 ncea = radix_tree_lookup(&sctx->name_cache, ino);
1782 if (!ncea)
1783 return NULL;
1784
1785 if (ncea[0] && ncea[0]->gen == gen)
1786 return ncea[0];
1787 else if (ncea[1] && ncea[1]->gen == gen)
1788 return ncea[1];
1789 return NULL;
1790}
1791
1792static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce)
1793{
1794 list_del(&nce->list);
1795 list_add_tail(&nce->list, &sctx->name_cache_list);
1796}
1797
1798static void name_cache_clean_unused(struct send_ctx *sctx)
1799{
1800 struct name_cache_entry *nce;
1801
1802 if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE)
1803 return;
1804
1805 while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) {
1806 nce = list_entry(sctx->name_cache_list.next,
1807 struct name_cache_entry, list);
1808 name_cache_delete(sctx, nce);
1809 kfree(nce);
1810 }
1811}
1812
1813static void name_cache_free(struct send_ctx *sctx)
1814{
1815 struct name_cache_entry *nce;
1816 struct name_cache_entry *tmp;
1817
1818 list_for_each_entry_safe(nce, tmp, &sctx->name_cache_list, list) {
1819 name_cache_delete(sctx, nce);
1820 }
1821}
1822
1823static int __get_cur_name_and_parent(struct send_ctx *sctx,
1824 u64 ino, u64 gen,
1825 u64 *parent_ino,
1826 u64 *parent_gen,
1827 struct fs_path *dest)
1828{
1829 int ret;
1830 int nce_ret;
1831 struct btrfs_path *path = NULL;
1832 struct name_cache_entry *nce = NULL;
1833
1834 nce = name_cache_search(sctx, ino, gen);
1835 if (nce) {
1836 if (ino < sctx->send_progress && nce->need_later_update) {
1837 name_cache_delete(sctx, nce);
1838 kfree(nce);
1839 nce = NULL;
1840 } else {
1841 name_cache_used(sctx, nce);
1842 *parent_ino = nce->parent_ino;
1843 *parent_gen = nce->parent_gen;
1844 ret = fs_path_add(dest, nce->name, nce->name_len);
1845 if (ret < 0)
1846 goto out;
1847 ret = nce->ret;
1848 goto out;
1849 }
1850 }
1851
1852 path = alloc_path_for_send();
1853 if (!path)
1854 return -ENOMEM;
1855
1856 ret = is_inode_existent(sctx, ino, gen);
1857 if (ret < 0)
1858 goto out;
1859
1860 if (!ret) {
1861 ret = gen_unique_name(sctx, ino, gen, dest);
1862 if (ret < 0)
1863 goto out;
1864 ret = 1;
1865 goto out_cache;
1866 }
1867
1868 if (ino < sctx->send_progress)
1869 ret = get_first_ref(sctx, sctx->send_root, ino,
1870 parent_ino, parent_gen, dest);
1871 else
1872 ret = get_first_ref(sctx, sctx->parent_root, ino,
1873 parent_ino, parent_gen, dest);
1874 if (ret < 0)
1875 goto out;
1876
1877 ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen,
1878 dest->start, dest->end - dest->start);
1879 if (ret < 0)
1880 goto out;
1881 if (ret) {
1882 fs_path_reset(dest);
1883 ret = gen_unique_name(sctx, ino, gen, dest);
1884 if (ret < 0)
1885 goto out;
1886 ret = 1;
1887 }
1888
1889out_cache:
1890 nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_NOFS);
1891 if (!nce) {
1892 ret = -ENOMEM;
1893 goto out;
1894 }
1895
1896 nce->ino = ino;
1897 nce->gen = gen;
1898 nce->parent_ino = *parent_ino;
1899 nce->parent_gen = *parent_gen;
1900 nce->name_len = fs_path_len(dest);
1901 nce->ret = ret;
1902 strcpy(nce->name, dest->start);
1903 memset(&nce->use_list, 0, sizeof(nce->use_list));
1904
1905 if (ino < sctx->send_progress)
1906 nce->need_later_update = 0;
1907 else
1908 nce->need_later_update = 1;
1909
1910 nce_ret = name_cache_insert(sctx, nce);
1911 if (nce_ret < 0)
1912 ret = nce_ret;
1913 name_cache_clean_unused(sctx);
1914
1915out:
1916 btrfs_free_path(path);
1917 return ret;
1918}
1919
1920/*
1921 * Magic happens here. This function returns the first ref to an inode as it
1922 * would look like while receiving the stream at this point in time.
1923 * We walk the path up to the root. For every inode in between, we check if it
1924 * was already processed/sent. If yes, we continue with the parent as found
1925 * in send_root. If not, we continue with the parent as found in parent_root.
1926 * If we encounter an inode that was deleted at this point in time, we use the
1927 * inodes "orphan" name instead of the real name and stop. Same with new inodes
1928 * that were not created yet and overwritten inodes/refs.
1929 *
1930 * When do we have have orphan inodes:
1931 * 1. When an inode is freshly created and thus no valid refs are available yet
1932 * 2. When a directory lost all it's refs (deleted) but still has dir items
1933 * inside which were not processed yet (pending for move/delete). If anyone
1934 * tried to get the path to the dir items, it would get a path inside that
1935 * orphan directory.
1936 * 3. When an inode is moved around or gets new links, it may overwrite the ref
1937 * of an unprocessed inode. If in that case the first ref would be
1938 * overwritten, the overwritten inode gets "orphanized". Later when we
1939 * process this overwritten inode, it is restored at a new place by moving
1940 * the orphan inode.
1941 *
1942 * sctx->send_progress tells this function at which point in time receiving
1943 * would be.
1944 */
1945static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
1946 struct fs_path *dest)
1947{
1948 int ret = 0;
1949 struct fs_path *name = NULL;
1950 u64 parent_inode = 0;
1951 u64 parent_gen = 0;
1952 int stop = 0;
1953
1954 name = fs_path_alloc(sctx);
1955 if (!name) {
1956 ret = -ENOMEM;
1957 goto out;
1958 }
1959
1960 dest->reversed = 1;
1961 fs_path_reset(dest);
1962
1963 while (!stop && ino != BTRFS_FIRST_FREE_OBJECTID) {
1964 fs_path_reset(name);
1965
1966 ret = __get_cur_name_and_parent(sctx, ino, gen,
1967 &parent_inode, &parent_gen, name);
1968 if (ret < 0)
1969 goto out;
1970 if (ret)
1971 stop = 1;
1972
1973 ret = fs_path_add_path(dest, name);
1974 if (ret < 0)
1975 goto out;
1976
1977 ino = parent_inode;
1978 gen = parent_gen;
1979 }
1980
1981out:
1982 fs_path_free(sctx, name);
1983 if (!ret)
1984 fs_path_unreverse(dest);
1985 return ret;
1986}
1987
1988/*
1989 * Called for regular files when sending extents data. Opens a struct file
1990 * to read from the file.
1991 */
1992static int open_cur_inode_file(struct send_ctx *sctx)
1993{
1994 int ret = 0;
1995 struct btrfs_key key;
1996 struct vfsmount *mnt;
1997 struct inode *inode;
1998 struct dentry *dentry;
1999 struct file *filp;
2000 int new = 0;
2001
2002 if (sctx->cur_inode_filp)
2003 goto out;
2004
2005 key.objectid = sctx->cur_ino;
2006 key.type = BTRFS_INODE_ITEM_KEY;
2007 key.offset = 0;
2008
2009 inode = btrfs_iget(sctx->send_root->fs_info->sb, &key, sctx->send_root,
2010 &new);
2011 if (IS_ERR(inode)) {
2012 ret = PTR_ERR(inode);
2013 goto out;
2014 }
2015
2016 dentry = d_obtain_alias(inode);
2017 inode = NULL;
2018 if (IS_ERR(dentry)) {
2019 ret = PTR_ERR(dentry);
2020 goto out;
2021 }
2022
2023 mnt = mntget(sctx->mnt);
2024 filp = dentry_open(dentry, mnt, O_RDONLY | O_LARGEFILE, current_cred());
2025 dentry = NULL;
2026 mnt = NULL;
2027 if (IS_ERR(filp)) {
2028 ret = PTR_ERR(filp);
2029 goto out;
2030 }
2031 sctx->cur_inode_filp = filp;
2032
2033out:
2034 /*
2035 * no xxxput required here as every vfs op
2036 * does it by itself on failure
2037 */
2038 return ret;
2039}
2040
2041/*
2042 * Closes the struct file that was created in open_cur_inode_file
2043 */
2044static int close_cur_inode_file(struct send_ctx *sctx)
2045{
2046 int ret = 0;
2047
2048 if (!sctx->cur_inode_filp)
2049 goto out;
2050
2051 ret = filp_close(sctx->cur_inode_filp, NULL);
2052 sctx->cur_inode_filp = NULL;
2053
2054out:
2055 return ret;
2056}
2057
2058/*
2059 * Sends a BTRFS_SEND_C_SUBVOL command/item to userspace
2060 */
2061static int send_subvol_begin(struct send_ctx *sctx)
2062{
2063 int ret;
2064 struct btrfs_root *send_root = sctx->send_root;
2065 struct btrfs_root *parent_root = sctx->parent_root;
2066 struct btrfs_path *path;
2067 struct btrfs_key key;
2068 struct btrfs_root_ref *ref;
2069 struct extent_buffer *leaf;
2070 char *name = NULL;
2071 int namelen;
2072
2073 path = alloc_path_for_send();
2074 if (!path)
2075 return -ENOMEM;
2076
2077 name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_NOFS);
2078 if (!name) {
2079 btrfs_free_path(path);
2080 return -ENOMEM;
2081 }
2082
2083 key.objectid = send_root->objectid;
2084 key.type = BTRFS_ROOT_BACKREF_KEY;
2085 key.offset = 0;
2086
2087 ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root,
2088 &key, path, 1, 0);
2089 if (ret < 0)
2090 goto out;
2091 if (ret) {
2092 ret = -ENOENT;
2093 goto out;
2094 }
2095
2096 leaf = path->nodes[0];
2097 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2098 if (key.type != BTRFS_ROOT_BACKREF_KEY ||
2099 key.objectid != send_root->objectid) {
2100 ret = -ENOENT;
2101 goto out;
2102 }
2103 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
2104 namelen = btrfs_root_ref_name_len(leaf, ref);
2105 read_extent_buffer(leaf, name, (unsigned long)(ref + 1), namelen);
2106 btrfs_release_path(path);
2107
2108 if (ret < 0)
2109 goto out;
2110
2111 if (parent_root) {
2112 ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT);
2113 if (ret < 0)
2114 goto out;
2115 } else {
2116 ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL);
2117 if (ret < 0)
2118 goto out;
2119 }
2120
2121 TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen);
2122 TLV_PUT_UUID(sctx, BTRFS_SEND_A_UUID,
2123 sctx->send_root->root_item.uuid);
2124 TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID,
2125 sctx->send_root->root_item.ctransid);
2126 if (parent_root) {
2127 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
2128 sctx->parent_root->root_item.uuid);
2129 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
2130 sctx->parent_root->root_item.ctransid);
2131 }
2132
2133 ret = send_cmd(sctx);
2134
2135tlv_put_failure:
2136out:
2137 btrfs_free_path(path);
2138 kfree(name);
2139 return ret;
2140}
2141
2142static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size)
2143{
2144 int ret = 0;
2145 struct fs_path *p;
2146
2147verbose_printk("btrfs: send_truncate %llu size=%llu\n", ino, size);
2148
2149 p = fs_path_alloc(sctx);
2150 if (!p)
2151 return -ENOMEM;
2152
2153 ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE);
2154 if (ret < 0)
2155 goto out;
2156
2157 ret = get_cur_path(sctx, ino, gen, p);
2158 if (ret < 0)
2159 goto out;
2160 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2161 TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size);
2162
2163 ret = send_cmd(sctx);
2164
2165tlv_put_failure:
2166out:
2167 fs_path_free(sctx, p);
2168 return ret;
2169}
2170
2171static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode)
2172{
2173 int ret = 0;
2174 struct fs_path *p;
2175
2176verbose_printk("btrfs: send_chmod %llu mode=%llu\n", ino, mode);
2177
2178 p = fs_path_alloc(sctx);
2179 if (!p)
2180 return -ENOMEM;
2181
2182 ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD);
2183 if (ret < 0)
2184 goto out;
2185
2186 ret = get_cur_path(sctx, ino, gen, p);
2187 if (ret < 0)
2188 goto out;
2189 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2190 TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777);
2191
2192 ret = send_cmd(sctx);
2193
2194tlv_put_failure:
2195out:
2196 fs_path_free(sctx, p);
2197 return ret;
2198}
2199
2200static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid)
2201{
2202 int ret = 0;
2203 struct fs_path *p;
2204
2205verbose_printk("btrfs: send_chown %llu uid=%llu, gid=%llu\n", ino, uid, gid);
2206
2207 p = fs_path_alloc(sctx);
2208 if (!p)
2209 return -ENOMEM;
2210
2211 ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN);
2212 if (ret < 0)
2213 goto out;
2214
2215 ret = get_cur_path(sctx, ino, gen, p);
2216 if (ret < 0)
2217 goto out;
2218 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2219 TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid);
2220 TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid);
2221
2222 ret = send_cmd(sctx);
2223
2224tlv_put_failure:
2225out:
2226 fs_path_free(sctx, p);
2227 return ret;
2228}
2229
2230static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen)
2231{
2232 int ret = 0;
2233 struct fs_path *p = NULL;
2234 struct btrfs_inode_item *ii;
2235 struct btrfs_path *path = NULL;
2236 struct extent_buffer *eb;
2237 struct btrfs_key key;
2238 int slot;
2239
2240verbose_printk("btrfs: send_utimes %llu\n", ino);
2241
2242 p = fs_path_alloc(sctx);
2243 if (!p)
2244 return -ENOMEM;
2245
2246 path = alloc_path_for_send();
2247 if (!path) {
2248 ret = -ENOMEM;
2249 goto out;
2250 }
2251
2252 key.objectid = ino;
2253 key.type = BTRFS_INODE_ITEM_KEY;
2254 key.offset = 0;
2255 ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0);
2256 if (ret < 0)
2257 goto out;
2258
2259 eb = path->nodes[0];
2260 slot = path->slots[0];
2261 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2262
2263 ret = begin_cmd(sctx, BTRFS_SEND_C_UTIMES);
2264 if (ret < 0)
2265 goto out;
2266
2267 ret = get_cur_path(sctx, ino, gen, p);
2268 if (ret < 0)
2269 goto out;
2270 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2271 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb,
2272 btrfs_inode_atime(ii));
2273 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb,
2274 btrfs_inode_mtime(ii));
2275 TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb,
2276 btrfs_inode_ctime(ii));
2277 /* TODO otime? */
2278
2279 ret = send_cmd(sctx);
2280
2281tlv_put_failure:
2282out:
2283 fs_path_free(sctx, p);
2284 btrfs_free_path(path);
2285 return ret;
2286}
2287
2288/*
2289 * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have
2290 * a valid path yet because we did not process the refs yet. So, the inode
2291 * is created as orphan.
2292 */
2293static int send_create_inode(struct send_ctx *sctx, struct btrfs_path *path,
2294 struct btrfs_key *key)
2295{
2296 int ret = 0;
2297 struct extent_buffer *eb = path->nodes[0];
2298 struct btrfs_inode_item *ii;
2299 struct fs_path *p;
2300 int slot = path->slots[0];
2301 int cmd;
2302 u64 mode;
2303
2304verbose_printk("btrfs: send_create_inode %llu\n", sctx->cur_ino);
2305
2306 p = fs_path_alloc(sctx);
2307 if (!p)
2308 return -ENOMEM;
2309
2310 ii = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
2311 mode = btrfs_inode_mode(eb, ii);
2312
2313 if (S_ISREG(mode))
2314 cmd = BTRFS_SEND_C_MKFILE;
2315 else if (S_ISDIR(mode))
2316 cmd = BTRFS_SEND_C_MKDIR;
2317 else if (S_ISLNK(mode))
2318 cmd = BTRFS_SEND_C_SYMLINK;
2319 else if (S_ISCHR(mode) || S_ISBLK(mode))
2320 cmd = BTRFS_SEND_C_MKNOD;
2321 else if (S_ISFIFO(mode))
2322 cmd = BTRFS_SEND_C_MKFIFO;
2323 else if (S_ISSOCK(mode))
2324 cmd = BTRFS_SEND_C_MKSOCK;
2325 else {
2326 printk(KERN_WARNING "btrfs: unexpected inode type %o",
2327 (int)(mode & S_IFMT));
2328 ret = -ENOTSUPP;
2329 goto out;
2330 }
2331
2332 ret = begin_cmd(sctx, cmd);
2333 if (ret < 0)
2334 goto out;
2335
2336 ret = gen_unique_name(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
2337 if (ret < 0)
2338 goto out;
2339
2340 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
2341 TLV_PUT_U64(sctx, BTRFS_SEND_A_INO, sctx->cur_ino);
2342
2343 if (S_ISLNK(mode)) {
2344 fs_path_reset(p);
2345 ret = read_symlink(sctx, sctx->send_root, sctx->cur_ino, p);
2346 if (ret < 0)
2347 goto out;
2348 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, p);
2349 } else if (S_ISCHR(mode) || S_ISBLK(mode) ||
2350 S_ISFIFO(mode) || S_ISSOCK(mode)) {
2351 TLV_PUT_U64(sctx, BTRFS_SEND_A_RDEV, btrfs_inode_rdev(eb, ii));
2352 }
2353
2354 ret = send_cmd(sctx);
2355 if (ret < 0)
2356 goto out;
2357
2358
2359tlv_put_failure:
2360out:
2361 fs_path_free(sctx, p);
2362 return ret;
2363}
2364
2365struct recorded_ref {
2366 struct list_head list;
2367 char *dir_path;
2368 char *name;
2369 struct fs_path *full_path;
2370 u64 dir;
2371 u64 dir_gen;
2372 int dir_path_len;
2373 int name_len;
2374};
2375
2376/*
2377 * We need to process new refs before deleted refs, but compare_tree gives us
2378 * everything mixed. So we first record all refs and later process them.
2379 * This function is a helper to record one ref.
2380 */
2381static int record_ref(struct list_head *head, u64 dir,
2382 u64 dir_gen, struct fs_path *path)
2383{
2384 struct recorded_ref *ref;
2385 char *tmp;
2386
2387 ref = kmalloc(sizeof(*ref), GFP_NOFS);
2388 if (!ref)
2389 return -ENOMEM;
2390
2391 ref->dir = dir;
2392 ref->dir_gen = dir_gen;
2393 ref->full_path = path;
2394
2395 tmp = strrchr(ref->full_path->start, '/');
2396 if (!tmp) {
2397 ref->name_len = ref->full_path->end - ref->full_path->start;
2398 ref->name = ref->full_path->start;
2399 ref->dir_path_len = 0;
2400 ref->dir_path = ref->full_path->start;
2401 } else {
2402 tmp++;
2403 ref->name_len = ref->full_path->end - tmp;
2404 ref->name = tmp;
2405 ref->dir_path = ref->full_path->start;
2406 ref->dir_path_len = ref->full_path->end -
2407 ref->full_path->start - 1 - ref->name_len;
2408 }
2409
2410 list_add_tail(&ref->list, head);
2411 return 0;
2412}
2413
2414static void __free_recorded_refs(struct send_ctx *sctx, struct list_head *head)
2415{
2416 struct recorded_ref *cur;
2417 struct recorded_ref *tmp;
2418
2419 list_for_each_entry_safe(cur, tmp, head, list) {
2420 fs_path_free(sctx, cur->full_path);
2421 kfree(cur);
2422 }
2423 INIT_LIST_HEAD(head);
2424}
2425
2426static void free_recorded_refs(struct send_ctx *sctx)
2427{
2428 __free_recorded_refs(sctx, &sctx->new_refs);
2429 __free_recorded_refs(sctx, &sctx->deleted_refs);
2430}
2431
2432/*
2433 * Renames/moves a file/dir to it's orphan name. Used when the first
2434 * ref of an unprocessed inode gets overwritten and for all non empty
2435 * directories.
2436 */
2437static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen,
2438 struct fs_path *path)
2439{
2440 int ret;
2441 struct fs_path *orphan;
2442
2443 orphan = fs_path_alloc(sctx);
2444 if (!orphan)
2445 return -ENOMEM;
2446
2447 ret = gen_unique_name(sctx, ino, gen, orphan);
2448 if (ret < 0)
2449 goto out;
2450
2451 ret = send_rename(sctx, path, orphan);
2452
2453out:
2454 fs_path_free(sctx, orphan);
2455 return ret;
2456}
2457
2458/*
2459 * Returns 1 if a directory can be removed at this point in time.
2460 * We check this by iterating all dir items and checking if the inode behind
2461 * the dir item was already processed.
2462 */
2463static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 send_progress)
2464{
2465 int ret = 0;
2466 struct btrfs_root *root = sctx->parent_root;
2467 struct btrfs_path *path;
2468 struct btrfs_key key;
2469 struct btrfs_key found_key;
2470 struct btrfs_key loc;
2471 struct btrfs_dir_item *di;
2472
2473 path = alloc_path_for_send();
2474 if (!path)
2475 return -ENOMEM;
2476
2477 key.objectid = dir;
2478 key.type = BTRFS_DIR_INDEX_KEY;
2479 key.offset = 0;
2480
2481 while (1) {
2482 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
2483 if (ret < 0)
2484 goto out;
2485 if (!ret) {
2486 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2487 path->slots[0]);
2488 }
2489 if (ret || found_key.objectid != key.objectid ||
2490 found_key.type != key.type) {
2491 break;
2492 }
2493
2494 di = btrfs_item_ptr(path->nodes[0], path->slots[0],
2495 struct btrfs_dir_item);
2496 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc);
2497
2498 if (loc.objectid > send_progress) {
2499 ret = 0;
2500 goto out;
2501 }
2502
2503 btrfs_release_path(path);
2504 key.offset = found_key.offset + 1;
2505 }
2506
2507 ret = 1;
2508
2509out:
2510 btrfs_free_path(path);
2511 return ret;
2512}
2513
2514struct finish_unordered_dir_ctx {
2515 struct send_ctx *sctx;
2516 struct fs_path *cur_path;
2517 struct fs_path *dir_path;
2518 u64 dir_ino;
2519 int need_delete;
2520 int delete_pass;
2521};
2522
2523int __finish_unordered_dir(int num, struct btrfs_key *di_key,
2524 const char *name, int name_len,
2525 const char *data, int data_len,
2526 u8 type, void *ctx)
2527{
2528 int ret = 0;
2529 struct finish_unordered_dir_ctx *fctx = ctx;
2530 struct send_ctx *sctx = fctx->sctx;
2531 u64 di_gen;
2532 u64 di_mode;
2533 int is_orphan = 0;
2534
2535 if (di_key->objectid >= fctx->dir_ino)
2536 goto out;
2537
2538 fs_path_reset(fctx->cur_path);
2539
2540 ret = get_inode_info(sctx->send_root, di_key->objectid,
2541 NULL, &di_gen, &di_mode, NULL, NULL);
2542 if (ret < 0)
2543 goto out;
2544
2545 ret = is_first_ref(sctx, sctx->send_root, di_key->objectid,
2546 fctx->dir_ino, name, name_len);
2547 if (ret < 0)
2548 goto out;
2549 if (ret) {
2550 is_orphan = 1;
2551 ret = gen_unique_name(sctx, di_key->objectid, di_gen,
2552 fctx->cur_path);
2553 } else {
2554 ret = get_cur_path(sctx, di_key->objectid, di_gen,
2555 fctx->cur_path);
2556 }
2557 if (ret < 0)
2558 goto out;
2559
2560 ret = fs_path_add(fctx->dir_path, name, name_len);
2561 if (ret < 0)
2562 goto out;
2563
2564 if (!fctx->delete_pass) {
2565 if (S_ISDIR(di_mode)) {
2566 ret = send_rename(sctx, fctx->cur_path,
2567 fctx->dir_path);
2568 } else {
2569 ret = send_link(sctx, fctx->dir_path,
2570 fctx->cur_path);
2571 if (is_orphan)
2572 fctx->need_delete = 1;
2573 }
2574 } else if (!S_ISDIR(di_mode)) {
2575 ret = send_unlink(sctx, fctx->cur_path);
2576 } else {
2577 ret = 0;
2578 }
2579
2580 fs_path_remove(fctx->dir_path);
2581
2582out:
2583 return ret;
2584}
2585
2586/*
2587 * Go through all dir items and see if we find refs which could not be created
2588 * in the past because the dir did not exist at that time.
2589 */
2590static int finish_outoforder_dir(struct send_ctx *sctx, u64 dir, u64 dir_gen)
2591{
2592 int ret = 0;
2593 struct btrfs_path *path = NULL;
2594 struct btrfs_key key;
2595 struct btrfs_key found_key;
2596 struct extent_buffer *eb;
2597 struct finish_unordered_dir_ctx fctx;
2598 int slot;
2599
2600 path = alloc_path_for_send();
2601 if (!path) {
2602 ret = -ENOMEM;
2603 goto out;
2604 }
2605
2606 memset(&fctx, 0, sizeof(fctx));
2607 fctx.sctx = sctx;
2608 fctx.cur_path = fs_path_alloc(sctx);
2609 fctx.dir_path = fs_path_alloc(sctx);
2610 if (!fctx.cur_path || !fctx.dir_path) {
2611 ret = -ENOMEM;
2612 goto out;
2613 }
2614 fctx.dir_ino = dir;
2615
2616 ret = get_cur_path(sctx, dir, dir_gen, fctx.dir_path);
2617 if (ret < 0)
2618 goto out;
2619
2620 /*
2621 * We do two passes. The first links in the new refs and the second
2622 * deletes orphans if required. Deletion of orphans is not required for
2623 * directory inodes, as we always have only one ref and use rename
2624 * instead of link for those.
2625 */
2626
2627again:
2628 key.objectid = dir;
2629 key.type = BTRFS_DIR_ITEM_KEY;
2630 key.offset = 0;
2631 while (1) {
2632 ret = btrfs_search_slot_for_read(sctx->send_root, &key, path,
2633 1, 0);
2634 if (ret < 0)
2635 goto out;
2636 eb = path->nodes[0];
2637 slot = path->slots[0];
2638 btrfs_item_key_to_cpu(eb, &found_key, slot);
2639
2640 if (found_key.objectid != key.objectid ||
2641 found_key.type != key.type) {
2642 btrfs_release_path(path);
2643 break;
2644 }
2645
2646 ret = iterate_dir_item(sctx, sctx->send_root, path,
2647 &found_key, __finish_unordered_dir,
2648 &fctx);
2649 if (ret < 0)
2650 goto out;
2651
2652 key.offset = found_key.offset + 1;
2653 btrfs_release_path(path);
2654 }
2655
2656 if (!fctx.delete_pass && fctx.need_delete) {
2657 fctx.delete_pass = 1;
2658 goto again;
2659 }
2660
2661out:
2662 btrfs_free_path(path);
2663 fs_path_free(sctx, fctx.cur_path);
2664 fs_path_free(sctx, fctx.dir_path);
2665 return ret;
2666}
2667
2668/*
2669 * This does all the move/link/unlink/rmdir magic.
2670 */
2671static int process_recorded_refs(struct send_ctx *sctx)
2672{
2673 int ret = 0;
2674 struct recorded_ref *cur;
2675 struct ulist *check_dirs = NULL;
2676 struct ulist_iterator uit;
2677 struct ulist_node *un;
2678 struct fs_path *valid_path = NULL;
2679 u64 ow_inode;
2680 u64 ow_gen;
2681 int did_overwrite = 0;
2682 int is_orphan = 0;
2683
2684verbose_printk("btrfs: process_recorded_refs %llu\n", sctx->cur_ino);
2685
2686 valid_path = fs_path_alloc(sctx);
2687 if (!valid_path) {
2688 ret = -ENOMEM;
2689 goto out;
2690 }
2691
2692 check_dirs = ulist_alloc(GFP_NOFS);
2693 if (!check_dirs) {
2694 ret = -ENOMEM;
2695 goto out;
2696 }
2697
2698 /*
2699 * First, check if the first ref of the current inode was overwritten
2700 * before. If yes, we know that the current inode was already orphanized
2701 * and thus use the orphan name. If not, we can use get_cur_path to
2702 * get the path of the first ref as it would like while receiving at
2703 * this point in time.
2704 * New inodes are always orphan at the beginning, so force to use the
2705 * orphan name in this case.
2706 * The first ref is stored in valid_path and will be updated if it
2707 * gets moved around.
2708 */
2709 if (!sctx->cur_inode_new) {
2710 ret = did_overwrite_first_ref(sctx, sctx->cur_ino,
2711 sctx->cur_inode_gen);
2712 if (ret < 0)
2713 goto out;
2714 if (ret)
2715 did_overwrite = 1;
2716 }
2717 if (sctx->cur_inode_new || did_overwrite) {
2718 ret = gen_unique_name(sctx, sctx->cur_ino,
2719 sctx->cur_inode_gen, valid_path);
2720 if (ret < 0)
2721 goto out;
2722 is_orphan = 1;
2723 } else {
2724 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen,
2725 valid_path);
2726 if (ret < 0)
2727 goto out;
2728 }
2729
2730 list_for_each_entry(cur, &sctx->new_refs, list) {
2731 /*
2732 * Check if this new ref would overwrite the first ref of
2733 * another unprocessed inode. If yes, orphanize the
2734 * overwritten inode. If we find an overwritten ref that is
2735 * not the first ref, simply unlink it.
2736 */
2737 ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2738 cur->name, cur->name_len,
2739 &ow_inode, &ow_gen);
2740 if (ret < 0)
2741 goto out;
2742 if (ret) {
2743 ret = is_first_ref(sctx, sctx->parent_root,
2744 ow_inode, cur->dir, cur->name,
2745 cur->name_len);
2746 if (ret < 0)
2747 goto out;
2748 if (ret) {
2749 ret = orphanize_inode(sctx, ow_inode, ow_gen,
2750 cur->full_path);
2751 if (ret < 0)
2752 goto out;
2753 } else {
2754 ret = send_unlink(sctx, cur->full_path);
2755 if (ret < 0)
2756 goto out;
2757 }
2758 }
2759
2760 /*
2761 * link/move the ref to the new place. If we have an orphan
2762 * inode, move it and update valid_path. If not, link or move
2763 * it depending on the inode mode.
2764 */
2765 if (is_orphan && !sctx->cur_inode_first_ref_orphan) {
2766 ret = send_rename(sctx, valid_path, cur->full_path);
2767 if (ret < 0)
2768 goto out;
2769 is_orphan = 0;
2770 ret = fs_path_copy(valid_path, cur->full_path);
2771 if (ret < 0)
2772 goto out;
2773 } else {
2774 if (S_ISDIR(sctx->cur_inode_mode)) {
2775 /*
2776 * Dirs can't be linked, so move it. For moved
2777 * dirs, we always have one new and one deleted
2778 * ref. The deleted ref is ignored later.
2779 */
2780 ret = send_rename(sctx, valid_path,
2781 cur->full_path);
2782 if (ret < 0)
2783 goto out;
2784 ret = fs_path_copy(valid_path, cur->full_path);
2785 if (ret < 0)
2786 goto out;
2787 } else {
2788 ret = send_link(sctx, cur->full_path,
2789 valid_path);
2790 if (ret < 0)
2791 goto out;
2792 }
2793 }
2794 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2795 GFP_NOFS);
2796 if (ret < 0)
2797 goto out;
2798 }
2799
2800 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_deleted) {
2801 /*
2802 * Check if we can already rmdir the directory. If not,
2803 * orphanize it. For every dir item inside that gets deleted
2804 * later, we do this check again and rmdir it then if possible.
2805 * See the use of check_dirs for more details.
2806 */
2807 ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_ino);
2808 if (ret < 0)
2809 goto out;
2810 if (ret) {
2811 ret = send_rmdir(sctx, valid_path);
2812 if (ret < 0)
2813 goto out;
2814 } else if (!is_orphan) {
2815 ret = orphanize_inode(sctx, sctx->cur_ino,
2816 sctx->cur_inode_gen, valid_path);
2817 if (ret < 0)
2818 goto out;
2819 is_orphan = 1;
2820 }
2821
2822 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2823 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2824 GFP_NOFS);
2825 if (ret < 0)
2826 goto out;
2827 }
2828 } else if (!S_ISDIR(sctx->cur_inode_mode)) {
2829 /*
2830 * We have a non dir inode. Go through all deleted refs and
2831 * unlink them if they were not already overwritten by other
2832 * inodes.
2833 */
2834 list_for_each_entry(cur, &sctx->deleted_refs, list) {
2835 ret = did_overwrite_ref(sctx, cur->dir, cur->dir_gen,
2836 sctx->cur_ino, sctx->cur_inode_gen,
2837 cur->name, cur->name_len);
2838 if (ret < 0)
2839 goto out;
2840 if (!ret) {
2841 /*
2842 * In case the inode was moved to a directory
2843 * that was not created yet (see
2844 * __record_new_ref), we can not unlink the ref
2845 * as it will be needed later when the parent
2846 * directory is created, so that we can move in
2847 * the inode to the new dir.
2848 */
2849 if (!is_orphan &&
2850 sctx->cur_inode_first_ref_orphan) {
2851 ret = orphanize_inode(sctx,
2852 sctx->cur_ino,
2853 sctx->cur_inode_gen,
2854 cur->full_path);
2855 if (ret < 0)
2856 goto out;
2857 ret = gen_unique_name(sctx,
2858 sctx->cur_ino,
2859 sctx->cur_inode_gen,
2860 valid_path);
2861 if (ret < 0)
2862 goto out;
2863 is_orphan = 1;
2864
2865 } else {
2866 ret = send_unlink(sctx, cur->full_path);
2867 if (ret < 0)
2868 goto out;
2869 }
2870 }
2871 ret = ulist_add(check_dirs, cur->dir, cur->dir_gen,
2872 GFP_NOFS);
2873 if (ret < 0)
2874 goto out;
2875 }
2876
2877 /*
2878 * If the inode is still orphan, unlink the orphan. This may
2879 * happen when a previous inode did overwrite the first ref
2880 * of this inode and no new refs were added for the current
2881 * inode.
2882 * We can however not delete the orphan in case the inode relies
2883 * in a directory that was not created yet (see
2884 * __record_new_ref)
2885 */
2886 if (is_orphan && !sctx->cur_inode_first_ref_orphan) {
2887 ret = send_unlink(sctx, valid_path);
2888 if (ret < 0)
2889 goto out;
2890 }
2891 }
2892
2893 /*
2894 * We did collect all parent dirs where cur_inode was once located. We
2895 * now go through all these dirs and check if they are pending for
2896 * deletion and if it's finally possible to perform the rmdir now.
2897 * We also update the inode stats of the parent dirs here.
2898 */
2899 ULIST_ITER_INIT(&uit);
2900 while ((un = ulist_next(check_dirs, &uit))) {
2901 if (un->val > sctx->cur_ino)
2902 continue;
2903
2904 ret = get_cur_inode_state(sctx, un->val, un->aux);
2905 if (ret < 0)
2906 goto out;
2907
2908 if (ret == inode_state_did_create ||
2909 ret == inode_state_no_change) {
2910 /* TODO delayed utimes */
2911 ret = send_utimes(sctx, un->val, un->aux);
2912 if (ret < 0)
2913 goto out;
2914 } else if (ret == inode_state_did_delete) {
2915 ret = can_rmdir(sctx, un->val, sctx->cur_ino);
2916 if (ret < 0)
2917 goto out;
2918 if (ret) {
2919 ret = get_cur_path(sctx, un->val, un->aux,
2920 valid_path);
2921 if (ret < 0)
2922 goto out;
2923 ret = send_rmdir(sctx, valid_path);
2924 if (ret < 0)
2925 goto out;
2926 }
2927 }
2928 }
2929
2930 /*
2931 * Current inode is now at it's new position, so we must increase
2932 * send_progress
2933 */
2934 sctx->send_progress = sctx->cur_ino + 1;
2935
2936 /*
2937 * We may have a directory here that has pending refs which could not
2938 * be created before (because the dir did not exist before, see
2939 * __record_new_ref). finish_outoforder_dir will link/move the pending
2940 * refs.
2941 */
2942 if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_new) {
2943 ret = finish_outoforder_dir(sctx, sctx->cur_ino,
2944 sctx->cur_inode_gen);
2945 if (ret < 0)
2946 goto out;
2947 }
2948
2949 ret = 0;
2950
2951out:
2952 free_recorded_refs(sctx);
2953 ulist_free(check_dirs);
2954 fs_path_free(sctx, valid_path);
2955 return ret;
2956}
2957
2958static int __record_new_ref(int num, u64 dir, int index,
2959 struct fs_path *name,
2960 void *ctx)
2961{
2962 int ret = 0;
2963 struct send_ctx *sctx = ctx;
2964 struct fs_path *p;
2965 u64 gen;
2966
2967 p = fs_path_alloc(sctx);
2968 if (!p)
2969 return -ENOMEM;
2970
2971 ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, NULL,
2972 NULL);
2973 if (ret < 0)
2974 goto out;
2975
2976 /*
2977 * The parent may be non-existent at this point in time. This happens
2978 * if the ino of the parent dir is higher then the current ino. In this
2979 * case, we can not process this ref until the parent dir is finally
2980 * created. If we reach the parent dir later, process_recorded_refs
2981 * will go through all dir items and process the refs that could not be
2982 * processed before. In case this is the first ref, we set
2983 * cur_inode_first_ref_orphan to 1 to inform process_recorded_refs to
2984 * keep an orphan of the inode so that it later can be used for
2985 * link/move
2986 */
2987 ret = is_inode_existent(sctx, dir, gen);
2988 if (ret < 0)
2989 goto out;
2990 if (!ret) {
2991 ret = is_first_ref(sctx, sctx->send_root, sctx->cur_ino, dir,
2992 name->start, fs_path_len(name));
2993 if (ret < 0)
2994 goto out;
2995 if (ret)
2996 sctx->cur_inode_first_ref_orphan = 1;
2997 ret = 0;
2998 goto out;
2999 }
3000
3001 ret = get_cur_path(sctx, dir, gen, p);
3002 if (ret < 0)
3003 goto out;
3004 ret = fs_path_add_path(p, name);
3005 if (ret < 0)
3006 goto out;
3007
3008 ret = record_ref(&sctx->new_refs, dir, gen, p);
3009
3010out:
3011 if (ret)
3012 fs_path_free(sctx, p);
3013 return ret;
3014}
3015
3016static int __record_deleted_ref(int num, u64 dir, int index,
3017 struct fs_path *name,
3018 void *ctx)
3019{
3020 int ret = 0;
3021 struct send_ctx *sctx = ctx;
3022 struct fs_path *p;
3023 u64 gen;
3024
3025 p = fs_path_alloc(sctx);
3026 if (!p)
3027 return -ENOMEM;
3028
3029 ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, NULL,
3030 NULL);
3031 if (ret < 0)
3032 goto out;
3033
3034 ret = get_cur_path(sctx, dir, gen, p);
3035 if (ret < 0)
3036 goto out;
3037 ret = fs_path_add_path(p, name);
3038 if (ret < 0)
3039 goto out;
3040
3041 ret = record_ref(&sctx->deleted_refs, dir, gen, p);
3042
3043out:
3044 if (ret)
3045 fs_path_free(sctx, p);
3046 return ret;
3047}
3048
3049static int record_new_ref(struct send_ctx *sctx)
3050{
3051 int ret;
3052
3053 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3054 sctx->cmp_key, 0, __record_new_ref, sctx);
3055 if (ret < 0)
3056 goto out;
3057 ret = 0;
3058
3059out:
3060 return ret;
3061}
3062
3063static int record_deleted_ref(struct send_ctx *sctx)
3064{
3065 int ret;
3066
3067 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3068 sctx->cmp_key, 0, __record_deleted_ref, sctx);
3069 if (ret < 0)
3070 goto out;
3071 ret = 0;
3072
3073out:
3074 return ret;
3075}
3076
3077struct find_ref_ctx {
3078 u64 dir;
3079 struct fs_path *name;
3080 int found_idx;
3081};
3082
3083static int __find_iref(int num, u64 dir, int index,
3084 struct fs_path *name,
3085 void *ctx_)
3086{
3087 struct find_ref_ctx *ctx = ctx_;
3088
3089 if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) &&
3090 strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) {
3091 ctx->found_idx = num;
3092 return 1;
3093 }
3094 return 0;
3095}
3096
3097static int find_iref(struct send_ctx *sctx,
3098 struct btrfs_root *root,
3099 struct btrfs_path *path,
3100 struct btrfs_key *key,
3101 u64 dir, struct fs_path *name)
3102{
3103 int ret;
3104 struct find_ref_ctx ctx;
3105
3106 ctx.dir = dir;
3107 ctx.name = name;
3108 ctx.found_idx = -1;
3109
3110 ret = iterate_inode_ref(sctx, root, path, key, 0, __find_iref, &ctx);
3111 if (ret < 0)
3112 return ret;
3113
3114 if (ctx.found_idx == -1)
3115 return -ENOENT;
3116
3117 return ctx.found_idx;
3118}
3119
3120static int __record_changed_new_ref(int num, u64 dir, int index,
3121 struct fs_path *name,
3122 void *ctx)
3123{
3124 int ret;
3125 struct send_ctx *sctx = ctx;
3126
3127 ret = find_iref(sctx, sctx->parent_root, sctx->right_path,
3128 sctx->cmp_key, dir, name);
3129 if (ret == -ENOENT)
3130 ret = __record_new_ref(num, dir, index, name, sctx);
3131 else if (ret > 0)
3132 ret = 0;
3133
3134 return ret;
3135}
3136
3137static int __record_changed_deleted_ref(int num, u64 dir, int index,
3138 struct fs_path *name,
3139 void *ctx)
3140{
3141 int ret;
3142 struct send_ctx *sctx = ctx;
3143
3144 ret = find_iref(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3145 dir, name);
3146 if (ret == -ENOENT)
3147 ret = __record_deleted_ref(num, dir, index, name, sctx);
3148 else if (ret > 0)
3149 ret = 0;
3150
3151 return ret;
3152}
3153
3154static int record_changed_ref(struct send_ctx *sctx)
3155{
3156 int ret = 0;
3157
3158 ret = iterate_inode_ref(sctx, sctx->send_root, sctx->left_path,
3159 sctx->cmp_key, 0, __record_changed_new_ref, sctx);
3160 if (ret < 0)
3161 goto out;
3162 ret = iterate_inode_ref(sctx, sctx->parent_root, sctx->right_path,
3163 sctx->cmp_key, 0, __record_changed_deleted_ref, sctx);
3164 if (ret < 0)
3165 goto out;
3166 ret = 0;
3167
3168out:
3169 return ret;
3170}
3171
3172/*
3173 * Record and process all refs at once. Needed when an inode changes the
3174 * generation number, which means that it was deleted and recreated.
3175 */
3176static int process_all_refs(struct send_ctx *sctx,
3177 enum btrfs_compare_tree_result cmd)
3178{
3179 int ret;
3180 struct btrfs_root *root;
3181 struct btrfs_path *path;
3182 struct btrfs_key key;
3183 struct btrfs_key found_key;
3184 struct extent_buffer *eb;
3185 int slot;
3186 iterate_inode_ref_t cb;
3187
3188 path = alloc_path_for_send();
3189 if (!path)
3190 return -ENOMEM;
3191
3192 if (cmd == BTRFS_COMPARE_TREE_NEW) {
3193 root = sctx->send_root;
3194 cb = __record_new_ref;
3195 } else if (cmd == BTRFS_COMPARE_TREE_DELETED) {
3196 root = sctx->parent_root;
3197 cb = __record_deleted_ref;
3198 } else {
3199 BUG();
3200 }
3201
3202 key.objectid = sctx->cmp_key->objectid;
3203 key.type = BTRFS_INODE_REF_KEY;
3204 key.offset = 0;
3205 while (1) {
3206 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3207 if (ret < 0) {
3208 btrfs_release_path(path);
3209 goto out;
3210 }
3211 if (ret) {
3212 btrfs_release_path(path);
3213 break;
3214 }
3215
3216 eb = path->nodes[0];
3217 slot = path->slots[0];
3218 btrfs_item_key_to_cpu(eb, &found_key, slot);
3219
3220 if (found_key.objectid != key.objectid ||
3221 found_key.type != key.type) {
3222 btrfs_release_path(path);
3223 break;
3224 }
3225
3226 ret = iterate_inode_ref(sctx, sctx->parent_root, path,
3227 &found_key, 0, cb, sctx);
3228 btrfs_release_path(path);
3229 if (ret < 0)
3230 goto out;
3231
3232 key.offset = found_key.offset + 1;
3233 }
3234
3235 ret = process_recorded_refs(sctx);
3236
3237out:
3238 btrfs_free_path(path);
3239 return ret;
3240}
3241
3242static int send_set_xattr(struct send_ctx *sctx,
3243 struct fs_path *path,
3244 const char *name, int name_len,
3245 const char *data, int data_len)
3246{
3247 int ret = 0;
3248
3249 ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR);
3250 if (ret < 0)
3251 goto out;
3252
3253 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3254 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3255 TLV_PUT(sctx, BTRFS_SEND_A_XATTR_DATA, data, data_len);
3256
3257 ret = send_cmd(sctx);
3258
3259tlv_put_failure:
3260out:
3261 return ret;
3262}
3263
3264static int send_remove_xattr(struct send_ctx *sctx,
3265 struct fs_path *path,
3266 const char *name, int name_len)
3267{
3268 int ret = 0;
3269
3270 ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR);
3271 if (ret < 0)
3272 goto out;
3273
3274 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path);
3275 TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len);
3276
3277 ret = send_cmd(sctx);
3278
3279tlv_put_failure:
3280out:
3281 return ret;
3282}
3283
3284static int __process_new_xattr(int num, struct btrfs_key *di_key,
3285 const char *name, int name_len,
3286 const char *data, int data_len,
3287 u8 type, void *ctx)
3288{
3289 int ret;
3290 struct send_ctx *sctx = ctx;
3291 struct fs_path *p;
3292 posix_acl_xattr_header dummy_acl;
3293
3294 p = fs_path_alloc(sctx);
3295 if (!p)
3296 return -ENOMEM;
3297
3298 /*
3299 * This hack is needed because empty acl's are stored as zero byte
3300 * data in xattrs. Problem with that is, that receiving these zero byte
3301 * acl's will fail later. To fix this, we send a dummy acl list that
3302 * only contains the version number and no entries.
3303 */
3304 if (!strncmp(name, XATTR_NAME_POSIX_ACL_ACCESS, name_len) ||
3305 !strncmp(name, XATTR_NAME_POSIX_ACL_DEFAULT, name_len)) {
3306 if (data_len == 0) {
3307 dummy_acl.a_version =
3308 cpu_to_le32(POSIX_ACL_XATTR_VERSION);
3309 data = (char *)&dummy_acl;
3310 data_len = sizeof(dummy_acl);
3311 }
3312 }
3313
3314 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3315 if (ret < 0)
3316 goto out;
3317
3318 ret = send_set_xattr(sctx, p, name, name_len, data, data_len);
3319
3320out:
3321 fs_path_free(sctx, p);
3322 return ret;
3323}
3324
3325static int __process_deleted_xattr(int num, struct btrfs_key *di_key,
3326 const char *name, int name_len,
3327 const char *data, int data_len,
3328 u8 type, void *ctx)
3329{
3330 int ret;
3331 struct send_ctx *sctx = ctx;
3332 struct fs_path *p;
3333
3334 p = fs_path_alloc(sctx);
3335 if (!p)
3336 return -ENOMEM;
3337
3338 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3339 if (ret < 0)
3340 goto out;
3341
3342 ret = send_remove_xattr(sctx, p, name, name_len);
3343
3344out:
3345 fs_path_free(sctx, p);
3346 return ret;
3347}
3348
3349static int process_new_xattr(struct send_ctx *sctx)
3350{
3351 int ret = 0;
3352
3353 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3354 sctx->cmp_key, __process_new_xattr, sctx);
3355
3356 return ret;
3357}
3358
3359static int process_deleted_xattr(struct send_ctx *sctx)
3360{
3361 int ret;
3362
3363 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3364 sctx->cmp_key, __process_deleted_xattr, sctx);
3365
3366 return ret;
3367}
3368
3369struct find_xattr_ctx {
3370 const char *name;
3371 int name_len;
3372 int found_idx;
3373 char *found_data;
3374 int found_data_len;
3375};
3376
3377static int __find_xattr(int num, struct btrfs_key *di_key,
3378 const char *name, int name_len,
3379 const char *data, int data_len,
3380 u8 type, void *vctx)
3381{
3382 struct find_xattr_ctx *ctx = vctx;
3383
3384 if (name_len == ctx->name_len &&
3385 strncmp(name, ctx->name, name_len) == 0) {
3386 ctx->found_idx = num;
3387 ctx->found_data_len = data_len;
3388 ctx->found_data = kmalloc(data_len, GFP_NOFS);
3389 if (!ctx->found_data)
3390 return -ENOMEM;
3391 memcpy(ctx->found_data, data, data_len);
3392 return 1;
3393 }
3394 return 0;
3395}
3396
3397static int find_xattr(struct send_ctx *sctx,
3398 struct btrfs_root *root,
3399 struct btrfs_path *path,
3400 struct btrfs_key *key,
3401 const char *name, int name_len,
3402 char **data, int *data_len)
3403{
3404 int ret;
3405 struct find_xattr_ctx ctx;
3406
3407 ctx.name = name;
3408 ctx.name_len = name_len;
3409 ctx.found_idx = -1;
3410 ctx.found_data = NULL;
3411 ctx.found_data_len = 0;
3412
3413 ret = iterate_dir_item(sctx, root, path, key, __find_xattr, &ctx);
3414 if (ret < 0)
3415 return ret;
3416
3417 if (ctx.found_idx == -1)
3418 return -ENOENT;
3419 if (data) {
3420 *data = ctx.found_data;
3421 *data_len = ctx.found_data_len;
3422 } else {
3423 kfree(ctx.found_data);
3424 }
3425 return ctx.found_idx;
3426}
3427
3428
3429static int __process_changed_new_xattr(int num, struct btrfs_key *di_key,
3430 const char *name, int name_len,
3431 const char *data, int data_len,
3432 u8 type, void *ctx)
3433{
3434 int ret;
3435 struct send_ctx *sctx = ctx;
3436 char *found_data = NULL;
3437 int found_data_len = 0;
3438 struct fs_path *p = NULL;
3439
3440 ret = find_xattr(sctx, sctx->parent_root, sctx->right_path,
3441 sctx->cmp_key, name, name_len, &found_data,
3442 &found_data_len);
3443 if (ret == -ENOENT) {
3444 ret = __process_new_xattr(num, di_key, name, name_len, data,
3445 data_len, type, ctx);
3446 } else if (ret >= 0) {
3447 if (data_len != found_data_len ||
3448 memcmp(data, found_data, data_len)) {
3449 ret = __process_new_xattr(num, di_key, name, name_len,
3450 data, data_len, type, ctx);
3451 } else {
3452 ret = 0;
3453 }
3454 }
3455
3456 kfree(found_data);
3457 fs_path_free(sctx, p);
3458 return ret;
3459}
3460
3461static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key,
3462 const char *name, int name_len,
3463 const char *data, int data_len,
3464 u8 type, void *ctx)
3465{
3466 int ret;
3467 struct send_ctx *sctx = ctx;
3468
3469 ret = find_xattr(sctx, sctx->send_root, sctx->left_path, sctx->cmp_key,
3470 name, name_len, NULL, NULL);
3471 if (ret == -ENOENT)
3472 ret = __process_deleted_xattr(num, di_key, name, name_len, data,
3473 data_len, type, ctx);
3474 else if (ret >= 0)
3475 ret = 0;
3476
3477 return ret;
3478}
3479
3480static int process_changed_xattr(struct send_ctx *sctx)
3481{
3482 int ret = 0;
3483
3484 ret = iterate_dir_item(sctx, sctx->send_root, sctx->left_path,
3485 sctx->cmp_key, __process_changed_new_xattr, sctx);
3486 if (ret < 0)
3487 goto out;
3488 ret = iterate_dir_item(sctx, sctx->parent_root, sctx->right_path,
3489 sctx->cmp_key, __process_changed_deleted_xattr, sctx);
3490
3491out:
3492 return ret;
3493}
3494
3495static int process_all_new_xattrs(struct send_ctx *sctx)
3496{
3497 int ret;
3498 struct btrfs_root *root;
3499 struct btrfs_path *path;
3500 struct btrfs_key key;
3501 struct btrfs_key found_key;
3502 struct extent_buffer *eb;
3503 int slot;
3504
3505 path = alloc_path_for_send();
3506 if (!path)
3507 return -ENOMEM;
3508
3509 root = sctx->send_root;
3510
3511 key.objectid = sctx->cmp_key->objectid;
3512 key.type = BTRFS_XATTR_ITEM_KEY;
3513 key.offset = 0;
3514 while (1) {
3515 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3516 if (ret < 0)
3517 goto out;
3518 if (ret) {
3519 ret = 0;
3520 goto out;
3521 }
3522
3523 eb = path->nodes[0];
3524 slot = path->slots[0];
3525 btrfs_item_key_to_cpu(eb, &found_key, slot);
3526
3527 if (found_key.objectid != key.objectid ||
3528 found_key.type != key.type) {
3529 ret = 0;
3530 goto out;
3531 }
3532
3533 ret = iterate_dir_item(sctx, root, path, &found_key,
3534 __process_new_xattr, sctx);
3535 if (ret < 0)
3536 goto out;
3537
3538 btrfs_release_path(path);
3539 key.offset = found_key.offset + 1;
3540 }
3541
3542out:
3543 btrfs_free_path(path);
3544 return ret;
3545}
3546
3547/*
3548 * Read some bytes from the current inode/file and send a write command to
3549 * user space.
3550 */
3551static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
3552{
3553 int ret = 0;
3554 struct fs_path *p;
3555 loff_t pos = offset;
3556 int readed;
3557 mm_segment_t old_fs;
3558
3559 p = fs_path_alloc(sctx);
3560 if (!p)
3561 return -ENOMEM;
3562
3563 /*
3564 * vfs normally only accepts user space buffers for security reasons.
3565 * we only read from the file and also only provide the read_buf buffer
3566 * to vfs. As this buffer does not come from a user space call, it's
3567 * ok to temporary allow kernel space buffers.
3568 */
3569 old_fs = get_fs();
3570 set_fs(KERNEL_DS);
3571
3572verbose_printk("btrfs: send_write offset=%llu, len=%d\n", offset, len);
3573
3574 ret = open_cur_inode_file(sctx);
3575 if (ret < 0)
3576 goto out;
3577
3578 ret = vfs_read(sctx->cur_inode_filp, sctx->read_buf, len, &pos);
3579 if (ret < 0)
3580 goto out;
3581 readed = ret;
3582 if (!readed)
3583 goto out;
3584
3585 ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
3586 if (ret < 0)
3587 goto out;
3588
3589 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3590 if (ret < 0)
3591 goto out;
3592
3593 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3594 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3595 TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, readed);
3596
3597 ret = send_cmd(sctx);
3598
3599tlv_put_failure:
3600out:
3601 fs_path_free(sctx, p);
3602 set_fs(old_fs);
3603 if (ret < 0)
3604 return ret;
3605 return readed;
3606}
3607
3608/*
3609 * Send a clone command to user space.
3610 */
3611static int send_clone(struct send_ctx *sctx,
3612 u64 offset, u32 len,
3613 struct clone_root *clone_root)
3614{
3615 int ret = 0;
3616 struct btrfs_root *clone_root2 = clone_root->root;
3617 struct fs_path *p;
3618 u64 gen;
3619
3620verbose_printk("btrfs: send_clone offset=%llu, len=%d, clone_root=%llu, "
3621 "clone_inode=%llu, clone_offset=%llu\n", offset, len,
3622 clone_root->root->objectid, clone_root->ino,
3623 clone_root->offset);
3624
3625 p = fs_path_alloc(sctx);
3626 if (!p)
3627 return -ENOMEM;
3628
3629 ret = begin_cmd(sctx, BTRFS_SEND_C_CLONE);
3630 if (ret < 0)
3631 goto out;
3632
3633 ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
3634 if (ret < 0)
3635 goto out;
3636
3637 TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
3638 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len);
3639 TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
3640
3641 if (clone_root2 == sctx->send_root) {
3642 ret = get_inode_info(sctx->send_root, clone_root->ino, NULL,
3643 &gen, NULL, NULL, NULL);
3644 if (ret < 0)
3645 goto out;
3646 ret = get_cur_path(sctx, clone_root->ino, gen, p);
3647 } else {
3648 ret = get_inode_path(sctx, clone_root2, clone_root->ino, p);
3649 }
3650 if (ret < 0)
3651 goto out;
3652
3653 TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID,
3654 clone_root2->root_item.uuid);
3655 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID,
3656 clone_root2->root_item.ctransid);
3657 TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p);
3658 TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET,
3659 clone_root->offset);
3660
3661 ret = send_cmd(sctx);
3662
3663tlv_put_failure:
3664out:
3665 fs_path_free(sctx, p);
3666 return ret;
3667}
3668
3669static int send_write_or_clone(struct send_ctx *sctx,
3670 struct btrfs_path *path,
3671 struct btrfs_key *key,
3672 struct clone_root *clone_root)
3673{
3674 int ret = 0;
3675 struct btrfs_file_extent_item *ei;
3676 u64 offset = key->offset;
3677 u64 pos = 0;
3678 u64 len;
3679 u32 l;
3680 u8 type;
3681
3682 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3683 struct btrfs_file_extent_item);
3684 type = btrfs_file_extent_type(path->nodes[0], ei);
3685 if (type == BTRFS_FILE_EXTENT_INLINE)
3686 len = btrfs_file_extent_inline_len(path->nodes[0], ei);
3687 else
3688 len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
3689
3690 if (offset + len > sctx->cur_inode_size)
3691 len = sctx->cur_inode_size - offset;
3692 if (len == 0) {
3693 ret = 0;
3694 goto out;
3695 }
3696
3697 if (!clone_root) {
3698 while (pos < len) {
3699 l = len - pos;
3700 if (l > BTRFS_SEND_READ_SIZE)
3701 l = BTRFS_SEND_READ_SIZE;
3702 ret = send_write(sctx, pos + offset, l);
3703 if (ret < 0)
3704 goto out;
3705 if (!ret)
3706 break;
3707 pos += ret;
3708 }
3709 ret = 0;
3710 } else {
3711 ret = send_clone(sctx, offset, len, clone_root);
3712 }
3713
3714out:
3715 return ret;
3716}
3717
3718static int is_extent_unchanged(struct send_ctx *sctx,
3719 struct btrfs_path *left_path,
3720 struct btrfs_key *ekey)
3721{
3722 int ret = 0;
3723 struct btrfs_key key;
3724 struct btrfs_path *path = NULL;
3725 struct extent_buffer *eb;
3726 int slot;
3727 struct btrfs_key found_key;
3728 struct btrfs_file_extent_item *ei;
3729 u64 left_disknr;
3730 u64 right_disknr;
3731 u64 left_offset;
3732 u64 right_offset;
3733 u64 left_offset_fixed;
3734 u64 left_len;
3735 u64 right_len;
3736 u8 left_type;
3737 u8 right_type;
3738
3739 path = alloc_path_for_send();
3740 if (!path)
3741 return -ENOMEM;
3742
3743 eb = left_path->nodes[0];
3744 slot = left_path->slots[0];
3745
3746 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3747 left_type = btrfs_file_extent_type(eb, ei);
3748 left_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3749 left_len = btrfs_file_extent_num_bytes(eb, ei);
3750 left_offset = btrfs_file_extent_offset(eb, ei);
3751
3752 if (left_type != BTRFS_FILE_EXTENT_REG) {
3753 ret = 0;
3754 goto out;
3755 }
3756
3757 /*
3758 * Following comments will refer to these graphics. L is the left
3759 * extents which we are checking at the moment. 1-8 are the right
3760 * extents that we iterate.
3761 *
3762 * |-----L-----|
3763 * |-1-|-2a-|-3-|-4-|-5-|-6-|
3764 *
3765 * |-----L-----|
3766 * |--1--|-2b-|...(same as above)
3767 *
3768 * Alternative situation. Happens on files where extents got split.
3769 * |-----L-----|
3770 * |-----------7-----------|-6-|
3771 *
3772 * Alternative situation. Happens on files which got larger.
3773 * |-----L-----|
3774 * |-8-|
3775 * Nothing follows after 8.
3776 */
3777
3778 key.objectid = ekey->objectid;
3779 key.type = BTRFS_EXTENT_DATA_KEY;
3780 key.offset = ekey->offset;
3781 ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0);
3782 if (ret < 0)
3783 goto out;
3784 if (ret) {
3785 ret = 0;
3786 goto out;
3787 }
3788
3789 /*
3790 * Handle special case where the right side has no extents at all.
3791 */
3792 eb = path->nodes[0];
3793 slot = path->slots[0];
3794 btrfs_item_key_to_cpu(eb, &found_key, slot);
3795 if (found_key.objectid != key.objectid ||
3796 found_key.type != key.type) {
3797 ret = 0;
3798 goto out;
3799 }
3800
3801 /*
3802 * We're now on 2a, 2b or 7.
3803 */
3804 key = found_key;
3805 while (key.offset < ekey->offset + left_len) {
3806 ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
3807 right_type = btrfs_file_extent_type(eb, ei);
3808 right_disknr = btrfs_file_extent_disk_bytenr(eb, ei);
3809 right_len = btrfs_file_extent_num_bytes(eb, ei);
3810 right_offset = btrfs_file_extent_offset(eb, ei);
3811
3812 if (right_type != BTRFS_FILE_EXTENT_REG) {
3813 ret = 0;
3814 goto out;
3815 }
3816
3817 /*
3818 * Are we at extent 8? If yes, we know the extent is changed.
3819 * This may only happen on the first iteration.
3820 */
3821 if (found_key.offset + right_len < ekey->offset) {
3822 ret = 0;
3823 goto out;
3824 }
3825
3826 left_offset_fixed = left_offset;
3827 if (key.offset < ekey->offset) {
3828 /* Fix the right offset for 2a and 7. */
3829 right_offset += ekey->offset - key.offset;
3830 } else {
3831 /* Fix the left offset for all behind 2a and 2b */
3832 left_offset_fixed += key.offset - ekey->offset;
3833 }
3834
3835 /*
3836 * Check if we have the same extent.
3837 */
3838 if (left_disknr + left_offset_fixed !=
3839 right_disknr + right_offset) {
3840 ret = 0;
3841 goto out;
3842 }
3843
3844 /*
3845 * Go to the next extent.
3846 */
3847 ret = btrfs_next_item(sctx->parent_root, path);
3848 if (ret < 0)
3849 goto out;
3850 if (!ret) {
3851 eb = path->nodes[0];
3852 slot = path->slots[0];
3853 btrfs_item_key_to_cpu(eb, &found_key, slot);
3854 }
3855 if (ret || found_key.objectid != key.objectid ||
3856 found_key.type != key.type) {
3857 key.offset += right_len;
3858 break;
3859 } else {
3860 if (found_key.offset != key.offset + right_len) {
3861 /* Should really not happen */
3862 ret = -EIO;
3863 goto out;
3864 }
3865 }
3866 key = found_key;
3867 }
3868
3869 /*
3870 * We're now behind the left extent (treat as unchanged) or at the end
3871 * of the right side (treat as changed).
3872 */
3873 if (key.offset >= ekey->offset + left_len)
3874 ret = 1;
3875 else
3876 ret = 0;
3877
3878
3879out:
3880 btrfs_free_path(path);
3881 return ret;
3882}
3883
3884static int process_extent(struct send_ctx *sctx,
3885 struct btrfs_path *path,
3886 struct btrfs_key *key)
3887{
3888 int ret = 0;
3889 struct clone_root *found_clone = NULL;
3890
3891 if (S_ISLNK(sctx->cur_inode_mode))
3892 return 0;
3893
3894 if (sctx->parent_root && !sctx->cur_inode_new) {
3895 ret = is_extent_unchanged(sctx, path, key);
3896 if (ret < 0)
3897 goto out;
3898 if (ret) {
3899 ret = 0;
3900 goto out;
3901 }
3902 }
3903
3904 ret = find_extent_clone(sctx, path, key->objectid, key->offset,
3905 sctx->cur_inode_size, &found_clone);
3906 if (ret != -ENOENT && ret < 0)
3907 goto out;
3908
3909 ret = send_write_or_clone(sctx, path, key, found_clone);
3910
3911out:
3912 return ret;
3913}
3914
3915static int process_all_extents(struct send_ctx *sctx)
3916{
3917 int ret;
3918 struct btrfs_root *root;
3919 struct btrfs_path *path;
3920 struct btrfs_key key;
3921 struct btrfs_key found_key;
3922 struct extent_buffer *eb;
3923 int slot;
3924
3925 root = sctx->send_root;
3926 path = alloc_path_for_send();
3927 if (!path)
3928 return -ENOMEM;
3929
3930 key.objectid = sctx->cmp_key->objectid;
3931 key.type = BTRFS_EXTENT_DATA_KEY;
3932 key.offset = 0;
3933 while (1) {
3934 ret = btrfs_search_slot_for_read(root, &key, path, 1, 0);
3935 if (ret < 0)
3936 goto out;
3937 if (ret) {
3938 ret = 0;
3939 goto out;
3940 }
3941
3942 eb = path->nodes[0];
3943 slot = path->slots[0];
3944 btrfs_item_key_to_cpu(eb, &found_key, slot);
3945
3946 if (found_key.objectid != key.objectid ||
3947 found_key.type != key.type) {
3948 ret = 0;
3949 goto out;
3950 }
3951
3952 ret = process_extent(sctx, path, &found_key);
3953 if (ret < 0)
3954 goto out;
3955
3956 btrfs_release_path(path);
3957 key.offset = found_key.offset + 1;
3958 }
3959
3960out:
3961 btrfs_free_path(path);
3962 return ret;
3963}
3964
3965static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end)
3966{
3967 int ret = 0;
3968
3969 if (sctx->cur_ino == 0)
3970 goto out;
3971 if (!at_end && sctx->cur_ino == sctx->cmp_key->objectid &&
3972 sctx->cmp_key->type <= BTRFS_INODE_REF_KEY)
3973 goto out;
3974 if (list_empty(&sctx->new_refs) && list_empty(&sctx->deleted_refs))
3975 goto out;
3976
3977 ret = process_recorded_refs(sctx);
3978
3979out:
3980 return ret;
3981}
3982
3983static int finish_inode_if_needed(struct send_ctx *sctx, int at_end)
3984{
3985 int ret = 0;
3986 u64 left_mode;
3987 u64 left_uid;
3988 u64 left_gid;
3989 u64 right_mode;
3990 u64 right_uid;
3991 u64 right_gid;
3992 int need_chmod = 0;
3993 int need_chown = 0;
3994
3995 ret = process_recorded_refs_if_needed(sctx, at_end);
3996 if (ret < 0)
3997 goto out;
3998
3999 if (sctx->cur_ino == 0 || sctx->cur_inode_deleted)
4000 goto out;
4001 if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino)
4002 goto out;
4003
4004 ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL,
4005 &left_mode, &left_uid, &left_gid);
4006 if (ret < 0)
4007 goto out;
4008
4009 if (!S_ISLNK(sctx->cur_inode_mode)) {
4010 if (!sctx->parent_root || sctx->cur_inode_new) {
4011 need_chmod = 1;
4012 need_chown = 1;
4013 } else {
4014 ret = get_inode_info(sctx->parent_root, sctx->cur_ino,
4015 NULL, NULL, &right_mode, &right_uid,
4016 &right_gid);
4017 if (ret < 0)
4018 goto out;
4019
4020 if (left_uid != right_uid || left_gid != right_gid)
4021 need_chown = 1;
4022 if (left_mode != right_mode)
4023 need_chmod = 1;
4024 }
4025 }
4026
4027 if (S_ISREG(sctx->cur_inode_mode)) {
4028 ret = send_truncate(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4029 sctx->cur_inode_size);
4030 if (ret < 0)
4031 goto out;
4032 }
4033
4034 if (need_chown) {
4035 ret = send_chown(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4036 left_uid, left_gid);
4037 if (ret < 0)
4038 goto out;
4039 }
4040 if (need_chmod) {
4041 ret = send_chmod(sctx, sctx->cur_ino, sctx->cur_inode_gen,
4042 left_mode);
4043 if (ret < 0)
4044 goto out;
4045 }
4046
4047 /*
4048 * Need to send that every time, no matter if it actually changed
4049 * between the two trees as we have done changes to the inode before.
4050 */
4051 ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen);
4052 if (ret < 0)
4053 goto out;
4054
4055out:
4056 return ret;
4057}
4058
4059static int changed_inode(struct send_ctx *sctx,
4060 enum btrfs_compare_tree_result result)
4061{
4062 int ret = 0;
4063 struct btrfs_key *key = sctx->cmp_key;
4064 struct btrfs_inode_item *left_ii = NULL;
4065 struct btrfs_inode_item *right_ii = NULL;
4066 u64 left_gen = 0;
4067 u64 right_gen = 0;
4068
4069 ret = close_cur_inode_file(sctx);
4070 if (ret < 0)
4071 goto out;
4072
4073 sctx->cur_ino = key->objectid;
4074 sctx->cur_inode_new_gen = 0;
4075 sctx->cur_inode_first_ref_orphan = 0;
4076 sctx->send_progress = sctx->cur_ino;
4077
4078 if (result == BTRFS_COMPARE_TREE_NEW ||
4079 result == BTRFS_COMPARE_TREE_CHANGED) {
4080 left_ii = btrfs_item_ptr(sctx->left_path->nodes[0],
4081 sctx->left_path->slots[0],
4082 struct btrfs_inode_item);
4083 left_gen = btrfs_inode_generation(sctx->left_path->nodes[0],
4084 left_ii);
4085 } else {
4086 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4087 sctx->right_path->slots[0],
4088 struct btrfs_inode_item);
4089 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4090 right_ii);
4091 }
4092 if (result == BTRFS_COMPARE_TREE_CHANGED) {
4093 right_ii = btrfs_item_ptr(sctx->right_path->nodes[0],
4094 sctx->right_path->slots[0],
4095 struct btrfs_inode_item);
4096
4097 right_gen = btrfs_inode_generation(sctx->right_path->nodes[0],
4098 right_ii);
4099 if (left_gen != right_gen)
4100 sctx->cur_inode_new_gen = 1;
4101 }
4102
4103 if (result == BTRFS_COMPARE_TREE_NEW) {
4104 sctx->cur_inode_gen = left_gen;
4105 sctx->cur_inode_new = 1;
4106 sctx->cur_inode_deleted = 0;
4107 sctx->cur_inode_size = btrfs_inode_size(
4108 sctx->left_path->nodes[0], left_ii);
4109 sctx->cur_inode_mode = btrfs_inode_mode(
4110 sctx->left_path->nodes[0], left_ii);
4111 if (sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID)
4112 ret = send_create_inode(sctx, sctx->left_path,
4113 sctx->cmp_key);
4114 } else if (result == BTRFS_COMPARE_TREE_DELETED) {
4115 sctx->cur_inode_gen = right_gen;
4116 sctx->cur_inode_new = 0;
4117 sctx->cur_inode_deleted = 1;
4118 sctx->cur_inode_size = btrfs_inode_size(
4119 sctx->right_path->nodes[0], right_ii);
4120 sctx->cur_inode_mode = btrfs_inode_mode(
4121 sctx->right_path->nodes[0], right_ii);
4122 } else if (result == BTRFS_COMPARE_TREE_CHANGED) {
4123 if (sctx->cur_inode_new_gen) {
4124 sctx->cur_inode_gen = right_gen;
4125 sctx->cur_inode_new = 0;
4126 sctx->cur_inode_deleted = 1;
4127 sctx->cur_inode_size = btrfs_inode_size(
4128 sctx->right_path->nodes[0], right_ii);
4129 sctx->cur_inode_mode = btrfs_inode_mode(
4130 sctx->right_path->nodes[0], right_ii);
4131 ret = process_all_refs(sctx,
4132 BTRFS_COMPARE_TREE_DELETED);
4133 if (ret < 0)
4134 goto out;
4135
4136 sctx->cur_inode_gen = left_gen;
4137 sctx->cur_inode_new = 1;
4138 sctx->cur_inode_deleted = 0;
4139 sctx->cur_inode_size = btrfs_inode_size(
4140 sctx->left_path->nodes[0], left_ii);
4141 sctx->cur_inode_mode = btrfs_inode_mode(
4142 sctx->left_path->nodes[0], left_ii);
4143 ret = send_create_inode(sctx, sctx->left_path,
4144 sctx->cmp_key);
4145 if (ret < 0)
4146 goto out;
4147
4148 ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW);
4149 if (ret < 0)
4150 goto out;
4151 ret = process_all_extents(sctx);
4152 if (ret < 0)
4153 goto out;
4154 ret = process_all_new_xattrs(sctx);
4155 if (ret < 0)
4156 goto out;
4157 } else {
4158 sctx->cur_inode_gen = left_gen;
4159 sctx->cur_inode_new = 0;
4160 sctx->cur_inode_new_gen = 0;
4161 sctx->cur_inode_deleted = 0;
4162 sctx->cur_inode_size = btrfs_inode_size(
4163 sctx->left_path->nodes[0], left_ii);
4164 sctx->cur_inode_mode = btrfs_inode_mode(
4165 sctx->left_path->nodes[0], left_ii);
4166 }
4167 }
4168
4169out:
4170 return ret;
4171}
4172
4173static int changed_ref(struct send_ctx *sctx,
4174 enum btrfs_compare_tree_result result)
4175{
4176 int ret = 0;
4177
4178 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4179
4180 if (!sctx->cur_inode_new_gen &&
4181 sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) {
4182 if (result == BTRFS_COMPARE_TREE_NEW)
4183 ret = record_new_ref(sctx);
4184 else if (result == BTRFS_COMPARE_TREE_DELETED)
4185 ret = record_deleted_ref(sctx);
4186 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4187 ret = record_changed_ref(sctx);
4188 }
4189
4190 return ret;
4191}
4192
4193static int changed_xattr(struct send_ctx *sctx,
4194 enum btrfs_compare_tree_result result)
4195{
4196 int ret = 0;
4197
4198 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4199
4200 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4201 if (result == BTRFS_COMPARE_TREE_NEW)
4202 ret = process_new_xattr(sctx);
4203 else if (result == BTRFS_COMPARE_TREE_DELETED)
4204 ret = process_deleted_xattr(sctx);
4205 else if (result == BTRFS_COMPARE_TREE_CHANGED)
4206 ret = process_changed_xattr(sctx);
4207 }
4208
4209 return ret;
4210}
4211
4212static int changed_extent(struct send_ctx *sctx,
4213 enum btrfs_compare_tree_result result)
4214{
4215 int ret = 0;
4216
4217 BUG_ON(sctx->cur_ino != sctx->cmp_key->objectid);
4218
4219 if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) {
4220 if (result != BTRFS_COMPARE_TREE_DELETED)
4221 ret = process_extent(sctx, sctx->left_path,
4222 sctx->cmp_key);
4223 }
4224
4225 return ret;
4226}
4227
4228
4229static int changed_cb(struct btrfs_root *left_root,
4230 struct btrfs_root *right_root,
4231 struct btrfs_path *left_path,
4232 struct btrfs_path *right_path,
4233 struct btrfs_key *key,
4234 enum btrfs_compare_tree_result result,
4235 void *ctx)
4236{
4237 int ret = 0;
4238 struct send_ctx *sctx = ctx;
4239
4240 sctx->left_path = left_path;
4241 sctx->right_path = right_path;
4242 sctx->cmp_key = key;
4243
4244 ret = finish_inode_if_needed(sctx, 0);
4245 if (ret < 0)
4246 goto out;
4247
4248 if (key->type == BTRFS_INODE_ITEM_KEY)
4249 ret = changed_inode(sctx, result);
4250 else if (key->type == BTRFS_INODE_REF_KEY)
4251 ret = changed_ref(sctx, result);
4252 else if (key->type == BTRFS_XATTR_ITEM_KEY)
4253 ret = changed_xattr(sctx, result);
4254 else if (key->type == BTRFS_EXTENT_DATA_KEY)
4255 ret = changed_extent(sctx, result);
4256
4257out:
4258 return ret;
4259}
4260
4261static int full_send_tree(struct send_ctx *sctx)
4262{
4263 int ret;
4264 struct btrfs_trans_handle *trans = NULL;
4265 struct btrfs_root *send_root = sctx->send_root;
4266 struct btrfs_key key;
4267 struct btrfs_key found_key;
4268 struct btrfs_path *path;
4269 struct extent_buffer *eb;
4270 int slot;
4271 u64 start_ctransid;
4272 u64 ctransid;
4273
4274 path = alloc_path_for_send();
4275 if (!path)
4276 return -ENOMEM;
4277
4278 spin_lock(&send_root->root_times_lock);
4279 start_ctransid = btrfs_root_ctransid(&send_root->root_item);
4280 spin_unlock(&send_root->root_times_lock);
4281
4282 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4283 key.type = BTRFS_INODE_ITEM_KEY;
4284 key.offset = 0;
4285
4286join_trans:
4287 /*
4288 * We need to make sure the transaction does not get committed
4289 * while we do anything on commit roots. Join a transaction to prevent
4290 * this.
4291 */
4292 trans = btrfs_join_transaction(send_root);
4293 if (IS_ERR(trans)) {
4294 ret = PTR_ERR(trans);
4295 trans = NULL;
4296 goto out;
4297 }
4298
4299 /*
4300 * Make sure the tree has not changed
4301 */
4302 spin_lock(&send_root->root_times_lock);
4303 ctransid = btrfs_root_ctransid(&send_root->root_item);
4304 spin_unlock(&send_root->root_times_lock);
4305
4306 if (ctransid != start_ctransid) {
4307 WARN(1, KERN_WARNING "btrfs: the root that you're trying to "
4308 "send was modified in between. This is "
4309 "probably a bug.\n");
4310 ret = -EIO;
4311 goto out;
4312 }
4313
4314 ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
4315 if (ret < 0)
4316 goto out;
4317 if (ret)
4318 goto out_finish;
4319
4320 while (1) {
4321 /*
4322 * When someone want to commit while we iterate, end the
4323 * joined transaction and rejoin.
4324 */
4325 if (btrfs_should_end_transaction(trans, send_root)) {
4326 ret = btrfs_end_transaction(trans, send_root);
4327 trans = NULL;
4328 if (ret < 0)
4329 goto out;
4330 btrfs_release_path(path);
4331 goto join_trans;
4332 }
4333
4334 eb = path->nodes[0];
4335 slot = path->slots[0];
4336 btrfs_item_key_to_cpu(eb, &found_key, slot);
4337
4338 ret = changed_cb(send_root, NULL, path, NULL,
4339 &found_key, BTRFS_COMPARE_TREE_NEW, sctx);
4340 if (ret < 0)
4341 goto out;
4342
4343 key.objectid = found_key.objectid;
4344 key.type = found_key.type;
4345 key.offset = found_key.offset + 1;
4346
4347 ret = btrfs_next_item(send_root, path);
4348 if (ret < 0)
4349 goto out;
4350 if (ret) {
4351 ret = 0;
4352 break;
4353 }
4354 }
4355
4356out_finish:
4357 ret = finish_inode_if_needed(sctx, 1);
4358
4359out:
4360 btrfs_free_path(path);
4361 if (trans) {
4362 if (!ret)
4363 ret = btrfs_end_transaction(trans, send_root);
4364 else
4365 btrfs_end_transaction(trans, send_root);
4366 }
4367 return ret;
4368}
4369
4370static int send_subvol(struct send_ctx *sctx)
4371{
4372 int ret;
4373
4374 ret = send_header(sctx);
4375 if (ret < 0)
4376 goto out;
4377
4378 ret = send_subvol_begin(sctx);
4379 if (ret < 0)
4380 goto out;
4381
4382 if (sctx->parent_root) {
4383 ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
4384 changed_cb, sctx);
4385 if (ret < 0)
4386 goto out;
4387 ret = finish_inode_if_needed(sctx, 1);
4388 if (ret < 0)
4389 goto out;
4390 } else {
4391 ret = full_send_tree(sctx);
4392 if (ret < 0)
4393 goto out;
4394 }
4395
4396out:
4397 if (!ret)
4398 ret = close_cur_inode_file(sctx);
4399 else
4400 close_cur_inode_file(sctx);
4401
4402 free_recorded_refs(sctx);
4403 return ret;
4404}
4405
4406long btrfs_ioctl_send(struct file *mnt_file, void __user *arg_)
4407{
4408 int ret = 0;
4409 struct btrfs_root *send_root;
4410 struct btrfs_root *clone_root;
4411 struct btrfs_fs_info *fs_info;
4412 struct btrfs_ioctl_send_args *arg = NULL;
4413 struct btrfs_key key;
4414 struct file *filp = NULL;
4415 struct send_ctx *sctx = NULL;
4416 u32 i;
4417 u64 *clone_sources_tmp = NULL;
4418
4419 if (!capable(CAP_SYS_ADMIN))
4420 return -EPERM;
4421
4422 send_root = BTRFS_I(fdentry(mnt_file)->d_inode)->root;
4423 fs_info = send_root->fs_info;
4424
4425 arg = memdup_user(arg_, sizeof(*arg));
4426 if (IS_ERR(arg)) {
4427 ret = PTR_ERR(arg);
4428 arg = NULL;
4429 goto out;
4430 }
4431
4432 if (!access_ok(VERIFY_READ, arg->clone_sources,
4433 sizeof(*arg->clone_sources *
4434 arg->clone_sources_count))) {
4435 ret = -EFAULT;
4436 goto out;
4437 }
4438
4439 sctx = kzalloc(sizeof(struct send_ctx), GFP_NOFS);
4440 if (!sctx) {
4441 ret = -ENOMEM;
4442 goto out;
4443 }
4444
4445 INIT_LIST_HEAD(&sctx->new_refs);
4446 INIT_LIST_HEAD(&sctx->deleted_refs);
4447 INIT_RADIX_TREE(&sctx->name_cache, GFP_NOFS);
4448 INIT_LIST_HEAD(&sctx->name_cache_list);
4449
4450 sctx->send_filp = fget(arg->send_fd);
4451 if (IS_ERR(sctx->send_filp)) {
4452 ret = PTR_ERR(sctx->send_filp);
4453 goto out;
4454 }
4455
4456 sctx->mnt = mnt_file->f_path.mnt;
4457
4458 sctx->send_root = send_root;
4459 sctx->clone_roots_cnt = arg->clone_sources_count;
4460
4461 sctx->send_max_size = BTRFS_SEND_BUF_SIZE;
4462 sctx->send_buf = vmalloc(sctx->send_max_size);
4463 if (!sctx->send_buf) {
4464 ret = -ENOMEM;
4465 goto out;
4466 }
4467
4468 sctx->read_buf = vmalloc(BTRFS_SEND_READ_SIZE);
4469 if (!sctx->read_buf) {
4470 ret = -ENOMEM;
4471 goto out;
4472 }
4473
4474 sctx->clone_roots = vzalloc(sizeof(struct clone_root) *
4475 (arg->clone_sources_count + 1));
4476 if (!sctx->clone_roots) {
4477 ret = -ENOMEM;
4478 goto out;
4479 }
4480
4481 if (arg->clone_sources_count) {
4482 clone_sources_tmp = vmalloc(arg->clone_sources_count *
4483 sizeof(*arg->clone_sources));
4484 if (!clone_sources_tmp) {
4485 ret = -ENOMEM;
4486 goto out;
4487 }
4488
4489 ret = copy_from_user(clone_sources_tmp, arg->clone_sources,
4490 arg->clone_sources_count *
4491 sizeof(*arg->clone_sources));
4492 if (ret) {
4493 ret = -EFAULT;
4494 goto out;
4495 }
4496
4497 for (i = 0; i < arg->clone_sources_count; i++) {
4498 key.objectid = clone_sources_tmp[i];
4499 key.type = BTRFS_ROOT_ITEM_KEY;
4500 key.offset = (u64)-1;
4501 clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
4502 if (!clone_root) {
4503 ret = -EINVAL;
4504 goto out;
4505 }
4506 if (IS_ERR(clone_root)) {
4507 ret = PTR_ERR(clone_root);
4508 goto out;
4509 }
4510 sctx->clone_roots[i].root = clone_root;
4511 }
4512 vfree(clone_sources_tmp);
4513 clone_sources_tmp = NULL;
4514 }
4515
4516 if (arg->parent_root) {
4517 key.objectid = arg->parent_root;
4518 key.type = BTRFS_ROOT_ITEM_KEY;
4519 key.offset = (u64)-1;
4520 sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
4521 if (!sctx->parent_root) {
4522 ret = -EINVAL;
4523 goto out;
4524 }
4525 }
4526
4527 /*
4528 * Clones from send_root are allowed, but only if the clone source
4529 * is behind the current send position. This is checked while searching
4530 * for possible clone sources.
4531 */
4532 sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
4533
4534 /* We do a bsearch later */
4535 sort(sctx->clone_roots, sctx->clone_roots_cnt,
4536 sizeof(*sctx->clone_roots), __clone_root_cmp_sort,
4537 NULL);
4538
4539 ret = send_subvol(sctx);
4540 if (ret < 0)
4541 goto out;
4542
4543 ret = begin_cmd(sctx, BTRFS_SEND_C_END);
4544 if (ret < 0)
4545 goto out;
4546 ret = send_cmd(sctx);
4547 if (ret < 0)
4548 goto out;
4549
4550out:
4551 if (filp)
4552 fput(filp);
4553 kfree(arg);
4554 vfree(clone_sources_tmp);
4555
4556 if (sctx) {
4557 if (sctx->send_filp)
4558 fput(sctx->send_filp);
4559
4560 vfree(sctx->clone_roots);
4561 vfree(sctx->send_buf);
4562 vfree(sctx->read_buf);
4563
4564 name_cache_free(sctx);
4565
4566 kfree(sctx);
4567 }
4568
4569 return ret;
4570}
diff --git a/fs/btrfs/send.h b/fs/btrfs/send.h
new file mode 100644
index 000000000000..9934e948e57f
--- /dev/null
+++ b/fs/btrfs/send.h
@@ -0,0 +1,133 @@
1/*
2 * Copyright (C) 2012 Alexander Block. All rights reserved.
3 * Copyright (C) 2012 STRATO. All rights reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public
7 * License v2 as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public
15 * License along with this program; if not, write to the
16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17 * Boston, MA 021110-1307, USA.
18 */
19
20#include "ctree.h"
21
22#define BTRFS_SEND_STREAM_MAGIC "btrfs-stream"
23#define BTRFS_SEND_STREAM_VERSION 1
24
25#define BTRFS_SEND_BUF_SIZE (1024 * 64)
26#define BTRFS_SEND_READ_SIZE (1024 * 48)
27
28enum btrfs_tlv_type {
29 BTRFS_TLV_U8,
30 BTRFS_TLV_U16,
31 BTRFS_TLV_U32,
32 BTRFS_TLV_U64,
33 BTRFS_TLV_BINARY,
34 BTRFS_TLV_STRING,
35 BTRFS_TLV_UUID,
36 BTRFS_TLV_TIMESPEC,
37};
38
39struct btrfs_stream_header {
40 char magic[sizeof(BTRFS_SEND_STREAM_MAGIC)];
41 __le32 version;
42} __attribute__ ((__packed__));
43
44struct btrfs_cmd_header {
45 /* len excluding the header */
46 __le32 len;
47 __le16 cmd;
48 /* crc including the header with zero crc field */
49 __le32 crc;
50} __attribute__ ((__packed__));
51
52struct btrfs_tlv_header {
53 __le16 tlv_type;
54 /* len excluding the header */
55 __le16 tlv_len;
56} __attribute__ ((__packed__));
57
58/* commands */
59enum btrfs_send_cmd {
60 BTRFS_SEND_C_UNSPEC,
61
62 BTRFS_SEND_C_SUBVOL,
63 BTRFS_SEND_C_SNAPSHOT,
64
65 BTRFS_SEND_C_MKFILE,
66 BTRFS_SEND_C_MKDIR,
67 BTRFS_SEND_C_MKNOD,
68 BTRFS_SEND_C_MKFIFO,
69 BTRFS_SEND_C_MKSOCK,
70 BTRFS_SEND_C_SYMLINK,
71
72 BTRFS_SEND_C_RENAME,
73 BTRFS_SEND_C_LINK,
74 BTRFS_SEND_C_UNLINK,
75 BTRFS_SEND_C_RMDIR,
76
77 BTRFS_SEND_C_SET_XATTR,
78 BTRFS_SEND_C_REMOVE_XATTR,
79
80 BTRFS_SEND_C_WRITE,
81 BTRFS_SEND_C_CLONE,
82
83 BTRFS_SEND_C_TRUNCATE,
84 BTRFS_SEND_C_CHMOD,
85 BTRFS_SEND_C_CHOWN,
86 BTRFS_SEND_C_UTIMES,
87
88 BTRFS_SEND_C_END,
89 __BTRFS_SEND_C_MAX,
90};
91#define BTRFS_SEND_C_MAX (__BTRFS_SEND_C_MAX - 1)
92
93/* attributes in send stream */
94enum {
95 BTRFS_SEND_A_UNSPEC,
96
97 BTRFS_SEND_A_UUID,
98 BTRFS_SEND_A_CTRANSID,
99
100 BTRFS_SEND_A_INO,
101 BTRFS_SEND_A_SIZE,
102 BTRFS_SEND_A_MODE,
103 BTRFS_SEND_A_UID,
104 BTRFS_SEND_A_GID,
105 BTRFS_SEND_A_RDEV,
106 BTRFS_SEND_A_CTIME,
107 BTRFS_SEND_A_MTIME,
108 BTRFS_SEND_A_ATIME,
109 BTRFS_SEND_A_OTIME,
110
111 BTRFS_SEND_A_XATTR_NAME,
112 BTRFS_SEND_A_XATTR_DATA,
113
114 BTRFS_SEND_A_PATH,
115 BTRFS_SEND_A_PATH_TO,
116 BTRFS_SEND_A_PATH_LINK,
117
118 BTRFS_SEND_A_FILE_OFFSET,
119 BTRFS_SEND_A_DATA,
120
121 BTRFS_SEND_A_CLONE_UUID,
122 BTRFS_SEND_A_CLONE_CTRANSID,
123 BTRFS_SEND_A_CLONE_PATH,
124 BTRFS_SEND_A_CLONE_OFFSET,
125 BTRFS_SEND_A_CLONE_LEN,
126
127 __BTRFS_SEND_A_MAX,
128};
129#define BTRFS_SEND_A_MAX (__BTRFS_SEND_A_MAX - 1)
130
131#ifdef __KERNEL__
132long btrfs_ioctl_send(struct file *mnt_file, void __user *arg);
133#endif
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index cc20e95ea289..7ac7cdcc294e 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -22,6 +22,7 @@
22#include <linux/writeback.h> 22#include <linux/writeback.h>
23#include <linux/pagemap.h> 23#include <linux/pagemap.h>
24#include <linux/blkdev.h> 24#include <linux/blkdev.h>
25#include <linux/uuid.h>
25#include "ctree.h" 26#include "ctree.h"
26#include "disk-io.h" 27#include "disk-io.h"
27#include "transaction.h" 28#include "transaction.h"
@@ -953,11 +954,13 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
953 struct dentry *dentry; 954 struct dentry *dentry;
954 struct extent_buffer *tmp; 955 struct extent_buffer *tmp;
955 struct extent_buffer *old; 956 struct extent_buffer *old;
957 struct timespec cur_time = CURRENT_TIME;
956 int ret; 958 int ret;
957 u64 to_reserve = 0; 959 u64 to_reserve = 0;
958 u64 index = 0; 960 u64 index = 0;
959 u64 objectid; 961 u64 objectid;
960 u64 root_flags; 962 u64 root_flags;
963 uuid_le new_uuid;
961 964
962 rsv = trans->block_rsv; 965 rsv = trans->block_rsv;
963 966
@@ -1051,6 +1054,20 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1051 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1054 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1052 btrfs_set_root_flags(new_root_item, root_flags); 1055 btrfs_set_root_flags(new_root_item, root_flags);
1053 1056
1057 btrfs_set_root_generation_v2(new_root_item,
1058 trans->transid);
1059 uuid_le_gen(&new_uuid);
1060 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1061 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1062 BTRFS_UUID_SIZE);
1063 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1064 new_root_item->otime.nsec = cpu_to_le64(cur_time.tv_nsec);
1065 btrfs_set_root_otransid(new_root_item, trans->transid);
1066 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1067 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1068 btrfs_set_root_stransid(new_root_item, 0);
1069 btrfs_set_root_rtransid(new_root_item, 0);
1070
1054 old = btrfs_lock_root_node(root); 1071 old = btrfs_lock_root_node(root);
1055 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); 1072 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1056 if (ret) { 1073 if (ret) {
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-01 00:24:09 -0500