~hc/RK356X_SDK_RELEASE.git

..	..	@@ -122,8 +122,6 @@
122	122
123	123	struct file_ra_state ra;
124	124
125		- char *read_buf;
126		-
127	125	/*
128	126	* We process inodes by their increasing order, so if before an
129	127	* incremental send we reverse the parent/child relationship of
..	..	@@ -268,6 +266,16 @@
268	266	int need_later_update;
269	267	int name_len;
270	268	char name[];
	269	+};
	270	+
	271	+#define ADVANCE 1
	272	+#define ADVANCE_ONLY_NEXT -1
	273	+
	274	+enum btrfs_compare_tree_result {
	275	+ BTRFS_COMPARE_TREE_NEW,
	276	+ BTRFS_COMPARE_TREE_DELETED,
	277	+ BTRFS_COMPARE_TREE_CHANGED,
	278	+ BTRFS_COMPARE_TREE_SAME,
271	279	};
272	280
273	281	__cold
..	..	@@ -570,8 +578,8 @@
570	578	return -EOVERFLOW;
571	579
572	580	hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size);
573		- hdr->tlv_type = cpu_to_le16(attr);
574		- hdr->tlv_len = cpu_to_le16(len);
	581	+ put_unaligned_le16(attr, &hdr->tlv_type);
	582	+ put_unaligned_le16(len, &hdr->tlv_len);
575	583	memcpy(hdr + 1, data, len);
576	584	sctx->send_size += total_len;
577	585
..	..	@@ -681,7 +689,7 @@
681	689
682	690	sctx->send_size += sizeof(*hdr);
683	691	hdr = (struct btrfs_cmd_header *)sctx->send_buf;
684		- hdr->cmd = cpu_to_le16(cmd);
	692	+ put_unaligned_le16(cmd, &hdr->cmd);
685	693
686	694	return 0;
687	695	}
..	..	@@ -693,17 +701,17 @@
693	701	u32 crc;
694	702
695	703	hdr = (struct btrfs_cmd_header *)sctx->send_buf;
696		- hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr));
697		- hdr->crc = 0;
	704	+ put_unaligned_le32(sctx->send_size - sizeof(*hdr), &hdr->len);
	705	+ put_unaligned_le32(0, &hdr->crc);
698	706
699		- crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
700		- hdr->crc = cpu_to_le32(crc);
	707	+ crc = btrfs_crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size);
	708	+ put_unaligned_le32(crc, &hdr->crc);
701	709
702	710	ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size,
703	711	&sctx->send_off);
704	712
705	713	sctx->total_send_size += sctx->send_size;
706		- sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size;
	714	+ sctx->cmd_send_size[get_unaligned_le16(&hdr->cmd)] += sctx->send_size;
707	715	sctx->send_size = 0;
708	716
709	717	return ret;
..	..	@@ -1170,7 +1178,6 @@
1170	1178	struct backref_ctx {
1171	1179	struct send_ctx *sctx;
1172	1180
1173		- struct btrfs_path *path;
1174	1181	/* number of total found references */
1175	1182	u64 found;
1176	1183
..	..	@@ -1196,9 +1203,9 @@
1196	1203	u64 root = (u64)(uintptr_t)key;
1197	1204	struct clone_root cr = (struct clone_root )elt;
1198	1205
1199		- if (root < cr->root->objectid)
	1206	+ if (root < cr->root->root_key.objectid)
1200	1207	return -1;
1201		- if (root > cr->root->objectid)
	1208	+ if (root > cr->root->root_key.objectid)
1202	1209	return 1;
1203	1210	return 0;
1204	1211	}
..	..	@@ -1208,9 +1215,9 @@
1208	1215	struct clone_root cr1 = (struct clone_root )e1;
1209	1216	struct clone_root cr2 = (struct clone_root )e2;
1210	1217
1211		- if (cr1->root->objectid < cr2->root->objectid)
	1218	+ if (cr1->root->root_key.objectid < cr2->root->root_key.objectid)
1212	1219	return -1;
1213		- if (cr1->root->objectid > cr2->root->objectid)
	1220	+ if (cr1->root->root_key.objectid > cr2->root->root_key.objectid)
1214	1221	return 1;
1215	1222	return 0;
1216	1223	}
..	..	@@ -1223,8 +1230,6 @@
1223	1230	{
1224	1231	struct backref_ctx *bctx = ctx_;
1225	1232	struct clone_root *found;
1226		- int ret;
1227		- u64 i_size;
1228	1233
1229	1234	/* First check if the root is in the list of accepted clone sources */
1230	1235	found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots,
..	..	@@ -1241,30 +1246,26 @@
1241	1246	}
1242	1247
1243	1248	/*
1244		- * There are inodes that have extents that lie behind its i_size. Don't
1245		- * accept clones from these extents.
1246		- */
1247		- ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL,
1248		- NULL, NULL, NULL);
1249		- btrfs_release_path(bctx->path);
1250		- if (ret < 0)
1251		- return ret;
1252		-
1253		- if (offset + bctx->data_offset + bctx->extent_len > i_size)
1254		- return 0;
1255		-
1256		- /*
1257	1249	* Make sure we don't consider clones from send_root that are
1258	1250	* behind the current inode/offset.
1259	1251	*/
1260	1252	if (found->root == bctx->sctx->send_root) {
1261	1253	/*
1262		- * TODO for the moment we don't accept clones from the inode
1263		- * that is currently send. We may change this when
1264		- * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same
1265		- * file.
	1254	+ * If the source inode was not yet processed we can't issue a
	1255	+ * clone operation, as the source extent does not exist yet at
	1256	+ * the destination of the stream.
1266	1257	*/
1267		- if (ino >= bctx->cur_objectid)
	1258	+ if (ino > bctx->cur_objectid)
	1259	+ return 0;
	1260	+ /*
	1261	+ * We clone from the inode currently being sent as long as the
	1262	+ * source extent is already processed, otherwise we could try
	1263	+ * to clone from an extent that does not exist yet at the
	1264	+ * destination of the stream.
	1265	+ */
	1266	+ if (ino == bctx->cur_objectid &&
	1267	+ offset + bctx->extent_len >
	1268	+ bctx->sctx->cur_inode_next_write_offset)
1268	1269	return 0;
1269	1270	}
1270	1271
..	..	@@ -1329,8 +1330,6 @@
1329	1330	ret = -ENOMEM;
1330	1331	goto out;
1331	1332	}
1332		-
1333		- backref_ctx->path = tmp_path;
1334	1333
1335	1334	if (data_offset >= ino_size) {
1336	1335	/*
..	..	@@ -1718,12 +1717,8 @@
1718	1717
1719	1718	di = btrfs_lookup_dir_item(NULL, root, path,
1720	1719	dir, name, name_len, 0);
1721		- if (!di) {
1722		- ret = -ENOENT;
1723		- goto out;
1724		- }
1725		- if (IS_ERR(di)) {
1726		- ret = PTR_ERR(di);
	1720	+ if (IS_ERR_OR_NULL(di)) {
	1721	+ ret = di ? PTR_ERR(di) : -ENOENT;
1727	1722	goto out;
1728	1723	}
1729	1724	btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
..	..	@@ -2267,7 +2262,7 @@
2267	2262	* inodes "orphan" name instead of the real name and stop. Same with new inodes
2268	2263	* that were not created yet and overwritten inodes/refs.
2269	2264	*
2270		- * When do we have have orphan inodes:
	2265	+ * When do we have orphan inodes:
2271	2266	* 1. When an inode is freshly created and thus no valid refs are available yet
2272	2267	* 2. When a directory lost all it's refs (deleted) but still has dir items
2273	2268	* inside which were not processed yet (pending for move/delete). If anyone
..	..	@@ -2371,7 +2366,7 @@
2371	2366	return -ENOMEM;
2372	2367	}
2373	2368
2374		- key.objectid = send_root->objectid;
	2369	+ key.objectid = send_root->root_key.objectid;
2375	2370	key.type = BTRFS_ROOT_BACKREF_KEY;
2376	2371	key.offset = 0;
2377	2372
..	..	@@ -2387,7 +2382,7 @@
2387	2382	leaf = path->nodes[0];
2388	2383	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2389	2384	if (key.type != BTRFS_ROOT_BACKREF_KEY \|\|
2390		- key.objectid != send_root->objectid) {
	2385	+ key.objectid != send_root->root_key.objectid) {
2391	2386	ret = -ENOENT;
2392	2387	goto out;
2393	2388	}
..	..	@@ -3957,52 +3952,56 @@
3957	3952	goto out;
3958	3953	}
3959	3954
	3955	+ /*
	3956	+ * Before doing any rename and link operations, do a first pass on the
	3957	+ * new references to orphanize any unprocessed inodes that may have a
	3958	+ * reference that conflicts with one of the new references of the current
	3959	+ * inode. This needs to happen first because a new reference may conflict
	3960	+ * with the old reference of a parent directory, so we must make sure
	3961	+ * that the path used for link and rename commands don't use an
	3962	+ * orphanized name when an ancestor was not yet orphanized.
	3963	+ *
	3964	+ * Example:
	3965	+ *
	3966	+ * Parent snapshot:
	3967	+ *
	3968	+ * . (ino 256)
	3969	+ * \|----- testdir/ (ino 259)
	3970	+ * \| \|----- a (ino 257)
	3971	+ * \|
	3972	+ * \|----- b (ino 258)
	3973	+ *
	3974	+ * Send snapshot:
	3975	+ *
	3976	+ * . (ino 256)
	3977	+ * \|----- testdir_2/ (ino 259)
	3978	+ * \| \|----- a (ino 260)
	3979	+ * \|
	3980	+ * \|----- testdir (ino 257)
	3981	+ * \|----- b (ino 257)
	3982	+ * \|----- b2 (ino 258)
	3983	+ *
	3984	+ * Processing the new reference for inode 257 with name "b" may happen
	3985	+ * before processing the new reference with name "testdir". If so, we
	3986	+ * must make sure that by the time we send a link command to create the
	3987	+ * hard link "b", inode 259 was already orphanized, since the generated
	3988	+ * path in "valid_path" already contains the orphanized name for 259.
	3989	+ * We are processing inode 257, so only later when processing 259 we do
	3990	+ * the rename operation to change its temporary (orphanized) name to
	3991	+ * "testdir_2".
	3992	+ */
3960	3993	list_for_each_entry(cur, &sctx->new_refs, list) {
3961		- /*
3962		- * We may have refs where the parent directory does not exist
3963		- * yet. This happens if the parent directories inum is higher
3964		- * the the current inum. To handle this case, we create the
3965		- * parent directory out of order. But we need to check if this
3966		- * did already happen before due to other refs in the same dir.
3967		- */
3968	3994	ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
3969	3995	if (ret < 0)
3970	3996	goto out;
3971		- if (ret == inode_state_will_create) {
3972		- ret = 0;
3973		- /*
3974		- * First check if any of the current inodes refs did
3975		- * already create the dir.
3976		- */
3977		- list_for_each_entry(cur2, &sctx->new_refs, list) {
3978		- if (cur == cur2)
3979		- break;
3980		- if (cur2->dir == cur->dir) {
3981		- ret = 1;
3982		- break;
3983		- }
3984		- }
3985		-
3986		- /*
3987		- * If that did not happen, check if a previous inode
3988		- * did already create the dir.
3989		- */
3990		- if (!ret)
3991		- ret = did_create_dir(sctx, cur->dir);
3992		- if (ret < 0)
3993		- goto out;
3994		- if (!ret) {
3995		- ret = send_create_inode(sctx, cur->dir);
3996		- if (ret < 0)
3997		- goto out;
3998		- }
3999		- }
	3997	+ if (ret == inode_state_will_create)
	3998	+ continue;
4000	3999
4001	4000	/*
4002		- * Check if this new ref would overwrite the first ref of
4003		- * another unprocessed inode. If yes, orphanize the
4004		- * overwritten inode. If we find an overwritten ref that is
4005		- * not the first ref, simply unlink it.
	4001	+ * Check if this new ref would overwrite the first ref of another
	4002	+ * unprocessed inode. If yes, orphanize the overwritten inode.
	4003	+ * If we find an overwritten ref that is not the first ref,
	4004	+ * simply unlink it.
4006	4005	*/
4007	4006	ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen,
4008	4007	cur->name, cur->name_len,
..	..	@@ -4093,6 +4092,49 @@
4093	4092	goto out;
4094	4093	}
4095	4094	ret = send_unlink(sctx, cur->full_path);
	4095	+ if (ret < 0)
	4096	+ goto out;
	4097	+ }
	4098	+ }
	4099	+
	4100	+ }
	4101	+
	4102	+ list_for_each_entry(cur, &sctx->new_refs, list) {
	4103	+ /*
	4104	+ * We may have refs where the parent directory does not exist
	4105	+ * yet. This happens if the parent directories inum is higher
	4106	+ * than the current inum. To handle this case, we create the
	4107	+ * parent directory out of order. But we need to check if this
	4108	+ * did already happen before due to other refs in the same dir.
	4109	+ */
	4110	+ ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen);
	4111	+ if (ret < 0)
	4112	+ goto out;
	4113	+ if (ret == inode_state_will_create) {
	4114	+ ret = 0;
	4115	+ /*
	4116	+ * First check if any of the current inodes refs did
	4117	+ * already create the dir.
	4118	+ */
	4119	+ list_for_each_entry(cur2, &sctx->new_refs, list) {
	4120	+ if (cur == cur2)
	4121	+ break;
	4122	+ if (cur2->dir == cur->dir) {
	4123	+ ret = 1;
	4124	+ break;
	4125	+ }
	4126	+ }
	4127	+
	4128	+ /*
	4129	+ * If that did not happen, check if a previous inode
	4130	+ * did already create the dir.
	4131	+ */
	4132	+ if (!ret)
	4133	+ ret = did_create_dir(sctx, cur->dir);
	4134	+ if (ret < 0)
	4135	+ goto out;
	4136	+ if (!ret) {
	4137	+ ret = send_create_inode(sctx, cur->dir);
4096	4138	if (ret < 0)
4097	4139	goto out;
4098	4140	}
..	..	@@ -4893,35 +4935,43 @@
4893	4935	return ret;
4894	4936	}
4895	4937
4896		-static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
	4938	+static inline u64 max_send_read_size(const struct send_ctx *sctx)
	4939	+{
	4940	+ return sctx->send_max_size - SZ_16K;
	4941	+}
	4942	+
	4943	+static int put_data_header(struct send_ctx *sctx, u32 len)
	4944	+{
	4945	+ struct btrfs_tlv_header *hdr;
	4946	+
	4947	+ if (sctx->send_max_size - sctx->send_size < sizeof(*hdr) + len)
	4948	+ return -EOVERFLOW;
	4949	+ hdr = (struct btrfs_tlv_header *)(sctx->send_buf + sctx->send_size);
	4950	+ put_unaligned_le16(BTRFS_SEND_A_DATA, &hdr->tlv_type);
	4951	+ put_unaligned_le16(len, &hdr->tlv_len);
	4952	+ sctx->send_size += sizeof(*hdr);
	4953	+ return 0;
	4954	+}
	4955	+
	4956	+static int put_file_data(struct send_ctx *sctx, u64 offset, u32 len)
4897	4957	{
4898	4958	struct btrfs_root *root = sctx->send_root;
4899	4959	struct btrfs_fs_info *fs_info = root->fs_info;
4900	4960	struct inode *inode;
4901	4961	struct page *page;
4902	4962	char *addr;
4903		- struct btrfs_key key;
4904	4963	pgoff_t index = offset >> PAGE_SHIFT;
4905	4964	pgoff_t last_index;
4906		- unsigned pg_offset = offset & ~PAGE_MASK;
4907		- ssize_t ret = 0;
	4965	+ unsigned pg_offset = offset_in_page(offset);
	4966	+ int ret;
4908	4967
4909		- key.objectid = sctx->cur_ino;
4910		- key.type = BTRFS_INODE_ITEM_KEY;
4911		- key.offset = 0;
	4968	+ ret = put_data_header(sctx, len);
	4969	+ if (ret)
	4970	+ return ret;
4912	4971
4913		- inode = btrfs_iget(fs_info->sb, &key, root, NULL);
	4972	+ inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root);
4914	4973	if (IS_ERR(inode))
4915	4974	return PTR_ERR(inode);
4916		-
4917		- if (offset + len > i_size_read(inode)) {
4918		- if (offset > i_size_read(inode))
4919		- len = 0;
4920		- else
4921		- len = offset - i_size_read(inode);
4922		- }
4923		- if (len == 0)
4924		- goto out;
4925	4975
4926	4976	last_index = (offset + len - 1) >> PAGE_SHIFT;
4927	4977
..	..	@@ -4967,16 +5017,16 @@
4967	5017	}
4968	5018
4969	5019	addr = kmap(page);
4970		- memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
	5020	+ memcpy(sctx->send_buf + sctx->send_size, addr + pg_offset,
	5021	+ cur_len);
4971	5022	kunmap(page);
4972	5023	unlock_page(page);
4973	5024	put_page(page);
4974	5025	index++;
4975	5026	pg_offset = 0;
4976	5027	len -= cur_len;
4977		- ret += cur_len;
	5028	+ sctx->send_size += cur_len;
4978	5029	}
4979		-out:
4980	5030	iput(inode);
4981	5031	return ret;
4982	5032	}
..	..	@@ -4990,20 +5040,12 @@
4990	5040	struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
4991	5041	int ret = 0;
4992	5042	struct fs_path *p;
4993		- ssize_t num_read = 0;
4994	5043
4995	5044	p = fs_path_alloc();
4996	5045	if (!p)
4997	5046	return -ENOMEM;
4998	5047
4999	5048	btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
5000		-
5001		- num_read = fill_read_buf(sctx, offset, len);
5002		- if (num_read <= 0) {
5003		- if (num_read < 0)
5004		- ret = num_read;
5005		- goto out;
5006		- }
5007	5049
5008	5050	ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5009	5051	if (ret < 0)
..	..	@@ -5015,16 +5057,16 @@
5015	5057
5016	5058	TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5017	5059	TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5018		- TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
	5060	+ ret = put_file_data(sctx, offset, len);
	5061	+ if (ret < 0)
	5062	+ goto out;
5019	5063
5020	5064	ret = send_cmd(sctx);
5021	5065
5022	5066	tlv_put_failure:
5023	5067	out:
5024	5068	fs_path_free(p);
5025		- if (ret < 0)
5026		- return ret;
5027		- return num_read;
	5069	+ return ret;
5028	5070	}
5029	5071
5030	5072	/*
..	..	@@ -5040,8 +5082,8 @@
5040	5082
5041	5083	btrfs_debug(sctx->send_root->fs_info,
5042	5084	"send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu",
5043		- offset, len, clone_root->root->objectid, clone_root->ino,
5044		- clone_root->offset);
	5085	+ offset, len, clone_root->root->root_key.objectid,
	5086	+ clone_root->ino, clone_root->offset);
5045	5087
5046	5088	p = fs_path_alloc();
5047	5089	if (!p)
..	..	@@ -5136,8 +5178,8 @@
5136	5178	static int send_hole(struct send_ctx *sctx, u64 end)
5137	5179	{
5138	5180	struct fs_path *p = NULL;
	5181	+ u64 read_size = max_send_read_size(sctx);
5139	5182	u64 offset = sctx->cur_inode_last_extent;
5140		- u64 len;
5141	5183	int ret = 0;
5142	5184
5143	5185	/*
..	..	@@ -5164,16 +5206,19 @@
5164	5206	ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p);
5165	5207	if (ret < 0)
5166	5208	goto tlv_put_failure;
5167		- memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE);
5168	5209	while (offset < end) {
5169		- len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE);
	5210	+ u64 len = min(end - offset, read_size);
5170	5211
5171	5212	ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
5172	5213	if (ret < 0)
5173	5214	break;
5174	5215	TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
5175	5216	TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
5176		- TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
	5217	+ ret = put_data_header(sctx, len);
	5218	+ if (ret < 0)
	5219	+ break;
	5220	+ memset(sctx->send_buf + sctx->send_size, 0, len);
	5221	+ sctx->send_size += len;
5177	5222	ret = send_cmd(sctx);
5178	5223	if (ret < 0)
5179	5224	break;
..	..	@@ -5189,23 +5234,20 @@
5189	5234	const u64 offset,
5190	5235	const u64 len)
5191	5236	{
	5237	+ u64 read_size = max_send_read_size(sctx);
5192	5238	u64 sent = 0;
5193	5239
5194	5240	if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
5195	5241	return send_update_extent(sctx, offset, len);
5196	5242
5197	5243	while (sent < len) {
5198		- u64 size = len - sent;
	5244	+ u64 size = min(len - sent, read_size);
5199	5245	int ret;
5200	5246
5201		- if (size > BTRFS_SEND_READ_SIZE)
5202		- size = BTRFS_SEND_READ_SIZE;
5203	5247	ret = send_write(sctx, offset + sent, size);
5204	5248	if (ret < 0)
5205	5249	return ret;
5206		- if (!ret)
5207		- break;
5208		- sent += ret;
	5250	+ sent += size;
5209	5251	}
5210	5252	return 0;
5211	5253	}
..	..	@@ -5278,6 +5320,7 @@
5278	5320	struct btrfs_path *path;
5279	5321	struct btrfs_key key;
5280	5322	int ret;
	5323	+ u64 clone_src_i_size = 0;
5281	5324
5282	5325	/*
5283	5326	* Prevent cloning from a zero offset with a length matching the sector
..	..	@@ -5301,6 +5344,16 @@
5301	5344	path = alloc_path_for_send();
5302	5345	if (!path)
5303	5346	return -ENOMEM;
	5347	+
	5348	+ /*
	5349	+ * There are inodes that have extents that lie behind its i_size. Don't
	5350	+ * accept clones from these extents.
	5351	+ */
	5352	+ ret = __get_inode_info(clone_root->root, path, clone_root->ino,
	5353	+ &clone_src_i_size, NULL, NULL, NULL, NULL, NULL);
	5354	+ btrfs_release_path(path);
	5355	+ if (ret < 0)
	5356	+ goto out;
5304	5357
5305	5358	/*
5306	5359	* We can't send a clone operation for the entire range if we find
..	..	@@ -5344,6 +5397,8 @@
5344	5397	u8 type;
5345	5398	u64 ext_len;
5346	5399	u64 clone_len;
	5400	+ u64 clone_data_offset;
	5401	+ bool crossed_src_i_size = false;
5347	5402
5348	5403	if (slot >= btrfs_header_nritems(leaf)) {
5349	5404	ret = btrfs_next_leaf(clone_root->root, path);
..	..	@@ -5397,13 +5452,94 @@
5397	5452	if (key.offset >= clone_root->offset + len)
5398	5453	break;
5399	5454
	5455	+ if (key.offset >= clone_src_i_size)
	5456	+ break;
	5457	+
	5458	+ if (key.offset + ext_len > clone_src_i_size) {
	5459	+ ext_len = clone_src_i_size - key.offset;
	5460	+ crossed_src_i_size = true;
	5461	+ }
	5462	+
	5463	+ clone_data_offset = btrfs_file_extent_offset(leaf, ei);
	5464	+ if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) {
	5465	+ clone_root->offset = key.offset;
	5466	+ if (clone_data_offset < data_offset &&
	5467	+ clone_data_offset + ext_len > data_offset) {
	5468	+ u64 extent_offset;
	5469	+
	5470	+ extent_offset = data_offset - clone_data_offset;
	5471	+ ext_len -= extent_offset;
	5472	+ clone_data_offset += extent_offset;
	5473	+ clone_root->offset += extent_offset;
	5474	+ }
	5475	+ }
	5476	+
5400	5477	clone_len = min_t(u64, ext_len, len);
5401	5478
5402	5479	if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte &&
5403		- btrfs_file_extent_offset(leaf, ei) == data_offset)
5404		- ret = send_clone(sctx, offset, clone_len, clone_root);
5405		- else
	5480	+ clone_data_offset == data_offset) {
	5481	+ const u64 src_end = clone_root->offset + clone_len;
	5482	+ const u64 sectorsize = SZ_64K;
	5483	+
	5484	+ /*
	5485	+ * We can't clone the last block, when its size is not
	5486	+ * sector size aligned, into the middle of a file. If we
	5487	+ * do so, the receiver will get a failure (-EINVAL) when
	5488	+ * trying to clone or will silently corrupt the data in
	5489	+ * the destination file if it's on a kernel without the
	5490	+ * fix introduced by commit ac765f83f1397646
	5491	+ * ("Btrfs: fix data corruption due to cloning of eof
	5492	+ * block).
	5493	+ *
	5494	+ * So issue a clone of the aligned down range plus a
	5495	+ * regular write for the eof block, if we hit that case.
	5496	+ *
	5497	+ * Also, we use the maximum possible sector size, 64K,
	5498	+ * because we don't know what's the sector size of the
	5499	+ * filesystem that receives the stream, so we have to
	5500	+ * assume the largest possible sector size.
	5501	+ */
	5502	+ if (src_end == clone_src_i_size &&
	5503	+ !IS_ALIGNED(src_end, sectorsize) &&
	5504	+ offset + clone_len < sctx->cur_inode_size) {
	5505	+ u64 slen;
	5506	+
	5507	+ slen = ALIGN_DOWN(src_end - clone_root->offset,
	5508	+ sectorsize);
	5509	+ if (slen > 0) {
	5510	+ ret = send_clone(sctx, offset, slen,
	5511	+ clone_root);
	5512	+ if (ret < 0)
	5513	+ goto out;
	5514	+ }
	5515	+ ret = send_extent_data(sctx, offset + slen,
	5516	+ clone_len - slen);
	5517	+ } else {
	5518	+ ret = send_clone(sctx, offset, clone_len,
	5519	+ clone_root);
	5520	+ }
	5521	+ } else if (crossed_src_i_size && clone_len < len) {
	5522	+ /*
	5523	+ * If we are at i_size of the clone source inode and we
	5524	+ * can not clone from it, terminate the loop. This is
	5525	+ * to avoid sending two write operations, one with a
	5526	+ * length matching clone_len and the final one after
	5527	+ * this loop with a length of len - clone_len.
	5528	+ *
	5529	+ * When using encoded writes (BTRFS_SEND_FLAG_COMPRESSED
	5530	+ * was passed to the send ioctl), this helps avoid
	5531	+ * sending an encoded write for an offset that is not
	5532	+ * sector size aligned, in case the i_size of the source
	5533	+ * inode is not sector size aligned. That will make the
	5534	+ * receiver fallback to decompression of the data and
	5535	+ * writing it using regular buffered IO, therefore while
	5536	+ * not incorrect, it's not optimal due decompression and
	5537	+ * possible re-compression at the receiver.
	5538	+ */
	5539	+ break;
	5540	+ } else {
5406	5541	ret = send_extent_data(sctx, offset, clone_len);
	5542	+ }
5407	5543
5408	5544	if (ret < 0)
5409	5545	goto out;
..	..	@@ -5413,6 +5549,21 @@
5413	5549	break;
5414	5550	offset += clone_len;
5415	5551	clone_root->offset += clone_len;
	5552	+
	5553	+ /*
	5554	+ * If we are cloning from the file we are currently processing,
	5555	+ * and using the send root as the clone root, we must stop once
	5556	+ * the current clone offset reaches the current eof of the file
	5557	+ * at the receiver, otherwise we would issue an invalid clone
	5558	+ * operation (source range going beyond eof) and cause the
	5559	+ * receiver to fail. So if we reach the current eof, bail out
	5560	+ * and fallback to a regular write.
	5561	+ */
	5562	+ if (clone_root->root == sctx->send_root &&
	5563	+ clone_root->ino == sctx->cur_ino &&
	5564	+ clone_root->offset >= sctx->cur_inode_next_write_offset)
	5565	+ break;
	5566	+
5416	5567	data_offset += clone_len;
5417	5568	next:
5418	5569	path->slots[0]++;
..	..	@@ -5433,51 +5584,29 @@
5433	5584	struct clone_root *clone_root)
5434	5585	{
5435	5586	int ret = 0;
5436		- struct btrfs_file_extent_item *ei;
5437	5587	u64 offset = key->offset;
5438		- u64 len;
5439		- u8 type;
	5588	+ u64 end;
5440	5589	u64 bs = sctx->send_root->fs_info->sb->s_blocksize;
5441	5590
5442		- ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
5443		- struct btrfs_file_extent_item);
5444		- type = btrfs_file_extent_type(path->nodes[0], ei);
5445		- if (type == BTRFS_FILE_EXTENT_INLINE) {
5446		- len = btrfs_file_extent_ram_bytes(path->nodes[0], ei);
5447		- /*
5448		- * it is possible the inline item won't cover the whole page,
5449		- * but there may be items after this page. Make
5450		- * sure to send the whole thing
5451		- */
5452		- len = PAGE_ALIGN(len);
5453		- } else {
5454		- len = btrfs_file_extent_num_bytes(path->nodes[0], ei);
5455		- }
	5591	+ end = min_t(u64, btrfs_file_extent_end(path), sctx->cur_inode_size);
	5592	+ if (offset >= end)
	5593	+ return 0;
5456	5594
5457		- if (offset >= sctx->cur_inode_size) {
5458		- ret = 0;
5459		- goto out;
5460		- }
5461		- if (offset + len > sctx->cur_inode_size)
5462		- len = sctx->cur_inode_size - offset;
5463		- if (len == 0) {
5464		- ret = 0;
5465		- goto out;
5466		- }
5467		-
5468		- if (clone_root && IS_ALIGNED(offset + len, bs)) {
	5595	+ if (clone_root && IS_ALIGNED(end, bs)) {
	5596	+ struct btrfs_file_extent_item *ei;
5469	5597	u64 disk_byte;
5470	5598	u64 data_offset;
5471	5599
	5600	+ ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
	5601	+ struct btrfs_file_extent_item);
5472	5602	disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei);
5473	5603	data_offset = btrfs_file_extent_offset(path->nodes[0], ei);
5474	5604	ret = clone_range(sctx, clone_root, disk_byte, data_offset,
5475		- offset, len);
	5605	+ offset, end - offset);
5476	5606	} else {
5477		- ret = send_extent_data(sctx, offset, len);
	5607	+ ret = send_extent_data(sctx, offset, end - offset);
5478	5608	}
5479		- sctx->cur_inode_next_write_offset = offset + len;
5480		-out:
	5609	+ sctx->cur_inode_next_write_offset = end;
5481	5610	return ret;
5482	5611	}
5483	5612
..	..	@@ -5675,10 +5804,7 @@
5675	5804	{
5676	5805	struct btrfs_path *path;
5677	5806	struct btrfs_root *root = sctx->send_root;
5678		- struct btrfs_file_extent_item *fi;
5679	5807	struct btrfs_key key;
5680		- u64 extent_end;
5681		- u8 type;
5682	5808	int ret;
5683	5809
5684	5810	path = alloc_path_for_send();
..	..	@@ -5698,18 +5824,7 @@
5698	5824	if (key.objectid != sctx->cur_ino \|\| key.type != BTRFS_EXTENT_DATA_KEY)
5699	5825	goto out;
5700	5826
5701		- fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5702		- struct btrfs_file_extent_item);
5703		- type = btrfs_file_extent_type(path->nodes[0], fi);
5704		- if (type == BTRFS_FILE_EXTENT_INLINE) {
5705		- u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5706		- extent_end = ALIGN(key.offset + size,
5707		- sctx->send_root->fs_info->sectorsize);
5708		- } else {
5709		- extent_end = key.offset +
5710		- btrfs_file_extent_num_bytes(path->nodes[0], fi);
5711		- }
5712		- sctx->cur_inode_last_extent = extent_end;
	5827	+ sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5713	5828	out:
5714	5829	btrfs_free_path(path);
5715	5830	return ret;
..	..	@@ -5763,16 +5878,7 @@
5763	5878	break;
5764	5879
5765	5880	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5766		- if (btrfs_file_extent_type(leaf, fi) ==
5767		- BTRFS_FILE_EXTENT_INLINE) {
5768		- u64 size = btrfs_file_extent_ram_bytes(leaf, fi);
5769		-
5770		- extent_end = ALIGN(key.offset + size,
5771		- root->fs_info->sectorsize);
5772		- } else {
5773		- extent_end = key.offset +
5774		- btrfs_file_extent_num_bytes(leaf, fi);
5775		- }
	5881	+ extent_end = btrfs_file_extent_end(path);
5776	5882	if (extent_end <= start)
5777	5883	goto next;
5778	5884	if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) {
..	..	@@ -5793,9 +5899,6 @@
5793	5899	static int maybe_send_hole(struct send_ctx sctx, struct btrfs_path path,
5794	5900	struct btrfs_key *key)
5795	5901	{
5796		- struct btrfs_file_extent_item *fi;
5797		- u64 extent_end;
5798		- u8 type;
5799	5902	int ret = 0;
5800	5903
5801	5904	if (sctx->cur_ino != key->objectid \|\| !need_send_hole(sctx))
..	..	@@ -5805,18 +5908,6 @@
5805	5908	ret = get_last_extent(sctx, key->offset - 1);
5806	5909	if (ret)
5807	5910	return ret;
5808		- }
5809		-
5810		- fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5811		- struct btrfs_file_extent_item);
5812		- type = btrfs_file_extent_type(path->nodes[0], fi);
5813		- if (type == BTRFS_FILE_EXTENT_INLINE) {
5814		- u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi);
5815		- extent_end = ALIGN(key->offset + size,
5816		- sctx->send_root->fs_info->sectorsize);
5817		- } else {
5818		- extent_end = key->offset +
5819		- btrfs_file_extent_num_bytes(path->nodes[0], fi);
5820	5911	}
5821	5912
5822	5913	if (path->slots[0] == 0 &&
..	..	@@ -5844,7 +5935,7 @@
5844	5935	else
5845	5936	ret = 0;
5846	5937	}
5847		- sctx->cur_inode_last_extent = extent_end;
	5938	+ sctx->cur_inode_last_extent = btrfs_file_extent_end(path);
5848	5939	return ret;
5849	5940	}
5850	5941
..	..	@@ -6654,6 +6745,365 @@
6654	6745	return ret;
6655	6746	}
6656	6747
	6748	+static int tree_move_down(struct btrfs_path path, int level)
	6749	+{
	6750	+ struct extent_buffer *eb;
	6751	+
	6752	+ BUG_ON(*level == 0);
	6753	+ eb = btrfs_read_node_slot(path->nodes[level], path->slots[level]);
	6754	+ if (IS_ERR(eb))
	6755	+ return PTR_ERR(eb);
	6756	+
	6757	+ path->nodes[*level - 1] = eb;
	6758	+ path->slots[*level - 1] = 0;
	6759	+ (*level)--;
	6760	+ return 0;
	6761	+}
	6762	+
	6763	+static int tree_move_next_or_upnext(struct btrfs_path *path,
	6764	+ int *level, int root_level)
	6765	+{
	6766	+ int ret = 0;
	6767	+ int nritems;
	6768	+ nritems = btrfs_header_nritems(path->nodes[*level]);
	6769	+
	6770	+ path->slots[*level]++;
	6771	+
	6772	+ while (path->slots[*level] >= nritems) {
	6773	+ if (*level == root_level)
	6774	+ return -1;
	6775	+
	6776	+ /* move upnext */
	6777	+ path->slots[*level] = 0;
	6778	+ free_extent_buffer(path->nodes[*level]);
	6779	+ path->nodes[*level] = NULL;
	6780	+ (*level)++;
	6781	+ path->slots[*level]++;
	6782	+
	6783	+ nritems = btrfs_header_nritems(path->nodes[*level]);
	6784	+ ret = 1;
	6785	+ }
	6786	+ return ret;
	6787	+}
	6788	+
	6789	+/*
	6790	+ * Returns 1 if it had to move up and next. 0 is returned if it moved only next
	6791	+ * or down.
	6792	+ */
	6793	+static int tree_advance(struct btrfs_path *path,
	6794	+ int *level, int root_level,
	6795	+ int allow_down,
	6796	+ struct btrfs_key *key)
	6797	+{
	6798	+ int ret;
	6799	+
	6800	+ if (*level == 0 \|\| !allow_down) {
	6801	+ ret = tree_move_next_or_upnext(path, level, root_level);
	6802	+ } else {
	6803	+ ret = tree_move_down(path, level);
	6804	+ }
	6805	+ if (ret >= 0) {
	6806	+ if (*level == 0)
	6807	+ btrfs_item_key_to_cpu(path->nodes[*level], key,
	6808	+ path->slots[*level]);
	6809	+ else
	6810	+ btrfs_node_key_to_cpu(path->nodes[*level], key,
	6811	+ path->slots[*level]);
	6812	+ }
	6813	+ return ret;
	6814	+}
	6815	+
	6816	+static int tree_compare_item(struct btrfs_path *left_path,
	6817	+ struct btrfs_path *right_path,
	6818	+ char *tmp_buf)
	6819	+{
	6820	+ int cmp;
	6821	+ int len1, len2;
	6822	+ unsigned long off1, off2;
	6823	+
	6824	+ len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
	6825	+ len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
	6826	+ if (len1 != len2)
	6827	+ return 1;
	6828	+
	6829	+ off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
	6830	+ off2 = btrfs_item_ptr_offset(right_path->nodes[0],
	6831	+ right_path->slots[0]);
	6832	+
	6833	+ read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
	6834	+
	6835	+ cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
	6836	+ if (cmp)
	6837	+ return 1;
	6838	+ return 0;
	6839	+}
	6840	+
	6841	+/*
	6842	+ * This function compares two trees and calls the provided callback for
	6843	+ * every changed/new/deleted item it finds.
	6844	+ * If shared tree blocks are encountered, whole subtrees are skipped, making
	6845	+ * the compare pretty fast on snapshotted subvolumes.
	6846	+ *
	6847	+ * This currently works on commit roots only. As commit roots are read only,
	6848	+ * we don't do any locking. The commit roots are protected with transactions.
	6849	+ * Transactions are ended and rejoined when a commit is tried in between.
	6850	+ *
	6851	+ * This function checks for modifications done to the trees while comparing.
	6852	+ * If it detects a change, it aborts immediately.
	6853	+ */
	6854	+static int btrfs_compare_trees(struct btrfs_root *left_root,
	6855	+ struct btrfs_root right_root, void ctx)
	6856	+{
	6857	+ struct btrfs_fs_info *fs_info = left_root->fs_info;
	6858	+ int ret;
	6859	+ int cmp;
	6860	+ struct btrfs_path *left_path = NULL;
	6861	+ struct btrfs_path *right_path = NULL;
	6862	+ struct btrfs_key left_key;
	6863	+ struct btrfs_key right_key;
	6864	+ char *tmp_buf = NULL;
	6865	+ int left_root_level;
	6866	+ int right_root_level;
	6867	+ int left_level;
	6868	+ int right_level;
	6869	+ int left_end_reached;
	6870	+ int right_end_reached;
	6871	+ int advance_left;
	6872	+ int advance_right;
	6873	+ u64 left_blockptr;
	6874	+ u64 right_blockptr;
	6875	+ u64 left_gen;
	6876	+ u64 right_gen;
	6877	+
	6878	+ left_path = btrfs_alloc_path();
	6879	+ if (!left_path) {
	6880	+ ret = -ENOMEM;
	6881	+ goto out;
	6882	+ }
	6883	+ right_path = btrfs_alloc_path();
	6884	+ if (!right_path) {
	6885	+ ret = -ENOMEM;
	6886	+ goto out;
	6887	+ }
	6888	+
	6889	+ tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
	6890	+ if (!tmp_buf) {
	6891	+ ret = -ENOMEM;
	6892	+ goto out;
	6893	+ }
	6894	+
	6895	+ left_path->search_commit_root = 1;
	6896	+ left_path->skip_locking = 1;
	6897	+ right_path->search_commit_root = 1;
	6898	+ right_path->skip_locking = 1;
	6899	+
	6900	+ /*
	6901	+ * Strategy: Go to the first items of both trees. Then do
	6902	+ *
	6903	+ * If both trees are at level 0
	6904	+ * Compare keys of current items
	6905	+ * If left < right treat left item as new, advance left tree
	6906	+ * and repeat
	6907	+ * If left > right treat right item as deleted, advance right tree
	6908	+ * and repeat
	6909	+ * If left == right do deep compare of items, treat as changed if
	6910	+ * needed, advance both trees and repeat
	6911	+ * If both trees are at the same level but not at level 0
	6912	+ * Compare keys of current nodes/leafs
	6913	+ * If left < right advance left tree and repeat
	6914	+ * If left > right advance right tree and repeat
	6915	+ * If left == right compare blockptrs of the next nodes/leafs
	6916	+ * If they match advance both trees but stay at the same level
	6917	+ * and repeat
	6918	+ * If they don't match advance both trees while allowing to go
	6919	+ * deeper and repeat
	6920	+ * If tree levels are different
	6921	+ * Advance the tree that needs it and repeat
	6922	+ *
	6923	+ * Advancing a tree means:
	6924	+ * If we are at level 0, try to go to the next slot. If that's not
	6925	+ * possible, go one level up and repeat. Stop when we found a level
	6926	+ * where we could go to the next slot. We may at this point be on a
	6927	+ * node or a leaf.
	6928	+ *
	6929	+ * If we are not at level 0 and not on shared tree blocks, go one
	6930	+ * level deeper.
	6931	+ *
	6932	+ * If we are not at level 0 and on shared tree blocks, go one slot to
	6933	+ * the right if possible or go up and right.
	6934	+ */
	6935	+
	6936	+ down_read(&fs_info->commit_root_sem);
	6937	+ left_level = btrfs_header_level(left_root->commit_root);
	6938	+ left_root_level = left_level;
	6939	+ left_path->nodes[left_level] =
	6940	+ btrfs_clone_extent_buffer(left_root->commit_root);
	6941	+ if (!left_path->nodes[left_level]) {
	6942	+ up_read(&fs_info->commit_root_sem);
	6943	+ ret = -ENOMEM;
	6944	+ goto out;
	6945	+ }
	6946	+
	6947	+ right_level = btrfs_header_level(right_root->commit_root);
	6948	+ right_root_level = right_level;
	6949	+ right_path->nodes[right_level] =
	6950	+ btrfs_clone_extent_buffer(right_root->commit_root);
	6951	+ if (!right_path->nodes[right_level]) {
	6952	+ up_read(&fs_info->commit_root_sem);
	6953	+ ret = -ENOMEM;
	6954	+ goto out;
	6955	+ }
	6956	+ up_read(&fs_info->commit_root_sem);
	6957	+
	6958	+ if (left_level == 0)
	6959	+ btrfs_item_key_to_cpu(left_path->nodes[left_level],
	6960	+ &left_key, left_path->slots[left_level]);
	6961	+ else
	6962	+ btrfs_node_key_to_cpu(left_path->nodes[left_level],
	6963	+ &left_key, left_path->slots[left_level]);
	6964	+ if (right_level == 0)
	6965	+ btrfs_item_key_to_cpu(right_path->nodes[right_level],
	6966	+ &right_key, right_path->slots[right_level]);
	6967	+ else
	6968	+ btrfs_node_key_to_cpu(right_path->nodes[right_level],
	6969	+ &right_key, right_path->slots[right_level]);
	6970	+
	6971	+ left_end_reached = right_end_reached = 0;
	6972	+ advance_left = advance_right = 0;
	6973	+
	6974	+ while (1) {
	6975	+ cond_resched();
	6976	+ if (advance_left && !left_end_reached) {
	6977	+ ret = tree_advance(left_path, &left_level,
	6978	+ left_root_level,
	6979	+ advance_left != ADVANCE_ONLY_NEXT,
	6980	+ &left_key);
	6981	+ if (ret == -1)
	6982	+ left_end_reached = ADVANCE;
	6983	+ else if (ret < 0)
	6984	+ goto out;
	6985	+ advance_left = 0;
	6986	+ }
	6987	+ if (advance_right && !right_end_reached) {
	6988	+ ret = tree_advance(right_path, &right_level,
	6989	+ right_root_level,
	6990	+ advance_right != ADVANCE_ONLY_NEXT,
	6991	+ &right_key);
	6992	+ if (ret == -1)
	6993	+ right_end_reached = ADVANCE;
	6994	+ else if (ret < 0)
	6995	+ goto out;
	6996	+ advance_right = 0;
	6997	+ }
	6998	+
	6999	+ if (left_end_reached && right_end_reached) {
	7000	+ ret = 0;
	7001	+ goto out;
	7002	+ } else if (left_end_reached) {
	7003	+ if (right_level == 0) {
	7004	+ ret = changed_cb(left_path, right_path,
	7005	+ &right_key,
	7006	+ BTRFS_COMPARE_TREE_DELETED,
	7007	+ ctx);
	7008	+ if (ret < 0)
	7009	+ goto out;
	7010	+ }
	7011	+ advance_right = ADVANCE;
	7012	+ continue;
	7013	+ } else if (right_end_reached) {
	7014	+ if (left_level == 0) {
	7015	+ ret = changed_cb(left_path, right_path,
	7016	+ &left_key,
	7017	+ BTRFS_COMPARE_TREE_NEW,
	7018	+ ctx);
	7019	+ if (ret < 0)
	7020	+ goto out;
	7021	+ }
	7022	+ advance_left = ADVANCE;
	7023	+ continue;
	7024	+ }
	7025	+
	7026	+ if (left_level == 0 && right_level == 0) {
	7027	+ cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
	7028	+ if (cmp < 0) {
	7029	+ ret = changed_cb(left_path, right_path,
	7030	+ &left_key,
	7031	+ BTRFS_COMPARE_TREE_NEW,
	7032	+ ctx);
	7033	+ if (ret < 0)
	7034	+ goto out;
	7035	+ advance_left = ADVANCE;
	7036	+ } else if (cmp > 0) {
	7037	+ ret = changed_cb(left_path, right_path,
	7038	+ &right_key,
	7039	+ BTRFS_COMPARE_TREE_DELETED,
	7040	+ ctx);
	7041	+ if (ret < 0)
	7042	+ goto out;
	7043	+ advance_right = ADVANCE;
	7044	+ } else {
	7045	+ enum btrfs_compare_tree_result result;
	7046	+
	7047	+ WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
	7048	+ ret = tree_compare_item(left_path, right_path,
	7049	+ tmp_buf);
	7050	+ if (ret)
	7051	+ result = BTRFS_COMPARE_TREE_CHANGED;
	7052	+ else
	7053	+ result = BTRFS_COMPARE_TREE_SAME;
	7054	+ ret = changed_cb(left_path, right_path,
	7055	+ &left_key, result, ctx);
	7056	+ if (ret < 0)
	7057	+ goto out;
	7058	+ advance_left = ADVANCE;
	7059	+ advance_right = ADVANCE;
	7060	+ }
	7061	+ } else if (left_level == right_level) {
	7062	+ cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
	7063	+ if (cmp < 0) {
	7064	+ advance_left = ADVANCE;
	7065	+ } else if (cmp > 0) {
	7066	+ advance_right = ADVANCE;
	7067	+ } else {
	7068	+ left_blockptr = btrfs_node_blockptr(
	7069	+ left_path->nodes[left_level],
	7070	+ left_path->slots[left_level]);
	7071	+ right_blockptr = btrfs_node_blockptr(
	7072	+ right_path->nodes[right_level],
	7073	+ right_path->slots[right_level]);
	7074	+ left_gen = btrfs_node_ptr_generation(
	7075	+ left_path->nodes[left_level],
	7076	+ left_path->slots[left_level]);
	7077	+ right_gen = btrfs_node_ptr_generation(
	7078	+ right_path->nodes[right_level],
	7079	+ right_path->slots[right_level]);
	7080	+ if (left_blockptr == right_blockptr &&
	7081	+ left_gen == right_gen) {
	7082	+ /*
	7083	+ * As we're on a shared block, don't
	7084	+ * allow to go deeper.
	7085	+ */
	7086	+ advance_left = ADVANCE_ONLY_NEXT;
	7087	+ advance_right = ADVANCE_ONLY_NEXT;
	7088	+ } else {
	7089	+ advance_left = ADVANCE;
	7090	+ advance_right = ADVANCE;
	7091	+ }
	7092	+ }
	7093	+ } else if (left_level < right_level) {
	7094	+ advance_right = ADVANCE;
	7095	+ } else {
	7096	+ advance_left = ADVANCE;
	7097	+ }
	7098	+ }
	7099	+
	7100	+out:
	7101	+ btrfs_free_path(left_path);
	7102	+ btrfs_free_path(right_path);
	7103	+ kvfree(tmp_buf);
	7104	+ return ret;
	7105	+}
	7106	+
6657	7107	static int send_subvol(struct send_ctx *sctx)
6658	7108	{
6659	7109	int ret;
..	..	@@ -6669,8 +7119,7 @@
6669	7119	goto out;
6670	7120
6671	7121	if (sctx->parent_root) {
6672		- ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root,
6673		- changed_cb, sctx);
	7122	+ ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx);
6674	7123	if (ret < 0)
6675	7124	goto out;
6676	7125	ret = finish_inode_if_needed(sctx, 1);
..	..	@@ -6779,20 +7228,25 @@
6779	7228	spin_unlock(&root->root_item_lock);
6780	7229	}
6781	7230
	7231	+static void dedupe_in_progress_warn(const struct btrfs_root *root)
	7232	+{
	7233	+ btrfs_warn_rl(root->fs_info,
	7234	+"cannot use root %llu for send while deduplications on it are in progress (%d in progress)",
	7235	+ root->root_key.objectid, root->dedupe_in_progress);
	7236	+}
	7237	+
6782	7238	long btrfs_ioctl_send(struct file mnt_file, struct btrfs_ioctl_send_args arg)
6783	7239	{
6784	7240	int ret = 0;
6785	7241	struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root;
6786	7242	struct btrfs_fs_info *fs_info = send_root->fs_info;
6787	7243	struct btrfs_root *clone_root;
6788		- struct btrfs_key key;
6789	7244	struct send_ctx *sctx = NULL;
6790	7245	u32 i;
6791	7246	u64 *clone_sources_tmp = NULL;
6792	7247	int clone_sources_to_rollback = 0;
6793		- unsigned alloc_size;
	7248	+ size_t alloc_size;
6794	7249	int sort_clone_roots = 0;
6795		- int index;
6796	7250
6797	7251	if (!capable(CAP_SYS_ADMIN))
6798	7252	return -EPERM;
..	..	@@ -6802,6 +7256,11 @@
6802	7256	* making it RW. This also protects against deletion.
6803	7257	*/
6804	7258	spin_lock(&send_root->root_item_lock);
	7259	+ if (btrfs_root_readonly(send_root) && send_root->dedupe_in_progress) {
	7260	+ dedupe_in_progress_warn(send_root);
	7261	+ spin_unlock(&send_root->root_item_lock);
	7262	+ return -EAGAIN;
	7263	+ }
6805	7264	send_root->send_in_progress++;
6806	7265	spin_unlock(&send_root->root_item_lock);
6807	7266
..	..	@@ -6817,18 +7276,11 @@
6817	7276	/*
6818	7277	* Check that we don't overflow at later allocations, we request
6819	7278	* clone_sources_count + 1 items, and compare to unsigned long inside
6820		- * access_ok.
	7279	+ * access_ok. Also set an upper limit for allocation size so this can't
	7280	+ * easily exhaust memory. Max number of clone sources is about 200K.
6821	7281	*/
6822		- if (arg->clone_sources_count >
6823		- ULONG_MAX / sizeof(struct clone_root) - 1) {
	7282	+ if (arg->clone_sources_count > SZ_8M / sizeof(struct clone_root)) {
6824	7283	ret = -EINVAL;
6825		- goto out;
6826		- }
6827		-
6828		- if (!access_ok(VERIFY_READ, arg->clone_sources,
6829		- sizeof(arg->clone_sources)
6830		- arg->clone_sources_count)) {
6831		- ret = -EFAULT;
6832	7284	goto out;
6833	7285	}
6834	7286
..	..	@@ -6875,25 +7327,20 @@
6875	7327	goto out;
6876	7328	}
6877	7329
6878		- sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL);
6879		- if (!sctx->read_buf) {
6880		- ret = -ENOMEM;
6881		- goto out;
6882		- }
6883		-
6884	7330	sctx->pending_dir_moves = RB_ROOT;
6885	7331	sctx->waiting_dir_moves = RB_ROOT;
6886	7332	sctx->orphan_dirs = RB_ROOT;
6887	7333
6888		- alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1);
6889		-
6890		- sctx->clone_roots = kvzalloc(alloc_size, GFP_KERNEL);
	7334	+ sctx->clone_roots = kvcalloc(sizeof(*sctx->clone_roots),
	7335	+ arg->clone_sources_count + 1,
	7336	+ GFP_KERNEL);
6891	7337	if (!sctx->clone_roots) {
6892	7338	ret = -ENOMEM;
6893	7339	goto out;
6894	7340	}
6895	7341
6896		- alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources);
	7342	+ alloc_size = array_size(sizeof(*arg->clone_sources),
	7343	+ arg->clone_sources_count);
6897	7344
6898	7345	if (arg->clone_sources_count) {
6899	7346	clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL);
..	..	@@ -6910,15 +7357,9 @@
6910	7357	}
6911	7358
6912	7359	for (i = 0; i < arg->clone_sources_count; i++) {
6913		- key.objectid = clone_sources_tmp[i];
6914		- key.type = BTRFS_ROOT_ITEM_KEY;
6915		- key.offset = (u64)-1;
6916		-
6917		- index = srcu_read_lock(&fs_info->subvol_srcu);
6918		-
6919		- clone_root = btrfs_read_fs_root_no_name(fs_info, &key);
	7360	+ clone_root = btrfs_get_fs_root(fs_info,
	7361	+ clone_sources_tmp[i], true);
6920	7362	if (IS_ERR(clone_root)) {
6921		- srcu_read_unlock(&fs_info->subvol_srcu, index);
6922	7363	ret = PTR_ERR(clone_root);
6923	7364	goto out;
6924	7365	}
..	..	@@ -6926,13 +7367,19 @@
6926	7367	if (!btrfs_root_readonly(clone_root) \|\|
6927	7368	btrfs_root_dead(clone_root)) {
6928	7369	spin_unlock(&clone_root->root_item_lock);
6929		- srcu_read_unlock(&fs_info->subvol_srcu, index);
	7370	+ btrfs_put_root(clone_root);
6930	7371	ret = -EPERM;
	7372	+ goto out;
	7373	+ }
	7374	+ if (clone_root->dedupe_in_progress) {
	7375	+ dedupe_in_progress_warn(clone_root);
	7376	+ spin_unlock(&clone_root->root_item_lock);
	7377	+ btrfs_put_root(clone_root);
	7378	+ ret = -EAGAIN;
6931	7379	goto out;
6932	7380	}
6933	7381	clone_root->send_in_progress++;
6934	7382	spin_unlock(&clone_root->root_item_lock);
6935		- srcu_read_unlock(&fs_info->subvol_srcu, index);
6936	7383
6937	7384	sctx->clone_roots[i].root = clone_root;
6938	7385	clone_sources_to_rollback = i + 1;
..	..	@@ -6942,15 +7389,9 @@
6942	7389	}
6943	7390
6944	7391	if (arg->parent_root) {
6945		- key.objectid = arg->parent_root;
6946		- key.type = BTRFS_ROOT_ITEM_KEY;
6947		- key.offset = (u64)-1;
6948		-
6949		- index = srcu_read_lock(&fs_info->subvol_srcu);
6950		-
6951		- sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key);
	7392	+ sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root,
	7393	+ true);
6952	7394	if (IS_ERR(sctx->parent_root)) {
6953		- srcu_read_unlock(&fs_info->subvol_srcu, index);
6954	7395	ret = PTR_ERR(sctx->parent_root);
6955	7396	goto out;
6956	7397	}
..	..	@@ -6960,13 +7401,16 @@
6960	7401	if (!btrfs_root_readonly(sctx->parent_root) \|\|
6961	7402	btrfs_root_dead(sctx->parent_root)) {
6962	7403	spin_unlock(&sctx->parent_root->root_item_lock);
6963		- srcu_read_unlock(&fs_info->subvol_srcu, index);
6964	7404	ret = -EPERM;
6965	7405	goto out;
6966	7406	}
	7407	+ if (sctx->parent_root->dedupe_in_progress) {
	7408	+ dedupe_in_progress_warn(sctx->parent_root);
	7409	+ spin_unlock(&sctx->parent_root->root_item_lock);
	7410	+ ret = -EAGAIN;
	7411	+ goto out;
	7412	+ }
6967	7413	spin_unlock(&sctx->parent_root->root_item_lock);
6968		-
6969		- srcu_read_unlock(&fs_info->subvol_srcu, index);
6970	7414	}
6971	7415
6972	7416	/*
..	..	@@ -6974,7 +7418,8 @@
6974	7418	* is behind the current send position. This is checked while searching
6975	7419	* for possible clone sources.
6976	7420	*/
6977		- sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root;
	7421	+ sctx->clone_roots[sctx->clone_roots_cnt++].root =
	7422	+ btrfs_grab_root(sctx->send_root);
6978	7423
6979	7424	/* We do a bsearch later */
6980	7425	sort(sctx->clone_roots, sctx->clone_roots_cnt,
..	..	@@ -6990,9 +7435,23 @@
6990	7435	if (ret)
6991	7436	goto out;
6992	7437
	7438	+ mutex_lock(&fs_info->balance_mutex);
	7439	+ if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
	7440	+ mutex_unlock(&fs_info->balance_mutex);
	7441	+ btrfs_warn_rl(fs_info,
	7442	+ "cannot run send because a balance operation is in progress");
	7443	+ ret = -EAGAIN;
	7444	+ goto out;
	7445	+ }
	7446	+ fs_info->send_in_progress++;
	7447	+ mutex_unlock(&fs_info->balance_mutex);
	7448	+
6993	7449	current->journal_info = BTRFS_SEND_TRANS_STUB;
6994	7450	ret = send_subvol(sctx);
6995	7451	current->journal_info = NULL;
	7452	+ mutex_lock(&fs_info->balance_mutex);
	7453	+ fs_info->send_in_progress--;
	7454	+ mutex_unlock(&fs_info->balance_mutex);
6996	7455	if (ret < 0)
6997	7456	goto out;
6998	7457
..	..	@@ -7045,18 +7504,24 @@
7045	7504	}
7046	7505
7047	7506	if (sort_clone_roots) {
7048		- for (i = 0; i < sctx->clone_roots_cnt; i++)
	7507	+ for (i = 0; i < sctx->clone_roots_cnt; i++) {
7049	7508	btrfs_root_dec_send_in_progress(
7050	7509	sctx->clone_roots[i].root);
	7510	+ btrfs_put_root(sctx->clone_roots[i].root);
	7511	+ }
7051	7512	} else {
7052		- for (i = 0; sctx && i < clone_sources_to_rollback; i++)
	7513	+ for (i = 0; sctx && i < clone_sources_to_rollback; i++) {
7053	7514	btrfs_root_dec_send_in_progress(
7054	7515	sctx->clone_roots[i].root);
	7516	+ btrfs_put_root(sctx->clone_roots[i].root);
	7517	+ }
7055	7518
7056	7519	btrfs_root_dec_send_in_progress(send_root);
7057	7520	}
7058		- if (sctx && !IS_ERR_OR_NULL(sctx->parent_root))
	7521	+ if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) {
7059	7522	btrfs_root_dec_send_in_progress(sctx->parent_root);
	7523	+ btrfs_put_root(sctx->parent_root);
	7524	+ }
7060	7525
7061	7526	kvfree(clone_sources_tmp);
7062	7527
..	..	@@ -7066,7 +7531,6 @@
7066	7531
7067	7532	kvfree(sctx->clone_roots);
7068	7533	kvfree(sctx->send_buf);
7069		- kvfree(sctx->read_buf);
7070	7534
7071	7535	name_cache_free(sctx);
7072	7536