btrfs: send: get rid of i_size logic in send_write()
authorOmar Sandoval <osandov@fb.com>
Fri, 21 Aug 2020 07:39:51 +0000 (00:39 -0700)
committerDavid Sterba <dsterba@suse.com>
Wed, 7 Oct 2020 10:13:17 +0000 (12:13 +0200)
send_write()/fill_read_buf() have some logic for avoiding reading past
i_size. However, everywhere that we call
send_write()/send_extent_data(), we've already clamped the length down
to i_size. Get rid of the i_size handling, which simplifies the next
change.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/send.c

index 7c7c09f..8af5e86 100644 (file)
@@ -4794,7 +4794,7 @@ out:
        return ret;
 }
 
-static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
+static int fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
 {
        struct btrfs_root *root = sctx->send_root;
        struct btrfs_fs_info *fs_info = root->fs_info;
@@ -4804,21 +4804,13 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
        pgoff_t index = offset >> PAGE_SHIFT;
        pgoff_t last_index;
        unsigned pg_offset = offset_in_page(offset);
-       ssize_t ret = 0;
+       int ret = 0;
+       size_t read = 0;
 
        inode = btrfs_iget(fs_info->sb, sctx->cur_ino, root);
        if (IS_ERR(inode))
                return PTR_ERR(inode);
 
-       if (offset + len > i_size_read(inode)) {
-               if (offset > i_size_read(inode))
-                       len = 0;
-               else
-                       len = offset - i_size_read(inode);
-       }
-       if (len == 0)
-               goto out;
-
        last_index = (offset + len - 1) >> PAGE_SHIFT;
 
        /* initial readahead */
@@ -4859,16 +4851,15 @@ static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len)
                }
 
                addr = kmap(page);
-               memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len);
+               memcpy(sctx->read_buf + read, addr + pg_offset, cur_len);
                kunmap(page);
                unlock_page(page);
                put_page(page);
                index++;
                pg_offset = 0;
                len -= cur_len;
-               ret += cur_len;
+               read += cur_len;
        }
-out:
        iput(inode);
        return ret;
 }
@@ -4882,7 +4873,6 @@ static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
        struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
        int ret = 0;
        struct fs_path *p;
-       ssize_t num_read = 0;
 
        p = fs_path_alloc();
        if (!p)
@@ -4890,12 +4880,9 @@ static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
 
        btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len);
 
-       num_read = fill_read_buf(sctx, offset, len);
-       if (num_read <= 0) {
-               if (num_read < 0)
-                       ret = num_read;
+       ret = fill_read_buf(sctx, offset, len);
+       if (ret < 0)
                goto out;
-       }
 
        ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE);
        if (ret < 0)
@@ -4907,16 +4894,14 @@ static int send_write(struct send_ctx *sctx, u64 offset, u32 len)
 
        TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p);
        TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset);
-       TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read);
+       TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len);
 
        ret = send_cmd(sctx);
 
 tlv_put_failure:
 out:
        fs_path_free(p);
-       if (ret < 0)
-               return ret;
-       return num_read;
+       return ret;
 }
 
 /*
@@ -5095,9 +5080,7 @@ static int send_extent_data(struct send_ctx *sctx,
                ret = send_write(sctx, offset + sent, size);
                if (ret < 0)
                        return ret;
-               if (!ret)
-                       break;
-               sent += ret;
+               sent += size;
        }
        return 0;
 }