Changeset ed903174 in mainline for uspace/srv/vfs/vfs_node.c

Timestamp:

2010-02-10T23:51:23Z (15 years ago)

Author:

Martin Decky <martin@…>

Branches:

lfn, master, serial, ticket/834-toolchain-update, topic/msim-upgrade, topic/simplify-dev-export

Children:

e70edd1

Parents:

b32c604f

Message:

implement support for 64bit file offsets

• the libc API is a small deviation from standard, but we have no reason to keep a strict backward compatibility with ancient code so far
  • the basic signed 64bit offset type is called off64_t
    • lseek() and fseek() take off64_t arguments (since the argument represents a relative offset)
    • ftell() returns off64_t values (since it is a wrapper of lseek())
    • vfs_seek() implementation supports negative offsets when SEEK_CUR and SEEK_END is used
  • aoff64_t is used for internal unsigned representation of sizes (in bytes, blocks, etc.) and absolute offsets
    • mmap() and ftruncate() take aoff64_t arguments (since the full range of the absolute file offset should be used here)
    • struct stat stores the file size as aoff64_t
  • in both cases the explicit range of the types shown in the names is helpful for proper filesystem and IPC interaction
  • note: size_t should be used only for representing in-memory sizes and offsets, not device and file-related information, and vice versa
    • the code base still needs a thorough revision with respect to this rule
  • PRIdOFF64 and PRIuOFF64 can be used for printing the offsets
• VFS_OUT_LOOKUP returns the 64bit file size in two IPC arguments
  • since all 5 IPC arguments have already been taken, the fs_handle is returned as the return value (fs_handle has only 16 bits, thus the return value can be used for both indicating errors as negative values and returning positive handles)
• VFS_OUT_READ and VFS_OUT_WRITE use aoff64_t absolute offsets split into two IPC arguments

replace bn_t with aoff64_t as a generic 64bit bytes/block counter type

note: filesystem drivers need to be revised with respect to make sure that all out-of-range checks are correct (especially w.r.t. file and block offsets)

File:

• uspace/srv/vfs/vfs_node.c (modified) (1 diff)

uspace/srv/vfs/vfs_node.c

@@ -242 +242 @@
 {
         vfs_node_t *node = hash_table_get_instance(item, vfs_node_t, nh_link);
-        return (node->fs_handle == key[KEY_FS_HANDLE]) &&
+        return (node->fs_handle == (fs_handle_t) key[KEY_FS_HANDLE]) &&
             (node->dev_handle == key[KEY_DEV_HANDLE]) &&
             (node->index == key[KEY_INDEX]);