Mirrors_Framework/nuttx
1
0
Fork 0
mirror of https://github.com/apache/nuttx.git synced 2026-05-20 12:33:27 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Add support for umount2(target, MNT_FORCE) in the FAT file system.

Browse Source
This commit is contained in:
Gregory Nutt
2015-03-15 07:45:19 -06:00
parent 43936a6a69
commit 9f7f258728
4 changed files with 274 additions and 256 deletions
Download Patch File Download Diff File Expand all files Collapse all files
TODO
+1 -67
View File
@@ -18,7 +18,7 @@ nuttx/
(12) Network (net/, drivers/net)
(4) USB (drivers/usbdev, drivers/usbhost)
(11) Libraries (libc/, libm/)
(13) File system/Generic drivers (fs/, drivers/)
(11) File system/Generic drivers (fs/, drivers/)
(9) Graphics subystem (graphics/)
(1) Pascal add-on (pcode/)
(1) Documentation (Documentation/)
@@ -1303,72 +1303,6 @@ o File system / Generic drivers (fs/, drivers/)
Status: Open
Priority: Medium
Title: UNMOUNT WITH OPEN FILES
Description: What is the policy for umounting files that have open file
references? Currently, umount() does the unmount
unconditionally and this has been noted to cause crashes in
some subsequent FAT file system operations. umount() is not
a standard OS interface and so have no specification. Here
are some possible behaviors or unmount() with open files:
1. Refuse to unmount() the file system if there are open
references to files on the file system (i.e., the file
system unbind() method returns a failure).
2. The file system knows that it has open references and so
does not unbind() immediately, but defers until the last
file is closed.
3. The file system performs the unbind() immediately, but
returns an error for an any subsequent operations on the
file system using one of the stale file handles.
4. In your application, do not un-mount until you have closed
all references to files or directories in the file system.
This would probably be a good practice anyway.
I lieu of any real specifications, I often just do what Linux
does. Here is now Linux does things:
http://man7.org/linux/man-pages/man2/umount.2.html .
Linux has have a umount2() that takes flags that control these
behaviors. They have a couple of other options as well.
It is not explicitly stated in the manpage, but it looks like
the Linux umount() does the first option: It should return
EBUSY meaning that the "target could not be unmounted because
it is busy." That one is pretty easy to implement within the
filesystem unbind() method.
I now think for an embedded system with mostly removable
storage devices, the option 3 is probably the most usable. For
example, NuttX has a little automounter at nuttx/fs/mount/fs_automount.c
That will automatically mount and unmount SD cards as they are
inserted or removed. I think that is a desire-able feature on
an embedded device. And I think it demands option 3 where the
umount occurs immediately, but then subsequent operations on
the volume fail.
The true lazy umount could also cause issues if the file is
never closed. Of course if the media is gone, I would hope
that the file access also fail in that case. But what if the
media is inserted and removed many times. Seems like the lazy
unmount could cause some problems and won't really buy you
anything anyway (because the file I/O will fail anyway).
Status: Open
Priority: Medium-High
Title: AUTMOUNTER BROKEN
Description: A bug was recently fixed in the FAT file system umount() logic.
FAT (and all other file systems) implement logic to fail the umount()
with EBUSY if there are open references to any file. For FAT, there
was a bug that prevented this from working so the busy check always
failed and FAT always permitted the umount(), whether there are open
file or not.
Unfortunately, the automounter depended on this bug to force the
un-mounting. I have changed the umount() call in the automounter to
umount2(MNT_FORCE), but the automounter will be broken until that
forced unmount feature is implemented.
Status: Open
Priority: Medium
o Graphics subsystem (graphics/)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
fs/binfs/fs_binfs.c
+1 -1
View File
@@ -100,7 +100,7 @@ static int binfs_stat(FAR struct inode *mountpt, FAR const char *relpath,
* Public Variables
****************************************************************************/
/* See fs_mount.c -- this structure is explicitly externed there.
/* See fs_mount.c -- this structure is explicitly extern'ed there.
* We use the old-fashioned kind of initializers so that this will compile
* with any compiler.
*/
fs/fat/fs_fat32.c
+266 -186
View File
File diff suppressed because it is too large Load Diff
fs/fat/fs_fat32.h
+6 -2
View File
@@ -271,12 +271,16 @@
#define FSTYPE_FAT16 1
#define FSTYPE_FAT32 2
/* File buffer flags */
/* File buffer flags (ff_bflags) */
#define FFBUFF_VALID 1
#define FFBUFF_DIRTY 2
#define FFBUFF_MODIFIED 4
/* Mount status flags (ff_bflags) */
#define UMOUNT_FORCED 8
/****************************************************************************
* These offset describe the FSINFO sector
*/
@@ -753,7 +757,7 @@ struct fat_mountpt_s
struct fat_file_s
{
struct fat_file_s *ff_next; /* Retained in a singly linked list */
uint8_t ff_bflags; /* The file buffer flags */
uint8_t ff_bflags; /* The file buffer/mount flags */
uint8_t ff_oflags; /* Flags provided when file was opened */
uint8_t ff_sectorsincluster; /* Sectors remaining in cluster */
uint16_t ff_dirindex; /* Index into ff_dirsector to directory entry */