Makes TIOCM ioctls for Data Carrier Detect & related functions
work like /drivers/serial/serial-core.c potentially needed
for pppd & similar user programs.
Signed-off-by: Denis Joseph Barrow <D.Barow@option.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:47:45 +0000 (13:47 +0000)]
tty: Fix the HSO termios handling a bit
Init the tty structure once
Don't set ->low_latency twice in a row
Don't force bits we should be leaving to the user
Don't allocate termios arrays as these are in fact allocated by the tty layer
for you and just overwrite the ones allocated in the driver
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:47:39 +0000 (13:47 +0000)]
hso: net driver using tty without locking
Checking tty == NULL doesn't help us unless we have a clear semantic for
the locking of the tty object in the driver. Use the tty kref objects so that
we can take references to the tty in the USB event handling paths.
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Thomas Pfaff [Fri, 2 Jan 2009 13:47:13 +0000 (13:47 +0000)]
tty: N_TTY SIGIO only works for read
The N_TTY ldisc layer does not send SIGIO POLL_OUTs correctly when output is
possible due to flawed handling of the TTY_DO_WRITE_WAKEUP bit. It will
either send no SIGIOs at all or on every tty wakeup.
The fix is to set the bit when the tty driver write would block and test
and clear it on write wakeup.
[Merged with existing N_TTY patches and a small buglet fixed -- Alan]
Signed-off-by: Thomas Pfaff <tpfaff@pcs.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:47:06 +0000 (13:47 +0000)]
tty: USB tty devices can block in tcdrain when unplugged
The underlying problem is that the device methods don't all correctly
handle disconnected status and some keep reporting bytes pending which
causes tcdrain to stall.
When the cable is unplugged they are definitely gone, and as this is true
for all USB cables we can fix it in the core usb serial code.
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Niels de Vos [Fri, 2 Jan 2009 13:46:58 +0000 (13:46 +0000)]
serial: set correct baud_base for Oxford Semiconductor Ltd EXSYS EX-41092 Dual 16950 Serial adapter
The PCI-card identified as "Oxford Semiconductor Ltd EXSYS EX-41092 Dual
16950 Serial adapter" is only usable with other devices (i.e. not the same
card) after doing a "setserial /dev/ttyS<n> baud_base 115200". This
baud_base should be default for this card.
Signed-off-by: Niels de Vos <niels.devos@wincor-nixdorf.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:45:36 +0000 (13:45 +0000)]
tty: relock generic_serial
Switch generic_serial to do port count locking via the tty_port structure
ready for moving to a common port wait routine. Keep the old driver lock for
internal calling so we don't risk messing up the drivers below until we
are ready.
Still needs kref conversions
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:45:26 +0000 (13:45 +0000)]
isicom: redo locking to use tty port locks
This helps set the basis for moving block_til_ready into common code. We also
introduce a tty_port_hangup helper as this will also be generally needed.
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
David Howells [Fri, 2 Jan 2009 13:44:49 +0000 (13:44 +0000)]
CRED: Wrap task credential accesses in the devpts filesystem
Wrap access to task credentials so that they can be separated more easily from
the task_struct during the introduction of COW creds.
Change most current->(|e|s|fs)[ug]id to current_(|e|s|fs)[ug]id().
Change some task->e?[ug]id to task_e?[ug]id(). In some places it makes more
sense to use RCU directly rather than a convenient wrapper; these will be
addressed by later patches.
Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Russell King [Fri, 2 Jan 2009 13:44:27 +0000 (13:44 +0000)]
And here's a patch (to be applied on top of the last) which prevents
this happening again by making use of 'const'.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Russell King [Fri, 2 Jan 2009 13:44:20 +0000 (13:44 +0000)]
Convert the oxsemi tornado special cases to use the quirk interface and not
scribble on its own reference structures.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alan Cox [Fri, 2 Jan 2009 13:44:04 +0000 (13:44 +0000)]
tty: Fix close races in USB serial
USB serial has always had races where the tty port usage count can hit zero
during a receive event. The internal locking is a mutex so we can't use
that in the IRQ handlers.
With krefs we can tackle this differently but we still need to be careful.
Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Joe Peterson [Fri, 2 Jan 2009 13:43:40 +0000 (13:43 +0000)]
n_tty: Output bells immediately on a full buffer
This patch causes "bell" (^G) characters (invoked when the input buffer
is full) to be immediately output rather than filling the echo buffer.
This is especially a problem when the tty is stopped and buffers fill, since
the bells do not serve their purpose of immediate notification that the
buffer cannot take further input, and they will flush all at once when the
tty is restarted.
Signed-off-by: Joe Peterson <joe@skyrush.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Joe Peterson [Fri, 2 Jan 2009 13:43:32 +0000 (13:43 +0000)]
n_tty: Fix hanfling of buffer full corner cases
Fix the handling of input characters when the tty buffer is full or nearly
full. This includes tests that are done in n_tty_receive_char() and handling
of PARMRK.
Problems with the buffer-full tests done in receive_char() caused characters to
be lost at times when the buffer(s) filled. Also, these full conditions
would often only be detected with echo on, and PARMRK was not accounted for
properly in all cases. One symptom of these problems, in addition to lost
characters, was early termination from unix commands like tr and cat when
^Q was used to break from a stopped tty with full buffers (note that breaking
out was often previously not possible, due to the pty getting in "gridlock",
which will be addressed in another patch). Note space is always reserved
at the end of the buffer for a newline (or EOF/EOL) in canonical mode.
Signed-off-by: Joe Peterson <joe@skyrush.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Joe Peterson [Fri, 2 Jan 2009 13:43:25 +0000 (13:43 +0000)]
n_tty: Fix handling of control characters and continuations
Fix process_output_block to detect continuation characters correctly
and to handle control characters even when O_OLCUC is enabled. Make
similar change to do_output_char().
Signed-off-by: Joe Peterson <joe@skyrush.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support containers, allow multiple instances of devpts filesystem, such
that indices of ptys allocated in one instance are independent of ptys
allocated in other instances of devpts.
But to preserve backward compatibility, enable this support for multiple
instances only if:
- CONFIG_DEVPTS_MULTIPLE_INSTANCES is set to Y, and
- '-o newinstance' mount option is specified while mounting devpts
To use multi-instance mount, a container startup script could:
where 'ns_exec -cm /bin/bash' is calls clone() with CLONE_NEWNS flag and execs
/bin/bash in the child process. A pty created by the sshd is not visible in
the original mount of /dev/pts.
USER-SPACE-IMPACT:
- See Documentation/fs/devpts.txt (included in next patch) for user-
space impact in multi-instance and mixed-mode operation.
TODO:
- Update mount(8), pts(4) man pages. Highlight impact of not
redirecting /dev/ptmx to /dev/pts/ptmx after a multi-instance mount.
Changelog[v6]:
- [Dave Hansen] Use new get_init_pts_sb() interface
- [Serge Hallyn] Don't bother displaying 'newinstance' in show_options
- [Serge Hallyn] Use macros (PARSE_REMOUNT/PARSE_MOUNT) instead of 0/1.
- [Serge Hallyn] Check error return from get_sb_single() (now
get_init_pts_sb())
- devpts_pty_kill(): don't dput error dentries
Changelog[v5]:
- Move get_sb_ref() definition to earlier patch
- Move usage info to Documentation/filesystems/devpts.txt (next patch)
- Make ptmx node even in init_pts_ns, now that default mode is 0000
(defined in earlier patch, enabled here).
- Cache ptmx dentry and use to update mode during remount
(defined in earlier patch, enabled here).
- Bugfix: explicitly ignore newinstance on remount (if newinstance was
specified on remount of initial mount, it would be ignored but
/proc/mounts would imply that the option was set)
Changelog[v4]:
- Update patch description to address H. Peter Anvin's comments
- Consolidate multi-instance mode code under new config token,
CONFIG_DEVPTS_MULTIPLE_INSTANCE.
- Move usage-details from patch description to
Documentation/fs/devpts.txt
Changelog[v3]:
- Rename new mount option to 'newinstance'
- Create ptmx nodes only in 'newinstance' mounts
- Bugfix: parse_mount_options() modifies @data but since we need to
parse the @data twice (once in devpts_get_sb() and once during
do_remount_sb()), parse a local copy of @data in devpts_get_sb().
(restructured code in devpts_get_sb() to fix this)
Changelog[v2]:
- Support both single-mount and multiple-mount semantics and
provide '-onewmnt' option to select the semantics.
/dev/ptmx is closely tied to the devpts filesystem. An open of /dev/ptmx,
allocates the next pty index and the associated device shows up in the
devpts fs as /dev/pts/n.
Wih multiple instancs of devpts filesystem, during an open of /dev/ptmx
we would be unable to determine which instance of the devpts is being
accessed.
So we move the 'ptmx' node into /dev/pts and use the inode of the 'ptmx'
node to identify the superblock and hence the devpts instance. This patch
adds ability for the kernel to internally create the [ptmx, c, 5:2] device
when mounting devpts filesystem. Since the ptmx node in devpts is new and
may surprise some userspace scripts, the default permissions for the new
node is 0000. These permissions can be changed either using chmod or by
remounting with the new '-o ptmxmode=0666' mount option.
Changelog[v5]:
- [Serge Hallyn bugfix]: Letting new_inode() assign inode number to
ptmx can collide with hand-assigning inode numbers to ptys. So,
hand-assign specific inode number to ptmx node also.
- [Serge Hallyn]: Maybe safer to grab root dentry mutex while creating
ptmx node
- [Bugfix with Serge Hallyn] Replace lookup_one_len() in mknod_ptmx()
wih d_alloc_name() (lookup during ->get_sb() locks up system). To
simplify patchset, fold the ptmx_dentry patch into this.
Changelog[v4]:
- Change default permissions of pts/ptmx node to 0000.
- Move code for ptmxmode under #ifdef CONFIG_DEVPTS_MULTIPLE_INSTANCES.
Changelog[v3]:
- Rename ptmx_mode to ptmxmode (for consistency with 'newinstance')
Changelog[v2]:
- [H. Peter Anvin] Remove mknod() system call support and create the
ptmx node internally.
Changelog[v1]:
- Earlier version of this patch enabled creating /dev/pts/tty as
well. As pointed out by Al Viro and H. Peter Anvin, that is not
really necessary.
With support for multiple mounts of devpts, the 'config' structure really
represents per-mount options rather than config parameters. Rename 'config'
structure to 'pts_mount_opts' and store it in the super-block.
To enable multiple mounts of devpts, 'allocated_ptys' must be a per-mount
variable rather than a global variable. Move 'allocated_ptys' into the
super_block's s_fs_info.
Changelog[v2]:
Define and use DEVPTS_SB() wrapper.
Remove the 'devpts_root' global variable and find the root dentry using
the super_block. The super-block can be found from the device inode, using
the new wrapper, pts_sb_from_inode().
Changelog: This patch is based on an earlier patchset from Serge Hallyn
and Matt Helsley.
Joe Peterson [Fri, 2 Jan 2009 13:40:53 +0000 (13:40 +0000)]
n_tty: Fix loss of echoed characters and remove bkl from n_tty
Fixes the loss of echoed (and other ldisc-generated characters) when
the tty is stopped or when the driver output buffer is full (happens
frequently for input during continuous program output, such as ^C)
and removes the Big Kernel Lock from the N_TTY line discipline.
Adds an "echo buffer" to the N_TTY line discipline that handles all
ldisc-generated output (including echoed characters). Along with the
loss of characters, this also fixes the associated loss of sync between
tty output and the ldisc state when characters cannot be immediately
written to the tty driver.
The echo buffer stores (in addition to characters) state operations that need
to be done at the time of character output (like management of the column
position). This allows echo to cooperate correctly with program output,
since the ldisc state remains consistent with actual characters written.
Since the echo buffer code now isolates the tty column state code
to the process_out* and process_echoes functions, we can remove the
Big Kernel Lock (BKL) and replace it with mutex locks.
Highlights are:
* Handles echo (and other ldisc output) when tty driver buffer is full
- continuous program output can block echo
* Saves echo when tty is in stopped state (e.g. ^S)
- (e.g.: ^Q will correctly cause held characters to be released for output)
* Control character pairs (e.g. "^C") are treated atomically and not
split up by interleaved program output
* Line discipline state is kept consistent with characters sent to
the tty driver
* Remove the big kernel lock (BKL) from N_TTY line discipline
Signed-off-by: Joe Peterson <joe@skyrush.com> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Graf Yang [Fri, 2 Jan 2009 13:40:22 +0000 (13:40 +0000)]
Blackfin Serial Driver: fix bug - SIR driver stop receiving randomly
Bug description:
The IRDA receiver may can't receiving any more after processed some signals.
To duplicate this issue is put three IRDA devices together, one blackfin,
two none blackfin, they will detect each other. Let one none blackfin devices
irdaping the blackfin devices, when it stopped print out ping information,
it is the time that blackfin stoped receiving, the time is random.
The related register bit is OK, the other devices is sending data continuously.
But no interrupt come.
Fixing:
I tried Michael's suggestion that request the UARTx error interrupt, and reset
the IRDA when found FE error. This method helps much, but it can't completely
avoid stop.
Reset the IRDA before every time sending the data is more safe.
Signed-off-by: Graf Yang <graf.yang@analog.com> Signed-off-by: Bryan Wu <cooloney@kernel.org> Signed-off-by: Alan Cox <alan@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Linus Torvalds [Wed, 31 Dec 2008 23:57:56 +0000 (15:57 -0800)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6: (34 commits)
nfsd race fixes: jfs
nfsd race fixes: reiserfs
nfsd race fixes: ext4
nfsd race fixes: ext3
nfsd race fixes: ext2
nfsd/create race fixes, infrastructure
filesystem notification: create fs/notify to contain all fs notification
fs/block_dev.c: __read_mostly improvement and sb_is_blkdev_sb utilization
kill ->dir_notify()
filp_cachep can be static in fs/file_table.c
fix f_count description in Documentation/filesystems/files.txt
make INIT_FS use the __RW_LOCK_UNLOCKED initialization
take init_fs to saner place
kill vfs_permission
pass a struct path * to may_open
kill walk_init_root
remove incorrect comment in inode_permission
expand some comments (d_path / seq_path)
correct wrong function name of d_put in kernel document and source comment
fix switch_names() breakage in short-to-short case
...
Al Viro [Tue, 30 Dec 2008 06:52:35 +0000 (01:52 -0500)]
nfsd race fixes: ext2
* make ext2_new_inode() put the inode into icache in locked state
* do not unlock until the inode is fully set up; otherwise nfsd
might pick it in half-baked state.
* make sure that ext2_new_inode() does *not* lead to two inodes with the
same inumber hashed at the same time; otherwise a bogus fhandle coming
from nfsd might race with inode creation:
nfsd: iget_locked() creates inode
nfsd: try to read from disk, block on that.
ext2_new_inode(): allocate inode with that inumber
ext2_new_inode(): insert it into icache, set it up and dirty
ext2_write_inode(): get the relevant part of inode table in cache,
set the entry for our inode (and start writing to disk)
nfsd: get CPU again, look into inode table, see nice and sane on-disk
inode, set the in-core inode from it
oops - we have two in-core inodes with the same inumber live in icache,
both used for IO. Welcome to fs corruption...
Al Viro [Tue, 30 Dec 2008 06:48:21 +0000 (01:48 -0500)]
nfsd/create race fixes, infrastructure
new helpers - insert_inode_locked() and insert_inode_locked4().
Hash new inode, making sure that there's no such inode in icache
already. If there is and it does not end up unhashed (as would
happen if we have nfsd trying to resolve a bogus fhandle), fail.
Otherwise insert our inode into hash and succeed.
In either case have i_state set to new+locked; cleanup ends up
being simpler with such calling conventions.
Eric Paris [Wed, 17 Dec 2008 18:59:41 +0000 (13:59 -0500)]
filesystem notification: create fs/notify to contain all fs notification
Creating a generic filesystem notification interface, fsnotify, which will be
used by inotify, dnotify, and eventually fanotify is really starting to
clutter the fs directory. This patch simply moves inotify and dnotify into
fs/notify/inotify and fs/notify/dnotify respectively to make both current fs/
and future notification tidier.
Signed-off-by: Eric Paris <eparis@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Al Viro [Fri, 26 Dec 2008 05:57:40 +0000 (00:57 -0500)]
kill ->dir_notify()
Remove the hopelessly misguided ->dir_notify(). The only instance (cifs)
has been broken by design from the very beginning; the objects it creates
are never destroyed, keep references to struct file they can outlive, nothing
that could possibly evict them exists on close(2) path *and* no locking
whatsoever is done to prevent races with close(), should the previous, er,
deficiencies someday be dealt with.
Eric Dumazet [Wed, 10 Dec 2008 17:35:45 +0000 (09:35 -0800)]
filp_cachep can be static in fs/file_table.c
Instead of creating the "filp" kmem_cache in vfs_caches_init(),
we can do it a litle be later in files_init(), so that filp_cachep
is static to fs/file_table.c
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Eric Dumazet [Wed, 10 Dec 2008 17:35:45 +0000 (09:35 -0800)]
fix f_count description in Documentation/filesystems/files.txt
Documentation/filesystems/files.txt was not updated when
f_count became an atomic_long_t.
atomic_long_inc_not_zero() is now used instead of atomic_inc_not_zero()
Steven Rostedt [Wed, 10 Dec 2008 23:37:28 +0000 (18:37 -0500)]
make INIT_FS use the __RW_LOCK_UNLOCKED initialization
[AV: rediffed on top of unification of init_fs]
Initialization of init_fs still uses the deprecated RW_LOCK_UNLOCKED macro.
This patch updates it to use the __RW_LOCK_UNLOCKED(lock) macro.
Signed-off-by: Steven Rostedt <srostedt@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
With all the nameidata removal there's no point anymore for this helper.
Of the three callers left two will go away with the next lookup series
anyway.
Also add proper kerneldoc to inode_permission as this is the main
permission check routine now.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Explain that you really need to use the return value of d_path rather than
the buffer you passed into it.
Also fix the comment for seq_path(), the function arguments changed
recently but the comment hadn't been updated in sync.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Zhaolei [Mon, 1 Dec 2008 22:34:58 +0000 (14:34 -0800)]
correct wrong function name of d_put in kernel document and source comment
no function named d_put(), it should be dput().
Impact: fix document and comment, no functionality changed
Signed-off-by: Zhao Lei <zhaolei@cn.fuijtsu.com> Signed-off-by: Randy Dunlap <rdunlap@xenotime.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Duane Griffin [Fri, 19 Dec 2008 20:47:12 +0000 (20:47 +0000)]
vfs: ensure page symlinks are NUL-terminated
On-disk data corruption could cause a page link to have its i_size set
to PAGE_SIZE (or a multiple thereof) and its contents all non-NUL.
NUL-terminate the link name to ensure this doesn't cause further
problems for the kernel.
Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Duane Griffin <duaneg@dghda.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Duane Griffin [Fri, 19 Dec 2008 20:47:11 +0000 (20:47 +0000)]
vfs: introduce helper function to safely NUL-terminate symlinks
A number of filesystems were potentially triggering kernel bugs due to
corrupted symlink names on disk. This function helps safely terminate
the names.
Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Duane Griffin <duaneg@dghda.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Duane Griffin [Fri, 19 Dec 2008 20:47:10 +0000 (20:47 +0000)]
eCryptfs: check readlink result was not an error before using it
The result from readlink is being used to index into the link name
buffer without checking whether it is a valid length. If readlink
returns an error this will fault or cause memory corruption.
Julia Lawall [Mon, 1 Dec 2008 22:34:51 +0000 (14:34 -0800)]
fs/namespace.c: drop code after return
The extra semicolon serves no purpose.
Signed-off-by: Julia Lawall <julia@diku.dk> Reviewed-by: Richard Genoud <richard.genoud@gmail.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Jan Engelhardt [Mon, 1 Dec 2008 22:34:50 +0000 (14:34 -0800)]
include: linux/fs.h: put declarations in __KERNEL__
include/linux/fs.h contains externs for a bunch of variables. That obviously
belongs under ifdef __KERNEL__.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Nick Piggin [Mon, 1 Dec 2008 08:33:43 +0000 (09:33 +0100)]
shrink struct dentry
struct dentry is one of the most critical structures in the kernel. So it's
sad to see it going neglected.
With CONFIG_PROFILING turned on (which is probably the common case at least
for distros and kernel developers), sizeof(struct dcache) == 208 here
(64-bit). This gives 19 objects per slab.
I packed d_mounted into a hole, and took another 4 bytes off the inline
name length to take the padding out from the end of the structure. This
shinks it to 200 bytes. I could have gone the other way and increased the
length to 40, but I'm aiming for a magic number, read on...
I then got rid of the d_cookie pointer. This shrinks it to 192 bytes. Rant:
why was this ever a good idea? The cookie system should increase its hash
size or use a tree or something if lookups are a problem. Also the "fast
dcookie lookups" in oprofile should be moved into the dcookie code -- how
can oprofile possibly care about the dcookie_mutex? It gets dropped after
get_dcookie() returns so it can't be providing any sort of protection.
At 192 bytes, 21 objects fit into a 4K page, saving about 3MB on my system
with ~140 000 entries allocated. 192 is also a multiple of 64, so we get
nice cacheline alignment on 64 and 32 byte line systems -- any given dentry
will now require 3 cachelines to touch all fields wheras previously it
would require 4.
I know the inline name size was chosen quite carefully, however with the
reduction in cacheline footprint, it should actually be just about as fast
to do a name lookup for a 36 character name as it was before the patch (and
faster for other sizes). The memory footprint savings for names which are
<= 32 or > 36 bytes long should more than make up for the memory cost for
33-36 byte names.
Richard Kennedy [Thu, 4 Dec 2008 11:17:47 +0000 (11:17 +0000)]
fs: reorder struct inotify_device on 64bits to remove padding
Reorder struct inotify_device to remove 8 bytes of padding on 64bit
builds, reducing size to 128 bytes . Therefore allocating from a smaller
slab & using one fewer cachelines.
Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
----
Hi,
patch against 2.6.28-rc7.
built & tested on AMDX2 desktop.
I've not been able to send this to the listed inotify maintainers, I
just get mail failures. So I guessed filesystem was the best home for
it, hope that's ok.
regards
Richard Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Linus Torvalds [Wed, 31 Dec 2008 17:00:59 +0000 (09:00 -0800)]
Merge branch 'irq-fixes-for-linus-4' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'irq-fixes-for-linus-4' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sparseirq: move __weak symbols into separate compilation unit
sparseirq: work around __weak alias bug
sparseirq: fix hang with !SPARSE_IRQ
sparseirq: set lock_class for legacy irq when sparse_irq is selected
sparseirq: work around compiler optimizing away __weak functions
sparseirq: fix desc->lock init
sparseirq: do not printk when migrating IRQ descriptors
sparseirq: remove duplicated arch_early_irq_init()
irq: simplify for_each_irq_desc() usage
proc: remove ifdef CONFIG_SPARSE_IRQ from stat.c
irq: for_each_irq_desc() move to irqnr.h
hrtimer: remove #include <linux/irq.h>
* git://git.kernel.org/pub/scm/linux/kernel/git/hirofumi/fatfs-2.6:
fat: make sure to set d_ops in fat_get_parent
fat: fix duplicate addition of ->llseek handler
fat: drop negative dentry on rename() path
Linus Torvalds [Wed, 31 Dec 2008 01:48:25 +0000 (17:48 -0800)]
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
* 'for-linus' of git://oss.sgi.com/xfs/xfs: (184 commits)
[XFS] Fix race in xfs_write() between direct and buffered I/O with DMAPI
[XFS] handle unaligned data in xfs_bmbt_disk_get_all
[XFS] avoid memory allocations in xfs_fs_vcmn_err
[XFS] Fix speculative allocation beyond eof
[XFS] Remove XFS_BUF_SHUT() and friends
[XFS] Use the incore inode size in xfs_file_readdir()
[XFS] set b_error from bio error in xfs_buf_bio_end_io
[XFS] use inode_change_ok for setattr permission checking
[XFS] add a FMODE flag to make XFS invisible I/O less hacky
[XFS] resync headers with libxfs
[XFS] simplify projid check in xfs_rename
[XFS] replace b_fspriv with b_mount
[XFS] Remove unused tracing code
[XFS] Remove unnecessary assertion
[XFS] Remove unused variable in ktrace_free()
[XFS] Check return value of xfs_buf_get_noaddr()
[XFS] Fix hang after disallowed rename across directory quota domains
[XFS] Fix compile with CONFIG_COMPAT enabled
move inode tracing out of xfs_vnode.
move vn_iowait / vn_iowake into xfs_aops.c
...
* git://git.linux-nfs.org/projects/trondmy/nfs-2.6: (70 commits)
fs/nfs/nfs4proc.c: make nfs4_map_errors() static
rpc: add service field to new upcall
rpc: add target field to new upcall
nfsd: support callbacks with gss flavors
rpc: allow gss callbacks to client
rpc: pass target name down to rpc level on callbacks
nfsd: pass client principal name in rsc downcall
rpc: implement new upcall
rpc: store pointer to pipe inode in gss upcall message
rpc: use count of pipe openers to wait for first open
rpc: track number of users of the gss upcall pipe
rpc: call release_pipe only on last close
rpc: add an rpc_pipe_open method
rpc: minor gss_alloc_msg cleanup
rpc: factor out warning code from gss_pipe_destroy_msg
rpc: remove unnecessary assignment
NFS: remove unused status from encode routines
NFS: increment number of operations in each encode routine
NFS: fix comment placement in nfs4xdr.c
NFS: fix tabs in nfs4xdr.c
...
Linus Torvalds [Wed, 31 Dec 2008 01:45:28 +0000 (17:45 -0800)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
IB/mlx4: Fix reading SL field out of cqe->sl_vid
RDMA/addr: Fix build breakage when IPv6 is disabled
projects / linux-beck.git / log