| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Add a new attribute flag ATTR_OPEN, with the meaning: "truncation was
initiated by open() due to the O_TRUNC flag".
This way filesystems wanting to implement truncation within their ->open()
method can ignore such truncate requests.
This is a quick & dirty hack, but it comes for free.
Signed-off-by: Miklos Szeredi <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Andreas Dilger <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Add necessary protocol changes for supplying a file handle with the getattr
operation. Step the API version to 7.9.
This patch doesn't actually supply the file handle, because that needs some
kind of VFS support, which we haven't yet been able to agree upon.
[[email protected]: coding-style fixes]
Signed-off-by: Miklos Szeredi <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
This patch set supports large block size(>4k, <=64k) in ext3 just enlarging
the block size limit. But it is NOT possible to have 64kB blocksize on
ext3 without some changes to the directory handling code. The reason is
that an empty 64kB directory block would have a rec_len == (__u16)2^16 ==
0, and this would cause an error to be hit in the filesystem. The proposed
solution is treat 64k rec_len with a an impossible value like rec_len =
0xffff to handle this.
The Patch-set consists of the following 2 patches.
[1/2] ext3: enlarge blocksize
- Allow blocksize up to pagesize
[2/2] ext3: fix rec_len overflow
- prevent rec_len from overflow with 64KB blocksize
Now on 64k page ppc64 box runs with this patch set we could create a 64k
block size ext3, and able to handle empty directory block.
Signed-off-by: Takashi Sato <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
Cc: <[email protected]>
Acked-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Convert bit_spin_lock to new locking bitops. Slub can use the non-atomic
store version to clear (Christoph?)
Signed-off-by: Nick Piggin <[email protected]>
Cc: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The return of the present "do {} while" based stub definition of
register_hotcpu_notifier() cannot be checked. This makes the stub
asymmetric w.r.t. the real HOTPLUG_CPU=y implementation that is
int-returning. So let us redefine this to be consistent with the full
version. Also do the same for unregister_hotcpu_notifier().
We cannot define these as static inline functions due to an existing GCC
bug (#33172). So define as macros that return appropriately instead (int
'0' for the register_hotcpu_notifier case and void for
unregister_hotcpu_notifier).
Signed-off-by: Satyam Sharma <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
This moves the new items to the end of the taskstats struct as
requested by Balbir and yourself.
Cc: Balbir Singh <[email protected]>
Cc: Jay Lan <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
This adds items to the taststats struct to account for user and system
time based on scaling the CPU frequency and instruction issue rates.
Adds account_(user|system)_time_scaled callbacks which architectures
can use to account for time using this mechanism.
Signed-off-by: Michael Neuling <[email protected]>
Cc: Balbir Singh <[email protected]>
Cc: Jay Lan <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Most of them are signedness, the rest unused function parameters.
Signed-off-by: Jiri Slaby <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The work done in bottom half doesn't cost much cpu time (e.g. tty_hangup
itself schedules its own bottom half), it's possible to do the work in isr
directly and save hence some .text.
Signed-off-by: Jiri Slaby <[email protected]>
Cc: Alan Cox <[email protected]>
Cc: Paul Fulghum <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The non-filesystem capability meaning of CAP_SETPCAP is that a process, p1,
can change the capabilities of another process, p2. This is not the
meaning that was intended for this capability at all, and this
implementation came about purely because, without filesystem capabilities,
there was no way to use capabilities without one process bestowing them on
another.
Since we now have a filesystem support for capabilities we can fix the
implementation of CAP_SETPCAP.
The most significant thing about this change is that, with it in effect, no
process can set the capabilities of another process.
The capabilities of a program are set via the capability convolution
rules:
pI(post-exec) = pI(pre-exec)
pP(post-exec) = (X(aka cap_bset) & fP) | (pI(post-exec) & fI)
pE(post-exec) = fE ? pP(post-exec) : 0
at exec() time. As such, the only influence the pre-exec() program can
have on the post-exec() program's capabilities are through the pI
capability set.
The correct implementation for CAP_SETPCAP (and that enabled by this patch)
is that it can be used to add extra pI capabilities to the current process
- to be picked up by subsequent exec()s when the above convolution rules
are applied.
Here is how it works:
Let's say we have a process, p. It has capability sets, pE, pP and pI.
Generally, p, can change the value of its own pI to pI' where
(pI' & ~pI) & ~pP = 0.
That is, the only new things in pI' that were not present in pI need to
be present in pP.
The role of CAP_SETPCAP is basically to permit changes to pI beyond
the above:
if (pE & CAP_SETPCAP) {
pI' = anything; /* ie., even (pI' & ~pI) & ~pP != 0 */
}
This capability is useful for things like login, which (say, via
pam_cap) might want to raise certain inheritable capabilities for use
by the children of the logged-in user's shell, but those capabilities
are not useful to or needed by the login program itself.
One such use might be to limit who can run ping. You set the
capabilities of the 'ping' program to be "= cap_net_raw+i", and then
only shells that have (pI & CAP_NET_RAW) will be able to run
it. Without CAP_SETPCAP implemented as described above, login(pam_cap)
would have to also have (pP & CAP_NET_RAW) in order to raise this
capability and pass it on through the inheritable set.
Signed-off-by: Andrew Morgan <[email protected]>
Signed-off-by: Serge E. Hallyn <[email protected]>
Cc: Stephen Smalley <[email protected]>
Cc: James Morris <[email protected]>
Cc: Casey Schaufler <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
After going through the kernels sysctl tables several times it has become
clear that code review and testing is just not effective in prevent
problematic sysctl tables from being used in the stable kernel. I certainly
can't seem to fix the problems as fast as they are introduced.
Therefore this patch adds sysctl_check_table which is called when a sysctl
table is registered and checks to see if we have a problematic sysctl table.
The biggest part of the code is the table of valid binary sysctl entries, but
since we have frozen our set of binary sysctls this table should not need to
change, and it makes it much easier to detect when someone unintentionally
adds a new binary sysctl value.
As best as I can determine all of the several hundred errors spewed on boot up
now are legitimate.
[[email protected]: kernel/sysctl_check.c must #include <linux/string.h>]
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Signed-off-by: Adrian Bunk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Grumble. These numbers should have been in sysctl.h from the beginning if we
ever expected anyone to use them. Oh well put them there now so we can find
them and make maintenance easier.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Samuel Ortiz <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The sysctl binary paths don't look as if they even code work, .data is not
filled in, and all of the proc_handlers look at extra1 and there is not
strategy routine.
So just kill the binary paths.
In addition this patch removes the setting of extra1 on directories. It
doesn't look like the parport code ever examines it, and it's bad sysctl form.
[[email protected]: remove parport_device_num()]
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Adrian Bunk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
There as been no easy way to wrap the default sysctl strategy routine except
for returning 0. Which is not always what we want. The few instances I have
seen that want different behaviour have written their own version of
sysctl_data. While not too hard it is unnecessary code and has the potential
for extra bugs.
So to make these situations easier and make that part of sysctl more symetric
I have factord sysctl_data out of do_sysctl_strategy and exported as a
function everyone can use.
Further having sysctl_data be an explicit function makes checking for badly
formed sysctl tables much easier.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
In sysctl.h the typedef struct ctl_table ctl_table violates coding style isn't
needed and is a bit of a nuisance because it makes it harder to recognize
ctl_table is a type name.
So this patch removes it from the generic sysctl code. Hopefully I will have
enough energy to send the rest of my patches will follow and to remove it from
the rest of the kernel.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Now that we have DMA_BIT_MASK(), these macros are pointless.
Cc: Jeremy Fitzhardinge <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Remove redundant DMA_..BIT_MASK definitions across two drivers. The
computation of the majority of the bitmasks is done by the compiler. The
initial split of the patch touching each a different file got removed due
to possible git bisect breakage.
Signed-off-by: Borislav Petkov <[email protected]>
Cc: Jeremy Fitzhardinge <[email protected]>
Cc: Muli Ben-Yehuda <[email protected]>
Cc: Jeff Garzik <[email protected]>
Cc: James Bottomley <[email protected]>
Reviewed-by: Satyam Sharma <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
On platforms that copy sys_tz into the vdso (currently only x86_64, soon to
include powerpc), it is possible for the vdso to get out of sync if a user
calls (admittedly unusual) settimeofday(NULL, ptr).
This patch adds a hook for architectures that set
CONFIG_GENERIC_TIME_VSYSCALL to ensure when sys_tz is updated they can also
updatee their copy in the vdso.
Signed-off-by: Tony Breeds <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Tony Luck <[email protected]>
Acked-by: John Stultz <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Hell knows what happened in commit 63b05203af57e7de4f3bb63b8b81d43bc196d32b
during 2.6.9 development. Commit introduced io_wait field which remained
write-only than and still remains write-only.
Also garbage collect macros which "use" io_wait.
Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The following scenario leads to total confusion of the platform firmware on
some boxes (eg. HPC nx6325):
* Hibernate with ACPI enabled
* Resume passing "acpi=off" to the boot kernel
To prevent this from happening it's necessary to check if ACPI is enabled (and
enable it if that's not the case) _right_ _after_ control has been transfered
from the boot kernel to the image kernel, before device_power_up() is called
(ie. with interrupts disabled). Enabling ACPI after calling
device_power_up() turns out to be insufficient.
For this reason, introduce new hibernation callback ->leave() that will be
executed before device_power_up() by the restored image kernel. To make it
work, it also is necessary to move swsusp_suspend() from swsusp.c to disk.c
(it's name is changed to "create_image", which is more up to the point).
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Currently, there's a CONFIG_DISABLE_CONSOLE_SUSPEND that allows one to stop
the serial console from being suspended when the rest of the machine goes
to sleep. This is incredibly useful for debugging power management-related
things; however, having it as a compile-time option has proved to be
incredibly inconvenient for us (OLPC). There are plenty of times that we
want serial console to not suspend, but for the most part we'd like serial
console to be suspended.
This drops CONFIG_DISABLE_CONSOLE_SUSPEND, and replaces it with a kernel
boot parameter (no_console_suspend). By default, the serial console will
be suspended along with the rest of the system; by passing
'no_console_suspend' to the kernel during boot, serial console will remain
alive during suspend.
For now, this is pretty serial console specific; further fixes could be
applied to make this work for things like netconsole.
Signed-off-by: Andres Salomon <[email protected]>
Acked-by: "Rafael J. Wysocki" <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Nigel Cunningham <[email protected]>
Cc: Russell King <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Introduce freezer-friendly wrappers around wait_event_interruptible() and
wait_event_interruptible_timeout(), originally defined in <linux/wait.h>, to
be used in freezable kernel threads. Make some of the freezable kernel
threads use them.
This is necessary for the freezer to stop sending signals to kernel threads,
which is implemented in the next patch.
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Nigel Cunningham <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Rename 'struct hibernation_ops' to 'struct platform_hibernation_ops' in
analogy with 'struct platform_suspend_ops'.
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
During hibernation we also need to tell the ACPI core that we're going to put
the system into the S4 sleep state. For this reason, an additional method in
'struct hibernation_ops' is needed, playing the role of set_target() in
'struct platform_suspend_operations'. Moreover, the role of the .prepare()
method is now different, so it's better to introduce another method, that in
general may be different from .prepare(), that will be used to prepare the
platform for creating the hibernation image (.prepare() is used anyway to
notify the platform that we're going to enter the low power state after the
image has been saved).
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The variable suspend_ops representing the set of global platform-specific
suspend-related operations, used by the PM core, need not be exported outside
of kernel/power/main.c . Make it static.
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
There is no reason why the .prepare() and .finish() methods in 'struct
platform_suspend_ops' should take any arguments, since architectures don't use
these methods' argument in any practically meaningful way (ie. either the
target system sleep state is conveyed to the platform by .set_target(), or
there is only one suspend state supported and it is indicated to the PM core
by .valid(), or .prepare() and .finish() aren't defined at all). There also
is no reason why .finish() should return any result.
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
The name of 'struct pm_ops' suggests that it is related to the power
management in general, but in fact it is only related to suspend. Moreover,
its name should indicate what this structure is used for, so it seems
reasonable to change it to 'struct platform_suspend_ops'. In that case, the
name of the global variable of this type used by the PM core and the names of
related functions should be changed accordingly.
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Move the definition of 'struct pm_ops' and related functions from <linux/pm.h>
to <linux/suspend.h> .
There are, at least, the following reasons to do that:
* 'struct pm_ops' is specifically related to suspend and not to the power
management in general.
* As long as 'struct pm_ops' is defined in <linux/pm.h>, any modification of it
causes the entire kernel to be recompiled, which is unnecessary and annoying.
* Some suspend-related features are already defined in <linux/suspend.h>, so it
is logical to move the definition of 'struct pm_ops' into there.
* 'struct hibernation_ops', being the hibernation-related counterpart of
'struct pm_ops', is defined in <linux/suspend.h> .
Signed-off-by: Rafael J. Wysocki <[email protected]>
Acked-by: Pavel Machek <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
|
|
Pull the copy_to_user out of hrtimer_nanosleep and into the callers
(common_nsleep, sys_nanosleep) in preparation for converting
compat_sys_nanosleep to use hrtimers.
Signed-off-by: Anton Blanchard <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
|
|
Short term, this works around a bug introduced by early sg-chaining
work.
Long term, removing this function eliminates a branch from a hot
path loop in each scatter/gather table build. Also, as this code
demonstrates, we don't need to _track_ the end of the s/g list, as
long as we mark it in some way. And doing so programatically is nice.
So its a useful cleanup, regardless of its short term effects.
Based conceptually on a quick patch by Jens Axboe.
Signed-off-by: Jeff Garzik <[email protected]>
|
|
schedstat is useful in investigating CPU scheduler behavior. Ideally,
I think it is beneficial to have it on all the time. However, the
cost of turning it on in production system is quite high, largely due
to number of events it collects and also due to its large memory
footprint.
Most of the fields probably don't need to be full 64-bit on 64-bit
arch. Rolling over 4 billion events will most like take a long time
and user space tool can be made to accommodate that. I'm proposing
kernel to cut back most of variable width on 64-bit system. (note,
the following patch doesn't affect 32-bit system).
Signed-off-by: Ken Chen <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
|
|
Macros like SCTP_CHUNKMAP_XXX(chukmap) require chukmap to be an array,
but match_packet() passes a pointer to these macros. Also remove the
ELEMCOUNT macro and fix a bug in SCTP_CHUNKMAP_COPY.
Signed-off-by: Li Zefan <[email protected]>
Signed-off-by: Patrick McHardy <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
[ This is kernel bugzilla 9174 "linux-2.6.23-git11 kernel panic" ]
The device in question is an IPv6-over-IPv4 tunnel, which doesn't have
any header_ops, so the crash happens in dev_parse_header when
dereferencing them.
Signed-off-by: Patrick McHardy <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Filter is attached in a separate function, so do the
same for filter detaching.
This also removes one variable sock_setsockopt().
Signed-off-by: Pavel Emelyanov <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Also of_unregister_driver. These will be shortly also used by the
PowerPC code.
Signed-off-by: Stephen Rothwell <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
|
|
Some drivers with shared NAPI need a synchronization barrier.
Also suggested by Benjamin Herrenschmidt for EMAC.
Signed-off-by: Stephen Hemminger <[email protected]>
Signed-off-by: Jeff Garzik <[email protected]>
|
|
Convert ext4_extent_idx.ei_leaf ext4_extent_idx.ei_leaf_lo
This helps in finding BUGs due to direct partial access of
these split 48 bit values.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
Convert ext4_extent.ee_start to ext4_extent.ee_start_lo
This helps in finding BUGs due to direct partial access of
these split 48 bit values
Also fix direct partial access in ext4 code
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
Convert s_r_blocks_count and s_free_blocks_count to
s_r_blocks_count_lo and s_free_blocks_count_lo
This helps in finding BUGs due to direct partial access of
these split 64 bit values
Also fix direct partial access in ext4 code
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
|
|
Convert s_blocks_count to s_blocks_count_lo
This helps in finding BUGs due to direct partial access of
these split 64 bit values
Also fix direct partial access in ext4 code
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
Convert bg_inode_bitmap and bg_inode_table to bg_inode_bitmap_lo
and bg_inode_table_lo. This helps in finding BUGs due to
direct partial access of these split 64 bit values
Also fix one direct partial access
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
Convert bg_block_bitmap to bg_block_bitmap_lo
This helps in catching some BUGS due to direct
partial access of these split fields.
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
This feature relaxes check restrictions on where each block groups meta
data is located within the storage media. This allows for the allocation
of bitmaps or inode tables outside the block group boundaries in cases
where bad blocks forces us to look for new blocks which the owning block
group can not satisfy. This will also allow for new meta-data allocation
schemes to improve performance and scalability.
Signed-off-by: Jose R. Santos <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
|
|
Signed-off-by: Aneesh Kumar K.V <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
|
|
In pass1 of e2fsck, every inode table in the fileystem is scanned and checked,
regardless of whether it is in use. This is this the most time consuming part
of the filesystem check. The unintialized block group feature can greatly
reduce e2fsck time by eliminating checking of uninitialized inodes.
With this feature, there is a a high water mark of used inodes for each block
group. Block and inode bitmaps can be uninitialized on disk via a flag in the
group descriptor to avoid reading or scanning them at e2fsck time. A checksum
of each group descriptor is used to ensure that corruption in the group
descriptor's bit flags does not cause incorrect operation.
The feature is enabled through a mkfs option
mke2fs /dev/ -O uninit_groups
A patch adding support for uninitialized block groups to e2fsprogs tools has
been posted to the linux-ext4 mailing list.
The patches have been stress tested with fsstress and fsx. In performance
tests testing e2fsck time, we have seen that e2fsck time on ext3 grows
linearly with the total number of inodes in the filesytem. In ext4 with the
uninitialized block groups feature, the e2fsck time is constant, based
solely on the number of used inodes rather than the total inode count.
Since typical ext4 filesystems only use 1-10% of their inodes, this feature can
greatly reduce e2fsck time for users. With performance improvement of 2-20
times, depending on how full the filesystem is.
The attached graph shows the major improvements in e2fsck times in filesystems
with a large total inode count, but few inodes in use.
In each group descriptor if we have
EXT4_BG_INODE_UNINIT set in bg_flags:
Inode table is not initialized/used in this group. So we can skip
the consistency check during fsck.
EXT4_BG_BLOCK_UNINIT set in bg_flags:
No block in the group is used. So we can skip the block bitmap
verification for this group.
We also add two new fields to group descriptor as a part of
uninitialized group patch.
__le16 bg_itable_unused; /* Unused inodes count */
__le16 bg_checksum; /* crc16(sb_uuid+group+desc) */
bg_itable_unused:
If we have EXT4_BG_INODE_UNINIT not set in bg_flags
then bg_itable_unused will give the offset within
the inode table till the inodes are used. This can be
used by fsck to skip list of inodes that are marked unused.
bg_checksum:
Now that we depend on bg_flags and bg_itable_unused to determine
the block and inode usage, we need to make sure group descriptor
is not corrupt. We add checksum to group descriptor to
detect corruption. If the descriptor is found to be corrupt, we
mark all the blocks and inodes in the group used.
Signed-off-by: Avantika Mathur <[email protected]>
Signed-off-by: Andreas Dilger <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
Signed-off-by: Aneesh Kumar K.V <[email protected]>
|
|
CONFIG_EXT4_INDEX is not an exposed config option in the kernel, and it is
unconditionally defined in ext4_fs.h. tune2fs is already able to turn off
dir indexing, so at this point it's just cluttering up the code. Remove
it.
Signed-off-by: Eric Sandeen <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
|
|
Fragment support in ext2/3/4 was never implemented, and it probably will
never be implemented. So remove it from ext4.
Signed-off-by: Coly Li <[email protected]>
Acked-by: Andreas Dilger <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
|
|
change JBD_XXX macros to JBD2_XXX in JBD2/Ext4
Signed-off-by: Mingming Cao <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
|
|
This patch cleans up jbd_kmalloc and replace it with kmalloc directly
Signed-off-by: Mingming Cao <[email protected]>
|
|
This patch cleans up jbd_kmalloc and replace it with kmalloc directly
Signed-off-by: Mingming Cao <[email protected]>
|