The check to whom a device is reserved is done by checking the path
state of the affected channel paths. If it turns out that one path is
flagged as reserved by someone else the whole device is marked as such.
However the meaning of the RESVD_ELSE bit is that the addressed device
is reserved to a different pathgroup (and not reserved to a different
LPAR). If we do this test on a path which is currently not a member of
the pathgroup we could erroneously mark the device as reserved to
someone else.
To fix this collect the reserved state for all potential members of the
pathgroup and only mark the device as reserved if all of those potential
members have the RESVD_ELSE bit set.
Acked-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com> Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Prior to memory slot sorting this loop compared all of the user memory
slots for overlap with new entries. With memory slot sorting, we're
just checking some number of entries in the array that may or may not
be user slots. Instead, walk all the slots with kvm_for_each_memslot,
which has the added benefit of terminating early when we hit the first
empty slot, and skip comparison to private slots.
The locking in update_vsyscall_tz() is not only unnecessary because the vdso
code copies the data unproteced in __kernel_gettimeofday() but also
introduces a hard to reproduce race condition between update_vsyscall()
and update_vsyscall_tz(), which causes user space process to loop
forever in vdso code.
The following patch removes the locking from update_vsyscall_tz().
Locking is not only unnecessary because the vdso code copies the data
unprotected in __kernel_gettimeofday() but also erroneous because updating
the tb_update_count is not atomic and introduces a hard to reproduce race
condition between update_vsyscall() and update_vsyscall_tz(), which further
causes user space process to loop forever in vdso code.
The below scenario describes the race condition,
x==0 Boot CPU other CPU
proc_P: x==0
timer interrupt
update_vsyscall
x==1 x++;sync settimeofday
update_vsyscall_tz
x==2 x++;sync
x==3 sync;x++
sync;x++
proc_P: x==3 (loops until x becomes even)
Because the ++ operator would be implemented as three instructions and not
atomic on powerpc.
A similar change was made for x86 in commit 6c260d58634
("x86: vdso: Remove bogus locking in update_vsyscall_tz")
Signed-off-by: Shan Hai <shan.hai@windriver.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If we build a kernel with CONFIG_RELOCATABLE=y CONFIG_CRASH_DUMP=n,
the kernel fails when we run at a non zero offset. It turns out
we were incorrectly wrapping some of the relocatable kernel code
with CONFIG_CRASH_DUMP.
Signed-off-by: Anton Blanchard <anton@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
I was chasing down a bug of random validity intercepts on s390.
(guest prefix page not mapped in the host virtual aspace). Turns out
that the problem was a wrong address space control element. The
cause was quite complex:
During paging activity a DAT protection during SIE caused a program
interrupt. Normally, the sie retry loop tries to catch all
interrupts during and shortly before sie to rerun the setup. The
problem is now that protection causes a suppressing program interrupt,
causing the PSW to point to the instruction AFTER SIE in case of DAT
protection. This confused the logic of the retry loop to not trigger,
instead we jumped directly back to SIE after return from
the program interrupt. (the protection fault handler itself did
a rewind of the psw). This usually works quite well, but:
If now the protection fault handler has to wait, another program
might be scheduled in. Later on the sie process will be schedules
in again. In that case the content of CR1 (primary address space)
will be wrong because switch_to will put the user space ASCE into CR1
and not the guest ASCE.
In addition the program parameter is also wrong for every protection
fault of a guest, since we dont issue the SPP instruction.
So lets also check for PSW == instruction after SIE in the program
check handler. Instead of expensively checking all program
interruption codes that might be suppressing we assume that a program
interrupt pointing after SIE was always a program interrupt in SIE.
(Otherwise we have a kernel bug anyway).
We also have to compensate the rewinding, since the C-level handlers
will do that. Therefore we need to add a nop with the same length
as SIE before the sie_loop.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> CC: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Newer kernels (linux-next with the transparent huge page patches)
use rrbm if the feature is announced via feature bit 66.
RRBM will cause intercepts, so KVM does not handle it right now,
causing an illegal instruction in the guest.
The easy solution is to disable the feature bit for the guest.
Reported-by: Stefan Lippers-Hollmann <s.l-h@gmx.de> Tested-by: Stefan Lippers-Hollmann <s.l-h@gmx.de> Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com> Acked-by: Gertjan van Wingerde <gwingerde@gmail.com> Acked-by: Stanislaw Gruszka <sgruszka@redhat.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We were using wrong IRQ number so clearing wasn't working at all.
Depending on a platform this could result in a one device having two
interrupts assigned. On BCM4706 this resulted in all IRQs being broken.
We know that we have issues with the fw in the reclaim path.
This is why iwl_reclaim doesn't complain too loud when it
happens since it is recoverable. Somehow, the caller of
iwl_reclaim however WARNed when it happens. This doesn't
make any sense.
When I digged into the history of that code, I discovered
that this bug occurs only when we receive a BA notification.
So move the W/A in the BA notification handling code where
it was before.
This patch addresses:
http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2387
This can lead to a panic if the driver isn't ready to
handle them. Since our interrupt line is shared, we can get
an interrupt at any time (and CONFIG_DEBUG_SHIRQ checks
that even when the interrupt is being freed).
If the op_mode has gone away, we musn't call it. To avoid
this the transport disables the interrupts when the hw is
stopped and the op_mode is leaving.
If there is an event that would cause an interrupt the INTA
register is updated regardless of the enablement of the
interrupts: even if the interrupts are disabled, the INTA
will be changed, but the device won't issue an interrupt.
But the ISR can be called at any time, so we ought ignore
the value in the INTA otherwise we can call the op_mode
after it was freed.
I found this bug when the op_mode_start failed, and called
iwl_trans_stop_hw(trans, true). Then I played with the
RFKILL button, and removed the module.
While removing the module, the IRQ is freed, and the ISR is
called (CONFIG_DEBUG_SHIRQ enabled). Panic.
ieee80211_free_txskb() needs to be used instead of dev_kfree_skb_any for
tx packets passed to the driver from mac80211
Signed-off-by: Felix Fietkau <nbd@openwrt.org> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Oliver reported that commit cd60042c caused his cifs mounts to
continually thrash through new inodes on readdir. His servers are not
sending inode numbers (or he's not using them), and the new test in
that function doesn't account for that sort of setup correctly.
If we're not using server inode numbers, then assume that the inode
attached to the dentry hasn't changed. Go ahead and update the
attributes in place, but keep the same inode number.
Reported-and-Tested-by: Oliver Mössinger <Oliver.Moessinger@ichaus.de> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The caller doesn't do anything with the dentry, so there's no point in
holding a reference to it on return. Also cifs_prime_dcache better
describes the actual purpose of the function.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <smfrench@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the restart timer is running due to BUS-OFF and the device is
disconnected an dev_put will decrease the usage counter to -1 thus
blocking the interface removal, resulting in the following dmesg
lines repeating every 10s:
can: notifier: receive list not found for dev can0
can: notifier: receive list not found for dev can0
can: notifier: receive list not found for dev can0
unregister_netdevice: waiting for can0 to become free. Usage count = -1
Signed-off-by: Alexander Stein <alexander.stein@systec-electronic.com> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 0a97e1e9f9a6 ('HID: apple: Add Apple wireless keyboard 2011 ANSI PID')
did not update the special driver list in hid-core.c, so hid-generic may
still bind to this device.
Reported-by: Ari Pollak <ari@scvngr.com>
References: http://bugs.debian.org/694546 Signed-off-by: Ben Hutchings <ben@decadent.org.uk> Signed-off-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 263a523d18bc ("linux/kernel.h: Fix warning seen with W=1 due to
change in DIV_ROUND_CLOSEST") fixes a warning seen with W=1 due to
change in DIV_ROUND_CLOSEST.
Unfortunately, the C compiler converts divide operations with unsigned
divisors to unsigned, even if the dividend is signed and negative (for
example, -10 / 5U = 858993457). The C standard says "If one operand has
unsigned int type, the other operand is converted to unsigned int", so
the compiler is not to blame. As a result, DIV_ROUND_CLOSEST(0, 2U) and
similar operations now return bad values, since the automatic conversion
of expressions such as "0 - 2U/2" to unsigned was not taken into
account.
Fix by checking for the divisor variable type when deciding which
operation to perform. This fixes DIV_ROUND_CLOSEST(0, 2U), but still
returns bad values for negative dividends divided by unsigned divisors.
Mark the latter case as unsupported.
One observed effect of this problem is that the s2c_hwmon driver reports
a value of 4198403 instead of 0 if the ADC reads 0.
Other impact is unpredictable. Problem is seen if the divisor is an
unsigned variable or constant and the dividend is less than (divisor/2).
Since commit e303297e6c3a ("mm: extended batches for generic
mmu_gather") we are batching pages to be freed until either
tlb_next_batch cannot allocate a new batch or we are done.
This works just fine most of the time but we can get in troubles with
non-preemptible kernel (CONFIG_PREEMPT_NONE or CONFIG_PREEMPT_VOLUNTARY)
on large machines where too aggressive batching might lead to soft
lockups during process exit path (exit_mmap) because there are no
scheduling points down the free_pages_and_swap_cache path and so the
freeing can take long enough to trigger the soft lockup.
The lockup is harmless except when the system is setup to panic on
softlockup which is not that unusual.
The simplest way to work around this issue is to limit the maximum
number of batches in a single mmu_gather. 10k of collected pages should
be safe to prevent from soft lockups (we would have 2ms for one) even if
they are all freed without an explicit scheduling point.
This patch doesn't add any new explicit scheduling points because it
relies on zap_pmd_range during page tables zapping which calls
cond_resched per PMD.
The following lockup has been reported for 3.0 kernel with a huge
process (in order of hundreds gigs but I do know any more details).
tm_mon is 0..11, whereas vt8500 expects 1..12 for the month field,
causing invalid date errors for January, and causing the day field to
roll over incorrectly.
The century flag is only handled in vt8500_rtc_read_time, but not set in
vt8500_rtc_set_time. This patch corrects the behaviour of the century
flag.
Signed-off-by: Edgar Toernig <froese@gmx.de> Signed-off-by: Tony Prisk <linux@prisktech.co.nz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Control register bitfield for 12H/24H mode is handled incorrectly.
Setting CR_24H actually enables 12H mode. This patch renames the define
and changes the initialization code to correctly set 24H mode.
Signed-off-by: Tony Prisk <linux@prisktech.co.nz> Cc: Edgar Toernig <froese@gmx.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We already perform the ida_simple_remove() in rtc_device_release(),
which is an appropriate place. Commit 2830a6d20 ("rtc: recycle id when
unloading a rtc driver") caused the kernel to emit
ida_remove called for id=0 which is not allocated.
warnings when rtc_device_release() tries to release an alread-released
ID.
Let's restore things to their previous state and then work out why
Vincent's kernel wasn't calling rtc_device_release() - presumably a bug
in a specific sub-driver.
Reported-by: Lothar Waßmann <LW@KARO-electronics.de> Acked-by: Alexander Holler <holler@ahsoftware.de> Cc: Vincent Palatin <vpalatin@chromium.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The inb/outb macros for CRIS are broken from a number of points of view,
missing () around parameters and they have an unprotected if statement
in them. This was breaking the compile of IPMI on CRIS and thus I was
being annoyed by build regressions, so I fixed them.
Plus I don't think they would have worked at all, since the data values
were missing "&" and the outsl had a "3" instead of a "4" for the size.
From what I can tell, this stuff is not used at all, so this can't be
any more broken than it was before, anyway.
We display a list of supplementary group for each process in
/proc/<pid>/status. However, we show only the first 32 groups, not all of
them.
Although this is rare, but sometimes processes do have more than 32
supplementary groups, and this kernel limitation breaks user-space apps
that rely on the group list in /proc/<pid>/status.
Number 32 comes from the internal NGROUPS_SMALL macro which defines the
length for the internal kernel "small" groups buffer. There is no
apparent reason to limit to this value.
This patch removes the 32 groups printing limit.
The Linux kernel limits the amount of supplementary groups by NGROUPS_MAX,
which is currently set to 65536. And this is the maximum count of groups
we may possibly print.
The atomic64 library uses a handful of static spin locks to implement
atomic 64-bit operations on architectures without support for atomic
64-bit instructions.
Unfortunately, the spinlocks are initialized in a pure initcall and that
is too late for the vfs namespace code which wants to use atomic64
operations before the initcall is run.
This became a problem as of commit 8823c079ba71: "vfs: Add setns support
for the mount namespace".
Commit 284f5f9 was intended to disable the "only_one_child()" optimization
on Stratus ftServer systems, but its DMI check is wrong. It looks for
DMI_SYS_VENDOR that contains "ftServer", when it should look for
DMI_SYS_VENDOR containing "Stratus" and DMI_PRODUCT_NAME containing
"ftServer".
Ulrich reported that his USB3 cardreader does not work reliably when
connected to the USB3 port. It turns out that USB3 controller failed to
awaken when plugging in the USB3 cardreader. Further experiments found
that the USB3 host controller can only be awakened via polling, not via PME
interrupt. But if the PCIe port to which the USB3 host controller is
connected is suspended, we cannot poll the controller because its config
space is not accessible when the PCIe port is in a low power state.
To solve the issue, the PCIe port will not be suspended if any subordinate
device needs PME polling.
[bhelgaas: use bool consistently rather than mixing int/bool]
Reference: http://lkml.kernel.org/r/50841CCC.9030809@uli-eckhardt.de Reported-by: Ulrich Eckhardt <usb@uli-eckhardt.de> Tested-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
- Always in D0 state, because some devices do not work again after
being put into D3 by the PCI bus.
- In SUSPENDED state if allowed, so that the parent devices can still
be put into low power state.
To satisfy these requirements, the runtime PM for the unbound PCI
devices are disabled and set to SUSPENDED state. One issue of this
solution is that the PCI devices will be put into SUSPENDED state even
if the SUSPENDED state is forbidden via the sysfs interface
(.../power/control) of the device. This is not an issue for most
devices, because most PCI devices are not used at all if unbound.
But there are exceptions. For example, unbound VGA card can be used
for display, but suspending its parents makes it stop working.
To fix the issue, we keep the runtime PM enabled when the PCI devices
are unbound. But the runtime PM callbacks will do nothing if the PCI
devices are unbound. This way, we can put the PCI devices into
SUSPENDED state without putting the PCI devices into D3 state.
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=48201 Signed-off-by: Huang Ying <ying.huang@intel.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Acked-by: Alan Stern <stern@rowland.harvard.edu> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We weren't clearing card->tx_skb[port] when processing the TX done interrupt.
If there wasn't another skb ready to transmit immediately, this led to a
double-free because we'd free it *again* next time we did have a packet to
send.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a call goes out, the signing code adjusts the sequence number
upward by two to account for the request and the response. An NT_CANCEL
however doesn't get a response of its own, it just hurries the server
along to get it to respond to the original request more quickly.
Therefore, we must adjust the sequence number back down by one after
signing a NT_CANCEL request.
Reported-by: Tim Perry <tdparmor-sambabugs@yahoo.com> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <smfrench@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This is likely due to a race vs. a reconnection event.
The current code checks for a NULL socket in smb_send_kvec, but that's
too late. By the time that check is done, the socket will already have
been passed to kernel_setsockopt. Move the check into smb_send_rqst, so
that it's checked earlier.
In truth, this is a bit of a half-assed fix. The -ENOTSOCK error
return here looks like it could bubble back up to userspace. The locking
rules around the ssocket pointer are really unclear as well. There are
cases where the ssocket pointer is changed without holding the srv_mutex,
but I'm not clear whether there's a potential race here yet or not.
This code seems like it could benefit from some fundamental re-think of
how the socket handling should behave. Until then though, this patch
should at least fix the above oops in most cases.
Reported-and-Tested-by: CAI Qian <caiqian@redhat.com> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <smfrench@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
File descriptors (even those for writing) do not hold freeze protection.
Thus mark_files_ro() must call __mnt_drop_write() to only drop protection
against remount read-only. Calling mnt_drop_write_file() as we do now
results in:
[ BUG: bad unlock balance detected! ] 3.7.0-rc6-00028-g88e75b6 #101 Not tainted
-------------------------------------
kworker/1:2/79 is trying to release lock (sb_writers) at:
[<ffffffff811b33b4>] mnt_drop_write+0x24/0x30
but there are no more locks to release!
Reported-by: Zdenek Kabelac <zkabelac@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
flush_cache_louis flushes the D-side caches to the point of unification
inner-shareable. On uniprocessor CPUs, this is defined as zero and
therefore no flushing will take place. Rather than invent a new interface
for UP systems, instead use our SMP_ON_UP patching code to read the
LoUU from the CLIDR instead.
Cc: Lorenzo Pieralisi <Lorenzo.Pieralisi@arm.com> Tested-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 34ae6c96a6a7 ("ARM: 7298/1: realview: fix mapping of MPCore
private memory region") accidentally broke the definition for the base
address of the private peripheral region on revision B Realview-EB
boards.
This patch uses the correct address for REALVIEW_EB11MP_PRIV_MEM_BASE.
Acked-by: Marc Zyngier <marc.zyngier@arm.com> Tested-by: Florian Fainelli <florian@openwrt.org> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
find_vma() is *not* safe when somebody else is removing vmas. Not just
the return value might get bogus just as you are getting it (this instance
doesn't try to dereference the resulting vma), the search itself can get
buggered in rather spectacular ways. IOW, ->mmap_sem really, really is
not optional here.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When updating the page protection map after calculating the user_pgprot
value, the base protection map is temporarily stored in an unsigned long
type, causing truncation of the protection bits when LPAE is enabled.
This effectively means that calls to mprotect() will corrupt the upper
page attributes, clearing the XN bit unconditionally.
This patch uses pteval_t to store the intermediate protection values,
preserving the upper bits for 64-bit descriptors.
Acked-by: Nicolas Pitre <nico@linaro.org> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Will Deacon <will.deacon@arm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When we fail to get a dquot lock during reclaim, we jump to an error
handler that unlocks the dquot. This is wrong as we didn't lock the
dquot, and unlocking it means who-ever is holding the lock has had
it silently taken away, and hence it results in a lock imbalance.
Found by inspection while modifying the code for the numa-lru
patchset. This fixes a random hang I've been seeing on xfstest 232
for the past several months.
Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ben Myers <bpm@sgi.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The direct IO path can do a nested transaction reservation when
writing past the EOF. The first transaction is the append
transaction for setting the filesize at IO completion, but we can
also need a transaction for allocation of blocks. If the log is low
on space due to reservations and small log, the append transaction
can be granted after wating for space as the only active transaction
in the system. This then attempts a reservation for an allocation,
which there isn't space in the log for, and the reservation sleeps.
The result is that there is nothing left in the system to wake up
all the processes waiting for log space to come free.
The stack trace that shows this deadlock is relatively innocuous:
This was discovered on a filesystem with a log of only 10MB, and a
log stripe unit of 256k whih increased the base reservations by
512k. Hence a allocation transaction requires 1.2MB of log space to
be available instead of only 260k, and so greatly increased the
chance that there wouldn't be enough log space available for the
nested transaction to succeed. The key to reproducing it is this
mkfs command:
The test case was a 1000 fsstress processes running with random
freeze and unfreezes every few seconds. Thanks to Eryu Guan
(eguan@redhat.com) for writing the test that found this on a system
with a somewhat unique default configuration....
Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Andrew Dahl <adahl@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 77097ae503b1 ("most of set_current_blocked() callers want
SIGKILL/SIGSTOP removed from set") removed the initialization of newmask
by accident, causing ltp to complain like this:
print_prefix() passes a NULL buf to print_time() to get the length of
the time prefix; when printk times are enabled, the current code just
returns the constant 15, which matches the format "[%5lu.%06lu] " used
to print the time value. However, this is obviously incorrect when the
whole seconds part of the time gets beyond 5 digits (100000 seconds is a
bit more than a day of uptime).
The simple fix is to use snprintf(NULL, 0, ...) to calculate the actual
length of the time prefix. This could be micro-optimized but it seems
better to have simpler, more readable code here.
The bug leads to the syslog system call miscomputing which messages fit
into the userspace buffer. If there are enough messages to fill
log_buf_len and some have a timestamp >= 100000, dmesg may fail with:
# dmesg
klogctl: Bad address
When this happens, strace shows that the failure is indeed EFAULT due to
the kernel mistakenly accessing past the end of dmesg's buffer, since
dmesg asks the kernel how big a buffer it needs, allocates a bit more,
and then gets an error when it asks the kernel to fill it:
As far as I can see, the bug has been there as long as print_time(),
which comes from commit 084681d14e42 ("printk: flush continuation lines
immediately to console") in 3.5-rc5.
Signed-off-by: Roland Dreier <roland@purestorage.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Joe Perches <joe@perches.com> Cc: Sylvain Munaut <s.munaut@whatever-company.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 35f9c09fe9c72e (tcp: tcp_sendpages() should call tcp_push() once)
added an internal flag : MSG_SENDPAGE_NOTLAST meant to be set on all
frags but the last one for a splice() call.
The condition used to set the flag in pipe_to_sendpage() relied on
splice() user passing the exact number of bytes present in the pipe,
or a smaller one.
But some programs pass an arbitrary high value, and the test fails.
The effect of this bug is a lack of tcp_push() at the end of a
splice(pipe -> socket) call, and possibly very slow or erratic TCP
sessions.
We should both test sd->total_len and fact that another fragment
is in the pipe (pipe->nrbufs > 1)
Many thanks to Willy for providing very clear bug report, bisection
and test programs.
Reported-by: Willy Tarreau <w@1wt.eu> Bisected-by: Willy Tarreau <w@1wt.eu> Tested-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 24cb81a6a (sctp: Push struct net down into all of the
state machine functions) introduced the net structure into all
state machine functions, but jsctp_sf_eat_sack was not updated,
hence when SCTP association probing is enabled in the kernel,
any simple SCTP client/server program from userspace will panic
the kernel.
Cc: Vlad Yasevich <vyasevich@gmail.com> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Acked-by: Vlad Yasevich <vyasevich@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Stefan Hasko <hasko.stevo@gmail.com> Acked-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When using nanosleep() in an userspace application we get a
ratelimit warning
NOHZ: local_softirq_pending 08
for 10 times.
This patch replaces netif_rx() with netif_rx_ni() which has
to be used from process/softirq context.
The process/softirq context will be called from fakelb driver.
Signed-off-by: Alexander Aring <alex.aring@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In function ndisc_redirect_rcv(), the skb->data points to the transport
header, but function icmpv6_notify() need the skb->data points to the
inner IP packet. So before using icmpv6_notify() to propagate redirect,
change skb->data to point the inner IP packet that triggered the sending
of the Redirect, and introduce struct rd_msg to make it easy.
Signed-off-by: Duan Jiong <djduanjiong@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This happens, because sk_clone_lock initializes sk_refcnt to 2, and thus
a single sock_put() is not enough to free the memory. Additionally, things
like xfrm, memcg, cookie_values,... may have been initialized.
We have to free them properly.
This is fixed by forcing a call to tcp_done(), ending up in
inet_csk_destroy_sock, doing the final sock_put(). tcp_done() is necessary,
because it ends up doing all the cleanup on xfrm, memcg, cookie_values,
xfrm,...
Before calling tcp_done, we have to set the socket to SOCK_DEAD, to
force it entering inet_csk_destroy_sock. To avoid the warning in
inet_csk_destroy_sock, inet_num has to be set to 0.
As inet_csk_destroy_sock does a dec on orphan_count, we first have to
increase it.
Calling tcp_done() allows us to remove the calls to
tcp_clear_xmit_timer() and tcp_cleanup_congestion_control().
A similar approach is taken for dccp by calling dccp_done().
This is in the kernel since 093d282321 (tproxy: fix hash locking issue
when using port redirection in __inet_inherit_port()), thus since
version >= 2.6.37.
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
But it actually gets 'orig_interval' or 'orig_interval - BATADV_JITTER'
because '%' and '*' have same precedence and associativity is
left-to-right.
This adds the parentheses at the appropriate position so that it matches
original intension.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com> Acked-by: Antonio Quartulli <ordex@autistici.org> Cc: Marek Lindner <lindner_marek@yahoo.de> Cc: Simon Wunderlich <siwu@hrz.tu-chemnitz.de> Cc: Antonio Quartulli <ordex@autistici.org> Cc: b.a.t.m.a.n@lists.open-mesh.org Cc: "David S. Miller" <davem@davemloft.net> Cc: netdev@vger.kernel.org Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The basic scheme of the block mode assembler is that we start by
enabling the FPU, loading the key into the floating point registers,
then iterate calling the encrypt/decrypt routine for each block.
For the 256-bit key cases, we run short on registers in the unrolled
loops.
So the {ENCRYPT,DECRYPT}_256_2() macros reload the key registers that
get clobbered.
The unrolled macros, {ENCRYPT,DECRYPT}_256(), are not mindful of this.
So if we have a mix of multi-block and single-block calls, the
single-block unrolled 256-bit encrypt/decrypt can run with some
of the key registers clobbered.
Handle this by always explicitly loading those registers before using
the non-unrolled 256-bit macro.
This was discovered thanks to all of the new test cases added by
Jussi Kivilinna.
Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Modifying the huge pte's requires that all the underlying pte's be
modified.
Version 2: added missing flush_tlb_page()
Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: sparclinux@vger.kernel.org Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A task is considered frozen enough between freezer_do_not_count() and
freezer_count() and freezers use freezer_should_skip() to test this
condition. This supposedly works because freezer_count() always calls
try_to_freezer() after clearing %PF_FREEZER_SKIP.
However, there currently is nothing which guarantees that
freezer_count() sees %true freezing() after clearing %PF_FREEZER_SKIP
when freezing is in progress, and vice-versa. A task can escape the
freezing condition in effect by freezer_count() seeing !freezing() and
freezer_should_skip() seeing %PF_FREEZER_SKIP.
This patch adds smp_mb()'s to freezer_count() and
freezer_should_skip() such that either %true freezing() is visible to
freezer_count() or !PF_FREEZER_SKIP is visible to
freezer_should_skip().
Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We were checking incorrectly if signatures were required to be sent,
so were always sending signatures after the initial session establishment.
For SMB3 mounts (vers=3.0) this was a problem because we were putting
SMB2 signatures in SMB3 requests which would cause access denied
on mount (the tree connection would fail).
This might also be worth considering for stable (for 3.7), as the
error message on mount (access denied) is confusing to users and
there is no workaround if the server is configured to only
support smb3.0. I am ok either way.
Signed-off-by: Steve French <smfrench@gmail.com> Reviewed-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Build kernel with CONFIG_HUGETLBFS=y,CONFIG_HUGETLB_PAGE=y and
CONFIG_CGROUP_HUGETLB=y, then specify hugepagesz=xx boot option, system
will fail to boot.
This failure is caused by following code path:
setup_hugepagesz
hugetlb_add_hstate
hugetlb_cgroup_file_init
cgroup_add_cftypes
kzalloc <--slab is *not available* yet
For this path, slab is not available yet, so memory allocated will be
failed, and cause WARN_ON() in hugetlb_cgroup_file_init().
So I move hugetlb_cgroup_file_init() into hugetlb_init().
in cgroup_add_file,when creating files for cgroup,
some of creation may be skipped. So we need to avoid
deleting these uncreated files in cgroup_rm_file,
otherwise the warning msg will be triggered.
"cgroup_addrm_files: failed to remove memory_pressure_enabled, err=-2"
cgroup_create_dir() does weird dancing with dentry refcnt. On
success, it gets and then puts it achieving nothing. On failure, it
puts but there isn't no matching get anywhere leading to the following
oops if cgroup_create_file() fails for whatever reason.
cgroup core has a bug which violates a basic rule about event
notifications - when a new entity needs to be added, you add that to
the notification list first and then make the new entity conform to
the current state. If done in the reverse order, an event happening
inbetween will be lost.
cgroup_subsys->fork() is invoked way before the new task is added to
the css_set. Currently, cgroup_freezer is the only user of ->fork()
and uses it to make new tasks conform to the current state of the
freezer. If FROZEN state is requested while fork is in progress
between cgroup_fork_callbacks() and cgroup_post_fork(), the child
could escape freezing - the cgroup isn't frozen when ->fork() is
called and the freezer couldn't see the new task on the css_set.
This patch moves cgroup_subsys->fork() invocation to
cgroup_post_fork() after the new task is added to the css_set.
cgroup_fork_callbacks() is removed.
Because now a task may be migrated during cgroup_subsys->fork(),
freezer_fork() is updated so that it adheres to the usual RCU locking
and the rather pointless comment on why locking can be different there
is removed (if it doesn't make anything simpler, why even bother?).
Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch should be backported to stable kernels as old as 3.0,
that contain commit 69e848c2090aebba5698a1620604c7dccb448684
"Intel xhci: Support EHCI/xHCI port switching."
Signed-off-by: Russell Webb <russell.webb@linux.intel.com> Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This minor patch creates a more stricter conditional for the Z1 sytems for applying
the Compliance Mode Patch, this to avoid the quirk to be applied to models that
contain a "Z1" in their dmi product string but are different from Z1 systems.
This patch should be backported to stable kernels as old as 3.2, that
contain the commit 71c731a296f1b08a3724bd1b514b64f1bda87a23 "usb: host:
xhci: Fix Compliance Mode on SN65LVPE502CP Hardware"
Signed-off-by: Alexis R. Cortes <alexis.cortes@ti.com> Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch (as1631) fixes a bug that shows up when a config change
fails for a device under an xHCI controller. The controller needs to
be told to disable the endpoints that have been enabled for the new
config. The existing code does this, but before storing the
information about which endpoints were enabled! As a result, any
second attempt to install the new config is doomed to fail because
xhci-hcd will refuse to enable an endpoint that is already enabled.
The patch optimistically initializes the new endpoints' device
structures before asking the device to switch to the new config. If
the request fails then the endpoint information is already stored, so
we can use usb_hcd_alloc_bandwidth() to disable the endpoints with no
trouble. The rest of the error path is slightly more complex now; we
have to disable the new interfaces and call put_device() rather than
simply deallocating them.
Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Reported-and-tested-by: Matthias Schniedermeyer <ms@citd.de> CC: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci_alloc_segments_for_ring() builds a list of xhci_segments and links
the tail to head at the end (forming a ring). When it bails out for OOM
reasons half-way through, it tries to destroy its half-built list with
xhci_free_segments_for_ring(), even though it is not a ring yet. This
causes a null-pointer dereference upon hitting the last element.
Furthermore, one of its callers (xhci_ring_alloc()) mistakenly believes
the output parameters to be valid upon this kind of OOM failure, and
calls xhci_ring_free() on them. Since the (incomplete) list/ring should
already be destroyed in that case, this would lead to a use after free.
This patch fixes those issues by having xhci_alloc_segments_for_ring()
destroy its half-built, non-circular list manually and destroying the
invalid struct xhci_ring in xhci_ring_alloc() with a plain kfree().
This patch should be backported to kernels as old as 2.6.31, that
contains the commit 0ebbab37422315a5d0cb29792271085bafdf38c0 "USB: xhci:
Ring allocation and initialization."
A separate patch will need to be developed for kernels older than 3.4,
since the ring allocation code was refactored in that kernel.
Signed-off-by: Julius Werner <jwerner@chromium.org> Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The xHCI 1.0 specification made a change to the TD Size field in TRBs.
The value is now the number of packets that remain to be sent in the TD,
not including this TRB. The TD Size value for the last TRB in a TD must
always be zero.
The xHCI function xhci_v1_0_td_remainder() attempts to calculate this,
but it gets it wrong. First, it erroneously reuses the old
xhci_td_remainder function, which will right shift the value by 10. The
xHCI 1.0 spec as of June 2011 says nothing about right shifting by 10.
Second, it does not set the TD size for the last TRB in a TD to zero.
Third, it uses roundup instead of DIV_ROUND_UP. The total packet count
is supposed to be the total number of bytes in this TD, divided by the
max packet size, rounded up. DIV_ROUND_UP is the right function to use
in that case.
With the old code, a TD on an endpoint with max packet size 1024 would
be set up like so:
TRB 1, TRB length = 600 bytes, TD size = 0
TRB 1, TRB length = 200 bytes, TD size = 0
TRB 1, TRB length = 100 bytes, TD size = 0
With the new code, the TD would be set up like this:
TRB 1, TRB length = 600 bytes, TD size = 1
TRB 1, TRB length = 200 bytes, TD size = 1
TRB 1, TRB length = 100 bytes, TD size = 0
This commit should be backported to kernels as old as 3.0, that contain
the commit 4da6e6f247a2601ab9f1e63424e4d944ed4124f3 "xhci 1.0: Update TD
size field format."
Above mentioned change was made along with multi usb phy change and
adding DT support for nop transceiver. But other two changes did not
make it to mainline. This in effect makes dsps musb wrapper unusable
even for single instance.
Hence revert it so that at least single instance can be supported.
Signed-off-by: Afzal Mohammed <afzal@ti.com> Signed-off-by: Felipe Balbi <balbi@ti.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Sankara reported that the genirq core code fails to adjust the
affinity of an interrupt thread in several cases:
1) On request/setup_irq() the call to setup_affinity() happens before
the new action is registered, so the new thread is not notified.
2) For secondary shared interrupts nothing notifies the new thread to
change its affinity.
3) Interrupts which have the IRQ_NO_BALANCE flag set are not moving
the thread either.
Fix this by setting the thread affinity flag right on thread creation
time. This ensures that under all circumstances the thread moves to
the right place. Requires a check in irq_thread_check_affinity for an
existing affinity mask (CONFIG_CPU_MASK_OFFSTACK=y)
If of_get_gpio_flags() returns an error (as in case when GPIO probe is
deferred) the driver would attempt to claim invalid GPIO. It should
propagate the error code up the stack instead so that the probe either
fails or will be retried later (in case of -EPROBE_DEFER).
Report only the position of the first finger as absolute non-MT coordinates,
instead of reporting both fingers alternatively. Actual MT events are
unaffected.
This fixes horizontal and improves vertical scrolling with the touchpad.
NFS appears to use d_obtain_alias() to create the root dentry rather than
d_make_root. This can cause 'prepend_path()' to complain that the root
has a weird name if an NFS filesystem is lazily unmounted. e.g. if
"/mnt" is an NFS mount then
{ cd /mnt; umount -l /mnt ; ls -l /proc/self/cwd; }
will cause a WARN message like
WARNING: at /home/git/linux/fs/dcache.c:2624 prepend_path+0x1d7/0x1e0()
...
Root dentry has weird name <>
to appear in kernel logs.
So change d_obtain_alias() to use "/" rather than "" as the anonymous
name.
Signed-off-by: NeilBrown <neilb@suse.de> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> 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> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch ensures that we free the rpc_task after the cleanup callbacks
are done in order to avoid a deadlock problem that can be triggered if
the callback needs to wait for another workqueue item to complete.
With NFSv4, if we create a file then open it we explicit avoid checking
the permissions on the file during the open because the fact that we
created it ensures we should be allow to open it (the create and the
open should appear to be a single operation).
However if the reply to an EXCLUSIVE create gets lots and the client
resends the create, the current code will perform the permission check -
because it doesn't realise that it did the open already..
This patch should fix this.
Note that I haven't actually seen this cause a problem. I was just
looking at the code trying to figure out a different EXCLUSIVE open
related issue, and this looked wrong.
(Fix confirmed with pynfs 4.0 test OPEN4--bfields)
Signed-off-by: NeilBrown <neilb@suse.de>
[bfields: use OWNER_OVERRIDE and update for 4.1] Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Very embarassing: 1091006c5eb15cba56785bd5b498a8d0b9546903 "nfsd: turn
on reply cache for NFSv4" missed a line, effectively leaving the reply
cache off in the v4 case. I thought I'd tested that, but I guess not.
This time, wrote a pynfs test to confirm it works.
Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The writeback code is already capable of passing errors back to user space
by means of the open_context->error. In the case of ENOSPC, Neil Brown
is reporting seeing 2 errors being returned.
Neil writes:
"e.g. if /mnt2/ if an nfs mounted filesystem that has no space then
reported Input/output error and the relevant parts of the strace output are:
write(1, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 512) = 512
fsync(1) = -1 EIO (Input/output error)
close(1) = -1 ENOSPC (No space left on device)"
Neil then shows that the duplication of error messages appears to be due to
the use of the PageError() mechanism, which causes filemap_fdatawait_range
to return the extra EIO. The regression was introduced by
commit 7b281ee026552f10862b617a2a51acf49c829554 (NFS: fsync() must exit
with an error if page writeback failed).
Fix this by removing the call to SetPageError(), and just relying on
open_context->error reporting the ENOSPC back to fsync().
Reported-by: Neil Brown <neilb@suse.de> Tested-by: Neil Brown <neilb@suse.de> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It is almost always wrong for NFS to call drop_nlink() after removing a
file. What we really want is to mark the inode's attributes for
revalidation, and we want to ensure that the VFS drops it if we're
reasonably sure that this is the final unlink().
Do the former using the usual cache validity flags, and the latter
by testing if inode->i_nlink == 1, and clearing it in that case.
This also fixes the following warning reported by Neil Brown and
Jeff Layton (among others).
In rare circumstances, nfs_clone_server() of a v2 or v3 server can get
an error between setting server->destory (to nfs_destroy_server), and
calling nfs_start_lockd (which will set server->nlm_host).
If this happens, nfs_clone_server will call nfs_free_server which
will call nfs_destroy_server and thence nlmclnt_done(NULL). This
causes the NULL to be dereferenced.
So add a guard to only call nlmclnt_done() if ->nlm_host is not NULL.
The other guards there are irrelevant as nlm_host can only be non-NULL
if one of these flags are set - so remove those tests. (Thanks to Trond
for this suggestion).
This is suitable for any stable kernel since 2.6.25.
Eryu provided a test program that would segfault when attempting to read
past the EOF on file that was opened O_DIRECT. The buffer given to the
read() call was on the stack, and when he attempted to read past it it
would scribble over the rest of the stack page.
If we hit the end of the file on a DIO READ request, then we don't want
to zero out the rest of the buffer. These aren't pagecache pages after
all, and there's no guarantee that the buffers that were passed in
represent entire pages.
Reported-by: Eryu Guan <eguan@redhat.com> Cc: Fred Isaman <iisaman@netapp.com> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 1f1ea6c "NFSv4: Fix buffer overflow checking in
__nfs4_get_acl_uncached" accidently dropped the checking for too small
result buffer length.
If someone uses getxattr on "system.nfs4_acl" on an NFSv4 mount
supporting ACLs, the ACL has not been cached and the buffer suplied is
too short, we still copy the complete ACL, resulting in kernel and user
space memory corruption.
While there's no locking involved here, the operations are serialized,
so CTO should prevent corruption.
The first write to the file is fine and writes 4 bytes. The file is then
extended on the server. When it's reopened a GETATTR is issued and the
size change is noticed. This causes NFS_INO_INVALID_DATA to be set on
the file. Because the file is opened for write only,
nfs_want_read_modify_write() returns 0 to nfs_write_begin().
nfs_updatepage then calls nfs_write_pageuptodate() to see if it should
extend the nfs_page to cover the whole page. NFS_INO_INVALID_DATA is
still set on the file at that point, but that flag is ignored and
nfs_pageuptodate erroneously extends the write to cover the whole page,
with the write done on the server side filled in with zeroes.
This patch just has that function check for NFS_INO_INVALID_DATA in
addition to NFS_INO_REVAL_PAGECACHE. This fixes the bug, but looking
over the code, I wonder if we might have a similar bug in
nfs_revalidate_size(). The difference between those two flags is very
subtle, so it seems like we ought to be checking for
NFS_INO_INVALID_DATA in most of the places that we look for
NFS_INO_REVAL_PAGECACHE.
I believe this is regression introduced by commit 8d197a568. The code
did check for NFS_INO_INVALID_DATA prior to that patch.
If I mount an NFS v4.1 server to a single client multiple times and then
run xfstests over each mountpoint I usually get the client into a state
where recovery deadlocks. The server informs the client of a
cb_path_down sequence error, the client then does a
bind_connection_to_session and checks the status of the lease.
I found that bind_connection_to_session sets the NFS4_SESSION_DRAINING
flag on the client, but this flag is never unset before
nfs4_check_lease() reaches nfs4_proc_sequence(). This causes the client
to deadlock, halting all NFS activity to the server. nfs4_proc_sequence()
is only called by the state manager, so I can change it to run in privileged
mode to bypass the NFS4_SESSION_DRAINING check and avoid the deadlock.
projects / karo-tx-linux.git / log