parisc: Define O_NONBLOCK to become 000200000
Helge Deller
maintainerRFC PATCH 1/1 parisc: Define O_NONBLOCK to become 000200000
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined O_NONBLOCK to 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).
To avoid such issues in the future, this patch changes O_NONBLOCK to become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.
Update the comment about SOCK_NONBLOCK too.
arch/parisc/include/asm/socket.h +2 -2
@@ -4,8 +4,8 @@#include <uapi/asm/socket.h> -/* O_NONBLOCK clashes with the bits used for socket types. Therefore we -* have to define SOCK_NONBLOCK to a different value here. +/* O_NONBLOCK clashed with the bits used for socket types. Therefore we +* had to define SOCK_NONBLOCK to a different value here. */ #define SOCK_NONBLOCK 0x40000000
arch/parisc/include/uapi/asm/fcntl.h +1 -1
@@ -8,7 +8,7 @@#define O_LARGEFILE 000004000#define __O_SYNC 000100000#define O_SYNC (__O_SYNC|O_DSYNC) -#define O_NONBLOCK 000200004/* HPUX has separate NDELAY & NONBLOCK */ +#define O_NONBLOCK 000200000#define O_NOCTTY 000400000/* not fcntl */ #define O_DSYNC 001000000#define O_NOATIME 004000000
Quoted 11 lines · On Sat, 29 Aug 2020 14:20:17 +0200 Helge Deller <[email protected]> wrote:
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined
O_NONBLOCKto 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).To avoid such issues in the future, this patch changes
O_NONBLOCKto become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.
I am seeing a problem with this exact commit in userland, so I think that last sentence is incorrect:
/usr/src/linux # git bisect good 75ae04206a4d0e4f541c1d692b7febd1c0fdb814 is the first bad commit commit 75ae04206a4d0e4f541c1d692b7febd1c0fdb814
Author: Helge Deller <[email protected]> Date: Sat Aug 29 14:11:35 2020 +0200
parisc: Define O_NONBLOCK to become 000200000
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined O_NONBLOCK to 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).
To avoid such issues in the future, this patch changes O_NONBLOCK to become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.
Update the comment about SOCK_NONBLOCK too.
arch/parisc/include/asm/socket.h | 4 ++-- arch/parisc/include/uapi/asm/fcntl.h | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
The first sign (in the boot process) that something is wrong is that idmapd fails to start:
* Starting idmapd ...
* make sure DNOTIFY support is enabled ...
[ !! ]
* ERROR: rpc.idmapd failed to start
* ERROR: cannot start nfsclient as rpc.idmapd would not start
Then, elogind reports an error when I ssh in as regular user:
[ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 298.040379][ T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument [ 298.040379][T4273] elogind-daemon[4273]: Failed to fully start up
daemon: Invalid argument
Yet the unprivileged user succeeds in logging in over SSH. Following that, sudo fails:
jeroen@karsten ~ $ sudo -i
sudo: unable to allocate memory
root can still login on the serial console and over SSH.
Would it make sense to rebuild libc against the newer kernel headers?
Or is this an unexpected result from the above commit and would it be useful to figure out what is going on while the bad kernel is running?
Regards,
jer
Helge Deller
maintainerQuoted 11 lines · Jeroen Roovers
Quoted 8 lines · On Sat, 29 Aug 2020 14:20:17 +0200 Helge Deller <[email protected]> wrote:
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined
O_NONBLOCKto 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).
Quoted 6 lines
Quoted 3 lines
To avoid such issues in the future, this patch changes
O_NONBLOCKto become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.I am seeing a problem with this exact commit in userland, so I think that last sentence is incorrect:
Thanks for testing and bisecting!!!
I'm fine with reverting the change, but we really need to analyze what is broken (and why).
In general the kernel sources seem ok as it's important, that code just check if bits are set, not if the value is equal to something e.g.
good: if (flags & O_NONBLOCK) { ... }
bad: if (flags == O_NONBLOCK) { .... }Quoted 8 lines
The first sign (in the boot process) that something is wrong is that idmapd fails to start:
* Starting idmapd ...
* make sure DNOTIFY support is enabled ...
[ !! ]
* ERROR: rpc.idmapd failed to start
* ERROR: cannot start nfsclient as rpc.idmapd would not start
Could you try an strace on it?
idmapd is from glibc, so I'll look into it too.
Quoted 10 lines
Then, elogind reports an error when I ssh in as regular user:
[ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 298.040379][ T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument [ 298.040379][T4273] elogind-daemon[4273]: Failed to fully start updaemon: Invalid argument
Quoted 7 lines
Yet the unprivileged user succeeds in logging in over SSH. Following that, sudo fails:
jeroen@karsten ~ $ sudo -i
sudo: unable to allocate memoryroot can still login on the serial console and over SSH.
At first thought I assume those issues are not related to the O_NONBLOCK patch. Can you try strace on the sudo too ?
Quoted 1 line
Would it make sense to rebuild libc against the newer kernel headers?
Yes, might make sense, but then my patch isn't compatible. So, I'd like to avoid that.
Quoted 2 lines
Or is this an unexpected result from the above commit and would it be useful to figure out what is going on while the bad kernel is running?
As I said, idmapd might be related, the elogind/sudo tests needs checking.
Helge
Quoted 34 lines · On Wed, 21 Oct 2020 08:07:15 +0200 Helge Deller <[email protected]> wrote:
Quoted 11 lines · Jeroen Roovers
Quoted 8 lines · On Sat, 29 Aug 2020 14:20:17 +0200 Helge Deller <[email protected]> wrote:
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined
O_NONBLOCKto 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).Quoted 6 lines
Quoted 3 lines
To avoid such issues in the future, this patch changes
O_NONBLOCKto become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.I am seeing a problem with this exact commit in userland, so I think that last sentence is incorrect:
Thanks for testing and bisecting!!!
I'm fine with reverting the change, but we really need to analyze what is broken (and why).
In general the kernel sources seem ok as it's important, that code just check if bits are set, not if the value is equal to something e.g.
good: if (flags & O_NONBLOCK) { ... } bad: if (flags == O_NONBLOCK) { .... }Quoted 3 lines
The first sign (in the boot process) that something is wrong is that idmapd fails to start:
Quoted 7 lines
Quoted 5 lines
* Starting idmapd ...
* make sure DNOTIFY support is enabled ...
[ !! ]
* ERROR: rpc.idmapd failed to start
* ERROR: cannot start nfsclient as rpc.idmapd would not startCould you try an strace on it?
[after editing the startup script to run strace -f .. rpc.idmapd:]
https://rooversj.home.xs4all.nl/rpc.idmapd.strace
Quoted 11 lines
idmapd is from glibc, so I'll look into it too.
Quoted 9 lines
Then, elogind reports an error when I ssh in as regular user:
[ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 297.825133][ T4273] elogind-daemon[4273]: Failed to registerSIGHUP handler: Invalid argument [ 298.040379][ T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument [ 298.040379][T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument
strace -f -o /tmp/elogind.strace /lib/elogind/elogind
https://rooversj.home.xs4all.nl/elogind.strace
Quoted 11 lines
Quoted 8 lines
Yet the unprivileged user succeeds in logging in over SSH. Following that, sudo fails:
jeroen@karsten ~ $ sudo -i
sudo: unable to allocate memoryroot can still login on the serial console and over SSH.
At first thought I assume those issues are not related to the
O_NONBLOCKpatch. Can you try strace on the sudo too ?
strace -f -u jeroen sudo -i
[...]pipe2(0x42f4712c, O_NONBLOCK|O_CLOEXEC) = -1 EINVAL (Invalid argument) openat(AT_FDCWD, 0xfadd9b50, O_RDONLY|O_CLOEXEC) = 3
fstat64(3, 0xfadd9e88) = 0 read(3, 0x42f47258, 4096) = 2998 read(3, "", 4096) = 0 close(3) = 0 openat(AT_FDCWD, 0x42f46bc0, O_RDONLY) = -1 ENOENT (No such file or
directory) openat(AT_FDCWD, 0x42f46f48, O_RDONLY) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, 0x42f46fc8, O_RDONLY) = -1 ENOENT (No such file or directory) openat(AT_FDCWD, 0x42f46f78, O_RDONLY) = -1 ENOENT (No such file or directory) ioctl(2, TCGETS, 0xfadd9c08)
= 0 write(2, 0xfadd7f46, 4sudo) = 4
ioctl(2, TCGETS, 0xfadd9c08) = 0write(2, 0xf8151e94, 2: ) = 2ioctl(2, TCGETS, 0xfadd9c08) = 0write(2, 0xfadd9608, 25unable to allocate memory) = 25ioctl(2, TCGETS, 0xfadd9c08) = 0
write(2, 0x42e39934, 2
) = 2exit_group(1) = ?
+++ exited with 1 +++
https://rooversj.home.xs4all.nl/sudo-i.strace
Regards,
jer
Helge Deller
maintainerQuoted 101 lines · Jeroen Roovers
Quoted 34 lines · On Wed, 21 Oct 2020 08:07:15 +0200 Helge Deller <[email protected]> wrote:
Quoted 11 lines · Jeroen Roovers
Quoted 8 lines · On Sat, 29 Aug 2020 14:20:17 +0200 Helge Deller <[email protected]> wrote:
HPUX has separate NDELAY & NONBLOCK values. In the past we wanted to be able to run HP-UX binaries natively on parisc Linux which is why we defined
O_NONBLOCKto 000200004 to distinguish NDELAY & NONBLOCK bits.
But with 2 bits set in this bitmask we often ran into compatibility issues with other Linux applications which often only test one bit (or even compare the values).Quoted 6 lines
Quoted 3 lines
To avoid such issues in the future, this patch changes
O_NONBLOCKto become 000200000. That way old programs will still be functional, and for new programs we now have only one bit set.I am seeing a problem with this exact commit in userland, so I think that last sentence is incorrect:
Thanks for testing and bisecting!!!
I'm fine with reverting the change, but we really need to analyze what is broken (and why).
In general the kernel sources seem ok as it's important, that code just check if bits are set, not if the value is equal to something e.g.
good: if (flags & O_NONBLOCK) { ... } bad: if (flags == O_NONBLOCK) { .... }Quoted 3 lines
The first sign (in the boot process) that something is wrong is that idmapd fails to start:
Quoted 7 lines
Quoted 5 lines
* Starting idmapd ...
* make sure DNOTIFY support is enabled ...
[ !! ]
* ERROR: rpc.idmapd failed to start
* ERROR: cannot start nfsclient as rpc.idmapd would not startCould you try an strace on it?
[after editing the startup script to run
strace -f .. rpc.idmapd:]https://rooversj.home.xs4all.nl/rpc.idmapd.strace
Quoted 11 lines
idmapd is from glibc, so I'll look into it too.
Quoted 9 lines
Then, elogind reports an error when I ssh in as regular user:
[ 297.825133][ T4273] elogind-daemon[4273]: Failed to register SIGHUP handler: Invalid argument [ 297.825133][ T4273] elogind-daemon[4273]: Failed to registerSIGHUP handler: Invalid argument [ 298.040379][ T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument [ 298.040379][T4273] elogind-daemon[4273]: Failed to fully start up daemon: Invalid argument
strace -f -o
/tmp/elogind.strace/lib/elogind/elogindhttps://rooversj.home.xs4all.nl/elogind.strace
Quoted 11 lines
Quoted 8 lines
Yet the unprivileged user succeeds in logging in over SSH. Following that, sudo fails:
jeroen@karsten ~ $ sudo -i
sudo: unable to allocate memoryroot can still login on the serial console and over SSH.
At first thought I assume those issues are not related to the
O_NONBLOCKpatch. Can you try strace on the sudo too ?strace -f -u jeroen sudo -i
[...]pipe2(0x42f4712c, O_NONBLOCK|O_CLOEXEC)= -1 EINVAL (Invalid argument)openat(AT_FDCWD, 0xfadd9b50, O_RDONLY|O_CLOEXEC)= 3fstat64(3, 0xfadd9e88) = 0 read(3, 0x42f47258, 4096) = 2998 read(3, "", 4096) = 0 close(3) = 0 openat(AT_FDCWD, 0x42f46bc0, O_RDONLY) = -1 ENOENT (No such file ordirectory)
openat(AT_FDCWD, 0x42f46f48, O_RDONLY)= -1 ENOENT (No such file or directory)openat(AT_FDCWD, 0x42f46fc8, O_RDONLY)= -1 ENOENT (No such file or directory)openat(AT_FDCWD, 0x42f46f78, O_RDONLY)= -1 ENOENT (No such file or directory)ioctl(2, TCGETS, 0xfadd9c08)= 0 write(2, 0xfadd7f46, 4sudo) = 4
ioctl(2, TCGETS, 0xfadd9c08)= 0write(2, 0xf8151e94, 2: )= 2ioctl(2, TCGETS, 0xfadd9c08)= 0write(2, 0xfadd9608, 25unable to allocate memory)= 25ioctl(2, TCGETS, 0xfadd9c08)= 0
write(2,0x42e39934, 2
) = 2exit_group(1)= ?
+++ exited with 1 +++
Quoted 1 line
Thanks!
I found the issue.
The syscalls timerfd_create(), signalfd4(), eventfd2(), userfaultfd() and pipe2() have a fcntl flags parameter which is checked hard. They return EINVAL and as such the programs fail.
I found systemd-udevd and udevadm failing because of this. sudo and elogind seem affected too.
I'm sending a RFC patch in a few minutes.
Helge
Helge Deller
maintainerRFC PATCH 1/1 parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
The commit 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000") changed the O_NONBLOCK constant to have only one bit set (like all other architectures). This change broke some existing userspace code (e.g. udevadm, systemd-udevd, elogind) which called specific syscalls which do strict value checking on their flag parameter.
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid O_NONBLOCK flags, reports in the syslog if the old O_NONBLOCK value was used and then calls the target syscall with the new O_NONBLOCK value.
arch/parisc/kernel/sys_parisc.c +69 -1
@@ -6,7 +6,7 @@* Copyright (C) 1999-2003 Matthew Wilcox <willy at parisc-linux.org> * Copyright (C) 2000-2003 Paul Bame <bame at parisc-linux.org> * Copyright (C) 2001 Thomas Bogendoerfer <tsbogend at parisc-linux.org> -* Copyright (C) 1999-2014 Helge Deller <[email protected]> +* Copyright (C) 1999-2020 Helge Deller <[email protected]> */ #include <linux/uaccess.h>
@@ -23,6 +23,8 @@#include <linux/utsname.h> #include <linux/personality.h> #include <linux/random.h> +#include <linux/compat.h> +#include <linux/version.h> /* we construct an artificial offset for the mapping based on the physical * address of the kernel mapping variable */
@@ -373,3 +375,69 @@ long parisc_personality(unsigned long personality)return err; } + +/* +* Up to kernel v5.9 we defined O_NONBLOCK as 000200004, +* since then O_NONBLOCK is defined as 000200000. +* +* The following wrapper functions mask out the old +* O_NDELAY bit from calls which use O_NONBLOCK. +*/ + +#if ((LINUX_VERSION_CODE >>16 ) &0x0ff ) >5 +#warning "Remove O_NONBLOCK compatibility wrappers now?" +#endif + +#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +} + +asmlinkage long parisc_timerfd_create (int clockid,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_timerfd_create (clockid, flags); +} + +asmlinkage long parisc_signalfd4 (int ufd,sigset_t __user *user_mask, +size_t sizemask,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_signalfd4 (ufd, user_mask, sizemask, flags); +} + +#ifdef CONFIG_COMPAT +asmlinkage long parisc_compat_signalfd4 (int ufd, +compat_sigset_t __user *user_mask, +compat_size_t sizemask,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return compat_sys_signalfd4 (ufd, user_mask, sizemask, flags); +} +#endif + +asmlinkage long parisc_eventfd2 (unsigned int count,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_eventfd2 (count, flags); +} + +asmlinkage long parisc_userfaultfd (int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_userfaultfd (flags); +} + +asmlinkage long parisc_pipe2 (int __user *fildes,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_pipe2 (fildes, flags); +}
arch/parisc/kernel/syscalls/syscall.tbl +5 -5
@@ -344,16 +344,16 @@ 304 common eventfd sys_eventfd 305 32 fallocate parisc_fallocate 305 64 fallocate sys_fallocate -306 common timerfd_create sys_timerfd_create +306 common timerfd_create parisc_timerfd_create 307 32 timerfd_settime sys_timerfd_settime32 307 64 timerfd_settime sys_timerfd_settime 308 32 timerfd_gettime sys_timerfd_gettime32 308 64 timerfd_gettime sys_timerfd_gettime -309 common signalfd4 sys_signalfd4 compat_sys_signalfd4 -310 common eventfd2 sys_eventfd2 +309 common signalfd4 parisc_signalfd4 parisc_compat_signalfd4 +310 common eventfd2 parisc_eventfd2 311 common epoll_create1 sys_epoll_create1 312 common dup3 sys_dup3 -313 common pipe2 sys_pipe2 +313 common pipe2 parisc_pipe2 314 common inotify_init1 sys_inotify_init1 315 common preadv sys_preadv compat_sys_preadv 316 common pwritev sys_pwritev compat_sys_pwritev
@@ -387,7 +387,7 @@ 341 common bpf sys_bpf 342 common execveat sys_execveat compat_sys_execveat 343 common membarrier sys_membarrier -344 common userfaultfd sys_userfaultfd +344 common userfaultfd parisc_userfaultfd 345 common mlock2 sys_mlock2 346 common copy_file_range sys_copy_file_range 347 common preadv2 sys_preadv2 compat_sys_preadv2
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
22.10.20 19:40 Helge Deller wrotw:
Quoted 7 lines
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid
O_NONBLOCKflags, reports in the syslog if the oldO_NONBLOCKvalue was used and then calls the target syscall with the newO_NONBLOCKvalue.Fixes 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000")Signed-off-by Helge Deller <[email protected]>
Works for me on RP2470 - boots up in full functionality and logs the recompilation warning about systemd-udevd and syslog-ng.
Signature · 1 line
Meelis Roos
Helge Deller
maintainerRe: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 7 lines · Meelis Roos
22.10.20 19:40 Helge Deller wrotw:
Quoted 5 lines
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid
O_NONBLOCKflags, reports in the syslog if the oldO_NONBLOCKvalue was used and then calls the target syscall with the newO_NONBLOCKvalue.
Quoted 7 lines
Quoted 2 lines
Fixes 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000")Signed-off-by Helge Deller <[email protected]>Works for me on RP2470 - boots up in full functionality and logs the recompilation warning about systemd-udevd and syslog-ng.
Tested-by Meelis Roos <[email protected]>
Thank you Meelis & Jeroen for testing!
The big question is:
We have two options
a) revert my original commit 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000"), or
b) apply and submit this new patch on top of it.
The benefit in b) is that we will get long term rid of the two-bit O_NONBLOCK define and thus will get more compatible to other Linux architectures. This comes with the downside of (at least for a few years) added overhead for those non-performance critical syscalls.
The benefit with a) is that we then step back again and stay compatible now. The downside is, that in the future we may run into other issues and need to keep special-handling in some other syscalls forever.
I'm still for going with b), and I hope we got all corner cases ruled out. Opinions?
Helge
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 24 lines · On Thu, 22 Oct 2020 22:29:18 +0200 Helge Deller <[email protected]> wrote:
Quoted 7 lines · Meelis Roos
22.10.20 19:40 Helge Deller wrotw:
Quoted 5 lines
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid
O_NONBLOCKflags, reports in the syslog if the oldO_NONBLOCKvalue was used and then calls the target syscall with the newO_NONBLOCKvalue.Quoted 7 lines
Quoted 2 lines
Fixes 75ae04206a4d ("parisc: Define O_NONBLOCK to become000200000") Signed-off-by: Helge Deller <[email protected]>
Works for me on RP2470 - boots up in full functionality and logs the recompilation warning about systemd-udevd and syslog-ng.
Tested-by Meelis Roos <[email protected]>Thank you Meelis & Jeroen for testing!
The big question is:
We have two options
a) revert my original commit75ae04206a4d("parisc: DefineO_NONBLOCKto become 000200000"), or b) apply and submit this new patch on top of it.
Quoted 4 lines
The benefit in b) is that we will get long term rid of the two-bit
O_NONBLOCKdefine and thus will get more compatible to other Linux architectures. This comes with the downside of (at least for a few years) added overhead for those non-performance critical syscalls.
The performance issue is resolved once the installed kernel headers/libc have been updated accordingly. I think after that the overhead should be minimal.
Quoted 8 lines
The benefit with a) is that we then step back again and stay compatible now. The downside is, that in the future we may run into other issues and need to keep special-handling in some other syscalls forever.
I'm still for going with b), and I hope we got all corner cases ruled out. Opinions?
I think the performance penalty isn't that great, but could be improved, so I'd go for b) with a small change. The warning that it issues seems redundant, because the immediate problem has already been "solved", and because the proposed solution does not work:
+{
+ if (flags & O_NONBLOCK_MASK_OUT) {
+ struct task_struct *tsk = current;
+ pr_warn("%s(%d) uses old O_NONBLOCK value. "
+ "Please recompile the application.\n",
+ tsk->comm, tsk->pid);
+ }
+ return flags & ~O_NONBLOCK_MASK_OUT;
+}
+
The text assumes that the officially packaged kernel headers are in sync with the kernel, which normally isn't the case as most distributions seem to either pick an LTS kernel for their toolchain, or keep the installed kernel in sync with the installed kernel headers, but do not prevent running newer kernels and may even encourage doing so. Simply recompiling the programs that use the old O_NONBLOCK value does not solve the problem in most cases.
If you'd remove that if() { pr_warn } entirely, you'd probably be rid of most of the performance penalty anyway.
Regards,
jer
Helge Deller
maintainerRe: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 27 lines · Jeroen Roovers
Quoted 24 lines · On Thu, 22 Oct 2020 22:29:18 +0200 Helge Deller <[email protected]> wrote:
Quoted 7 lines · Meelis Roos
22.10.20 19:40 Helge Deller wrotw:
Quoted 5 lines
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid
O_NONBLOCKflags, reports in the syslog if the oldO_NONBLOCKvalue was used and then calls the target syscall with the newO_NONBLOCKvalue.Quoted 7 lines
Quoted 2 lines
Fixes 75ae04206a4d ("parisc: Define O_NONBLOCK to become000200000") Signed-off-by: Helge Deller <[email protected]>
Works for me on RP2470 - boots up in full functionality and logs the recompilation warning about systemd-udevd and syslog-ng.
Tested-by Meelis Roos <[email protected]>Thank you Meelis & Jeroen for testing!
The big question is:
We have two options
a) revert my original commit75ae04206a4d("parisc: DefineO_NONBLOCKto become 000200000"), or b) apply and submit this new patch on top of it.
Quoted 8 lines
Quoted 4 lines
The benefit in b) is that we will get long term rid of the two-bit
O_NONBLOCKdefine and thus will get more compatible to other Linux architectures. This comes with the downside of (at least for a few years) added overhead for those non-performance critical syscalls.The performance issue is resolved once the installed kernel headers/libc have been updated accordingly. I think after that the overhead should be minimal.
Yes, probably.
Quoted 6 lines
Quoted 6 lines
The benefit with a) is that we then step back again and stay compatible now. The downside is, that in the future we may run into other issues and need to keep special-handling in some other syscalls forever.
I'm still for going with b), and I hope we got all corner cases ruled
I meant "proposing to go with option b)"....
Quoted 18 lines
Quoted 1 line
out. Opinions?
I think the performance penalty isn't that great, but could be improved, so I'd go for b) with a small change. The warning that it issues seems redundant, because the immediate problem has already been "solved", and because the proposed solution does not work:
arch/parisc/kernel/sys_parisc.c +10
-0+{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +} +
Quoted 7 lines
The text assumes that the officially packaged kernel headers are in sync with the kernel, which normally isn't the case as most distributions seem to either pick an LTS kernel for their toolchain, or keep the installed kernel in sync with the installed kernel headers, but do not prevent running newer kernels and may even encourage doing so. Simply recompiling the programs that use the old
O_NONBLOCKvalue does not solve the problem in most cases.
True.
Quoted 2 lines
If you'd remove that
if(){pr_warn} entirely, you'd probably be rid of most of the performance penalty anyway.
I think it makes sense to keep at least one warning. Rolf Eike proposed a pr_warn_once() which seems the best compromise as it keeps us at least informed which packages needs updating and how relevant it is to keep those wrappers....
Helge
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 4 lines · On Fri, 23 Oct 2020 10:35:34 +0200 Helge Deller <[email protected]> wrote:
I think it makes sense to keep at least one warning. Rolf Eike proposed a
pr_warn_once()which seems the best compromise as it keeps us at least informed which packages needs updating and how relevant it is to keep those wrappers....
I agree that if we are to keep the warning, pr_warn_once is a better choice.
In the rpc.idmapd strace I still see
4635 inotify_init1(IN_NONBLOCK) = -1 EINVAL (Invalid argument)
so maybe you want to wrap around inotify_init1 as well?
Regards,
jer
Helge Deller
maintainerRe: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 8 lines · Jeroen Roovers
Quoted 4 lines · On Fri, 23 Oct 2020 10:35:34 +0200 Helge Deller <[email protected]> wrote:
I think it makes sense to keep at least one warning. Rolf Eike proposed a
pr_warn_once()which seems the best compromise as it keeps us at least informed which packages needs updating and how relevant it is to keep those wrappers....
Quoted 8 lines
I agree that if we are to keep the warning,
pr_warn_onceis a better choice.In the rpc.idmapd strace I still see
4635
inotify_init1(IN_NONBLOCK)= -1 EINVAL (Invalid argument)so maybe you want to wrap around
inotify_init1as well?
Done.
Added in the new patch version.
Helge
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
In reply to Helge DellerQuoted 7 lines · On Thu, 22 Oct 2020 18:40:07 +0200 Helge Deller <[email protected]> wrote:
The commit
75ae04206a4d("parisc: DefineO_NONBLOCKto become 000200000") changed theO_NONBLOCKconstant to have only one bit set (like all other architectures). This change broke some existing userspace code (e.g. udevadm, systemd-udevd, elogind) which called specific syscalls which do strict value checking on their flag parameter.
Quoted 7 lines
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid
O_NONBLOCKflags, reports in the syslog if the oldO_NONBLOCKvalue was used and then calls the target syscall with the newO_NONBLOCKvalue.Fixes 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000")Signed-off-by Helge Deller <[email protected]>
That fixes all the previous problems.
Regards,
jer
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
In reply to Helge Dellerarch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 402
Quoted 12 lines
arch/parisc/kernel/sys_parisc.c +12
-0+#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}
This will also trigger if I just pass 0x4 in flags, no? The check should be
if ((flags & O_NONBLOCK_OLD) == O_NONBLOCK_OLD)
because that would correctly reject a bare 0x4, at least I hope that this would already happen with the strict checking you mentioned.
Would a pr_warn_once make sense? Otherwise your log may get flooded by them if e.g. sudo is the problem and my nagios comes every minute to check something.
Eike
Helge Deller
maintainerRe: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 15 lines · Rolf Eike Beer
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 402
Quoted 12 lines
arch/parisc/kernel/sys_parisc.c +12
-0+#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}This will also trigger if I just pass
0x4in flags, no? The check should be
Quoted 1 line
if ((flags & O_NONBLOCK_OLD) == O_NONBLOCK_OLD)
RIGHT!
That's a very good point.
I was thinking about what would happen if over time a new (unrelated) define gets created which then gets 0x4 as value. My code would then have wrongly modified it.
I'll fix this.
Quoted 5 lines
because that would correctly reject a bare
0x4, at least I hope that this would already happen with the strict checking you mentioned.Would a
pr_warn_oncemake sense? Otherwise your log may get flooded by them if e.g. sudo is the problem and my nagios comes every minute to check something.
I like the idea.
Helge
Helge Deller
maintainerRe: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 16 lines · Helge Deller
Quoted 15 lines · Rolf Eike Beer
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 402
Quoted 12 lines
arch/parisc/kernel/sys_parisc.c +12
-0+#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}This will also trigger if I just pass
0x4in flags, no? The check should be
Quoted 7 lines
Quoted 1 line
if ((flags & O_NONBLOCK_OLD) == O_NONBLOCK_OLD)RIGHT!
That's a very good point.
I was thinking about what would happen if over time a new (unrelated) define gets created which then gets0x4as value. My code would then have wrongly modified it.
After some more thinking....
It's not that easy.
Let's assume there will be another new flag at some time with value 0x4. Now, the caller sets this flag (0x4) and new O_NONBLOCK (000200000), so you end up with 000200004 again, which then triggers your check:
if ((flags & O_NONBLOCK_OLD) == O_NONBLOCK_OLD)
So, my check to test only the mask for 0x4 was basically better, because it would prevent the usage of 0x4 as any new value. In any way, it seems we need to avoid using 0x4 for a long time...
Helge
Re: [RFC PATCH] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Am Freitag, 23. Oktober 2020, 10:31:14 CEST schrieb Helge Deller:
Quoted 34 lines · Helge Deller
Quoted 17 lines · Rolf Eike Beer
Quoted 16 lines
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 402
Quoted 12 lines
arch/parisc/kernel/sys_parisc.c +12
-0+#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn ("%s(%d) uses old O_NONBLOCK value. " +"Please recompile the application.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}This will also trigger if I just pass
0x4in flags, no? The check should beQuoted 7 lines
Quoted 1 line
if ((flags & O_NONBLOCK_OLD) == O_NONBLOCK_OLD)RIGHT!
That's a very good point.
I was thinking about what would happen if over time a new (unrelated) define gets created which then gets0x4as value. My code would then have wrongly modified it.After some more thinking....
It's not that easy.
Let's assume there will be another new flag at some time with value0x4. Now, the caller sets this flag (0x4) and newO_NONBLOCK(000200000), so you end up with 000200004 again, which then triggers your check:
if ((flags &O_NONBLOCK_OLD) ==O_NONBLOCK_OLD)
Quoted 3 lines
So, my check to test only the mask for
0x4was basically better, because it would prevent the usage of0x4as any new value. In any way, it seems we need to avoid using0x4for a long time...
Then maybe just add that as an explicit comment in the header so noone picks it by accident, and add a comment pointing to these wrappers so they are more likely to be kept in sync.
And for extra fun we can just scrap all this for 64 bit userspace as we don't need the compat there ;)
Eike
Helge Deller
maintainerRFC PATCH v3 1/1 parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
In reply to Helge DellerThe commit 75ae04206a4d ("parisc: Define O_NONBLOCK to become 000200000") changed the O_NONBLOCK constant to have only one bit set (like all other architectures). This change broke some existing userspace code (e.g. udevadm, systemd-udevd, elogind) which called specific syscalls which do strict value checking on their flag parameter.
This patch adds wrapper functions for the relevant syscalls. The wrappers masks out any old invalid O_NONBLOCK flags, reports in the syslog if the old O_NONBLOCK value was used and then calls the target syscall with the new O_NONBLOCK value.
--
v3: Added inotify_init1() syscall wrapper
Changed warning to pr_warn_once()
arch/parisc/kernel/sys_parisc.c +72 -1
@@ -6,7 +6,7 @@* Copyright (C) 1999-2003 Matthew Wilcox <willy at parisc-linux.org> * Copyright (C) 2000-2003 Paul Bame <bame at parisc-linux.org> * Copyright (C) 2001 Thomas Bogendoerfer <tsbogend at parisc-linux.org> -* Copyright (C) 1999-2014 Helge Deller <[email protected]> +* Copyright (C) 1999-2020 Helge Deller <[email protected]> */ #include <linux/uaccess.h>
@@ -23,6 +23,7 @@#include <linux/utsname.h> #include <linux/personality.h> #include <linux/random.h> +#include <linux/compat.h> /* we construct an artificial offset for the mapping based on the physical * address of the kernel mapping variable */
@@ -373,3 +374,73 @@ long parisc_personality(unsigned long personality)return err; } + +/* +* Up to kernel v5.9 we defined O_NONBLOCK as 000200004, +* since then O_NONBLOCK is defined as 000200000. +* +* The following wrapper functions mask out the old +* O_NDELAY bit from calls which use O_NONBLOCK. +* +* XXX: Remove those in year 2022 (or later)? +*/ + +#define O_NONBLOCK_OLD 000200004 +#define O_NONBLOCK_MASK_OUT (O_NONBLOCK_OLD & ~O_NONBLOCK) + +static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; +pr_warn_once ("%s(%d) uses a deprecated O_NONBLOCK value.\n" , + tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +} + +asmlinkage long parisc_timerfd_create (int clockid,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_timerfd_create (clockid, flags); +} + +asmlinkage long parisc_signalfd4 (int ufd,sigset_t __user *user_mask, +size_t sizemask,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_signalfd4 (ufd, user_mask, sizemask, flags); +} + +#ifdef CONFIG_COMPAT +asmlinkage long parisc_compat_signalfd4 (int ufd, +compat_sigset_t __user *user_mask, +compat_size_t sizemask,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return compat_sys_signalfd4 (ufd, user_mask, sizemask, flags); +} +#endif + +asmlinkage long parisc_eventfd2 (unsigned int count,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_eventfd2 (count, flags); +} + +asmlinkage long parisc_userfaultfd (int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_userfaultfd (flags); +} + +asmlinkage long parisc_pipe2 (int __user *fildes,int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_pipe2 (fildes, flags); +} + +asmlinkage long parisc_inotify_init1 (int flags) +{ + flags =FIX_O_NONBLOCK (flags); +return sys_inotify_init1 (flags); +}
arch/parisc/kernel/syscalls/syscall.tbl +6 -6
@@ -344,17 +344,17 @@ 304 common eventfd sys_eventfd 305 32 fallocate parisc_fallocate 305 64 fallocate sys_fallocate -306 common timerfd_create sys_timerfd_create +306 common timerfd_create parisc_timerfd_create 307 32 timerfd_settime sys_timerfd_settime32 307 64 timerfd_settime sys_timerfd_settime 308 32 timerfd_gettime sys_timerfd_gettime32 308 64 timerfd_gettime sys_timerfd_gettime -309 common signalfd4 sys_signalfd4 compat_sys_signalfd4 -310 common eventfd2 sys_eventfd2 +309 common signalfd4 parisc_signalfd4 parisc_compat_signalfd4 +310 common eventfd2 parisc_eventfd2 311 common epoll_create1 sys_epoll_create1 312 common dup3 sys_dup3 -313 common pipe2 sys_pipe2 -314 common inotify_init1 sys_inotify_init1 +313 common pipe2 parisc_pipe2 +314 common inotify_init1 parisc_inotify_init1 315 common preadv sys_preadv compat_sys_preadv 316 common pwritev sys_pwritev compat_sys_pwritev 317 common rt_tgsigqueueinfo sys_rt_tgsigqueueinfo compat_sys_rt_tgsigqueueinfo
@@ -387,7 +387,7 @@ 341 common bpf sys_bpf 342 common execveat sys_execveat compat_sys_execveat 343 common membarrier sys_membarrier -344 common userfaultfd sys_userfaultfd +344 common userfaultfd parisc_userfaultfd 345 common mlock2 sys_mlock2 346 common copy_file_range sys_copy_file_range 347 common preadv2 sys_preadv2 compat_sys_preadv2
Re: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 399
Quoted 10 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c +5
-0+static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; + pr_warn_once("%s(%d) uses a deprecated O_NONBLOCKvalue.\n",
arch/parisc/kernel/sys_parisc.c +4
-0+ tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}
Would it be interesting to additionally report the calling function in search for other syscalls that might not be covered yet?
Regards,
jer
Re: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 16 lines · On Sat, 24 Oct 2020 10:22:18 +0200 Jeroen Roovers <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 399
Quoted 10 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c +5
-0+static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; + pr_warn_once("%s(%d) uses a deprecated O_NONBLOCKvalue.\n",
arch/parisc/kernel/sys_parisc.c +4
-0+ tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}Would it be interesting to additionally report the calling function in search for other syscalls that might not be covered yet?
Wait, that doesn't make sense, does it?
Regards,
jer
Helge Deller
maintainerRe: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 20 lines · Jeroen Roovers
Quoted 16 lines · On Sat, 24 Oct 2020 10:22:18 +0200 Jeroen Roovers <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 399
Quoted 10 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c +5
-0+static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; + pr_warn_once("%s(%d) uses a deprecated O_NONBLOCKvalue.\n",
arch/parisc/kernel/sys_parisc.c +4
-0+ tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}Would it be interesting to additionally report the calling function in search for other syscalls that might not be covered yet?
Quoted 1 line
Wait, that doesn't make sense, does it?
makes no sense :-)
The function is only called by syscalls where we know they are affected.
Helge
Re: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 19 lines · Helge Deller
Quoted 18 lines · Jeroen Roovers
Quoted 15 lines · On Sat, 24 Oct 2020 10:22:18 +0200 Jeroen Roovers <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c long parisc_personality(unsigned long personality) · line 399
Quoted 10 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
arch/parisc/kernel/sys_parisc.c +5
-0+static int FIX_O_NONBLOCK (int flags) +{ +if (flags &O_NONBLOCK_MASK_OUT ) { +struct task_struct *tsk = current; + pr_warn_once("%s(%d) uses a deprecated O_NONBLOCKvalue.\n",
arch/parisc/kernel/sys_parisc.c +4
-0+ tsk->comm, tsk->pid); + } +return flags & ~O_NONBLOCK_MASK_OUT ; +}Would it be interesting to additionally report the calling function in search for other syscalls that might not be covered yet?
Quoted 3 lines
Quoted 1 line
Wait, that doesn't make sense, does it?
makes no sense :-)
The function is only called by syscalls where we know they are affected.
I tend to think the warning is annoying. We will have to keep compatibility with old binaries forever.
Dave
Signature · 1 line
John David Anglin [email protected]
Re: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
In reply to Helge DellerQuoted 2 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
v3: Added
inotify_init1()syscall wrapper
Changed warning topr_warn_once()
I do not know how pr_warn_once decides when "once" has already occurred, but with this patch I do not see any warnings at all.
Regards,
jer
Helge Deller
maintainerRe: [RFC PATCH v3] parisc: Add wrapper syscalls to fix O_NONBLOCK flag usage
Quoted 9 lines · Jeroen Roovers
Quoted 2 lines · On Fri, 23 Oct 2020 20:18:47 +0200 Helge Deller <[email protected]> wrote:
v3: Added
inotify_init1()syscall wrapper
Changed warning topr_warn_once()I do not know how
pr_warn_oncedecides when "once" has already occurred
It's a static variable....
Quoted 1 line
, but with this patch I do not see any warnings at all.
Works for me:
[ 19.944173] systemd[1]: systemd 246-2 running in system mode. (+PAM +AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP +GCRYPT +GNUTLS +ACL +XZ +LZ4 +ZSTD -SECCOMP +BLKID +ELFUTILS +KMOD +IDN2 -IDN +PCRE2 default-hierarchy=hybrid) [ 19.954344] systemd[1]: Detected architecture parisc. Welcome to Debian GNU/Linux bullseye/sid! [ 20.254667] systemd[1]: Set hostname to <debian>. [ 20.374343] systemd(1) uses a deprecated O_NONBLOCK value.
....
Helge