LSM hooks for IORING_OP_URING_CMD
Subsystems (4)
Paul Moore
maintainercover LSM hooks for IORING_OP_URING_CMD
This patchset includes three patches: one to add a new LSM hook for the IORING_OP_URING_CMD operation, one to add the SELinux implementation for the new hook, and one to enable IORING_OP_URING_CMD for /dev/null. The last patch, the /dev/null support, is obviously not critical but it makes testing so much easier and I believe is in keeping with the general motivation behind /dev/null.
Luis' patch has already been vetted by Jens and the io_uring folks, so the only new bits are the SELinux implementation and the trivial /dev/null implementation of IORING_OP_URING_CMD. Assuming no one has any objections over the next few days, I'll plan on sending this up to Linus during the v6.0-rcX cycle.
I believe Casey is also currently working on Smack support for the IORING_OP_URING_CMD hook, and as soon as he is ready I can add it to this patchset (or Casey can send it up himself).
-Paul
---
Luis Chamberlain (1):
lsm,io_uring: add LSM hooks for the new uring_cmd file op
Paul Moore (2):
selinux: implement the security_uring_cmd() LSM hook /dev/null: add IORING_OP_URING_CMD support
drivers/char/mem.c | 6 ++++++ include/linux/lsm_hook_defs.h | 1 + include/linux/lsm_hooks.h | 3 +++ include/linux/security.h | 5 +++++ io_uring/uring_cmd.c | 5 +++++ security/security.c | 4 ++++ security/selinux/hooks.c | 24 ++++++++++++++++++++++++ security/selinux/include/classmap.h | 2 +- 8 files changed, 49 insertions(+), 1 deletion(-)
Paul Moore
maintainerPATCH 1/3 lsm,io_uring: add LSM hooks for the new uring_cmd file op
From: Luis Chamberlain <[email protected]>
io-uring cmd support was added through ee692a21e9bf ("fs,io_uring:
add infrastructure for uring-cmd"), this extended the struct file_operations to allow a new command which each subsystem can use to enable command passthrough. Add an LSM specific for the command passthrough which enables LSMs to inspect the command details.
This was discussed long ago without no clear pointer for something conclusive, so this enables LSMs to at least reject this new file operation.
[0] https://lkml.kernel.org/r/[email protected]
--- include/linux/lsm_hook_defs.h | 1 + include/linux/lsm_hooks.h | 3 +++ include/linux/security.h | 5 +++++ io_uring/uring_cmd.c | 5 +++++ security/security.c | 4 ++++ 5 files changed, 18 insertions(+)
include/linux/lsm_hook_defs.h +1 -0
@@ -407,4 +407,5 @@ LSM_HOOK(int, 0, perf_event_write, struct perf_event *event)#ifdef CONFIG_IO_URING LSM_HOOK(int, 0, uring_override_creds, const struct cred *new) LSM_HOOK(int, 0, uring_sqpoll, void) +LSM_HOOK(int, 0, uring_cmd, struct io_uring_cmd *ioucmd)#endif /* CONFIG_IO_URING */
include/linux/lsm_hooks.h +3 -0
@@ -1582,6 +1582,9 @@* Check whether the current task is allowed to spawn a io_uring polling * thread (IORING_SETUP_SQPOLL). * +* @uring_cmd: +* Check whether the file_operations uring_cmd is allowed to run. +* */ union security_list_options {#define LSM_HOOK (RET ,DEFAULT ,NAME , ...) RET (*NAME)(__VA_ARGS__);
include/linux/security.h +5 -0
@@ -2060,6 +2060,7 @@ static inline int security_perf_event_write(struct perf_event *event)#ifdef CONFIG_SECURITY extern int security_uring_override_creds (const struct cred *new);extern int security_uring_sqpoll (void ); +extern int security_uring_cmd (struct io_uring_cmd *ioucmd);#else static inline int security_uring_override_creds (const struct cred *new) {
@@ -2069,6 +2070,10 @@ static inline int security_uring_sqpoll(void) {return 0 ; } +static inline int security_uring_cmd (struct io_uring_cmd *ioucmd) +{ +return 0 ; +}#endif /* CONFIG_SECURITY */ #endif /* CONFIG_IO_URING */
io_uring/uring_cmd.c +5 -0
@@ -3,6 +3,7 @@#include <linux/errno.h> #include <linux/file.h> #include <linux/io_uring.h> +#include <linux/security.h> #include <uapi/linux/io_uring.h>
@@ -88,6 +89,10 @@ int io_uring_cmd(struct io_kiocb *req, unsigned int issue_flags)if (!req->file->f_op->uring_cmd)return -EOPNOTSUPP ; + ret =security_uring_cmd (ioucmd); +if (ret) +return ret; +if (ctx->flags &IORING_SETUP_SQE128 ) issue_flags |=IO_URING_F_SQE128 ;if (ctx->flags &IORING_SETUP_CQE32 )
security/security.c +4 -0
@@ -2660,4 +2660,8 @@ int security_uring_sqpoll(void) {return call_int_hook (uring_sqpoll,0 ); } +int security_uring_cmd (struct io_uring_cmd *ioucmd) +{ +return call_int_hook (uring_cmd,0 , ioucmd); +}#endif /* CONFIG_IO_URING */
Greg Kroah-Hartman
maintainerRe: [PATCH 1/3] lsm,io_uring: add LSM hooks for the new uring_cmd file op
Quoted 8 lines · Paul Moore
From: Luis Chamberlain <[email protected]>
io-uring cmd support was added through
ee692a21e9bf("fs,io_uring:
add infrastructure for uring-cmd"), this extended the structfile_operationsto allow a new command which each subsystem can use to enable command passthrough. Add an LSM specific for the command passthrough which enables LSMs to inspect the command details.
Quoted 7 lines
This was discussed long ago without no clear pointer for something conclusive, so this enables LSMs to at least reject this new file operation.
[0] https://lkml.kernel.org/r/[email protected]
Fixes ee692a21e9bf ("fs,io_uring: add infrastructure for uring-cmd")
You are not "fixing" anything, you are adding new functionality. Careful with using "Fixes:" for something like this, you will trigger the bug-detection scripts and have to fend off stable bot emails for a long time for stuff that should not be backported to stable trees.
thanks,
greg k-h
Paul Moore
maintainerRe: [PATCH 1/3] lsm,io_uring: add LSM hooks for the new uring_cmd file op
Quoted 13 lines · Greg Kroah-Hartman
Quoted 12 lines · Paul Moore
From: Luis Chamberlain <[email protected]>
io-uring cmd support was added through
ee692a21e9bf("fs,io_uring:
add infrastructure for uring-cmd"), this extended the structfile_operationsto allow a new command which each subsystem can use to enable command passthrough. Add an LSM specific for the command passthrough which enables LSMs to inspect the command details.This was discussed long ago without no clear pointer for something conclusive, so this enables LSMs to at least reject this new file operation.
Quoted 8 lines
Quoted 3 lines
[0] https://lkml.kernel.org/r/[email protected]
Fixes ee692a21e9bf ("fs,io_uring: add infrastructure for uring-cmd")You are not "fixing" anything, you are adding new functionality. Careful with using "Fixes:" for something like this, you will trigger the bug-detection scripts and have to fend off stable bot emails for a long time for stuff that should not be backported to stable trees.
This patch, as well as the SELinux and (soon to come) Smack hook implementations, fix a LSM access control regression that occured when the IORING_OP_URING_CMD functionality was merged in v5.19. You may disagree about this being a regression Greg, but there are at least three people with their name on this patch that believe it is
important: Luis (patch author), Jens (io_uring maintainer), and myself (LSM, SELinux maintainer).
--
paul-moore.com
Greg Kroah-Hartman
maintainerRe: [PATCH 1/3] lsm,io_uring: add LSM hooks for the new uring_cmd file op
Quoted 24 lines · Paul Moore
Quoted 13 lines · Greg Kroah-Hartman
Quoted 12 lines · Paul Moore
From: Luis Chamberlain <[email protected]>
io-uring cmd support was added through
ee692a21e9bf("fs,io_uring:
add infrastructure for uring-cmd"), this extended the structfile_operationsto allow a new command which each subsystem can use to enable command passthrough. Add an LSM specific for the command passthrough which enables LSMs to inspect the command details.This was discussed long ago without no clear pointer for something conclusive, so this enables LSMs to at least reject this new file operation.
Quoted 8 lines
Quoted 3 lines
[0] https://lkml.kernel.org/r/[email protected]
Fixes ee692a21e9bf ("fs,io_uring: add infrastructure for uring-cmd")You are not "fixing" anything, you are adding new functionality. Careful with using "Fixes:" for something like this, you will trigger the bug-detection scripts and have to fend off stable bot emails for a long time for stuff that should not be backported to stable trees.
Quoted 7 lines
This patch, as well as the SELinux and (soon to come) Smack hook implementations, fix a LSM access control regression that occured when the
IORING_OP_URING_CMDfunctionality was merged in v5.19. You may disagree about this being a regression Greg, but there are at least three people with their name on this patch that believe it is
important: Luis (patch author), Jens (io_uringmaintainer), and myself (LSM, SELinux maintainer).
Ok, I'll let it be, but note that "Fixes:" tags do not mean that a patch will ever get backported to a stable tree, so I guess we don't have to worry about it :)
thanks,
greg k-h
Paul Moore
maintainerRe: [PATCH 1/3] lsm,io_uring: add LSM hooks for the new uring_cmd file op
Quoted 33 lines · Greg Kroah-Hartman
Quoted 24 lines · Paul Moore
Quoted 13 lines · Greg Kroah-Hartman
Quoted 12 lines · Paul Moore
From: Luis Chamberlain <[email protected]>
io-uring cmd support was added through
ee692a21e9bf("fs,io_uring:
add infrastructure for uring-cmd"), this extended the structfile_operationsto allow a new command which each subsystem can use to enable command passthrough. Add an LSM specific for the command passthrough which enables LSMs to inspect the command details.This was discussed long ago without no clear pointer for something conclusive, so this enables LSMs to at least reject this new file operation.
Quoted 8 lines
Quoted 3 lines
[0] https://lkml.kernel.org/r/[email protected]
Fixes ee692a21e9bf ("fs,io_uring: add infrastructure for uring-cmd")You are not "fixing" anything, you are adding new functionality. Careful with using "Fixes:" for something like this, you will trigger the bug-detection scripts and have to fend off stable bot emails for a long time for stuff that should not be backported to stable trees.
Quoted 7 lines
This patch, as well as the SELinux and (soon to come) Smack hook implementations, fix a LSM access control regression that occured when the
IORING_OP_URING_CMDfunctionality was merged in v5.19. You may disagree about this being a regression Greg, but there are at least three people with their name on this patch that believe it is
important: Luis (patch author), Jens (io_uringmaintainer), and myself (LSM, SELinux maintainer).
Quoted 3 lines
Ok, I'll let it be, but note that "Fixes:" tags do not mean that a patch will ever get backported to a stable tree, so I guess we don't have to worry about it :)
Ha! Now that's the *proper* LSM dismissing GregKH comment this thread was missing :)
Signature · 1 line
paul-moore.com
Paul Moore
maintainerPATCH 2/3 selinux: implement the security_uring_cmd() LSM hook
In reply to Paul MooreAdd a SELinux access control for the iouring IORING_OP_URING_CMD command. This includes the addition of a new permission in the existing "io_uring" object class: "cmd". The subject of the new permission check is the domain of the process requesting access, the object is the open file which points to the device/file that is the target of the IORING_OP_URING_CMD operation. A sample policy rule is shown below:
allow <domain> <file>:io_uring { cmd };
--- security/selinux/hooks.c | 24 ++++++++++++++++++++++++ security/selinux/include/classmap.h | 2 +- 2 files changed, 25 insertions(+), 1 deletion(-)
security/selinux/hooks.c +24 -0
@@ -91,6 +91,7 @@#include <uapi/linux/mount.h> #include <linux/fsnotify.h> #include <linux/fanotify.h> +#include <linux/io_uring.h> #include "avc.h" #include "objsec.h"
@@ -6987,6 +6988,28 @@ static int selinux_uring_sqpoll(void)return avc_has_perm (&selinux_state, sid, sid, SECCLASS_IO_URING, IO_URING__SQPOLL, NULL); } + +/** +* selinux_uring_cmd - check if IORING_OP_URING_CMD is allowed +* @ioucmd: the io_uring command structure +* +* Check to see if the current domain is allowed to execute an +* IORING_OP_URING_CMD against the device/file specified in @ioucmd. +* +*/ +static int selinux_uring_cmd (struct io_uring_cmd *ioucmd) +{ +struct file *file = ioucmd->file; +struct inode *inode =file_inode (file); +struct inode_security_struct *isec =selinux_inode (inode); +struct common_audit_data ad; + + ad.type =LSM_AUDIT_DATA_FILE ; + ad.u.file = file; + +return avc_has_perm (&selinux_state,current_sid (), isec->sid, +SECCLASS_IO_URING ,IO_URING__CMD , &ad); +}#endif /* CONFIG_IO_URING */ /*
@@ -7231,6 +7254,7 @@ static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {#ifdef CONFIG_IO_URING LSM_HOOK_INIT (uring_override_creds, selinux_uring_override_creds),LSM_HOOK_INIT (uring_sqpoll, selinux_uring_sqpoll), + LSM_HOOK_INIT(uring_cmd, selinux_uring_cmd), #endif/*
security/selinux/include/classmap.h +1 -1
@@ -253,7 +253,7 @@ const struct security_class_mapping secclass_map[] = { {"anon_inode" , { COMMON_FILE_PERMS, NULL } }, {"io_uring" , - { "override_creds", "sqpoll", NULL } }, + { "override_creds", "sqpoll", "cmd", NULL } }, { NULL } };
Greg Kroah-Hartman
maintainerRe: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
Quoted 11 lines · Paul Moore
Add a SELinux access control for the iouring
IORING_OP_URING_CMDcommand. This includes the addition of a new permission in the existing "io_uring" object class: "cmd". The subject of the new permission check is the domain of the process requesting access, the object is the open file which points to the device/file that is the target of theIORING_OP_URING_CMDoperation. A sample policy rule is shown below:allow <domain> <file>:io_uring { cmd };
This is not stable material as you are adding a new feature. Please read the stable documentation for what is and is not allowed.
thanks,
greg k-h
Paul Moore
maintainerRe: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
Quoted 9 lines · Greg Kroah-Hartman
Quoted 8 lines · Paul Moore
Add a SELinux access control for the iouring
IORING_OP_URING_CMDcommand. This includes the addition of a new permission in the existing "io_uring" object class: "cmd". The subject of the new permission check is the domain of the process requesting access, the object is the open file which points to the device/file that is the target of theIORING_OP_URING_CMDoperation. A sample policy rule is shown below:
Quoted 6 lines
Quoted 3 lines
allow <domain> <file>:io_uring { cmd };
This is not stable material as you are adding a new feature. Please read the stable documentation for what is and is not allowed.
Strongly disagree, see my comments on patch 1/3 in this patchset.
Signature · 1 line
paul-moore.com
Re: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
In reply to Paul MooreHey Paul
I realize that you have already sent this upstream but I wanted to share the Selinux part of the testing that we did to see if there is any feedback.
With my tests I see that the selinux_uring_cmd hook is run and it results in a "avc : denied" when I run it with selinux in permissive mode with an unpriviledged user. I assume that this is the expected behavior. Here is how I tested
*** With the patch:
* I ran the io_uring_passthrough.c test on a char device with an unpriviledged user.
* I took care of changing the permissions of /dev/ng0n1 to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* Prior to executing the test, I also placed a breakpoint inside selinux_uring_cmd to make sure that it was executed.
* This is the output of the audit when I executed the test:
[ 136.615924] audit: type=1400 audit(1662043624.701:94): avc: denied { create } for pid=263 comm="io_uring_passth" anonclass=[io_uring] scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1
[ 136.621036] audit: type=1300 audit(1662043624.701:94): arch=c000003e syscall=425 success=yes exit=3 a0=40 a1=7ffca29835a0 a2=7ffca29835a0 a3=561529be2300 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null)
[ 136.624812] audit: type=1327 audit(1662043624.701:94): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31
[ 136.626074] audit: type=1400 audit(1662043624.702:95): avc: denied { map } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1
[ 136.628012] audit: type=1400 audit(1662043624.702:95): avc: denied { read write } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1
[ 136.629873] audit: type=1300 audit(1662043624.702:95): arch=c000003e syscall=9 success=yes exit=140179765297152 a0=0 a1=1380 a2=3 a3=8001 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null)
[ 136.632415] audit: type=1327 audit(1662043624.702:95): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31
[ 136.633652] audit: type=1400 audit(1662043624.705:96): avc: denied { cmd } for pid=263 comm="io_uring_passth" path="/dev/ng0n1" dev="devtmpfs" ino=120 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:device_t tclass=io_uring permissive=1
[ 136.635384] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null)
[ 136.636863] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null)* From the output on time 136.633652 I see that the access should have been denied had selinux been enforcing.
* I also saw that the breakpoint hit.
*** Without the patch:
* I ran the io_uring_passthrough.c test on a char device with an unpriviledged user.
* I took care of changing the permissions of /dev/ng0n1 to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* There were no audit messages when I executed the test.
As with my smack tests I would really appreciate feecback on the approach I took to testing and it's validity.
Thx in advance
Best
Quoted 81 lines · Paul Moore
Add a SELinux access control for the iouring
IORING_OP_URING_CMDcommand. This includes the addition of a new permission in the existing "io_uring" object class: "cmd". The subject of the new permission check is the domain of the process requesting access, the object is the open file which points to the device/file that is the target of theIORING_OP_URING_CMDoperation. A sample policy rule is shown below:allow <domain> <file>:io_uring { cmd };
Fixes ee692a21e9bf ("fs,io_uring: add infrastructure for uring-cmd")Signed-off-by Paul Moore <[email protected]>--- security/selinux/hooks.c | 24 ++++++++++++++++++++++++ security/selinux/include/classmap.h | 2 +- 2 files changed, 25 insertions(+), 1 deletion(-)security/selinux/hooks.c +24
-0@@ -91,6 +91,7 @@#include <uapi/linux/mount.h> #include <linux/fsnotify.h> #include <linux/fanotify.h> +#include <linux/io_uring.h> #include "avc.h" #include "objsec.h" @@ -6987,6 +6988,28 @@ static int selinux_uring_sqpoll(void)return avc_has_perm (&selinux_state, sid, sid, SECCLASS_IO_URING, IO_URING__SQPOLL, NULL); } + +/** +* selinux_uring_cmd - check if IORING_OP_URING_CMD is allowed +* @ioucmd: the io_uring command structure +* +* Check to see if the current domain is allowed to execute an +* IORING_OP_URING_CMD against the device/file specified in @ioucmd. +* +*/ +static int selinux_uring_cmd (struct io_uring_cmd *ioucmd) +{ +struct file *file = ioucmd->file; +struct inode *inode =file_inode (file); +struct inode_security_struct *isec =selinux_inode (inode); +struct common_audit_data ad; + + ad.type =LSM_AUDIT_DATA_FILE ; + ad.u.file = file; + +return avc_has_perm (&selinux_state,current_sid (), isec->sid, +SECCLASS_IO_URING ,IO_URING__CMD , &ad); +}#endif /* CONFIG_IO_URING */ /* @@ -7231,6 +7254,7 @@ static struct security_hook_list selinux_hooks[] __lsm_ro_after_init = {#ifdef CONFIG_IO_URING LSM_HOOK_INIT (uring_override_creds, selinux_uring_override_creds),LSM_HOOK_INIT (uring_sqpoll, selinux_uring_sqpoll), + LSM_HOOK_INIT(uring_cmd, selinux_uring_cmd), #endif/* security/selinux/include/classmap.h +1
-1@@ -253,7 +253,7 @@ const struct security_class_mapping secclass_map[] = { {"anon_inode" , { COMMON_FILE_PERMS, NULL } }, {"io_uring" , - { "override_creds", "sqpoll", NULL } }, + { "override_creds", "sqpoll", "cmd", NULL } }, { NULL } };
Paul Moore
maintainerRe: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
Quoted 50 lines · Joel Granados
Hey Paul
I realize that you have already sent this upstream but I wanted to share the Selinux part of the testing that we did to see if there is any feedback.
With my tests I see that the
selinux_uring_cmdhook is run and it results in a "avc : denied" when I run it with selinux in permissive mode with an unpriviledged user. I assume that this is the expected behavior. Here is how I tested*** With the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* Prior to executing the test, I also placed a breakpoint insideselinux_uring_cmdto make sure that it was executed.
* This is the output of the audit when I executed the test:[ 136.615924] audit: type=1400 audit(1662043624.701:94): avc: denied { create } for pid=263 comm="io_uring_passth" anonclass=[io_uring] scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.621036] audit: type=1300 audit(1662043624.701:94): arch=c000003e syscall=425 success=yes exit=3 a0=40 a1=7ffca29835a0 a2=7ffca29835a0 a3=561529be2300 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.624812] audit: type=1327 audit(1662043624.701:94): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.626074] audit: type=1400 audit(1662043624.702:95): avc: denied { map } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.628012] audit: type=1400 audit(1662043624.702:95): avc: denied { read write } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.629873] audit: type=1300 audit(1662043624.702:95): arch=c000003e syscall=9 success=yes exit=140179765297152 a0=0 a1=1380 a2=3 a3=8001 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.632415] audit: type=1327 audit(1662043624.702:95): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.633652] audit: type=1400 audit(1662043624.705:96): avc: denied { cmd } for pid=263 comm="io_uring_passth" path="/dev/ng0n1" dev="devtmpfs" ino=120 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:device_t tclass=io_uring permissive=1 [ 136.635384] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null) [ 136.636863] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null)* From the output on time 136.633652 I see that the access should have been denied had selinux been enforcing.
* I also saw that the breakpoint hit.*** Without the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* There were no audit messages when I executed the test.
Quoted 2 lines
As with my smack tests I would really appreciate feecback on the approach I took to testing and it's validity.
Hi Joel,
Thanks for the additional testing and verification! Work like this is always welcome, regardless if the patch has already been merged upstream.
As far as you test approach is concerned, I think you are on the right track, I might suggest resolving the other SELinux/AVC denials you are seeing with your test application to help reduce the noise in the logs. Are you familiar with the selinux-testsuite (link below)?
* https://github.com/SELinuxProject/selinux-testsuite
Signature · 1 line
paul-moore.com
Re: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
Hey Paul
Quoted 51 lines · Paul Moore
Quoted 50 lines · Joel Granados
Hey Paul
I realize that you have already sent this upstream but I wanted to share the Selinux part of the testing that we did to see if there is any feedback.
With my tests I see that the
selinux_uring_cmdhook is run and it results in a "avc : denied" when I run it with selinux in permissive mode with an unpriviledged user. I assume that this is the expected behavior. Here is how I tested*** With the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* Prior to executing the test, I also placed a breakpoint insideselinux_uring_cmdto make sure that it was executed.
* This is the output of the audit when I executed the test:[ 136.615924] audit: type=1400 audit(1662043624.701:94): avc: denied { create } for pid=263 comm="io_uring_passth" anonclass=[io_uring] scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.621036] audit: type=1300 audit(1662043624.701:94): arch=c000003e syscall=425 success=yes exit=3 a0=40 a1=7ffca29835a0 a2=7ffca29835a0 a3=561529be2300 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.624812] audit: type=1327 audit(1662043624.701:94): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.626074] audit: type=1400 audit(1662043624.702:95): avc: denied { map } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.628012] audit: type=1400 audit(1662043624.702:95): avc: denied { read write } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.629873] audit: type=1300 audit(1662043624.702:95): arch=c000003e syscall=9 success=yes exit=140179765297152 a0=0 a1=1380 a2=3 a3=8001 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.632415] audit: type=1327 audit(1662043624.702:95): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.633652] audit: type=1400 audit(1662043624.705:96): avc: denied { cmd } for pid=263 comm="io_uring_passth" path="/dev/ng0n1" dev="devtmpfs" ino=120 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:device_t tclass=io_uring permissive=1 [ 136.635384] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null) [ 136.636863] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null)* From the output on time 136.633652 I see that the access should have been denied had selinux been enforcing.
* I also saw that the breakpoint hit.*** Without the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* There were no audit messages when I executed the test.
Quoted 8 lines
Quoted 2 lines
As with my smack tests I would really appreciate feecback on the approach I took to testing and it's validity.
Hi Joel,
Thanks for the additional testing and verification! Work like this is always welcome, regardless if the patch has already been merged upstream.
np
Quoted 7 lines
As far as you test approach is concerned, I think you are on the right track, I might suggest resolving the other SELinux/AVC denials you are seeing with your test application to help reduce the noise in the logs. Are you familiar with the selinux-testsuite (link below)?
Thx. Could not figure out how to remove the AVC from a quick look at the page, but I'll probably figures something out :).
ATM, I'm doing a performance test on the io_uring_passtrhough path to see how much, if any, perf we loose.
Thx again
Best
Joel
Quoted 3 lines
Signature · 2 lines
-- paul-moore.com
Re: [PATCH 2/3] selinux: implement the security_uring_cmd() LSM hook
Hello.
Took the time to re-run my performance tests on the current LSM patch for io_uring_cmd. I'll explain what I did and then give the results:
How I ran it:
I took a version of the kernel with the patch a0d2212773d1 and then compiled two versions: The first was the vanilla kernel and the other was the same except for the LSM hook called from io_uring_cmd removed. Same kernel configurations. For my tests I used one of the test files from FIO called t/io_uring.c which is basically a READ test. I ran my tests on both an nvme device and the null device (/dev/null). For the first I did not change io_uring.c and for the second I replaced the admin calls with dummy data that was not really needed for testing with /dev/null. These are the arguments I used for the test "./t/io_uring -b4096 -d128 -c32 -s32 -p0 -F1 -B0 -O0 -P1 -u1 -n1" Finally, I'm taking the max of several samples.
Results:
+-------------------------------------+------------------------+-----------------------+ | Name | /dev/ng0n1 (BW: MiB/s) | /dev/null (BW: GiB/s) | +-------------------------------------+------------------------+-----------------------+ | (A) for-next (vanilla) | 1341 | 30.16 | | (B) for-next (no io_uring_cmd hook) | 1362 | 40.61 | | [1-(A/B)] * 100 | 1.54185022 | 25.732578183 | +-------------------------------------+------------------------+-----------------------+
So on a device (dev/ng0n1) there is a 1% performance difference on a read. Whereas on the null device (dev/null) there is a 25% difference on a read.
This difference is interesting and expected as there is a lot more stuff happening when we go through the actual device.
Best
Joel
Quoted 86 lines · Joel Granados
Hey Paul
Quoted 51 lines · Paul Moore
Quoted 50 lines · Joel Granados
Hey Paul
I realize that you have already sent this upstream but I wanted to share the Selinux part of the testing that we did to see if there is any feedback.
With my tests I see that the
selinux_uring_cmdhook is run and it results in a "avc : denied" when I run it with selinux in permissive mode with an unpriviledged user. I assume that this is the expected behavior. Here is how I tested*** With the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* Prior to executing the test, I also placed a breakpoint insideselinux_uring_cmdto make sure that it was executed.
* This is the output of the audit when I executed the test:[ 136.615924] audit: type=1400 audit(1662043624.701:94): avc: denied { create } for pid=263 comm="io_uring_passth" anonclass=[io_uring] scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.621036] audit: type=1300 audit(1662043624.701:94): arch=c000003e syscall=425 success=yes exit=3 a0=40 a1=7ffca29835a0 a2=7ffca29835a0 a3=561529be2300 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.624812] audit: type=1327 audit(1662043624.701:94): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.626074] audit: type=1400 audit(1662043624.702:95): avc: denied { map } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.628012] audit: type=1400 audit(1662043624.702:95): avc: denied { read write } for pid=263 comm="io_uring_passth" path="anon_inode:[io_uring]" dev="anon_inodefs" ino=11715 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:kernel_t tclass=anon_inode permissive=1 [ 136.629873] audit: type=1300 audit(1662043624.702:95): arch=c000003e syscall=9 success=yes exit=140179765297152 a0=0 a1=1380 a2=3 a3=8001 items=0 ppid=252 pid=263 auid=1001 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 tty=pts1 ses=3 comm="io_uring_passth" exe="/mnt/src/liburing/test/io_uring_passthrough.t" subj=system_u:system_r:kernel_t key=(null) [ 136.632415] audit: type=1327 audit(1662043624.702:95): proctitle=2F6D6E742F7372632F6C69627572696E672F746573742F696F5F7572696E675F706173737468726F7567682E74002F6465762F6E67306E31 [ 136.633652] audit: type=1400 audit(1662043624.705:96): avc: denied { cmd } for pid=263 comm="io_uring_passth" path="/dev/ng0n1" dev="devtmpfs" ino=120 scontext=system_u:system_r:kernel_t tcontext=system_u:object_r:device_t tclass=io_uring permissive=1 [ 136.635384] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null) [ 136.636863] audit: type=1336 audit(1662043624.705:96): uring_op=46 items=0 ppid=252 pid=263 uid=1001 gid=1002 euid=1001 suid=1001 fsuid=1001 egid=1002 sgid=1002 fsgid=1002 subj=system_u:system_r:kernel_t key=(null)* From the output on time 136.633652 I see that the access should have been denied had selinux been enforcing.
* I also saw that the breakpoint hit.*** Without the patch:
* I ran theio_uring_passthrough.ctest on a char device with an unpriviledged user.
* I took care of changing the permissions of/dev/ng0n1to 666 prior to any testing.
* made sure that Selinux was in permissive mode.
* Made sure to have audit activated by passing "audit=1" to the kernel
* After noticing that some audit messages where getting lost I upped the backlog limit to 256
* There were no audit messages when I executed the test.Quoted 8 lines
Quoted 2 lines
As with my smack tests I would really appreciate feecback on the approach I took to testing and it's validity.
Hi Joel,
Thanks for the additional testing and verification! Work like this is always welcome, regardless if the patch has already been merged upstream.
np
Quoted 7 lines
As far as you test approach is concerned, I think you are on the right track, I might suggest resolving the other SELinux/AVC denials you are seeing with your test application to help reduce the noise in the logs. Are you familiar with the selinux-testsuite (link below)?
Thx. Could not figure out how to remove the AVC from a quick look at the page, but I'll probably figures something out :).
ATM, I'm doing a performance test on the
io_uring_passtrhoughpath to see how much, if any, perf we loose.Thx again
Best
Joel
Quoted 3 lines
Signature · 2 lines
-- paul-moore.com
Paul Moore
maintainerPATCH 3/3 /dev/null: add IORING_OP_URING_CMD support
In reply to Paul MooreThis patch adds support for the io_uring command pass through, aka IORING_OP_URING_CMD, to the /dev/null driver. As with all of the /dev/null functionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simple IORING_OP_URING_CMD test device that doesn't require any special hardware.
--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)
drivers/char/mem.c +6 -0
@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +} +static ssize_t read_iter_zero (struct kiocb *iocb,struct iov_iter *iter) {size_t written =0 ;
@@ -663,6 +668,7 @@ static const struct file_operations null_fops = { .read_iter = read_iter_null, .write_iter = write_iter_null, .splice_write = splice_write_null, + .uring_cmd = uring_cmd_null, };static const struct file_operations __maybe_unused port_fops = {
Re: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0
drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}
This would be better as:
return IOU_OK;
using the proper return values for the uring_cmd hook. With that:
Signature · 1 line
Jens Axboe
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 30 lines · Jens Axboe
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}This would be better as:
Quoted 3 lines
return IOU_OK;using the proper return values for the
uring_cmdhook.
The only problem I see with that is that IOU_OK is defined under io_uring/io_uring.h and not include/linux/io_uring.h so the #include macro is kinda ugly:
#include "../../io_uring/io_uring.h"
I'm not sure I want to submit that upstream looking like that. Are you okay with leaving the return code as 0 for now and changing it at a later date? I'm trying to keep this patchset relatively small since we are in the -rcX stage, but if you're okay with a simple cut-n-paste of the enum to linux/io_uring.h I can do that.
Quoted 3 lines
With that:
Acked-by Jens Axboe <[email protected]>
Signature · 1 line
paul-moore.com
Jens Axboe
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 39 lines · Paul Moore
Quoted 30 lines · Jens Axboe
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}This would be better as:
Quoted 3 lines
return IOU_OK;using the proper return values for the
uring_cmdhook.The only problem I see with that is that
IOU_OKis defined underio_uring/io_uring.hand notinclude/linux/io_uring.hso the #include macro is kinda ugly:
Quoted 7 lines
#include "../../io_uring/io_uring.h"I'm not sure I want to submit that upstream looking like that. Are you okay with leaving the return code as 0 for now and changing it at a later date? I'm trying to keep this patchset relatively small since we are in the -rcX stage, but if you're okay with a simple cut-n-paste of the enum to
linux/io_uring.hI can do that.
Ugh yes, that should move into the general domain. Yeah I'm fine with it as it is, we can fix that up (and them nvme as well) at a later point.
Signature · 1 line
Jens Axboe
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 42 lines · Jens Axboe
Quoted 41 lines · Paul Moore
Quoted 30 lines · Jens Axboe
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}This would be better as:
Quoted 3 lines
return IOU_OK;using the proper return values for the
uring_cmdhook.The only problem I see with that is that
IOU_OKis defined underio_uring/io_uring.hand notinclude/linux/io_uring.hso the #include macro is kinda ugly:#include "../../io_uring/io_uring.h"
Quoted 8 lines
Quoted 5 lines
I'm not sure I want to submit that upstream looking like that. Are you okay with leaving the return code as 0 for now and changing it at a later date? I'm trying to keep this patchset relatively small since we are in the -rcX stage, but if you're okay with a simple cut-n-paste of the enum to
linux/io_uring.hI can do that.Ugh yes, that should move into the general domain. Yeah I'm fine with it as it is, we can fix that up (and them nvme as well) at a later point.
Okay, sounds good, I'll leave it as-is. Is it okay to still add your ACK?
Signature · 1 line
paul-moore.com
Jens Axboe
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 49 lines · Paul Moore
Quoted 48 lines · Jens Axboe
Quoted 46 lines · Paul Moore
Quoted 30 lines · Jens Axboe
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}This would be better as:
Quoted 3 lines
return IOU_OK;using the proper return values for the
uring_cmdhook.The only problem I see with that is that
IOU_OKis defined underio_uring/io_uring.hand notinclude/linux/io_uring.hso the #include macro is kinda ugly:#include "../../io_uring/io_uring.h"I'm not sure I want to submit that upstream looking like that. Are you okay with leaving the return code as 0 for now and changing it at a later date? I'm trying to keep this patchset relatively small since we are in the -rcX stage, but if you're okay with a simple cut-n-paste of the enum to
linux/io_uring.hI can do that.
Quoted 4 lines
Quoted 2 lines
Ugh yes, that should move into the general domain. Yeah I'm fine with it as it is, we can fix that up (and them nvme as well) at a later point.
Okay, sounds good, I'll leave it as-is. Is it okay to still add your ACK?
Yep, all things considered, for 6.0 I think that's the way to go.
Signature · 1 line
Jens Axboe
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 51 lines · Jens Axboe
Quoted 49 lines · Paul Moore
Quoted 48 lines · Jens Axboe
Quoted 46 lines · Paul Moore
Quoted 30 lines · Jens Axboe
Quoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 486
Quoted 7 lines
drivers/char/mem.c +4
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ; +}This would be better as:
Quoted 3 lines
return IOU_OK;using the proper return values for the
uring_cmdhook.The only problem I see with that is that
IOU_OKis defined underio_uring/io_uring.hand notinclude/linux/io_uring.hso the #include macro is kinda ugly:#include "../../io_uring/io_uring.h"I'm not sure I want to submit that upstream looking like that. Are you okay with leaving the return code as 0 for now and changing it at a later date? I'm trying to keep this patchset relatively small since we are in the -rcX stage, but if you're okay with a simple cut-n-paste of the enum to
linux/io_uring.hI can do that.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Ugh yes, that should move into the general domain. Yeah I'm fine with it as it is, we can fix that up (and them nvme as well) at a later point.
Okay, sounds good, I'll leave it as-is. Is it okay to still add your ACK?
Yep, all things considered, for 6.0 I think that's the way to go.
Great, thanks.
Signature · 1 line
paul-moore.com
Greg Kroah-Hartman
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
In reply to Paul MooreQuoted 19 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0
drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 485
Quoted 6 lines
drivers/char/mem.c +3
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); } +static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ;
If a callback just returns 0, that implies it is not needed at all and can be removed and then you are back at the original file before your commit :)
thanks,
greg k-h
Jens Axboe
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 23 lines · Greg Kroah-Hartman
Quoted 22 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); }
Quoted 7 lines
drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 485
Quoted 3 lines
drivers/char/mem.c +3
-0+static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ;If a callback just returns 0, that implies it is not needed at all and can be removed and then you are back at the original file before your commit :)
In theory you are correct, but the empty hook is needed so that submitting an io_uring cmd to the file type is attempted. If not it's just errored upfront.
Paul, is it strictly needed to test the selinux uring cmd policy? If the operation would've been attempted but null doesn't support it, you'd get -1/EOPNOTSUPP - and supposedly you'd get EACCES/EPERM or something if it's filtered?
Signature · 1 line
Jens Axboe
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 32 lines · Jens Axboe
Quoted 23 lines · Greg Kroah-Hartman
Quoted 22 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Cc Arnd Bergmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Signed-off-by Paul Moore <[email protected]>--- drivers/char/mem.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/char/mem.c +0
-0@@ -480,6 +480,11 @@ static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,return splice_from_pipe (pipe, out, ppos, len, flags, pipe_to_null); }Quoted 7 lines
drivers/char/mem.c static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out, · line 485
Quoted 3 lines
drivers/char/mem.c +3
-0+static int uring_cmd_null (struct io_uring_cmd *ioucmd,unsigned int issue_flags) +{ +return 0 ;If a callback just returns 0, that implies it is not needed at all and can be removed and then you are back at the original file before your commit :)
Quoted 8 lines
In theory you are correct, but the empty hook is needed so that submitting an
io_uringcmd to the file type is attempted. If not it's just errored upfront.Paul, is it strictly needed to test the selinux uring cmd policy? If the operation would've been attempted but null doesn't support it, you'd get -1/EOPNOTSUPP - and supposedly you'd get EACCES/EPERM or something if it's filtered?
I haven't built a kernel without this patch to test, but yes, you are probably correct that it wouldn't be strictly necessary, but considering the extreme simplicity of this addition, what is the real harm here? Wouldn't it be nice to have a rather simple IORING_OP_URING_CMD target?
Also, just so we are clear, I didn't mark this patch with the stable/fixes tags because I don't believe this should go into -stable; while I believe it is a nice addition, it is definitely not critical.
Signature · 1 line
paul-moore.com
Greg Kroah-Hartman
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
In reply to Paul MooreQuoted 6 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.
Also, shouldn't you document this somewhere?
At least in the code itself saying "this is here so that /dev/null works as a io_uring sink" or something like that? Otherwise it just looks like it does nothing at all.
thanks,
greg k-h
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 8 lines · Greg Kroah-Hartman
Quoted 6 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.
Quoted 5 lines
Also, shouldn't you document this somewhere?
At least in the code itself saying "this is here so that
/dev/nullworks as aio_uringsink" or something like that? Otherwise it just looks like it does nothing at all.
What about read_null() and write_null()? I can definitely add a comment (there is no /dev/null documentation in the kernel source tree that I can see), but there is clearly precedence for /dev/null having "do nothing" file_operations functions. If nothing else, it's pretty much in the *name*; we're adding the "uring_cmd_null()" member function to the "null_fops" struct for the "null" device ... come on Greg :)
Signature · 1 line
paul-moore.com
Greg Kroah-Hartman
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 12 lines · Paul Moore
Quoted 10 lines · Greg Kroah-Hartman
Quoted 6 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Also, shouldn't you document this somewhere?
Quoted 8 lines
Quoted 3 lines
At least in the code itself saying "this is here so that
/dev/nullworks as aio_uringsink" or something like that? Otherwise it just looks like it does nothing at all.What about
read_null()andwrite_null()? I can definitely add a comment (there is no/dev/nulldocumentation in the kernel source tree that I can see), but there is clearly precedence for/dev/nullhaving "do nothing"file_operationsfunctions.
Yes, they should "do nothing". write_null() does report that it consumed everything, why doesn't this function have to also do that?
thanks,
greg k-h
Paul Moore
maintainerRe: [PATCH 3/3] /dev/null: add IORING_OP_URING_CMD support
Quoted 17 lines · Greg Kroah-Hartman
Quoted 16 lines · Paul Moore
Quoted 13 lines · Greg Kroah-Hartman
Quoted 6 lines · Paul Moore
This patch adds support for the
io_uringcommand pass through, akaIORING_OP_URING_CMD, to the/dev/nulldriver. As with all of the/dev/nullfunctionality, the implementation is just a simple sink where commands go to die, but it should be useful for developers who need a simpleIORING_OP_URING_CMDtest device that doesn't require any special hardware.Also, shouldn't you document this somewhere?
At least in the code itself saying "this is here so that
/dev/nullworks as aio_uringsink" or something like that? Otherwise it just looks like it does nothing at all.
Quoted 6 lines
Quoted 4 lines
What about
read_null()andwrite_null()? I can definitely add a comment (there is no/dev/nulldocumentation in the kernel source tree that I can see), but there is clearly precedence for/dev/nullhaving "do nothing"file_operationsfunctions.Yes, they should "do nothing".
Right, I don't think anyone was disputing that. You were asking for a comment for the new function that effectively says "this function does nothing", which seems a little silly given the simplicity of the function, the name, and the context of it all.
Quoted 2 lines
write_null()does report that it
consumed everything, why doesn't this function have to also do that?
Because a file write (file_operations->write) and a IORING_OP_URING_CMD (file_operations->uring_cmd) are fundamentally different operations; uring_cmd_null() returns 0, which is the success return code for this file op (not to mention a significant number of kernel functions that return an int).
Signature · 1 line
paul-moore.com
Paul Moore
maintainerQuoted 31 lines · Paul Moore
This patchset includes three patches: one to add a new LSM hook for the
IORING_OP_URING_CMDoperation, one to add the SELinux implementation for the new hook, and one to enableIORING_OP_URING_CMDfor/dev/null. The last patch, the/dev/nullsupport, is obviously not critical but it makes testing so much easier and I believe is in keeping with the general motivation behind/dev/null.Luis' patch has already been vetted by Jens and the
io_uringfolks, so the only new bits are the SELinux implementation and the trivial/dev/nullimplementation ofIORING_OP_URING_CMD. Assuming no one has any objections over the next few days, I'll plan on sending this up to Linus during the v6.0-rcX cycle.I believe Casey is also currently working on Smack support for the
IORING_OP_URING_CMDhook, and as soon as he is ready I can add it to this patchset (or Casey can send it up himself).-Paul
---
Luis Chamberlain (1):
lsm,io_uring: add LSM hooks for the newuring_cmdfile opPaul Moore (2):
selinux: implement thesecurity_uring_cmd()LSM hook/dev/null: addIORING_OP_URING_CMDsupport
Quoted 9 lines
drivers/char/mem.c | 6 ++++++ include/linux/lsm_hook_defs.h | 1 + include/linux/lsm_hooks.h | 3 +++ include/linux/security.h | 5 +++++ io_uring/uring_cmd.c | 5 +++++ security/security.c | 4 ++++ security/selinux/hooks.c | 24 ++++++++++++++++++++++++ security/selinux/include/classmap.h | 2 +- 8 files changed, 49 insertions(+), 1 deletion(-)
FYI, I just merged this into lsm/stable-6.0 and once the automated testing completes and we sort out the Smack patch I'll send this up to Linus.
Signature · 1 line
paul-moore.com