wine fails to start with seccomp updates for v5.9-rc1
A message that is not in the archive (sent off-list or not yet received): 1596812929.lz7fuo8r2w.none.ref@localhost
Hi,
On Linus' master, wine fails to start with the following error:
wine client error:0: write: Bad file descriptor
This issue is not present on 5.8. It appears to be caused by failure to write to a pipe FD received via SCM_RIGHTS. Therefore, I tried reverting 9ecc6ea491f0, which resolved the issue.
Thanks,
Alex.
Quoted 8 lines · On Fri, Aug 7, 2020 at 8:19 AM Alex Xu (Hello71) <[email protected]> wrote:
On Linus' master, wine fails to start with the following error:
wine client error:0: write: Bad file descriptor
This issue is not present on 5.8. It appears to be caused by failure to write to a pipe FD received via
SCM_RIGHTS. Therefore, I tried reverting9ecc6ea491f0, which resolved the issue.
Would you mind trying to bisect exactly where it happens?
I don't think any of the commits in that pull are supposed to change semantics, and while reverting the whole merge shows that yes, that's what brought in the problems, it would be good to pinpoint just which change breaks so that we can fix just that thing.
Kees, ideas?
Linus
Quoted 4 lines · Linus Torvalds
Quoted 3 lines · On Fri, Aug 7, 2020 at 8:19 AM Alex Xu (Hello71) <[email protected]> wrote:
On Linus' master, wine fails to start with the following error:
Quoted 8 lines
Quoted 5 lines
wine client error:0: write: Bad file descriptor
This issue is not present on 5.8. It appears to be caused by failure to write to a pipe FD received via
SCM_RIGHTS. Therefore, I tried reverting9ecc6ea491f0, which resolved the issue.Would you mind trying to bisect exactly where it happens?
This report [1] seemed related and pointed out at c0029de50982 ("net/scm:
Regularize compat handling of scm_detach_fds()"). The use of CMSG_USER_DATA instead of CMSG_COMPAT_DATA seems fishy.
Alex, can you try applying the patch below?
Cascardo.
[1] https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216156.html
Quoted 8 lines
I don't think any of the commits in that pull are supposed to change semantics, and while reverting the whole merge shows that yes, that's what brought in the problems, it would be good to pinpoint just which change breaks so that we can fix just that thing.
Kees, ideas?
Linus
---
net/compat.c +1 -1
@@ -294,7 +294,7 @@ void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm) (struct compat_cmsghdr __user *)msg->msg_control;unsigned int o_flags = (msg->msg_flags &MSG_CMSG_CLOEXEC ) ?O_CLOEXEC :0 ;int fdmax =min_t (int,scm_max_fds_compat (msg), scm->fp->count); -int __user *cmsg_data =CMSG_USER_DATA (cm); +int __user *cmsg_data =CMSG_COMPAT_DATA (cm);int err =0 , i;for (i =0 ; i < fdmax; i++) {
Quoted 11 lines · Thadeu Lima de Souza Cascardo
Quoted 10 lines · Linus Torvalds
Quoted 8 lines · On Fri, Aug 7, 2020 at 8:19 AM Alex Xu (Hello71) <[email protected]> wrote:
On Linus' master, wine fails to start with the following error:
wine client error:0: write: Bad file descriptor
This issue is not present on 5.8. It appears to be caused by failure to write to a pipe FD received via
SCM_RIGHTS. Therefore, I tried reverting9ecc6ea491f0, which resolved the issue.
Quoted 6 lines
Quoted 2 lines
Would you mind trying to bisect exactly where it happens?
This report [1] seemed related and pointed out at
c0029de50982("net/scm:
Regularize compat handling ofscm_detach_fds()"). The use ofCMSG_USER_DATAinstead ofCMSG_COMPAT_DATAseems fishy.
Argh; yes. Thank you for finding that! That's what I get for trying to regularize the compat path. :(
Quoted 21 lines
Alex, can you try applying the patch below?
Cascardo.
[1] https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216156.html
Quoted 8 lines
I don't think any of the commits in that pull are supposed to change semantics, and while reverting the whole merge shows that yes, that's what brought in the problems, it would be good to pinpoint just which change breaks so that we can fix just that thing.
Kees, ideas?
Linus
---
net/compat.c +0
-0
net/compat.c void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm) · line 300
Quoted 8 lines
net/compat.c +1
-1@@ -294,7 +294,7 @@ void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm) (struct compat_cmsghdr __user *)msg->msg_control;unsigned int o_flags = (msg->msg_flags &MSG_CMSG_CLOEXEC ) ?O_CLOEXEC :0 ;int fdmax =min_t (int,scm_max_fds_compat (msg), scm->fp->count); -int __user *cmsg_data =CMSG_USER_DATA (cm); +int __user *cmsg_data =CMSG_COMPAT_DATA (cm);int err =0 , i;for (i =0 ; i < fdmax; i++) {
That almost certainly will fix the problem.
Signature · 1 line
Kees Cook
Alex Xu (Hello71)
Tested-byExcerpts from Thadeu Lima de Souza Cascardo's message of August 7, 2020 1:36 pm:
Quoted 40 lines · Linus Torvalds
Quoted 12 lines · On Fri, Aug 7, 2020 at 8:19 AM Alex Xu (Hello71) <[email protected]> wrote:
Quoted 8 lines
On Linus' master, wine fails to start with the following error:
wine client error:0: write: Bad file descriptor
This issue is not present on 5.8. It appears to be caused by failure to write to a pipe FD received via
SCM_RIGHTS. Therefore, I tried reverting9ecc6ea491f0, which resolved the issue.Would you mind trying to bisect exactly where it happens?
This report [1] seemed related and pointed out at
c0029de50982("net/scm:
Regularize compat handling ofscm_detach_fds()"). The use ofCMSG_USER_DATAinstead ofCMSG_COMPAT_DATAseems fishy.Alex, can you try applying the patch below?
Cascardo.
[1] https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216156.html
Quoted 8 lines
I don't think any of the commits in that pull are supposed to change semantics, and while reverting the whole merge shows that yes, that's what brought in the problems, it would be good to pinpoint just which change breaks so that we can fix just that thing.
Kees, ideas?
Linus
---
net/compat.c +0
-0
net/compat.c void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm) · line 300
Quoted 8 lines
net/compat.c +1
-1@@ -294,7 +294,7 @@ void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm) (struct compat_cmsghdr __user *)msg->msg_control;unsigned int o_flags = (msg->msg_flags &MSG_CMSG_CLOEXEC ) ?O_CLOEXEC :0 ;int fdmax =min_t (int,scm_max_fds_compat (msg), scm->fp->count); -int __user *cmsg_data =CMSG_USER_DATA (cm); +int __user *cmsg_data =CMSG_COMPAT_DATA (cm);int err =0 , i;for (i =0 ; i < fdmax; i++) {
Yes, this seems to work.
Thanks!