Mount a tmpfs at /dev/shm in the sandbox. #3263
Conversation
|
Looks like it's failing the same way in CI as it s on my machine: |
|
According to the man page for |
|
I added |
|
Oh woops, I thought I looked to see if you had added it and didn't see it, but given it's the very first thing in the diff I don't know how I could have missed it. But... I dunno then. |
|
As a note, this will also help get .NET Core working on Sandstorm: dotnet/runtime#2534 (comment) |
...To get shm_open() and friends working. Fixes sandstorm-io#3196.
|
Fixed. Just had a bogus leading slash in front of |
zenhack
changed the title
|
@ocdtrekkie, feel free to tag this one ready-for-review. |
| // so we don't have to strictly separate their storage capacity. We could mount | ||
| // a single tmpfs somewhere invisible, create subdirectories, and then bind-mount | ||
| // them to their final destinations. | ||
| KJ_SYSCALL(mkdir("dev/shm", 0700)); |
There was a problem hiding this comment.
It looks like the /dev filesystem itself is created with 0755 permission, any particular reason to use 0700 here? I guess it ultimately doesn't mater since there's only one user in the sandbox and they own everything?
|
Quoting Kenton Varda (2020-04-05 18:55:11)
It looks like the /dev filesystem itself is created with 0755
permission, any particular reason to use 0700 here? I guess it
ultimately doesn't mater since there's only one user in the sandbox and
they own everything?
I tend to default to being conservative with permissions, but yeah it
probably doesn't matter.
|
...To get shm_open() and friends working.
Fixes #3196.
Note that the tests are currently not passing on my machine -- for some reason
shm_openis returningENOSYS. Based on the links @abliss posted at:#3196 (comment)
and the fact that there are two alternate implementations of
shm_open, depending on build flags:https://sourceware.org/git/?p=glibc.git;a=blob;f=rt/shm_open.c;h=b4623d3cdaf5a26e4fd795a866c12aa1b29eaa9a;hb=HEAD
...one possible explanation is that my system's glibc is built without support for this. So I'd be curious as to whether the tests pass for others. Grepping through the postgres source suggests it has some fallbacks if shm_open is not available, so another possibility is that the errors @orblivion saw re:
/dev/shmweren't actually going throughshm_open.We should investigate and work out what's going on there before merging this.