pussuw opened a new pull request, #16326:
URL: https://github.com/apache/nuttx/pull/16326
## Summary
There are two race conditions present in the current implementation:
1. File is partially open / getting closed and dup2/3 is called
- This condition arises when one thread opens and a second thread duplicates
a fixed file descriptor number that happens to collide with the file being
opened.
2. File I/O is active but someone calls close / dup2/3 on the descriptor.
- This condition occurs when one thread writes and another calls close or
dup2/3 on an open descriptor. If the write operation is not finished before
closing the file, it will violate the NuttX f_refs counter which protects the
file I/O from getting its file struct wiped from under its feet.
Fix these extremely rare race conditions by:
1. Returning an error (-EBUSY) to the user
2. Waiting for the file I/O to finish before closing and re-using the
descriptor "newfd"
This would not be an issue if only the userspace would panic due to totally
nonsensical use of dup2/3, but in our case the kernel will crash as the file
description behind newfd changes unexpectedly -> when the device tries to
access the file description it cannot find its inode nor its private data.
## Impact
Attempt to fix dup2() and dup3() system calls. Currently they have a race
condition that causes the kernel to crash.
No impact to other APIs, documentation, user etc.
## Testing
MPFS target with BUILD_KERNEL and several processes / threads using dup2,
write. Before this patch we get random crashes net_local and socket APIs due to
disappearing file descriptors. With this patch we no longer crash.
rv-virt:ksmp64 ostest
rv-virt:knsh64 ostest
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscr...@nuttx.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
