1.7 -> tmpfs vs fork or execvp

[nmwael]

If a parent enclave writes to tmpfs and then creates a child with either fork or execvp, does the instance of tmpfs survive or is a new instance?

[dimakuv]

Current tmpfs implementation says this:

gramine/libos/src/fs/tmpfs/fs.c

Lines 8 to 10 in a20d5b8

	* This file contains code for implementation of 'tmpfs' filesystem. The tmpfs files are *not*
	* cloned during fork/clone (except for files currently open) and cannot be synchronized between
	* processes.

So, the child process/enclave will only get those tmpfs files that are currently being open.

To illustrate with an example, file1 will be visible in the child, but file2 will not be visible in the child:

int fd1 = open("tmpfs/file1");
write(fd1, "some text");

int fd2 = open("tmpfs/file2");
write(fd2, "some other text");
close(fd2);

fork();

// child gets only fd1 and its associated file "file1"

[nmwael]

Ahh, ok. Thanks for the explanation and "workaround"

[nmwael]

what happens if I close fd1 after fork, just to be clear?

int fd1 = open("tmpfs/file1");
write(fd1, "some text");

int fd2 = open("tmpfs/file2");
write(fd2, "some other text");
close(fd2);

fork();
close(fd1);

[dimakuv]

what happens if I close fd1 after fork, just to be clear?

Nothing bad. The file "file1" with its contents will still be there in the child.