uipc_shm: Implements fspacectl(2) support

[khng]

Summary

This implements fspacectl(2) support on shared memory objects. The
semantic of SPACECTL_DEALLOC is equivalent to clearing the backing
store and free the pages within the affected range. If the call
succeeds, subsequent reads on the affected range return all zero.

tests/sys/posixshm/posixshm_tests.c is expanded to include a
fspacectl(2) functional test.

Sponsored by: The FreeBSD Foundation

[khng]

Return when shm_partial_page_invalidate fails.

[khng]

Returns the offset shm_deallocate was failing at.

[kib]

sys/kern/uipc_shm.c

634

The note about vm object lock is really redundant. The assert at the start of the static function already tells the story.

646

I suspect this is not a good assert. You want to assert that base + end is on the same page as base.

[khng]

Comment and assert suggestions from kib@

[kib]

I think that DEALLOC is same as write, so you should honor F_SEAL_WRITE.
I am not sure should DEALLOC be disabled if there are kernel mappings. You may end up removing kernel-wired pages from the object' queue.

[markj]

In D31490#709942, @kib wrote:

I am not sure should DEALLOC be disabled if there are kernel mappings. You may end up removing kernel-wired pages from the object' queue.

vm_object_page_remove() will not free wired pages from the object, it only removes managed mappings and marks the page invalid.

sys/kern/uipc_shm.c

1898

rl_cookie and td are unused.

1935

Suppose off == OFF_MAX - PAGE_MASK and len == PAGE_MASK. Then startofs == 0 && pi == piend && pistart == piend, so this function will do nothing, I believe.

tests/sys/posixshm/posixshm_test.c

1156

Extra newline.

[kevans]

sys/kern/uipc_shm.c

1984

Do we want this to be an __assert_unreachable() instead (with error previously initialized to EINVAL)? This is generally a nop if we were passed a cmd that we didn't handle, so I don't know that it's worth taking down the machine when we could toss back an error instead on non-debug kernels.

tests/sys/posixshm/posixshm_test.c

3

Kind of nitpicky, but to curtail someone pointing it out after the fact: the 'All rights reserved.' should be moved to the line above it with rwatson unless you've obtained his permission to drop it.

[imp]

tests/sys/posixshm/posixshm_test.c

3

I'm pretty sure rwatson said it was OK to drop all rights reserved on his code.

[khng]

• rl_cookie and td are unused, remove them
• off == OFF_MAX - PAGE_MASK and len == PAGE_MASK can be fixed by simply using vm_ooffset_t.
• Remove extra newline.
• panic -> __assert_unreachable for unreachable path in shm_fspacectl
• Restore copyright notice.

[khng]

tests/sys/posixshm/posixshm_test.c

3

I will play safe for now.

[kevans]

sys/kern/uipc_shm.c

1959

I think this should be EINVAL in case we hit the unreachable branch on a non-debug kernel. All of the success paths will set it so.

[khng]

initialize error to EINVAL in shm_fspacectl

[kib]

Something similar should work and should be done to tmpfs.

[khng]

• Fix shm_partial_page_invalidation assert condition