Implementing VOP_DEALLOCATE to allow hole-punching in the same manner as
POSIX shared memory's fspacectl(SPACECTL_DEALLOC) support.
Sponsored by: The FreeBSD Foundation
Details
Details
- Reviewers
markj kib - Commits
- rG8d7cd10ba633: tmpfs: Implement VOP_DEALLOCATE
Diff Detail
Diff Detail
- Repository
- rG FreeBSD src repository
- Lint
Lint Not Applicable - Unit
Tests Not Applicable
Event Timeline
| sys/fs/tmpfs/tmpfs_subr.c | ||
|---|---|---|
| 1798 | I would assert that tn_type == VREG. | |
| 1816 | There should be a blank line before multi-line comment | |
| 1824 | error != 0 | |
| 1834 | vm_object_page_remove() leaves wired pages alone. This is fine for truncation, but probably not for deallocation. I think you need to do a pass over the leftover pages and zero them? BTW, similar issue probably exists with shmfd. | |
| 1846 | Same | |
| sys/fs/tmpfs/tmpfs_vnops.c | ||
| 703 | These are rather silly asserts from GSoC times, they are a) not useful b) done wrong (ASSERT_VOP_LOCKED should be used if anything) | |
| sys/fs/tmpfs/tmpfs_subr.c | ||
|---|---|---|
| 1834 | Replying to myself, no need to zero them. The pages are marked as invalid, and I checked that all paths to get the page content (read/mmap) ensure that the page is validated before. | |
Comment Actions
- Styles
- KASSERT message changes
- KASSERT node type be VREG in tmpfs_reg_punch_hole
- Remove the MPASS(VOP_ISLOCKED(vp)) from tmpfs_deallocate. This is guarded by ASSERT_VOP_LOCKED already.