| commit | d89ce96a0788eaec175450e82a898f3764b99441 | |
| author | Matthew Dillon <dillon@apollo.backplane.com> | |
| Sat, 13 Feb 2010 09:58:02 +0000 (13 01:58 -0800) | ||
| committer | Matthew Dillon <dillon@apollo.backplane.com> | |
| Sat, 13 Feb 2010 09:58:02 +0000 (13 01:58 -0800) | ||
| tree | e971add2faf0e1cac0f3596077ccdbb169329b1c | treesnapshot (tar.gz zip) |
| parent | b815802035a205b934ffd7e3b2487c7440e7e138 | commitdiff |
kernel - TMPFS - Bug fixing pass - paging to/from swap, vnode recycling
* vm_page_set_validdirty() needs to flag the underlying object or
the vfsync() code for the vnode will not know that its dirty.
* Any paging activity has to sync to swap because the kernel marks the
VM pages as clean afterwords even if we [re]dirty them during the I/O.
Testing the uio for UIO_NOCOPY is sufficient to detect when a tmpfs_write()
is due to VM paging or not.
This fixes vnode recycling - where the data is forced out to swap, and
later can be paged back in, as well as paging due to memory pressure.
* vm_page_set_validdirty() needs to flag the underlying object or
the vfsync() code for the vnode will not know that its dirty.
* Any paging activity has to sync to swap because the kernel marks the
VM pages as clean afterwords even if we [re]dirty them during the I/O.
Testing the uio for UIO_NOCOPY is sufficient to detect when a tmpfs_write()
is due to VM paging or not.
This fixes vnode recycling - where the data is forced out to swap, and
later can be paged back in, as well as paging due to memory pressure.
| sys/vfs/tmpfs/tmpfs_vnops.c | diffblobblamehistory | |
| sys/vm/vm_page.c | diffblobblamehistory |