diff options
author | dillon <dillon@FreeBSD.org> | 2001-12-14 01:16:57 +0000 |
---|---|---|
committer | dillon <dillon@FreeBSD.org> | 2001-12-14 01:16:57 +0000 |
commit | cd4d323ad300ef689d4b7dae113791a0f82ee65e (patch) | |
tree | 41ee3cc72a843eef0d91fff1ed8c59b176341b5b /sys/vm | |
parent | 637ec33540e81a939866bcbced0bbbff709ae333 (diff) | |
download | FreeBSD-src-cd4d323ad300ef689d4b7dae113791a0f82ee65e.zip FreeBSD-src-cd4d323ad300ef689d4b7dae113791a0f82ee65e.tar.gz |
This fixes a large number of bugs in our NFS client side code. A recent
commit by Kirk also fixed a softupdates bug that could easily be triggered
by server side NFS.
* An edge case with shared R+W mmap()'s and truncate whereby
the system would inappropriately clear the dirty bits on
still-dirty data. (applicable to all filesystems)
THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING.
see vm/vm_page.c line 1641
* The straddle case for VM pages and buffer cache buffers when
truncating. (applicable to NFS client side)
* Possible SMP database corruption due to vm_pager_unmap_page()
not clearing the TLB for the other cpu's. (applicable to NFS
client side but could effect all filesystems). Note: not
considered serious since the corruption occurs beyond the file
EOF.
* When flusing a dirty buffer due to B_CACHE getting cleared,
we were accidently setting B_CACHE again (that is, bwrite() sets
B_CACHE), when we really want it to stay clear after the write
is complete. This resulted in a corrupt buffer. (applicable
to all filesystems but probably only triggered by NFS)
* We have to call vtruncbuf() when ftruncate()ing to remove
any buffer cache buffers. This is still tentitive, I may
be able to remove it due to the second bug fix. (applicable
to NFS client side)
* vnode_pager_setsize() race against nfs_vinvalbuf()... we have
to set n_size before calling nfs_vinvalbuf or the NFS code
may recursively vnode_pager_setsize() to the original value
before the truncate. This is what was causing the user mmap
bus faults in the nfs tester program. (applicable to NFS
client side)
* Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made
by Kirk).
Testing program written by: Avadis Tevanian, Jr.
Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step')
MFC after: 1 week
Diffstat (limited to 'sys/vm')
-rw-r--r-- | sys/vm/vm_page.c | 14 | ||||
-rw-r--r-- | sys/vm/vnode_pager.c | 29 |
2 files changed, 41 insertions, 2 deletions
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c index 0eb06fc..abc4194 100644 --- a/sys/vm/vm_page.c +++ b/sys/vm/vm_page.c @@ -1630,10 +1630,24 @@ vm_page_set_validclean(vm_page_t m, int base, int size) * use this opportunity to clear the PG_NOSYNC flag. If a process * takes a write fault on a MAP_NOSYNC memory area the flag will * be set again. + * + * We set valid bits inclusive of any overlap, but we can only + * clear dirty bits for DEV_BSIZE chunks that are fully within + * the range. */ pagebits = vm_page_bits(base, size); m->valid |= pagebits; +#if 0 /* NOT YET */ + if ((frag = base & (DEV_BSIZE - 1)) != 0) { + frag = DEV_BSIZE - frag; + base += frag; + size -= frag; + if (size < 0) + size = 0; + } + pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); +#endif m->dirty &= ~pagebits; if (base == 0 && size == PAGE_SIZE) { pmap_clear_modify(m); diff --git a/sys/vm/vnode_pager.c b/sys/vm/vnode_pager.c index 9e6363b..b0348c6 100644 --- a/sys/vm/vnode_pager.c +++ b/sys/vm/vnode_pager.c @@ -298,14 +298,18 @@ vnode_pager_setsize(vp, nsize) } /* * this gets rid of garbage at the end of a page that is now - * only partially backed by the vnode... + * only partially backed by the vnode. + * + * XXX for some reason (I don't know yet), if we take a + * completely invalid page and mark it partially valid + * it can screw up NFS reads, so we don't allow the case. */ if (nsize & PAGE_MASK) { vm_offset_t kva; vm_page_t m; m = vm_page_lookup(object, OFF_TO_IDX(nsize)); - if (m) { + if (m && m->valid) { int base = (int)nsize & PAGE_MASK; int size = PAGE_SIZE - base; @@ -318,6 +322,20 @@ vnode_pager_setsize(vp, nsize) vm_pager_unmap_page(kva); /* + * XXX work around SMP data integrity race + * by unmapping the page from user processes. + * The garbage we just cleared may be mapped + * to a user process running on another cpu + * and this code is not running through normal + * I/O channels which handle SMP issues for + * us, so unmap page to synchronize all cpus. + * + * XXX should vm_pager_unmap_page() have + * dealt with this? + */ + vm_page_protect(m, VM_PROT_NONE); + + /* * Clear out partial-page dirty bits. This * has the side effect of setting the valid * bits, but that is ok. There are a bunch @@ -325,6 +343,10 @@ vnode_pager_setsize(vp, nsize) * m->dirty == VM_PAGE_BITS_ALL. The file EOF * case is one of them. If the page is still * partially dirty, make it fully dirty. + * + * note that we do not clear out the valid + * bits. This would prevent bogus_page + * replacement from working properly. */ vm_page_set_validclean(m, base, size); if (m->dirty != 0) @@ -965,6 +987,9 @@ vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals) * may not properly clear the dirty bits for the entire page (which * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). * With the page locked we are free to fix-up the dirty bits here. + * + * We do not under any circumstances truncate the valid bits, as + * this will screw up bogus page replacement. */ if (maxsize + poffset > object->un_pager.vnp.vnp_size) { if (object->un_pager.vnp.vnp_size > poffset) { |