freebsd-skq/sys/vm
Konstantin Belousov 5975e53d40 Fix a race in vm_page_busy_sleep(9).
Suppose that we have an exclusively busy page, and a thread which can
accept shared-busy page.  In this case, typical code waiting for the
page xbusy state to pass is
again:
	VM_OBJECT_WLOCK(object);
	...
	if (vm_page_xbusied(m)) {
		vm_page_lock(m);
 		VM_OBJECT_WUNLOCK(object);    <---1
		vm_page_busy_sleep(p, "vmopax");
 		goto again;
	}

Suppose that the xbusy state owner locked the object, unbusied the
page and unlocked the object after we are at the line [1], but before we
executed the load of the busy_lock word in vm_page_busy_sleep().  If it
happens that there is still no waiters recorded for the busy state,
the xbusy owner did not acquired the page lock, so it proceeded.

More, suppose that some other thread happen to share-busy the page
after xbusy state was relinquished but before the m->busy_lock is read
in vm_page_busy_sleep().  Again, that thread only needs vm_object lock
to proceed.  Then, vm_page_busy_sleep() reads busy_lock value equal to
the VPB_SHARERS_WORD(1).

In this case, all tests in vm_page_busy_sleep(9) pass and we are going
to sleep, despite the page being share-busied.

Update check for m->busy_lock == VPB_UNBUSIED in vm_page_busy_sleep(9)
to also accept shared-busy state if we only wait for the xbusy state to
pass.

Merge sequential if()s with the same 'then' clause in
vm_page_busy_sleep().

Note that the current code does not share-busy pages from parallel
threads, the only way to have more that one sbusy owner is right now
is to recurse.

Reported and tested by:	pho (previous version)
Reviewed by:	alc, markj
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D8196
2016-10-13 14:41:05 +00:00
..
_vm_radix.h
default_pager.c A change to KPI of vm_pager_get_pages() and underlying VOP_GETPAGES(). 2015-12-16 21:30:45 +00:00
device_pager.c Avoid duplicated calls to pmap_page_get_memattr(). 2016-05-01 17:48:43 +00:00
memguard.c Include sys/_task.h into uma_int.h, so that taskqueue.h isn't a 2016-02-09 20:22:35 +00:00
memguard.h
phys_pager.c sys: use our roundup2/rounddown2() macros when param.h is available. 2016-04-21 19:57:40 +00:00
pmap.h Various changes to pmap_ts_referenced() 2016-09-10 16:49:25 +00:00
redzone.c
redzone.h
sg_pager.c Plug a vm_page leak introduced in r292373. 2016-02-05 19:35:53 +00:00
swap_pager.c Respect the caller's hints when performing swap readahead. 2016-09-04 00:25:49 +00:00
swap_pager.h
uma_core.c Release the second critical section in uma_zfree_arg() slightly earlier. 2016-07-20 01:01:50 +00:00
uma_dbg.c Fix memguard(9) in kernels with INVARIANTS enabled. 2016-06-01 22:31:35 +00:00
uma_dbg.h Move uma_dbg_alloc() and uma_dbg_free() into uma_core.c, which allows 2016-02-03 22:02:36 +00:00
uma_int.h Autotune the number of pages set aside for UMA startup based on the number 2016-07-07 18:37:12 +00:00
uma.h Replace a number of conflations of mp_ncpus and mp_maxid with either 2016-07-06 14:09:49 +00:00
vm_domain.c Move vm_domain_rr_selectdomain() under #ifdef VM_NUMA_ALLOC. 2016-05-10 22:25:55 +00:00
vm_domain.h
vm_extern.h Various cleanups to the main function for AIO kernel processes: 2016-01-19 21:37:51 +00:00
vm_fault.c Add a comment describing the 'fast path' that was introduced in r270011. 2016-07-20 17:20:22 +00:00
vm_glue.c Eliminate unneeded vm_page_xbusy() and vm_page_xunbusy() operations when 2016-08-14 22:00:45 +00:00
vm_init.c Conditionally move initial vfs bio alloc above 4G 2016-10-03 13:23:43 +00:00
vm_kern.c Introduce a new mechanism for relocating virtual pages to a new physical 2015-12-19 18:42:50 +00:00
vm_kern.h
vm_map.c Change the type of the map entry's next_read field from a vm_pindex_t to a 2016-07-07 20:58:16 +00:00
vm_map.h Change the type of the map entry's next_read field from a vm_pindex_t to a 2016-07-07 20:58:16 +00:00
vm_meter.c Remove support for idle page zeroing. 2016-09-03 20:38:13 +00:00
vm_mmap.c When mmap(2) is used with a vnode, capture vnode attributes in the 2016-07-10 11:49:10 +00:00
vm_object.c Fix a race in vm_page_busy_sleep(9). 2016-10-13 14:41:05 +00:00
vm_object.h The vmtotal sysctl handler marks active vm objects to calculate 2016-06-21 17:49:33 +00:00
vm_page.c Fix a race in vm_page_busy_sleep(9). 2016-10-13 14:41:05 +00:00
vm_page.h Fix a race in vm_page_busy_sleep(9). 2016-10-13 14:41:05 +00:00
vm_pageout.c Make the page daemon's notion of what kind of pass is being performed 2016-10-05 17:32:06 +00:00
vm_pageout.h The flag "vm_pages_needed" has long served two distinct purposes: (1) to 2016-05-27 19:15:45 +00:00
vm_pager.c Remove slightly used const values that can be replaced with nitems(). 2016-04-21 15:38:28 +00:00
vm_pager.h A change to KPI of vm_pager_get_pages() and underlying VOP_GETPAGES(). 2015-12-16 21:30:45 +00:00
vm_param.h Remove the v_cache_min and v_cache_max sysctls. They are unused and have 2015-09-11 03:00:20 +00:00
vm_phys.c Remove support for idle page zeroing. 2016-09-03 20:38:13 +00:00
vm_phys.h Remove support for idle page zeroing. 2016-09-03 20:38:13 +00:00
vm_radix.c Cleanup redundant parenthesis from existing howmany()/roundup() macro uses. 2016-04-22 16:57:42 +00:00
vm_radix.h
vm_reserv.c Introduce a new mechanism for relocating virtual pages to a new physical 2015-12-19 18:42:50 +00:00
vm_reserv.h Introduce a new mechanism for relocating virtual pages to a new physical 2015-12-19 18:42:50 +00:00
vm_unix.c
vm.h Add new bus methods for mapping resources. 2016-05-20 17:57:47 +00:00
vnode_pager.c Clarify the vnode_destroy_vobject() logic handling for already terminated 2016-07-05 11:21:02 +00:00
vnode_pager.h A change to KPI of vm_pager_get_pages() and underlying VOP_GETPAGES(). 2015-12-16 21:30:45 +00:00