Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 14 Jun 2018 18:07:52 -0500
From:      Alan Cox <alc@rice.edu>
To:        Steven Hartland <steven.hartland@multiplay.co.uk>
Cc:        Konstantin Belousov <kib@FreeBSD.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   Re: svn commit: r335171 - head/sys/vm
Message-ID:  <F220E356-D55E-48B9-9AF0-ABBF41A74FC7@rice.edu>
In-Reply-To: <c241bd08-eb94-152e-c1f1-d77dc6987908@multiplay.co.uk>
References:  <201806141941.w5EJf2qa069373@repo.freebsd.org> <c241bd08-eb94-152e-c1f1-d77dc6987908@multiplay.co.uk>

next in thread | previous in thread | raw e-mail | index | archive | help

> On Jun 14, 2018, at 5:54 PM, Steven Hartland =
<steven.hartland@multiplay.co.uk> wrote:
>=20
> Out of interest, how would this exhibit itself?
>=20

A panic in vm_page_insert_after().

> On 14/06/2018 20:41, Konstantin Belousov wrote:
>> Author: kib
>> Date: Thu Jun 14 19:41:02 2018
>> New Revision: 335171
>> URL: https://svnweb.freebsd.org/changeset/base/335171 =
<https://svnweb.freebsd.org/changeset/base/335171>;
>>=20
>> Log:
>>   Handle the race between fork/vm_object_split() and faults.
>>  =20
>>   If fault started before vmspace_fork() locked the map, and then =
during
>>   fork, vm_map_copy_entry()->vm_object_split() is executed, it is
>>   possible that the fault instantiate the page into the original =
object
>>   when the page was already copied into the new object (see
>>   vm_map_split() for the orig/new objects terminology). This can =
happen
>>   if split found a busy page (e.g. from the fault) and slept dropping
>>   the objects lock, which allows the swap pager to instantiate
>>   read-behind pages for the fault.  Then the restart of the scan can =
see
>>   a page in the scanned range, where it was already copied to the =
upper
>>   object.
>>  =20
>>   Fix it by instantiating the read-ahead pages before
>>   swap_pager_getpages() method drops the lock to allocate pbuf.  The
>>   object scan would see the whole range prefilled with the busy pages
>>   and not proceed the range.
>>  =20
>>   Note that vm_fault rechecks the map generation count after the =
object
>>   unlock, so that it restarts the handling if raced with split, and
>>   re-lookups the right page from the upper object.
>>  =20
>>   In collaboration with:	alc
>>   Tested by:	pho
>>   Sponsored by:	The FreeBSD Foundation
>>   MFC after:	1 week
>>=20
>> Modified:
>>   head/sys/vm/swap_pager.c
>>=20
>> Modified: head/sys/vm/swap_pager.c
>> =
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D
>> --- head/sys/vm/swap_pager.c	Thu Jun 14 19:01:40 2018	=
(r335170)
>> +++ head/sys/vm/swap_pager.c	Thu Jun 14 19:41:02 2018	=
(r335171)
>> @@ -1096,21 +1096,24 @@ swap_pager_getpages(vm_object_t object, =
vm_page_t *ma,
>>      int *rahead)
>>  {
>>  	struct buf *bp;
>> -	vm_page_t mpred, msucc, p;
>> +	vm_page_t bm, mpred, msucc, p;
>>  	vm_pindex_t pindex;
>>  	daddr_t blk;
>> -	int i, j, maxahead, maxbehind, reqcount, shift;
>> +	int i, maxahead, maxbehind, reqcount;
>> =20
>>  	reqcount =3D count;
>> =20
>> -	VM_OBJECT_WUNLOCK(object);
>> -	bp =3D getpbuf(&nsw_rcount);
>> -	VM_OBJECT_WLOCK(object);
>> -
>> -	if (!swap_pager_haspage(object, ma[0]->pindex, &maxbehind, =
&maxahead)) {
>> -		relpbuf(bp, &nsw_rcount);
>> +	/*
>> +	 * Determine the final number of read-behind pages and
>> +	 * allocate them BEFORE releasing the object lock.  Otherwise,
>> +	 * there can be a problematic race with vm_object_split().
>> +	 * Specifically, vm_object_split() might first transfer pages
>> +	 * that precede ma[0] in the current object to a new object,
>> +	 * and then this function incorrectly recreates those pages as
>> +	 * read-behind pages in the current object.
>> +	 */
>> +	if (!swap_pager_haspage(object, ma[0]->pindex, &maxbehind, =
&maxahead))
>>  		return (VM_PAGER_FAIL);
>> -	}
>> =20
>>  	/*
>>  	 * Clip the readahead and readbehind ranges to exclude resident =
pages.
>> @@ -1132,35 +1135,31 @@ swap_pager_getpages(vm_object_t object, =
vm_page_t *ma,
>>  			*rbehind =3D pindex - mpred->pindex - 1;
>>  	}
>> =20
>> +	bm =3D ma[0];
>> +	for (i =3D 0; i < count; i++)
>> +		ma[i]->oflags |=3D VPO_SWAPINPROG;
>> +
>>  	/*
>>  	 * Allocate readahead and readbehind pages.
>>  	 */
>> -	shift =3D rbehind !=3D NULL ? *rbehind : 0;
>> -	if (shift !=3D 0) {
>> -		for (i =3D 1; i <=3D shift; i++) {
>> +	if (rbehind !=3D NULL) {
>> +		for (i =3D 1; i <=3D *rbehind; i++) {
>>  			p =3D vm_page_alloc(object, ma[0]->pindex - i,
>>  			    VM_ALLOC_NORMAL);
>> -			if (p =3D=3D NULL) {
>> -				/* Shift allocated pages to the left. */
>> -				for (j =3D 0; j < i - 1; j++)
>> -					bp->b_pages[j] =3D
>> -					    bp->b_pages[j + shift - i + =
1];
>> +			if (p =3D=3D NULL)
>>  				break;
>> -			}
>> -			bp->b_pages[shift - i] =3D p;
>> +			p->oflags |=3D VPO_SWAPINPROG;
>> +			bm =3D p;
>>  		}
>> -		shift =3D i - 1;
>> -		*rbehind =3D shift;
>> +		*rbehind =3D i - 1;
>>  	}
>> -	for (i =3D 0; i < reqcount; i++)
>> -		bp->b_pages[i + shift] =3D ma[i];
>>  	if (rahead !=3D NULL) {
>>  		for (i =3D 0; i < *rahead; i++) {
>>  			p =3D vm_page_alloc(object,
>>  			    ma[reqcount - 1]->pindex + i + 1, =
VM_ALLOC_NORMAL);
>>  			if (p =3D=3D NULL)
>>  				break;
>> -			bp->b_pages[shift + reqcount + i] =3D p;
>> +			p->oflags |=3D VPO_SWAPINPROG;
>>  		}
>>  		*rahead =3D i;
>>  	}
>> @@ -1171,15 +1170,18 @@ swap_pager_getpages(vm_object_t object, =
vm_page_t *ma,
>> =20
>>  	vm_object_pip_add(object, count);
>> =20
>> -	for (i =3D 0; i < count; i++)
>> -		bp->b_pages[i]->oflags |=3D VPO_SWAPINPROG;
>> -
>> -	pindex =3D bp->b_pages[0]->pindex;
>> +	pindex =3D bm->pindex;
>>  	blk =3D swp_pager_meta_ctl(object, pindex, 0);
>>  	KASSERT(blk !=3D SWAPBLK_NONE,
>>  	    ("no swap blocking containing %p(%jx)", object, =
(uintmax_t)pindex));
>> =20
>>  	VM_OBJECT_WUNLOCK(object);
>> +	bp =3D getpbuf(&nsw_rcount);
>> +	/* Pages cannot leave the object while busy. */
>> +	for (i =3D 0, p =3D bm; i < count; i++, p =3D TAILQ_NEXT(p, =
listq)) {
>> +		MPASS(p->pindex =3D=3D bm->pindex + i);
>> +		bp->b_pages[i] =3D p;
>> +	}
>> =20
>>  	bp->b_flags |=3D B_PAGING;
>>  	bp->b_iocmd =3D BIO_READ;
>>=20
>=20




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?F220E356-D55E-48B9-9AF0-ABBF41A74FC7>