svn commit: r335171 - head/sys/vm

[Alan Cox]

> On Jun 14, 2018, at 6:07 PM, Alan Cox <alc at rice.edu> wrote:
> 
> 
>> On Jun 14, 2018, at 5:54 PM, Steven Hartland <steven.hartland at multiplay.co.uk <mailto:steven.hartland at multiplay.co.uk>> wrote:
>> 
>> Out of interest, how would this exhibit itself?
>> 
> 
> A panic in vm_page_insert_after().
> 

I should add that a non-debug kernel will panic a little later in vm_radix_insert().

>> 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>
>>> 
>>> Log:
>>>   Handle the race between fork/vm_object_split() and faults.
>>>   
>>>   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.
>>>   
>>>   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.
>>>   
>>>   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.
>>>   
>>>   In collaboration with:	alc
>>>   Tested by:	pho
>>>   Sponsored by:	The FreeBSD Foundation
>>>   MFC after:	1 week
>>> 
>>> Modified:
>>>   head/sys/vm/swap_pager.c
>>> 
>>> Modified: head/sys/vm/swap_pager.c
>>> ==============================================================================
>>> --- 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;
>>>  
>>>  	reqcount = count;
>>>  
>>> -	VM_OBJECT_WUNLOCK(object);
>>> -	bp = 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);
>>> -	}
>>>  
>>>  	/*
>>>  	 * 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 = pindex - mpred->pindex - 1;
>>>  	}
>>>  
>>> +	bm = ma[0];
>>> +	for (i = 0; i < count; i++)
>>> +		ma[i]->oflags |= VPO_SWAPINPROG;
>>> +
>>>  	/*
>>>  	 * Allocate readahead and readbehind pages.
>>>  	 */
>>> -	shift = rbehind != NULL ? *rbehind : 0;
>>> -	if (shift != 0) {
>>> -		for (i = 1; i <= shift; i++) {
>>> +	if (rbehind != NULL) {
>>> +		for (i = 1; i <= *rbehind; i++) {
>>>  			p = vm_page_alloc(object, ma[0]->pindex - i,
>>>  			    VM_ALLOC_NORMAL);
>>> -			if (p == NULL) {
>>> -				/* Shift allocated pages to the left. */
>>> -				for (j = 0; j < i - 1; j++)
>>> -					bp->b_pages[j] =
>>> -					    bp->b_pages[j + shift - i + 1];
>>> +			if (p == NULL)
>>>  				break;
>>> -			}
>>> -			bp->b_pages[shift - i] = p;
>>> +			p->oflags |= VPO_SWAPINPROG;
>>> +			bm = p;
>>>  		}
>>> -		shift = i - 1;
>>> -		*rbehind = shift;
>>> +		*rbehind = i - 1;
>>>  	}
>>> -	for (i = 0; i < reqcount; i++)
>>> -		bp->b_pages[i + shift] = ma[i];
>>>  	if (rahead != NULL) {
>>>  		for (i = 0; i < *rahead; i++) {
>>>  			p = vm_page_alloc(object,
>>>  			    ma[reqcount - 1]->pindex + i + 1, VM_ALLOC_NORMAL);
>>>  			if (p == NULL)
>>>  				break;
>>> -			bp->b_pages[shift + reqcount + i] = p;
>>> +			p->oflags |= VPO_SWAPINPROG;
>>>  		}
>>>  		*rahead = i;
>>>  	}
>>> @@ -1171,15 +1170,18 @@ swap_pager_getpages(vm_object_t object, vm_page_t *ma,
>>>  
>>>  	vm_object_pip_add(object, count);
>>>  
>>> -	for (i = 0; i < count; i++)
>>> -		bp->b_pages[i]->oflags |= VPO_SWAPINPROG;
>>> -
>>> -	pindex = bp->b_pages[0]->pindex;
>>> +	pindex = bm->pindex;
>>>  	blk = swp_pager_meta_ctl(object, pindex, 0);
>>>  	KASSERT(blk != SWAPBLK_NONE,
>>>  	    ("no swap blocking containing %p(%jx)", object, (uintmax_t)pindex));
>>>  
>>>  	VM_OBJECT_WUNLOCK(object);
>>> +	bp = getpbuf(&nsw_rcount);
>>> +	/* Pages cannot leave the object while busy. */
>>> +	for (i = 0, p = bm; i < count; i++, p = TAILQ_NEXT(p, listq)) {
>>> +		MPASS(p->pindex == bm->pindex + i);
>>> +		bp->b_pages[i] = p;
>>> +	}
>>>  
>>>  	bp->b_flags |= B_PAGING;
>>>  	bp->b_iocmd = BIO_READ;
>>> 
>> 
>