i386/74327: [PATCH] mlock() causes physical memory leakage

[Denis Koreshkov]

>Number:         74327
>Category:       i386
>Synopsis:       mlock() causes physical memory leakage
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    freebsd-i386
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Nov 24 18:20:26 GMT 2004
>Closed-Date:
>Last-Modified:
>Originator:     Denis Koreshkov
>Release:        FreeBSD 4.10-RELEASE i386
>Organization:
>Environment:
System: FreeBSD doscrypt 4.8-RELEASE FreeBSD 4.8-RELEASE #0: Sat Oct 30 02:38:01 MSD 2004 root at doscrypt:/usr/obj/usr/src/sys/LITE i386

>Description:
        On FreeBSD/i386, mlock()'ing an memory area cause underlying
        pages to become unusable (to both kernel and applications)
        following exit(), when the following conditions are met:        
                1) The area had already been accessed before
                    a call to mlock(), and
                2) No corresponding call to munlock() had been
                    issued before exit().
>How-To-Repeat:
        1) malloc() a few megs of memory,
        2) bzero() it,
        3) mlock() it,
        4) exit().

        sysctl vm.stats.vm.v_wire_count will have grown
        by the amount of pages locked
>Fix:
        The bug is confined in pmap_enter() (file sys/i386/i386/pmap.c)
        One should toggle PG_W bit of pte to reflect wiring
        changes, as vm_fault_user_wire() does vm_fault() to
        bring pages into physmem.

*** sys/i386/i386/pmap.c.orig   Sun Nov 14 11:12:34 2004
--- sys/i386/i386/pmap.c        Sun Nov 14 11:21:25 2004
***************
*** 1962,1971 ****
                 * are valid mappings in them. Hence, if a user page is wired,
                 * the PT page will be also.
                 */
!               if (wired && ((origpte & PG_W) == 0))
                        pmap->pm_stats.wired_count++;
!               else if (!wired && (origpte & PG_W))
                        pmap->pm_stats.wired_count--;
  
  #if defined(PMAP_DIAGNOSTIC)
                if (pmap_nw_modified((pt_entry_t) origpte)) {
--- 1962,1974 ----
                 * are valid mappings in them. Hence, if a user page is wired,
                 * the PT page will be also.
                 */
!               if (wired && ((origpte & PG_W) == 0)) {
!                       *pte |= PG_W;
                        pmap->pm_stats.wired_count++;
!               } else if (!wired && (origpte & PG_W)) {
!                       *pte &= ~PG_W;
                        pmap->pm_stats.wired_count--;
+               }
  
  #if defined(PMAP_DIAGNOSTIC)
                if (pmap_nw_modified((pt_entry_t) origpte)) {

>Release-Note:
>Audit-Trail:
>Unformatted: