freebsd-dev/sys/vm/vnode_pager.c

1415 lines
38 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1990 University of Utah.
* Copyright (c) 1991 The Regents of the University of California.
* All rights reserved.
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
* Copyright (c) 1993, 1994 John S. Dyson
* Copyright (c) 1995, David Greenman
1994-05-24 10:09:53 +00:00
*
* This code is derived from software contributed to Berkeley by
* the Systems Programming Group of the University of Utah Computer
* Science Department.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
1994-05-24 10:09:53 +00:00
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91
1994-05-24 10:09:53 +00:00
*/
/*
* Page to/from files (vnodes).
*/
/*
* TODO:
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
* Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will
* greatly re-simplify the vnode_pager.
*/
2003-06-11 23:50:51 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_vm.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/vnode.h>
#include <sys/mount.h>
#include <sys/bio.h>
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
#include <sys/buf.h>
#include <sys/vmmeter.h>
#include <sys/limits.h>
#include <sys/conf.h>
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
#include <sys/rwlock.h>
#include <sys/sf_buf.h>
1994-05-24 10:09:53 +00:00
#include <machine/atomic.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_object.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm_page.h>
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
#include <vm/vm_pager.h>
#include <vm/vm_map.h>
1994-05-24 10:09:53 +00:00
#include <vm/vnode_pager.h>
#include <vm/vm_extern.h>
1994-05-24 10:09:53 +00:00
static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address,
daddr_t *rtaddress, int *run);
2002-03-19 22:20:14 +00:00
static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m);
static int vnode_pager_input_old(vm_object_t object, vm_page_t m);
static void vnode_pager_dealloc(vm_object_t);
static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
static int vnode_pager_getpages_async(vm_object_t, vm_page_t *, int, int *,
int *, vop_getpages_iodone_t, void *);
static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
2002-03-19 22:20:14 +00:00
static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
vm_ooffset_t, struct ucred *cred);
static int vnode_pager_generic_getpages_done(struct buf *);
static void vnode_pager_generic_getpages_done_async(struct buf *);
1994-05-24 10:09:53 +00:00
struct pagerops vnodepagerops = {
.pgo_alloc = vnode_pager_alloc,
.pgo_dealloc = vnode_pager_dealloc,
.pgo_getpages = vnode_pager_getpages,
.pgo_getpages_async = vnode_pager_getpages_async,
.pgo_putpages = vnode_pager_putpages,
.pgo_haspage = vnode_pager_haspage,
1994-05-24 10:09:53 +00:00
};
int vnode_pbuf_freecnt;
int vnode_async_pbuf_freecnt;
/* Create the VM system backing object for this vnode */
int
vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td)
{
vm_object_t object;
vm_ooffset_t size = isize;
struct vattr va;
if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
return (0);
while ((object = vp->v_object) != NULL) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
if (!(object->flags & OBJ_DEAD)) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return (0);
}
VOP_UNLOCK(vp, 0);
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vodead", 0);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
}
if (size == 0) {
if (vn_isdisk(vp, NULL)) {
size = IDX_TO_OFF(INT_MAX);
} else {
if (VOP_GETATTR(vp, &va, td->td_ucred))
return (0);
size = va.va_size;
}
}
object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred);
/*
* Dereference the reference we just created. This assumes
* that the object is associated with the vp.
*/
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
object->ref_count--;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
vrele(vp);
KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
return (0);
}
void
vnode_destroy_vobject(struct vnode *vp)
{
struct vm_object *obj;
obj = vp->v_object;
if (obj == NULL)
return;
ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
umtx_shm_object_terminated(obj);
if (obj->ref_count == 0) {
/*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0) {
vm_object_terminate(obj);
} else {
/*
* Waiters were already handled during object
* termination. The exclusive vnode lock hopefully
* prevented new waiters from referencing the dying
* object.
*/
KASSERT((obj->flags & OBJ_DISCONNECTWNT) == 0,
("OBJ_DISCONNECTWNT set obj %p flags %x",
obj, obj->flags));
vp->v_object = NULL;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
}
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
}
KASSERT(vp->v_object == NULL, ("vp %p obj %p", vp, vp->v_object));
}
1994-05-24 10:09:53 +00:00
/*
* Allocate (or lookup) pager for a vnode.
* Handle is a vnode pointer.
*
* MPSAFE
1994-05-24 10:09:53 +00:00
*/
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
vm_object_t
vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t offset, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
vm_object_t object;
1994-05-24 10:09:53 +00:00
struct vnode *vp;
/*
* Pageout to vnode, no can do yet.
*/
if (handle == NULL)
return (NULL);
1994-05-24 10:09:53 +00:00
vp = (struct vnode *) handle;
/*
* If the object is being terminated, wait for it to
* go away.
*/
retry:
while ((object = vp->v_object) != NULL) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
2008-07-30 21:18:08 +00:00
if ((object->flags & OBJ_DEAD) == 0)
break;
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0);
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference"));
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
if (object == NULL) {
1994-05-24 10:09:53 +00:00
/*
* Add an object of the appropriate size
1994-05-24 10:09:53 +00:00
*/
object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
object->un_pager.vnp.vnp_size = size;
object->un_pager.vnp.writemappings = 0;
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
object->handle = handle;
VI_LOCK(vp);
if (vp->v_object != NULL) {
/*
* Object has been created while we were sleeping
*/
VI_UNLOCK(vp);
VM_OBJECT_WLOCK(object);
KASSERT(object->ref_count == 1,
("leaked ref %p %d", object, object->ref_count));
object->type = OBJT_DEAD;
object->ref_count = 0;
VM_OBJECT_WUNLOCK(object);
vm_object_destroy(object);
goto retry;
}
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
vp->v_object = object;
VI_UNLOCK(vp);
} else {
object->ref_count++;
#if VM_NRESERVLEVEL > 0
vm_object_color(object, 0);
#endif
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
}
vref(vp);
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
return (object);
1994-05-24 10:09:53 +00:00
}
/*
* The object must be locked.
*/
static void
2014-01-20 18:47:56 +00:00
vnode_pager_dealloc(vm_object_t object)
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
{
struct vnode *vp;
int refs;
1994-05-24 10:09:53 +00:00
vp = object->handle;
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
if (vp == NULL)
panic("vnode_pager_dealloc: pager already dealloced");
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
vm_object_pip_wait(object, "vnpdea");
refs = object->ref_count;
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
object->handle = NULL;
object->type = OBJT_DEAD;
if (object->flags & OBJ_DISCONNECTWNT) {
vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
wakeup(object);
}
ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
if (object->un_pager.vnp.writemappings > 0) {
object->un_pager.vnp.writemappings = 0;
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
vp->v_object = NULL;
VOP_UNSET_TEXT(vp);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
while (refs-- > 0)
vunref(vp);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
1994-05-24 10:09:53 +00:00
}
static boolean_t
2014-01-20 18:47:56 +00:00
vnode_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
int *after)
1994-05-24 10:09:53 +00:00
{
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
struct vnode *vp = object->handle;
daddr_t bn;
int err;
daddr_t reqblock;
int poff;
int bsize;
int pagesperblock, blocksperpage;
1994-05-24 10:09:53 +00:00
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
/*
* If no vp or vp is doomed or marked transparent to VM, we do not
* have the page.
*/
if (vp == NULL || vp->v_iflag & VI_DOOMED)
return FALSE;
1994-05-24 10:09:53 +00:00
/*
* If the offset is beyond end of file we do
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
* not have the page.
1994-05-24 10:09:53 +00:00
*/
if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
return FALSE;
1994-05-24 10:09:53 +00:00
bsize = vp->v_mount->mnt_stat.f_iosize;
pagesperblock = bsize / PAGE_SIZE;
blocksperpage = 0;
if (pagesperblock > 0) {
reqblock = pindex / pagesperblock;
} else {
blocksperpage = (PAGE_SIZE / bsize);
reqblock = pindex * blocksperpage;
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
1995-05-30 08:16:23 +00:00
if (err)
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
return TRUE;
if (bn == -1)
return FALSE;
if (pagesperblock > 0) {
poff = pindex - (reqblock * pagesperblock);
if (before) {
*before *= pagesperblock;
*before += poff;
}
if (after) {
/*
* The BMAP vop can report a partial block in the
* 'after', but must not report blocks after EOF.
* Assert the latter, and truncate 'after' in case
* of the former.
*/
KASSERT((reqblock + *after) * pagesperblock <
roundup2(object->size, pagesperblock),
("%s: reqblock %jd after %d size %ju", __func__,
(intmax_t )reqblock, *after,
(uintmax_t )object->size));
*after *= pagesperblock;
*after += pagesperblock - (poff + 1);
if (pindex + *after >= object->size)
*after = object->size - 1 - pindex;
}
} else {
if (before) {
*before /= blocksperpage;
}
if (after) {
*after /= blocksperpage;
}
}
return TRUE;
1994-05-24 10:09:53 +00:00
}
/*
* Lets the VM system know about a change in size for a file.
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
* We adjust our own internal size and flush any cached pages in
1994-05-24 10:09:53 +00:00
* the associated object that are affected by the size change.
*
* Note: this routine may be invoked as a result of a pager put
* operation (possibly at object termination time), so we must be careful.
*/
void
2014-01-20 18:47:56 +00:00
vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize)
1994-05-24 10:09:53 +00:00
{
vm_object_t object;
vm_page_t m;
vm_pindex_t nobjsize;
if ((object = vp->v_object) == NULL)
1994-05-24 10:09:53 +00:00
return;
/* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
if (object->type == OBJT_DEAD) {
VM_OBJECT_WUNLOCK(object);
return;
}
KASSERT(object->type == OBJT_VNODE,
("not vnode-backed object %p", object));
if (nsize == object->un_pager.vnp.vnp_size) {
/*
* Hasn't changed size
*/
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
1994-05-24 10:09:53 +00:00
return;
}
nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
if (nsize < object->un_pager.vnp.vnp_size) {
/*
* File has shrunk. Toss any cached pages beyond the new EOF.
*/
if (nobjsize < object->size)
vm_object_page_remove(object, nobjsize, object->size,
0);
/*
* this gets rid of garbage at the end of a page that is now
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
2001-12-14 01:16:57 +00:00
* 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) &&
(m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
m->valid != 0) {
int base = (int)nsize & PAGE_MASK;
int size = PAGE_SIZE - base;
/*
* Clear out partial-page garbage in case
* the page has been mapped.
*/
pmap_zero_page_area(m, base, size);
/*
* Update the valid bits to reflect the blocks that
* have been zeroed. Some of these valid bits may
* have already been set.
*/
vm_page_set_valid_range(m, base, size);
/*
* Round "base" to the next block boundary so that the
* dirty bit for a partially zeroed block is not
* cleared.
*/
base = roundup2(base, DEV_BSIZE);
/*
* Clear out partial-page dirty bits.
*
* note that we do not clear out the valid
* bits. This would prevent bogus_page
* replacement from working properly.
*/
vm_page_clear_dirty(m, base, PAGE_SIZE - base);
}
1994-05-24 10:09:53 +00:00
}
object->un_pager.vnp.vnp_size = nsize;
object->size = nobjsize;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
1994-05-24 10:09:53 +00:00
}
/*
* calculate the linear (byte) disk address of specified virtual
* file address
*/
static int
vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
int *run)
{
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
int bsize;
int err;
daddr_t vblock;
daddr_t voffset;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (address < 0)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
return -1;
if (vp->v_iflag & VI_DOOMED)
return -1;
bsize = vp->v_mount->mnt_stat.f_iosize;
vblock = address / bsize;
voffset = address % bsize;
err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
if (err == 0) {
if (*rtaddress != -1)
*rtaddress += voffset / DEV_BSIZE;
if (run) {
*run += 1;
*run *= bsize/PAGE_SIZE;
*run -= voffset/PAGE_SIZE;
}
}
return (err);
}
/*
2002-05-16 21:28:32 +00:00
* small block filesystem vnode pager input
*/
static int
2014-01-20 18:47:56 +00:00
vnode_pager_input_smlfs(vm_object_t object, vm_page_t m)
{
struct vnode *vp;
struct bufobj *bo;
struct buf *bp;
struct sf_buf *sf;
daddr_t fileaddr;
vm_offset_t bsize;
vm_page_bits_t bits;
int error, i;
error = 0;
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct proc or any VM system structure will have to be rebuilt!!! Much needed overhaul of the VM system. Included in this first round of changes: 1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages, haspage, and sync operations are supported. The haspage interface now provides information about clusterability. All pager routines now take struct vm_object's instead of "pagers". 2) Improved data structures. In the previous paradigm, there is constant confusion caused by pagers being both a data structure ("allocate a pager") and a collection of routines. The idea of a pager structure has escentially been eliminated. Objects now have types, and this type is used to index the appropriate pager. In most cases, items in the pager structure were duplicated in the object data structure and thus were unnecessary. In the few cases that remained, a un_pager structure union was created in the object to contain these items. 3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now be removed. For instance, vm_object_enter(), vm_object_lookup(), vm_object_remove(), and the associated object hash list were some of the things that were removed. 4) simple_lock's removed. Discussion with several people reveals that the SMP locking primitives used in the VM system aren't likely the mechanism that we'll be adopting. Even if it were, the locking that was in the code was very inadequate and would have to be mostly re-done anyway. The locking in a uni-processor kernel was a no-op but went a long way toward making the code difficult to read and debug. 5) Places that attempted to kludge-up the fact that we don't have kernel thread support have been fixed to reflect the reality that we are really dealing with processes, not threads. The VM system didn't have complete thread support, so the comments and mis-named routines were just wrong. We now use tsleep and wakeup directly in the lock routines, for instance. 6) Where appropriate, the pagers have been improved, especially in the pager_alloc routines. Most of the pager_allocs have been rewritten and are now faster and easier to maintain. 7) The pagedaemon pageout clustering algorithm has been rewritten and now tries harder to output an even number of pages before and after the requested page. This is sort of the reverse of the ideal pagein algorithm and should provide better overall performance. 8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup have been removed. Some other unnecessary casts have also been removed. 9) Some almost useless debugging code removed. 10) Terminology of shadow objects vs. backing objects straightened out. The fact that the vm_object data structure escentially had this backwards really confused things. The use of "shadow" and "backing object" throughout the code is now internally consistent and correct in the Mach terminology. 11) Several minor bug fixes, including one in the vm daemon that caused 0 RSS objects to not get purged as intended. 12) A "default pager" has now been created which cleans up the transition of objects to the "swap" type. The previous checks throughout the code for swp->pg_data != NULL were really ugly. This change also provides the rudiments for future backing of "anonymous" memory by something other than the swap pager (via the vnode pager, for example), and it allows the decision about which of these pagers to use to be made dynamically (although will need some additional decision code to do this, of course). 13) (dyson) MAP_COPY has been deprecated and the corresponding "copy object" code has been removed. MAP_COPY was undocumented and non- standard. It was furthermore broken in several ways which caused its behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will continue to work correctly, but via the slightly different semantics of MAP_PRIVATE. 14) (dyson) Sharing maps have been removed. It's marginal usefulness in a threads design can be worked around in other ways. Both #12 and #13 were done to simplify the code and improve readability and maintain- ability. (As were most all of these changes) TODO: 1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing this will reduce the vnode pager to a mere fraction of its current size. 2) Rewrite vm_fault and the swap/vnode pagers to use the clustering information provided by the new haspage pager interface. This will substantially reduce the overhead by eliminating a large number of VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be improved to provide both a "behind" and "ahead" indication of contiguousness. 3) Implement the extended features of pager_haspage in swap_pager_haspage(). It currently just says 0 pages ahead/behind. 4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps via a much more general mechanism that could also be used for disk striping of regular filesystems. 5) Do something to improve the architecture of vm_object_collapse(). The fact that it makes calls into the swap pager and knows too much about how the swap pager operates really bothers me. It also doesn't allow for collapsing of non-swap pager objects ("unnamed" objects backed by other pagers).
1995-07-13 08:48:48 +00:00
vp = object->handle;
if (vp->v_iflag & VI_DOOMED)
return VM_PAGER_BAD;
bsize = vp->v_mount->mnt_stat.f_iosize;
VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
sf = sf_buf_alloc(m, 0);
for (i = 0; i < PAGE_SIZE / bsize; i++) {
vm_ooffset_t address;
bits = vm_page_bits(i * bsize, bsize);
if (m->valid & bits)
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
continue;
address = IDX_TO_OFF(m->pindex) + i * bsize;
if (address >= object->un_pager.vnp.vnp_size) {
fileaddr = -1;
} else {
error = vnode_pager_addr(vp, address, &fileaddr, NULL);
if (error)
break;
}
if (fileaddr != -1) {
bp = getpbuf(&vnode_pbuf_freecnt);
/* build a minimal buffer header */
bp->b_iocmd = BIO_READ;
bp->b_iodone = bdone;
KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
bp->b_rcred = crhold(curthread->td_ucred);
bp->b_wcred = crhold(curthread->td_ucred);
bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
bp->b_blkno = fileaddr;
pbgetbo(bo, bp);
bp->b_vp = vp;
bp->b_bcount = bsize;
bp->b_bufsize = bsize;
bp->b_runningbufspace = bp->b_bufsize;
atomic_add_long(&runningbufspace, bp->b_runningbufspace);
/* do the input */
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
bwait(bp, PVM, "vnsrd");
if ((bp->b_ioflags & BIO_ERROR) != 0)
error = EIO;
/*
* free the buffer header back to the swap buffer pool
*/
bp->b_vp = NULL;
pbrelbo(bp);
relpbuf(bp, &vnode_pbuf_freecnt);
if (error)
break;
} else
bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
KASSERT((m->dirty & bits) == 0,
("vnode_pager_input_smlfs: page %p is dirty", m));
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
m->valid |= bits;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
}
sf_buf_free(sf);
if (error) {
return VM_PAGER_ERROR;
}
return VM_PAGER_OK;
}
/*
2004-12-25 21:30:41 +00:00
* old style vnode pager input routine
*/
static int
2014-01-20 18:47:56 +00:00
vnode_pager_input_old(vm_object_t object, vm_page_t m)
{
1994-05-24 10:09:53 +00:00
struct uio auio;
struct iovec aiov;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
int error;
int size;
struct sf_buf *sf;
struct vnode *vp;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
error = 0;
/*
* Return failure if beyond current EOF
*/
if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
return VM_PAGER_BAD;
} else {
size = PAGE_SIZE;
if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
vp = object->handle;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
/*
* Allocate a kernel virtual address and initialize so that
* we can use VOP_READ/WRITE routines.
*/
sf = sf_buf_alloc(m, 0);
aiov.iov_base = (caddr_t)sf_buf_kva(sf);
aiov.iov_len = size;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = IDX_TO_OFF(m->pindex);
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
auio.uio_resid = size;
auio.uio_td = curthread;
error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
if (!error) {
2001-07-04 19:00:13 +00:00
int count = size - auio.uio_resid;
if (count == 0)
error = EINVAL;
else if (count != PAGE_SIZE)
bzero((caddr_t)sf_buf_kva(sf) + count,
PAGE_SIZE - count);
}
sf_buf_free(sf);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
}
KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
if (!error)
m->valid = VM_PAGE_BITS_ALL;
return error ? VM_PAGER_ERROR : VM_PAGER_OK;
}
/*
* generic vnode pager input routine
*/
/*
* Local media VFS's that do not implement their own VOP_GETPAGES
* should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
* to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_GETPAGES.
*/
static int
vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
int *rahead)
{
struct vnode *vp;
int rtval;
vp = object->handle;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
rtval = VOP_GETPAGES(vp, m, count, rbehind, rahead);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: FS getpages not implemented\n"));
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
return rtval;
}
static int
vnode_pager_getpages_async(vm_object_t object, vm_page_t *m, int count,
int *rbehind, int *rahead, vop_getpages_iodone_t iodone, void *arg)
{
struct vnode *vp;
int rtval;
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_GETPAGES_ASYNC(vp, m, count, rbehind, rahead, iodone, arg);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: FS getpages_async not implemented\n"));
VM_OBJECT_WLOCK(object);
return (rtval);
}
/*
* The implementation of VOP_GETPAGES() and VOP_GETPAGES_ASYNC() for
* local filesystems, where partially valid pages can only occur at
* the end of file.
*/
int
vnode_pager_local_getpages(struct vop_getpages_args *ap)
{
return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
ap->a_rbehind, ap->a_rahead, NULL, NULL));
}
int
vnode_pager_local_getpages_async(struct vop_getpages_async_args *ap)
{
return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg));
}
/*
* This is now called from local media FS's to operate against their
* own vnodes if they fail to implement VOP_GETPAGES.
*/
int
vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int count,
int *a_rbehind, int *a_rahead, vop_getpages_iodone_t iodone, void *arg)
{
vm_object_t object;
struct bufobj *bo;
struct buf *bp;
off_t foff;
#ifdef INVARIANTS
off_t blkno0;
#endif
int bsize, pagesperblock, *freecnt;
int error, before, after, rbehind, rahead, poff, i;
int bytecount, secmask;
KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
("%s does not support devices", __func__));
if (vp->v_iflag & VI_DOOMED)
return (VM_PAGER_BAD);
object = vp->v_object;
foff = IDX_TO_OFF(m[0]->pindex);
bsize = vp->v_mount->mnt_stat.f_iosize;
pagesperblock = bsize / PAGE_SIZE;
KASSERT(foff < object->un_pager.vnp.vnp_size,
("%s: page %p offset beyond vp %p size", __func__, m[0], vp));
KASSERT(count <= sizeof(bp->b_pages),
("%s: requested %d pages", __func__, count));
/*
* The last page has valid blocks. Invalid part can only
* exist at the end of file, and the page is made fully valid
* by zeroing in vm_pager_get_pages().
*/
if (m[count - 1]->valid != 0 && --count == 0) {
if (iodone != NULL)
iodone(arg, m, 1, 0);
return (VM_PAGER_OK);
}
1994-05-24 10:09:53 +00:00
/*
* Synchronous and asynchronous paging operations use different
* free pbuf counters. This is done to avoid asynchronous requests
* to consume all pbufs.
* Allocate the pbuf at the very beginning of the function, so that
* if we are low on certain kind of pbufs don't even proceed to BMAP,
* but sleep.
*/
freecnt = iodone != NULL ?
&vnode_async_pbuf_freecnt : &vnode_pbuf_freecnt;
bp = getpbuf(freecnt);
1994-05-24 10:09:53 +00:00
/*
* Get the underlying device blocks for the file with VOP_BMAP().
* If the file system doesn't support VOP_BMAP, use old way of
* getting pages via VOP_READ.
1994-05-24 10:09:53 +00:00
*/
error = VOP_BMAP(vp, foff / bsize, &bo, &bp->b_blkno, &after, &before);
if (error == EOPNOTSUPP) {
relpbuf(bp, freecnt);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
for (i = 0; i < count; i++) {
PCPU_INC(cnt.v_vnodein);
PCPU_INC(cnt.v_vnodepgsin);
error = vnode_pager_input_old(object, m[i]);
if (error)
break;
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return (error);
} else if (error != 0) {
relpbuf(bp, freecnt);
return (VM_PAGER_ERROR);
}
/*
* If the file system supports BMAP, but blocksize is smaller
* than a page size, then use special small filesystem code.
*/
if (pagesperblock == 0) {
relpbuf(bp, freecnt);
for (i = 0; i < count; i++) {
PCPU_INC(cnt.v_vnodein);
PCPU_INC(cnt.v_vnodepgsin);
error = vnode_pager_input_smlfs(object, m[i]);
if (error)
break;
}
return (error);
}
/*
* A sparse file can be encountered only for a single page request,
* which may not be preceded by call to vm_pager_haspage().
*/
if (bp->b_blkno == -1) {
KASSERT(count == 1,
("%s: array[%d] request to a sparse file %p", __func__,
count, vp));
relpbuf(bp, freecnt);
pmap_zero_page(m[0]);
KASSERT(m[0]->dirty == 0, ("%s: page %p is dirty",
__func__, m[0]));
VM_OBJECT_WLOCK(object);
m[0]->valid = VM_PAGE_BITS_ALL;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return (VM_PAGER_OK);
}
#ifdef INVARIANTS
blkno0 = bp->b_blkno;
#endif
bp->b_blkno += (foff % bsize) / DEV_BSIZE;
/* Recalculate blocks available after/before to pages. */
poff = (foff % bsize) / PAGE_SIZE;
before *= pagesperblock;
before += poff;
after *= pagesperblock;
after += pagesperblock - (poff + 1);
if (m[0]->pindex + after >= object->size)
after = object->size - 1 - m[0]->pindex;
KASSERT(count <= after + 1, ("%s: %d pages asked, can do only %d",
__func__, count, after + 1));
after -= count - 1;
/* Trim requested rbehind/rahead to possible values. */
rbehind = a_rbehind ? *a_rbehind : 0;
rahead = a_rahead ? *a_rahead : 0;
rbehind = min(rbehind, before);
rbehind = min(rbehind, m[0]->pindex);
rahead = min(rahead, after);
rahead = min(rahead, object->size - m[count - 1]->pindex);
/*
* Check that total amount of pages fit into buf. Trim rbehind and
* rahead evenly if not.
*/
if (rbehind + rahead + count > nitems(bp->b_pages)) {
int trim, sum;
trim = rbehind + rahead + count - nitems(bp->b_pages) + 1;
sum = rbehind + rahead;
if (rbehind == before) {
/* Roundup rbehind trim to block size. */
rbehind -= roundup(trim * rbehind / sum, pagesperblock);
if (rbehind < 0)
rbehind = 0;
} else
rbehind -= trim * rbehind / sum;
rahead -= trim * rahead / sum;
}
KASSERT(rbehind + rahead + count <= nitems(bp->b_pages),
("%s: behind %d ahead %d count %d", __func__,
rbehind, rahead, count));
/*
* Fill in the bp->b_pages[] array with requested and optional
* read behind or read ahead pages. Read behind pages are looked
* up in a backward direction, down to a first cached page. Same
* for read ahead pages, but there is no need to shift the array
* in case of encountering a cached page.
*/
i = bp->b_npages = 0;
if (rbehind) {
vm_pindex_t startpindex, tpindex;
vm_page_t p;
VM_OBJECT_WLOCK(object);
startpindex = m[0]->pindex - rbehind;
if ((p = TAILQ_PREV(m[0], pglist, listq)) != NULL &&
p->pindex >= startpindex)
startpindex = p->pindex + 1;
/* tpindex is unsigned; beware of numeric underflow. */
for (tpindex = m[0]->pindex - 1;
tpindex >= startpindex && tpindex < m[0]->pindex;
tpindex--, i++) {
p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
if (p == NULL) {
/* Shift the array. */
for (int j = 0; j < i; j++)
bp->b_pages[j] = bp->b_pages[j +
tpindex + 1 - startpindex];
break;
}
bp->b_pages[tpindex - startpindex] = p;
}
bp->b_pgbefore = i;
bp->b_npages += i;
bp->b_blkno -= IDX_TO_OFF(i) / DEV_BSIZE;
} else
bp->b_pgbefore = 0;
/* Requested pages. */
for (int j = 0; j < count; j++, i++)
bp->b_pages[i] = m[j];
bp->b_npages += count;
if (rahead) {
vm_pindex_t endpindex, tpindex;
vm_page_t p;
if (!VM_OBJECT_WOWNED(object))
VM_OBJECT_WLOCK(object);
endpindex = m[count - 1]->pindex + rahead + 1;
if ((p = TAILQ_NEXT(m[count - 1], listq)) != NULL &&
p->pindex < endpindex)
endpindex = p->pindex;
if (endpindex > object->size)
endpindex = object->size;
for (tpindex = m[count - 1]->pindex + 1;
tpindex < endpindex; i++, tpindex++) {
p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL);
if (p == NULL)
break;
bp->b_pages[i] = p;
}
bp->b_pgafter = i - bp->b_npages;
bp->b_npages = i;
} else
bp->b_pgafter = 0;
if (VM_OBJECT_WOWNED(object))
VM_OBJECT_WUNLOCK(object);
/* Report back actual behind/ahead read. */
if (a_rbehind)
*a_rbehind = bp->b_pgbefore;
if (a_rahead)
*a_rahead = bp->b_pgafter;
#ifdef INVARIANTS
KASSERT(bp->b_npages <= nitems(bp->b_pages),
("%s: buf %p overflowed", __func__, bp));
for (int j = 1; j < bp->b_npages; j++)
KASSERT(bp->b_pages[j]->pindex - 1 ==
bp->b_pages[j - 1]->pindex,
("%s: pages array not consecutive, bp %p", __func__, bp));
#endif
/*
* Recalculate first offset and bytecount with regards to read behind.
* Truncate bytecount to vnode real size and round up physical size
* for real devices.
*/
foff = IDX_TO_OFF(bp->b_pages[0]->pindex);
bytecount = bp->b_npages << PAGE_SHIFT;
if ((foff + bytecount) > object->un_pager.vnp.vnp_size)
bytecount = object->un_pager.vnp.vnp_size - foff;
secmask = bo->bo_bsize - 1;
KASSERT(secmask < PAGE_SIZE && secmask > 0,
("%s: sector size %d too large", __func__, secmask + 1));
bytecount = (bytecount + secmask) & ~secmask;
/*
* And map the pages to be read into the kva, if the filesystem
* requires mapped buffers.
*/
if ((vp->v_mount->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 &&
unmapped_buf_allowed) {
bp->b_data = unmapped_buf;
bp->b_offset = 0;
} else {
bp->b_data = bp->b_kvabase;
pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages);
}
/* Build a minimal buffer header. */
bp->b_iocmd = BIO_READ;
KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
bp->b_rcred = crhold(curthread->td_ucred);
bp->b_wcred = crhold(curthread->td_ucred);
pbgetbo(bo, bp);
bp->b_vp = vp;
bp->b_bcount = bp->b_bufsize = bp->b_runningbufspace = bytecount;
bp->b_iooffset = dbtob(bp->b_blkno);
KASSERT(IDX_TO_OFF(m[0]->pindex - bp->b_pages[0]->pindex) ==
(blkno0 - bp->b_blkno) * DEV_BSIZE +
IDX_TO_OFF(m[0]->pindex) % bsize,
("wrong offsets bsize %d m[0] %ju b_pages[0] %ju "
"blkno0 %ju b_blkno %ju", bsize,
(uintmax_t)m[0]->pindex, (uintmax_t)bp->b_pages[0]->pindex,
(uintmax_t)blkno0, (uintmax_t)bp->b_blkno));
atomic_add_long(&runningbufspace, bp->b_runningbufspace);
PCPU_INC(cnt.v_vnodein);
PCPU_ADD(cnt.v_vnodepgsin, bp->b_npages);
if (iodone != NULL) { /* async */
bp->b_pgiodone = iodone;
bp->b_caller1 = arg;
bp->b_iodone = vnode_pager_generic_getpages_done_async;
bp->b_flags |= B_ASYNC;
BUF_KERNPROC(bp);
bstrategy(bp);
return (VM_PAGER_OK);
} else {
bp->b_iodone = bdone;
bstrategy(bp);
bwait(bp, PVM, "vnread");
error = vnode_pager_generic_getpages_done(bp);
for (i = 0; i < bp->b_npages; i++)
bp->b_pages[i] = NULL;
bp->b_vp = NULL;
pbrelbo(bp);
relpbuf(bp, &vnode_pbuf_freecnt);
return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK);
}
}
static void
vnode_pager_generic_getpages_done_async(struct buf *bp)
{
int error;
error = vnode_pager_generic_getpages_done(bp);
/* Run the iodone upon the requested range. */
bp->b_pgiodone(bp->b_caller1, bp->b_pages + bp->b_pgbefore,
bp->b_npages - bp->b_pgbefore - bp->b_pgafter, error);
for (int i = 0; i < bp->b_npages; i++)
bp->b_pages[i] = NULL;
bp->b_vp = NULL;
pbrelbo(bp);
relpbuf(bp, &vnode_async_pbuf_freecnt);
}
static int
vnode_pager_generic_getpages_done(struct buf *bp)
{
vm_object_t object;
off_t tfoff, nextoff;
int i, error;
error = (bp->b_ioflags & BIO_ERROR) != 0 ? EIO : 0;
object = bp->b_vp->v_object;
if (error == 0 && bp->b_bcount != bp->b_npages * PAGE_SIZE) {
if (!buf_mapped(bp)) {
bp->b_data = bp->b_kvabase;
pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages,
bp->b_npages);
}
bzero(bp->b_data + bp->b_bcount,
PAGE_SIZE * bp->b_npages - bp->b_bcount);
}
if (buf_mapped(bp)) {
pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages);
bp->b_data = unmapped_buf;
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
for (i = 0, tfoff = IDX_TO_OFF(bp->b_pages[0]->pindex);
i < bp->b_npages; i++, tfoff = nextoff) {
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
vm_page_t mt;
nextoff = tfoff + PAGE_SIZE;
mt = bp->b_pages[i];
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
if (nextoff <= object->un_pager.vnp.vnp_size) {
/*
* Read filled up entire page.
*/
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
mt->valid = VM_PAGE_BITS_ALL;
KASSERT(mt->dirty == 0,
("%s: page %p is dirty", __func__, mt));
KASSERT(!pmap_page_is_mapped(mt),
("%s: page %p is mapped", __func__, mt));
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
} else {
/*
* Read did not fill up entire page.
*
* Currently we do not set the entire page valid,
* we just try to clear the piece that we couldn't
* read.
*/
vm_page_set_valid_range(mt, 0,
object->un_pager.vnp.vnp_size - tfoff);
KASSERT((mt->dirty & vm_page_bits(0,
object->un_pager.vnp.vnp_size - tfoff)) == 0,
("%s: page %p is dirty", __func__, mt));
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
}
if (i < bp->b_pgbefore || i >= bp->b_npages - bp->b_pgafter)
vm_page_readahead_finish(mt);
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
if (error != 0)
printf("%s: I/O read error %d\n", __func__, error);
return (error);
}
/*
* EOPNOTSUPP is no longer legal. For local media VFS's that do not
* implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
* vnode_pager_generic_putpages() to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_PUTPAGES.
*/
static void
2014-01-20 18:47:56 +00:00
vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
int flags, int *rtvals)
{
int rtval;
struct vnode *vp;
int bytes = count * PAGE_SIZE;
/*
* Force synchronous operation if we are extremely low on memory
* to prevent a low-memory deadlock. VOP operations often need to
* allocate more memory to initiate the I/O ( i.e. do a BMAP
* operation ). The swapper handles the case by limiting the amount
* of asynchronous I/O, but that sort of solution doesn't scale well
* for the vnode pager without a lot of work.
*
* Also, the backing vnode's iodone routine may not wake the pageout
* daemon up. This should be probably be addressed XXX.
*/
if (vm_cnt.v_free_count < vm_cnt.v_pageout_free_min)
flags |= VM_PAGER_PUT_SYNC;
/*
* Call device-specific putpages function
*/
vp = object->handle;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: stale FS putpages\n"));
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
}
/*
* This is now called from local media FS's to operate against their
1999-03-27 02:39:01 +00:00
* own vnodes if they fail to implement VOP_PUTPAGES.
*
* This is typically called indirectly via the pageout daemon and
* clustering has already typically occurred, so in general we ask the
* underlying filesystem to write the data out asynchronously rather
* then delayed.
*/
int
vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
int flags, int *rtvals)
{
int i;
vm_object_t object;
vm_page_t m;
int count;
int maxsize, ncount;
vm_ooffset_t poffset;
struct uio auio;
struct iovec aiov;
int error;
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
int ioflags;
int ppscheck = 0;
static struct timeval lastfail;
static int curfail;
object = vp->v_object;
count = bytecount / PAGE_SIZE;
for (i = 0; i < count; i++)
rtvals[i] = VM_PAGER_ERROR;
if ((int64_t)ma[0]->pindex < 0) {
printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
(long)ma[0]->pindex, (u_long)ma[0]->dirty);
rtvals[0] = VM_PAGER_BAD;
return VM_PAGER_BAD;
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
maxsize = count * PAGE_SIZE;
ncount = count;
1994-05-24 10:09:53 +00:00
poffset = IDX_TO_OFF(ma[0]->pindex);
/*
* If the page-aligned write is larger then the actual file we
* have to invalidate pages occurring beyond the file EOF. However,
* there is an edge case where a file may not be page-aligned where
* the last page is partially invalid. In this case the filesystem
* 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.
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
2001-12-14 01:16:57 +00:00
*
* We do not under any circumstances truncate the valid bits, as
* this will screw up bogus page replacement.
*/
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
if (object->un_pager.vnp.vnp_size > poffset) {
int pgoff;
maxsize = object->un_pager.vnp.vnp_size - poffset;
ncount = btoc(maxsize);
if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
/*
* If the object is locked and the following
* conditions hold, then the page's dirty
* field cannot be concurrently changed by a
* pmap operation.
*/
m = ma[ncount - 1];
vm_page_assert_sbusied(m);
KASSERT(!pmap_page_is_write_mapped(m),
("vnode_pager_generic_putpages: page %p is not read-only", m));
vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
pgoff);
}
} else {
maxsize = 0;
ncount = 0;
}
if (ncount < count) {
for (i = ncount; i < count; i++) {
rtvals[i] = VM_PAGER_BAD;
}
}
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
/*
2011-10-14 11:48:32 +00:00
* pageouts are already clustered, use IO_ASYNC to force a bawrite()
* rather then a bdwrite() to prevent paging I/O from saturating
* the buffer cache. Dummy-up the sequential heuristic to cause
* large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
* the system decides how to cluster.
*/
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
ioflags = IO_VMIO;
if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
ioflags |= IO_SYNC;
else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
ioflags |= IO_ASYNC;
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
ioflags |= (flags & VM_PAGER_PUT_NOREUSE) ? IO_NOREUSE : 0;
ioflags |= IO_SEQMAX << IO_SEQSHIFT;
aiov.iov_base = (caddr_t) 0;
aiov.iov_len = maxsize;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = poffset;
auio.uio_segflg = UIO_NOCOPY;
auio.uio_rw = UIO_WRITE;
auio.uio_resid = maxsize;
auio.uio_td = (struct thread *) 0;
error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, ncount);
if (error) {
if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
printf("vnode_pager_putpages: I/O error %d\n", error);
}
if (auio.uio_resid) {
if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
auio.uio_resid, (u_long)ma[0]->pindex);
}
for (i = 0; i < ncount; i++) {
rtvals[i] = VM_PAGER_OK;
}
return rtvals[0];
}
void
vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
{
vm_object_t obj;
int i, pos;
if (written == 0)
return;
obj = ma[0]->object;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
if (pos < trunc_page(written)) {
rtvals[i] = VM_PAGER_OK;
vm_page_undirty(ma[i]);
} else {
/* Partially written page. */
rtvals[i] = VM_PAGER_AGAIN;
vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
}
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
}
void
vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
vm_offset_t end)
{
struct vnode *vp;
vm_ooffset_t old_wm;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
if (object->type != OBJT_VNODE) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return;
}
old_wm = object->un_pager.vnp.writemappings;
object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
vp = object->handle;
if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
VOP_ADD_WRITECOUNT(vp, 1);
CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
__func__, vp, vp->v_writecount);
} else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
}
void
vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
vm_offset_t end)
{
struct vnode *vp;
struct mount *mp;
vm_offset_t inc;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(object);
/*
* First, recheck the object type to account for the race when
* the vnode is reclaimed.
*/
if (object->type != OBJT_VNODE) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return;
}
/*
* Optimize for the case when writemappings is not going to
* zero.
*/
inc = end - start;
if (object->un_pager.vnp.writemappings != inc) {
object->un_pager.vnp.writemappings -= inc;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
return;
}
vp = object->handle;
vhold(vp);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(object);
mp = NULL;
vn_start_write(vp, &mp, V_WAIT);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Decrement the object's writemappings, by swapping the start
* and end arguments for vnode_pager_update_writecount(). If
* there was not a race with vnode reclaimation, then the
* vnode's v_writecount is decremented.
*/
vnode_pager_update_writecount(object, end, start);
VOP_UNLOCK(vp, 0);
vdrop(vp);
if (mp != NULL)
vn_finished_write(mp);
}