2 * Copyright (c) 1998 Matthew Dillon,
3 * Copyright (c) 1994 John S. Dyson
4 * Copyright (c) 1990 University of Utah.
5 * Copyright (c) 1982, 1986, 1989, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 * Radix Bitmap 'blists'.
45 * - The new swapper uses the new radix bitmap code. This should scale
46 * to arbitrarily small or arbitrarily large swap spaces and an almost
47 * arbitrary degree of fragmentation.
51 * - on the fly reallocation of swap during putpages. The new system
52 * does not try to keep previously allocated swap blocks for dirty
55 * - on the fly deallocation of swap
57 * - No more garbage collection required. Unnecessarily allocated swap
58 * blocks only exist for dirty vm_page_t's now and these are already
59 * cycled (in a high-load system) by the pager. We also do on-the-fly
60 * removal of invalidated swap blocks when a page is destroyed
63 * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$
65 * @(#)swap_pager.c 8.9 (Berkeley) 3/21/94
66 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94
69 #include <sys/cdefs.h>
70 __FBSDID("$FreeBSD$");
75 #include <sys/param.h>
76 #include <sys/systm.h>
78 #include <sys/kernel.h>
84 #include <sys/fcntl.h>
85 #include <sys/mount.h>
86 #include <sys/namei.h>
87 #include <sys/vnode.h>
88 #include <sys/malloc.h>
89 #include <sys/racct.h>
90 #include <sys/resource.h>
91 #include <sys/resourcevar.h>
92 #include <sys/rwlock.h>
93 #include <sys/sysctl.h>
94 #include <sys/sysproto.h>
95 #include <sys/blist.h>
98 #include <sys/vmmeter.h>
100 #include <security/mac/mac_framework.h>
104 #include <vm/vm_map.h>
105 #include <vm/vm_kern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_page.h>
108 #include <vm/vm_pager.h>
109 #include <vm/vm_pageout.h>
110 #include <vm/vm_param.h>
111 #include <vm/swap_pager.h>
112 #include <vm/vm_extern.h>
115 #include <geom/geom.h>
118 * SWB_NPAGES must be a power of 2. It may be set to 1, 2, 4, 8, 16
119 * or 32 pages per allocation.
120 * The 32-page limit is due to the radix code (kern/subr_blist.c).
122 #ifndef MAX_PAGEOUT_CLUSTER
123 #define MAX_PAGEOUT_CLUSTER 16
126 #if !defined(SWB_NPAGES)
127 #define SWB_NPAGES MAX_PAGEOUT_CLUSTER
131 * The swblock structure maps an object and a small, fixed-size range
132 * of page indices to disk addresses within a swap area.
133 * The collection of these mappings is implemented as a hash table.
134 * Unused disk addresses within a swap area are allocated and managed
137 #define SWCORRECT(n) (sizeof(void *) * (n) / sizeof(daddr_t))
138 #define SWAP_META_PAGES (SWB_NPAGES * 2)
139 #define SWAP_META_MASK (SWAP_META_PAGES - 1)
142 struct swblock *swb_hnext;
143 vm_object_t swb_object;
144 vm_pindex_t swb_index;
146 daddr_t swb_pages[SWAP_META_PAGES];
149 static MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "swap pager private data");
150 static struct mtx sw_dev_mtx;
151 static TAILQ_HEAD(, swdevt) swtailq = TAILQ_HEAD_INITIALIZER(swtailq);
152 static struct swdevt *swdevhd; /* Allocate from here next */
153 static int nswapdev; /* Number of swap devices */
154 int swap_pager_avail;
155 static int swdev_syscall_active = 0; /* serialize swap(on|off) */
157 static vm_ooffset_t swap_total;
158 SYSCTL_QUAD(_vm, OID_AUTO, swap_total, CTLFLAG_RD, &swap_total, 0,
159 "Total amount of available swap storage.");
160 static vm_ooffset_t swap_reserved;
161 SYSCTL_QUAD(_vm, OID_AUTO, swap_reserved, CTLFLAG_RD, &swap_reserved, 0,
162 "Amount of swap storage needed to back all allocated anonymous memory.");
163 static int overcommit = 0;
164 SYSCTL_INT(_vm, OID_AUTO, overcommit, CTLFLAG_RW, &overcommit, 0,
165 "Configure virtual memory overcommit behavior. See tuning(7) "
168 /* bits from overcommit */
169 #define SWAP_RESERVE_FORCE_ON (1 << 0)
170 #define SWAP_RESERVE_RLIMIT_ON (1 << 1)
171 #define SWAP_RESERVE_ALLOW_NONWIRED (1 << 2)
174 swap_reserve(vm_ooffset_t incr)
177 return (swap_reserve_by_cred(incr, curthread->td_ucred));
181 swap_reserve_by_cred(vm_ooffset_t incr, struct ucred *cred)
186 static struct timeval lastfail;
189 uip = cred->cr_ruidinfo;
191 if (incr & PAGE_MASK)
192 panic("swap_reserve: & PAGE_MASK");
196 error = racct_add(curproc, RACCT_SWAP, incr);
197 PROC_UNLOCK(curproc);
203 mtx_lock(&sw_dev_mtx);
204 r = swap_reserved + incr;
205 if (overcommit & SWAP_RESERVE_ALLOW_NONWIRED) {
206 s = cnt.v_page_count - cnt.v_free_reserved - cnt.v_wire_count;
211 if ((overcommit & SWAP_RESERVE_FORCE_ON) == 0 || r <= s ||
212 (error = priv_check(curthread, PRIV_VM_SWAP_NOQUOTA)) == 0) {
216 mtx_unlock(&sw_dev_mtx);
220 UIDINFO_VMSIZE_LOCK(uip);
221 if ((overcommit & SWAP_RESERVE_RLIMIT_ON) != 0 &&
222 uip->ui_vmsize + incr > lim_cur(curproc, RLIMIT_SWAP) &&
223 priv_check(curthread, PRIV_VM_SWAP_NORLIMIT))
226 uip->ui_vmsize += incr;
227 UIDINFO_VMSIZE_UNLOCK(uip);
228 PROC_UNLOCK(curproc);
230 mtx_lock(&sw_dev_mtx);
231 swap_reserved -= incr;
232 mtx_unlock(&sw_dev_mtx);
235 if (!res && ppsratecheck(&lastfail, &curfail, 1)) {
236 printf("uid %d, pid %d: swap reservation for %jd bytes failed\n",
237 uip->ui_uid, curproc->p_pid, incr);
243 racct_sub(curproc, RACCT_SWAP, incr);
244 PROC_UNLOCK(curproc);
252 swap_reserve_force(vm_ooffset_t incr)
256 mtx_lock(&sw_dev_mtx);
257 swap_reserved += incr;
258 mtx_unlock(&sw_dev_mtx);
262 racct_add_force(curproc, RACCT_SWAP, incr);
263 PROC_UNLOCK(curproc);
266 uip = curthread->td_ucred->cr_ruidinfo;
268 UIDINFO_VMSIZE_LOCK(uip);
269 uip->ui_vmsize += incr;
270 UIDINFO_VMSIZE_UNLOCK(uip);
271 PROC_UNLOCK(curproc);
275 swap_release(vm_ooffset_t decr)
280 cred = curthread->td_ucred;
281 swap_release_by_cred(decr, cred);
282 PROC_UNLOCK(curproc);
286 swap_release_by_cred(vm_ooffset_t decr, struct ucred *cred)
290 uip = cred->cr_ruidinfo;
292 if (decr & PAGE_MASK)
293 panic("swap_release: & PAGE_MASK");
295 mtx_lock(&sw_dev_mtx);
296 if (swap_reserved < decr)
297 panic("swap_reserved < decr");
298 swap_reserved -= decr;
299 mtx_unlock(&sw_dev_mtx);
301 UIDINFO_VMSIZE_LOCK(uip);
302 if (uip->ui_vmsize < decr)
303 printf("negative vmsize for uid = %d\n", uip->ui_uid);
304 uip->ui_vmsize -= decr;
305 UIDINFO_VMSIZE_UNLOCK(uip);
307 racct_sub_cred(cred, RACCT_SWAP, decr);
310 static void swapdev_strategy(struct buf *, struct swdevt *sw);
312 #define SWM_FREE 0x02 /* free, period */
313 #define SWM_POP 0x04 /* pop out */
315 int swap_pager_full = 2; /* swap space exhaustion (task killing) */
316 static int swap_pager_almost_full = 1; /* swap space exhaustion (w/hysteresis)*/
317 static int nsw_rcount; /* free read buffers */
318 static int nsw_wcount_sync; /* limit write buffers / synchronous */
319 static int nsw_wcount_async; /* limit write buffers / asynchronous */
320 static int nsw_wcount_async_max;/* assigned maximum */
321 static int nsw_cluster_max; /* maximum VOP I/O allowed */
323 static struct swblock **swhash;
324 static int swhash_mask;
325 static struct mtx swhash_mtx;
327 static int swap_async_max = 4; /* maximum in-progress async I/O's */
328 static struct sx sw_alloc_sx;
331 SYSCTL_INT(_vm, OID_AUTO, swap_async_max,
332 CTLFLAG_RW, &swap_async_max, 0, "Maximum running async swap ops");
335 * "named" and "unnamed" anon region objects. Try to reduce the overhead
336 * of searching a named list by hashing it just a little.
341 #define NOBJLIST(handle) \
342 (&swap_pager_object_list[((int)(intptr_t)handle >> 4) & (NOBJLISTS-1)])
344 static struct mtx sw_alloc_mtx; /* protect list manipulation */
345 static struct pagerlst swap_pager_object_list[NOBJLISTS];
346 static uma_zone_t swap_zone;
347 static struct vm_object swap_zone_obj;
350 * pagerops for OBJT_SWAP - "swap pager". Some ops are also global procedure
351 * calls hooked from other parts of the VM system and do not appear here.
352 * (see vm/swap_pager.h).
355 swap_pager_alloc(void *handle, vm_ooffset_t size,
356 vm_prot_t prot, vm_ooffset_t offset, struct ucred *);
357 static void swap_pager_dealloc(vm_object_t object);
358 static int swap_pager_getpages(vm_object_t, vm_page_t *, int, int);
359 static void swap_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *);
361 swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after);
362 static void swap_pager_init(void);
363 static void swap_pager_unswapped(vm_page_t);
364 static void swap_pager_swapoff(struct swdevt *sp);
366 struct pagerops swappagerops = {
367 .pgo_init = swap_pager_init, /* early system initialization of pager */
368 .pgo_alloc = swap_pager_alloc, /* allocate an OBJT_SWAP object */
369 .pgo_dealloc = swap_pager_dealloc, /* deallocate an OBJT_SWAP object */
370 .pgo_getpages = swap_pager_getpages, /* pagein */
371 .pgo_putpages = swap_pager_putpages, /* pageout */
372 .pgo_haspage = swap_pager_haspage, /* get backing store status for page */
373 .pgo_pageunswapped = swap_pager_unswapped, /* remove swap related to page */
377 * dmmax is in page-sized chunks with the new swap system. It was
378 * dev-bsized chunks in the old. dmmax is always a power of 2.
380 * swap_*() routines are externally accessible. swp_*() routines are
384 static int nswap_lowat = 128; /* in pages, swap_pager_almost_full warn */
385 static int nswap_hiwat = 512; /* in pages, swap_pager_almost_full warn */
387 SYSCTL_INT(_vm, OID_AUTO, dmmax,
388 CTLFLAG_RD, &dmmax, 0, "Maximum size of a swap block");
390 static void swp_sizecheck(void);
391 static void swp_pager_async_iodone(struct buf *bp);
392 static int swapongeom(struct thread *, struct vnode *);
393 static int swaponvp(struct thread *, struct vnode *, u_long);
394 static int swapoff_one(struct swdevt *sp, struct ucred *cred);
397 * Swap bitmap functions
399 static void swp_pager_freeswapspace(daddr_t blk, int npages);
400 static daddr_t swp_pager_getswapspace(int npages);
405 static struct swblock **swp_pager_hash(vm_object_t object, vm_pindex_t index);
406 static void swp_pager_meta_build(vm_object_t, vm_pindex_t, daddr_t);
407 static void swp_pager_meta_free(vm_object_t, vm_pindex_t, daddr_t);
408 static void swp_pager_meta_free_all(vm_object_t);
409 static daddr_t swp_pager_meta_ctl(vm_object_t, vm_pindex_t, int);
412 swp_pager_free_nrpage(vm_page_t m)
416 if (m->wire_count == 0)
422 * SWP_SIZECHECK() - update swap_pager_full indication
424 * update the swap_pager_almost_full indication and warn when we are
425 * about to run out of swap space, using lowat/hiwat hysteresis.
427 * Clear swap_pager_full ( task killing ) indication when lowat is met.
429 * No restrictions on call
430 * This routine may not block.
436 if (swap_pager_avail < nswap_lowat) {
437 if (swap_pager_almost_full == 0) {
438 printf("swap_pager: out of swap space\n");
439 swap_pager_almost_full = 1;
443 if (swap_pager_avail > nswap_hiwat)
444 swap_pager_almost_full = 0;
449 * SWP_PAGER_HASH() - hash swap meta data
451 * This is an helper function which hashes the swapblk given
452 * the object and page index. It returns a pointer to a pointer
453 * to the object, or a pointer to a NULL pointer if it could not
456 static struct swblock **
457 swp_pager_hash(vm_object_t object, vm_pindex_t index)
459 struct swblock **pswap;
460 struct swblock *swap;
462 index &= ~(vm_pindex_t)SWAP_META_MASK;
463 pswap = &swhash[(index ^ (int)(intptr_t)object) & swhash_mask];
464 while ((swap = *pswap) != NULL) {
465 if (swap->swb_object == object &&
466 swap->swb_index == index
470 pswap = &swap->swb_hnext;
476 * SWAP_PAGER_INIT() - initialize the swap pager!
478 * Expected to be started from system init. NOTE: This code is run
479 * before much else so be careful what you depend on. Most of the VM
480 * system has yet to be initialized at this point.
483 swap_pager_init(void)
486 * Initialize object lists
490 for (i = 0; i < NOBJLISTS; ++i)
491 TAILQ_INIT(&swap_pager_object_list[i]);
492 mtx_init(&sw_alloc_mtx, "swap_pager list", NULL, MTX_DEF);
493 mtx_init(&sw_dev_mtx, "swapdev", NULL, MTX_DEF);
496 * Device Stripe, in PAGE_SIZE'd blocks
498 dmmax = SWB_NPAGES * 2;
502 * SWAP_PAGER_SWAP_INIT() - swap pager initialization from pageout process
504 * Expected to be started from pageout process once, prior to entering
508 swap_pager_swap_init(void)
513 * Number of in-transit swap bp operations. Don't
514 * exhaust the pbufs completely. Make sure we
515 * initialize workable values (0 will work for hysteresis
516 * but it isn't very efficient).
518 * The nsw_cluster_max is constrained by the bp->b_pages[]
519 * array (MAXPHYS/PAGE_SIZE) and our locally defined
520 * MAX_PAGEOUT_CLUSTER. Also be aware that swap ops are
521 * constrained by the swap device interleave stripe size.
523 * Currently we hardwire nsw_wcount_async to 4. This limit is
524 * designed to prevent other I/O from having high latencies due to
525 * our pageout I/O. The value 4 works well for one or two active swap
526 * devices but is probably a little low if you have more. Even so,
527 * a higher value would probably generate only a limited improvement
528 * with three or four active swap devices since the system does not
529 * typically have to pageout at extreme bandwidths. We will want
530 * at least 2 per swap devices, and 4 is a pretty good value if you
531 * have one NFS swap device due to the command/ack latency over NFS.
532 * So it all works out pretty well.
534 nsw_cluster_max = min((MAXPHYS/PAGE_SIZE), MAX_PAGEOUT_CLUSTER);
537 nsw_rcount = (nswbuf + 1) / 2;
538 nsw_wcount_sync = (nswbuf + 3) / 4;
539 nsw_wcount_async = 4;
540 nsw_wcount_async_max = nsw_wcount_async;
541 mtx_unlock(&pbuf_mtx);
544 * Initialize our zone. Right now I'm just guessing on the number
545 * we need based on the number of pages in the system. Each swblock
546 * can hold 16 pages, so this is probably overkill. This reservation
547 * is typically limited to around 32MB by default.
549 n = cnt.v_page_count / 2;
550 if (maxswzone && n > maxswzone / sizeof(struct swblock))
551 n = maxswzone / sizeof(struct swblock);
553 swap_zone = uma_zcreate("SWAPMETA", sizeof(struct swblock), NULL, NULL,
554 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_VM);
555 if (swap_zone == NULL)
556 panic("failed to create swap_zone.");
558 if (uma_zone_set_obj(swap_zone, &swap_zone_obj, n))
561 * if the allocation failed, try a zone two thirds the
562 * size of the previous attempt.
567 printf("Swap zone entries reduced from %d to %d.\n", n2, n);
571 * Initialize our meta-data hash table. The swapper does not need to
572 * be quite as efficient as the VM system, so we do not use an
573 * oversized hash table.
575 * n: size of hash table, must be power of 2
576 * swhash_mask: hash table index mask
578 for (n = 1; n < n2 / 8; n *= 2)
580 swhash = malloc(sizeof(struct swblock *) * n, M_VMPGDATA, M_WAITOK | M_ZERO);
582 mtx_init(&swhash_mtx, "swap_pager swhash", NULL, MTX_DEF);
586 * SWAP_PAGER_ALLOC() - allocate a new OBJT_SWAP VM object and instantiate
587 * its metadata structures.
589 * This routine is called from the mmap and fork code to create a new
590 * OBJT_SWAP object. We do this by creating an OBJT_DEFAULT object
591 * and then converting it with swp_pager_meta_build().
593 * This routine may block in vm_object_allocate() and create a named
594 * object lookup race, so we must interlock.
599 swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
600 vm_ooffset_t offset, struct ucred *cred)
605 pindex = OFF_TO_IDX(offset + PAGE_MASK + size);
609 * Reference existing named region or allocate new one. There
610 * should not be a race here against swp_pager_meta_build()
611 * as called from vm_page_remove() in regards to the lookup
614 sx_xlock(&sw_alloc_sx);
615 object = vm_pager_object_lookup(NOBJLIST(handle), handle);
616 if (object == NULL) {
618 if (!swap_reserve_by_cred(size, cred)) {
619 sx_xunlock(&sw_alloc_sx);
625 object = vm_object_allocate(OBJT_DEFAULT, pindex);
626 VM_OBJECT_WLOCK(object);
627 object->handle = handle;
630 object->charge = size;
632 swp_pager_meta_build(object, 0, SWAPBLK_NONE);
633 VM_OBJECT_WUNLOCK(object);
635 sx_xunlock(&sw_alloc_sx);
639 if (!swap_reserve_by_cred(size, cred))
643 object = vm_object_allocate(OBJT_DEFAULT, pindex);
644 VM_OBJECT_WLOCK(object);
647 object->charge = size;
649 swp_pager_meta_build(object, 0, SWAPBLK_NONE);
650 VM_OBJECT_WUNLOCK(object);
656 * SWAP_PAGER_DEALLOC() - remove swap metadata from object
658 * The swap backing for the object is destroyed. The code is
659 * designed such that we can reinstantiate it later, but this
660 * routine is typically called only when the entire object is
661 * about to be destroyed.
663 * The object must be locked.
666 swap_pager_dealloc(vm_object_t object)
670 * Remove from list right away so lookups will fail if we block for
671 * pageout completion.
673 if (object->handle != NULL) {
674 mtx_lock(&sw_alloc_mtx);
675 TAILQ_REMOVE(NOBJLIST(object->handle), object, pager_object_list);
676 mtx_unlock(&sw_alloc_mtx);
679 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
680 vm_object_pip_wait(object, "swpdea");
683 * Free all remaining metadata. We only bother to free it from
684 * the swap meta data. We do not attempt to free swapblk's still
685 * associated with vm_page_t's for this object. We do not care
686 * if paging is still in progress on some objects.
688 swp_pager_meta_free_all(object);
691 /************************************************************************
692 * SWAP PAGER BITMAP ROUTINES *
693 ************************************************************************/
696 * SWP_PAGER_GETSWAPSPACE() - allocate raw swap space
698 * Allocate swap for the requested number of pages. The starting
699 * swap block number (a page index) is returned or SWAPBLK_NONE
700 * if the allocation failed.
702 * Also has the side effect of advising that somebody made a mistake
703 * when they configured swap and didn't configure enough.
705 * This routine may not sleep.
707 * We allocate in round-robin fashion from the configured devices.
710 swp_pager_getswapspace(int npages)
717 mtx_lock(&sw_dev_mtx);
719 for (i = 0; i < nswapdev; i++) {
721 sp = TAILQ_FIRST(&swtailq);
722 if (!(sp->sw_flags & SW_CLOSING)) {
723 blk = blist_alloc(sp->sw_blist, npages);
724 if (blk != SWAPBLK_NONE) {
726 sp->sw_used += npages;
727 swap_pager_avail -= npages;
729 swdevhd = TAILQ_NEXT(sp, sw_list);
733 sp = TAILQ_NEXT(sp, sw_list);
735 if (swap_pager_full != 2) {
736 printf("swap_pager_getswapspace(%d): failed\n", npages);
738 swap_pager_almost_full = 1;
742 mtx_unlock(&sw_dev_mtx);
747 swp_pager_isondev(daddr_t blk, struct swdevt *sp)
750 return (blk >= sp->sw_first && blk < sp->sw_end);
754 swp_pager_strategy(struct buf *bp)
758 mtx_lock(&sw_dev_mtx);
759 TAILQ_FOREACH(sp, &swtailq, sw_list) {
760 if (bp->b_blkno >= sp->sw_first && bp->b_blkno < sp->sw_end) {
761 mtx_unlock(&sw_dev_mtx);
762 sp->sw_strategy(bp, sp);
766 panic("Swapdev not found");
771 * SWP_PAGER_FREESWAPSPACE() - free raw swap space
773 * This routine returns the specified swap blocks back to the bitmap.
775 * This routine may not sleep.
778 swp_pager_freeswapspace(daddr_t blk, int npages)
782 mtx_lock(&sw_dev_mtx);
783 TAILQ_FOREACH(sp, &swtailq, sw_list) {
784 if (blk >= sp->sw_first && blk < sp->sw_end) {
785 sp->sw_used -= npages;
787 * If we are attempting to stop swapping on
788 * this device, we don't want to mark any
789 * blocks free lest they be reused.
791 if ((sp->sw_flags & SW_CLOSING) == 0) {
792 blist_free(sp->sw_blist, blk - sp->sw_first,
794 swap_pager_avail += npages;
797 mtx_unlock(&sw_dev_mtx);
801 panic("Swapdev not found");
805 * SWAP_PAGER_FREESPACE() - frees swap blocks associated with a page
806 * range within an object.
808 * This is a globally accessible routine.
810 * This routine removes swapblk assignments from swap metadata.
812 * The external callers of this routine typically have already destroyed
813 * or renamed vm_page_t's associated with this range in the object so
817 swap_pager_freespace(vm_object_t object, vm_pindex_t start, vm_size_t size)
820 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
821 swp_pager_meta_free(object, start, size);
825 * SWAP_PAGER_RESERVE() - reserve swap blocks in object
827 * Assigns swap blocks to the specified range within the object. The
828 * swap blocks are not zerod. Any previous swap assignment is destroyed.
830 * Returns 0 on success, -1 on failure.
833 swap_pager_reserve(vm_object_t object, vm_pindex_t start, vm_size_t size)
836 daddr_t blk = SWAPBLK_NONE;
837 vm_pindex_t beg = start; /* save start index */
839 VM_OBJECT_WLOCK(object);
843 while ((blk = swp_pager_getswapspace(n)) == SWAPBLK_NONE) {
846 swp_pager_meta_free(object, beg, start - beg);
847 VM_OBJECT_WUNLOCK(object);
852 swp_pager_meta_build(object, start, blk);
858 swp_pager_meta_free(object, start, n);
859 VM_OBJECT_WUNLOCK(object);
864 * SWAP_PAGER_COPY() - copy blocks from source pager to destination pager
865 * and destroy the source.
867 * Copy any valid swapblks from the source to the destination. In
868 * cases where both the source and destination have a valid swapblk,
869 * we keep the destination's.
871 * This routine is allowed to sleep. It may sleep allocating metadata
872 * indirectly through swp_pager_meta_build() or if paging is still in
873 * progress on the source.
875 * The source object contains no vm_page_t's (which is just as well)
877 * The source object is of type OBJT_SWAP.
879 * The source and destination objects must be locked.
880 * Both object locks may temporarily be released.
883 swap_pager_copy(vm_object_t srcobject, vm_object_t dstobject,
884 vm_pindex_t offset, int destroysource)
888 VM_OBJECT_LOCK_ASSERT(srcobject, RA_WLOCKED);
889 VM_OBJECT_LOCK_ASSERT(dstobject, RA_WLOCKED);
892 * If destroysource is set, we remove the source object from the
893 * swap_pager internal queue now.
896 if (srcobject->handle != NULL) {
897 mtx_lock(&sw_alloc_mtx);
899 NOBJLIST(srcobject->handle),
903 mtx_unlock(&sw_alloc_mtx);
908 * transfer source to destination.
910 for (i = 0; i < dstobject->size; ++i) {
914 * Locate (without changing) the swapblk on the destination,
915 * unless it is invalid in which case free it silently, or
916 * if the destination is a resident page, in which case the
917 * source is thrown away.
919 dstaddr = swp_pager_meta_ctl(dstobject, i, 0);
921 if (dstaddr == SWAPBLK_NONE) {
923 * Destination has no swapblk and is not resident,
928 srcaddr = swp_pager_meta_ctl(
934 if (srcaddr != SWAPBLK_NONE) {
936 * swp_pager_meta_build() can sleep.
938 vm_object_pip_add(srcobject, 1);
939 VM_OBJECT_WUNLOCK(srcobject);
940 vm_object_pip_add(dstobject, 1);
941 swp_pager_meta_build(dstobject, i, srcaddr);
942 vm_object_pip_wakeup(dstobject);
943 VM_OBJECT_WLOCK(srcobject);
944 vm_object_pip_wakeup(srcobject);
948 * Destination has valid swapblk or it is represented
949 * by a resident page. We destroy the sourceblock.
952 swp_pager_meta_ctl(srcobject, i + offset, SWM_FREE);
957 * Free left over swap blocks in source.
959 * We have to revert the type to OBJT_DEFAULT so we do not accidently
960 * double-remove the object from the swap queues.
963 swp_pager_meta_free_all(srcobject);
965 * Reverting the type is not necessary, the caller is going
966 * to destroy srcobject directly, but I'm doing it here
967 * for consistency since we've removed the object from its
970 srcobject->type = OBJT_DEFAULT;
975 * SWAP_PAGER_HASPAGE() - determine if we have good backing store for
976 * the requested page.
978 * We determine whether good backing store exists for the requested
979 * page and return TRUE if it does, FALSE if it doesn't.
981 * If TRUE, we also try to determine how much valid, contiguous backing
982 * store exists before and after the requested page within a reasonable
983 * distance. We do not try to restrict it to the swap device stripe
984 * (that is handled in getpages/putpages). It probably isn't worth
988 swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after)
992 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
994 * do we have good backing store at the requested index ?
996 blk0 = swp_pager_meta_ctl(object, pindex, 0);
998 if (blk0 == SWAPBLK_NONE) {
1007 * find backwards-looking contiguous good backing store
1009 if (before != NULL) {
1012 for (i = 1; i < (SWB_NPAGES/2); ++i) {
1017 blk = swp_pager_meta_ctl(object, pindex - i, 0);
1018 if (blk != blk0 - i)
1025 * find forward-looking contiguous good backing store
1027 if (after != NULL) {
1030 for (i = 1; i < (SWB_NPAGES/2); ++i) {
1033 blk = swp_pager_meta_ctl(object, pindex + i, 0);
1034 if (blk != blk0 + i)
1043 * SWAP_PAGER_PAGE_UNSWAPPED() - remove swap backing store related to page
1045 * This removes any associated swap backing store, whether valid or
1046 * not, from the page.
1048 * This routine is typically called when a page is made dirty, at
1049 * which point any associated swap can be freed. MADV_FREE also
1050 * calls us in a special-case situation
1052 * NOTE!!! If the page is clean and the swap was valid, the caller
1053 * should make the page dirty before calling this routine. This routine
1054 * does NOT change the m->dirty status of the page. Also: MADV_FREE
1057 * This routine may not sleep.
1060 swap_pager_unswapped(vm_page_t m)
1063 VM_OBJECT_LOCK_ASSERT(m->object, RA_WLOCKED);
1064 swp_pager_meta_ctl(m->object, m->pindex, SWM_FREE);
1068 * SWAP_PAGER_GETPAGES() - bring pages in from swap
1070 * Attempt to retrieve (m, count) pages from backing store, but make
1071 * sure we retrieve at least m[reqpage]. We try to load in as large
1072 * a chunk surrounding m[reqpage] as is contiguous in swap and which
1073 * belongs to the same object.
1075 * The code is designed for asynchronous operation and
1076 * immediate-notification of 'reqpage' but tends not to be
1077 * used that way. Please do not optimize-out this algorithmic
1078 * feature, I intend to improve on it in the future.
1080 * The parent has a single vm_object_pip_add() reference prior to
1081 * calling us and we should return with the same.
1083 * The parent has BUSY'd the pages. We should return with 'm'
1084 * left busy, but the others adjusted.
1087 swap_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
1097 KASSERT(mreq->object == object,
1098 ("swap_pager_getpages: object mismatch %p/%p",
1099 object, mreq->object));
1102 * Calculate range to retrieve. The pages have already been assigned
1103 * their swapblks. We require a *contiguous* range but we know it to
1104 * not span devices. If we do not supply it, bad things
1105 * happen. Note that blk, iblk & jblk can be SWAPBLK_NONE, but the
1106 * loops are set up such that the case(s) are handled implicitly.
1108 * The swp_*() calls must be made with the object locked.
1110 blk = swp_pager_meta_ctl(mreq->object, mreq->pindex, 0);
1112 for (i = reqpage - 1; i >= 0; --i) {
1115 iblk = swp_pager_meta_ctl(m[i]->object, m[i]->pindex, 0);
1116 if (blk != iblk + (reqpage - i))
1121 for (j = reqpage + 1; j < count; ++j) {
1124 jblk = swp_pager_meta_ctl(m[j]->object, m[j]->pindex, 0);
1125 if (blk != jblk - (j - reqpage))
1130 * free pages outside our collection range. Note: we never free
1131 * mreq, it must remain busy throughout.
1133 if (0 < i || j < count) {
1136 for (k = 0; k < i; ++k)
1137 swp_pager_free_nrpage(m[k]);
1138 for (k = j; k < count; ++k)
1139 swp_pager_free_nrpage(m[k]);
1143 * Return VM_PAGER_FAIL if we have nothing to do. Return mreq
1144 * still busy, but the others unbusied.
1146 if (blk == SWAPBLK_NONE)
1147 return (VM_PAGER_FAIL);
1150 * Getpbuf() can sleep.
1152 VM_OBJECT_WUNLOCK(object);
1154 * Get a swap buffer header to perform the IO
1156 bp = getpbuf(&nsw_rcount);
1157 bp->b_flags |= B_PAGING;
1160 * map our page(s) into kva for input
1162 pmap_qenter((vm_offset_t)bp->b_data, m + i, j - i);
1164 bp->b_iocmd = BIO_READ;
1165 bp->b_iodone = swp_pager_async_iodone;
1166 bp->b_rcred = crhold(thread0.td_ucred);
1167 bp->b_wcred = crhold(thread0.td_ucred);
1168 bp->b_blkno = blk - (reqpage - i);
1169 bp->b_bcount = PAGE_SIZE * (j - i);
1170 bp->b_bufsize = PAGE_SIZE * (j - i);
1171 bp->b_pager.pg_reqpage = reqpage - i;
1173 VM_OBJECT_WLOCK(object);
1177 for (k = i; k < j; ++k) {
1178 bp->b_pages[k - i] = m[k];
1179 m[k]->oflags |= VPO_SWAPINPROG;
1182 bp->b_npages = j - i;
1184 PCPU_INC(cnt.v_swapin);
1185 PCPU_ADD(cnt.v_swappgsin, bp->b_npages);
1188 * We still hold the lock on mreq, and our automatic completion routine
1189 * does not remove it.
1191 vm_object_pip_add(object, bp->b_npages);
1192 VM_OBJECT_WUNLOCK(object);
1195 * perform the I/O. NOTE!!! bp cannot be considered valid after
1196 * this point because we automatically release it on completion.
1197 * Instead, we look at the one page we are interested in which we
1198 * still hold a lock on even through the I/O completion.
1200 * The other pages in our m[] array are also released on completion,
1201 * so we cannot assume they are valid anymore either.
1203 * NOTE: b_blkno is destroyed by the call to swapdev_strategy
1206 swp_pager_strategy(bp);
1209 * wait for the page we want to complete. VPO_SWAPINPROG is always
1210 * cleared on completion. If an I/O error occurs, SWAPBLK_NONE
1211 * is set in the meta-data.
1213 VM_OBJECT_WLOCK(object);
1214 while ((mreq->oflags & VPO_SWAPINPROG) != 0) {
1215 mreq->oflags |= VPO_WANTED;
1216 PCPU_INC(cnt.v_intrans);
1217 if (VM_OBJECT_SLEEP(mreq, object, PSWP, "swread", hz * 20)) {
1219 "swap_pager: indefinite wait buffer: bufobj: %p, blkno: %jd, size: %ld\n",
1220 bp->b_bufobj, (intmax_t)bp->b_blkno, bp->b_bcount);
1225 * mreq is left busied after completion, but all the other pages
1226 * are freed. If we had an unrecoverable read error the page will
1229 if (mreq->valid != VM_PAGE_BITS_ALL) {
1230 return (VM_PAGER_ERROR);
1232 return (VM_PAGER_OK);
1236 * A final note: in a low swap situation, we cannot deallocate swap
1237 * and mark a page dirty here because the caller is likely to mark
1238 * the page clean when we return, causing the page to possibly revert
1239 * to all-zero's later.
1244 * swap_pager_putpages:
1246 * Assign swap (if necessary) and initiate I/O on the specified pages.
1248 * We support both OBJT_DEFAULT and OBJT_SWAP objects. DEFAULT objects
1249 * are automatically converted to SWAP objects.
1251 * In a low memory situation we may block in VOP_STRATEGY(), but the new
1252 * vm_page reservation system coupled with properly written VFS devices
1253 * should ensure that no low-memory deadlock occurs. This is an area
1256 * The parent has N vm_object_pip_add() references prior to
1257 * calling us and will remove references for rtvals[] that are
1258 * not set to VM_PAGER_PEND. We need to remove the rest on I/O
1261 * The parent has soft-busy'd the pages it passes us and will unbusy
1262 * those whos rtvals[] entry is not set to VM_PAGER_PEND on return.
1263 * We need to unbusy the rest on I/O completion.
1266 swap_pager_putpages(vm_object_t object, vm_page_t *m, int count,
1267 boolean_t sync, int *rtvals)
1272 if (count && m[0]->object != object) {
1273 panic("swap_pager_putpages: object mismatch %p/%p",
1282 * Turn object into OBJT_SWAP
1283 * check for bogus sysops
1284 * force sync if not pageout process
1286 if (object->type != OBJT_SWAP)
1287 swp_pager_meta_build(object, 0, SWAPBLK_NONE);
1288 VM_OBJECT_WUNLOCK(object);
1290 if (curproc != pageproc)
1296 * Update nsw parameters from swap_async_max sysctl values.
1297 * Do not let the sysop crash the machine with bogus numbers.
1299 mtx_lock(&pbuf_mtx);
1300 if (swap_async_max != nsw_wcount_async_max) {
1306 if ((n = swap_async_max) > nswbuf / 2)
1313 * Adjust difference ( if possible ). If the current async
1314 * count is too low, we may not be able to make the adjustment
1317 n -= nsw_wcount_async_max;
1318 if (nsw_wcount_async + n >= 0) {
1319 nsw_wcount_async += n;
1320 nsw_wcount_async_max += n;
1321 wakeup(&nsw_wcount_async);
1324 mtx_unlock(&pbuf_mtx);
1329 * Assign swap blocks and issue I/O. We reallocate swap on the fly.
1330 * The page is left dirty until the pageout operation completes
1333 for (i = 0; i < count; i += n) {
1339 * Maximum I/O size is limited by a number of factors.
1341 n = min(BLIST_MAX_ALLOC, count - i);
1342 n = min(n, nsw_cluster_max);
1345 * Get biggest block of swap we can. If we fail, fall
1346 * back and try to allocate a smaller block. Don't go
1347 * overboard trying to allocate space if it would overly
1351 (blk = swp_pager_getswapspace(n)) == SWAPBLK_NONE &&
1356 if (blk == SWAPBLK_NONE) {
1357 for (j = 0; j < n; ++j)
1358 rtvals[i+j] = VM_PAGER_FAIL;
1363 * All I/O parameters have been satisfied, build the I/O
1364 * request and assign the swap space.
1367 bp = getpbuf(&nsw_wcount_sync);
1369 bp = getpbuf(&nsw_wcount_async);
1370 bp->b_flags = B_ASYNC;
1372 bp->b_flags |= B_PAGING;
1373 bp->b_iocmd = BIO_WRITE;
1375 pmap_qenter((vm_offset_t)bp->b_data, &m[i], n);
1377 bp->b_rcred = crhold(thread0.td_ucred);
1378 bp->b_wcred = crhold(thread0.td_ucred);
1379 bp->b_bcount = PAGE_SIZE * n;
1380 bp->b_bufsize = PAGE_SIZE * n;
1383 VM_OBJECT_WLOCK(object);
1384 for (j = 0; j < n; ++j) {
1385 vm_page_t mreq = m[i+j];
1387 swp_pager_meta_build(
1392 vm_page_dirty(mreq);
1393 rtvals[i+j] = VM_PAGER_OK;
1395 mreq->oflags |= VPO_SWAPINPROG;
1396 bp->b_pages[j] = mreq;
1398 VM_OBJECT_WUNLOCK(object);
1401 * Must set dirty range for NFS to work.
1404 bp->b_dirtyend = bp->b_bcount;
1406 PCPU_INC(cnt.v_swapout);
1407 PCPU_ADD(cnt.v_swappgsout, bp->b_npages);
1412 * NOTE: b_blkno is destroyed by the call to swapdev_strategy
1414 if (sync == FALSE) {
1415 bp->b_iodone = swp_pager_async_iodone;
1417 swp_pager_strategy(bp);
1419 for (j = 0; j < n; ++j)
1420 rtvals[i+j] = VM_PAGER_PEND;
1421 /* restart outter loop */
1428 * NOTE: b_blkno is destroyed by the call to swapdev_strategy
1430 bp->b_iodone = bdone;
1431 swp_pager_strategy(bp);
1434 * Wait for the sync I/O to complete, then update rtvals.
1435 * We just set the rtvals[] to VM_PAGER_PEND so we can call
1436 * our async completion routine at the end, thus avoiding a
1439 bwait(bp, PVM, "swwrt");
1440 for (j = 0; j < n; ++j)
1441 rtvals[i+j] = VM_PAGER_PEND;
1443 * Now that we are through with the bp, we can call the
1444 * normal async completion, which frees everything up.
1446 swp_pager_async_iodone(bp);
1448 VM_OBJECT_WLOCK(object);
1452 * swp_pager_async_iodone:
1454 * Completion routine for asynchronous reads and writes from/to swap.
1455 * Also called manually by synchronous code to finish up a bp.
1457 * For READ operations, the pages are VPO_BUSY'd. For WRITE operations,
1458 * the pages are vm_page_t->busy'd. For READ operations, we VPO_BUSY
1459 * unbusy all pages except the 'main' request page. For WRITE
1460 * operations, we vm_page_t->busy'd unbusy all pages ( we can do this
1461 * because we marked them all VM_PAGER_PEND on return from putpages ).
1463 * This routine may not sleep.
1466 swp_pager_async_iodone(struct buf *bp)
1469 vm_object_t object = NULL;
1474 if (bp->b_ioflags & BIO_ERROR) {
1476 "swap_pager: I/O error - %s failed; blkno %ld,"
1477 "size %ld, error %d\n",
1478 ((bp->b_iocmd == BIO_READ) ? "pagein" : "pageout"),
1486 * remove the mapping for kernel virtual
1488 pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages);
1491 object = bp->b_pages[0]->object;
1492 VM_OBJECT_WLOCK(object);
1496 * cleanup pages. If an error occurs writing to swap, we are in
1497 * very serious trouble. If it happens to be a disk error, though,
1498 * we may be able to recover by reassigning the swap later on. So
1499 * in this case we remove the m->swapblk assignment for the page
1500 * but do not free it in the rlist. The errornous block(s) are thus
1501 * never reallocated as swap. Redirty the page and continue.
1503 for (i = 0; i < bp->b_npages; ++i) {
1504 vm_page_t m = bp->b_pages[i];
1506 m->oflags &= ~VPO_SWAPINPROG;
1508 if (bp->b_ioflags & BIO_ERROR) {
1510 * If an error occurs I'd love to throw the swapblk
1511 * away without freeing it back to swapspace, so it
1512 * can never be used again. But I can't from an
1515 if (bp->b_iocmd == BIO_READ) {
1517 * When reading, reqpage needs to stay
1518 * locked for the parent, but all other
1519 * pages can be freed. We still want to
1520 * wakeup the parent waiting on the page,
1521 * though. ( also: pg_reqpage can be -1 and
1522 * not match anything ).
1524 * We have to wake specifically requested pages
1525 * up too because we cleared VPO_SWAPINPROG and
1526 * someone may be waiting for that.
1528 * NOTE: for reads, m->dirty will probably
1529 * be overridden by the original caller of
1530 * getpages so don't play cute tricks here.
1533 if (i != bp->b_pager.pg_reqpage)
1534 swp_pager_free_nrpage(m);
1538 * If i == bp->b_pager.pg_reqpage, do not wake
1539 * the page up. The caller needs to.
1543 * If a write error occurs, reactivate page
1544 * so it doesn't clog the inactive list,
1545 * then finish the I/O.
1549 vm_page_activate(m);
1551 vm_page_io_finish(m);
1553 } else if (bp->b_iocmd == BIO_READ) {
1555 * NOTE: for reads, m->dirty will probably be
1556 * overridden by the original caller of getpages so
1557 * we cannot set them in order to free the underlying
1558 * swap in a low-swap situation. I don't think we'd
1559 * want to do that anyway, but it was an optimization
1560 * that existed in the old swapper for a time before
1561 * it got ripped out due to precisely this problem.
1563 * If not the requested page then deactivate it.
1565 * Note that the requested page, reqpage, is left
1566 * busied, but we still have to wake it up. The
1567 * other pages are released (unbusied) by
1570 KASSERT(!pmap_page_is_mapped(m),
1571 ("swp_pager_async_iodone: page %p is mapped", m));
1572 m->valid = VM_PAGE_BITS_ALL;
1573 KASSERT(m->dirty == 0,
1574 ("swp_pager_async_iodone: page %p is dirty", m));
1577 * We have to wake specifically requested pages
1578 * up too because we cleared VPO_SWAPINPROG and
1579 * could be waiting for it in getpages. However,
1580 * be sure to not unbusy getpages specifically
1581 * requested page - getpages expects it to be
1584 if (i != bp->b_pager.pg_reqpage) {
1586 vm_page_deactivate(m);
1593 * For write success, clear the dirty
1594 * status, then finish the I/O ( which decrements the
1595 * busy count and possibly wakes waiter's up ).
1597 KASSERT(!pmap_page_is_write_mapped(m),
1598 ("swp_pager_async_iodone: page %p is not write"
1601 vm_page_io_finish(m);
1602 if (vm_page_count_severe()) {
1604 vm_page_try_to_cache(m);
1611 * adjust pip. NOTE: the original parent may still have its own
1612 * pip refs on the object.
1614 if (object != NULL) {
1615 vm_object_pip_wakeupn(object, bp->b_npages);
1616 VM_OBJECT_WUNLOCK(object);
1620 * swapdev_strategy() manually sets b_vp and b_bufobj before calling
1621 * bstrategy(). Set them back to NULL now we're done with it, or we'll
1622 * trigger a KASSERT in relpbuf().
1626 bp->b_bufobj = NULL;
1629 * release the physical I/O buffer
1633 ((bp->b_iocmd == BIO_READ) ? &nsw_rcount :
1634 ((bp->b_flags & B_ASYNC) ?
1643 * swap_pager_isswapped:
1645 * Return 1 if at least one page in the given object is paged
1646 * out to the given swap device.
1648 * This routine may not sleep.
1651 swap_pager_isswapped(vm_object_t object, struct swdevt *sp)
1657 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
1658 if (object->type != OBJT_SWAP)
1661 mtx_lock(&swhash_mtx);
1662 for (bcount = 0; bcount < object->un_pager.swp.swp_bcount; bcount++) {
1663 struct swblock *swap;
1665 if ((swap = *swp_pager_hash(object, index)) != NULL) {
1666 for (i = 0; i < SWAP_META_PAGES; ++i) {
1667 if (swp_pager_isondev(swap->swb_pages[i], sp)) {
1668 mtx_unlock(&swhash_mtx);
1673 index += SWAP_META_PAGES;
1675 mtx_unlock(&swhash_mtx);
1680 * SWP_PAGER_FORCE_PAGEIN() - force a swap block to be paged in
1682 * This routine dissociates the page at the given index within a
1683 * swap block from its backing store, paging it in if necessary.
1684 * If the page is paged in, it is placed in the inactive queue,
1685 * since it had its backing store ripped out from under it.
1686 * We also attempt to swap in all other pages in the swap block,
1687 * we only guarantee that the one at the specified index is
1690 * XXX - The code to page the whole block in doesn't work, so we
1691 * revert to the one-by-one behavior for now. Sigh.
1694 swp_pager_force_pagein(vm_object_t object, vm_pindex_t pindex)
1698 vm_object_pip_add(object, 1);
1699 m = vm_page_grab(object, pindex, VM_ALLOC_NORMAL|VM_ALLOC_RETRY);
1700 if (m->valid == VM_PAGE_BITS_ALL) {
1701 vm_object_pip_subtract(object, 1);
1704 vm_page_activate(m);
1707 vm_pager_page_unswapped(m);
1711 if (swap_pager_getpages(object, &m, 1, 0) != VM_PAGER_OK)
1712 panic("swap_pager_force_pagein: read from swap failed");/*XXX*/
1713 vm_object_pip_subtract(object, 1);
1716 vm_page_deactivate(m);
1719 vm_pager_page_unswapped(m);
1723 * swap_pager_swapoff:
1725 * Page in all of the pages that have been paged out to the
1726 * given device. The corresponding blocks in the bitmap must be
1727 * marked as allocated and the device must be flagged SW_CLOSING.
1728 * There may be no processes swapped out to the device.
1730 * This routine may block.
1733 swap_pager_swapoff(struct swdevt *sp)
1735 struct swblock *swap;
1742 mtx_lock(&swhash_mtx);
1743 for (i = 0; i <= swhash_mask; i++) { /* '<=' is correct here */
1745 for (swap = swhash[i]; swap != NULL; swap = swap->swb_hnext) {
1746 vm_object_t object = swap->swb_object;
1747 vm_pindex_t pindex = swap->swb_index;
1748 for (j = 0; j < SWAP_META_PAGES; ++j) {
1749 if (swp_pager_isondev(swap->swb_pages[j], sp)) {
1750 /* avoid deadlock */
1751 if (!VM_OBJECT_TRYWLOCK(object)) {
1754 mtx_unlock(&swhash_mtx);
1755 swp_pager_force_pagein(object,
1757 VM_OBJECT_WUNLOCK(object);
1758 mtx_lock(&swhash_mtx);
1765 mtx_unlock(&swhash_mtx);
1768 * Objects may be locked or paging to the device being
1769 * removed, so we will miss their pages and need to
1770 * make another pass. We have marked this device as
1771 * SW_CLOSING, so the activity should finish soon.
1774 if (retries > 100) {
1775 panic("swapoff: failed to locate %d swap blocks",
1778 pause("swpoff", hz / 20);
1783 /************************************************************************
1785 ************************************************************************
1787 * These routines manipulate the swap metadata stored in the
1790 * Swap metadata is implemented with a global hash and not directly
1791 * linked into the object. Instead the object simply contains
1792 * appropriate tracking counters.
1796 * SWP_PAGER_META_BUILD() - add swap block to swap meta data for object
1798 * We first convert the object to a swap object if it is a default
1801 * The specified swapblk is added to the object's swap metadata. If
1802 * the swapblk is not valid, it is freed instead. Any previously
1803 * assigned swapblk is freed.
1806 swp_pager_meta_build(vm_object_t object, vm_pindex_t pindex, daddr_t swapblk)
1808 static volatile int exhausted;
1809 struct swblock *swap;
1810 struct swblock **pswap;
1813 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
1815 * Convert default object to swap object if necessary
1817 if (object->type != OBJT_SWAP) {
1818 object->type = OBJT_SWAP;
1819 object->un_pager.swp.swp_bcount = 0;
1821 if (object->handle != NULL) {
1822 mtx_lock(&sw_alloc_mtx);
1824 NOBJLIST(object->handle),
1828 mtx_unlock(&sw_alloc_mtx);
1833 * Locate hash entry. If not found create, but if we aren't adding
1834 * anything just return. If we run out of space in the map we wait
1835 * and, since the hash table may have changed, retry.
1838 mtx_lock(&swhash_mtx);
1839 pswap = swp_pager_hash(object, pindex);
1841 if ((swap = *pswap) == NULL) {
1844 if (swapblk == SWAPBLK_NONE)
1847 swap = *pswap = uma_zalloc(swap_zone, M_NOWAIT);
1849 mtx_unlock(&swhash_mtx);
1850 VM_OBJECT_WUNLOCK(object);
1851 if (uma_zone_exhausted(swap_zone)) {
1852 if (atomic_cmpset_int(&exhausted, 0, 1))
1853 printf("swap zone exhausted, "
1854 "increase kern.maxswzone\n");
1855 vm_pageout_oom(VM_OOM_SWAPZ);
1856 pause("swzonex", 10);
1859 VM_OBJECT_WLOCK(object);
1863 if (atomic_cmpset_int(&exhausted, 1, 0))
1864 printf("swap zone ok\n");
1866 swap->swb_hnext = NULL;
1867 swap->swb_object = object;
1868 swap->swb_index = pindex & ~(vm_pindex_t)SWAP_META_MASK;
1869 swap->swb_count = 0;
1871 ++object->un_pager.swp.swp_bcount;
1873 for (i = 0; i < SWAP_META_PAGES; ++i)
1874 swap->swb_pages[i] = SWAPBLK_NONE;
1878 * Delete prior contents of metadata
1880 idx = pindex & SWAP_META_MASK;
1882 if (swap->swb_pages[idx] != SWAPBLK_NONE) {
1883 swp_pager_freeswapspace(swap->swb_pages[idx], 1);
1888 * Enter block into metadata
1890 swap->swb_pages[idx] = swapblk;
1891 if (swapblk != SWAPBLK_NONE)
1894 mtx_unlock(&swhash_mtx);
1898 * SWP_PAGER_META_FREE() - free a range of blocks in the object's swap metadata
1900 * The requested range of blocks is freed, with any associated swap
1901 * returned to the swap bitmap.
1903 * This routine will free swap metadata structures as they are cleaned
1904 * out. This routine does *NOT* operate on swap metadata associated
1905 * with resident pages.
1908 swp_pager_meta_free(vm_object_t object, vm_pindex_t index, daddr_t count)
1911 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
1912 if (object->type != OBJT_SWAP)
1916 struct swblock **pswap;
1917 struct swblock *swap;
1919 mtx_lock(&swhash_mtx);
1920 pswap = swp_pager_hash(object, index);
1922 if ((swap = *pswap) != NULL) {
1923 daddr_t v = swap->swb_pages[index & SWAP_META_MASK];
1925 if (v != SWAPBLK_NONE) {
1926 swp_pager_freeswapspace(v, 1);
1927 swap->swb_pages[index & SWAP_META_MASK] =
1929 if (--swap->swb_count == 0) {
1930 *pswap = swap->swb_hnext;
1931 uma_zfree(swap_zone, swap);
1932 --object->un_pager.swp.swp_bcount;
1938 int n = SWAP_META_PAGES - (index & SWAP_META_MASK);
1942 mtx_unlock(&swhash_mtx);
1947 * SWP_PAGER_META_FREE_ALL() - destroy all swap metadata associated with object
1949 * This routine locates and destroys all swap metadata associated with
1953 swp_pager_meta_free_all(vm_object_t object)
1957 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
1958 if (object->type != OBJT_SWAP)
1961 while (object->un_pager.swp.swp_bcount) {
1962 struct swblock **pswap;
1963 struct swblock *swap;
1965 mtx_lock(&swhash_mtx);
1966 pswap = swp_pager_hash(object, index);
1967 if ((swap = *pswap) != NULL) {
1970 for (i = 0; i < SWAP_META_PAGES; ++i) {
1971 daddr_t v = swap->swb_pages[i];
1972 if (v != SWAPBLK_NONE) {
1974 swp_pager_freeswapspace(v, 1);
1977 if (swap->swb_count != 0)
1978 panic("swap_pager_meta_free_all: swb_count != 0");
1979 *pswap = swap->swb_hnext;
1980 uma_zfree(swap_zone, swap);
1981 --object->un_pager.swp.swp_bcount;
1983 mtx_unlock(&swhash_mtx);
1984 index += SWAP_META_PAGES;
1989 * SWP_PAGER_METACTL() - misc control of swap and vm_page_t meta data.
1991 * This routine is capable of looking up, popping, or freeing
1992 * swapblk assignments in the swap meta data or in the vm_page_t.
1993 * The routine typically returns the swapblk being looked-up, or popped,
1994 * or SWAPBLK_NONE if the block was freed, or SWAPBLK_NONE if the block
1995 * was invalid. This routine will automatically free any invalid
1996 * meta-data swapblks.
1998 * It is not possible to store invalid swapblks in the swap meta data
1999 * (other then a literal 'SWAPBLK_NONE'), so we don't bother checking.
2001 * When acting on a busy resident page and paging is in progress, we
2002 * have to wait until paging is complete but otherwise can act on the
2005 * SWM_FREE remove and free swap block from metadata
2006 * SWM_POP remove from meta data but do not free.. pop it out
2009 swp_pager_meta_ctl(vm_object_t object, vm_pindex_t pindex, int flags)
2011 struct swblock **pswap;
2012 struct swblock *swap;
2016 VM_OBJECT_LOCK_ASSERT(object, RA_WLOCKED);
2018 * The meta data only exists of the object is OBJT_SWAP
2019 * and even then might not be allocated yet.
2021 if (object->type != OBJT_SWAP)
2022 return (SWAPBLK_NONE);
2025 mtx_lock(&swhash_mtx);
2026 pswap = swp_pager_hash(object, pindex);
2028 if ((swap = *pswap) != NULL) {
2029 idx = pindex & SWAP_META_MASK;
2030 r1 = swap->swb_pages[idx];
2032 if (r1 != SWAPBLK_NONE) {
2033 if (flags & SWM_FREE) {
2034 swp_pager_freeswapspace(r1, 1);
2037 if (flags & (SWM_FREE|SWM_POP)) {
2038 swap->swb_pages[idx] = SWAPBLK_NONE;
2039 if (--swap->swb_count == 0) {
2040 *pswap = swap->swb_hnext;
2041 uma_zfree(swap_zone, swap);
2042 --object->un_pager.swp.swp_bcount;
2047 mtx_unlock(&swhash_mtx);
2052 * System call swapon(name) enables swapping on device name,
2053 * which must be in the swdevsw. Return EBUSY
2054 * if already swapping on this device.
2056 #ifndef _SYS_SYSPROTO_H_
2057 struct swapon_args {
2067 sys_swapon(struct thread *td, struct swapon_args *uap)
2071 struct nameidata nd;
2074 error = priv_check(td, PRIV_SWAPON);
2079 while (swdev_syscall_active)
2080 tsleep(&swdev_syscall_active, PUSER - 1, "swpon", 0);
2081 swdev_syscall_active = 1;
2084 * Swap metadata may not fit in the KVM if we have physical
2087 if (swap_zone == NULL) {
2092 NDINIT(&nd, LOOKUP, ISOPEN | FOLLOW | AUDITVNODE1, UIO_USERSPACE,
2098 NDFREE(&nd, NDF_ONLY_PNBUF);
2101 if (vn_isdisk(vp, &error)) {
2102 error = swapongeom(td, vp);
2103 } else if (vp->v_type == VREG &&
2104 (vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 &&
2105 (error = VOP_GETATTR(vp, &attr, td->td_ucred)) == 0) {
2107 * Allow direct swapping to NFS regular files in the same
2108 * way that nfs_mountroot() sets up diskless swapping.
2110 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE);
2116 swdev_syscall_active = 0;
2117 wakeup_one(&swdev_syscall_active);
2123 * Check that the total amount of swap currently configured does not
2124 * exceed half the theoretical maximum. If it does, print a warning
2125 * message and return -1; otherwise, return 0.
2128 swapon_check_swzone(unsigned long npages)
2130 unsigned long maxpages;
2132 /* absolute maximum we can handle assuming 100% efficiency */
2133 maxpages = uma_zone_get_max(swap_zone) * SWAP_META_PAGES;
2135 /* recommend using no more than half that amount */
2136 if (npages > maxpages / 2) {
2137 printf("warning: total configured swap (%lu pages) "
2138 "exceeds maximum recommended amount (%lu pages).\n",
2139 npages, maxpages / 2);
2140 printf("warning: increase kern.maxswzone "
2141 "or reduce amount of swap.\n");
2148 swaponsomething(struct vnode *vp, void *id, u_long nblks, sw_strategy_t *strategy, sw_close_t *close, dev_t dev)
2150 struct swdevt *sp, *tsp;
2155 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
2156 * First chop nblks off to page-align it, then convert.
2158 * sw->sw_nblks is in page-sized chunks now too.
2160 nblks &= ~(ctodb(1) - 1);
2161 nblks = dbtoc(nblks);
2164 * If we go beyond this, we get overflows in the radix
2167 mblocks = 0x40000000 / BLIST_META_RADIX;
2168 if (nblks > mblocks) {
2170 "WARNING: reducing swap size to maximum of %luMB per unit\n",
2171 mblocks / 1024 / 1024 * PAGE_SIZE);
2175 sp = malloc(sizeof *sp, M_VMPGDATA, M_WAITOK | M_ZERO);
2180 sp->sw_nblks = nblks;
2182 sp->sw_strategy = strategy;
2183 sp->sw_close = close;
2185 sp->sw_blist = blist_create(nblks, M_WAITOK);
2187 * Do not free the first two block in order to avoid overwriting
2188 * any bsd label at the front of the partition
2190 blist_free(sp->sw_blist, 2, nblks - 2);
2193 mtx_lock(&sw_dev_mtx);
2194 TAILQ_FOREACH(tsp, &swtailq, sw_list) {
2195 if (tsp->sw_end >= dvbase) {
2197 * We put one uncovered page between the devices
2198 * in order to definitively prevent any cross-device
2201 dvbase = tsp->sw_end + 1;
2204 sp->sw_first = dvbase;
2205 sp->sw_end = dvbase + nblks;
2206 TAILQ_INSERT_TAIL(&swtailq, sp, sw_list);
2208 swap_pager_avail += nblks;
2209 swap_total += (vm_ooffset_t)nblks * PAGE_SIZE;
2210 swapon_check_swzone(swap_total / PAGE_SIZE);
2212 mtx_unlock(&sw_dev_mtx);
2216 * SYSCALL: swapoff(devname)
2218 * Disable swapping on the given device.
2220 * XXX: Badly designed system call: it should use a device index
2221 * rather than filename as specification. We keep sw_vp around
2222 * only to make this work.
2224 #ifndef _SYS_SYSPROTO_H_
2225 struct swapoff_args {
2235 sys_swapoff(struct thread *td, struct swapoff_args *uap)
2238 struct nameidata nd;
2242 error = priv_check(td, PRIV_SWAPOFF);
2247 while (swdev_syscall_active)
2248 tsleep(&swdev_syscall_active, PUSER - 1, "swpoff", 0);
2249 swdev_syscall_active = 1;
2251 NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name,
2256 NDFREE(&nd, NDF_ONLY_PNBUF);
2259 mtx_lock(&sw_dev_mtx);
2260 TAILQ_FOREACH(sp, &swtailq, sw_list) {
2261 if (sp->sw_vp == vp)
2264 mtx_unlock(&sw_dev_mtx);
2269 error = swapoff_one(sp, td->td_ucred);
2271 swdev_syscall_active = 0;
2272 wakeup_one(&swdev_syscall_active);
2278 swapoff_one(struct swdevt *sp, struct ucred *cred)
2280 u_long nblks, dvbase;
2285 mtx_assert(&Giant, MA_OWNED);
2287 (void) vn_lock(sp->sw_vp, LK_EXCLUSIVE | LK_RETRY);
2288 error = mac_system_check_swapoff(cred, sp->sw_vp);
2289 (void) VOP_UNLOCK(sp->sw_vp, 0);
2293 nblks = sp->sw_nblks;
2296 * We can turn off this swap device safely only if the
2297 * available virtual memory in the system will fit the amount
2298 * of data we will have to page back in, plus an epsilon so
2299 * the system doesn't become critically low on swap space.
2301 if (cnt.v_free_count + cnt.v_cache_count + swap_pager_avail <
2302 nblks + nswap_lowat) {
2307 * Prevent further allocations on this device.
2309 mtx_lock(&sw_dev_mtx);
2310 sp->sw_flags |= SW_CLOSING;
2311 for (dvbase = 0; dvbase < sp->sw_end; dvbase += dmmax) {
2312 swap_pager_avail -= blist_fill(sp->sw_blist,
2315 swap_total -= (vm_ooffset_t)nblks * PAGE_SIZE;
2316 mtx_unlock(&sw_dev_mtx);
2319 * Page in the contents of the device and close it.
2321 swap_pager_swapoff(sp);
2323 sp->sw_close(curthread, sp);
2325 mtx_lock(&sw_dev_mtx);
2326 TAILQ_REMOVE(&swtailq, sp, sw_list);
2328 if (nswapdev == 0) {
2329 swap_pager_full = 2;
2330 swap_pager_almost_full = 1;
2334 mtx_unlock(&sw_dev_mtx);
2335 blist_destroy(sp->sw_blist);
2336 free(sp, M_VMPGDATA);
2343 struct swdevt *sp, *spt;
2344 const char *devname;
2348 while (swdev_syscall_active)
2349 tsleep(&swdev_syscall_active, PUSER - 1, "swpoff", 0);
2350 swdev_syscall_active = 1;
2352 mtx_lock(&sw_dev_mtx);
2353 TAILQ_FOREACH_SAFE(sp, &swtailq, sw_list, spt) {
2354 mtx_unlock(&sw_dev_mtx);
2355 if (vn_isdisk(sp->sw_vp, NULL))
2356 devname = devtoname(sp->sw_vp->v_rdev);
2359 error = swapoff_one(sp, thread0.td_ucred);
2361 printf("Cannot remove swap device %s (error=%d), "
2362 "skipping.\n", devname, error);
2363 } else if (bootverbose) {
2364 printf("Swap device %s removed.\n", devname);
2366 mtx_lock(&sw_dev_mtx);
2368 mtx_unlock(&sw_dev_mtx);
2370 swdev_syscall_active = 0;
2371 wakeup_one(&swdev_syscall_active);
2376 swap_pager_status(int *total, int *used)
2382 mtx_lock(&sw_dev_mtx);
2383 TAILQ_FOREACH(sp, &swtailq, sw_list) {
2384 *total += sp->sw_nblks;
2385 *used += sp->sw_used;
2387 mtx_unlock(&sw_dev_mtx);
2391 swap_dev_info(int name, struct xswdev *xs, char *devname, size_t len)
2394 const char *tmp_devname;
2399 mtx_lock(&sw_dev_mtx);
2400 TAILQ_FOREACH(sp, &swtailq, sw_list) {
2405 xs->xsw_version = XSWDEV_VERSION;
2406 xs->xsw_dev = sp->sw_dev;
2407 xs->xsw_flags = sp->sw_flags;
2408 xs->xsw_nblks = sp->sw_nblks;
2409 xs->xsw_used = sp->sw_used;
2410 if (devname != NULL) {
2411 if (vn_isdisk(sp->sw_vp, NULL))
2412 tmp_devname = devtoname(sp->sw_vp->v_rdev);
2414 tmp_devname = "[file]";
2415 strncpy(devname, tmp_devname, len);
2420 mtx_unlock(&sw_dev_mtx);
2425 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS)
2430 if (arg2 != 1) /* name length */
2432 error = swap_dev_info(*(int *)arg1, &xs, NULL, 0);
2435 error = SYSCTL_OUT(req, &xs, sizeof(xs));
2439 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswapdev, 0,
2440 "Number of swap devices");
2441 SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD, sysctl_vm_swap_info,
2442 "Swap statistics by device");
2445 * vmspace_swap_count() - count the approximate swap usage in pages for a
2448 * The map must be locked.
2450 * Swap usage is determined by taking the proportional swap used by
2451 * VM objects backing the VM map. To make up for fractional losses,
2452 * if the VM object has any swap use at all the associated map entries
2453 * count for at least 1 swap page.
2456 vmspace_swap_count(struct vmspace *vmspace)
2463 map = &vmspace->vm_map;
2466 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
2467 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2468 (object = cur->object.vm_object) != NULL) {
2469 VM_OBJECT_WLOCK(object);
2470 if (object->type == OBJT_SWAP &&
2471 object->un_pager.swp.swp_bcount != 0) {
2472 n = (cur->end - cur->start) / PAGE_SIZE;
2473 count += object->un_pager.swp.swp_bcount *
2474 SWAP_META_PAGES * n / object->size + 1;
2476 VM_OBJECT_WUNLOCK(object);
2485 * Swapping onto disk devices.
2489 static g_orphan_t swapgeom_orphan;
2491 static struct g_class g_swap_class = {
2493 .version = G_VERSION,
2494 .orphan = swapgeom_orphan,
2497 DECLARE_GEOM_CLASS(g_swap_class, g_class);
2501 swapgeom_done(struct bio *bp2)
2505 bp = bp2->bio_caller2;
2506 bp->b_ioflags = bp2->bio_flags;
2508 bp->b_ioflags |= BIO_ERROR;
2509 bp->b_resid = bp->b_bcount - bp2->bio_completed;
2510 bp->b_error = bp2->bio_error;
2516 swapgeom_strategy(struct buf *bp, struct swdevt *sp)
2519 struct g_consumer *cp;
2523 bp->b_error = ENXIO;
2524 bp->b_ioflags |= BIO_ERROR;
2528 if (bp->b_iocmd == BIO_WRITE)
2531 bio = g_alloc_bio();
2533 bp->b_error = ENOMEM;
2534 bp->b_ioflags |= BIO_ERROR;
2539 bio->bio_caller2 = bp;
2540 bio->bio_cmd = bp->b_iocmd;
2541 bio->bio_data = bp->b_data;
2542 bio->bio_offset = (bp->b_blkno - sp->sw_first) * PAGE_SIZE;
2543 bio->bio_length = bp->b_bcount;
2544 bio->bio_done = swapgeom_done;
2545 g_io_request(bio, cp);
2550 swapgeom_orphan(struct g_consumer *cp)
2554 mtx_lock(&sw_dev_mtx);
2555 TAILQ_FOREACH(sp, &swtailq, sw_list)
2556 if (sp->sw_id == cp)
2557 sp->sw_flags |= SW_CLOSING;
2558 mtx_unlock(&sw_dev_mtx);
2562 swapgeom_close_ev(void *arg, int flags)
2564 struct g_consumer *cp;
2567 g_access(cp, -1, -1, 0);
2569 g_destroy_consumer(cp);
2573 swapgeom_close(struct thread *td, struct swdevt *sw)
2576 /* XXX: direct call when Giant untangled */
2577 g_waitfor_event(swapgeom_close_ev, sw->sw_id, M_WAITOK, NULL);
2588 swapongeom_ev(void *arg, int flags)
2591 struct g_provider *pp;
2592 struct g_consumer *cp;
2593 static struct g_geom *gp;
2600 pp = g_dev_getprovider(swh->dev);
2602 swh->error = ENODEV;
2605 mtx_lock(&sw_dev_mtx);
2606 TAILQ_FOREACH(sp, &swtailq, sw_list) {
2608 if (cp != NULL && cp->provider == pp) {
2609 mtx_unlock(&sw_dev_mtx);
2614 mtx_unlock(&sw_dev_mtx);
2616 gp = g_new_geomf(&g_swap_class, "swap");
2617 cp = g_new_consumer(gp);
2620 * XXX: Everytime you think you can improve the margin for
2621 * footshooting, somebody depends on the ability to do so:
2622 * savecore(8) wants to write to our swapdev so we cannot
2623 * set an exclusive count :-(
2625 error = g_access(cp, 1, 1, 0);
2628 g_destroy_consumer(cp);
2632 nblks = pp->mediasize / DEV_BSIZE;
2633 swaponsomething(swh->vp, cp, nblks, swapgeom_strategy,
2634 swapgeom_close, dev2udev(swh->dev));
2640 swapongeom(struct thread *td, struct vnode *vp)
2645 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2647 swh.dev = vp->v_rdev;
2650 /* XXX: direct call when Giant untangled */
2651 error = g_waitfor_event(swapongeom_ev, &swh, M_WAITOK, NULL);
2661 * This is used mainly for network filesystem (read: probably only tested
2662 * with NFS) swapfiles.
2667 swapdev_strategy(struct buf *bp, struct swdevt *sp)
2671 bp->b_blkno = ctodb(bp->b_blkno - sp->sw_first);
2675 if (bp->b_iocmd == BIO_WRITE) {
2677 bufobj_wdrop(bp->b_bufobj);
2678 bufobj_wref(&vp2->v_bufobj);
2680 if (bp->b_bufobj != &vp2->v_bufobj)
2681 bp->b_bufobj = &vp2->v_bufobj;
2683 bp->b_iooffset = dbtob(bp->b_blkno);
2689 swapdev_close(struct thread *td, struct swdevt *sp)
2692 VOP_CLOSE(sp->sw_vp, FREAD | FWRITE, td->td_ucred, td);
2698 swaponvp(struct thread *td, struct vnode *vp, u_long nblks)
2705 mtx_lock(&sw_dev_mtx);
2706 TAILQ_FOREACH(sp, &swtailq, sw_list) {
2707 if (sp->sw_id == vp) {
2708 mtx_unlock(&sw_dev_mtx);
2712 mtx_unlock(&sw_dev_mtx);
2714 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2716 error = mac_system_check_swapon(td->td_ucred, vp);
2719 error = VOP_OPEN(vp, FREAD | FWRITE, td->td_ucred, td, NULL);
2720 (void) VOP_UNLOCK(vp, 0);
2724 swaponsomething(vp, vp, nblks, swapdev_strategy, swapdev_close,