FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_swap.c
1 /*
2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94
34 * $FreeBSD: releng/5.1/sys/vm/vm_swap.c 111936 2003-03-05 23:50:15Z rwatson $
35 */
36
37 #include "opt_mac.h"
38 #include "opt_swap.h"
39 #include "opt_vm.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/bio.h>
45 #include <sys/buf.h>
46 #include <sys/proc.h>
47 #include <sys/namei.h>
48 #include <sys/vnode.h>
49 #include <sys/fcntl.h>
50 #include <sys/blist.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/conf.h>
54 #include <sys/disk.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/mac.h>
58 #include <sys/mount.h>
59 #include <vm/vm.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_param.h>
62 #include <vm/vm_pageout.h>
63 #include <vm/swap_pager.h>
64 #include <vm/uma.h>
65
66 /*
67 * Indirect driver for multi-controller paging.
68 */
69
70 typedef int32_t swblk_t; /* swap offset */
71
72 #ifndef NSWAPDEV
73 #define NSWAPDEV 4
74 #endif
75 static struct swdevt should_be_malloced[NSWAPDEV];
76 struct swdevt *swdevt = should_be_malloced;
77 static int nswap; /* first block after the interleaved devs */
78 int nswdev = NSWAPDEV;
79 int vm_swap_size;
80 static int swdev_syscall_active = 0; /* serialize swap(on|off) */
81
82
83 static int swapdev_strategy(struct vop_strategy_args *ap);
84 struct vnode *swapdev_vp;
85
86 /*
87 * swapdev_strategy:
88 *
89 * VOP_STRATEGY() for swapdev_vp.
90 * Perform swap strategy interleave device selection.
91 *
92 * The bp is expected to be locked and *not* B_DONE on call.
93 */
94 static int
95 swapdev_strategy(ap)
96 struct vop_strategy_args /* {
97 struct vnode *a_vp;
98 struct buf *a_bp;
99 } */ *ap;
100 {
101 int s, sz, off, seg, index;
102 struct swdevt *sp;
103 struct vnode *vp;
104 struct buf *bp;
105
106 bp = ap->a_bp;
107 sz = howmany(bp->b_bcount, PAGE_SIZE);
108
109 /*
110 * Convert interleaved swap into per-device swap. Note that
111 * the block size is left in PAGE_SIZE'd chunks (for the newswap)
112 * here.
113 */
114 if (nswdev > 1) {
115 off = bp->b_blkno % dmmax;
116 if (off + sz > dmmax) {
117 bp->b_error = EINVAL;
118 bp->b_ioflags |= BIO_ERROR;
119 bufdone(bp);
120 return 0;
121 }
122 seg = bp->b_blkno / dmmax;
123 index = seg % nswdev;
124 seg /= nswdev;
125 bp->b_blkno = seg * dmmax + off;
126 } else {
127 index = 0;
128 }
129 sp = &swdevt[index];
130 if (bp->b_blkno + sz > sp->sw_nblks) {
131 bp->b_error = EINVAL;
132 bp->b_ioflags |= BIO_ERROR;
133 bufdone(bp);
134 return 0;
135 }
136 bp->b_dev = sp->sw_device;
137 if (sp->sw_vp == NULL) {
138 bp->b_error = ENODEV;
139 bp->b_ioflags |= BIO_ERROR;
140 bufdone(bp);
141 return 0;
142 }
143
144 /*
145 * Convert from PAGE_SIZE'd to DEV_BSIZE'd chunks for the actual I/O
146 */
147 bp->b_blkno = ctodb(bp->b_blkno);
148
149 vhold(sp->sw_vp);
150 s = splvm();
151 if (bp->b_iocmd == BIO_WRITE) {
152 vp = bp->b_vp;
153 if (vp) {
154 VI_LOCK(vp);
155 vp->v_numoutput--;
156 if ((vp->v_iflag & VI_BWAIT) && vp->v_numoutput <= 0) {
157 vp->v_iflag &= ~VI_BWAIT;
158 wakeup(&vp->v_numoutput);
159 }
160 VI_UNLOCK(vp);
161 }
162 VI_LOCK(sp->sw_vp);
163 sp->sw_vp->v_numoutput++;
164 VI_UNLOCK(sp->sw_vp);
165 }
166 bp->b_vp = sp->sw_vp;
167 splx(s);
168 if (bp->b_vp->v_type == VCHR)
169 VOP_SPECSTRATEGY(bp->b_vp, bp);
170 else
171 VOP_STRATEGY(bp->b_vp, bp);
172 return 0;
173 }
174
175 /*
176 * Create a special vnode op vector for swapdev_vp - we only use
177 * VOP_STRATEGY() and reclaim; everything else returns an error.
178 */
179 vop_t **swapdev_vnodeop_p;
180 static struct vnodeopv_entry_desc swapdev_vnodeop_entries[] = {
181 { &vop_default_desc, (vop_t *) vop_defaultop },
182 { &vop_reclaim_desc, (vop_t *) vop_null },
183 { &vop_strategy_desc, (vop_t *) swapdev_strategy },
184 { NULL, NULL }
185 };
186 static struct vnodeopv_desc swapdev_vnodeop_opv_desc =
187 { &swapdev_vnodeop_p, swapdev_vnodeop_entries };
188
189 VNODEOP_SET(swapdev_vnodeop_opv_desc);
190
191 /*
192 * System call swapon(name) enables swapping on device name,
193 * which must be in the swdevsw. Return EBUSY
194 * if already swapping on this device.
195 */
196 #ifndef _SYS_SYSPROTO_H_
197 struct swapon_args {
198 char *name;
199 };
200 #endif
201
202 /*
203 * MPSAFE
204 */
205 /* ARGSUSED */
206 int
207 swapon(td, uap)
208 struct thread *td;
209 struct swapon_args *uap;
210 {
211 struct vattr attr;
212 struct vnode *vp;
213 struct nameidata nd;
214 int error;
215
216 mtx_lock(&Giant);
217 error = suser(td);
218 if (error)
219 goto done2;
220
221 while (swdev_syscall_active)
222 tsleep(&swdev_syscall_active, PUSER - 1, "swpon", 0);
223 swdev_syscall_active = 1;
224
225 /*
226 * Swap metadata may not fit in the KVM if we have physical
227 * memory of >1GB.
228 */
229 if (swap_zone == NULL) {
230 error = ENOMEM;
231 goto done;
232 }
233
234 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, uap->name, td);
235 error = namei(&nd);
236 if (error)
237 goto done;
238
239 NDFREE(&nd, NDF_ONLY_PNBUF);
240 vp = nd.ni_vp;
241
242 if (vn_isdisk(vp, &error))
243 error = swaponvp(td, vp, vp->v_rdev, 0);
244 else if (vp->v_type == VREG &&
245 (vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 &&
246 (error = VOP_GETATTR(vp, &attr, td->td_ucred, td)) == 0) {
247 /*
248 * Allow direct swapping to NFS regular files in the same
249 * way that nfs_mountroot() sets up diskless swapping.
250 */
251 error = swaponvp(td, vp, NODEV, attr.va_size / DEV_BSIZE);
252 }
253
254 if (error)
255 vrele(vp);
256 done:
257 swdev_syscall_active = 0;
258 wakeup_one(&swdev_syscall_active);
259 done2:
260 mtx_unlock(&Giant);
261 return (error);
262 }
263
264 /*
265 * Swfree(index) frees the index'th portion of the swap map.
266 * Each of the nswdev devices provides 1/nswdev'th of the swap
267 * space, which is laid out with blocks of dmmax pages circularly
268 * among the devices.
269 *
270 * The new swap code uses page-sized blocks. The old swap code used
271 * DEV_BSIZE'd chunks.
272 */
273 int
274 swaponvp(td, vp, dev, nblks)
275 struct thread *td;
276 struct vnode *vp;
277 dev_t dev;
278 u_long nblks;
279 {
280 int index;
281 struct swdevt *sp;
282 swblk_t vsbase;
283 long blk;
284 swblk_t dvbase;
285 int error;
286 u_long aligned_nblks;
287 off_t mediasize;
288
289 if (!swapdev_vp) {
290 error = getnewvnode("none", NULL, swapdev_vnodeop_p,
291 &swapdev_vp);
292 if (error)
293 panic("Cannot get vnode for swapdev");
294 swapdev_vp->v_type = VNON; /* Untyped */
295 }
296
297 ASSERT_VOP_UNLOCKED(vp, "swaponvp");
298 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
299 if (sp->sw_vp == vp)
300 return EBUSY;
301 if (!sp->sw_vp)
302 goto found;
303
304 }
305 return EINVAL;
306 found:
307 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
308 #ifdef MAC
309 error = mac_check_system_swapon(td->td_ucred, vp);
310 if (error == 0)
311 #endif
312 error = VOP_OPEN(vp, FREAD | FWRITE, td->td_ucred, td);
313 (void) VOP_UNLOCK(vp, 0, td);
314 if (error)
315 return (error);
316
317 error = VOP_IOCTL(vp, DIOCGMEDIASIZE, (caddr_t)&mediasize,
318 FREAD, td->td_ucred, td);
319 if (error == 0)
320 nblks = mediasize / DEV_BSIZE;
321 else
322 nblks = 0;
323 /*
324 * XXX: We should also check that the sectorsize makes sense
325 * XXX: it should be a power of two, no larger than the page size.
326 */
327 if (nblks == 0) {
328 (void) VOP_CLOSE(vp, FREAD | FWRITE, td->td_ucred, td);
329 return (ENXIO);
330 }
331
332 /*
333 * If we go beyond this, we get overflows in the radix
334 * tree bitmap code.
335 */
336 if (nblks > 0x40000000 / BLIST_META_RADIX / nswdev) {
337 printf("exceeded maximum of %d blocks per swap unit\n",
338 0x40000000 / BLIST_META_RADIX / nswdev);
339 (void) VOP_CLOSE(vp, FREAD | FWRITE, td->td_ucred, td);
340 return (ENXIO);
341 }
342 /*
343 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks.
344 * First chop nblks off to page-align it, then convert.
345 *
346 * sw->sw_nblks is in page-sized chunks now too.
347 */
348 nblks &= ~(ctodb(1) - 1);
349 nblks = dbtoc(nblks);
350
351 sp->sw_vp = vp;
352 sp->sw_dev = dev2udev(dev);
353 sp->sw_device = dev;
354 sp->sw_flags = SW_FREED;
355 sp->sw_nblks = nblks;
356 sp->sw_used = 0;
357
358 /*
359 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not
360 * DEV_BSIZE'd. aligned_nblks is used to calculate the
361 * size of the swap bitmap, taking into account the stripe size.
362 */
363 aligned_nblks = (nblks + (dmmax -1)) & ~(u_long)(dmmax -1);
364
365 if (aligned_nblks * nswdev > nswap)
366 nswap = aligned_nblks * nswdev;
367
368 if (swapblist == NULL)
369 swapblist = blist_create(nswap);
370 else
371 blist_resize(&swapblist, nswap, 0);
372
373 for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
374 blk = min(nblks - dvbase, dmmax);
375 vsbase = index * dmmax + dvbase * nswdev;
376 blist_free(swapblist, vsbase, blk);
377 vm_swap_size += blk;
378 }
379
380 swap_pager_full = 0;
381
382 return (0);
383 }
384
385 /*
386 * SYSCALL: swapoff(devname)
387 *
388 * Disable swapping on the given device.
389 */
390 #ifndef _SYS_SYSPROTO_H_
391 struct swapoff_args {
392 char *name;
393 };
394 #endif
395
396 /*
397 * MPSAFE
398 */
399 /* ARGSUSED */
400 int
401 swapoff(td, uap)
402 struct thread *td;
403 struct swapoff_args *uap;
404 {
405 struct vnode *vp;
406 struct nameidata nd;
407 struct swdevt *sp;
408 swblk_t dvbase, vsbase;
409 u_long nblks, aligned_nblks, blk;
410 int error, index;
411
412 mtx_lock(&Giant);
413
414 error = suser(td);
415 if (error)
416 goto done2;
417
418 while (swdev_syscall_active)
419 tsleep(&swdev_syscall_active, PUSER - 1, "swpoff", 0);
420 swdev_syscall_active = 1;
421
422 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, uap->name, td);
423 error = namei(&nd);
424 if (error)
425 goto done;
426 NDFREE(&nd, NDF_ONLY_PNBUF);
427 vp = nd.ni_vp;
428
429 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) {
430 if (sp->sw_vp == vp)
431 goto found;
432 }
433 error = EINVAL;
434 goto done;
435 found:
436 #ifdef MAC
437 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
438 error = mac_check_system_swapoff(td->td_ucred, vp);
439 (void) VOP_UNLOCK(vp, 0, td);
440 if (error != 0)
441 goto done;
442 #endif
443
444 nblks = sp->sw_nblks;
445
446 /*
447 * We can turn off this swap device safely only if the
448 * available virtual memory in the system will fit the amount
449 * of data we will have to page back in, plus an epsilon so
450 * the system doesn't become critically low on swap space.
451 */
452 if (cnt.v_free_count + cnt.v_cache_count + vm_swap_size <
453 nblks + nswap_lowat) {
454 error = ENOMEM;
455 goto done;
456 }
457
458 /*
459 * Prevent further allocations on this device.
460 */
461 sp->sw_flags |= SW_CLOSING;
462 for (dvbase = dmmax; dvbase < nblks; dvbase += dmmax) {
463 blk = min(nblks - dvbase, dmmax);
464 vsbase = index * dmmax + dvbase * nswdev;
465 vm_swap_size -= blist_fill(swapblist, vsbase, blk);
466 }
467
468 /*
469 * Page in the contents of the device and close it.
470 */
471 #ifndef NO_SWAPPING
472 vm_proc_swapin_all(index);
473 #endif /* !NO_SWAPPING */
474 swap_pager_swapoff(index, &sp->sw_used);
475
476 VOP_CLOSE(vp, FREAD | FWRITE, td->td_ucred, td);
477 vrele(vp);
478 sp->sw_vp = NULL;
479
480 /*
481 * Resize the bitmap based on the new largest swap device,
482 * or free the bitmap if there are no more devices.
483 */
484 for (sp = swdevt, nblks = 0; sp < swdevt + nswdev; sp++) {
485 if (sp->sw_vp == NULL)
486 continue;
487 nblks = max(nblks, sp->sw_nblks);
488 }
489
490 aligned_nblks = (nblks + (dmmax -1)) & ~(u_long)(dmmax -1);
491 nswap = aligned_nblks * nswdev;
492
493 if (nswap == 0) {
494 blist_destroy(swapblist);
495 swapblist = NULL;
496 vrele(swapdev_vp);
497 swapdev_vp = NULL;
498 } else
499 blist_resize(&swapblist, nswap, 0);
500
501 done:
502 swdev_syscall_active = 0;
503 wakeup_one(&swdev_syscall_active);
504 done2:
505 mtx_unlock(&Giant);
506 return (error);
507 }
508
509 static int
510 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS)
511 {
512 int *name = (int *)arg1;
513 int error, i, n;
514 struct xswdev xs;
515 struct swdevt *sp;
516
517 if (arg2 != 1) /* name length */
518 return (EINVAL);
519
520 for (sp = swdevt, i = 0, n = 0 ; i < nswdev; i++, sp++) {
521 if (sp->sw_vp) {
522 if (n == *name) {
523 xs.xsw_version = XSWDEV_VERSION;
524 xs.xsw_dev = sp->sw_dev;
525 xs.xsw_flags = sp->sw_flags;
526 xs.xsw_nblks = sp->sw_nblks;
527 xs.xsw_used = sp->sw_used;
528
529 error = SYSCTL_OUT(req, &xs, sizeof(xs));
530 return (error);
531 }
532 n++;
533 }
534
535 }
536 return (ENOENT);
537 }
538
539 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswdev, 0,
540 "Number of swap devices");
541 SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD, sysctl_vm_swap_info,
542 "Swap statistics by device");
Cache object: ea0e5de57e9a0ab4b0689f5bfc55b645
|