1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94 36 * $FreeBSD: src/sys/vm/vm_swap.c,v 1.96.2.2 2001/10/14 18:46:47 iedowse Exp $ 37 * $DragonFly: src/sys/vm/vm_swap.c,v 1.36 2007/07/20 17:21:54 dillon Exp $ 38 */ 39 40 #include "opt_swap.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/sysproto.h> 45 #include <sys/buf.h> 46 #include <sys/proc.h> 47 #include <sys/priv.h> 48 #include <sys/nlookup.h> 49 #include <sys/sysctl.h> 50 #include <sys/dmap.h> /* XXX */ 51 #include <sys/vnode.h> 52 #include <sys/fcntl.h> 53 #include <sys/blist.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/conf.h> 57 #include <sys/stat.h> 58 59 #include <vm/vm.h> 60 #include <vm/vm_extern.h> 61 #include <vm/swap_pager.h> 62 #include <vm/vm_zone.h> 63 #include <vm/vm_param.h> 64 65 #include <sys/thread2.h> 66 #include <sys/mplock2.h> 67 #include <sys/mutex2.h> 68 #include <sys/spinlock2.h> 69 70 /* 71 * Indirect driver for multi-controller paging. 72 */ 73 74 #ifndef NSWAPDEV 75 #define NSWAPDEV 4 76 #endif 77 static struct swdevt should_be_malloced[NSWAPDEV]; 78 struct swdevt *swdevt = should_be_malloced; /* exported to pstat/systat */ 79 static swblk_t nswap; /* first block after the interleaved devs */ 80 static struct mtx swap_mtx = MTX_INITIALIZER; 81 int nswdev = NSWAPDEV; /* exported to pstat/systat */ 82 int vm_swap_size; 83 int vm_swap_max; 84 85 static int swapoff_one(int index); 86 struct vnode *swapdev_vp; 87 88 /* 89 * (struct vnode *a_vp, struct bio *b_bio) 90 * 91 * vn_strategy() for swapdev_vp. Perform swap strategy interleave device 92 * selection. 93 * 94 * No requirements. 95 */ 96 static int 97 swapdev_strategy(struct vop_strategy_args *ap) 98 { 99 struct bio *bio = ap->a_bio; 100 struct bio *nbio; 101 struct buf *bp = bio->bio_buf; 102 int sz, off, seg, index, blkno, nblkno; 103 struct swdevt *sp; 104 struct vnode *vp; 105 106 vp = ap->a_vp; 107 sz = howmany(bp->b_bcount, PAGE_SIZE); 108 blkno = (int)(bio->bio_offset >> PAGE_SHIFT); 109 110 /* 111 * Convert interleaved swap into per-device swap. Note that 112 * the block size is left in PAGE_SIZE'd chunks (for the newswap) 113 * here. 114 */ 115 nbio = push_bio(bio); 116 if (nswdev > 1) { 117 off = blkno % dmmax; 118 if (off + sz > dmmax) { 119 bp->b_error = EINVAL; 120 bp->b_flags |= B_ERROR; 121 biodone(bio); 122 return 0; 123 } 124 seg = blkno / dmmax; 125 index = seg % nswdev; 126 seg /= nswdev; 127 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT; 128 } else { 129 index = 0; 130 nbio->bio_offset = bio->bio_offset; 131 } 132 nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT); 133 sp = &swdevt[index]; 134 if (nblkno + sz > sp->sw_nblks) { 135 bp->b_error = EINVAL; 136 bp->b_flags |= B_ERROR; 137 /* I/O was never started on nbio, must biodone(bio) */ 138 biodone(bio); 139 return 0; 140 } 141 if (sp->sw_vp == NULL) { 142 bp->b_error = ENODEV; 143 bp->b_flags |= B_ERROR; 144 /* I/O was never started on nbio, must biodone(bio) */ 145 biodone(bio); 146 return 0; 147 } 148 149 /* 150 * Issue a strategy call on the appropriate swap vnode. Note that 151 * bp->b_vp is not modified. Strategy code is always supposed to 152 * use the passed vp. 153 * 154 * We have to use vn_strategy() here even if we know we have a 155 * device in order to properly break up requests which exceed the 156 * device's DMA limits. 157 */ 158 vn_strategy(sp->sw_vp, nbio); 159 return 0; 160 } 161 162 static int 163 swapdev_inactive(struct vop_inactive_args *ap) 164 { 165 vrecycle(ap->a_vp); 166 return(0); 167 } 168 169 static int 170 swapdev_reclaim(struct vop_reclaim_args *ap) 171 { 172 return(0); 173 } 174 175 /* 176 * Create a special vnode op vector for swapdev_vp - we only use 177 * vn_strategy(), everything else returns an error. 178 */ 179 static struct vop_ops swapdev_vnode_vops = { 180 .vop_default = vop_defaultop, 181 .vop_strategy = swapdev_strategy, 182 .vop_inactive = swapdev_inactive, 183 .vop_reclaim = swapdev_reclaim 184 }; 185 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops; 186 187 VNODEOP_SET(swapdev_vnode_vops); 188 189 /* 190 * swapon_args(char *name) 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 * No requirements. 197 */ 198 int 199 sys_swapon(struct swapon_args *uap) 200 { 201 struct thread *td = curthread; 202 struct vattr attr; 203 struct vnode *vp; 204 struct nlookupdata nd; 205 int error; 206 struct ucred *cred; 207 208 cred = td->td_ucred; 209 210 error = priv_check(td, PRIV_ROOT); 211 if (error) 212 return (error); 213 214 mtx_lock(&swap_mtx); 215 get_mplock(); 216 vp = NULL; 217 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 218 if (error == 0) 219 error = nlookup(&nd); 220 if (error == 0) 221 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 222 nlookup_done(&nd); 223 if (error) { 224 rel_mplock(); 225 mtx_unlock(&swap_mtx); 226 return (error); 227 } 228 229 if (vn_isdisk(vp, &error)) { 230 error = swaponvp(td, vp, 0); 231 } else if (vp->v_type == VREG && vp->v_tag == VT_NFS && 232 (error = VOP_GETATTR(vp, &attr)) == 0) { 233 /* 234 * Allow direct swapping to NFS regular files in the same 235 * way that nfs_mountroot() sets up diskless swapping. 236 */ 237 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); 238 } 239 if (error) 240 vrele(vp); 241 rel_mplock(); 242 mtx_unlock(&swap_mtx); 243 244 return (error); 245 } 246 247 /* 248 * Swfree(index) frees the index'th portion of the swap map. 249 * Each of the nswdev devices provides 1/nswdev'th of the swap 250 * space, which is laid out with blocks of dmmax pages circularly 251 * among the devices. 252 * 253 * The new swap code uses page-sized blocks. The old swap code used 254 * DEV_BSIZE'd chunks. 255 * 256 * XXX locking when multiple swapon's run in parallel 257 */ 258 int 259 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks) 260 { 261 swblk_t aligned_nblks; 262 int64_t dpsize; 263 struct ucred *cred; 264 struct swdevt *sp; 265 swblk_t vsbase; 266 swblk_t dvbase; 267 cdev_t dev; 268 int index; 269 int error; 270 swblk_t blk; 271 272 cred = td->td_ucred; 273 274 lwkt_gettoken(&vm_token); /* needed for vm_swap_size and blist */ 275 mtx_lock(&swap_mtx); 276 277 if (!swapdev_vp) { 278 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p, 279 &swapdev_vp, 0, 0); 280 if (error) 281 panic("Cannot get vnode for swapdev"); 282 swapdev_vp->v_type = VNON; /* Untyped */ 283 vx_unlock(swapdev_vp); 284 } 285 286 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) { 287 if (sp->sw_vp == vp) { 288 error = EBUSY; 289 goto done; 290 } 291 if (!sp->sw_vp) 292 goto found; 293 294 } 295 error = EINVAL; 296 goto done; 297 found: 298 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 299 error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL); 300 vn_unlock(vp); 301 if (error) 302 goto done; 303 304 /* 305 * v_rdev is not valid until after the VOP_OPEN() call. dev_psize() 306 * must be supported if a character device has been specified. 307 */ 308 if (vp->v_type == VCHR) 309 dev = vp->v_rdev; 310 else 311 dev = NULL; 312 313 if (nblks == 0 && dev != NULL) { 314 dpsize = dev_dpsize(dev); 315 if (dpsize == -1) { 316 VOP_CLOSE(vp, FREAD | FWRITE); 317 error = ENXIO; 318 goto done; 319 } 320 nblks = (u_quad_t)dpsize; 321 } 322 if (nblks == 0) { 323 VOP_CLOSE(vp, FREAD | FWRITE); 324 error = ENXIO; 325 goto done; 326 } 327 328 /* 329 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks. 330 * First chop nblks off to page-align it, then convert. 331 * 332 * sw->sw_nblks is in page-sized chunks now too. 333 */ 334 nblks &= ~(u_quad_t)(ctodb(1) - 1); 335 nblks = dbtoc(nblks); 336 337 /* 338 * Post-conversion nblks must not be >= BLIST_MAXBLKS, and 339 * we impose a 4-swap-device limit so we have to divide it out 340 * further. Going beyond this will result in overflows in the 341 * blist code. 342 * 343 * Post-conversion nblks must fit within a (swblk_t), which 344 * this test also ensures. 345 */ 346 if (nblks > BLIST_MAXBLKS / nswdev) { 347 kprintf("exceeded maximum of %d blocks per swap unit\n", 348 (int)BLIST_MAXBLKS / nswdev); 349 VOP_CLOSE(vp, FREAD | FWRITE); 350 error = ENXIO; 351 goto done; 352 } 353 354 sp->sw_vp = vp; 355 sp->sw_dev = dev2udev(dev); 356 sp->sw_device = dev; 357 sp->sw_flags = SW_FREED; 358 sp->sw_nused = 0; 359 360 /* 361 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not 362 * DEV_BSIZE'd. aligned_nblks is used to calculate the 363 * size of the swap bitmap, taking into account the stripe size. 364 */ 365 aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1)); 366 sp->sw_nblks = aligned_nblks; 367 368 if (aligned_nblks * nswdev > nswap) 369 nswap = aligned_nblks * nswdev; 370 371 if (swapblist == NULL) 372 swapblist = blist_create(nswap); 373 else 374 blist_resize(&swapblist, nswap, 0); 375 376 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) { 377 blk = min(aligned_nblks - dvbase, dmmax); 378 vsbase = index * dmmax + dvbase * nswdev; 379 blist_free(swapblist, vsbase, blk); 380 vm_swap_size += blk; 381 vm_swap_max += blk; 382 } 383 swap_pager_newswap(); 384 error = 0; 385 done: 386 mtx_unlock(&swap_mtx); 387 lwkt_reltoken(&vm_token); 388 return (error); 389 } 390 391 /* 392 * swapoff_args(char *name) 393 * 394 * System call swapoff(name) disables swapping on device name, 395 * which must be an active swap device. Return ENOMEM 396 * if there is not enough memory to page in the contents of 397 * the given device. 398 * 399 * No requirements. 400 */ 401 int 402 sys_swapoff(struct swapoff_args *uap) 403 { 404 struct vnode *vp; 405 struct nlookupdata nd; 406 struct swdevt *sp; 407 int error, index; 408 409 error = priv_check(curthread, PRIV_ROOT); 410 if (error) 411 return (error); 412 413 mtx_lock(&swap_mtx); 414 get_mplock(); 415 vp = NULL; 416 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 417 if (error == 0) 418 error = nlookup(&nd); 419 if (error == 0) 420 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 421 nlookup_done(&nd); 422 if (error) 423 goto done; 424 425 for (sp = swdevt, index = 0; index < nswdev; index++, sp++) { 426 if (sp->sw_vp == vp) 427 goto found; 428 } 429 error = EINVAL; 430 goto done; 431 found: 432 error = swapoff_one(index); 433 434 done: 435 rel_mplock(); 436 mtx_unlock(&swap_mtx); 437 return (error); 438 } 439 440 static int 441 swapoff_one(int index) 442 { 443 swblk_t blk, aligned_nblks; 444 swblk_t dvbase, vsbase; 445 u_int pq_active_clean, pq_inactive_clean; 446 struct swdevt *sp; 447 struct vm_page marker; 448 vm_page_t m; 449 int q; 450 451 mtx_lock(&swap_mtx); 452 453 sp = &swdevt[index]; 454 aligned_nblks = sp->sw_nblks; 455 pq_active_clean = pq_inactive_clean = 0; 456 457 /* 458 * We can turn off this swap device safely only if the 459 * available virtual memory in the system will fit the amount 460 * of data we will have to page back in, plus an epsilon so 461 * the system doesn't become critically low on swap space. 462 */ 463 for (q = 0; q < PQ_L2_SIZE; ++q) { 464 bzero(&marker, sizeof(marker)); 465 marker.flags = PG_BUSY | PG_FICTITIOUS | PG_MARKER; 466 marker.queue = PQ_ACTIVE + q; 467 marker.pc = q; 468 marker.wire_count = 1; 469 470 vm_page_queues_spin_lock(marker.queue); 471 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl, 472 &marker, pageq); 473 474 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) { 475 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, 476 &marker, pageq); 477 TAILQ_INSERT_AFTER(&vm_page_queues[marker.queue].pl, m, 478 &marker, pageq); 479 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 480 continue; 481 482 if (vm_page_busy_try(m, FALSE) == 0) { 483 vm_page_queues_spin_unlock(marker.queue); 484 if (m->dirty == 0) { 485 vm_page_test_dirty(m); 486 if (m->dirty == 0) 487 ++pq_active_clean; 488 } 489 vm_page_wakeup(m); 490 vm_page_queues_spin_lock(marker.queue); 491 } 492 } 493 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, &marker, pageq); 494 vm_page_queues_spin_unlock(marker.queue); 495 496 marker.queue = PQ_INACTIVE + q; 497 marker.pc = q; 498 vm_page_queues_spin_lock(marker.queue); 499 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl, 500 &marker, pageq); 501 502 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) { 503 TAILQ_REMOVE( 504 &vm_page_queues[marker.queue].pl, 505 &marker, pageq); 506 TAILQ_INSERT_AFTER( 507 &vm_page_queues[marker.queue].pl, 508 m, &marker, pageq); 509 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 510 continue; 511 512 if (vm_page_busy_try(m, FALSE) == 0) { 513 vm_page_queues_spin_unlock(marker.queue); 514 if (m->dirty == 0) { 515 vm_page_test_dirty(m); 516 if (m->dirty == 0) 517 ++pq_inactive_clean; 518 } 519 vm_page_wakeup(m); 520 vm_page_queues_spin_lock(marker.queue); 521 } 522 } 523 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, 524 &marker, pageq); 525 vm_page_queues_spin_unlock(marker.queue); 526 } 527 528 if (vmstats.v_free_count + vmstats.v_cache_count + pq_active_clean + 529 pq_inactive_clean + vm_swap_size < aligned_nblks + nswap_lowat) { 530 mtx_unlock(&swap_mtx); 531 return (ENOMEM); 532 } 533 534 /* 535 * Prevent further allocations on this device 536 */ 537 sp->sw_flags |= SW_CLOSING; 538 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) { 539 blk = min(aligned_nblks - dvbase, dmmax); 540 vsbase = index * dmmax + dvbase * nswdev; 541 vm_swap_size -= blist_fill(swapblist, vsbase, blk); 542 vm_swap_max -= blk; 543 } 544 545 /* 546 * Page in the contents of the device and close it. 547 */ 548 if (swap_pager_swapoff(index)) { 549 mtx_unlock(&swap_mtx); 550 return (EINTR); 551 } 552 553 VOP_CLOSE(sp->sw_vp, FREAD | FWRITE); 554 vrele(sp->sw_vp); 555 bzero(swdevt + index, sizeof(struct swdevt)); 556 557 /* 558 * Resize the bitmap based on the nem largest swap device, 559 * or free the bitmap if there are no more devices. 560 */ 561 for (sp = swdevt, aligned_nblks = 0; sp < swdevt + nswdev; sp++) { 562 if (sp->sw_vp) 563 aligned_nblks = max(aligned_nblks, sp->sw_nblks); 564 } 565 566 nswap = aligned_nblks * nswdev; 567 568 if (nswap == 0) { 569 blist_destroy(swapblist); 570 swapblist = NULL; 571 vrele(swapdev_vp); 572 swapdev_vp = NULL; 573 } else { 574 blist_resize(&swapblist, nswap, 0); 575 } 576 577 mtx_unlock(&swap_mtx); 578 return (0); 579 } 580 581 /* 582 * Account for swap space in individual swdevt's. The caller ensures 583 * that the provided range falls into a single swdevt. 584 * 585 * +count space freed 586 * -count space allocated 587 */ 588 void 589 swapacctspace(swblk_t base, swblk_t count) 590 { 591 int index; 592 int seg; 593 594 vm_swap_size += count; 595 seg = base / dmmax; 596 index = seg % nswdev; 597 swdevt[index].sw_nused -= count; 598 } 599 600 /* 601 * Retrieve swap info 602 */ 603 static int 604 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) 605 { 606 struct xswdev xs; 607 struct swdevt *sp; 608 int error; 609 int n; 610 611 error = 0; 612 for (n = 0; n < nswdev; ++n) { 613 sp = &swdevt[n]; 614 615 xs.xsw_size = sizeof(xs); 616 xs.xsw_version = XSWDEV_VERSION; 617 xs.xsw_blksize = PAGE_SIZE; 618 xs.xsw_dev = sp->sw_dev; 619 xs.xsw_flags = sp->sw_flags; 620 xs.xsw_nblks = sp->sw_nblks; 621 xs.xsw_used = sp->sw_nused; 622 623 error = SYSCTL_OUT(req, &xs, sizeof(xs)); 624 if (error) 625 break; 626 } 627 return (error); 628 } 629 630 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswdev, 0, 631 "Number of swap devices"); 632 SYSCTL_NODE(_vm, OID_AUTO, swap_info_array, CTLFLAG_RD, sysctl_vm_swap_info, 633 "Swap statistics by device"); 634