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. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94 32 * $FreeBSD: src/sys/vm/vm_swap.c,v 1.96.2.2 2001/10/14 18:46:47 iedowse Exp $ 33 */ 34 35 #include "opt_swap.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/sysproto.h> 40 #include <sys/buf.h> 41 #include <sys/proc.h> 42 #include <sys/priv.h> 43 #include <sys/nlookup.h> 44 #include <sys/sysctl.h> 45 #include <sys/dmap.h> /* XXX */ 46 #include <sys/vnode.h> 47 #include <sys/fcntl.h> 48 #include <sys/blist.h> 49 #include <sys/kernel.h> 50 #include <sys/lock.h> 51 #include <sys/conf.h> 52 #include <sys/stat.h> 53 54 #include <vm/vm.h> 55 #include <vm/vm_extern.h> 56 #include <vm/swap_pager.h> 57 #include <vm/vm_zone.h> 58 #include <vm/vm_param.h> 59 60 #include <sys/thread2.h> 61 #include <sys/mplock2.h> 62 #include <sys/mutex2.h> 63 #include <sys/spinlock2.h> 64 65 /* 66 * Indirect driver for multi-controller paging. 67 */ 68 69 #ifndef NSWAPDEV 70 #define NSWAPDEV 4 71 #endif 72 static struct swdevt should_be_malloced[NSWAPDEV]; 73 struct swdevt *swdevt = should_be_malloced; /* exported to pstat/systat */ 74 static swblk_t nswap; /* first block after the interleaved devs */ 75 static struct mtx swap_mtx = MTX_INITIALIZER("swpmtx"); 76 int nswdev = NSWAPDEV; /* exported to pstat/systat */ 77 int vm_swap_size; 78 int vm_swap_max; 79 80 static int swapoff_one(int index); 81 struct vnode *swapdev_vp; 82 83 /* 84 * (struct vnode *a_vp, struct bio *b_bio) 85 * 86 * vn_strategy() for swapdev_vp. Perform swap strategy interleave device 87 * selection. 88 * 89 * No requirements. 90 */ 91 static int 92 swapdev_strategy(struct vop_strategy_args *ap) 93 { 94 struct bio *bio = ap->a_bio; 95 struct bio *nbio; 96 struct buf *bp = bio->bio_buf; 97 int sz, off, seg, index, blkno, nblkno; 98 struct swdevt *sp; 99 sz = howmany(bp->b_bcount, PAGE_SIZE); 100 blkno = (int)(bio->bio_offset >> PAGE_SHIFT); 101 102 /* 103 * Convert interleaved swap into per-device swap. Note that 104 * the block size is left in PAGE_SIZE'd chunks (for the newswap) 105 * here. 106 */ 107 nbio = push_bio(bio); 108 if (nswdev > 1) { 109 off = blkno % dmmax; 110 if (off + sz > dmmax) { 111 bp->b_error = EINVAL; 112 bp->b_flags |= B_ERROR; 113 biodone(bio); 114 return 0; 115 } 116 seg = blkno / dmmax; 117 index = seg % nswdev; 118 seg /= nswdev; 119 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT; 120 } else { 121 index = 0; 122 nbio->bio_offset = bio->bio_offset; 123 } 124 nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT); 125 sp = &swdevt[index]; 126 if (nblkno + sz > sp->sw_nblks) { 127 bp->b_error = EINVAL; 128 bp->b_flags |= B_ERROR; 129 /* I/O was never started on nbio, must biodone(bio) */ 130 biodone(bio); 131 return 0; 132 } 133 if (sp->sw_vp == NULL) { 134 bp->b_error = ENODEV; 135 bp->b_flags |= B_ERROR; 136 /* I/O was never started on nbio, must biodone(bio) */ 137 biodone(bio); 138 return 0; 139 } 140 141 /* 142 * Issue a strategy call on the appropriate swap vnode. Note that 143 * bp->b_vp is not modified. Strategy code is always supposed to 144 * use the passed vp. 145 * 146 * We have to use vn_strategy() here even if we know we have a 147 * device in order to properly break up requests which exceed the 148 * device's DMA limits. 149 */ 150 vn_strategy(sp->sw_vp, nbio); 151 return 0; 152 } 153 154 static int 155 swapdev_inactive(struct vop_inactive_args *ap) 156 { 157 vrecycle(ap->a_vp); 158 return(0); 159 } 160 161 static int 162 swapdev_reclaim(struct vop_reclaim_args *ap) 163 { 164 return(0); 165 } 166 167 /* 168 * Create a special vnode op vector for swapdev_vp - we only use 169 * vn_strategy(), everything else returns an error. 170 */ 171 static struct vop_ops swapdev_vnode_vops = { 172 .vop_default = vop_defaultop, 173 .vop_strategy = swapdev_strategy, 174 .vop_inactive = swapdev_inactive, 175 .vop_reclaim = swapdev_reclaim 176 }; 177 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops; 178 179 VNODEOP_SET(swapdev_vnode_vops); 180 181 /* 182 * swapon_args(char *name) 183 * 184 * System call swapon(name) enables swapping on device name, 185 * which must be in the swdevsw. Return EBUSY 186 * if already swapping on this device. 187 * 188 * No requirements. 189 */ 190 int 191 sys_swapon(struct swapon_args *uap) 192 { 193 struct thread *td = curthread; 194 struct vattr attr; 195 struct vnode *vp; 196 struct nlookupdata nd; 197 int error; 198 199 error = priv_check(td, PRIV_ROOT); 200 if (error) 201 return (error); 202 203 mtx_lock(&swap_mtx); 204 get_mplock(); 205 vp = NULL; 206 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 207 if (error == 0) 208 error = nlookup(&nd); 209 if (error == 0) 210 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 211 nlookup_done(&nd); 212 if (error) { 213 rel_mplock(); 214 mtx_unlock(&swap_mtx); 215 return (error); 216 } 217 218 if (vn_isdisk(vp, &error)) { 219 error = swaponvp(td, vp, 0); 220 } else if (vp->v_type == VREG && vp->v_tag == VT_NFS && 221 (error = VOP_GETATTR(vp, &attr)) == 0) { 222 /* 223 * Allow direct swapping to NFS regular files in the same 224 * way that nfs_mountroot() sets up diskless swapping. 225 */ 226 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); 227 } 228 if (error) 229 vrele(vp); 230 rel_mplock(); 231 mtx_unlock(&swap_mtx); 232 233 return (error); 234 } 235 236 /* 237 * Swfree(index) frees the index'th portion of the swap map. 238 * Each of the nswdev devices provides 1/nswdev'th of the swap 239 * space, which is laid out with blocks of dmmax pages circularly 240 * among the devices. 241 * 242 * The new swap code uses page-sized blocks. The old swap code used 243 * DEV_BSIZE'd chunks. 244 * 245 * XXX locking when multiple swapon's run in parallel 246 */ 247 int 248 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks) 249 { 250 swblk_t aligned_nblks; 251 int64_t dpsize; 252 struct ucred *cred; 253 struct swdevt *sp; 254 swblk_t vsbase; 255 swblk_t dvbase; 256 cdev_t dev; 257 int index; 258 int error; 259 swblk_t blk; 260 261 cred = td->td_ucred; 262 263 lwkt_gettoken(&vm_token); /* needed for vm_swap_size and blist */ 264 mtx_lock(&swap_mtx); 265 266 if (!swapdev_vp) { 267 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p, 268 &swapdev_vp, 0, 0); 269 if (error) 270 panic("Cannot get vnode for swapdev"); 271 swapdev_vp->v_type = VNON; /* Untyped */ 272 vx_unlock(swapdev_vp); 273 } 274 275 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) { 276 if (sp->sw_vp == vp) { 277 error = EBUSY; 278 goto done; 279 } 280 if (!sp->sw_vp) 281 goto found; 282 283 } 284 error = EINVAL; 285 goto done; 286 found: 287 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 288 error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL); 289 vn_unlock(vp); 290 if (error) 291 goto done; 292 293 /* 294 * v_rdev is not valid until after the VOP_OPEN() call. dev_psize() 295 * must be supported if a character device has been specified. 296 */ 297 if (vp->v_type == VCHR) 298 dev = vp->v_rdev; 299 else 300 dev = NULL; 301 302 if (nblks == 0 && dev != NULL) { 303 dpsize = dev_dpsize(dev); 304 if (dpsize == -1) { 305 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 306 VOP_CLOSE(vp, FREAD | FWRITE, NULL); 307 vn_unlock(vp); 308 error = ENXIO; 309 goto done; 310 } 311 nblks = (u_quad_t)dpsize; 312 } 313 if (nblks == 0) { 314 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 315 VOP_CLOSE(vp, FREAD | FWRITE, NULL); 316 vn_unlock(vp); 317 error = ENXIO; 318 goto done; 319 } 320 321 /* 322 * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks. 323 * First chop nblks off to page-align it, then convert. 324 * 325 * sw->sw_nblks is in page-sized chunks now too. 326 */ 327 nblks &= ~(u_quad_t)(ctodb(1) - 1); 328 nblks = dbtoc(nblks); 329 330 /* 331 * Post-conversion nblks must not be >= BLIST_MAXBLKS, and 332 * we impose a 4-swap-device limit so we have to divide it out 333 * further. Going beyond this will result in overflows in the 334 * blist code. 335 * 336 * Post-conversion nblks must fit within a (swblk_t), which 337 * this test also ensures. 338 */ 339 if (nblks > BLIST_MAXBLKS / nswdev) { 340 kprintf("exceeded maximum of %d blocks per swap unit\n", 341 (int)BLIST_MAXBLKS / nswdev); 342 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 343 VOP_CLOSE(vp, FREAD | FWRITE, NULL); 344 vn_unlock(vp); 345 error = ENXIO; 346 goto done; 347 } 348 349 sp->sw_vp = vp; 350 sp->sw_dev = dev2udev(dev); 351 sp->sw_device = dev; 352 sp->sw_flags = SW_FREED; 353 sp->sw_nused = 0; 354 355 /* 356 * nblks, nswap, and dmmax are PAGE_SIZE'd parameters now, not 357 * DEV_BSIZE'd. aligned_nblks is used to calculate the 358 * size of the swap bitmap, taking into account the stripe size. 359 */ 360 aligned_nblks = (swblk_t)((nblks + (dmmax - 1)) & ~(u_long)(dmmax - 1)); 361 sp->sw_nblks = aligned_nblks; 362 363 if (aligned_nblks * nswdev > nswap) 364 nswap = aligned_nblks * nswdev; 365 366 if (swapblist == NULL) 367 swapblist = blist_create(nswap); 368 else 369 blist_resize(&swapblist, nswap, 0); 370 371 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) { 372 blk = min(aligned_nblks - dvbase, dmmax); 373 vsbase = index * dmmax + dvbase * nswdev; 374 blist_free(swapblist, vsbase, blk); 375 vm_swap_size += blk; 376 vm_swap_max += blk; 377 } 378 swap_pager_newswap(); 379 error = 0; 380 done: 381 mtx_unlock(&swap_mtx); 382 lwkt_reltoken(&vm_token); 383 return (error); 384 } 385 386 /* 387 * swapoff_args(char *name) 388 * 389 * System call swapoff(name) disables swapping on device name, 390 * which must be an active swap device. Return ENOMEM 391 * if there is not enough memory to page in the contents of 392 * the given device. 393 * 394 * No requirements. 395 */ 396 int 397 sys_swapoff(struct swapoff_args *uap) 398 { 399 struct vnode *vp; 400 struct nlookupdata nd; 401 struct swdevt *sp; 402 int error, index; 403 404 error = priv_check(curthread, PRIV_ROOT); 405 if (error) 406 return (error); 407 408 mtx_lock(&swap_mtx); 409 get_mplock(); 410 vp = NULL; 411 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 412 if (error == 0) 413 error = nlookup(&nd); 414 if (error == 0) 415 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 416 nlookup_done(&nd); 417 if (error) 418 goto done; 419 420 for (sp = swdevt, index = 0; index < nswdev; index++, sp++) { 421 if (sp->sw_vp == vp) 422 goto found; 423 } 424 error = EINVAL; 425 goto done; 426 found: 427 error = swapoff_one(index); 428 429 done: 430 rel_mplock(); 431 mtx_unlock(&swap_mtx); 432 return (error); 433 } 434 435 static int 436 swapoff_one(int index) 437 { 438 swblk_t blk, aligned_nblks; 439 swblk_t dvbase, vsbase; 440 u_int pq_active_clean, pq_inactive_clean; 441 struct swdevt *sp; 442 struct vm_page marker; 443 vm_page_t m; 444 int q; 445 446 mtx_lock(&swap_mtx); 447 448 sp = &swdevt[index]; 449 aligned_nblks = sp->sw_nblks; 450 pq_active_clean = pq_inactive_clean = 0; 451 452 /* 453 * We can turn off this swap device safely only if the 454 * available virtual memory in the system will fit the amount 455 * of data we will have to page back in, plus an epsilon so 456 * the system doesn't become critically low on swap space. 457 */ 458 for (q = 0; q < PQ_L2_SIZE; ++q) { 459 bzero(&marker, sizeof(marker)); 460 marker.flags = PG_BUSY | PG_FICTITIOUS | PG_MARKER; 461 marker.queue = PQ_ACTIVE + q; 462 marker.pc = q; 463 marker.wire_count = 1; 464 465 vm_page_queues_spin_lock(marker.queue); 466 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl, 467 &marker, pageq); 468 469 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) { 470 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, 471 &marker, pageq); 472 TAILQ_INSERT_AFTER(&vm_page_queues[marker.queue].pl, m, 473 &marker, pageq); 474 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 475 continue; 476 477 if (vm_page_busy_try(m, FALSE) == 0) { 478 vm_page_queues_spin_unlock(marker.queue); 479 if (m->dirty == 0) { 480 vm_page_test_dirty(m); 481 if (m->dirty == 0) 482 ++pq_active_clean; 483 } 484 vm_page_wakeup(m); 485 vm_page_queues_spin_lock(marker.queue); 486 } 487 } 488 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, &marker, pageq); 489 vm_page_queues_spin_unlock(marker.queue); 490 491 marker.queue = PQ_INACTIVE + q; 492 marker.pc = q; 493 vm_page_queues_spin_lock(marker.queue); 494 TAILQ_INSERT_HEAD(&vm_page_queues[marker.queue].pl, 495 &marker, pageq); 496 497 while ((m = TAILQ_NEXT(&marker, pageq)) != NULL) { 498 TAILQ_REMOVE( 499 &vm_page_queues[marker.queue].pl, 500 &marker, pageq); 501 TAILQ_INSERT_AFTER( 502 &vm_page_queues[marker.queue].pl, 503 m, &marker, pageq); 504 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 505 continue; 506 507 if (vm_page_busy_try(m, FALSE) == 0) { 508 vm_page_queues_spin_unlock(marker.queue); 509 if (m->dirty == 0) { 510 vm_page_test_dirty(m); 511 if (m->dirty == 0) 512 ++pq_inactive_clean; 513 } 514 vm_page_wakeup(m); 515 vm_page_queues_spin_lock(marker.queue); 516 } 517 } 518 TAILQ_REMOVE(&vm_page_queues[marker.queue].pl, 519 &marker, pageq); 520 vm_page_queues_spin_unlock(marker.queue); 521 } 522 523 if (vmstats.v_free_count + vmstats.v_cache_count + pq_active_clean + 524 pq_inactive_clean + vm_swap_size < aligned_nblks + nswap_lowat) { 525 mtx_unlock(&swap_mtx); 526 return (ENOMEM); 527 } 528 529 /* 530 * Prevent further allocations on this device 531 */ 532 sp->sw_flags |= SW_CLOSING; 533 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) { 534 blk = min(aligned_nblks - dvbase, dmmax); 535 vsbase = index * dmmax + dvbase * nswdev; 536 vm_swap_size -= blist_fill(swapblist, vsbase, blk); 537 vm_swap_max -= blk; 538 } 539 540 /* 541 * Page in the contents of the device and close it. 542 */ 543 if (swap_pager_swapoff(index) && swap_pager_swapoff(index)) { 544 mtx_unlock(&swap_mtx); 545 return (EINTR); 546 } 547 548 vn_lock(sp->sw_vp, LK_EXCLUSIVE | LK_RETRY); 549 VOP_CLOSE(sp->sw_vp, FREAD | FWRITE, NULL); 550 vn_unlock(sp->sw_vp); 551 vrele(sp->sw_vp); 552 bzero(swdevt + index, sizeof(struct swdevt)); 553 554 /* 555 * Resize the bitmap based on the nem largest swap device, 556 * or free the bitmap if there are no more devices. 557 */ 558 for (sp = swdevt, aligned_nblks = 0; sp < swdevt + nswdev; sp++) { 559 if (sp->sw_vp) 560 aligned_nblks = max(aligned_nblks, sp->sw_nblks); 561 } 562 563 nswap = aligned_nblks * nswdev; 564 565 if (nswap == 0) { 566 blist_destroy(swapblist); 567 swapblist = NULL; 568 vrele(swapdev_vp); 569 swapdev_vp = NULL; 570 } else { 571 blist_resize(&swapblist, nswap, 0); 572 } 573 574 mtx_unlock(&swap_mtx); 575 return (0); 576 } 577 578 /* 579 * Account for swap space in individual swdevt's. The caller ensures 580 * that the provided range falls into a single swdevt. 581 * 582 * +count space freed 583 * -count space allocated 584 */ 585 void 586 swapacctspace(swblk_t base, swblk_t count) 587 { 588 int index; 589 int seg; 590 591 vm_swap_size += count; 592 seg = base / dmmax; 593 index = seg % nswdev; 594 swdevt[index].sw_nused -= count; 595 } 596 597 /* 598 * Retrieve swap info 599 */ 600 static int 601 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) 602 { 603 struct xswdev xs; 604 struct swdevt *sp; 605 int error; 606 int n; 607 608 error = 0; 609 for (n = 0; n < nswdev; ++n) { 610 sp = &swdevt[n]; 611 612 xs.xsw_size = sizeof(xs); 613 xs.xsw_version = XSWDEV_VERSION; 614 xs.xsw_blksize = PAGE_SIZE; 615 xs.xsw_dev = sp->sw_dev; 616 xs.xsw_flags = sp->sw_flags; 617 xs.xsw_nblks = sp->sw_nblks; 618 xs.xsw_used = sp->sw_nused; 619 620 error = SYSCTL_OUT(req, &xs, sizeof(xs)); 621 if (error) 622 break; 623 } 624 return (error); 625 } 626 627 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswdev, 0, 628 "Number of swap devices"); 629 SYSCTL_NODE(_vm, OID_AUTO, swap_info_array, CTLFLAG_RD, sysctl_vm_swap_info, 630 "Swap statistics by device"); 631