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