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 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 struct vnode *vp; 104 105 vp = ap->a_vp; 106 sz = howmany(bp->b_bcount, PAGE_SIZE); 107 blkno = (int)(bio->bio_offset >> PAGE_SHIFT); 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 nbio = push_bio(bio); 115 if (nswdev > 1) { 116 off = blkno % dmmax; 117 if (off + sz > dmmax) { 118 bp->b_error = EINVAL; 119 bp->b_flags |= B_ERROR; 120 biodone(bio); 121 return 0; 122 } 123 seg = blkno / dmmax; 124 index = seg % nswdev; 125 seg /= nswdev; 126 nbio->bio_offset = (off_t)(seg * dmmax + off) << PAGE_SHIFT; 127 } else { 128 index = 0; 129 nbio->bio_offset = bio->bio_offset; 130 } 131 nblkno = (int)(nbio->bio_offset >> PAGE_SHIFT); 132 sp = &swdevt[index]; 133 if (nblkno + sz > sp->sw_nblks) { 134 bp->b_error = EINVAL; 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 if (sp->sw_vp == NULL) { 141 bp->b_error = ENODEV; 142 bp->b_flags |= B_ERROR; 143 /* I/O was never started on nbio, must biodone(bio) */ 144 biodone(bio); 145 return 0; 146 } 147 148 /* 149 * Issue a strategy call on the appropriate swap vnode. Note that 150 * bp->b_vp is not modified. Strategy code is always supposed to 151 * use the passed vp. 152 * 153 * We have to use vn_strategy() here even if we know we have a 154 * device in order to properly break up requests which exceed the 155 * device's DMA limits. 156 */ 157 vn_strategy(sp->sw_vp, nbio); 158 return 0; 159 } 160 161 static int 162 swapdev_inactive(struct vop_inactive_args *ap) 163 { 164 vrecycle(ap->a_vp); 165 return(0); 166 } 167 168 static int 169 swapdev_reclaim(struct vop_reclaim_args *ap) 170 { 171 return(0); 172 } 173 174 /* 175 * Create a special vnode op vector for swapdev_vp - we only use 176 * vn_strategy(), everything else returns an error. 177 */ 178 static struct vop_ops swapdev_vnode_vops = { 179 .vop_default = vop_defaultop, 180 .vop_strategy = swapdev_strategy, 181 .vop_inactive = swapdev_inactive, 182 .vop_reclaim = swapdev_reclaim 183 }; 184 static struct vop_ops *swapdev_vnode_vops_p = &swapdev_vnode_vops; 185 186 VNODEOP_SET(swapdev_vnode_vops); 187 188 /* 189 * swapon_args(char *name) 190 * 191 * System call swapon(name) enables swapping on device name, 192 * which must be in the swdevsw. Return EBUSY 193 * if already swapping on this device. 194 * 195 * No requirements. 196 */ 197 int 198 sys_swapon(struct swapon_args *uap) 199 { 200 struct thread *td = curthread; 201 struct vattr attr; 202 struct vnode *vp; 203 struct nlookupdata nd; 204 int error; 205 struct ucred *cred; 206 207 cred = td->td_ucred; 208 209 error = priv_check(td, PRIV_ROOT); 210 if (error) 211 return (error); 212 213 mtx_lock(&swap_mtx); 214 get_mplock(); 215 vp = NULL; 216 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 217 if (error == 0) 218 error = nlookup(&nd); 219 if (error == 0) 220 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 221 nlookup_done(&nd); 222 if (error) { 223 rel_mplock(); 224 mtx_unlock(&swap_mtx); 225 return (error); 226 } 227 228 if (vn_isdisk(vp, &error)) { 229 error = swaponvp(td, vp, 0); 230 } else if (vp->v_type == VREG && vp->v_tag == VT_NFS && 231 (error = VOP_GETATTR(vp, &attr)) == 0) { 232 /* 233 * Allow direct swapping to NFS regular files in the same 234 * way that nfs_mountroot() sets up diskless swapping. 235 */ 236 error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); 237 } 238 if (error) 239 vrele(vp); 240 rel_mplock(); 241 mtx_unlock(&swap_mtx); 242 243 return (error); 244 } 245 246 /* 247 * Swfree(index) frees the index'th portion of the swap map. 248 * Each of the nswdev devices provides 1/nswdev'th of the swap 249 * space, which is laid out with blocks of dmmax pages circularly 250 * among the devices. 251 * 252 * The new swap code uses page-sized blocks. The old swap code used 253 * DEV_BSIZE'd chunks. 254 * 255 * XXX locking when multiple swapon's run in parallel 256 */ 257 int 258 swaponvp(struct thread *td, struct vnode *vp, u_quad_t nblks) 259 { 260 swblk_t aligned_nblks; 261 int64_t dpsize; 262 struct ucred *cred; 263 struct swdevt *sp; 264 swblk_t vsbase; 265 swblk_t dvbase; 266 cdev_t dev; 267 int index; 268 int error; 269 swblk_t blk; 270 271 cred = td->td_ucred; 272 273 mtx_lock(&swap_mtx); 274 275 if (!swapdev_vp) { 276 error = getspecialvnode(VT_NON, NULL, &swapdev_vnode_vops_p, 277 &swapdev_vp, 0, 0); 278 if (error) 279 panic("Cannot get vnode for swapdev"); 280 swapdev_vp->v_type = VNON; /* Untyped */ 281 vx_unlock(swapdev_vp); 282 } 283 284 for (sp = swdevt, index = 0 ; index < nswdev; index++, sp++) { 285 if (sp->sw_vp == vp) { 286 mtx_unlock(&swap_mtx); 287 return EBUSY; 288 } 289 if (!sp->sw_vp) 290 goto found; 291 292 } 293 mtx_unlock(&swap_mtx); 294 return EINVAL; 295 found: 296 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 297 error = VOP_OPEN(vp, FREAD | FWRITE, cred, NULL); 298 vn_unlock(vp); 299 if (error) { 300 mtx_unlock(&swap_mtx); 301 return (error); 302 } 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 mtx_unlock(&swap_mtx); 318 return (ENXIO); 319 } 320 nblks = (u_quad_t)dpsize; 321 } 322 if (nblks == 0) { 323 VOP_CLOSE(vp, FREAD | FWRITE); 324 mtx_unlock(&swap_mtx); 325 return (ENXIO); 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 mtx_unlock(&swap_mtx); 351 return (ENXIO); 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(); 384 385 mtx_unlock(&swap_mtx); 386 return (0); 387 } 388 389 /* 390 * swapoff_args(char *name) 391 * 392 * System call swapoff(name) disables swapping on device name, 393 * which must be an active swap device. Return ENOMEM 394 * if there is not enough memory to page in the contents of 395 * the given device. 396 * 397 * No requirements. 398 */ 399 int 400 sys_swapoff(struct swapoff_args *uap) 401 { 402 struct vnode *vp; 403 struct nlookupdata nd; 404 struct swdevt *sp; 405 int error, index; 406 407 error = priv_check(curthread, PRIV_ROOT); 408 if (error) 409 return (error); 410 411 mtx_lock(&swap_mtx); 412 get_mplock(); 413 vp = NULL; 414 error = nlookup_init(&nd, uap->name, UIO_USERSPACE, NLC_FOLLOW); 415 if (error == 0) 416 error = nlookup(&nd); 417 if (error == 0) 418 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 419 nlookup_done(&nd); 420 if (error) 421 goto done; 422 423 for (sp = swdevt, index = 0; index < nswdev; index++, sp++) { 424 if (sp->sw_vp == vp) 425 goto found; 426 } 427 error = EINVAL; 428 goto done; 429 found: 430 error = swapoff_one(index); 431 432 done: 433 rel_mplock(); 434 mtx_unlock(&swap_mtx); 435 return (error); 436 } 437 438 static int 439 swapoff_one(int index) 440 { 441 swblk_t blk, aligned_nblks; 442 swblk_t dvbase, vsbase; 443 u_int pq_active_clean, pq_inactive_clean; 444 struct swdevt *sp; 445 vm_page_t m; 446 447 mtx_lock(&swap_mtx); 448 449 sp = &swdevt[index]; 450 aligned_nblks = sp->sw_nblks; 451 pq_active_clean = pq_inactive_clean = 0; 452 453 /* 454 * We can turn off this swap device safely only if the 455 * available virtual memory in the system will fit the amount 456 * of data we will have to page back in, plus an epsilon so 457 * the system doesn't become critically low on swap space. 458 */ 459 lwkt_gettoken(&vm_token); 460 TAILQ_FOREACH(m, &vm_page_queues[PQ_ACTIVE].pl, pageq) { 461 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 462 continue; 463 464 if (m->dirty == 0) { 465 vm_page_test_dirty(m); 466 if (m->dirty == 0) 467 ++pq_active_clean; 468 } 469 } 470 TAILQ_FOREACH(m, &vm_page_queues[PQ_INACTIVE].pl, pageq) { 471 if (m->flags & (PG_MARKER | PG_FICTITIOUS)) 472 continue; 473 474 if (m->dirty == 0) { 475 vm_page_test_dirty(m); 476 if (m->dirty == 0) 477 ++pq_inactive_clean; 478 } 479 } 480 lwkt_reltoken(&vm_token); 481 482 if (vmstats.v_free_count + vmstats.v_cache_count + pq_active_clean + 483 pq_inactive_clean + vm_swap_size < aligned_nblks + nswap_lowat) { 484 mtx_unlock(&swap_mtx); 485 return (ENOMEM); 486 } 487 488 /* 489 * Prevent further allocations on this device 490 */ 491 sp->sw_flags |= SW_CLOSING; 492 for (dvbase = dmmax; dvbase < aligned_nblks; dvbase += dmmax) { 493 blk = min(aligned_nblks - dvbase, dmmax); 494 vsbase = index * dmmax + dvbase * nswdev; 495 vm_swap_size -= blist_fill(swapblist, vsbase, blk); 496 vm_swap_max -= blk; 497 } 498 499 /* 500 * Page in the contents of the device and close it. 501 */ 502 if (swap_pager_swapoff(index)) { 503 mtx_unlock(&swap_mtx); 504 return (EINTR); 505 } 506 507 VOP_CLOSE(sp->sw_vp, FREAD | FWRITE); 508 vrele(sp->sw_vp); 509 bzero(swdevt + index, sizeof(struct swdevt)); 510 511 /* 512 * Resize the bitmap based on the nem largest swap device, 513 * or free the bitmap if there are no more devices. 514 */ 515 for (sp = swdevt, aligned_nblks = 0; sp < swdevt + nswdev; sp++) { 516 if (sp->sw_vp) 517 aligned_nblks = max(aligned_nblks, sp->sw_nblks); 518 } 519 520 nswap = aligned_nblks * nswdev; 521 522 if (nswap == 0) { 523 blist_destroy(swapblist); 524 swapblist = NULL; 525 vrele(swapdev_vp); 526 swapdev_vp = NULL; 527 } else { 528 blist_resize(&swapblist, nswap, 0); 529 } 530 531 mtx_unlock(&swap_mtx); 532 return (0); 533 } 534 535 /* 536 * Account for swap space in individual swdevt's. The caller ensures 537 * that the provided range falls into a single swdevt. 538 * 539 * +count space freed 540 * -count space allocated 541 */ 542 void 543 swapacctspace(swblk_t base, swblk_t count) 544 { 545 int index; 546 int seg; 547 548 vm_swap_size += count; 549 seg = base / dmmax; 550 index = seg % nswdev; 551 swdevt[index].sw_nused -= count; 552 } 553 554 /* 555 * Retrieve swap info 556 */ 557 static int 558 sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) 559 { 560 struct xswdev xs; 561 struct swdevt *sp; 562 int error; 563 int n; 564 565 error = 0; 566 for (n = 0; n < nswdev; ++n) { 567 sp = &swdevt[n]; 568 569 xs.xsw_size = sizeof(xs); 570 xs.xsw_version = XSWDEV_VERSION; 571 xs.xsw_blksize = PAGE_SIZE; 572 xs.xsw_dev = sp->sw_dev; 573 xs.xsw_flags = sp->sw_flags; 574 xs.xsw_nblks = sp->sw_nblks; 575 xs.xsw_used = sp->sw_nused; 576 577 error = SYSCTL_OUT(req, &xs, sizeof(xs)); 578 if (error) 579 break; 580 } 581 return (error); 582 } 583 584 SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswdev, 0, 585 "Number of swap devices"); 586 SYSCTL_NODE(_vm, OID_AUTO, swap_info_array, CTLFLAG_RD, sysctl_vm_swap_info, 587 "Swap statistics by device"); 588