1 /* $NetBSD: uvm_vnode.c,v 1.63 2004/03/24 07:55:01 junyoung Exp $ */ 2 3 /* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993 6 * The Regents of the University of California. 7 * Copyright (c) 1990 University of Utah. 8 * 9 * All rights reserved. 10 * 11 * This code is derived from software contributed to Berkeley by 12 * the Systems Programming Group of the University of Utah Computer 13 * Science Department. 14 * 15 * Redistribution and use in source and binary forms, with or without 16 * modification, are permitted provided that the following conditions 17 * are met: 18 * 1. Redistributions of source code must retain the above copyright 19 * notice, this list of conditions and the following disclaimer. 20 * 2. Redistributions in binary form must reproduce the above copyright 21 * notice, this list of conditions and the following disclaimer in the 22 * documentation and/or other materials provided with the distribution. 23 * 3. All advertising materials mentioning features or use of this software 24 * must display the following acknowledgement: 25 * This product includes software developed by Charles D. Cranor, 26 * Washington University, the University of California, Berkeley and 27 * its contributors. 28 * 4. Neither the name of the University nor the names of its contributors 29 * may be used to endorse or promote products derived from this software 30 * without specific prior written permission. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 33 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 34 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 35 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 36 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 37 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 38 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 39 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 40 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 41 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 42 * SUCH DAMAGE. 43 * 44 * @(#)vnode_pager.c 8.8 (Berkeley) 2/13/94 45 * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp 46 */ 47 48 /* 49 * uvm_vnode.c: the vnode pager. 50 */ 51 52 #include <sys/cdefs.h> 53 __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.63 2004/03/24 07:55:01 junyoung Exp $"); 54 55 #include "fs_nfs.h" 56 #include "opt_uvmhist.h" 57 #include "opt_ddb.h" 58 59 #include <sys/param.h> 60 #include <sys/systm.h> 61 #include <sys/kernel.h> 62 #include <sys/proc.h> 63 #include <sys/malloc.h> 64 #include <sys/vnode.h> 65 #include <sys/disklabel.h> 66 #include <sys/ioctl.h> 67 #include <sys/fcntl.h> 68 #include <sys/conf.h> 69 #include <sys/pool.h> 70 #include <sys/mount.h> 71 72 #include <miscfs/specfs/specdev.h> 73 74 #include <uvm/uvm.h> 75 76 /* 77 * functions 78 */ 79 80 void uvn_detach(struct uvm_object *); 81 int uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *, int, 82 vm_prot_t, int, int); 83 int uvn_put(struct uvm_object *, voff_t, voff_t, int); 84 void uvn_reference(struct uvm_object *); 85 86 int uvn_findpage(struct uvm_object *, voff_t, struct vm_page **, int); 87 88 /* 89 * master pager structure 90 */ 91 92 struct uvm_pagerops uvm_vnodeops = { 93 NULL, 94 uvn_reference, 95 uvn_detach, 96 NULL, 97 uvn_get, 98 uvn_put, 99 }; 100 101 /* 102 * the ops! 103 */ 104 105 /* 106 * uvn_attach 107 * 108 * attach a vnode structure to a VM object. if the vnode is already 109 * attached, then just bump the reference count by one and return the 110 * VM object. if not already attached, attach and return the new VM obj. 111 * the "accessprot" tells the max access the attaching thread wants to 112 * our pages. 113 * 114 * => caller must _not_ already be holding the lock on the uvm_object. 115 * => in fact, nothing should be locked so that we can sleep here. 116 * => note that uvm_object is first thing in vnode structure, so their 117 * pointers are equiv. 118 */ 119 120 struct uvm_object * 121 uvn_attach(arg, accessprot) 122 void *arg; 123 vm_prot_t accessprot; 124 { 125 struct vnode *vp = arg; 126 struct uvm_object *uobj = &vp->v_uobj; 127 struct vattr vattr; 128 const struct bdevsw *bdev; 129 int result; 130 struct partinfo pi; 131 voff_t used_vnode_size; 132 UVMHIST_FUNC("uvn_attach"); UVMHIST_CALLED(maphist); 133 134 UVMHIST_LOG(maphist, "(vn=0x%x)", arg,0,0,0); 135 used_vnode_size = (voff_t)0; 136 137 /* 138 * first get a lock on the uobj. 139 */ 140 141 simple_lock(&uobj->vmobjlock); 142 while (vp->v_flag & VXLOCK) { 143 vp->v_flag |= VXWANT; 144 UVMHIST_LOG(maphist, " SLEEPING on blocked vn",0,0,0,0); 145 UVM_UNLOCK_AND_WAIT(uobj, &uobj->vmobjlock, FALSE, 146 "uvn_attach", 0); 147 simple_lock(&uobj->vmobjlock); 148 UVMHIST_LOG(maphist," WOKE UP",0,0,0,0); 149 } 150 151 /* 152 * if we're mapping a BLK device, make sure it is a disk. 153 */ 154 if (vp->v_type == VBLK) { 155 bdev = bdevsw_lookup(vp->v_rdev); 156 if (bdev == NULL || bdev->d_type != D_DISK) { 157 simple_unlock(&uobj->vmobjlock); 158 UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)", 159 0,0,0,0); 160 return(NULL); 161 } 162 } 163 KASSERT(vp->v_type == VREG || vp->v_type == VBLK); 164 165 /* 166 * set up our idea of the size 167 * if this hasn't been done already. 168 */ 169 if (vp->v_size == VSIZENOTSET) { 170 171 172 vp->v_flag |= VXLOCK; 173 simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */ 174 /* XXX: curproc? */ 175 if (vp->v_type == VBLK) { 176 /* 177 * We could implement this as a specfs getattr call, but: 178 * 179 * (1) VOP_GETATTR() would get the file system 180 * vnode operation, not the specfs operation. 181 * 182 * (2) All we want is the size, anyhow. 183 */ 184 bdev = bdevsw_lookup(vp->v_rdev); 185 if (bdev != NULL) { 186 result = (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART, 187 (caddr_t)&pi, FREAD, curproc); 188 } else { 189 result = ENXIO; 190 } 191 if (result == 0) { 192 /* XXX should remember blocksize */ 193 used_vnode_size = (voff_t)pi.disklab->d_secsize * 194 (voff_t)pi.part->p_size; 195 } 196 } else { 197 result = VOP_GETATTR(vp, &vattr, curproc->p_ucred, curproc); 198 if (result == 0) 199 used_vnode_size = vattr.va_size; 200 } 201 202 /* relock object */ 203 simple_lock(&uobj->vmobjlock); 204 205 if (vp->v_flag & VXWANT) { 206 wakeup(vp); 207 } 208 vp->v_flag &= ~(VXLOCK|VXWANT); 209 210 if (result != 0) { 211 simple_unlock(&uobj->vmobjlock); 212 UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0); 213 return(NULL); 214 } 215 vp->v_size = used_vnode_size; 216 217 } 218 219 simple_unlock(&uobj->vmobjlock); 220 UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount, 221 0, 0, 0); 222 return uobj; 223 } 224 225 226 /* 227 * uvn_reference 228 * 229 * duplicate a reference to a VM object. Note that the reference 230 * count must already be at least one (the passed in reference) so 231 * there is no chance of the uvn being killed or locked out here. 232 * 233 * => caller must call with object unlocked. 234 * => caller must be using the same accessprot as was used at attach time 235 */ 236 237 void 238 uvn_reference(uobj) 239 struct uvm_object *uobj; 240 { 241 VREF((struct vnode *)uobj); 242 } 243 244 245 /* 246 * uvn_detach 247 * 248 * remove a reference to a VM object. 249 * 250 * => caller must call with object unlocked and map locked. 251 */ 252 253 void 254 uvn_detach(uobj) 255 struct uvm_object *uobj; 256 { 257 vrele((struct vnode *)uobj); 258 } 259 260 /* 261 * uvn_put: flush page data to backing store. 262 * 263 * => object must be locked on entry! VOP_PUTPAGES must unlock it. 264 * => flags: PGO_SYNCIO -- use sync. I/O 265 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed) 266 */ 267 268 int 269 uvn_put(uobj, offlo, offhi, flags) 270 struct uvm_object *uobj; 271 voff_t offlo; 272 voff_t offhi; 273 int flags; 274 { 275 struct vnode *vp = (struct vnode *)uobj; 276 int error; 277 278 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 279 error = VOP_PUTPAGES(vp, offlo, offhi, flags); 280 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock)); 281 return error; 282 } 283 284 285 /* 286 * uvn_get: get pages (synchronously) from backing store 287 * 288 * => prefer map unlocked (not required) 289 * => object must be locked! we will _unlock_ it before starting any I/O. 290 * => flags: PGO_ALLPAGES: get all of the pages 291 * PGO_LOCKED: fault data structures are locked 292 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx] 293 * => NOTE: caller must check for released pages!! 294 */ 295 296 int 297 uvn_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags) 298 struct uvm_object *uobj; 299 voff_t offset; 300 struct vm_page **pps; /* IN/OUT */ 301 int *npagesp; /* IN (OUT if PGO_LOCKED) */ 302 int centeridx; 303 vm_prot_t access_type; 304 int advice, flags; 305 { 306 struct vnode *vp = (struct vnode *)uobj; 307 int error; 308 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist); 309 310 UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0); 311 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx, 312 access_type, advice, flags); 313 return error; 314 } 315 316 317 /* 318 * uvn_findpages: 319 * return the page for the uobj and offset requested, allocating if needed. 320 * => uobj must be locked. 321 * => returned pages will be BUSY. 322 */ 323 324 int 325 uvn_findpages(uobj, offset, npagesp, pgs, flags) 326 struct uvm_object *uobj; 327 voff_t offset; 328 int *npagesp; 329 struct vm_page **pgs; 330 int flags; 331 { 332 int i, count, found, npages, rv; 333 334 count = found = 0; 335 npages = *npagesp; 336 if (flags & UFP_BACKWARD) { 337 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) { 338 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 339 if (rv == 0) { 340 if (flags & UFP_DIRTYONLY) 341 break; 342 } else 343 found++; 344 count++; 345 } 346 } else { 347 for (i = 0; i < npages; i++, offset += PAGE_SIZE) { 348 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 349 if (rv == 0) { 350 if (flags & UFP_DIRTYONLY) 351 break; 352 } else 353 found++; 354 count++; 355 } 356 } 357 *npagesp = count; 358 return (found); 359 } 360 361 int 362 uvn_findpage(uobj, offset, pgp, flags) 363 struct uvm_object *uobj; 364 voff_t offset; 365 struct vm_page **pgp; 366 int flags; 367 { 368 struct vm_page *pg; 369 boolean_t dirty; 370 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist); 371 UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0); 372 373 if (*pgp != NULL) { 374 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0); 375 return 0; 376 } 377 for (;;) { 378 /* look for an existing page */ 379 pg = uvm_pagelookup(uobj, offset); 380 381 /* nope? allocate one now */ 382 if (pg == NULL) { 383 if (flags & UFP_NOALLOC) { 384 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0); 385 return 0; 386 } 387 pg = uvm_pagealloc(uobj, offset, NULL, 0); 388 if (pg == NULL) { 389 if (flags & UFP_NOWAIT) { 390 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 391 return 0; 392 } 393 simple_unlock(&uobj->vmobjlock); 394 uvm_wait("uvn_fp1"); 395 simple_lock(&uobj->vmobjlock); 396 continue; 397 } 398 UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0); 399 break; 400 } else if (flags & UFP_NOCACHE) { 401 UVMHIST_LOG(ubchist, "nocache",0,0,0,0); 402 return 0; 403 } 404 405 /* page is there, see if we need to wait on it */ 406 if ((pg->flags & PG_BUSY) != 0) { 407 if (flags & UFP_NOWAIT) { 408 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 409 return 0; 410 } 411 pg->flags |= PG_WANTED; 412 UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0); 413 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, 414 "uvn_fp2", 0); 415 simple_lock(&uobj->vmobjlock); 416 continue; 417 } 418 419 /* skip PG_RDONLY pages if requested */ 420 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) { 421 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0); 422 return 0; 423 } 424 425 /* stop on clean pages if requested */ 426 if (flags & UFP_DIRTYONLY) { 427 dirty = pmap_clear_modify(pg) || 428 (pg->flags & PG_CLEAN) == 0; 429 pg->flags |= PG_CLEAN; 430 if (!dirty) { 431 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0); 432 return 0; 433 } 434 } 435 436 /* mark the page BUSY and we're done. */ 437 pg->flags |= PG_BUSY; 438 UVM_PAGE_OWN(pg, "uvn_findpage"); 439 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0); 440 break; 441 } 442 *pgp = pg; 443 return 1; 444 } 445 446 /* 447 * uvm_vnp_setsize: grow or shrink a vnode uobj 448 * 449 * grow => just update size value 450 * shrink => toss un-needed pages 451 * 452 * => we assume that the caller has a reference of some sort to the 453 * vnode in question so that it will not be yanked out from under 454 * us. 455 */ 456 457 void 458 uvm_vnp_setsize(vp, newsize) 459 struct vnode *vp; 460 voff_t newsize; 461 { 462 struct uvm_object *uobj = &vp->v_uobj; 463 voff_t pgend = round_page(newsize); 464 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist); 465 466 simple_lock(&uobj->vmobjlock); 467 UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x", 468 vp, vp->v_size, newsize, 0); 469 470 /* 471 * now check if the size has changed: if we shrink we had better 472 * toss some pages... 473 */ 474 475 if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) { 476 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO); 477 } else { 478 simple_unlock(&uobj->vmobjlock); 479 } 480 vp->v_size = newsize; 481 } 482 483 /* 484 * uvm_vnp_zerorange: set a range of bytes in a file to zero. 485 */ 486 487 void 488 uvm_vnp_zerorange(vp, off, len) 489 struct vnode *vp; 490 off_t off; 491 size_t len; 492 { 493 void *win; 494 495 /* 496 * XXXUBC invent kzero() and use it 497 */ 498 499 while (len) { 500 vsize_t bytelen = len; 501 502 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UBC_WRITE); 503 memset(win, 0, bytelen); 504 ubc_release(win, 0); 505 506 off += bytelen; 507 len -= bytelen; 508 } 509 } 510