1 /* $NetBSD: uvm_vnode.c,v 1.58 2002/05/17 22:00:50 enami 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 1.58 2002/05/17 22:00:50 enami 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 __P((struct uvm_object *)); 81 int uvn_get __P((struct uvm_object *, voff_t, struct vm_page **, int *, int, 82 vm_prot_t, int, int)); 83 int uvn_put __P((struct uvm_object *, voff_t, voff_t, int)); 84 void uvn_reference __P((struct uvm_object *)); 85 86 int uvn_findpage __P((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 int result; 129 struct partinfo pi; 130 voff_t used_vnode_size; 131 UVMHIST_FUNC("uvn_attach"); UVMHIST_CALLED(maphist); 132 133 UVMHIST_LOG(maphist, "(vn=0x%x)", arg,0,0,0); 134 used_vnode_size = (voff_t)0; 135 136 /* 137 * first get a lock on the uobj. 138 */ 139 140 simple_lock(&uobj->vmobjlock); 141 while (vp->v_flag & VXLOCK) { 142 vp->v_flag |= VXWANT; 143 UVMHIST_LOG(maphist, " SLEEPING on blocked vn",0,0,0,0); 144 UVM_UNLOCK_AND_WAIT(uobj, &uobj->vmobjlock, FALSE, 145 "uvn_attach", 0); 146 simple_lock(&uobj->vmobjlock); 147 UVMHIST_LOG(maphist," WOKE UP",0,0,0,0); 148 } 149 150 /* 151 * if we're mapping a BLK device, make sure it is a disk. 152 */ 153 if (vp->v_type == VBLK && bdevsw[major(vp->v_rdev)].d_type != D_DISK) { 154 simple_unlock(&uobj->vmobjlock); 155 UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)", 0,0,0,0); 156 return(NULL); 157 } 158 KASSERT(vp->v_type == VREG || vp->v_type == VBLK); 159 160 /* 161 * set up our idea of the size 162 * if this hasn't been done already. 163 */ 164 if (vp->v_size == VSIZENOTSET) { 165 166 167 vp->v_flag |= VXLOCK; 168 simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */ 169 /* XXX: curproc? */ 170 if (vp->v_type == VBLK) { 171 /* 172 * We could implement this as a specfs getattr call, but: 173 * 174 * (1) VOP_GETATTR() would get the file system 175 * vnode operation, not the specfs operation. 176 * 177 * (2) All we want is the size, anyhow. 178 */ 179 result = (*bdevsw[major(vp->v_rdev)].d_ioctl)(vp->v_rdev, 180 DIOCGPART, (caddr_t)&pi, FREAD, curproc); 181 if (result == 0) { 182 /* XXX should remember blocksize */ 183 used_vnode_size = (voff_t)pi.disklab->d_secsize * 184 (voff_t)pi.part->p_size; 185 } 186 } else { 187 result = VOP_GETATTR(vp, &vattr, curproc->p_ucred, curproc); 188 if (result == 0) 189 used_vnode_size = vattr.va_size; 190 } 191 192 /* relock object */ 193 simple_lock(&uobj->vmobjlock); 194 195 if (vp->v_flag & VXWANT) { 196 wakeup(vp); 197 } 198 vp->v_flag &= ~(VXLOCK|VXWANT); 199 200 if (result != 0) { 201 simple_unlock(&uobj->vmobjlock); 202 UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0); 203 return(NULL); 204 } 205 vp->v_size = used_vnode_size; 206 207 } 208 209 simple_unlock(&uobj->vmobjlock); 210 UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount, 211 0, 0, 0); 212 return uobj; 213 } 214 215 216 /* 217 * uvn_reference 218 * 219 * duplicate a reference to a VM object. Note that the reference 220 * count must already be at least one (the passed in reference) so 221 * there is no chance of the uvn being killed or locked out here. 222 * 223 * => caller must call with object unlocked. 224 * => caller must be using the same accessprot as was used at attach time 225 */ 226 227 void 228 uvn_reference(uobj) 229 struct uvm_object *uobj; 230 { 231 VREF((struct vnode *)uobj); 232 } 233 234 235 /* 236 * uvn_detach 237 * 238 * remove a reference to a VM object. 239 * 240 * => caller must call with object unlocked and map locked. 241 */ 242 243 void 244 uvn_detach(uobj) 245 struct uvm_object *uobj; 246 { 247 vrele((struct vnode *)uobj); 248 } 249 250 /* 251 * uvn_put: flush page data to backing store. 252 * 253 * => object must be locked on entry! VOP_PUTPAGES must unlock it. 254 * => flags: PGO_SYNCIO -- use sync. I/O 255 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed) 256 */ 257 258 int 259 uvn_put(uobj, offlo, offhi, flags) 260 struct uvm_object *uobj; 261 voff_t offlo; 262 voff_t offhi; 263 int flags; 264 { 265 struct vnode *vp = (struct vnode *)uobj; 266 int error; 267 268 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 269 error = VOP_PUTPAGES(vp, offlo, offhi, flags); 270 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock)); 271 return error; 272 } 273 274 275 /* 276 * uvn_get: get pages (synchronously) from backing store 277 * 278 * => prefer map unlocked (not required) 279 * => object must be locked! we will _unlock_ it before starting any I/O. 280 * => flags: PGO_ALLPAGES: get all of the pages 281 * PGO_LOCKED: fault data structures are locked 282 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx] 283 * => NOTE: caller must check for released pages!! 284 */ 285 286 int 287 uvn_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags) 288 struct uvm_object *uobj; 289 voff_t offset; 290 struct vm_page **pps; /* IN/OUT */ 291 int *npagesp; /* IN (OUT if PGO_LOCKED) */ 292 int centeridx; 293 vm_prot_t access_type; 294 int advice, flags; 295 { 296 struct vnode *vp = (struct vnode *)uobj; 297 int error; 298 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist); 299 300 UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0); 301 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx, 302 access_type, advice, flags); 303 return error; 304 } 305 306 307 /* 308 * uvn_findpages: 309 * return the page for the uobj and offset requested, allocating if needed. 310 * => uobj must be locked. 311 * => returned pages will be BUSY. 312 */ 313 314 int 315 uvn_findpages(uobj, offset, npagesp, pgs, flags) 316 struct uvm_object *uobj; 317 voff_t offset; 318 int *npagesp; 319 struct vm_page **pgs; 320 int flags; 321 { 322 int i, count, found, npages, rv; 323 324 count = found = 0; 325 npages = *npagesp; 326 if (flags & UFP_BACKWARD) { 327 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) { 328 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 329 if (rv == 0) { 330 if (flags & UFP_DIRTYONLY) 331 break; 332 } else 333 found++; 334 count++; 335 } 336 } else { 337 for (i = 0; i < npages; 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 } 347 *npagesp = count; 348 return (found); 349 } 350 351 int 352 uvn_findpage(uobj, offset, pgp, flags) 353 struct uvm_object *uobj; 354 voff_t offset; 355 struct vm_page **pgp; 356 int flags; 357 { 358 struct vm_page *pg; 359 boolean_t dirty; 360 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist); 361 UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0); 362 363 if (*pgp != NULL) { 364 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0); 365 return 0; 366 } 367 for (;;) { 368 /* look for an existing page */ 369 pg = uvm_pagelookup(uobj, offset); 370 371 /* nope? allocate one now */ 372 if (pg == NULL) { 373 if (flags & UFP_NOALLOC) { 374 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0); 375 return 0; 376 } 377 pg = uvm_pagealloc(uobj, offset, NULL, 0); 378 if (pg == NULL) { 379 if (flags & UFP_NOWAIT) { 380 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 381 return 0; 382 } 383 simple_unlock(&uobj->vmobjlock); 384 uvm_wait("uvn_fp1"); 385 simple_lock(&uobj->vmobjlock); 386 continue; 387 } 388 if (UVM_OBJ_IS_VTEXT(uobj)) { 389 uvmexp.execpages++; 390 } else { 391 uvmexp.filepages++; 392 } 393 UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0); 394 break; 395 } else if (flags & UFP_NOCACHE) { 396 UVMHIST_LOG(ubchist, "nocache",0,0,0,0); 397 return 0; 398 } 399 400 /* page is there, see if we need to wait on it */ 401 if ((pg->flags & PG_BUSY) != 0) { 402 if (flags & UFP_NOWAIT) { 403 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 404 return 0; 405 } 406 pg->flags |= PG_WANTED; 407 UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0); 408 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, 409 "uvn_fp2", 0); 410 simple_lock(&uobj->vmobjlock); 411 continue; 412 } 413 414 /* skip PG_RDONLY pages if requested */ 415 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) { 416 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0); 417 return 0; 418 } 419 420 /* stop on clean pages if requested */ 421 if (flags & UFP_DIRTYONLY) { 422 dirty = pmap_clear_modify(pg) || 423 (pg->flags & PG_CLEAN) == 0; 424 pg->flags |= PG_CLEAN; 425 if (!dirty) { 426 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0); 427 return 0; 428 } 429 } 430 431 /* mark the page BUSY and we're done. */ 432 pg->flags |= PG_BUSY; 433 UVM_PAGE_OWN(pg, "uvn_findpage"); 434 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0); 435 break; 436 } 437 *pgp = pg; 438 return 1; 439 } 440 441 /* 442 * uvm_vnp_setsize: grow or shrink a vnode uobj 443 * 444 * grow => just update size value 445 * shrink => toss un-needed pages 446 * 447 * => we assume that the caller has a reference of some sort to the 448 * vnode in question so that it will not be yanked out from under 449 * us. 450 */ 451 452 void 453 uvm_vnp_setsize(vp, newsize) 454 struct vnode *vp; 455 voff_t newsize; 456 { 457 struct uvm_object *uobj = &vp->v_uobj; 458 voff_t pgend = round_page(newsize); 459 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist); 460 461 simple_lock(&uobj->vmobjlock); 462 UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x", 463 vp, vp->v_size, newsize, 0); 464 465 /* 466 * now check if the size has changed: if we shrink we had better 467 * toss some pages... 468 */ 469 470 if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) { 471 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO); 472 } else { 473 simple_unlock(&uobj->vmobjlock); 474 } 475 vp->v_size = newsize; 476 } 477 478 /* 479 * uvm_vnp_zerorange: set a range of bytes in a file to zero. 480 */ 481 482 void 483 uvm_vnp_zerorange(vp, off, len) 484 struct vnode *vp; 485 off_t off; 486 size_t len; 487 { 488 void *win; 489 490 /* 491 * XXXUBC invent kzero() and use it 492 */ 493 494 while (len) { 495 vsize_t bytelen = len; 496 497 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UBC_WRITE); 498 memset(win, 0, bytelen); 499 ubc_release(win, 0); 500 501 off += bytelen; 502 len -= bytelen; 503 } 504 } 505