uvm_vnode.c revision 1.46.2.9
1 /* $NetBSD: uvm_vnode.c,v 1.46.2.9 2002/06/20 03:50:47 nathanw 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 1.46.2.9 2002/06/20 03:50:47 nathanw 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/lwp.h> 62 #include <sys/kernel.h> 63 #include <sys/proc.h> 64 #include <sys/malloc.h> 65 #include <sys/vnode.h> 66 #include <sys/disklabel.h> 67 #include <sys/ioctl.h> 68 #include <sys/fcntl.h> 69 #include <sys/conf.h> 70 #include <sys/pool.h> 71 #include <sys/mount.h> 72 73 #include <miscfs/specfs/specdev.h> 74 75 #include <uvm/uvm.h> 76 77 /* 78 * functions 79 */ 80 81 void uvn_detach __P((struct uvm_object *)); 82 int uvn_get __P((struct uvm_object *, voff_t, struct vm_page **, int *, int, 83 vm_prot_t, int, int)); 84 int uvn_put __P((struct uvm_object *, voff_t, voff_t, int)); 85 void uvn_reference __P((struct uvm_object *)); 86 87 int uvn_findpage __P((struct uvm_object *, voff_t, struct vm_page **, int)); 88 89 /* 90 * master pager structure 91 */ 92 93 struct uvm_pagerops uvm_vnodeops = { 94 NULL, 95 uvn_reference, 96 uvn_detach, 97 NULL, 98 uvn_get, 99 uvn_put, 100 }; 101 102 /* 103 * the ops! 104 */ 105 106 /* 107 * uvn_attach 108 * 109 * attach a vnode structure to a VM object. if the vnode is already 110 * attached, then just bump the reference count by one and return the 111 * VM object. if not already attached, attach and return the new VM obj. 112 * the "accessprot" tells the max access the attaching thread wants to 113 * our pages. 114 * 115 * => caller must _not_ already be holding the lock on the uvm_object. 116 * => in fact, nothing should be locked so that we can sleep here. 117 * => note that uvm_object is first thing in vnode structure, so their 118 * pointers are equiv. 119 */ 120 121 struct uvm_object * 122 uvn_attach(arg, accessprot) 123 void *arg; 124 vm_prot_t accessprot; 125 { 126 struct vnode *vp = arg; 127 struct uvm_object *uobj = &vp->v_uobj; 128 struct vattr vattr; 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 && bdevsw[major(vp->v_rdev)].d_type != D_DISK) { 155 simple_unlock(&uobj->vmobjlock); 156 UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)", 0,0,0,0); 157 return(NULL); 158 } 159 KASSERT(vp->v_type == VREG || vp->v_type == VBLK); 160 161 /* 162 * set up our idea of the size 163 * if this hasn't been done already. 164 */ 165 if (vp->v_size == VSIZENOTSET) { 166 167 168 vp->v_flag |= VXLOCK; 169 simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */ 170 /* XXX: curproc? */ 171 if (vp->v_type == VBLK) { 172 /* 173 * We could implement this as a specfs getattr call, but: 174 * 175 * (1) VOP_GETATTR() would get the file system 176 * vnode operation, not the specfs operation. 177 * 178 * (2) All we want is the size, anyhow. 179 */ 180 result = (*bdevsw[major(vp->v_rdev)].d_ioctl)(vp->v_rdev, 181 DIOCGPART, (caddr_t)&pi, FREAD, curproc->l_proc); 182 if (result == 0) { 183 /* XXX should remember blocksize */ 184 used_vnode_size = (voff_t)pi.disklab->d_secsize * 185 (voff_t)pi.part->p_size; 186 } 187 } else { 188 result = VOP_GETATTR(vp, &vattr, curproc->l_proc->p_ucred, 189 curproc->l_proc); 190 if (result == 0) 191 used_vnode_size = vattr.va_size; 192 } 193 194 /* relock object */ 195 simple_lock(&uobj->vmobjlock); 196 197 if (vp->v_flag & VXWANT) { 198 wakeup(vp); 199 } 200 vp->v_flag &= ~(VXLOCK|VXWANT); 201 202 if (result != 0) { 203 simple_unlock(&uobj->vmobjlock); 204 UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0); 205 return(NULL); 206 } 207 vp->v_size = used_vnode_size; 208 209 } 210 211 simple_unlock(&uobj->vmobjlock); 212 UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount, 213 0, 0, 0); 214 return uobj; 215 } 216 217 218 /* 219 * uvn_reference 220 * 221 * duplicate a reference to a VM object. Note that the reference 222 * count must already be at least one (the passed in reference) so 223 * there is no chance of the uvn being killed or locked out here. 224 * 225 * => caller must call with object unlocked. 226 * => caller must be using the same accessprot as was used at attach time 227 */ 228 229 void 230 uvn_reference(uobj) 231 struct uvm_object *uobj; 232 { 233 VREF((struct vnode *)uobj); 234 } 235 236 237 /* 238 * uvn_detach 239 * 240 * remove a reference to a VM object. 241 * 242 * => caller must call with object unlocked and map locked. 243 */ 244 245 void 246 uvn_detach(uobj) 247 struct uvm_object *uobj; 248 { 249 vrele((struct vnode *)uobj); 250 } 251 252 /* 253 * uvn_put: flush page data to backing store. 254 * 255 * => object must be locked on entry! VOP_PUTPAGES must unlock it. 256 * => flags: PGO_SYNCIO -- use sync. I/O 257 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed) 258 */ 259 260 int 261 uvn_put(uobj, offlo, offhi, flags) 262 struct uvm_object *uobj; 263 voff_t offlo; 264 voff_t offhi; 265 int flags; 266 { 267 struct vnode *vp = (struct vnode *)uobj; 268 int error; 269 270 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 271 error = VOP_PUTPAGES(vp, offlo, offhi, flags); 272 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock)); 273 return error; 274 } 275 276 277 /* 278 * uvn_get: get pages (synchronously) from backing store 279 * 280 * => prefer map unlocked (not required) 281 * => object must be locked! we will _unlock_ it before starting any I/O. 282 * => flags: PGO_ALLPAGES: get all of the pages 283 * PGO_LOCKED: fault data structures are locked 284 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx] 285 * => NOTE: caller must check for released pages!! 286 */ 287 288 int 289 uvn_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags) 290 struct uvm_object *uobj; 291 voff_t offset; 292 struct vm_page **pps; /* IN/OUT */ 293 int *npagesp; /* IN (OUT if PGO_LOCKED) */ 294 int centeridx; 295 vm_prot_t access_type; 296 int advice, flags; 297 { 298 struct vnode *vp = (struct vnode *)uobj; 299 int error; 300 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist); 301 302 UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0); 303 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx, 304 access_type, advice, flags); 305 return error; 306 } 307 308 309 /* 310 * uvn_findpages: 311 * return the page for the uobj and offset requested, allocating if needed. 312 * => uobj must be locked. 313 * => returned pages will be BUSY. 314 */ 315 316 int 317 uvn_findpages(uobj, offset, npagesp, pgs, flags) 318 struct uvm_object *uobj; 319 voff_t offset; 320 int *npagesp; 321 struct vm_page **pgs; 322 int flags; 323 { 324 int i, count, found, npages, rv; 325 326 count = found = 0; 327 npages = *npagesp; 328 if (flags & UFP_BACKWARD) { 329 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) { 330 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 331 if (rv == 0) { 332 if (flags & UFP_DIRTYONLY) 333 break; 334 } else 335 found++; 336 count++; 337 } 338 } else { 339 for (i = 0; i < npages; i++, offset += PAGE_SIZE) { 340 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 341 if (rv == 0) { 342 if (flags & UFP_DIRTYONLY) 343 break; 344 } else 345 found++; 346 count++; 347 } 348 } 349 *npagesp = count; 350 return (found); 351 } 352 353 int 354 uvn_findpage(uobj, offset, pgp, flags) 355 struct uvm_object *uobj; 356 voff_t offset; 357 struct vm_page **pgp; 358 int flags; 359 { 360 struct vm_page *pg; 361 boolean_t dirty; 362 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist); 363 UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0); 364 365 if (*pgp != NULL) { 366 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0); 367 return 0; 368 } 369 for (;;) { 370 /* look for an existing page */ 371 pg = uvm_pagelookup(uobj, offset); 372 373 /* nope? allocate one now */ 374 if (pg == NULL) { 375 if (flags & UFP_NOALLOC) { 376 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0); 377 return 0; 378 } 379 pg = uvm_pagealloc(uobj, offset, NULL, 0); 380 if (pg == NULL) { 381 if (flags & UFP_NOWAIT) { 382 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 383 return 0; 384 } 385 simple_unlock(&uobj->vmobjlock); 386 uvm_wait("uvn_fp1"); 387 simple_lock(&uobj->vmobjlock); 388 continue; 389 } 390 if (UVM_OBJ_IS_VTEXT(uobj)) { 391 uvmexp.execpages++; 392 } else { 393 uvmexp.filepages++; 394 } 395 UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0); 396 break; 397 } else if (flags & UFP_NOCACHE) { 398 UVMHIST_LOG(ubchist, "nocache",0,0,0,0); 399 return 0; 400 } 401 402 /* page is there, see if we need to wait on it */ 403 if ((pg->flags & PG_BUSY) != 0) { 404 if (flags & UFP_NOWAIT) { 405 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 406 return 0; 407 } 408 pg->flags |= PG_WANTED; 409 UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0); 410 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0, 411 "uvn_fp2", 0); 412 simple_lock(&uobj->vmobjlock); 413 continue; 414 } 415 416 /* skip PG_RDONLY pages if requested */ 417 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) { 418 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0); 419 return 0; 420 } 421 422 /* stop on clean pages if requested */ 423 if (flags & UFP_DIRTYONLY) { 424 dirty = pmap_clear_modify(pg) || 425 (pg->flags & PG_CLEAN) == 0; 426 pg->flags |= PG_CLEAN; 427 if (!dirty) { 428 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0); 429 return 0; 430 } 431 } 432 433 /* mark the page BUSY and we're done. */ 434 pg->flags |= PG_BUSY; 435 UVM_PAGE_OWN(pg, "uvn_findpage"); 436 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0); 437 break; 438 } 439 *pgp = pg; 440 return 1; 441 } 442 443 /* 444 * uvm_vnp_setsize: grow or shrink a vnode uobj 445 * 446 * grow => just update size value 447 * shrink => toss un-needed pages 448 * 449 * => we assume that the caller has a reference of some sort to the 450 * vnode in question so that it will not be yanked out from under 451 * us. 452 */ 453 454 void 455 uvm_vnp_setsize(vp, newsize) 456 struct vnode *vp; 457 voff_t newsize; 458 { 459 struct uvm_object *uobj = &vp->v_uobj; 460 voff_t pgend = round_page(newsize); 461 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist); 462 463 simple_lock(&uobj->vmobjlock); 464 UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x", 465 vp, vp->v_size, newsize, 0); 466 467 /* 468 * now check if the size has changed: if we shrink we had better 469 * toss some pages... 470 */ 471 472 if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) { 473 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO); 474 } else { 475 simple_unlock(&uobj->vmobjlock); 476 } 477 vp->v_size = newsize; 478 } 479 480 /* 481 * uvm_vnp_zerorange: set a range of bytes in a file to zero. 482 */ 483 484 void 485 uvm_vnp_zerorange(vp, off, len) 486 struct vnode *vp; 487 off_t off; 488 size_t len; 489 { 490 void *win; 491 492 /* 493 * XXXUBC invent kzero() and use it 494 */ 495 496 while (len) { 497 vsize_t bytelen = len; 498 499 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UBC_WRITE); 500 memset(win, 0, bytelen); 501 ubc_release(win, 0); 502 503 off += bytelen; 504 len -= bytelen; 505 } 506 } 507
Indexes created Sat Oct 11 13:09:53 GMT 2025