1 /* $NetBSD: uvm_vnode.c,v 1.96 2011/06/12 03:36:04 rmind 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. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * @(#)vnode_pager.c 8.8 (Berkeley) 2/13/94 40 * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp 41 */ 42 43 /* 44 * uvm_vnode.c: the vnode pager. 45 */ 46 47 #include <sys/cdefs.h> 48 __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.96 2011/06/12 03:36:04 rmind Exp $"); 49 50 #include "opt_uvmhist.h" 51 52 #include <sys/param.h> 53 #include <sys/systm.h> 54 #include <sys/kernel.h> 55 #include <sys/vnode.h> 56 #include <sys/disklabel.h> 57 #include <sys/ioctl.h> 58 #include <sys/fcntl.h> 59 #include <sys/conf.h> 60 #include <sys/pool.h> 61 #include <sys/mount.h> 62 63 #include <miscfs/specfs/specdev.h> 64 65 #include <uvm/uvm.h> 66 #include <uvm/uvm_readahead.h> 67 68 /* 69 * functions 70 */ 71 72 static void uvn_detach(struct uvm_object *); 73 static int uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *, 74 int, vm_prot_t, int, int); 75 static int uvn_put(struct uvm_object *, voff_t, voff_t, int); 76 static void uvn_reference(struct uvm_object *); 77 78 static int uvn_findpage(struct uvm_object *, voff_t, struct vm_page **, 79 int); 80 81 /* 82 * master pager structure 83 */ 84 85 const struct uvm_pagerops uvm_vnodeops = { 86 .pgo_reference = uvn_reference, 87 .pgo_detach = uvn_detach, 88 .pgo_get = uvn_get, 89 .pgo_put = uvn_put, 90 }; 91 92 /* 93 * the ops! 94 */ 95 96 /* 97 * uvn_reference 98 * 99 * duplicate a reference to a VM object. Note that the reference 100 * count must already be at least one (the passed in reference) so 101 * there is no chance of the uvn being killed or locked out here. 102 * 103 * => caller must call with object unlocked. 104 * => caller must be using the same accessprot as was used at attach time 105 */ 106 107 static void 108 uvn_reference(struct uvm_object *uobj) 109 { 110 vref((struct vnode *)uobj); 111 } 112 113 114 /* 115 * uvn_detach 116 * 117 * remove a reference to a VM object. 118 * 119 * => caller must call with object unlocked and map locked. 120 */ 121 122 static void 123 uvn_detach(struct uvm_object *uobj) 124 { 125 vrele((struct vnode *)uobj); 126 } 127 128 /* 129 * uvn_put: flush page data to backing store. 130 * 131 * => object must be locked on entry! VOP_PUTPAGES must unlock it. 132 * => flags: PGO_SYNCIO -- use sync. I/O 133 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed) 134 */ 135 136 static int 137 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags) 138 { 139 struct vnode *vp = (struct vnode *)uobj; 140 int error; 141 142 KASSERT(mutex_owned(vp->v_interlock)); 143 error = VOP_PUTPAGES(vp, offlo, offhi, flags); 144 145 return error; 146 } 147 148 149 /* 150 * uvn_get: get pages (synchronously) from backing store 151 * 152 * => prefer map unlocked (not required) 153 * => object must be locked! we will _unlock_ it before starting any I/O. 154 * => flags: PGO_ALLPAGES: get all of the pages 155 * PGO_LOCKED: fault data structures are locked 156 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx] 157 * => NOTE: caller must check for released pages!! 158 */ 159 160 static int 161 uvn_get(struct uvm_object *uobj, voff_t offset, 162 struct vm_page **pps /* IN/OUT */, 163 int *npagesp /* IN (OUT if PGO_LOCKED)*/, 164 int centeridx, vm_prot_t access_type, int advice, int flags) 165 { 166 struct vnode *vp = (struct vnode *)uobj; 167 int error; 168 169 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist); 170 171 UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0); 172 173 if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) { 174 vn_ra_allocctx(vp); 175 uvm_ra_request(vp->v_ractx, advice, uobj, offset, 176 *npagesp << PAGE_SHIFT); 177 } 178 179 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx, 180 access_type, advice, flags); 181 182 KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(vp->v_interlock)) || 183 (flags & PGO_LOCKED) == 0); 184 return error; 185 } 186 187 188 /* 189 * uvn_findpages: 190 * return the page for the uobj and offset requested, allocating if needed. 191 * => uobj must be locked. 192 * => returned pages will be BUSY. 193 */ 194 195 int 196 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp, 197 struct vm_page **pgs, int flags) 198 { 199 int i, count, found, npages, rv; 200 201 count = found = 0; 202 npages = *npagesp; 203 if (flags & UFP_BACKWARD) { 204 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) { 205 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 206 if (rv == 0) { 207 if (flags & UFP_DIRTYONLY) 208 break; 209 } else 210 found++; 211 count++; 212 } 213 } else { 214 for (i = 0; i < npages; i++, offset += PAGE_SIZE) { 215 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 216 if (rv == 0) { 217 if (flags & UFP_DIRTYONLY) 218 break; 219 } else 220 found++; 221 count++; 222 } 223 } 224 *npagesp = count; 225 return (found); 226 } 227 228 static int 229 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp, 230 int flags) 231 { 232 struct vm_page *pg; 233 bool dirty; 234 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist); 235 UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0); 236 237 KASSERT(mutex_owned(uobj->vmobjlock)); 238 239 if (*pgp != NULL) { 240 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0); 241 return 0; 242 } 243 for (;;) { 244 /* look for an existing page */ 245 pg = uvm_pagelookup(uobj, offset); 246 247 /* nope? allocate one now */ 248 if (pg == NULL) { 249 if (flags & UFP_NOALLOC) { 250 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0); 251 return 0; 252 } 253 pg = uvm_pagealloc(uobj, offset, NULL, 0); 254 if (pg == NULL) { 255 if (flags & UFP_NOWAIT) { 256 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 257 return 0; 258 } 259 mutex_exit(uobj->vmobjlock); 260 uvm_wait("uvn_fp1"); 261 mutex_enter(uobj->vmobjlock); 262 continue; 263 } 264 UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0); 265 break; 266 } else if (flags & UFP_NOCACHE) { 267 UVMHIST_LOG(ubchist, "nocache",0,0,0,0); 268 return 0; 269 } 270 271 /* page is there, see if we need to wait on it */ 272 if ((pg->flags & PG_BUSY) != 0) { 273 if (flags & UFP_NOWAIT) { 274 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 275 return 0; 276 } 277 pg->flags |= PG_WANTED; 278 UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0); 279 UVM_UNLOCK_AND_WAIT(pg, uobj->vmobjlock, 0, 280 "uvn_fp2", 0); 281 mutex_enter(uobj->vmobjlock); 282 continue; 283 } 284 285 /* skip PG_RDONLY pages if requested */ 286 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) { 287 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0); 288 return 0; 289 } 290 291 /* stop on clean pages if requested */ 292 if (flags & UFP_DIRTYONLY) { 293 dirty = pmap_clear_modify(pg) || 294 (pg->flags & PG_CLEAN) == 0; 295 pg->flags |= PG_CLEAN; 296 if (!dirty) { 297 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0); 298 return 0; 299 } 300 } 301 302 /* mark the page BUSY and we're done. */ 303 pg->flags |= PG_BUSY; 304 UVM_PAGE_OWN(pg, "uvn_findpage"); 305 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0); 306 break; 307 } 308 *pgp = pg; 309 return 1; 310 } 311 312 /* 313 * uvm_vnp_setsize: grow or shrink a vnode uobj 314 * 315 * grow => just update size value 316 * shrink => toss un-needed pages 317 * 318 * => we assume that the caller has a reference of some sort to the 319 * vnode in question so that it will not be yanked out from under 320 * us. 321 */ 322 323 void 324 uvm_vnp_setsize(struct vnode *vp, voff_t newsize) 325 { 326 struct uvm_object *uobj = &vp->v_uobj; 327 voff_t pgend = round_page(newsize); 328 voff_t oldsize; 329 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist); 330 331 mutex_enter(uobj->vmobjlock); 332 UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x", 333 vp, vp->v_size, newsize, 0); 334 335 /* 336 * now check if the size has changed: if we shrink we had better 337 * toss some pages... 338 */ 339 340 KASSERT(newsize != VSIZENOTSET); 341 KASSERT(vp->v_size <= vp->v_writesize); 342 KASSERT(vp->v_size == vp->v_writesize || 343 newsize == vp->v_writesize || newsize <= vp->v_size); 344 345 oldsize = vp->v_writesize; 346 KASSERT(oldsize != VSIZENOTSET || pgend > oldsize); 347 348 if (oldsize > pgend) { 349 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO); 350 mutex_enter(uobj->vmobjlock); 351 } 352 vp->v_size = vp->v_writesize = newsize; 353 mutex_exit(uobj->vmobjlock); 354 } 355 356 void 357 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize) 358 { 359 360 mutex_enter(vp->v_interlock); 361 KASSERT(newsize != VSIZENOTSET); 362 KASSERT(vp->v_size != VSIZENOTSET); 363 KASSERT(vp->v_writesize != VSIZENOTSET); 364 KASSERT(vp->v_size <= vp->v_writesize); 365 KASSERT(vp->v_size <= newsize); 366 vp->v_writesize = newsize; 367 mutex_exit(vp->v_interlock); 368 } 369 370 bool 371 uvn_text_p(struct uvm_object *uobj) 372 { 373 struct vnode *vp = (struct vnode *)uobj; 374 375 return (vp->v_iflag & VI_EXECMAP) != 0; 376 } 377 378 bool 379 uvn_clean_p(struct uvm_object *uobj) 380 { 381 struct vnode *vp = (struct vnode *)uobj; 382 383 return (vp->v_iflag & VI_ONWORKLST) == 0; 384 } 385 386 bool 387 uvn_needs_writefault_p(struct uvm_object *uobj) 388 { 389 struct vnode *vp = (struct vnode *)uobj; 390 391 return uvn_clean_p(uobj) || 392 (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP; 393 } 394