1 /* $NetBSD: uvm_vnode.c,v 1.103 2017/10/28 00:37:13 pgoyette 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 1.103 2017/10/28 00:37:13 pgoyette Exp $"); 49 50 #ifdef _KERNEL_OPT 51 #include "opt_uvmhist.h" 52 #endif 53 54 #include <sys/param.h> 55 #include <sys/systm.h> 56 #include <sys/kernel.h> 57 #include <sys/vnode.h> 58 #include <sys/disklabel.h> 59 #include <sys/ioctl.h> 60 #include <sys/fcntl.h> 61 #include <sys/conf.h> 62 #include <sys/pool.h> 63 #include <sys/mount.h> 64 65 #include <miscfs/specfs/specdev.h> 66 67 #include <uvm/uvm.h> 68 #include <uvm/uvm_readahead.h> 69 70 #ifdef UVMHIST 71 UVMHIST_DEFINE(ubchist); 72 #endif 73 74 /* 75 * functions 76 */ 77 78 static void uvn_detach(struct uvm_object *); 79 static int uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *, 80 int, vm_prot_t, int, int); 81 static int uvn_put(struct uvm_object *, voff_t, voff_t, int); 82 static void uvn_reference(struct uvm_object *); 83 84 static int uvn_findpage(struct uvm_object *, voff_t, struct vm_page **, 85 int); 86 87 /* 88 * master pager structure 89 */ 90 91 const struct uvm_pagerops uvm_vnodeops = { 92 .pgo_reference = uvn_reference, 93 .pgo_detach = uvn_detach, 94 .pgo_get = uvn_get, 95 .pgo_put = uvn_put, 96 }; 97 98 /* 99 * the ops! 100 */ 101 102 /* 103 * uvn_reference 104 * 105 * duplicate a reference to a VM object. Note that the reference 106 * count must already be at least one (the passed in reference) so 107 * there is no chance of the uvn being killed or locked out here. 108 * 109 * => caller must call with object unlocked. 110 * => caller must be using the same accessprot as was used at attach time 111 */ 112 113 static void 114 uvn_reference(struct uvm_object *uobj) 115 { 116 vref((struct vnode *)uobj); 117 } 118 119 120 /* 121 * uvn_detach 122 * 123 * remove a reference to a VM object. 124 * 125 * => caller must call with object unlocked and map locked. 126 */ 127 128 static void 129 uvn_detach(struct uvm_object *uobj) 130 { 131 vrele((struct vnode *)uobj); 132 } 133 134 /* 135 * uvn_put: flush page data to backing store. 136 * 137 * => object must be locked on entry! VOP_PUTPAGES must unlock it. 138 * => flags: PGO_SYNCIO -- use sync. I/O 139 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed) 140 */ 141 142 static int 143 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags) 144 { 145 struct vnode *vp = (struct vnode *)uobj; 146 int error; 147 148 KASSERT(mutex_owned(vp->v_interlock)); 149 error = VOP_PUTPAGES(vp, offlo, offhi, flags); 150 151 return error; 152 } 153 154 155 /* 156 * uvn_get: get pages (synchronously) from backing store 157 * 158 * => prefer map unlocked (not required) 159 * => object must be locked! we will _unlock_ it before starting any I/O. 160 * => flags: PGO_ALLPAGES: get all of the pages 161 * PGO_LOCKED: fault data structures are locked 162 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx] 163 * => NOTE: caller must check for released pages!! 164 */ 165 166 static int 167 uvn_get(struct uvm_object *uobj, voff_t offset, 168 struct vm_page **pps /* IN/OUT */, 169 int *npagesp /* IN (OUT if PGO_LOCKED)*/, 170 int centeridx, vm_prot_t access_type, int advice, int flags) 171 { 172 struct vnode *vp = (struct vnode *)uobj; 173 int error; 174 175 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist); 176 177 UVMHIST_LOG(ubchist, "vp %#jx off 0x%jx", (uintptr_t)vp, (int)offset, 178 0, 0); 179 180 if (vp->v_type == VREG && (access_type & VM_PROT_WRITE) == 0 181 && (flags & PGO_LOCKED) == 0) { 182 vn_ra_allocctx(vp); 183 uvm_ra_request(vp->v_ractx, advice, uobj, offset, 184 *npagesp << PAGE_SHIFT); 185 } 186 187 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx, 188 access_type, advice, flags); 189 190 KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(vp->v_interlock)) || 191 (flags & PGO_LOCKED) == 0); 192 return error; 193 } 194 195 196 /* 197 * uvn_findpages: 198 * return the page for the uobj and offset requested, allocating if needed. 199 * => uobj must be locked. 200 * => returned pages will be BUSY. 201 */ 202 203 int 204 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp, 205 struct vm_page **pgs, int flags) 206 { 207 int i, count, found, npages, rv; 208 209 count = found = 0; 210 npages = *npagesp; 211 if (flags & UFP_BACKWARD) { 212 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) { 213 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 214 if (rv == 0) { 215 if (flags & UFP_DIRTYONLY) 216 break; 217 } else 218 found++; 219 count++; 220 } 221 } else { 222 for (i = 0; i < npages; i++, offset += PAGE_SIZE) { 223 rv = uvn_findpage(uobj, offset, &pgs[i], flags); 224 if (rv == 0) { 225 if (flags & UFP_DIRTYONLY) 226 break; 227 } else 228 found++; 229 count++; 230 } 231 } 232 *npagesp = count; 233 return (found); 234 } 235 236 static int 237 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp, 238 int flags) 239 { 240 struct vm_page *pg; 241 bool dirty; 242 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist); 243 UVMHIST_LOG(ubchist, "vp %#jx off 0x%jx", (uintptr_t)uobj, offset, 244 0, 0); 245 246 KASSERT(mutex_owned(uobj->vmobjlock)); 247 248 if (*pgp != NULL) { 249 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0); 250 return 0; 251 } 252 for (;;) { 253 /* look for an existing page */ 254 pg = uvm_pagelookup(uobj, offset); 255 256 /* nope? allocate one now */ 257 if (pg == NULL) { 258 if (flags & UFP_NOALLOC) { 259 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0); 260 return 0; 261 } 262 pg = uvm_pagealloc(uobj, offset, NULL, 263 UVM_FLAG_COLORMATCH); 264 if (pg == NULL) { 265 if (flags & UFP_NOWAIT) { 266 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 267 return 0; 268 } 269 mutex_exit(uobj->vmobjlock); 270 uvm_wait("uvn_fp1"); 271 mutex_enter(uobj->vmobjlock); 272 continue; 273 } 274 UVMHIST_LOG(ubchist, "alloced %#jx (color %ju)", 275 (uintptr_t)pg, VM_PGCOLOR_BUCKET(pg), 0, 0); 276 break; 277 } else if (flags & UFP_NOCACHE) { 278 UVMHIST_LOG(ubchist, "nocache",0,0,0,0); 279 return 0; 280 } 281 282 /* page is there, see if we need to wait on it */ 283 if ((pg->flags & PG_BUSY) != 0) { 284 if (flags & UFP_NOWAIT) { 285 UVMHIST_LOG(ubchist, "nowait",0,0,0,0); 286 return 0; 287 } 288 pg->flags |= PG_WANTED; 289 UVMHIST_LOG(ubchist, "wait %#jx (color %ju)", 290 (uintptr_t)pg, VM_PGCOLOR_BUCKET(pg), 0, 0); 291 UVM_UNLOCK_AND_WAIT(pg, uobj->vmobjlock, 0, 292 "uvn_fp2", 0); 293 mutex_enter(uobj->vmobjlock); 294 continue; 295 } 296 297 /* skip PG_RDONLY pages if requested */ 298 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) { 299 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0); 300 return 0; 301 } 302 303 /* stop on clean pages if requested */ 304 if (flags & UFP_DIRTYONLY) { 305 dirty = pmap_clear_modify(pg) || 306 (pg->flags & PG_CLEAN) == 0; 307 pg->flags |= PG_CLEAN; 308 if (!dirty) { 309 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0); 310 return 0; 311 } 312 } 313 314 /* mark the page BUSY and we're done. */ 315 pg->flags |= PG_BUSY; 316 UVM_PAGE_OWN(pg, "uvn_findpage"); 317 UVMHIST_LOG(ubchist, "found %#jx (color %ju)", 318 (uintptr_t)pg, VM_PGCOLOR_BUCKET(pg), 0, 0); 319 break; 320 } 321 *pgp = pg; 322 return 1; 323 } 324 325 /* 326 * uvm_vnp_setsize: grow or shrink a vnode uobj 327 * 328 * grow => just update size value 329 * shrink => toss un-needed pages 330 * 331 * => we assume that the caller has a reference of some sort to the 332 * vnode in question so that it will not be yanked out from under 333 * us. 334 */ 335 336 void 337 uvm_vnp_setsize(struct vnode *vp, voff_t newsize) 338 { 339 struct uvm_object *uobj = &vp->v_uobj; 340 voff_t pgend = round_page(newsize); 341 voff_t oldsize; 342 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist); 343 344 mutex_enter(uobj->vmobjlock); 345 UVMHIST_LOG(ubchist, "vp %#jx old 0x%jx new 0x%jx", 346 (uintptr_t)vp, vp->v_size, newsize, 0); 347 348 /* 349 * now check if the size has changed: if we shrink we had better 350 * toss some pages... 351 */ 352 353 KASSERT(newsize != VSIZENOTSET && newsize >= 0); 354 KASSERT(vp->v_size <= vp->v_writesize); 355 KASSERT(vp->v_size == vp->v_writesize || 356 newsize == vp->v_writesize || newsize <= vp->v_size); 357 358 oldsize = vp->v_writesize; 359 360 /* 361 * check whether size shrinks 362 * if old size hasn't been set, there are no pages to drop 363 * if there was an integer overflow in pgend, then this is no shrink 364 */ 365 if (oldsize > pgend && oldsize != VSIZENOTSET && pgend >= 0) { 366 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO); 367 mutex_enter(uobj->vmobjlock); 368 } 369 vp->v_size = vp->v_writesize = newsize; 370 mutex_exit(uobj->vmobjlock); 371 } 372 373 void 374 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize) 375 { 376 377 mutex_enter(vp->v_interlock); 378 KASSERT(newsize != VSIZENOTSET && newsize >= 0); 379 KASSERT(vp->v_size != VSIZENOTSET); 380 KASSERT(vp->v_writesize != VSIZENOTSET); 381 KASSERT(vp->v_size <= vp->v_writesize); 382 KASSERT(vp->v_size <= newsize); 383 vp->v_writesize = newsize; 384 mutex_exit(vp->v_interlock); 385 } 386 387 bool 388 uvn_text_p(struct uvm_object *uobj) 389 { 390 struct vnode *vp = (struct vnode *)uobj; 391 392 return (vp->v_iflag & VI_EXECMAP) != 0; 393 } 394 395 bool 396 uvn_clean_p(struct uvm_object *uobj) 397 { 398 struct vnode *vp = (struct vnode *)uobj; 399 400 return (vp->v_iflag & VI_ONWORKLST) == 0; 401 } 402 403 bool 404 uvn_needs_writefault_p(struct uvm_object *uobj) 405 { 406 struct vnode *vp = (struct vnode *)uobj; 407 408 return uvn_clean_p(uobj) || 409 (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP; 410 } 411