sys/vm/vm_vmspace.c

   1 /*
   2  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  * $DragonFly: src/sys/vm/vm_vmspace.c,v 1.11 2007/04/29 18:25:41 dillon Exp $
  35  */
  36 #include "opt_ddb.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/kernel.h>
  40 #include <sys/systm.h>
  41 #include <sys/sysproto.h>
  42 #include <sys/kern_syscall.h>
  43 #include <sys/mman.h>
  44 #include <sys/proc.h>
  45 #include <sys/malloc.h>
  46 #include <sys/sysctl.h>
  47 #include <sys/vkernel.h>
  48 #include <sys/vmspace.h>
  49
  50 #include <vm/vm_extern.h>
  51 #include <vm/pmap.h>
  52 #include <ddb/ddb.h>
  53
  54 #include <machine/vmparam.h>
  55
  56 #include <sys/spinlock2.h>
  57 #include <sys/sysref2.h>
  58
  59 static struct vmspace_entry *vkernel_find_vmspace(struct vkernel_common *vc,
  60                                                   void *id);
  61 static void vmspace_entry_delete(struct vmspace_entry *ve,
  62                                  struct vkernel_common *vc);
  63
  64 static MALLOC_DEFINE(M_VKERNEL, "vkernel", "VKernel structures");
  65
  66 /*
  67  * vmspace_create (void *id, int type, void *data)
  68  *
  69  * Create a VMSPACE under the control of the caller with the specified id.
  70  * An id of NULL cannot be used.  The type and data fields must currently
  71  * be 0.
  72  *
  73  * The vmspace starts out completely empty.  Memory may be mapped into the
  74  * VMSPACE with vmspace_mmap() and MAP_VPAGETABLE section(s) controlled
  75  * with vmspace_mcontrol().
  76  */
  77 int
  78 sys_vmspace_create(struct vmspace_create_args *uap)
  79 {
  80         struct vkernel_common *vc;
  81         struct vmspace_entry *ve;
  82         struct vkernel *vk;
  83
  84         if (vkernel_enable == 0)
  85                 return (EOPNOTSUPP);
  86
  87         /*
  88          * Create a virtual kernel side-structure for the process if one
  89          * does not exist.
  90          */
  91         if ((vk = curproc->p_vkernel) == NULL) {
  92                 vk = kmalloc(sizeof(*vk), M_VKERNEL, M_WAITOK|M_ZERO);
  93                 vc = kmalloc(sizeof(*vc), M_VKERNEL, M_WAITOK|M_ZERO);
  94                 vc->vc_refs = 1;
  95                 spin_init(&vc->vc_spin);
  96                 RB_INIT(&vc->vc_root);
  97                 vk->vk_common = vc;
  98                 curproc->p_vkernel = vk;
  99         }
 100         vc = vk->vk_common;
 101
 102         /*
 103          * Create a new VMSPACE
 104          */
 105         if (vkernel_find_vmspace(vc, uap->id))
 106                 return (EEXIST);
 107         ve = kmalloc(sizeof(struct vmspace_entry), M_VKERNEL, M_WAITOK|M_ZERO);
 108         ve->vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 109         ve->id = uap->id;
 110         pmap_pinit2(vmspace_pmap(ve->vmspace));
 111         RB_INSERT(vmspace_rb_tree, &vc->vc_root, ve);
 112         return (0);
 113 }
 114
 115 /*
 116  * vmspace_destroy (void *id)
 117  *
 118  * Destroy a VMSPACE.
 119  */
 120 int
 121 sys_vmspace_destroy(struct vmspace_destroy_args *uap)
 122 {
 123         struct vkernel_common *vc;
 124         struct vmspace_entry *ve;
 125         struct vkernel *vk;
 126
 127         if ((vk = curproc->p_vkernel) == NULL)
 128                 return (EINVAL);
 129         vc = vk->vk_common;
 130         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 131                 return (ENOENT);
 132         if (ve->refs)
 133                 return (EBUSY);
 134         vmspace_entry_delete(ve, vc);
 135         return(0);
 136 }
 137
 138 /*
 139  * vmspace_ctl (void *id, int cmd, struct trapframe *tframe,
 140  *              struct vextframe *vframe);
 141  *
 142  * Transfer control to a VMSPACE.  Control is returned after the specified
 143  * number of microseconds or if a page fault, signal, trap, or system call
 144  * occurs.  The context is updated as appropriate.
 145  */
 146 int
 147 sys_vmspace_ctl(struct vmspace_ctl_args *uap)
 148 {
 149         struct vkernel_common *vc;
 150         struct vmspace_entry *ve;
 151         struct vkernel *vk;
 152         struct proc *p;
 153         int framesz;
 154         int error;
 155
 156         if ((vk = curproc->p_vkernel) == NULL)
 157                 return (EINVAL);
 158         vc = vk->vk_common;
 159         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 160                 return (ENOENT);
 161
 162         /*
 163          * Signal mailbox interlock
 164          */
 165         if (curproc->p_flag & P_MAILBOX) {
 166                 curproc->p_flag &= ~P_MAILBOX;
 167                 return (EINTR);
 168         }
 169
 170         switch(uap->cmd) {
 171         case VMSPACE_CTL_RUN:
 172                 /*
 173                  * Save the caller's register context, swap VM spaces, and
 174                  * install the passed register context.  Return with
 175                  * EJUSTRETURN so the syscall code doesn't adjust the context.
 176                  */
 177                 p = curproc;
 178                 ++ve->refs;
 179                 framesz = sizeof(struct trapframe);
 180                 vk->vk_current = ve;
 181                 vk->vk_save_vmspace = p->p_vmspace;
 182                 vk->vk_user_trapframe = uap->tframe;
 183                 vk->vk_user_vextframe = uap->vframe;
 184                 bcopy(uap->sysmsg_frame, &vk->vk_save_trapframe, framesz);
 185                 bcopy(&curthread->td_tls, &vk->vk_save_vextframe.vx_tls,
 186                       sizeof(vk->vk_save_vextframe.vx_tls));
 187                 error = copyin(uap->tframe, uap->sysmsg_frame, framesz);
 188                 if (error == 0)
 189                         error = copyin(&uap->vframe->vx_tls, &curthread->td_tls, sizeof(struct savetls));
 190                 if (error == 0)
 191                         error = cpu_sanitize_frame(uap->sysmsg_frame);
 192                 if (error == 0)
 193                         error = cpu_sanitize_tls(&curthread->td_tls);
 194                 if (error) {
 195                         bcopy(&vk->vk_save_trapframe, uap->sysmsg_frame, framesz);
 196                         bcopy(&vk->vk_save_vextframe.vx_tls, &curthread->td_tls,
 197                               sizeof(vk->vk_save_vextframe.vx_tls));
 198                         set_user_TLS();
 199                         vk->vk_current = NULL;
 200                         vk->vk_save_vmspace = NULL;
 201                         --ve->refs;
 202                 } else {
 203                         pmap_replacevm(p, ve->vmspace, 0);
 204                         set_user_TLS();
 205                         set_vkernel_fp(uap->sysmsg_frame);
 206                         error = EJUSTRETURN;
 207                 }
 208                 break;
 209         default:
 210                 error = EOPNOTSUPP;
 211                 break;
 212         }
 213         return(error);
 214 }
 215
 216 /*
 217  * vmspace_mmap(id, addr, len, prot, flags, fd, offset)
 218  *
 219  * map memory within a VMSPACE.  This function is just like a normal mmap()
 220  * but operates on the vmspace's memory map.  Most callers use this to create
 221  * a MAP_VPAGETABLE mapping.
 222  */
 223 int
 224 sys_vmspace_mmap(struct vmspace_mmap_args *uap)
 225 {
 226         struct vkernel_common *vc;
 227         struct vmspace_entry *ve;
 228         struct vkernel *vk;
 229         int error;
 230
 231         if ((vk = curproc->p_vkernel) == NULL)
 232                 return (EINVAL);
 233         vc = vk->vk_common;
 234         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 235                 return (ENOENT);
 236         error = kern_mmap(ve->vmspace, uap->addr, uap->len,
 237                           uap->prot, uap->flags,
 238                           uap->fd, uap->offset, &uap->sysmsg_resultp);
 239         return (error);
 240 }
 241
 242 /*
 243  * vmspace_munmap(id, addr, len)
 244  *
 245  * unmap memory within a VMSPACE.
 246  */
 247 int
 248 sys_vmspace_munmap(struct vmspace_munmap_args *uap)
 249 {
 250         struct vkernel_common *vc;
 251         struct vmspace_entry *ve;
 252         struct vkernel *vk;
 253         vm_offset_t addr;
 254         vm_size_t size, pageoff;
 255         vm_map_t map;
 256
 257         if ((vk = curproc->p_vkernel) == NULL)
 258                 return (EINVAL);
 259         vc = vk->vk_common;
 260         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 261                 return (ENOENT);
 262
 263         /*
 264          * Copied from sys_munmap()
 265          */
 266         addr = (vm_offset_t)uap->addr;
 267         size = uap->len;
 268
 269         pageoff = (addr & PAGE_MASK);
 270         addr -= pageoff;
 271         size += pageoff;
 272         size = (vm_size_t)round_page(size);
 273         if (addr + size < addr)
 274                 return (EINVAL);
 275         if (size == 0)
 276                 return (0);
 277
 278         if (VM_MAX_USER_ADDRESS > 0 && addr + size > VM_MAX_USER_ADDRESS)
 279                 return (EINVAL);
 280         if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
 281                 return (EINVAL);
 282         map = &ve->vmspace->vm_map;
 283         if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
 284                 return (EINVAL);
 285         vm_map_remove(map, addr, addr + size);
 286         return (0);
 287 }
 288
 289 /*
 290  * vmspace_pread(id, buf, nbyte, flags, offset)
 291  *
 292  * Read data from a vmspace.  The number of bytes read is returned or
 293  * -1 if an unrecoverable error occured.  If the number of bytes read is
 294  * less then the request size, a page fault occured in the VMSPACE which
 295  * the caller must resolve in order to proceed.
 296  */
 297 int
 298 sys_vmspace_pread(struct vmspace_pread_args *uap)
 299 {
 300         struct vkernel_common *vc;
 301         struct vmspace_entry *ve;
 302         struct vkernel *vk;
 303
 304         if ((vk = curproc->p_vkernel) == NULL)
 305                 return (EINVAL);
 306         vc = vk->vk_common;
 307         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 308                 return (ENOENT);
 309         return (EINVAL);
 310 }
 311
 312 /*
 313  * vmspace_pwrite(id, buf, nbyte, flags, offset)
 314  *
 315  * Write data to a vmspace.  The number of bytes written is returned or
 316  * -1 if an unrecoverable error occured.  If the number of bytes written is
 317  * less then the request size, a page fault occured in the VMSPACE which
 318  * the caller must resolve in order to proceed.
 319  */
 320 int
 321 sys_vmspace_pwrite(struct vmspace_pwrite_args *uap)
 322 {
 323         struct vkernel_common *vc;
 324         struct vmspace_entry *ve;
 325         struct vkernel *vk;
 326
 327         if ((vk = curproc->p_vkernel) == NULL)
 328                 return (EINVAL);
 329         vc = vk->vk_common;
 330         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 331                 return (ENOENT);
 332         return (EINVAL);
 333 }
 334
 335 /*
 336  * vmspace_mcontrol(id, addr, len, behav, value)
 337  *
 338  * madvise/mcontrol support for a vmspace.
 339  */
 340 int
 341 sys_vmspace_mcontrol(struct vmspace_mcontrol_args *uap)
 342 {
 343         struct vkernel_common *vc;
 344         struct vmspace_entry *ve;
 345         struct vkernel *vk;
 346         vm_offset_t start, end;
 347
 348         if ((vk = curproc->p_vkernel) == NULL)
 349                 return (EINVAL);
 350         vc = vk->vk_common;
 351         if ((ve = vkernel_find_vmspace(vc, uap->id)) == NULL)
 352                 return (ENOENT);
 353
 354         /*
 355          * This code is basically copied from sys_mcontrol()
 356          */
 357         if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
 358                 return (EINVAL);
 359
 360         if (VM_MAX_USER_ADDRESS > 0 &&
 361                 ((vm_offset_t) uap->addr + uap->len) > VM_MAX_USER_ADDRESS)
 362                 return (EINVAL);
 363         if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
 364                 return (EINVAL);
 365         if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
 366                 return (EINVAL);
 367
 368         start = trunc_page((vm_offset_t) uap->addr);
 369         end = round_page((vm_offset_t) uap->addr + uap->len);
 370
 371         return (vm_map_madvise(&ve->vmspace->vm_map, start, end,
 372                                 uap->behav, uap->value));
 373 }
 374
 375 /*
 376  * Red black tree functions
 377  */
 378 static int rb_vmspace_compare(struct vmspace_entry *, struct vmspace_entry *);
 379 RB_GENERATE(vmspace_rb_tree, vmspace_entry, rb_entry, rb_vmspace_compare);
 380
 381 /* a->start is address, and the only field has to be initialized */
 382 static int
 383 rb_vmspace_compare(struct vmspace_entry *a, struct vmspace_entry *b)
 384 {
 385         if ((char *)a->id < (char *)b->id)
 386                 return(-1);
 387         else if ((char *)a->id > (char *)b->id)
 388                 return(1);
 389         return(0);
 390 }
 391
 392 static
 393 int
 394 rb_vmspace_delete(struct vmspace_entry *ve, void *data)
 395 {
 396         struct vkernel_common *vc = data;
 397
 398         KKASSERT(ve->refs == 0);
 399         vmspace_entry_delete(ve, vc);
 400         return(0);
 401 }
 402
 403 /*
 404  * Remove a vmspace_entry from the RB tree and destroy it.  We have to clean
 405  * up the pmap, the vm_map, then destroy the vmspace.
 406  */
 407 static
 408 void
 409 vmspace_entry_delete(struct vmspace_entry *ve, struct vkernel_common *vc)
 410 {
 411         RB_REMOVE(vmspace_rb_tree, &vc->vc_root, ve);
 412
 413         pmap_remove_pages(vmspace_pmap(ve->vmspace),
 414                           VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 415         vm_map_remove(&ve->vmspace->vm_map,
 416                       VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 417         sysref_put(&ve->vmspace->vm_sysref);
 418         kfree(ve, M_VKERNEL);
 419 }
 420
 421
 422 static
 423 struct vmspace_entry *
 424 vkernel_find_vmspace(struct vkernel_common *vc, void *id)
 425 {
 426         struct vmspace_entry *ve;
 427         struct vmspace_entry key;
 428
 429         key.id = id;
 430         ve = RB_FIND(vmspace_rb_tree, &vc->vc_root, &key);
 431         return (ve);
 432 }
 433
 434 /*
 435  * Manage vkernel refs, used by the kernel when fork()ing or exit()ing
 436  * a vkernel process.
 437  */
 438 void
 439 vkernel_inherit(struct proc *p1, struct proc *p2)
 440 {
 441         struct vkernel_common *vc;
 442         struct vkernel *vk;
 443
 444         vk = p1->p_vkernel;
 445         vc = vk->vk_common;
 446         KKASSERT(vc->vc_refs > 0);
 447         atomic_add_int(&vc->vc_refs, 1);
 448         vk = kmalloc(sizeof(*vk), M_VKERNEL, M_WAITOK|M_ZERO);
 449         p2->p_vkernel = vk;
 450         vk->vk_common = vc;
 451 }
 452
 453 void
 454 vkernel_exit(struct proc *p)
 455 {
 456         struct vkernel_common *vc;
 457         struct vmspace_entry *ve;
 458         struct vkernel *vk;
 459         int freeme = 0;
 460
 461         vk = p->p_vkernel;
 462         p->p_vkernel = NULL;
 463         vc = vk->vk_common;
 464         vk->vk_common = NULL;
 465
 466         /*
 467          * Restore the original VM context if we are killed while running
 468          * a different one.
 469          *
 470          * This isn't supposed to happen.  What is supposed to happen is
 471          * that the process should enter vkernel_trap() before the handling
 472          * the signal.
 473          */
 474         if ((ve = vk->vk_current) != NULL) {
 475                 kprintf("Killed with active VC, notify kernel list\n");
 476 #ifdef DDB
 477                 db_print_backtrace();
 478 #endif
 479                 vk->vk_current = NULL;
 480                 pmap_replacevm(p, vk->vk_save_vmspace, 0);
 481                 vk->vk_save_vmspace = NULL;
 482                 KKASSERT(ve->refs > 0);
 483                 --ve->refs;
 484         }
 485
 486         /*
 487          * Dereference the common area
 488          */
 489         KKASSERT(vc->vc_refs > 0);
 490         spin_lock_wr(&vc->vc_spin);
 491         if (--vc->vc_refs == 0)
 492                 freeme = 1;
 493         spin_unlock_wr(&vc->vc_spin);
 494
 495         if (freeme) {
 496                 RB_SCAN(vmspace_rb_tree, &vc->vc_root, NULL,
 497                         rb_vmspace_delete, vc);
 498                 kfree(vc, M_VKERNEL);
 499         }
 500         kfree(vk, M_VKERNEL);
 501 }
 502
 503 /*
 504  * A VM space under virtual kernel control trapped out or made a system call
 505  * or otherwise needs to return control to the virtual kernel context.
 506  */
 507 int
 508 vkernel_trap(struct proc *p, struct trapframe *frame)
 509 {
 510         struct vmspace_entry *ve;
 511         struct vkernel *vk;
 512         int error;
 513
 514         /*
 515          * Which vmspace entry was running?
 516          */
 517         vk = p->p_vkernel;
 518         ve = vk->vk_current;
 519         vk->vk_current = NULL;
 520         KKASSERT(ve != NULL);
 521
 522         /*
 523          * Switch the process context back to the virtual kernel's VM space.
 524          */
 525         pmap_replacevm(p, vk->vk_save_vmspace, 0);
 526         vk->vk_save_vmspace = NULL;
 527         KKASSERT(ve->refs > 0);
 528         --ve->refs;
 529
 530         /*
 531          * Copy the emulated process frame to the virtual kernel process.
 532          * The emulated process cannot change TLS descriptors so don't
 533          * bother saving them, we already have a copy.
 534          *
 535          * Restore the virtual kernel's saved context so the virtual kernel
 536          * process can resume.
 537          */
 538         error = copyout(frame, vk->vk_user_trapframe, sizeof(*frame));
 539         bcopy(&vk->vk_save_trapframe, frame, sizeof(*frame));
 540         bcopy(&vk->vk_save_vextframe.vx_tls, &curthread->td_tls,
 541               sizeof(vk->vk_save_vextframe.vx_tls));
 542         set_user_TLS();
 543         return(error);
 544 }
 545