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 1.14 2007/08/15 03:15:07 dillon Exp $
  35  */
  36
  37 #include <sys/param.h>
  38 #include <sys/kernel.h>
  39 #include <sys/systm.h>
  40 #include <sys/sysproto.h>
  41 #include <sys/kern_syscall.h>
  42 #include <sys/mman.h>
  43 #include <sys/proc.h>
  44 #include <sys/malloc.h>
  45 #include <sys/sysctl.h>
  46 #include <sys/vkernel.h>
  47 #include <sys/vmspace.h>
  48
  49 #include <vm/vm_extern.h>
  50 #include <vm/pmap.h>
  51
  52 #include <machine/vmparam.h>
  53
  54 #include <sys/spinlock2.h>
  55 #include <sys/sysref2.h>
  56 #include <sys/mplock2.h>
  57
  58 static struct vmspace_entry *vkernel_find_vmspace(struct vkernel_proc *vkp,
  59                                                   void *id);
  60 static void vmspace_entry_delete(struct vmspace_entry *ve,
  61                                  struct vkernel_proc *vkp);
  62
  63 static MALLOC_DEFINE(M_VKERNEL, "vkernel", "VKernel structures");
  64
  65 /*
  66  * vmspace_create (void *id, int type, void *data)
  67  *
  68  * Create a VMSPACE under the control of the caller with the specified id.
  69  * An id of NULL cannot be used.  The type and data fields must currently
  70  * be 0.
  71  *
  72  * The vmspace starts out completely empty.  Memory may be mapped into the
  73  * VMSPACE with vmspace_mmap() and MAP_VPAGETABLE section(s) controlled
  74  * with vmspace_mcontrol().
  75  *
  76  * MPALMOSTSAFE
  77  */
  78 int
  79 sys_vmspace_create(struct vmspace_create_args *uap)
  80 {
  81         struct vmspace_entry *ve;
  82         struct vkernel_proc *vkp;
  83         int error;
  84
  85         if (vkernel_enable == 0)
  86                 return (EOPNOTSUPP);
  87
  88         /*
  89          * Create a virtual kernel side-structure for the process if one
  90          * does not exist.
  91          */
  92         get_mplock();
  93         if ((vkp = curproc->p_vkernel) == NULL) {
  94                 vkp = kmalloc(sizeof(*vkp), M_VKERNEL, M_WAITOK|M_ZERO);
  95                 vkp->refs = 1;
  96                 spin_init(&vkp->spin);
  97                 RB_INIT(&vkp->root);
  98                 curproc->p_vkernel = vkp;
  99         }
 100
 101         /*
 102          * Create a new VMSPACE
 103          *
 104          * XXX race if kmalloc blocks
 105          */
 106         if (vkernel_find_vmspace(vkp, uap->id)) {
 107                 error = EEXIST;
 108                 goto done;
 109         }
 110         ve = kmalloc(sizeof(struct vmspace_entry), M_VKERNEL, M_WAITOK|M_ZERO);
 111         ve->vmspace = vmspace_alloc(VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 112         ve->id = uap->id;
 113         pmap_pinit2(vmspace_pmap(ve->vmspace));
 114         RB_INSERT(vmspace_rb_tree, &vkp->root, ve);
 115         error = 0;
 116 done:
 117         rel_mplock();
 118         return (error);
 119 }
 120
 121 /*
 122  * vmspace_destroy (void *id)
 123  *
 124  * Destroy a VMSPACE.
 125  *
 126  * MPALMOSTSAFE
 127  */
 128 int
 129 sys_vmspace_destroy(struct vmspace_destroy_args *uap)
 130 {
 131         struct vkernel_proc *vkp;
 132         struct vmspace_entry *ve;
 133         int error;
 134
 135         get_mplock();
 136         if ((vkp = curproc->p_vkernel) == NULL) {
 137                 error = EINVAL;
 138                 goto done;
 139         }
 140         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 141                 error = ENOENT;
 142                 goto done;
 143         }
 144         if (ve->refs) {
 145                 error = EBUSY;
 146                 goto done;
 147         }
 148         vmspace_entry_delete(ve, vkp);
 149         error = 0;
 150 done:
 151         rel_mplock();
 152         return(error);
 153 }
 154
 155 /*
 156  * vmspace_ctl (void *id, int cmd, struct trapframe *tframe,
 157  *              struct vextframe *vframe);
 158  *
 159  * Transfer control to a VMSPACE.  Control is returned after the specified
 160  * number of microseconds or if a page fault, signal, trap, or system call
 161  * occurs.  The context is updated as appropriate.
 162  *
 163  * MPALMOSTSAFE
 164  */
 165 int
 166 sys_vmspace_ctl(struct vmspace_ctl_args *uap)
 167 {
 168         struct vkernel_proc *vkp;
 169         struct vkernel_lwp *vklp;
 170         struct vmspace_entry *ve;
 171         struct lwp *lp;
 172         struct proc *p;
 173         int framesz;
 174         int error;
 175
 176         lp = curthread->td_lwp;
 177         p = lp->lwp_proc;
 178
 179         get_mplock();
 180         if ((vkp = p->p_vkernel) == NULL) {
 181                 error = EINVAL;
 182                 goto done;
 183         }
 184         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 185                 error = ENOENT;
 186                 goto done;
 187         }
 188
 189         /*
 190          * Signal mailbox interlock
 191          */
 192         if (p->p_flag & P_MAILBOX) {
 193                 p->p_flag &= ~P_MAILBOX;
 194                 error = EINTR;
 195                 goto done;
 196         }
 197
 198         switch(uap->cmd) {
 199         case VMSPACE_CTL_RUN:
 200                 /*
 201                  * Save the caller's register context, swap VM spaces, and
 202                  * install the passed register context.  Return with
 203                  * EJUSTRETURN so the syscall code doesn't adjust the context.
 204                  */
 205                 atomic_add_int(&ve->refs, 1);
 206                 framesz = sizeof(struct trapframe);
 207                 if ((vklp = lp->lwp_vkernel) == NULL) {
 208                         vklp = kmalloc(sizeof(*vklp), M_VKERNEL,
 209                                        M_WAITOK|M_ZERO);
 210                         lp->lwp_vkernel = vklp;
 211                 }
 212                 vklp->user_trapframe = uap->tframe;
 213                 vklp->user_vextframe = uap->vframe;
 214                 bcopy(uap->sysmsg_frame, &vklp->save_trapframe, framesz);
 215                 bcopy(&curthread->td_tls, &vklp->save_vextframe.vx_tls,
 216                       sizeof(vklp->save_vextframe.vx_tls));
 217                 error = copyin(uap->tframe, uap->sysmsg_frame, framesz);
 218                 if (error == 0)
 219                         error = copyin(&uap->vframe->vx_tls, &curthread->td_tls, sizeof(struct savetls));
 220                 if (error == 0)
 221                         error = cpu_sanitize_frame(uap->sysmsg_frame);
 222                 if (error == 0)
 223                         error = cpu_sanitize_tls(&curthread->td_tls);
 224                 if (error) {
 225                         bcopy(&vklp->save_trapframe, uap->sysmsg_frame, framesz);
 226                         bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
 227                               sizeof(vklp->save_vextframe.vx_tls));
 228                         set_user_TLS();
 229                         atomic_subtract_int(&ve->refs, 1);
 230                 } else {
 231                         vklp->ve = ve;
 232                         pmap_setlwpvm(lp, ve->vmspace);
 233                         set_user_TLS();
 234                         set_vkernel_fp(uap->sysmsg_frame);
 235                         error = EJUSTRETURN;
 236                 }
 237                 break;
 238         default:
 239                 error = EOPNOTSUPP;
 240                 break;
 241         }
 242 done:
 243         rel_mplock();
 244         return(error);
 245 }
 246
 247 /*
 248  * vmspace_mmap(id, addr, len, prot, flags, fd, offset)
 249  *
 250  * map memory within a VMSPACE.  This function is just like a normal mmap()
 251  * but operates on the vmspace's memory map.  Most callers use this to create
 252  * a MAP_VPAGETABLE mapping.
 253  *
 254  * MPALMOSTSAFE
 255  */
 256 int
 257 sys_vmspace_mmap(struct vmspace_mmap_args *uap)
 258 {
 259         struct vkernel_proc *vkp;
 260         struct vmspace_entry *ve;
 261         int error;
 262
 263         get_mplock();
 264         if ((vkp = curproc->p_vkernel) == NULL) {
 265                 error = EINVAL;
 266                 goto done;
 267         }
 268         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 269                 error = ENOENT;
 270                 goto done;
 271         }
 272         error = kern_mmap(ve->vmspace, uap->addr, uap->len,
 273                           uap->prot, uap->flags,
 274                           uap->fd, uap->offset, &uap->sysmsg_resultp);
 275 done:
 276         rel_mplock();
 277         return (error);
 278 }
 279
 280 /*
 281  * vmspace_munmap(id, addr, len)
 282  *
 283  * unmap memory within a VMSPACE.
 284  *
 285  * MPALMOSTSAFE
 286  */
 287 int
 288 sys_vmspace_munmap(struct vmspace_munmap_args *uap)
 289 {
 290         struct vkernel_proc *vkp;
 291         struct vmspace_entry *ve;
 292         vm_offset_t addr;
 293         vm_offset_t tmpaddr;
 294         vm_size_t size, pageoff;
 295         vm_map_t map;
 296         int error;
 297
 298         get_mplock();
 299         if ((vkp = curproc->p_vkernel) == NULL) {
 300                 error = EINVAL;
 301                 goto done;
 302         }
 303         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 304                 error = ENOENT;
 305                 goto done;
 306         }
 307
 308         /*
 309          * Copied from sys_munmap()
 310          */
 311         addr = (vm_offset_t)uap->addr;
 312         size = uap->len;
 313
 314         pageoff = (addr & PAGE_MASK);
 315         addr -= pageoff;
 316         size += pageoff;
 317         size = (vm_size_t)round_page(size);
 318         if (size < uap->len) {          /* wrap */
 319                 error = EINVAL;
 320                 goto done;
 321         }
 322         tmpaddr = addr + size;          /* workaround gcc4 opt */
 323         if (tmpaddr < addr) {           /* wrap */
 324                 error = EINVAL;
 325                 goto done;
 326         }
 327         if (size == 0) {
 328                 error = 0;
 329                 goto done;
 330         }
 331
 332         if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
 333                 error = EINVAL;
 334                 goto done;
 335         }
 336         if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) {
 337                 error = EINVAL;
 338                 goto done;
 339         }
 340         map = &ve->vmspace->vm_map;
 341         if (!vm_map_check_protection(map, addr, tmpaddr, VM_PROT_NONE)) {
 342                 error = EINVAL;
 343                 goto done;
 344         }
 345         vm_map_remove(map, addr, addr + size);
 346         error = 0;
 347 done:
 348         rel_mplock();
 349         return (error);
 350 }
 351
 352 /*
 353  * vmspace_pread(id, buf, nbyte, flags, offset)
 354  *
 355  * Read data from a vmspace.  The number of bytes read is returned or
 356  * -1 if an unrecoverable error occured.  If the number of bytes read is
 357  * less then the request size, a page fault occured in the VMSPACE which
 358  * the caller must resolve in order to proceed.
 359  *
 360  * (not implemented yet)
 361  *
 362  * MPALMOSTSAFE
 363  */
 364 int
 365 sys_vmspace_pread(struct vmspace_pread_args *uap)
 366 {
 367         struct vkernel_proc *vkp;
 368         struct vmspace_entry *ve;
 369         int error;
 370
 371         get_mplock();
 372         if ((vkp = curproc->p_vkernel) == NULL) {
 373                 error = EINVAL;
 374                 goto done;
 375         }
 376         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 377                 error = ENOENT;
 378                 goto done;
 379         }
 380         error = EINVAL;
 381 done:
 382         rel_mplock();
 383         return (error);
 384 }
 385
 386 /*
 387  * vmspace_pwrite(id, buf, nbyte, flags, offset)
 388  *
 389  * Write data to a vmspace.  The number of bytes written is returned or
 390  * -1 if an unrecoverable error occured.  If the number of bytes written is
 391  * less then the request size, a page fault occured in the VMSPACE which
 392  * the caller must resolve in order to proceed.
 393  *
 394  * (not implemented yet)
 395  *
 396  * MPALMOSTSAFE
 397  */
 398 int
 399 sys_vmspace_pwrite(struct vmspace_pwrite_args *uap)
 400 {
 401         struct vkernel_proc *vkp;
 402         struct vmspace_entry *ve;
 403         int error;
 404
 405         get_mplock();
 406         if ((vkp = curproc->p_vkernel) == NULL) {
 407                 error = EINVAL;
 408                 goto done;
 409         }
 410         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 411                 error = ENOENT;
 412                 goto done;
 413         }
 414         error = EINVAL;
 415 done:
 416         rel_mplock();
 417         return (error);
 418 }
 419
 420 /*
 421  * vmspace_mcontrol(id, addr, len, behav, value)
 422  *
 423  * madvise/mcontrol support for a vmspace.
 424  *
 425  * MPALMOSTSAFE
 426  */
 427 int
 428 sys_vmspace_mcontrol(struct vmspace_mcontrol_args *uap)
 429 {
 430         struct vkernel_proc *vkp;
 431         struct vmspace_entry *ve;
 432         vm_offset_t start, end;
 433         vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
 434         int error;
 435
 436         get_mplock();
 437         if ((vkp = curproc->p_vkernel) == NULL) {
 438                 error = EINVAL;
 439                 goto done;
 440         }
 441         if ((ve = vkernel_find_vmspace(vkp, uap->id)) == NULL) {
 442                 error = ENOENT;
 443                 goto done;
 444         }
 445
 446         /*
 447          * This code is basically copied from sys_mcontrol()
 448          */
 449         if (uap->behav < 0 || uap->behav > MADV_CONTROL_END) {
 450                 error = EINVAL;
 451                 goto done;
 452         }
 453
 454         if (tmpaddr < (vm_offset_t)uap->addr) {
 455                 error = EINVAL;
 456                 goto done;
 457         }
 458         if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) {
 459                 error = EINVAL;
 460                 goto done;
 461         }
 462         if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS) {
 463                 error = EINVAL;
 464                 goto done;
 465         }
 466
 467         start = trunc_page((vm_offset_t) uap->addr);
 468         end = round_page(tmpaddr);
 469
 470         error = vm_map_madvise(&ve->vmspace->vm_map, start, end,
 471                                 uap->behav, uap->value);
 472 done:
 473         rel_mplock();
 474         return (error);
 475 }
 476
 477 /*
 478  * Red black tree functions
 479  */
 480 static int rb_vmspace_compare(struct vmspace_entry *, struct vmspace_entry *);
 481 RB_GENERATE(vmspace_rb_tree, vmspace_entry, rb_entry, rb_vmspace_compare);
 482
 483 /* a->start is address, and the only field has to be initialized */
 484 static int
 485 rb_vmspace_compare(struct vmspace_entry *a, struct vmspace_entry *b)
 486 {
 487         if ((char *)a->id < (char *)b->id)
 488                 return(-1);
 489         else if ((char *)a->id > (char *)b->id)
 490                 return(1);
 491         return(0);
 492 }
 493
 494 static
 495 int
 496 rb_vmspace_delete(struct vmspace_entry *ve, void *data)
 497 {
 498         struct vkernel_proc *vkp = data;
 499
 500         KKASSERT(ve->refs == 0);
 501         vmspace_entry_delete(ve, vkp);
 502         return(0);
 503 }
 504
 505 /*
 506  * Remove a vmspace_entry from the RB tree and destroy it.  We have to clean
 507  * up the pmap, the vm_map, then destroy the vmspace.
 508  */
 509 static
 510 void
 511 vmspace_entry_delete(struct vmspace_entry *ve, struct vkernel_proc *vkp)
 512 {
 513         RB_REMOVE(vmspace_rb_tree, &vkp->root, ve);
 514
 515         pmap_remove_pages(vmspace_pmap(ve->vmspace),
 516                           VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 517         vm_map_remove(&ve->vmspace->vm_map,
 518                       VM_MIN_USER_ADDRESS, VM_MAX_USER_ADDRESS);
 519         sysref_put(&ve->vmspace->vm_sysref);
 520         kfree(ve, M_VKERNEL);
 521 }
 522
 523
 524 static
 525 struct vmspace_entry *
 526 vkernel_find_vmspace(struct vkernel_proc *vkp, void *id)
 527 {
 528         struct vmspace_entry *ve;
 529         struct vmspace_entry key;
 530
 531         key.id = id;
 532         ve = RB_FIND(vmspace_rb_tree, &vkp->root, &key);
 533         return (ve);
 534 }
 535
 536 /*
 537  * Manage vkernel refs, used by the kernel when fork()ing or exit()ing
 538  * a vkernel process.
 539  */
 540 void
 541 vkernel_inherit(struct proc *p1, struct proc *p2)
 542 {
 543         struct vkernel_proc *vkp;
 544
 545         vkp = p1->p_vkernel;
 546         KKASSERT(vkp->refs > 0);
 547         atomic_add_int(&vkp->refs, 1);
 548         p2->p_vkernel = vkp;
 549 }
 550
 551 void
 552 vkernel_exit(struct proc *p)
 553 {
 554         struct vkernel_proc *vkp;
 555         struct lwp *lp;
 556         int freeme = 0;
 557
 558         vkp = p->p_vkernel;
 559         /*
 560          * Restore the original VM context if we are killed while running
 561          * a different one.
 562          *
 563          * This isn't supposed to happen.  What is supposed to happen is
 564          * that the process should enter vkernel_trap() before the handling
 565          * the signal.
 566          */
 567         RB_FOREACH(lp, lwp_rb_tree, &p->p_lwp_tree) {
 568                 vkernel_lwp_exit(lp);
 569         }
 570
 571         /*
 572          * Dereference the common area
 573          */
 574         p->p_vkernel = NULL;
 575         KKASSERT(vkp->refs > 0);
 576         spin_lock_wr(&vkp->spin);
 577         if (--vkp->refs == 0)
 578                 freeme = 1;
 579         spin_unlock_wr(&vkp->spin);
 580
 581         if (freeme) {
 582                 RB_SCAN(vmspace_rb_tree, &vkp->root, NULL,
 583                         rb_vmspace_delete, vkp);
 584                 kfree(vkp, M_VKERNEL);
 585         }
 586 }
 587
 588 void
 589 vkernel_lwp_exit(struct lwp *lp)
 590 {
 591         struct vkernel_lwp *vklp;
 592         struct vmspace_entry *ve;
 593
 594         if ((vklp = lp->lwp_vkernel) != NULL) {
 595                 if ((ve = vklp->ve) != NULL) {
 596                         kprintf("Warning, pid %d killed with "
 597                                 "active VC!\n", lp->lwp_proc->p_pid);
 598                         print_backtrace();
 599                         pmap_setlwpvm(lp, lp->lwp_proc->p_vmspace);
 600                         vklp->ve = NULL;
 601                         KKASSERT(ve->refs > 0);
 602                         atomic_subtract_int(&ve->refs, 1);
 603                 }
 604                 lp->lwp_vkernel = NULL;
 605                 kfree(vklp, M_VKERNEL);
 606         }
 607 }
 608
 609 /*
 610  * A VM space under virtual kernel control trapped out or made a system call
 611  * or otherwise needs to return control to the virtual kernel context.
 612  */
 613 void
 614 vkernel_trap(struct lwp *lp, struct trapframe *frame)
 615 {
 616         struct proc *p = lp->lwp_proc;
 617         struct vmspace_entry *ve;
 618         struct vkernel_lwp *vklp;
 619         int error;
 620
 621         /*
 622          * Which vmspace entry was running?
 623          */
 624         vklp = lp->lwp_vkernel;
 625         KKASSERT(vklp);
 626         ve = vklp->ve;
 627         KKASSERT(ve != NULL);
 628
 629         /*
 630          * Switch the LWP vmspace back to the virtual kernel's VM space.
 631          */
 632         vklp->ve = NULL;
 633         pmap_setlwpvm(lp, p->p_vmspace);
 634         KKASSERT(ve->refs > 0);
 635         atomic_subtract_int(&ve->refs, 1);
 636
 637         /*
 638          * Copy the emulated process frame to the virtual kernel process.
 639          * The emulated process cannot change TLS descriptors so don't
 640          * bother saving them, we already have a copy.
 641          *
 642          * Restore the virtual kernel's saved context so the virtual kernel
 643          * process can resume.
 644          */
 645         error = copyout(frame, vklp->user_trapframe, sizeof(*frame));
 646         bcopy(&vklp->save_trapframe, frame, sizeof(*frame));
 647         bcopy(&vklp->save_vextframe.vx_tls, &curthread->td_tls,
 648               sizeof(vklp->save_vextframe.vx_tls));
 649         set_user_TLS();
 650         cpu_vkernel_trap(frame, error);
 651 }
 652