bsd/kern/kern_core.c

   1 /*
   2  * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /* Copyright (c) 1991 NeXT Computer, Inc.  All rights reserved.
  29  *
  30  *      File:   bsd/kern/kern_core.c
  31  *
  32  *      This file contains machine independent code for performing core dumps.
  33  *
  34  */
  35 #if CONFIG_COREDUMP
  36
  37 #include <mach/vm_param.h>
  38 #include <mach/thread_status.h>
  39 #include <sys/content_protection.h>
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/signalvar.h>
  43 #include <sys/resourcevar.h>
  44 #include <sys/namei.h>
  45 #include <sys/vnode_internal.h>
  46 #include <sys/proc_internal.h>
  47 #include <sys/kauth.h>
  48 #include <sys/timeb.h>
  49 #include <sys/times.h>
  50 #include <sys/acct.h>
  51 #include <sys/file_internal.h>
  52 #include <sys/uio.h>
  53 #include <sys/kernel.h>
  54 #include <sys/stat.h>
  55
  56 #include <mach-o/loader.h>
  57 #include <mach/vm_region.h>
  58 #include <mach/vm_statistics.h>
  59
  60 #include <vm/vm_kern.h>
  61 #include <vm/vm_protos.h> /* last */
  62 #include <vm/vm_map.h>          /* current_map() */
  63 #include <mach/mach_vm.h>       /* mach_vm_region_recurse() */
  64 #include <mach/task.h>          /* task_suspend() */
  65 #include <kern/task.h>          /* get_task_numacts() */
  66
  67 #include <security/audit/audit.h>
  68
  69 #if CONFIG_MACF
  70 #include <security/mac_framework.h>
  71 #endif /* CONFIG_MACF */
  72
  73 typedef struct {
  74         int     flavor;                 /* the number for this flavor */
  75         mach_msg_type_number_t  count;  /* count of ints in this flavor */
  76 } mythread_state_flavor_t;
  77
  78 #if defined (__i386__) || defined (__x86_64__)
  79 mythread_state_flavor_t thread_flavor_array[] = {
  80         {x86_THREAD_STATE, x86_THREAD_STATE_COUNT},
  81         {x86_FLOAT_STATE, x86_FLOAT_STATE_COUNT},
  82         {x86_EXCEPTION_STATE, x86_EXCEPTION_STATE_COUNT},
  83 };
  84 int mynum_flavors = 3;
  85 #elif defined (__arm__)
  86 mythread_state_flavor_t thread_flavor_array[] = {
  87         {ARM_THREAD_STATE, ARM_THREAD_STATE_COUNT},
  88         {ARM_VFP_STATE, ARM_VFP_STATE_COUNT},
  89         {ARM_EXCEPTION_STATE, ARM_EXCEPTION_STATE_COUNT}
  90 };
  91 int mynum_flavors = 3;
  92
  93 #elif defined (__arm64__)
  94 mythread_state_flavor_t thread_flavor_array[] = {
  95         {ARM_THREAD_STATE64, ARM_THREAD_STATE64_COUNT},
  96         /* ARM64_TODO: VFP */
  97         {ARM_EXCEPTION_STATE64, ARM_EXCEPTION_STATE64_COUNT}
  98 };
  99 int mynum_flavors = 2;
 100 #else
 101 #error architecture not supported
 102 #endif
 103
 104
 105 typedef struct {
 106         vm_offset_t header;
 107         size_t hoffset;
 108         mythread_state_flavor_t *flavors;
 109         size_t tstate_size;
 110         size_t flavor_count;
 111 } tir_t;
 112
 113 extern int freespace_mb(vnode_t vp);
 114
 115 /* XXX not in a Mach header anywhere */
 116 kern_return_t thread_getstatus(thread_t act, int flavor,
 117     thread_state_t tstate, mach_msg_type_number_t *count);
 118 void task_act_iterate_wth_args(task_t, void (*)(thread_t, void *), void *);
 119
 120 #ifdef SECURE_KERNEL
 121 __XNU_PRIVATE_EXTERN int do_coredump = 0;       /* default: don't dump cores */
 122 #else
 123 __XNU_PRIVATE_EXTERN int do_coredump = 1;       /* default: dump cores */
 124 #endif
 125 __XNU_PRIVATE_EXTERN int sugid_coredump = 0; /* default: but not SGUID binaries */
 126
 127
 128 /* cpu_type returns only the most generic indication of the current CPU. */
 129 /* in a core we want to know the kind of process. */
 130
 131 static cpu_type_t
 132 process_cpu_type(proc_t core_proc)
 133 {
 134         cpu_type_t what_we_think;
 135 #if defined (__i386__) || defined (__x86_64__)
 136         if (IS_64BIT_PROCESS(core_proc)) {
 137                 what_we_think = CPU_TYPE_X86_64;
 138         } else {
 139                 what_we_think = CPU_TYPE_I386;
 140         }
 141 #elif defined (__arm__) || defined(__arm64__)
 142         if (IS_64BIT_PROCESS(core_proc)) {
 143                 what_we_think = CPU_TYPE_ARM64;
 144         } else {
 145                 what_we_think = CPU_TYPE_ARM;
 146         }
 147 #endif
 148
 149         return what_we_think;
 150 }
 151
 152 static cpu_type_t
 153 process_cpu_subtype(proc_t core_proc)
 154 {
 155         cpu_type_t what_we_think;
 156 #if defined (__i386__) || defined (__x86_64__)
 157         if (IS_64BIT_PROCESS(core_proc)) {
 158                 what_we_think = CPU_SUBTYPE_X86_64_ALL;
 159         } else {
 160                 what_we_think = CPU_SUBTYPE_I386_ALL;
 161         }
 162 #elif defined (__arm__) || defined(__arm64__)
 163         if (IS_64BIT_PROCESS(core_proc)) {
 164                 what_we_think = CPU_SUBTYPE_ARM64_ALL;
 165         } else {
 166                 what_we_think = CPU_SUBTYPE_ARM_ALL;
 167         }
 168 #endif
 169         return what_we_think;
 170 }
 171
 172 static void
 173 collectth_state(thread_t th_act, void *tirp)
 174 {
 175         vm_offset_t     header;
 176         size_t  hoffset, i;
 177         mythread_state_flavor_t *flavors;
 178         struct thread_command   *tc;
 179         tir_t *t = (tir_t *)tirp;
 180
 181         /*
 182          *      Fill in thread command structure.
 183          */
 184         header = t->header;
 185         hoffset = t->hoffset;
 186         flavors = t->flavors;
 187
 188         tc = (struct thread_command *) (header + hoffset);
 189         tc->cmd = LC_THREAD;
 190         tc->cmdsize = (uint32_t)(sizeof(struct thread_command)
 191             + t->tstate_size);
 192         hoffset += sizeof(struct thread_command);
 193         /*
 194          * Follow with a struct thread_state_flavor and
 195          * the appropriate thread state struct for each
 196          * thread state flavor.
 197          */
 198         for (i = 0; i < t->flavor_count; i++) {
 199                 *(mythread_state_flavor_t *)(header + hoffset) =
 200                     flavors[i];
 201                 hoffset += sizeof(mythread_state_flavor_t);
 202                 thread_getstatus(th_act, flavors[i].flavor,
 203                     (thread_state_t)(header + hoffset),
 204                     &flavors[i].count);
 205                 hoffset += flavors[i].count * sizeof(int);
 206         }
 207
 208         t->hoffset = hoffset;
 209 }
 210
 211 /*
 212  * coredump
 213  *
 214  * Description: Create a core image on the file "core" for the process
 215  *              indicated
 216  *
 217  * Parameters:  core_proc                       Process to dump core [*]
 218  *                              reserve_mb                      If non-zero, leave filesystem with
 219  *                                                                      at least this much free space.
 220  *                              coredump_flags  Extra options (ignore rlimit, run fsync)
 221  *
 222  * Returns:     0                               Success
 223  *              !0                              Failure errno
 224  *
 225  * IMPORTANT:   This function can only be called on the current process, due
 226  *              to assumptions below; see variable declaration section for
 227  *              details.
 228  */
 229 #define MAX_TSTATE_FLAVORS      10
 230 int
 231 coredump(proc_t core_proc, uint32_t reserve_mb, int coredump_flags)
 232 {
 233 /* Begin assumptions that limit us to only the current process */
 234         vfs_context_t ctx = vfs_context_current();
 235         vm_map_t        map = current_map();
 236         task_t          task = current_task();
 237 /* End assumptions */
 238         kauth_cred_t cred = vfs_context_ucred(ctx);
 239         int error = 0;
 240         struct vnode_attr va;
 241         size_t          thread_count, segment_count;
 242         size_t          command_size, header_size, tstate_size;
 243         size_t          hoffset;
 244         off_t           foffset;
 245         mach_vm_offset_t vmoffset;
 246         vm_offset_t     header;
 247         mach_vm_size_t  vmsize;
 248         vm_prot_t       prot;
 249         vm_prot_t       maxprot;
 250         vm_inherit_t    inherit;
 251         int             error1 = 0;
 252         char            stack_name[MAXCOMLEN + 6];
 253         char            *alloced_name = NULL;
 254         char            *name = NULL;
 255         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 256         vm_size_t       mapsize;
 257         size_t          i;
 258         uint32_t nesting_depth = 0;
 259         kern_return_t   kret;
 260         struct vm_region_submap_info_64 vbr;
 261         mach_msg_type_number_t vbrcount = 0;
 262         tir_t tir1;
 263         struct vnode * vp;
 264         struct mach_header      *mh = NULL;     /* protected by is_64 */
 265         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 266         int             is_64 = 0;
 267         size_t          mach_header_sz = sizeof(struct mach_header);
 268         size_t          segment_command_sz = sizeof(struct segment_command);
 269
 270         if (current_proc() != core_proc) {
 271                 panic("coredump() called against proc that is not current_proc: %p", core_proc);
 272         }
 273
 274         if (do_coredump == 0 ||         /* Not dumping at all */
 275             ((sugid_coredump == 0) &&   /* Not dumping SUID/SGID binaries */
 276             ((kauth_cred_getsvuid(cred) != kauth_cred_getruid(cred)) ||
 277             (kauth_cred_getsvgid(cred) != kauth_cred_getrgid(cred))))) {
 278                 error = EFAULT;
 279                 goto out2;
 280         }
 281
 282 #if CONFIG_MACF
 283         error = mac_proc_check_dump_core(core_proc);
 284         if (error != 0) {
 285                 goto out2;
 286         }
 287 #endif
 288
 289         if (IS_64BIT_PROCESS(core_proc)) {
 290                 is_64 = 1;
 291                 mach_header_sz = sizeof(struct mach_header_64);
 292                 segment_command_sz = sizeof(struct segment_command_64);
 293         }
 294
 295         mapsize = get_vmmap_size(map);
 296
 297         if (((coredump_flags & COREDUMP_IGNORE_ULIMIT) == 0) &&
 298             (mapsize >= proc_limitgetcur(core_proc, RLIMIT_CORE, FALSE))) {
 299                 error = EFAULT;
 300                 goto out2;
 301         }
 302
 303         (void) task_suspend_internal(task);
 304
 305         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 306
 307         /* create name according to sysctl'able format string */
 308         /* if name creation fails, fall back to historical behaviour... */
 309         if (alloced_name == NULL ||
 310             proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 311             core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 312                 snprintf(stack_name, sizeof(stack_name),
 313                     "/cores/core.%d", core_proc->p_pid);
 314                 name = stack_name;
 315         } else {
 316                 name = alloced_name;
 317         }
 318
 319         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx))) {
 320                 goto out2;
 321         }
 322
 323         VATTR_INIT(&va);
 324         VATTR_WANTED(&va, va_nlink);
 325         /* Don't dump to non-regular files or files with links. */
 326         if (vp->v_type != VREG ||
 327             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 328                 error = EFAULT;
 329                 goto out;
 330         }
 331
 332         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 333         VATTR_SET(&va, va_data_size, 0);
 334         if (core_proc == initproc) {
 335                 VATTR_SET(&va, va_dataprotect_class, PROTECTION_CLASS_D);
 336         }
 337         vnode_setattr(vp, &va, ctx);
 338         core_proc->p_acflag |= ACORE;
 339
 340         if ((reserve_mb > 0) &&
 341             ((freespace_mb(vp) - (mapsize >> 20)) < reserve_mb)) {
 342                 error = ENOSPC;
 343                 goto out;
 344         }
 345
 346         /*
 347          *      If the task is modified while dumping the file
 348          *      (e.g., changes in threads or VM, the resulting
 349          *      file will not necessarily be correct.
 350          */
 351
 352         thread_count = get_task_numacts(task);
 353         segment_count = get_vmmap_entries(map); /* XXX */
 354         tir1.flavor_count = sizeof(thread_flavor_array) / sizeof(mythread_state_flavor_t);
 355         bcopy(thread_flavor_array, flavors, sizeof(thread_flavor_array));
 356         tstate_size = 0;
 357         for (i = 0; i < tir1.flavor_count; i++) {
 358                 tstate_size += sizeof(mythread_state_flavor_t) +
 359                     (flavors[i].count * sizeof(int));
 360         }
 361
 362         {
 363                 size_t lhs;
 364                 size_t rhs;
 365
 366                 /* lhs = segment_count * segment_command_sz */
 367                 if (os_mul_overflow(segment_count, segment_command_sz, &lhs)) {
 368                         error = ENOMEM;
 369                         goto out;
 370                 }
 371
 372                 /* rhs = (tstate_size + sizeof(struct thread_command)) * thread_count */
 373                 if (os_add_and_mul_overflow(tstate_size, sizeof(struct thread_command), thread_count, &rhs)) {
 374                         error = ENOMEM;
 375                         goto out;
 376                 }
 377
 378                 /* command_size = lhs + rhs */
 379                 if (os_add_overflow(lhs, rhs, &command_size)) {
 380                         error = ENOMEM;
 381                         goto out;
 382                 }
 383         }
 384
 385         if (os_add_overflow(command_size, mach_header_sz, &header_size)) {
 386                 error = ENOMEM;
 387                 goto out;
 388         }
 389
 390         if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
 391                 error = ENOMEM;
 392                 goto out;
 393         }
 394
 395         /*
 396          *      Set up Mach-O header.
 397          */
 398         if (is_64) {
 399                 mh64 = (struct mach_header_64 *)header;
 400                 mh64->magic = MH_MAGIC_64;
 401                 mh64->cputype = process_cpu_type(core_proc);
 402                 mh64->cpusubtype = process_cpu_subtype(core_proc);
 403                 mh64->filetype = MH_CORE;
 404                 mh64->ncmds = (uint32_t)(segment_count + thread_count);
 405                 mh64->sizeofcmds = (uint32_t)command_size;
 406                 mh64->reserved = 0;             /* 8 byte alignment */
 407         } else {
 408                 mh = (struct mach_header *)header;
 409                 mh->magic = MH_MAGIC;
 410                 mh->cputype = process_cpu_type(core_proc);
 411                 mh->cpusubtype = process_cpu_subtype(core_proc);
 412                 mh->filetype = MH_CORE;
 413                 mh->ncmds = (uint32_t)(segment_count + thread_count);
 414                 mh->sizeofcmds = (uint32_t)command_size;
 415         }
 416
 417         hoffset = mach_header_sz;       /* offset into header */
 418         foffset = round_page(header_size);      /* offset into file */
 419         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 420
 421         /*
 422          * We use to check for an error, here, now we try and get
 423          * as much as we can
 424          */
 425         while (segment_count > 0) {
 426                 struct segment_command          *sc;
 427                 struct segment_command_64       *sc64;
 428
 429                 /*
 430                  *      Get region information for next region.
 431                  */
 432
 433                 while (1) {
 434                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 435                         if ((kret = mach_vm_region_recurse(map,
 436                             &vmoffset, &vmsize, &nesting_depth,
 437                             (vm_region_recurse_info_t)&vbr,
 438                             &vbrcount)) != KERN_SUCCESS) {
 439                                 break;
 440                         }
 441                         /*
 442                          * If we get a valid mapping back, but we're dumping
 443                          * a 32 bit process,  and it's over the allowable
 444                          * address space of a 32 bit process, it's the same
 445                          * as if mach_vm_region_recurse() failed.
 446                          */
 447                         if (!(is_64) &&
 448                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 449                                 kret = KERN_INVALID_ADDRESS;
 450                                 break;
 451                         }
 452                         if (vbr.is_submap) {
 453                                 nesting_depth++;
 454                                 continue;
 455                         } else {
 456                                 break;
 457                         }
 458                 }
 459                 if (kret != KERN_SUCCESS) {
 460                         break;
 461                 }
 462
 463                 prot = vbr.protection;
 464                 maxprot = vbr.max_protection;
 465                 inherit = vbr.inheritance;
 466                 /*
 467                  *      Fill in segment command structure.
 468                  */
 469                 if (is_64) {
 470                         sc64 = (struct segment_command_64 *)(header + hoffset);
 471                         sc64->cmd = LC_SEGMENT_64;
 472                         sc64->cmdsize = sizeof(struct segment_command_64);
 473                         /* segment name is zeroed by kmem_alloc */
 474                         sc64->segname[0] = 0;
 475                         sc64->vmaddr = vmoffset;
 476                         sc64->vmsize = vmsize;
 477                         sc64->fileoff = foffset;
 478                         sc64->filesize = vmsize;
 479                         sc64->maxprot = maxprot;
 480                         sc64->initprot = prot;
 481                         sc64->nsects = 0;
 482                         sc64->flags = 0;
 483                 } else {
 484                         sc = (struct segment_command *) (header + hoffset);
 485                         sc->cmd = LC_SEGMENT;
 486                         sc->cmdsize = sizeof(struct segment_command);
 487                         /* segment name is zeroed by kmem_alloc */
 488                         sc->segname[0] = 0;
 489                         sc->vmaddr = CAST_DOWN_EXPLICIT(uint32_t, vmoffset);
 490                         sc->vmsize = CAST_DOWN_EXPLICIT(uint32_t, vmsize);
 491                         sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t, foffset); /* will never truncate */
 492                         sc->filesize = CAST_DOWN_EXPLICIT(uint32_t, vmsize); /* will never truncate */
 493                         sc->maxprot = maxprot;
 494                         sc->initprot = prot;
 495                         sc->nsects = 0;
 496                         sc->flags = 0;
 497                 }
 498
 499                 /*
 500                  *      Write segment out.  Try as hard as possible to
 501                  *      get read access to the data.
 502                  */
 503                 if ((prot & VM_PROT_READ) == 0) {
 504                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 505                             prot | VM_PROT_READ);
 506                 }
 507                 /*
 508                  *      Only actually perform write if we can read.
 509                  *      Note: if we can't read, then we end up with
 510                  *      a hole in the file.
 511                  */
 512                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 513                     && vbr.user_tag != VM_MEMORY_IOKIT
 514                     && coredumpok(map, vmoffset)) {
 515                         error = vn_rdwr_64(UIO_WRITE, vp, vmoffset, vmsize, foffset,
 516                             (IS_64BIT_PROCESS(core_proc) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 517                             IO_NOCACHE | IO_NODELOCKED | IO_UNIT, cred, (int64_t *) 0, core_proc);
 518                 }
 519
 520                 hoffset += segment_command_sz;
 521                 foffset += vmsize;
 522                 vmoffset += vmsize;
 523                 segment_count--;
 524         }
 525
 526         /*
 527          * If there are remaining segments which have not been written
 528          * out because break in the loop above, then they were not counted
 529          * because they exceed the real address space of the executable
 530          * type: remove them from the header's count.  This is OK, since
 531          * we are allowed to have a sparse area following the segments.
 532          */
 533         if (is_64) {
 534                 mh64->ncmds -= segment_count;
 535                 mh64->sizeofcmds -= segment_count * segment_command_sz;
 536         } else {
 537                 mh->ncmds -= segment_count;
 538                 mh->sizeofcmds -= segment_count * segment_command_sz;
 539         }
 540
 541         tir1.header = header;
 542         tir1.hoffset = hoffset;
 543         tir1.flavors = flavors;
 544         tir1.tstate_size = tstate_size;
 545         task_act_iterate_wth_args(task, collectth_state, &tir1);
 546
 547         /*
 548          *      Write out the Mach header at the beginning of the
 549          *      file.  OK to use a 32 bit write for this.
 550          */
 551         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, (int)MIN(header_size, INT_MAX), (off_t)0,
 552             UIO_SYSSPACE, IO_NOCACHE | IO_NODELOCKED | IO_UNIT, cred, (int *) 0, core_proc);
 553         kmem_free(kernel_map, header, header_size);
 554
 555         if ((coredump_flags & COREDUMP_FULLFSYNC) && error == 0) {
 556                 error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
 557         }
 558 out:
 559         error1 = vnode_close(vp, FWRITE, ctx);
 560 out2:
 561 #if CONFIG_AUDIT
 562         audit_proc_coredump(core_proc, name, error);
 563 #endif
 564         if (alloced_name != NULL) {
 565                 FREE(alloced_name, M_TEMP);
 566         }
 567         if (error == 0) {
 568                 error = error1;
 569         }
 570
 571         return error;
 572 }
 573
 574 #else /* CONFIG_COREDUMP */
 575
 576 /* When core dumps aren't needed, no need to compile this file at all */
 577
 578 #error assertion failed: this section is not compiled
 579
 580 #endif /* CONFIG_COREDUMP */