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