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