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
  36 #include <mach/vm_param.h>
  37 #include <mach/thread_status.h>
  38
  39 #include <sys/param.h>
  40 #include <sys/systm.h>
  41 #include <sys/signalvar.h>
  42 #include <sys/resourcevar.h>
  43 #include <sys/namei.h>
  44 #include <sys/vnode_internal.h>
  45 #include <sys/proc_internal.h>
  46 #include <sys/kauth.h>
  47 #include <sys/timeb.h>
  48 #include <sys/times.h>
  49 #include <sys/acct.h>
  50 #include <sys/file_internal.h>
  51 #include <sys/uio.h>
  52 #include <sys/kernel.h>
  53 #include <sys/stat.h>
  54
  55 #include <mach-o/loader.h>
  56 #include <mach/vm_region.h>
  57 #include <mach/vm_statistics.h>
  58
  59 #include <vm/vm_kern.h>
  60 #include <vm/vm_protos.h> /* last */
  61 #include <vm/vm_map.h>          /* current_map() */
  62 #include <mach/mach_vm.h>       /* mach_vm_region_recurse() */
  63 #include <mach/task.h>          /* task_suspend() */
  64 #include <kern/task.h>          /* get_task_numacts() */
  65
  66 #include <security/audit/audit.h>
  67
  68 typedef struct {
  69         int     flavor;                 /* the number for this flavor */
  70         mach_msg_type_number_t  count;  /* count of ints in this flavor */
  71 } mythread_state_flavor_t;
  72
  73 #if defined (__ppc__)
  74 /* 64 bit */
  75 mythread_state_flavor_t thread_flavor_array64[]={
  76                 {PPC_THREAD_STATE64 , PPC_THREAD_STATE64_COUNT},
  77                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  78                 {PPC_EXCEPTION_STATE64, PPC_EXCEPTION_STATE64_COUNT},
  79                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  80                 };
  81
  82 /* 32 bit */
  83 mythread_state_flavor_t thread_flavor_array[]={
  84                 {PPC_THREAD_STATE , PPC_THREAD_STATE_COUNT},
  85                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  86                 {PPC_EXCEPTION_STATE, PPC_EXCEPTION_STATE_COUNT},
  87                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  88                 };
  89
  90 #elif defined (__i386__) || defined (__x86_64__)
  91 mythread_state_flavor_t thread_flavor_array [] = {
  92                 {x86_THREAD_STATE, x86_THREAD_STATE_COUNT},
  93                 {x86_FLOAT_STATE, x86_FLOAT_STATE_COUNT},
  94                 {x86_EXCEPTION_STATE, x86_EXCEPTION_STATE_COUNT},
  95                 };
  96 int mynum_flavors=3;
  97 #else
  98 #error architecture not supported
  99 #endif
 100
 101
 102 typedef struct {
 103         vm_offset_t header;
 104         int  hoffset;
 105         mythread_state_flavor_t *flavors;
 106         int tstate_size;
 107         int flavor_count;
 108 } tir_t;
 109
 110 /* XXX should be static */
 111 void collectth_state(thread_t th_act, void *tirp);
 112
 113 /* XXX not in a Mach header anywhere */
 114 kern_return_t thread_getstatus(register thread_t act, int flavor,
 115         thread_state_t tstate, mach_msg_type_number_t *count);
 116 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 117
 118 static cpu_type_t process_cpu_type(proc_t proc);
 119 static cpu_type_t process_cpu_subtype(proc_t proc);
 120
 121 #ifdef SECURE_KERNEL
 122 __private_extern__ int do_coredump = 0; /* default: don't dump cores */
 123 #else
 124 __private_extern__ int do_coredump = 1; /* default: dump cores */
 125 #endif
 126 __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 (__ppc__)
 143         #pragma unused(core_proc)
 144         what_we_think = CPU_TYPE_POWERPC;
 145 #endif
 146         return what_we_think;
 147 }
 148
 149 static cpu_type_t
 150 process_cpu_subtype(proc_t core_proc)
 151 {
 152         cpu_type_t what_we_think;
 153 #if defined (__i386__) || defined (__x86_64__)
 154     if (IS_64BIT_PROCESS(core_proc)) {
 155                 what_we_think = CPU_SUBTYPE_X86_64_ALL;
 156         } else {
 157                 what_we_think = CPU_SUBTYPE_I386_ALL;
 158         }
 159 #elif defined (__ppc__)
 160         #pragma unused(core_proc)
 161         what_we_think = CPU_SUBTYPE_POWERPC_ALL;
 162 #endif
 163         return what_we_think;
 164 }
 165
 166 void
 167 collectth_state(thread_t th_act, void *tirp)
 168 {
 169         vm_offset_t     header;
 170         int  hoffset, i ;
 171         mythread_state_flavor_t *flavors;
 172         struct thread_command   *tc;
 173         tir_t *t = (tir_t *)tirp;
 174
 175                 /*
 176                  *      Fill in thread command structure.
 177                  */
 178                 header = t->header;
 179                 hoffset = t->hoffset;
 180                 flavors = t->flavors;
 181
 182                 tc = (struct thread_command *) (header + hoffset);
 183                 tc->cmd = LC_THREAD;
 184                 tc->cmdsize = sizeof(struct thread_command)
 185                                 + t->tstate_size;
 186                 hoffset += sizeof(struct thread_command);
 187                 /*
 188                  * Follow with a struct thread_state_flavor and
 189                  * the appropriate thread state struct for each
 190                  * thread state flavor.
 191                  */
 192                 for (i = 0; i < t->flavor_count; i++) {
 193                         *(mythread_state_flavor_t *)(header+hoffset) =
 194                           flavors[i];
 195                         hoffset += sizeof(mythread_state_flavor_t);
 196                         thread_getstatus(th_act, flavors[i].flavor,
 197                                         (thread_state_t)(header+hoffset),
 198                                         &flavors[i].count);
 199                         hoffset += flavors[i].count*sizeof(int);
 200                 }
 201
 202                 t->hoffset = hoffset;
 203 }
 204
 205
 206 /*
 207  * coredump
 208  *
 209  * Description: Create a core image on the file "core" for the process
 210  *              indicated
 211  *
 212  * Parameters:  core_proc                       Process to dump core [*]
 213  *
 214  * Returns:     0                               Success
 215  *              EFAULT                          Failed
 216  *
 217  * IMPORTANT:   This function can only be called on the current process, due
 218  *              to assumptions below; see variable declaration section for
 219  *              details.
 220  */
 221 #define MAX_TSTATE_FLAVORS      10
 222 int
 223 coredump(proc_t core_proc)
 224 {
 225 /* Begin assumptions that limit us to only the current process */
 226         vfs_context_t ctx = vfs_context_current();
 227         vm_map_t        map = current_map();
 228         task_t          task = current_task();
 229 /* End assumptions */
 230         kauth_cred_t cred = vfs_context_ucred(ctx);
 231         int error = 0;
 232         struct vnode_attr va;
 233         int             thread_count, segment_count;
 234         int             command_size, header_size, tstate_size;
 235         int             hoffset;
 236         off_t           foffset;
 237         vm_map_offset_t vmoffset;
 238         vm_offset_t     header;
 239         vm_map_size_t   vmsize;
 240         vm_prot_t       prot;
 241         vm_prot_t       maxprot;
 242         vm_inherit_t    inherit;
 243         int             error1 = 0;
 244         char            stack_name[MAXCOMLEN+6];
 245         char            *alloced_name = NULL;
 246         char            *name;
 247         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 248         vm_size_t       mapsize;
 249         int             i;
 250         uint32_t nesting_depth = 0;
 251         kern_return_t   kret;
 252         struct vm_region_submap_info_64 vbr;
 253         mach_msg_type_number_t vbrcount = 0;
 254         tir_t tir1;
 255         struct vnode * vp;
 256         struct mach_header      *mh = NULL;     /* protected by is_64 */
 257         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 258         int             is_64 = 0;
 259         size_t          mach_header_sz = sizeof(struct mach_header);
 260         size_t          segment_command_sz = sizeof(struct segment_command);
 261
 262         if (do_coredump == 0 ||         /* Not dumping at all */
 263             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 264               ( (cred->cr_svuid != cred->cr_ruid) ||
 265                 (cred->cr_svgid != cred->cr_rgid)))) {
 266
 267 #if CONFIG_AUDIT
 268                 audit_proc_coredump(core_proc, NULL, EFAULT);
 269 #endif
 270                 return (EFAULT);
 271         }
 272
 273         if (IS_64BIT_PROCESS(core_proc)) {
 274                 is_64 = 1;
 275                 mach_header_sz = sizeof(struct mach_header_64);
 276                 segment_command_sz = sizeof(struct segment_command_64);
 277         }
 278
 279         mapsize = get_vmmap_size(map);
 280
 281         if (mapsize >=  core_proc->p_rlimit[RLIMIT_CORE].rlim_cur)
 282                 return (EFAULT);
 283         (void) task_suspend(task);
 284
 285         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 286
 287         /* create name according to sysctl'able format string */
 288         /* if name creation fails, fall back to historical behaviour... */
 289         if (alloced_name == NULL ||
 290             proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 291                            core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 292                 snprintf(stack_name, sizeof(stack_name),
 293                          "/cores/core.%d", core_proc->p_pid);
 294                 name = stack_name;
 295         } else
 296                 name = alloced_name;
 297
 298         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx)))
 299                 goto out2;
 300
 301         VATTR_INIT(&va);
 302         VATTR_WANTED(&va, va_nlink);
 303         /* Don't dump to non-regular files or files with links. */
 304         if (vp->v_type != VREG ||
 305             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 306                 error = EFAULT;
 307                 goto out;
 308         }
 309
 310         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 311         VATTR_SET(&va, va_data_size, 0);
 312         vnode_setattr(vp, &va, ctx);
 313         core_proc->p_acflag |= ACORE;
 314
 315         /*
 316          *      If the task is modified while dumping the file
 317          *      (e.g., changes in threads or VM, the resulting
 318          *      file will not necessarily be correct.
 319          */
 320
 321         thread_count = get_task_numacts(task);
 322         segment_count = get_vmmap_entries(map); /* XXX */
 323 #if defined (__ppc__)
 324         if (is_64) {
 325                 tir1.flavor_count = sizeof(thread_flavor_array64)/sizeof(mythread_state_flavor_t);
 326                 bcopy(thread_flavor_array64, flavors,sizeof(thread_flavor_array64));
 327         } else {
 328 #endif  /* __ppc __ */
 329                 tir1.flavor_count = sizeof(thread_flavor_array)/sizeof(mythread_state_flavor_t);
 330                 bcopy(thread_flavor_array, flavors,sizeof(thread_flavor_array));
 331 #if defined (__ppc__)
 332         }
 333 #endif  /* __ppc __ */
 334         tstate_size = 0;
 335         for (i = 0; i < tir1.flavor_count; i++)
 336                 tstate_size += sizeof(mythread_state_flavor_t) +
 337                   (flavors[i].count * sizeof(int));
 338         command_size = segment_count * segment_command_sz +
 339           thread_count*sizeof(struct thread_command) +
 340           tstate_size*thread_count;
 341
 342         header_size = command_size + mach_header_sz;
 343
 344         if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size) != KERN_SUCCESS) {
 345                 error = ENOMEM;
 346                 goto out;
 347         }
 348
 349         /*
 350          *      Set up Mach-O header.
 351          */
 352         if (is_64) {
 353                 mh64 = (struct mach_header_64 *)header;
 354                 mh64->magic = MH_MAGIC_64;
 355                 mh64->cputype = process_cpu_type(core_proc);
 356                 mh64->cpusubtype = process_cpu_subtype(core_proc);
 357                 mh64->filetype = MH_CORE;
 358                 mh64->ncmds = segment_count + thread_count;
 359                 mh64->sizeofcmds = command_size;
 360                 mh64->reserved = 0;             /* 8 byte alignment */
 361         } else {
 362                 mh = (struct mach_header *)header;
 363                 mh->magic = MH_MAGIC;
 364                 mh->cputype = process_cpu_type(core_proc);
 365                 mh->cpusubtype = process_cpu_subtype(core_proc);
 366                 mh->filetype = MH_CORE;
 367                 mh->ncmds = segment_count + thread_count;
 368                 mh->sizeofcmds = command_size;
 369         }
 370
 371         hoffset = mach_header_sz;       /* offset into header */
 372         foffset = round_page(header_size);      /* offset into file */
 373         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 374
 375         /*
 376          * We use to check for an error, here, now we try and get
 377          * as much as we can
 378          */
 379         while (segment_count > 0) {
 380                 struct segment_command          *sc;
 381                 struct segment_command_64       *sc64;
 382
 383                 /*
 384                  *      Get region information for next region.
 385                  */
 386
 387                 while (1) {
 388                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 389                         if((kret = mach_vm_region_recurse(map,
 390                                         &vmoffset, &vmsize, &nesting_depth,
 391                                         (vm_region_recurse_info_t)&vbr,
 392                                         &vbrcount)) != KERN_SUCCESS) {
 393                                 break;
 394                         }
 395                         /*
 396                          * If we get a valid mapping back, but we're dumping
 397                          * a 32 bit process,  and it's over the allowable
 398                          * address space of a 32 bit process, it's the same
 399                          * as if mach_vm_region_recurse() failed.
 400                          */
 401                         if (!(is_64) &&
 402                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 403                                 kret = KERN_INVALID_ADDRESS;
 404                                 break;
 405                         }
 406                         if(vbr.is_submap) {
 407                                 nesting_depth++;
 408                                 continue;
 409                         } else {
 410                                 break;
 411                         }
 412                 }
 413                 if(kret != KERN_SUCCESS)
 414                         break;
 415
 416                 prot = vbr.protection;
 417                 maxprot = vbr.max_protection;
 418                 inherit = vbr.inheritance;
 419                 /*
 420                  *      Fill in segment command structure.
 421                  */
 422                 if (is_64) {
 423                         sc64 = (struct segment_command_64 *)(header + hoffset);
 424                         sc64->cmd = LC_SEGMENT_64;
 425                         sc64->cmdsize = sizeof(struct segment_command_64);
 426                         /* segment name is zeroed by kmem_alloc */
 427                         sc64->segname[0] = 0;
 428                         sc64->vmaddr = vmoffset;
 429                         sc64->vmsize = vmsize;
 430                         sc64->fileoff = foffset;
 431                         sc64->filesize = vmsize;
 432                         sc64->maxprot = maxprot;
 433                         sc64->initprot = prot;
 434                         sc64->nsects = 0;
 435                 } else  {
 436                         sc = (struct segment_command *) (header + hoffset);
 437                         sc->cmd = LC_SEGMENT;
 438                         sc->cmdsize = sizeof(struct segment_command);
 439                         /* segment name is zeroed by kmem_alloc */
 440                         sc->segname[0] = 0;
 441                         sc->vmaddr = CAST_DOWN_EXPLICIT(vm_offset_t,vmoffset);
 442                         sc->vmsize = CAST_DOWN_EXPLICIT(vm_size_t,vmsize);
 443                         sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t,foffset); /* will never truncate */
 444                         sc->filesize = CAST_DOWN_EXPLICIT(uint32_t,vmsize); /* will never truncate */
 445                         sc->maxprot = maxprot;
 446                         sc->initprot = prot;
 447                         sc->nsects = 0;
 448                 }
 449
 450                 /*
 451                  *      Write segment out.  Try as hard as possible to
 452                  *      get read access to the data.
 453                  */
 454                 if ((prot & VM_PROT_READ) == 0) {
 455                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 456                                            prot|VM_PROT_READ);
 457                 }
 458                 /*
 459                  *      Only actually perform write if we can read.
 460                  *      Note: if we can't read, then we end up with
 461                  *      a hole in the file.
 462                  */
 463                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 464                         && vbr.user_tag != VM_MEMORY_IOKIT
 465                         && coredumpok(map,vmoffset)) {
 466
 467                         error = vn_rdwr_64(UIO_WRITE, vp, vmoffset, vmsize, foffset,
 468                                         (IS_64BIT_PROCESS(core_proc) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 469                                         IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int64_t *) 0, core_proc);
 470
 471                 }
 472
 473                 hoffset += segment_command_sz;
 474                 foffset += vmsize;
 475                 vmoffset += vmsize;
 476                 segment_count--;
 477         }
 478
 479         /*
 480          * If there are remaining segments which have not been written
 481          * out because break in the loop above, then they were not counted
 482          * because they exceed the real address space of the executable
 483          * type: remove them from the header's count.  This is OK, since
 484          * we are allowed to have a sparse area following the segments.
 485          */
 486         if (is_64) {
 487                 mh64->ncmds -= segment_count;
 488                 mh64->sizeofcmds -= segment_count * segment_command_sz;
 489         } else {
 490                 mh->ncmds -= segment_count;
 491                 mh->sizeofcmds -= segment_count * segment_command_sz;
 492         }
 493
 494         tir1.header = header;
 495         tir1.hoffset = hoffset;
 496         tir1.flavors = flavors;
 497         tir1.tstate_size = tstate_size;
 498         task_act_iterate_wth_args(task, collectth_state,&tir1);
 499
 500         /*
 501          *      Write out the Mach header at the beginning of the
 502          *      file.  OK to use a 32 bit write for this.
 503          */
 504         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 505                         UIO_SYSSPACE, IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int *) 0, core_proc);
 506         kmem_free(kernel_map, header, header_size);
 507 out:
 508         error1 = vnode_close(vp, FWRITE, ctx);
 509 out2:
 510 #if CONFIG_AUDIT
 511         audit_proc_coredump(core_proc, name, error);
 512 #endif
 513         if (alloced_name != NULL)
 514                 FREE(alloced_name, M_TEMP);
 515         if (error == 0)
 516                 error = error1;
 517
 518         return (error);
 519 }