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