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 extern int freespace_mb(vnode_t vp);
  97
  98 /* XXX not in a Mach header anywhere */
  99 kern_return_t thread_getstatus(register thread_t act, int flavor,
 100         thread_state_t tstate, mach_msg_type_number_t *count);
 101 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 102 extern kern_return_t task_suspend_internal(task_t);
 103
 104 static cpu_type_t process_cpu_type(proc_t proc);
 105 static cpu_type_t process_cpu_subtype(proc_t proc);
 106
 107 #ifdef SECURE_KERNEL
 108 __XNU_PRIVATE_EXTERN int do_coredump = 0;       /* default: don't dump cores */
 109 #else
 110 __XNU_PRIVATE_EXTERN int do_coredump = 1;       /* default: dump cores */
 111 #endif
 112 __XNU_PRIVATE_EXTERN int sugid_coredump = 0; /* default: but not SGUID binaries */
 113
 114
 115 /* cpu_type returns only the most generic indication of the current CPU. */
 116 /* in a core we want to know the kind of process. */
 117
 118 static cpu_type_t
 119 process_cpu_type(proc_t core_proc)
 120 {
 121         cpu_type_t what_we_think;
 122 #if defined (__i386__) || defined (__x86_64__)
 123     if (IS_64BIT_PROCESS(core_proc)) {
 124                 what_we_think = CPU_TYPE_X86_64;
 125         } else {
 126                 what_we_think = CPU_TYPE_I386;
 127         }
 128 #endif
 129         return what_we_think;
 130 }
 131
 132 static cpu_type_t
 133 process_cpu_subtype(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_SUBTYPE_X86_64_ALL;
 139         } else {
 140                 what_we_think = CPU_SUBTYPE_I386_ALL;
 141         }
 142 #endif
 143         return what_we_think;
 144 }
 145
 146 void
 147 collectth_state(thread_t th_act, void *tirp)
 148 {
 149         vm_offset_t     header;
 150         int  hoffset, i ;
 151         mythread_state_flavor_t *flavors;
 152         struct thread_command   *tc;
 153         tir_t *t = (tir_t *)tirp;
 154
 155                 /*
 156                  *      Fill in thread command structure.
 157                  */
 158                 header = t->header;
 159                 hoffset = t->hoffset;
 160                 flavors = t->flavors;
 161
 162                 tc = (struct thread_command *) (header + hoffset);
 163                 tc->cmd = LC_THREAD;
 164                 tc->cmdsize = sizeof(struct thread_command)
 165                                 + t->tstate_size;
 166                 hoffset += sizeof(struct thread_command);
 167                 /*
 168                  * Follow with a struct thread_state_flavor and
 169                  * the appropriate thread state struct for each
 170                  * thread state flavor.
 171                  */
 172                 for (i = 0; i < t->flavor_count; i++) {
 173                         *(mythread_state_flavor_t *)(header+hoffset) =
 174                           flavors[i];
 175                         hoffset += sizeof(mythread_state_flavor_t);
 176                         thread_getstatus(th_act, flavors[i].flavor,
 177                                         (thread_state_t)(header+hoffset),
 178                                         &flavors[i].count);
 179                         hoffset += flavors[i].count*sizeof(int);
 180                 }
 181
 182                 t->hoffset = hoffset;
 183 }
 184
 185
 186 /*
 187  * coredump
 188  *
 189  * Description: Create a core image on the file "core" for the process
 190  *              indicated
 191  *
 192  * Parameters:  core_proc                       Process to dump core [*]
 193  *                              reserve_mb                      If non-zero, leave filesystem with
 194  *                                                                      at least this much free space.
 195  *                              ignore_ulimit           If set, ignore the process's core file ulimit.
 196  *
 197  * Returns:     0                               Success
 198  *              EFAULT                          Failed
 199  *
 200  * IMPORTANT:   This function can only be called on the current process, due
 201  *              to assumptions below; see variable declaration section for
 202  *              details.
 203  */
 204 #define MAX_TSTATE_FLAVORS      10
 205 int
 206 coredump(proc_t core_proc, uint32_t reserve_mb, int ignore_ulimit)
 207 {
 208 /* Begin assumptions that limit us to only the current process */
 209         vfs_context_t ctx = vfs_context_current();
 210         vm_map_t        map = current_map();
 211         task_t          task = current_task();
 212 /* End assumptions */
 213         kauth_cred_t cred = vfs_context_ucred(ctx);
 214         int error = 0;
 215         struct vnode_attr va;
 216         int             thread_count, segment_count;
 217         int             command_size, header_size, tstate_size;
 218         int             hoffset;
 219         off_t           foffset;
 220         mach_vm_offset_t vmoffset;
 221         vm_offset_t     header;
 222         mach_vm_size_t  vmsize;
 223         vm_prot_t       prot;
 224         vm_prot_t       maxprot;
 225         vm_inherit_t    inherit;
 226         int             error1 = 0;
 227         char            stack_name[MAXCOMLEN+6];
 228         char            *alloced_name = NULL;
 229         char            *name;
 230         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 231         vm_size_t       mapsize;
 232         int             i;
 233         uint32_t nesting_depth = 0;
 234         kern_return_t   kret;
 235         struct vm_region_submap_info_64 vbr;
 236         mach_msg_type_number_t vbrcount = 0;
 237         tir_t tir1;
 238         struct vnode * vp;
 239         struct mach_header      *mh = NULL;     /* protected by is_64 */
 240         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 241         int             is_64 = 0;
 242         size_t          mach_header_sz = sizeof(struct mach_header);
 243         size_t          segment_command_sz = sizeof(struct segment_command);
 244
 245         if (current_proc() != core_proc) {
 246                 panic("coredump() called against proc that is not current_proc: %p", core_proc);
 247         }
 248
 249         if (do_coredump == 0 ||         /* Not dumping at all */
 250             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 251               ( (kauth_cred_getsvuid(cred) != kauth_cred_getruid(cred)) ||
 252                 (kauth_cred_getsvgid(cred) != kauth_cred_getrgid(cred))))) {
 253
 254 #if CONFIG_AUDIT
 255                 audit_proc_coredump(core_proc, NULL, EFAULT);
 256 #endif
 257                 return (EFAULT);
 258         }
 259
 260         if (IS_64BIT_PROCESS(core_proc)) {
 261                 is_64 = 1;
 262                 mach_header_sz = sizeof(struct mach_header_64);
 263                 segment_command_sz = sizeof(struct segment_command_64);
 264         }
 265
 266         mapsize = get_vmmap_size(map);
 267
 268         if ((mapsize >=  core_proc->p_rlimit[RLIMIT_CORE].rlim_cur) && (ignore_ulimit == 0))
 269                 return (EFAULT);
 270         (void) task_suspend_internal(task);
 271
 272         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 273
 274         /* create name according to sysctl'able format string */
 275         /* if name creation fails, fall back to historical behaviour... */
 276         if (alloced_name == NULL ||
 277             proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 278                            core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 279                 snprintf(stack_name, sizeof(stack_name),
 280                          "/cores/core.%d", core_proc->p_pid);
 281                 name = stack_name;
 282         } else
 283                 name = alloced_name;
 284
 285         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx)))
 286                 goto out2;
 287
 288         VATTR_INIT(&va);
 289         VATTR_WANTED(&va, va_nlink);
 290         /* Don't dump to non-regular files or files with links. */
 291         if (vp->v_type != VREG ||
 292             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 293                 error = EFAULT;
 294                 goto out;
 295         }
 296
 297         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 298         VATTR_SET(&va, va_data_size, 0);
 299         vnode_setattr(vp, &va, ctx);
 300         core_proc->p_acflag |= ACORE;
 301
 302         if ((reserve_mb > 0) &&
 303             ((freespace_mb(vp) - (mapsize >> 20)) < reserve_mb)) {
 304                 error = ENOSPC;
 305                 goto out;
 306         }
 307
 308         /*
 309          *      If the task is modified while dumping the file
 310          *      (e.g., changes in threads or VM, the resulting
 311          *      file will not necessarily be correct.
 312          */
 313
 314         thread_count = get_task_numacts(task);
 315         segment_count = get_vmmap_entries(map); /* XXX */
 316         tir1.flavor_count = sizeof(thread_flavor_array)/sizeof(mythread_state_flavor_t);
 317         bcopy(thread_flavor_array, flavors,sizeof(thread_flavor_array));
 318         tstate_size = 0;
 319         for (i = 0; i < tir1.flavor_count; i++)
 320                 tstate_size += sizeof(mythread_state_flavor_t) +
 321                   (flavors[i].count * sizeof(int));
 322         command_size = segment_count * segment_command_sz +
 323           thread_count*sizeof(struct thread_command) +
 324           tstate_size*thread_count;
 325
 326         header_size = command_size + mach_header_sz;
 327
 328         if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size) != KERN_SUCCESS) {
 329                 error = ENOMEM;
 330                 goto out;
 331         }
 332
 333         /*
 334          *      Set up Mach-O header.
 335          */
 336         if (is_64) {
 337                 mh64 = (struct mach_header_64 *)header;
 338                 mh64->magic = MH_MAGIC_64;
 339                 mh64->cputype = process_cpu_type(core_proc);
 340                 mh64->cpusubtype = process_cpu_subtype(core_proc);
 341                 mh64->filetype = MH_CORE;
 342                 mh64->ncmds = segment_count + thread_count;
 343                 mh64->sizeofcmds = command_size;
 344                 mh64->reserved = 0;             /* 8 byte alignment */
 345         } else {
 346                 mh = (struct mach_header *)header;
 347                 mh->magic = MH_MAGIC;
 348                 mh->cputype = process_cpu_type(core_proc);
 349                 mh->cpusubtype = process_cpu_subtype(core_proc);
 350                 mh->filetype = MH_CORE;
 351                 mh->ncmds = segment_count + thread_count;
 352                 mh->sizeofcmds = command_size;
 353         }
 354
 355         hoffset = mach_header_sz;       /* offset into header */
 356         foffset = round_page(header_size);      /* offset into file */
 357         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 358
 359         /*
 360          * We use to check for an error, here, now we try and get
 361          * as much as we can
 362          */
 363         while (segment_count > 0) {
 364                 struct segment_command          *sc;
 365                 struct segment_command_64       *sc64;
 366
 367                 /*
 368                  *      Get region information for next region.
 369                  */
 370
 371                 while (1) {
 372                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 373                         if((kret = mach_vm_region_recurse(map,
 374                                         &vmoffset, &vmsize, &nesting_depth,
 375                                         (vm_region_recurse_info_t)&vbr,
 376                                         &vbrcount)) != KERN_SUCCESS) {
 377                                 break;
 378                         }
 379                         /*
 380                          * If we get a valid mapping back, but we're dumping
 381                          * a 32 bit process,  and it's over the allowable
 382                          * address space of a 32 bit process, it's the same
 383                          * as if mach_vm_region_recurse() failed.
 384                          */
 385                         if (!(is_64) &&
 386                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 387                                 kret = KERN_INVALID_ADDRESS;
 388                                 break;
 389                         }
 390                         if(vbr.is_submap) {
 391                                 nesting_depth++;
 392                                 continue;
 393                         } else {
 394                                 break;
 395                         }
 396                 }
 397                 if(kret != KERN_SUCCESS)
 398                         break;
 399
 400                 prot = vbr.protection;
 401                 maxprot = vbr.max_protection;
 402                 inherit = vbr.inheritance;
 403                 /*
 404                  *      Fill in segment command structure.
 405                  */
 406                 if (is_64) {
 407                         sc64 = (struct segment_command_64 *)(header + hoffset);
 408                         sc64->cmd = LC_SEGMENT_64;
 409                         sc64->cmdsize = sizeof(struct segment_command_64);
 410                         /* segment name is zeroed by kmem_alloc */
 411                         sc64->segname[0] = 0;
 412                         sc64->vmaddr = vmoffset;
 413                         sc64->vmsize = vmsize;
 414                         sc64->fileoff = foffset;
 415                         sc64->filesize = vmsize;
 416                         sc64->maxprot = maxprot;
 417                         sc64->initprot = prot;
 418                         sc64->nsects = 0;
 419                 } else  {
 420                         sc = (struct segment_command *) (header + hoffset);
 421                         sc->cmd = LC_SEGMENT;
 422                         sc->cmdsize = sizeof(struct segment_command);
 423                         /* segment name is zeroed by kmem_alloc */
 424                         sc->segname[0] = 0;
 425                         sc->vmaddr = CAST_DOWN_EXPLICIT(vm_offset_t,vmoffset);
 426                         sc->vmsize = CAST_DOWN_EXPLICIT(vm_size_t,vmsize);
 427                         sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t,foffset); /* will never truncate */
 428                         sc->filesize = CAST_DOWN_EXPLICIT(uint32_t,vmsize); /* will never truncate */
 429                         sc->maxprot = maxprot;
 430                         sc->initprot = prot;
 431                         sc->nsects = 0;
 432                 }
 433
 434                 /*
 435                  *      Write segment out.  Try as hard as possible to
 436                  *      get read access to the data.
 437                  */
 438                 if ((prot & VM_PROT_READ) == 0) {
 439                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 440                                            prot|VM_PROT_READ);
 441                 }
 442                 /*
 443                  *      Only actually perform write if we can read.
 444                  *      Note: if we can't read, then we end up with
 445                  *      a hole in the file.
 446                  */
 447                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 448                         && vbr.user_tag != VM_MEMORY_IOKIT
 449                         && coredumpok(map,vmoffset)) {
 450
 451                         error = vn_rdwr_64(UIO_WRITE, vp, vmoffset, vmsize, foffset,
 452                                         (IS_64BIT_PROCESS(core_proc) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 453                                         IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int64_t *) 0, core_proc);
 454
 455                 }
 456
 457                 hoffset += segment_command_sz;
 458                 foffset += vmsize;
 459                 vmoffset += vmsize;
 460                 segment_count--;
 461         }
 462
 463         /*
 464          * If there are remaining segments which have not been written
 465          * out because break in the loop above, then they were not counted
 466          * because they exceed the real address space of the executable
 467          * type: remove them from the header's count.  This is OK, since
 468          * we are allowed to have a sparse area following the segments.
 469          */
 470         if (is_64) {
 471                 mh64->ncmds -= segment_count;
 472                 mh64->sizeofcmds -= segment_count * segment_command_sz;
 473         } else {
 474                 mh->ncmds -= segment_count;
 475                 mh->sizeofcmds -= segment_count * segment_command_sz;
 476         }
 477
 478         tir1.header = header;
 479         tir1.hoffset = hoffset;
 480         tir1.flavors = flavors;
 481         tir1.tstate_size = tstate_size;
 482         task_act_iterate_wth_args(task, collectth_state,&tir1);
 483
 484         /*
 485          *      Write out the Mach header at the beginning of the
 486          *      file.  OK to use a 32 bit write for this.
 487          */
 488         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 489                         UIO_SYSSPACE, IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int *) 0, core_proc);
 490         kmem_free(kernel_map, header, header_size);
 491 out:
 492         error1 = vnode_close(vp, FWRITE, ctx);
 493 out2:
 494 #if CONFIG_AUDIT
 495         audit_proc_coredump(core_proc, name, error);
 496 #endif
 497         if (alloced_name != NULL)
 498                 FREE(alloced_name, M_TEMP);
 499         if (error == 0)
 500                 error = error1;
 501
 502         return (error);
 503 }