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
  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 #else
  87 #error architecture not supported
  88 #endif
  89
  90
  91 typedef struct {
  92         vm_offset_t header;
  93         int  hoffset;
  94         mythread_state_flavor_t *flavors;
  95         int tstate_size;
  96         int flavor_count;
  97 } tir_t;
  98
  99 extern int freespace_mb(vnode_t vp);
 100
 101 /* XXX not in a Mach header anywhere */
 102 kern_return_t thread_getstatus(thread_t act, int flavor,
 103         thread_state_t tstate, mach_msg_type_number_t *count);
 104 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 105
 106 #ifdef SECURE_KERNEL
 107 __XNU_PRIVATE_EXTERN int do_coredump = 0;       /* default: don't dump cores */
 108 #else
 109 __XNU_PRIVATE_EXTERN int do_coredump = 1;       /* default: dump cores */
 110 #endif
 111 __XNU_PRIVATE_EXTERN int sugid_coredump = 0; /* default: but not SGUID binaries */
 112
 113
 114 /* cpu_type returns only the most generic indication of the current CPU. */
 115 /* in a core we want to know the kind of process. */
 116
 117 static cpu_type_t
 118 process_cpu_type(proc_t core_proc)
 119 {
 120         cpu_type_t what_we_think;
 121 #if defined (__i386__) || defined (__x86_64__)
 122     if (IS_64BIT_PROCESS(core_proc)) {
 123                 what_we_think = CPU_TYPE_X86_64;
 124         } else {
 125                 what_we_think = CPU_TYPE_I386;
 126         }
 127 #endif
 128         return what_we_think;
 129 }
 130
 131 static cpu_type_t
 132 process_cpu_subtype(proc_t core_proc)
 133 {
 134         cpu_type_t what_we_think;
 135 #if defined (__i386__) || defined (__x86_64__)
 136     if (IS_64BIT_PROCESS(core_proc)) {
 137                 what_we_think = CPU_SUBTYPE_X86_64_ALL;
 138         } else {
 139                 what_we_think = CPU_SUBTYPE_I386_ALL;
 140         }
 141 #endif
 142         return what_we_think;
 143 }
 144
 145 static void
 146 collectth_state(thread_t th_act, void *tirp)
 147 {
 148         vm_offset_t     header;
 149         int  hoffset, i ;
 150         mythread_state_flavor_t *flavors;
 151         struct thread_command   *tc;
 152         tir_t *t = (tir_t *)tirp;
 153
 154                 /*
 155                  *      Fill in thread command structure.
 156                  */
 157                 header = t->header;
 158                 hoffset = t->hoffset;
 159                 flavors = t->flavors;
 160
 161                 tc = (struct thread_command *) (header + hoffset);
 162                 tc->cmd = LC_THREAD;
 163                 tc->cmdsize = sizeof(struct thread_command)
 164                                 + t->tstate_size;
 165                 hoffset += sizeof(struct thread_command);
 166                 /*
 167                  * Follow with a struct thread_state_flavor and
 168                  * the appropriate thread state struct for each
 169                  * thread state flavor.
 170                  */
 171                 for (i = 0; i < t->flavor_count; i++) {
 172                         *(mythread_state_flavor_t *)(header+hoffset) =
 173                           flavors[i];
 174                         hoffset += sizeof(mythread_state_flavor_t);
 175                         thread_getstatus(th_act, flavors[i].flavor,
 176                                         (thread_state_t)(header+hoffset),
 177                                         &flavors[i].count);
 178                         hoffset += flavors[i].count*sizeof(int);
 179                 }
 180
 181                 t->hoffset = hoffset;
 182 }
 183
 184 /*
 185  * coredump
 186  *
 187  * Description: Create a core image on the file "core" for the process
 188  *              indicated
 189  *
 190  * Parameters:  core_proc                       Process to dump core [*]
 191  *                              reserve_mb                      If non-zero, leave filesystem with
 192  *                                                                      at least this much free space.
 193  *                              coredump_flags  Extra options (ignore rlimit, run fsync)
 194  *
 195  * Returns:     0                               Success
 196  *              EFAULT                          Failed
 197  *
 198  * IMPORTANT:   This function can only be called on the current process, due
 199  *              to assumptions below; see variable declaration section for
 200  *              details.
 201  */
 202 #define MAX_TSTATE_FLAVORS      10
 203 int
 204 coredump(proc_t core_proc, uint32_t reserve_mb, int coredump_flags)
 205 {
 206 /* Begin assumptions that limit us to only the current process */
 207         vfs_context_t ctx = vfs_context_current();
 208         vm_map_t        map = current_map();
 209         task_t          task = current_task();
 210 /* End assumptions */
 211         kauth_cred_t cred = vfs_context_ucred(ctx);
 212         int error = 0;
 213         struct vnode_attr va;
 214         int             thread_count, segment_count;
 215         int             command_size, header_size, tstate_size;
 216         int             hoffset;
 217         off_t           foffset;
 218         mach_vm_offset_t vmoffset;
 219         vm_offset_t     header;
 220         mach_vm_size_t  vmsize;
 221         vm_prot_t       prot;
 222         vm_prot_t       maxprot;
 223         vm_inherit_t    inherit;
 224         int             error1 = 0;
 225         char            stack_name[MAXCOMLEN+6];
 226         char            *alloced_name = NULL;
 227         char            *name;
 228         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 229         vm_size_t       mapsize;
 230         int             i;
 231         uint32_t nesting_depth = 0;
 232         kern_return_t   kret;
 233         struct vm_region_submap_info_64 vbr;
 234         mach_msg_type_number_t vbrcount = 0;
 235         tir_t tir1;
 236         struct vnode * vp;
 237         struct mach_header      *mh = NULL;     /* protected by is_64 */
 238         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 239         int             is_64 = 0;
 240         size_t          mach_header_sz = sizeof(struct mach_header);
 241         size_t          segment_command_sz = sizeof(struct segment_command);
 242
 243         if (current_proc() != core_proc) {
 244                 panic("coredump() called against proc that is not current_proc: %p", core_proc);
 245         }
 246
 247         if (do_coredump == 0 ||         /* Not dumping at all */
 248             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 249               ( (kauth_cred_getsvuid(cred) != kauth_cred_getruid(cred)) ||
 250                 (kauth_cred_getsvgid(cred) != kauth_cred_getrgid(cred))))) {
 251
 252 #if CONFIG_AUDIT
 253                 audit_proc_coredump(core_proc, NULL, EFAULT);
 254 #endif
 255                 return (EFAULT);
 256         }
 257
 258 #if CONFIG_CSR
 259         /* If the process is restricted, CSR isn't configured to allow
 260          * restricted processes to be debugged, and CSR isn't configured in
 261          * AppleInternal mode, then don't dump core. */
 262         if (cs_restricted(core_proc) &&
 263             csr_check(CSR_ALLOW_TASK_FOR_PID) &&
 264             csr_check(CSR_ALLOW_APPLE_INTERNAL)) {
 265 #if CONFIG_AUDIT
 266                 audit_proc_coredump(core_proc, NULL, EFAULT);
 267 #endif
 268                 return (EFAULT);
 269         }
 270 #endif
 271
 272         if (IS_64BIT_PROCESS(core_proc)) {
 273                 is_64 = 1;
 274                 mach_header_sz = sizeof(struct mach_header_64);
 275                 segment_command_sz = sizeof(struct segment_command_64);
 276         }
 277
 278         mapsize = get_vmmap_size(map);
 279
 280         if (((coredump_flags & COREDUMP_IGNORE_ULIMIT) == 0) &&
 281             (mapsize >=  core_proc->p_rlimit[RLIMIT_CORE].rlim_cur))
 282                 return (EFAULT);
 283
 284         (void) task_suspend_internal(task);
 285
 286         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 287
 288         /* create name according to sysctl'able format string */
 289         /* if name creation fails, fall back to historical behaviour... */
 290         if (alloced_name == NULL ||
 291             proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 292                            core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 293                 snprintf(stack_name, sizeof(stack_name),
 294                          "/cores/core.%d", core_proc->p_pid);
 295                 name = stack_name;
 296         } else
 297                 name = alloced_name;
 298
 299         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx)))
 300                 goto out2;
 301
 302         VATTR_INIT(&va);
 303         VATTR_WANTED(&va, va_nlink);
 304         /* Don't dump to non-regular files or files with links. */
 305         if (vp->v_type != VREG ||
 306             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 307                 error = EFAULT;
 308                 goto out;
 309         }
 310
 311         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 312         VATTR_SET(&va, va_data_size, 0);
 313         vnode_setattr(vp, &va, ctx);
 314         core_proc->p_acflag |= ACORE;
 315
 316         if ((reserve_mb > 0) &&
 317             ((freespace_mb(vp) - (mapsize >> 20)) < reserve_mb)) {
 318                 error = ENOSPC;
 319                 goto out;
 320         }
 321
 322         /*
 323          *      If the task is modified while dumping the file
 324          *      (e.g., changes in threads or VM, the resulting
 325          *      file will not necessarily be correct.
 326          */
 327
 328         thread_count = get_task_numacts(task);
 329         segment_count = get_vmmap_entries(map); /* XXX */
 330         tir1.flavor_count = sizeof(thread_flavor_array)/sizeof(mythread_state_flavor_t);
 331         bcopy(thread_flavor_array, flavors,sizeof(thread_flavor_array));
 332         tstate_size = 0;
 333         for (i = 0; i < tir1.flavor_count; i++)
 334                 tstate_size += sizeof(mythread_state_flavor_t) +
 335                   (flavors[i].count * sizeof(int));
 336         command_size = segment_count * segment_command_sz +
 337           thread_count*sizeof(struct thread_command) +
 338           tstate_size*thread_count;
 339
 340         header_size = command_size + mach_header_sz;
 341
 342         if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
 343                 error = ENOMEM;
 344                 goto out;
 345         }
 346
 347         /*
 348          *      Set up Mach-O header.
 349          */
 350         if (is_64) {
 351                 mh64 = (struct mach_header_64 *)header;
 352                 mh64->magic = MH_MAGIC_64;
 353                 mh64->cputype = process_cpu_type(core_proc);
 354                 mh64->cpusubtype = process_cpu_subtype(core_proc);
 355                 mh64->filetype = MH_CORE;
 356                 mh64->ncmds = segment_count + thread_count;
 357                 mh64->sizeofcmds = command_size;
 358                 mh64->reserved = 0;             /* 8 byte alignment */
 359         } else {
 360                 mh = (struct mach_header *)header;
 361                 mh->magic = MH_MAGIC;
 362                 mh->cputype = process_cpu_type(core_proc);
 363                 mh->cpusubtype = process_cpu_subtype(core_proc);
 364                 mh->filetype = MH_CORE;
 365                 mh->ncmds = segment_count + thread_count;
 366                 mh->sizeofcmds = command_size;
 367         }
 368
 369         hoffset = mach_header_sz;       /* offset into header */
 370         foffset = round_page(header_size);      /* offset into file */
 371         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 372
 373         /*
 374          * We use to check for an error, here, now we try and get
 375          * as much as we can
 376          */
 377         while (segment_count > 0) {
 378                 struct segment_command          *sc;
 379                 struct segment_command_64       *sc64;
 380
 381                 /*
 382                  *      Get region information for next region.
 383                  */
 384
 385                 while (1) {
 386                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 387                         if((kret = mach_vm_region_recurse(map,
 388                                         &vmoffset, &vmsize, &nesting_depth,
 389                                         (vm_region_recurse_info_t)&vbr,
 390                                         &vbrcount)) != KERN_SUCCESS) {
 391                                 break;
 392                         }
 393                         /*
 394                          * If we get a valid mapping back, but we're dumping
 395                          * a 32 bit process,  and it's over the allowable
 396                          * address space of a 32 bit process, it's the same
 397                          * as if mach_vm_region_recurse() failed.
 398                          */
 399                         if (!(is_64) &&
 400                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 401                                 kret = KERN_INVALID_ADDRESS;
 402                                 break;
 403                         }
 404                         if(vbr.is_submap) {
 405                                 nesting_depth++;
 406                                 continue;
 407                         } else {
 408                                 break;
 409                         }
 410                 }
 411                 if(kret != KERN_SUCCESS)
 412                         break;
 413
 414                 prot = vbr.protection;
 415                 maxprot = vbr.max_protection;
 416                 inherit = vbr.inheritance;
 417                 /*
 418                  *      Fill in segment command structure.
 419                  */
 420                 if (is_64) {
 421                         sc64 = (struct segment_command_64 *)(header + hoffset);
 422                         sc64->cmd = LC_SEGMENT_64;
 423                         sc64->cmdsize = sizeof(struct segment_command_64);
 424                         /* segment name is zeroed by kmem_alloc */
 425                         sc64->segname[0] = 0;
 426                         sc64->vmaddr = vmoffset;
 427                         sc64->vmsize = vmsize;
 428                         sc64->fileoff = foffset;
 429                         sc64->filesize = vmsize;
 430                         sc64->maxprot = maxprot;
 431                         sc64->initprot = prot;
 432                         sc64->nsects = 0;
 433                         sc64->flags = 0;
 434                 } else  {
 435                         sc = (struct segment_command *) (header + hoffset);
 436                         sc->cmd = LC_SEGMENT;
 437                         sc->cmdsize = sizeof(struct segment_command);
 438                         /* segment name is zeroed by kmem_alloc */
 439                         sc->segname[0] = 0;
 440                         sc->vmaddr = CAST_DOWN_EXPLICIT(vm_offset_t,vmoffset);
 441                         sc->vmsize = CAST_DOWN_EXPLICIT(vm_size_t,vmsize);
 442                         sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t,foffset); /* will never truncate */
 443                         sc->filesize = CAST_DOWN_EXPLICIT(uint32_t,vmsize); /* will never truncate */
 444                         sc->maxprot = maxprot;
 445                         sc->initprot = prot;
 446                         sc->nsects = 0;
 447                         sc->flags = 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
 508         if ((coredump_flags & COREDUMP_FULLFSYNC) && error == 0)
 509                 error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
 510 out:
 511         error1 = vnode_close(vp, FWRITE, ctx);
 512 out2:
 513 #if CONFIG_AUDIT
 514         audit_proc_coredump(core_proc, name, error);
 515 #endif
 516         if (alloced_name != NULL)
 517                 FREE(alloced_name, M_TEMP);
 518         if (error == 0)
 519                 error = error1;
 520
 521         return (error);
 522 }
 523
 524 #else /* CONFIG_COREDUMP */
 525
 526 /* When core dumps aren't needed, no need to compile this file at all */
 527
 528 #error assertion failed: this section is not compiled
 529
 530 #endif /* CONFIG_COREDUMP */