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
 103 static cpu_type_t process_cpu_type(proc_t proc);
 104 static cpu_type_t process_cpu_subtype(proc_t proc);
 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 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 /*
 186  * coredump
 187  *
 188  * Description: Create a core image on the file "core" for the process
 189  *              indicated
 190  *
 191  * Parameters:  core_proc                       Process to dump core [*]
 192  *                              reserve_mb                      If non-zero, leave filesystem with
 193  *                                                                      at least this much free space.
 194  *                              coredump_flags  Extra options (ignore rlimit, run fsync)
 195  *
 196  * Returns:     0                               Success
 197  *              EFAULT                          Failed
 198  *
 199  * IMPORTANT:   This function can only be called on the current process, due
 200  *              to assumptions below; see variable declaration section for
 201  *              details.
 202  */
 203 #define MAX_TSTATE_FLAVORS      10
 204 int
 205 coredump(proc_t core_proc, uint32_t reserve_mb, int coredump_flags)
 206 {
 207 /* Begin assumptions that limit us to only the current process */
 208         vfs_context_t ctx = vfs_context_current();
 209         vm_map_t        map = current_map();
 210         task_t          task = current_task();
 211 /* End assumptions */
 212         kauth_cred_t cred = vfs_context_ucred(ctx);
 213         int error = 0;
 214         struct vnode_attr va;
 215         int             thread_count, segment_count;
 216         int             command_size, header_size, tstate_size;
 217         int             hoffset;
 218         off_t           foffset;
 219         mach_vm_offset_t vmoffset;
 220         vm_offset_t     header;
 221         mach_vm_size_t  vmsize;
 222         vm_prot_t       prot;
 223         vm_prot_t       maxprot;
 224         vm_inherit_t    inherit;
 225         int             error1 = 0;
 226         char            stack_name[MAXCOMLEN+6];
 227         char            *alloced_name = NULL;
 228         char            *name;
 229         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 230         vm_size_t       mapsize;
 231         int             i;
 232         uint32_t nesting_depth = 0;
 233         kern_return_t   kret;
 234         struct vm_region_submap_info_64 vbr;
 235         mach_msg_type_number_t vbrcount = 0;
 236         tir_t tir1;
 237         struct vnode * vp;
 238         struct mach_header      *mh = NULL;     /* protected by is_64 */
 239         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 240         int             is_64 = 0;
 241         size_t          mach_header_sz = sizeof(struct mach_header);
 242         size_t          segment_command_sz = sizeof(struct segment_command);
 243
 244         if (current_proc() != core_proc) {
 245                 panic("coredump() called against proc that is not current_proc: %p", core_proc);
 246         }
 247
 248         if (do_coredump == 0 ||         /* Not dumping at all */
 249             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 250               ( (kauth_cred_getsvuid(cred) != kauth_cred_getruid(cred)) ||
 251                 (kauth_cred_getsvgid(cred) != kauth_cred_getrgid(cred))))) {
 252
 253 #if CONFIG_AUDIT
 254                 audit_proc_coredump(core_proc, NULL, EFAULT);
 255 #endif
 256                 return (EFAULT);
 257         }
 258
 259         if (IS_64BIT_PROCESS(core_proc)) {
 260                 is_64 = 1;
 261                 mach_header_sz = sizeof(struct mach_header_64);
 262                 segment_command_sz = sizeof(struct segment_command_64);
 263         }
 264
 265         mapsize = get_vmmap_size(map);
 266
 267         if (((coredump_flags & COREDUMP_IGNORE_ULIMIT) == 0) &&
 268             (mapsize >=  core_proc->p_rlimit[RLIMIT_CORE].rlim_cur))
 269                 return (EFAULT);
 270
 271         (void) task_suspend_internal(task);
 272
 273         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 274
 275         /* create name according to sysctl'able format string */
 276         /* if name creation fails, fall back to historical behaviour... */
 277         if (alloced_name == NULL ||
 278             proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 279                            core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 280                 snprintf(stack_name, sizeof(stack_name),
 281                          "/cores/core.%d", core_proc->p_pid);
 282                 name = stack_name;
 283         } else
 284                 name = alloced_name;
 285
 286         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx)))
 287                 goto out2;
 288
 289         VATTR_INIT(&va);
 290         VATTR_WANTED(&va, va_nlink);
 291         /* Don't dump to non-regular files or files with links. */
 292         if (vp->v_type != VREG ||
 293             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 294                 error = EFAULT;
 295                 goto out;
 296         }
 297
 298         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 299         VATTR_SET(&va, va_data_size, 0);
 300         vnode_setattr(vp, &va, ctx);
 301         core_proc->p_acflag |= ACORE;
 302
 303         if ((reserve_mb > 0) &&
 304             ((freespace_mb(vp) - (mapsize >> 20)) < reserve_mb)) {
 305                 error = ENOSPC;
 306                 goto out;
 307         }
 308
 309         /*
 310          *      If the task is modified while dumping the file
 311          *      (e.g., changes in threads or VM, the resulting
 312          *      file will not necessarily be correct.
 313          */
 314
 315         thread_count = get_task_numacts(task);
 316         segment_count = get_vmmap_entries(map); /* XXX */
 317         tir1.flavor_count = sizeof(thread_flavor_array)/sizeof(mythread_state_flavor_t);
 318         bcopy(thread_flavor_array, flavors,sizeof(thread_flavor_array));
 319         tstate_size = 0;
 320         for (i = 0; i < tir1.flavor_count; i++)
 321                 tstate_size += sizeof(mythread_state_flavor_t) +
 322                   (flavors[i].count * sizeof(int));
 323         command_size = segment_count * segment_command_sz +
 324           thread_count*sizeof(struct thread_command) +
 325           tstate_size*thread_count;
 326
 327         header_size = command_size + mach_header_sz;
 328
 329         if (kmem_alloc(kernel_map, &header, (vm_size_t)header_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
 330                 error = ENOMEM;
 331                 goto out;
 332         }
 333
 334         /*
 335          *      Set up Mach-O header.
 336          */
 337         if (is_64) {
 338                 mh64 = (struct mach_header_64 *)header;
 339                 mh64->magic = MH_MAGIC_64;
 340                 mh64->cputype = process_cpu_type(core_proc);
 341                 mh64->cpusubtype = process_cpu_subtype(core_proc);
 342                 mh64->filetype = MH_CORE;
 343                 mh64->ncmds = segment_count + thread_count;
 344                 mh64->sizeofcmds = command_size;
 345                 mh64->reserved = 0;             /* 8 byte alignment */
 346         } else {
 347                 mh = (struct mach_header *)header;
 348                 mh->magic = MH_MAGIC;
 349                 mh->cputype = process_cpu_type(core_proc);
 350                 mh->cpusubtype = process_cpu_subtype(core_proc);
 351                 mh->filetype = MH_CORE;
 352                 mh->ncmds = segment_count + thread_count;
 353                 mh->sizeofcmds = command_size;
 354         }
 355
 356         hoffset = mach_header_sz;       /* offset into header */
 357         foffset = round_page(header_size);      /* offset into file */
 358         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 359
 360         /*
 361          * We use to check for an error, here, now we try and get
 362          * as much as we can
 363          */
 364         while (segment_count > 0) {
 365                 struct segment_command          *sc;
 366                 struct segment_command_64       *sc64;
 367
 368                 /*
 369                  *      Get region information for next region.
 370                  */
 371
 372                 while (1) {
 373                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 374                         if((kret = mach_vm_region_recurse(map,
 375                                         &vmoffset, &vmsize, &nesting_depth,
 376                                         (vm_region_recurse_info_t)&vbr,
 377                                         &vbrcount)) != KERN_SUCCESS) {
 378                                 break;
 379                         }
 380                         /*
 381                          * If we get a valid mapping back, but we're dumping
 382                          * a 32 bit process,  and it's over the allowable
 383                          * address space of a 32 bit process, it's the same
 384                          * as if mach_vm_region_recurse() failed.
 385                          */
 386                         if (!(is_64) &&
 387                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 388                                 kret = KERN_INVALID_ADDRESS;
 389                                 break;
 390                         }
 391                         if(vbr.is_submap) {
 392                                 nesting_depth++;
 393                                 continue;
 394                         } else {
 395                                 break;
 396                         }
 397                 }
 398                 if(kret != KERN_SUCCESS)
 399                         break;
 400
 401                 prot = vbr.protection;
 402                 maxprot = vbr.max_protection;
 403                 inherit = vbr.inheritance;
 404                 /*
 405                  *      Fill in segment command structure.
 406                  */
 407                 if (is_64) {
 408                         sc64 = (struct segment_command_64 *)(header + hoffset);
 409                         sc64->cmd = LC_SEGMENT_64;
 410                         sc64->cmdsize = sizeof(struct segment_command_64);
 411                         /* segment name is zeroed by kmem_alloc */
 412                         sc64->segname[0] = 0;
 413                         sc64->vmaddr = vmoffset;
 414                         sc64->vmsize = vmsize;
 415                         sc64->fileoff = foffset;
 416                         sc64->filesize = vmsize;
 417                         sc64->maxprot = maxprot;
 418                         sc64->initprot = prot;
 419                         sc64->nsects = 0;
 420                         sc64->flags = 0;
 421                 } else  {
 422                         sc = (struct segment_command *) (header + hoffset);
 423                         sc->cmd = LC_SEGMENT;
 424                         sc->cmdsize = sizeof(struct segment_command);
 425                         /* segment name is zeroed by kmem_alloc */
 426                         sc->segname[0] = 0;
 427                         sc->vmaddr = CAST_DOWN_EXPLICIT(vm_offset_t,vmoffset);
 428                         sc->vmsize = CAST_DOWN_EXPLICIT(vm_size_t,vmsize);
 429                         sc->fileoff = CAST_DOWN_EXPLICIT(uint32_t,foffset); /* will never truncate */
 430                         sc->filesize = CAST_DOWN_EXPLICIT(uint32_t,vmsize); /* will never truncate */
 431                         sc->maxprot = maxprot;
 432                         sc->initprot = prot;
 433                         sc->nsects = 0;
 434                         sc->flags = 0;
 435                 }
 436
 437                 /*
 438                  *      Write segment out.  Try as hard as possible to
 439                  *      get read access to the data.
 440                  */
 441                 if ((prot & VM_PROT_READ) == 0) {
 442                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 443                                            prot|VM_PROT_READ);
 444                 }
 445                 /*
 446                  *      Only actually perform write if we can read.
 447                  *      Note: if we can't read, then we end up with
 448                  *      a hole in the file.
 449                  */
 450                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 451                         && vbr.user_tag != VM_MEMORY_IOKIT
 452                         && coredumpok(map,vmoffset)) {
 453
 454                         error = vn_rdwr_64(UIO_WRITE, vp, vmoffset, vmsize, foffset,
 455                                         (IS_64BIT_PROCESS(core_proc) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 456                                         IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int64_t *) 0, core_proc);
 457
 458                 }
 459
 460                 hoffset += segment_command_sz;
 461                 foffset += vmsize;
 462                 vmoffset += vmsize;
 463                 segment_count--;
 464         }
 465
 466         /*
 467          * If there are remaining segments which have not been written
 468          * out because break in the loop above, then they were not counted
 469          * because they exceed the real address space of the executable
 470          * type: remove them from the header's count.  This is OK, since
 471          * we are allowed to have a sparse area following the segments.
 472          */
 473         if (is_64) {
 474                 mh64->ncmds -= segment_count;
 475                 mh64->sizeofcmds -= segment_count * segment_command_sz;
 476         } else {
 477                 mh->ncmds -= segment_count;
 478                 mh->sizeofcmds -= segment_count * segment_command_sz;
 479         }
 480
 481         tir1.header = header;
 482         tir1.hoffset = hoffset;
 483         tir1.flavors = flavors;
 484         tir1.tstate_size = tstate_size;
 485         task_act_iterate_wth_args(task, collectth_state,&tir1);
 486
 487         /*
 488          *      Write out the Mach header at the beginning of the
 489          *      file.  OK to use a 32 bit write for this.
 490          */
 491         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 492                         UIO_SYSSPACE, IO_NOCACHE|IO_NODELOCKED|IO_UNIT, cred, (int *) 0, core_proc);
 493         kmem_free(kernel_map, header, header_size);
 494
 495         if ((coredump_flags & COREDUMP_FULLFSYNC) && error == 0)
 496                 error = VNOP_IOCTL(vp, F_FULLFSYNC, (caddr_t)NULL, 0, ctx);
 497 out:
 498         error1 = vnode_close(vp, FWRITE, ctx);
 499 out2:
 500 #if CONFIG_AUDIT
 501         audit_proc_coredump(core_proc, name, error);
 502 #endif
 503         if (alloced_name != NULL)
 504                 FREE(alloced_name, M_TEMP);
 505         if (error == 0)
 506                 error = error1;
 507
 508         return (error);
 509 }