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 typedef struct {
  67         int     flavor;                 /* the number for this flavor */
  68         mach_msg_type_number_t  count;  /* count of ints in this flavor */
  69 } mythread_state_flavor_t;
  70
  71 #if defined (__ppc__)
  72 /* 64 bit */
  73 mythread_state_flavor_t thread_flavor_array64[]={
  74                 {PPC_THREAD_STATE64 , PPC_THREAD_STATE64_COUNT},
  75                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  76                 {PPC_EXCEPTION_STATE64, PPC_EXCEPTION_STATE64_COUNT},
  77                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  78                 };
  79
  80 /* 32 bit */
  81 mythread_state_flavor_t thread_flavor_array[]={
  82                 {PPC_THREAD_STATE , PPC_THREAD_STATE_COUNT},
  83                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  84                 {PPC_EXCEPTION_STATE, PPC_EXCEPTION_STATE_COUNT},
  85                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  86                 };
  87
  88 #elif defined (__i386__)
  89 mythread_state_flavor_t thread_flavor_array [] = {
  90                 {x86_THREAD_STATE, x86_THREAD_STATE_COUNT},
  91                 {x86_FLOAT_STATE, x86_FLOAT_STATE_COUNT},
  92                 {x86_EXCEPTION_STATE, x86_EXCEPTION_STATE_COUNT},
  93                 };
  94 int mynum_flavors=3;
  95 #else
  96 #error architecture not supported
  97 #endif
  98
  99
 100 typedef struct {
 101         vm_offset_t header;
 102         int  hoffset;
 103         mythread_state_flavor_t *flavors;
 104         int tstate_size;
 105         int flavor_count;
 106 } tir_t;
 107
 108 /* XXX should be static */
 109 void collectth_state(thread_t th_act, void *tirp);
 110
 111 /* XXX not in a Mach header anywhere */
 112 kern_return_t thread_getstatus(register thread_t act, int flavor,
 113         thread_state_t tstate, mach_msg_type_number_t *count);
 114 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 115
 116
 117 __private_extern__ int do_coredump = 1; /* default: dump cores */
 118 __private_extern__ int sugid_coredump = 0; /* default: but not SGUID binaries */
 119
 120 void
 121 collectth_state(thread_t th_act, void *tirp)
 122 {
 123         vm_offset_t     header;
 124         int  hoffset, i ;
 125         mythread_state_flavor_t *flavors;
 126         struct thread_command   *tc;
 127         tir_t *t = (tir_t *)tirp;
 128
 129                 /*
 130                  *      Fill in thread command structure.
 131                  */
 132                 header = t->header;
 133                 hoffset = t->hoffset;
 134                 flavors = t->flavors;
 135
 136                 tc = (struct thread_command *) (header + hoffset);
 137                 tc->cmd = LC_THREAD;
 138                 tc->cmdsize = sizeof(struct thread_command)
 139                                 + t->tstate_size;
 140                 hoffset += sizeof(struct thread_command);
 141                 /*
 142                  * Follow with a struct thread_state_flavor and
 143                  * the appropriate thread state struct for each
 144                  * thread state flavor.
 145                  */
 146                 for (i = 0; i < t->flavor_count; i++) {
 147                         *(mythread_state_flavor_t *)(header+hoffset) =
 148                           flavors[i];
 149                         hoffset += sizeof(mythread_state_flavor_t);
 150                         thread_getstatus(th_act, flavors[i].flavor,
 151                                         (thread_state_t)(header+hoffset),
 152                                         &flavors[i].count);
 153                         hoffset += flavors[i].count*sizeof(int);
 154                 }
 155
 156                 t->hoffset = hoffset;
 157 }
 158
 159
 160 /*
 161  * coredump
 162  *
 163  * Description: Create a core image on the file "core" for the process
 164  *              indicated
 165  *
 166  * Parameters:  core_proc                       Process to dump core [*]
 167  *
 168  * Returns:     0                               Success
 169  *              EFAULT                          Failed
 170  *
 171  * IMPORTANT:   This function can only be called on the current process, due
 172  *              to assumptions below; see variable declaration section for
 173  *              details.
 174  */
 175 #define MAX_TSTATE_FLAVORS      10
 176 int
 177 coredump(proc_t core_proc)
 178 {
 179 /* Begin assumptions that limit us to only the current process */
 180         vfs_context_t ctx = vfs_context_current();
 181         vm_map_t        map = current_map();
 182         task_t          task = current_task();
 183 /* End assumptions */
 184         kauth_cred_t cred = vfs_context_ucred(ctx);
 185         int error = 0;
 186         struct vnode_attr va;
 187         int             thread_count, segment_count;
 188         int             command_size, header_size, tstate_size;
 189         int             hoffset;
 190         off_t           foffset;
 191         vm_map_offset_t vmoffset;
 192         vm_offset_t     header;
 193         vm_map_size_t   vmsize;
 194         vm_prot_t       prot;
 195         vm_prot_t       maxprot;
 196         vm_inherit_t    inherit;
 197         int             error1 = 0;
 198         char            stack_name[MAXCOMLEN+6];
 199         char            *alloced_name = NULL;
 200         char            *name;
 201         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 202         vm_size_t       mapsize;
 203         int             i;
 204         uint32_t nesting_depth = 0;
 205         kern_return_t   kret;
 206         struct vm_region_submap_info_64 vbr;
 207         mach_msg_type_number_t vbrcount = 0;
 208         tir_t tir1;
 209         struct vnode * vp;
 210         struct mach_header      *mh = NULL;     /* protected by is_64 */
 211         struct mach_header_64   *mh64 = NULL;   /* protected by is_64 */
 212         int             is_64 = 0;
 213         size_t          mach_header_sz = sizeof(struct mach_header);
 214         size_t          segment_command_sz = sizeof(struct segment_command);
 215
 216         if (do_coredump == 0 ||         /* Not dumping at all */
 217             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 218               ( (cred->cr_svuid != cred->cr_ruid) ||
 219                 (cred->cr_svgid != cred->cr_rgid)))) {
 220
 221                 return (EFAULT);
 222         }
 223
 224         if (IS_64BIT_PROCESS(core_proc)) {
 225                 is_64 = 1;
 226                 mach_header_sz = sizeof(struct mach_header_64);
 227                 segment_command_sz = sizeof(struct segment_command_64);
 228         }
 229
 230         mapsize = get_vmmap_size(map);
 231
 232         if (mapsize >=  core_proc->p_rlimit[RLIMIT_CORE].rlim_cur)
 233                 return (EFAULT);
 234         (void) task_suspend(task);
 235
 236         MALLOC(alloced_name, char *, MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
 237
 238         /* create name according to sysctl'able format string */
 239         /* if name creation fails, fall back to historical behaviour... */
 240         if (proc_core_name(core_proc->p_comm, kauth_cred_getuid(cred),
 241                            core_proc->p_pid, alloced_name, MAXPATHLEN)) {
 242                 snprintf(stack_name, sizeof(stack_name),
 243                          "/cores/core.%d", core_proc->p_pid);
 244                 name = stack_name;
 245         } else
 246                 name = alloced_name;
 247
 248         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, ctx)))
 249                 goto out2;
 250
 251         VATTR_INIT(&va);
 252         VATTR_WANTED(&va, va_nlink);
 253         /* Don't dump to non-regular files or files with links. */
 254         if (vp->v_type != VREG ||
 255             vnode_getattr(vp, &va, ctx) || va.va_nlink != 1) {
 256                 error = EFAULT;
 257                 goto out;
 258         }
 259
 260         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 261         VATTR_SET(&va, va_data_size, 0);
 262         vnode_setattr(vp, &va, ctx);
 263         core_proc->p_acflag |= ACORE;
 264
 265         /*
 266          *      If the task is modified while dumping the file
 267          *      (e.g., changes in threads or VM, the resulting
 268          *      file will not necessarily be correct.
 269          */
 270
 271         thread_count = get_task_numacts(task);
 272         segment_count = get_vmmap_entries(map); /* XXX */
 273 #if defined (__ppc__)
 274         if (is_64) {
 275                 tir1.flavor_count = sizeof(thread_flavor_array64)/sizeof(mythread_state_flavor_t);
 276                 bcopy(thread_flavor_array64, flavors,sizeof(thread_flavor_array64));
 277         } else {
 278 #endif  /* __ppc __ */
 279                 tir1.flavor_count = sizeof(thread_flavor_array)/sizeof(mythread_state_flavor_t);
 280                 bcopy(thread_flavor_array, flavors,sizeof(thread_flavor_array));
 281 #if defined (__ppc__)
 282         }
 283 #endif  /* __ppc __ */
 284         tstate_size = 0;
 285         for (i = 0; i < tir1.flavor_count; i++)
 286                 tstate_size += sizeof(mythread_state_flavor_t) +
 287                   (flavors[i].count * sizeof(int));
 288         command_size = segment_count * segment_command_sz +
 289           thread_count*sizeof(struct thread_command) +
 290           tstate_size*thread_count;
 291
 292         header_size = command_size + mach_header_sz;
 293
 294         (void) kmem_alloc(kernel_map,
 295                                     (vm_offset_t *)&header,
 296                                     (vm_size_t)header_size);
 297
 298         /*
 299          *      Set up Mach-O header.
 300          */
 301         if (is_64) {
 302                 mh64 = (struct mach_header_64 *)header;
 303                 mh64->magic = MH_MAGIC_64;
 304                 mh64->cputype = cpu_type();
 305                 mh64->cpusubtype = cpu_subtype();
 306                 mh64->filetype = MH_CORE;
 307                 mh64->ncmds = segment_count + thread_count;
 308                 mh64->sizeofcmds = command_size;
 309                 mh64->reserved = 0;             /* 8 byte alignment */
 310         } else {
 311                 mh = (struct mach_header *)header;
 312                 mh->magic = MH_MAGIC;
 313                 mh->cputype = cpu_type();
 314                 mh->cpusubtype = cpu_subtype();
 315                 mh->filetype = MH_CORE;
 316                 mh->ncmds = segment_count + thread_count;
 317                 mh->sizeofcmds = command_size;
 318         }
 319
 320         hoffset = mach_header_sz;       /* offset into header */
 321         foffset = round_page(header_size);      /* offset into file */
 322         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 323
 324         /*
 325          * We use to check for an error, here, now we try and get
 326          * as much as we can
 327          */
 328         while (segment_count > 0) {
 329                 struct segment_command          *sc;
 330                 struct segment_command_64       *sc64;
 331
 332                 /*
 333                  *      Get region information for next region.
 334                  */
 335
 336                 while (1) {
 337                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 338                         if((kret = mach_vm_region_recurse(map,
 339                                         &vmoffset, &vmsize, &nesting_depth,
 340                                         (vm_region_recurse_info_t)&vbr,
 341                                         &vbrcount)) != KERN_SUCCESS) {
 342                                 break;
 343                         }
 344                         /*
 345                          * If we get a valid mapping back, but we're dumping
 346                          * a 32 bit process,  and it's over the allowable
 347                          * address space of a 32 bit process, it's the same
 348                          * as if mach_vm_region_recurse() failed.
 349                          */
 350                         if (!(is_64) &&
 351                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 352                                 kret = KERN_INVALID_ADDRESS;
 353                                 break;
 354                         }
 355                         if(vbr.is_submap) {
 356                                 nesting_depth++;
 357                                 continue;
 358                         } else {
 359                                 break;
 360                         }
 361                 }
 362                 if(kret != KERN_SUCCESS)
 363                         break;
 364
 365                 prot = vbr.protection;
 366                 maxprot = vbr.max_protection;
 367                 inherit = vbr.inheritance;
 368                 /*
 369                  *      Fill in segment command structure.
 370                  */
 371                 if (is_64) {
 372                         sc64 = (struct segment_command_64 *)(header + hoffset);
 373                         sc64->cmd = LC_SEGMENT_64;
 374                         sc64->cmdsize = sizeof(struct segment_command_64);
 375                         /* segment name is zeroed by kmem_alloc */
 376                         sc64->segname[0] = 0;
 377                         sc64->vmaddr = vmoffset;
 378                         sc64->vmsize = vmsize;
 379                         sc64->fileoff = foffset;
 380                         sc64->filesize = vmsize;
 381                         sc64->maxprot = maxprot;
 382                         sc64->initprot = prot;
 383                         sc64->nsects = 0;
 384                 } else  {
 385                         sc = (struct segment_command *) (header + hoffset);
 386                         sc->cmd = LC_SEGMENT;
 387                         sc->cmdsize = sizeof(struct segment_command);
 388                         /* segment name is zeroed by kmem_alloc */
 389                         sc->segname[0] = 0;
 390                         sc->vmaddr = CAST_DOWN(vm_offset_t,vmoffset);
 391                         sc->vmsize = CAST_DOWN(vm_size_t,vmsize);
 392                         sc->fileoff = CAST_DOWN(uint32_t,foffset);
 393                         sc->filesize = CAST_DOWN(uint32_t,vmsize);
 394                         sc->maxprot = maxprot;
 395                         sc->initprot = prot;
 396                         sc->nsects = 0;
 397                 }
 398
 399                 /*
 400                  *      Write segment out.  Try as hard as possible to
 401                  *      get read access to the data.
 402                  */
 403                 if ((prot & VM_PROT_READ) == 0) {
 404                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 405                                            prot|VM_PROT_READ);
 406                 }
 407                 /*
 408                  *      Only actually perform write if we can read.
 409                  *      Note: if we can't read, then we end up with
 410                  *      a hole in the file.
 411                  */
 412                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 413                         && vbr.user_tag != VM_MEMORY_IOKIT
 414                         && coredumpok(map,vmoffset)) {
 415                         vm_map_size_t   tmp_vmsize = vmsize;
 416                         off_t           xfer_foffset = foffset;
 417
 418                         //LP64todo - works around vn_rdwr_64() 2G limit
 419                         while (tmp_vmsize > 0) {
 420                                 vm_map_size_t   xfer_vmsize = tmp_vmsize;
 421                                 if (xfer_vmsize > INT_MAX)
 422                                         xfer_vmsize = INT_MAX;
 423                                 error = vn_rdwr_64(UIO_WRITE, vp,
 424                                                 vmoffset, xfer_vmsize, xfer_foffset,
 425                                         (IS_64BIT_PROCESS(core_proc) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 426                                         IO_NODELOCKED|IO_UNIT, cred, (int *) 0, core_proc);
 427                                 tmp_vmsize -= xfer_vmsize;
 428                                 xfer_foffset += xfer_vmsize;
 429                         }
 430                 }
 431
 432                 hoffset += segment_command_sz;
 433                 foffset += vmsize;
 434                 vmoffset += vmsize;
 435                 segment_count--;
 436         }
 437
 438         /*
 439          * If there are remaining segments which have not been written
 440          * out because break in the loop above, then they were not counted
 441          * because they exceed the real address space of the executable
 442          * type: remove them from the header's count.  This is OK, since
 443          * we are allowed to have a sparse area following the segments.
 444          */
 445         if (is_64) {
 446                 mh64->ncmds -= segment_count;
 447                 mh64->sizeofcmds -= segment_count * segment_command_sz;
 448         } else {
 449                 mh->ncmds -= segment_count;
 450                 mh->sizeofcmds -= segment_count * segment_command_sz;
 451         }
 452
 453         tir1.header = header;
 454         tir1.hoffset = hoffset;
 455         tir1.flavors = flavors;
 456         tir1.tstate_size = tstate_size;
 457         task_act_iterate_wth_args(task, collectth_state,&tir1);
 458
 459         /*
 460          *      Write out the Mach header at the beginning of the
 461          *      file.  OK to use a 32 bit write for this.
 462          */
 463         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 464                         UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, core_proc);
 465         kmem_free(kernel_map, header, header_size);
 466 out:
 467         error1 = vnode_close(vp, FWRITE, ctx);
 468 out2:
 469         if (alloced_name != NULL)
 470                 FREE(alloced_name, M_TEMP);
 471         if (error == 0)
 472                 error = error1;
 473
 474         return (error);
 475 }