bsd/kern/kern_core.c

   1 /*
   2  * Copyright (c) 2000-2004 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         int     count;                  /* count of ints in this flavor */
  69 } mythread_state_flavor_t;
  70
  71 #if defined (__ppc__)
  72
  73 mythread_state_flavor_t thread_flavor_array[]={
  74                 {PPC_THREAD_STATE , PPC_THREAD_STATE_COUNT},
  75                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  76                 {PPC_EXCEPTION_STATE, PPC_EXCEPTION_STATE_COUNT},
  77                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  78                 };
  79 int mynum_flavors=4;
  80 #elif defined (__i386__)
  81 mythread_state_flavor_t thread_flavor_array [] = {
  82                 {i386_THREAD_STATE, i386_THREAD_STATE_COUNT},
  83                 {i386_THREAD_FPSTATE, i386_THREAD_FPSTATE_COUNT},
  84                 {i386_THREAD_EXCEPTSTATE, i386_THREAD_EXCEPTSTATE_COUNT},
  85                 {i386_THREAD_CTHREADSTATE, i386_THREAD_CTHREADSTATE_COUNT},
  86                 {i386_NEW_THREAD_STATE, i386_NEW_THREAD_STATE_COUNT},
  87                 {i386_FLOAT_STATE, i386_FLOAT_STATE_COUNT},
  88                 {i386_ISA_PORT_MAP_STATE, i386_ISA_PORT_MAP_STATE_COUNT},
  89                 {i386_V86_ASSIST_STATE, i386_V86_ASSIST_STATE_COUNT},
  90                 {THREAD_SYSCALL_STATE, i386_THREAD_SYSCALL_STATE_COUNT}
  91                 };
  92 int mynum_flavors=9;
  93
  94 #else
  95 #error architecture not supported
  96 #endif
  97
  98
  99 typedef struct {
 100         vm_offset_t header;
 101         int  hoffset;
 102         mythread_state_flavor_t *flavors;
 103         int tstate_size;
 104 } tir_t;
 105
 106 /* XXX should be static */
 107 void collectth_state(thread_t th_act, void *tirp);
 108
 109 /* XXX not in a Mach header anywhere */
 110 kern_return_t thread_getstatus(register thread_t act, int flavor,
 111         thread_state_t tstate, mach_msg_type_number_t *count);
 112 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 113
 114
 115 __private_extern__ int do_coredump = 1; /* default: dump cores */
 116 __private_extern__ int sugid_coredump = 0; /* default: but not SGUID binaries */
 117
 118 void
 119 collectth_state(thread_t th_act, void *tirp)
 120 {
 121         vm_offset_t     header;
 122         int  hoffset, i ;
 123         mythread_state_flavor_t *flavors;
 124         struct thread_command   *tc;
 125         tir_t *t = (tir_t *)tirp;
 126
 127                 /*
 128                  *      Fill in thread command structure.
 129                  */
 130                 header = t->header;
 131                 hoffset = t->hoffset;
 132                 flavors = t->flavors;
 133
 134                 tc = (struct thread_command *) (header + hoffset);
 135                 tc->cmd = LC_THREAD;
 136                 tc->cmdsize = sizeof(struct thread_command)
 137                                 + t->tstate_size;
 138                 hoffset += sizeof(struct thread_command);
 139                 /*
 140                  * Follow with a struct thread_state_flavor and
 141                  * the appropriate thread state struct for each
 142                  * thread state flavor.
 143                  */
 144                 for (i = 0; i < mynum_flavors; i++) {
 145                         *(mythread_state_flavor_t *)(header+hoffset) =
 146                           flavors[i];
 147                         hoffset += sizeof(mythread_state_flavor_t);
 148                         thread_getstatus(th_act, flavors[i].flavor,
 149                                         (thread_state_t)(header+hoffset),
 150                                         &flavors[i].count);
 151                         hoffset += flavors[i].count*sizeof(int);
 152                 }
 153
 154                 t->hoffset = hoffset;
 155 }
 156
 157 /*
 158  * Create a core image on the file "core".
 159  */
 160 #define MAX_TSTATE_FLAVORS      10
 161 int
 162 coredump(struct proc *p)
 163 {
 164         int error=0;
 165         kauth_cred_t cred = kauth_cred_get();
 166         struct vnode_attr va;
 167         struct vfs_context context;
 168         vm_map_t        map;
 169         int             thread_count, segment_count;
 170         int             command_size, header_size, tstate_size;
 171         int             hoffset;
 172         off_t           foffset;
 173         vm_map_offset_t vmoffset;
 174         vm_offset_t     header;
 175         vm_map_size_t   vmsize;
 176         vm_prot_t       prot;
 177         vm_prot_t       maxprot;
 178         vm_inherit_t    inherit;
 179         int             error1;
 180         task_t          task;
 181         char            core_name[MAXCOMLEN+6];
 182         char            *name;
 183         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 184         vm_size_t       mapsize;
 185         int             i;
 186         int nesting_depth = 0;
 187         kern_return_t   kret;
 188         struct vm_region_submap_info_64 vbr;
 189         int vbrcount=0;
 190         tir_t tir1;
 191         struct vnode * vp;
 192         struct mach_header      *mh;
 193         struct mach_header_64   *mh64;
 194         int             is_64 = 0;
 195         size_t          mach_header_sz = sizeof(struct mach_header);
 196         size_t          segment_command_sz = sizeof(struct segment_command);
 197
 198         if (do_coredump == 0 ||         /* Not dumping at all */
 199             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 200               ( (cred->cr_svuid != cred->cr_ruid) ||
 201                 (cred->cr_svgid != cred->cr_rgid)))) {
 202
 203                 return (EFAULT);
 204         }
 205
 206         if (IS_64BIT_PROCESS(p)) {
 207                 is_64 = 1;
 208                 mach_header_sz = sizeof(struct mach_header_64);
 209                 segment_command_sz = sizeof(struct segment_command_64);
 210         }
 211
 212         task = current_task();
 213         map = current_map();
 214         mapsize = get_vmmap_size(map);
 215
 216         if (mapsize >=  p->p_rlimit[RLIMIT_CORE].rlim_cur)
 217                 return (EFAULT);
 218         (void) task_suspend(task);
 219
 220         /* create name according to sysctl'able format string */
 221         name = proc_core_name(p->p_comm, kauth_cred_getuid(cred), p->p_pid);
 222
 223         /* if name creation fails, fall back to historical behaviour... */
 224         if (name == NULL) {
 225                 sprintf(core_name, "/cores/core.%d", p->p_pid);
 226                 name = core_name;
 227         }
 228         context.vc_proc = p;
 229         context.vc_ucred = cred;
 230
 231         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, &context)))
 232                 return (error);
 233
 234         VATTR_INIT(&va);
 235         VATTR_WANTED(&va, va_nlink);
 236         /* Don't dump to non-regular files or files with links. */
 237         if (vp->v_type != VREG ||
 238             vnode_getattr(vp, &va, &context) || va.va_nlink != 1) {
 239                 error = EFAULT;
 240                 goto out;
 241         }
 242
 243         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 244         VATTR_SET(&va, va_data_size, 0);
 245         vnode_setattr(vp, &va, &context);
 246         p->p_acflag |= ACORE;
 247
 248         /*
 249          *      If the task is modified while dumping the file
 250          *      (e.g., changes in threads or VM, the resulting
 251          *      file will not necessarily be correct.
 252          */
 253
 254         thread_count = get_task_numacts(task);
 255         segment_count = get_vmmap_entries(map); /* XXX */
 256         bcopy(thread_flavor_array,flavors,sizeof(thread_flavor_array));
 257         tstate_size = 0;
 258         for (i = 0; i < mynum_flavors; i++)
 259                 tstate_size += sizeof(mythread_state_flavor_t) +
 260                   (flavors[i].count * sizeof(int));
 261
 262         command_size = segment_count * segment_command_sz +
 263           thread_count*sizeof(struct thread_command) +
 264           tstate_size*thread_count;
 265
 266         header_size = command_size + mach_header_sz;
 267
 268         (void) kmem_alloc(kernel_map,
 269                                     (vm_offset_t *)&header,
 270                                     (vm_size_t)header_size);
 271
 272         /*
 273          *      Set up Mach-O header.
 274          */
 275         if (is_64) {
 276                 mh64 = (struct mach_header_64 *)header;
 277                 mh64->magic = MH_MAGIC_64;
 278                 mh64->cputype = cpu_type();
 279                 mh64->cpusubtype = cpu_subtype();
 280                 mh64->filetype = MH_CORE;
 281                 mh64->ncmds = segment_count + thread_count;
 282                 mh64->sizeofcmds = command_size;
 283                 mh64->reserved = 0;             /* 8 byte alignment */
 284         } else {
 285                 mh = (struct mach_header *)header;
 286                 mh->magic = MH_MAGIC;
 287                 mh->cputype = cpu_type();
 288                 mh->cpusubtype = cpu_subtype();
 289                 mh->filetype = MH_CORE;
 290                 mh->ncmds = segment_count + thread_count;
 291                 mh->sizeofcmds = command_size;
 292         }
 293
 294         hoffset = mach_header_sz;       /* offset into header */
 295         foffset = round_page(header_size);      /* offset into file */
 296         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 297
 298         /*
 299          * We use to check for an error, here, now we try and get
 300          * as much as we can
 301          */
 302         while (segment_count > 0) {
 303                 struct segment_command          *sc;
 304                 struct segment_command_64       *sc64;
 305
 306                 /*
 307                  *      Get region information for next region.
 308                  */
 309
 310                 while (1) {
 311                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 312                         if((kret = mach_vm_region_recurse(map,
 313                                         &vmoffset, &vmsize, &nesting_depth,
 314                                         (vm_region_recurse_info_t)&vbr,
 315                                         &vbrcount)) != KERN_SUCCESS) {
 316                                 break;
 317                         }
 318                         /*
 319                          * If we get a valid mapping back, but we're dumping
 320                          * a 32 bit process,  and it's over the allowable
 321                          * address space of a 32 bit process, it's the same
 322                          * as if mach_vm_region_recurse() failed.
 323                          */
 324                         if (!(is_64) &&
 325                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 326                                 kret = KERN_INVALID_ADDRESS;
 327                                 break;
 328                         }
 329                         if(vbr.is_submap) {
 330                                 nesting_depth++;
 331                                 continue;
 332                         } else {
 333                                 break;
 334                         }
 335                 }
 336                 if(kret != KERN_SUCCESS)
 337                         break;
 338
 339                 prot = vbr.protection;
 340                 maxprot = vbr.max_protection;
 341                 inherit = vbr.inheritance;
 342                 /*
 343                  *      Fill in segment command structure.
 344                  */
 345                 if (is_64) {
 346                         sc64 = (struct segment_command_64 *)(header + hoffset);
 347                         sc64->cmd = LC_SEGMENT_64;
 348                         sc64->cmdsize = sizeof(struct segment_command_64);
 349                         /* segment name is zeroed by kmem_alloc */
 350                         sc64->segname[0] = 0;
 351                         sc64->vmaddr = vmoffset;
 352                         sc64->vmsize = vmsize;
 353                         sc64->fileoff = foffset;
 354                         sc64->filesize = vmsize;
 355                         sc64->maxprot = maxprot;
 356                         sc64->initprot = prot;
 357                         sc64->nsects = 0;
 358                 } else  {
 359                         sc = (struct segment_command *) (header + hoffset);
 360                         sc->cmd = LC_SEGMENT;
 361                         sc->cmdsize = sizeof(struct segment_command);
 362                         /* segment name is zeroed by kmem_alloc */
 363                         sc->segname[0] = 0;
 364                         sc->vmaddr = CAST_DOWN(vm_offset_t,vmoffset);
 365                         sc->vmsize = CAST_DOWN(vm_size_t,vmsize);
 366                         sc->fileoff = CAST_DOWN(uint32_t,foffset);
 367                         sc->filesize = CAST_DOWN(uint32_t,vmsize);
 368                         sc->maxprot = maxprot;
 369                         sc->initprot = prot;
 370                         sc->nsects = 0;
 371                 }
 372
 373                 /*
 374                  *      Write segment out.  Try as hard as possible to
 375                  *      get read access to the data.
 376                  */
 377                 if ((prot & VM_PROT_READ) == 0) {
 378                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 379                                            prot|VM_PROT_READ);
 380                 }
 381                 /*
 382                  *      Only actually perform write if we can read.
 383                  *      Note: if we can't read, then we end up with
 384                  *      a hole in the file.
 385                  */
 386                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 387                         && vbr.user_tag != VM_MEMORY_IOKIT
 388                         && coredumpok(map,vmoffset)) {
 389                         vm_map_size_t   tmp_vmsize = vmsize;
 390                         off_t           xfer_foffset = foffset;
 391
 392                         //LP64todo - works around vn_rdwr_64() 2G limit
 393                         while (tmp_vmsize > 0) {
 394                                 vm_map_size_t   xfer_vmsize = tmp_vmsize;
 395                                 if (xfer_vmsize > INT_MAX)
 396                                         xfer_vmsize = INT_MAX;
 397                                 error = vn_rdwr_64(UIO_WRITE, vp,
 398                                                 vmoffset, xfer_vmsize, xfer_foffset,
 399                                         (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 400                                         IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 401                                 tmp_vmsize -= xfer_vmsize;
 402                                 xfer_foffset += xfer_vmsize;
 403                         }
 404                 }
 405
 406                 hoffset += segment_command_sz;
 407                 foffset += vmsize;
 408                 vmoffset += vmsize;
 409                 segment_count--;
 410         }
 411
 412         /*
 413          * If there are remaining segments which have not been written
 414          * out because break in the loop above, then they were not counted
 415          * because they exceed the real address space of the executable
 416          * type: remove them from the header's count.  This is OK, since
 417          * we are allowed to have a sparse area following the segments.
 418          */
 419         if (is_64) {
 420                 mh64->ncmds -= segment_count;
 421         } else {
 422                 mh->ncmds -= segment_count;
 423         }
 424
 425         tir1.header = header;
 426         tir1.hoffset = hoffset;
 427         tir1.flavors = flavors;
 428         tir1.tstate_size = tstate_size;
 429         task_act_iterate_wth_args(task, collectth_state,&tir1);
 430
 431         /*
 432          *      Write out the Mach header at the beginning of the
 433          *      file.  OK to use a 32 bit write for this.
 434          */
 435         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 436                         UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 437         kmem_free(kernel_map, header, header_size);
 438 out:
 439         error1 = vnode_close(vp, FWRITE, &context);
 440         if (error == 0)
 441                 error = error1;
 442
 443         return (error);
 444 }