bsd/kern/kern_core.c

   1 /*
   2  * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /* Copyright (c) 1991 NeXT Computer, Inc.  All rights reserved.
  23  *
  24  *      File:   bsd/kern/kern_core.c
  25  *
  26  *      This file contains machine independent code for performing core dumps.
  27  *
  28  */
  29
  30 #include <mach/vm_param.h>
  31 #include <mach/thread_status.h>
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/signalvar.h>
  36 #include <sys/resourcevar.h>
  37 #include <sys/namei.h>
  38 #include <sys/vnode_internal.h>
  39 #include <sys/proc_internal.h>
  40 #include <sys/kauth.h>
  41 #include <sys/timeb.h>
  42 #include <sys/times.h>
  43 #include <sys/acct.h>
  44 #include <sys/file_internal.h>
  45 #include <sys/uio.h>
  46 #include <sys/kernel.h>
  47 #include <sys/stat.h>
  48
  49 #include <mach-o/loader.h>
  50 #include <mach/vm_region.h>
  51 #include <mach/vm_statistics.h>
  52
  53 #include <vm/vm_kern.h>
  54 #include <vm/vm_protos.h> /* last */
  55 #include <vm/vm_map.h>          /* current_map() */
  56 #include <mach/mach_vm.h>       /* mach_vm_region_recurse() */
  57 #include <mach/task.h>          /* task_suspend() */
  58 #include <kern/task.h>          /* get_task_numacts() */
  59
  60 typedef struct {
  61         int     flavor;                 /* the number for this flavor */
  62         int     count;                  /* count of ints in this flavor */
  63 } mythread_state_flavor_t;
  64
  65 #if defined (__ppc__)
  66
  67 mythread_state_flavor_t thread_flavor_array[]={
  68                 {PPC_THREAD_STATE , PPC_THREAD_STATE_COUNT},
  69                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  70                 {PPC_EXCEPTION_STATE, PPC_EXCEPTION_STATE_COUNT},
  71                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  72                 };
  73 int mynum_flavors=4;
  74 #elif defined (__i386__)
  75 mythread_state_flavor_t thread_flavor_array [] = {
  76                 {x86_THREAD_STATE, x86_THREAD_STATE_COUNT},
  77                 {x86_FLOAT_STATE, x86_FLOAT_STATE_COUNT},
  78                 {x86_EXCEPTION_STATE, x86_EXCEPTION_STATE_COUNT},
  79                 };
  80 int mynum_flavors=3;
  81
  82 #else
  83 #error architecture not supported
  84 #endif
  85
  86
  87 typedef struct {
  88         vm_offset_t header;
  89         int  hoffset;
  90         mythread_state_flavor_t *flavors;
  91         int tstate_size;
  92 } tir_t;
  93
  94 /* XXX should be static */
  95 void collectth_state(thread_t th_act, void *tirp);
  96
  97 /* XXX not in a Mach header anywhere */
  98 kern_return_t thread_getstatus(register thread_t act, int flavor,
  99         thread_state_t tstate, mach_msg_type_number_t *count);
 100 void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *);
 101
 102
 103 __private_extern__ int do_coredump = 1; /* default: dump cores */
 104 __private_extern__ int sugid_coredump = 0; /* default: but not SGUID binaries */
 105
 106 void
 107 collectth_state(thread_t th_act, void *tirp)
 108 {
 109         vm_offset_t     header;
 110         int  hoffset, i ;
 111         mythread_state_flavor_t *flavors;
 112         struct thread_command   *tc;
 113         tir_t *t = (tir_t *)tirp;
 114
 115                 /*
 116                  *      Fill in thread command structure.
 117                  */
 118                 header = t->header;
 119                 hoffset = t->hoffset;
 120                 flavors = t->flavors;
 121
 122                 tc = (struct thread_command *) (header + hoffset);
 123                 tc->cmd = LC_THREAD;
 124                 tc->cmdsize = sizeof(struct thread_command)
 125                                 + t->tstate_size;
 126                 hoffset += sizeof(struct thread_command);
 127                 /*
 128                  * Follow with a struct thread_state_flavor and
 129                  * the appropriate thread state struct for each
 130                  * thread state flavor.
 131                  */
 132                 for (i = 0; i < mynum_flavors; i++) {
 133                         *(mythread_state_flavor_t *)(header+hoffset) =
 134                           flavors[i];
 135                         hoffset += sizeof(mythread_state_flavor_t);
 136                         thread_getstatus(th_act, flavors[i].flavor,
 137                                         (thread_state_t)(header+hoffset),
 138                                         &flavors[i].count);
 139                         hoffset += flavors[i].count*sizeof(int);
 140                 }
 141
 142                 t->hoffset = hoffset;
 143 }
 144
 145 /*
 146  * Create a core image on the file "core".
 147  */
 148 #define MAX_TSTATE_FLAVORS      10
 149 int
 150 coredump(struct proc *p)
 151 {
 152         int error=0;
 153         kauth_cred_t cred = kauth_cred_get();
 154         struct vnode_attr va;
 155         struct vfs_context context;
 156         vm_map_t        map;
 157         int             thread_count, segment_count;
 158         int             command_size, header_size, tstate_size;
 159         int             hoffset;
 160         off_t           foffset;
 161         vm_map_offset_t vmoffset;
 162         vm_offset_t     header;
 163         vm_map_size_t   vmsize;
 164         vm_prot_t       prot;
 165         vm_prot_t       maxprot;
 166         vm_inherit_t    inherit;
 167         int             error1;
 168         task_t          task;
 169         char            core_name[MAXCOMLEN+6];
 170         char            *name;
 171         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 172         vm_size_t       mapsize;
 173         int             i;
 174         int nesting_depth = 0;
 175         kern_return_t   kret;
 176         struct vm_region_submap_info_64 vbr;
 177         int vbrcount=0;
 178         tir_t tir1;
 179         struct vnode * vp;
 180         struct mach_header      *mh;
 181         struct mach_header_64   *mh64;
 182         int             is_64 = 0;
 183         size_t          mach_header_sz = sizeof(struct mach_header);
 184         size_t          segment_command_sz = sizeof(struct segment_command);
 185
 186         if (do_coredump == 0 ||         /* Not dumping at all */
 187             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 188               ( (cred->cr_svuid != cred->cr_ruid) ||
 189                 (cred->cr_svgid != cred->cr_rgid)))) {
 190
 191                 return (EFAULT);
 192         }
 193
 194         if (IS_64BIT_PROCESS(p)) {
 195                 is_64 = 1;
 196                 mach_header_sz = sizeof(struct mach_header_64);
 197                 segment_command_sz = sizeof(struct segment_command_64);
 198         }
 199
 200         task = current_task();
 201         map = current_map();
 202         mapsize = get_vmmap_size(map);
 203
 204         if (mapsize >=  p->p_rlimit[RLIMIT_CORE].rlim_cur)
 205                 return (EFAULT);
 206         (void) task_suspend(task);
 207
 208         /* create name according to sysctl'able format string */
 209         name = proc_core_name(p->p_comm, kauth_cred_getuid(cred), p->p_pid);
 210
 211         /* if name creation fails, fall back to historical behaviour... */
 212         if (name == NULL) {
 213                 sprintf(core_name, "/cores/core.%d", p->p_pid);
 214                 name = core_name;
 215         }
 216         context.vc_proc = p;
 217         context.vc_ucred = cred;
 218
 219         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, &context)))
 220                 return (error);
 221
 222         VATTR_INIT(&va);
 223         VATTR_WANTED(&va, va_nlink);
 224         /* Don't dump to non-regular files or files with links. */
 225         if (vp->v_type != VREG ||
 226             vnode_getattr(vp, &va, &context) || va.va_nlink != 1) {
 227                 error = EFAULT;
 228                 goto out;
 229         }
 230
 231         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 232         VATTR_SET(&va, va_data_size, 0);
 233         vnode_setattr(vp, &va, &context);
 234         p->p_acflag |= ACORE;
 235
 236         /*
 237          *      If the task is modified while dumping the file
 238          *      (e.g., changes in threads or VM, the resulting
 239          *      file will not necessarily be correct.
 240          */
 241
 242         thread_count = get_task_numacts(task);
 243         segment_count = get_vmmap_entries(map); /* XXX */
 244         bcopy(thread_flavor_array,flavors,sizeof(thread_flavor_array));
 245         tstate_size = 0;
 246         for (i = 0; i < mynum_flavors; i++)
 247                 tstate_size += sizeof(mythread_state_flavor_t) +
 248                   (flavors[i].count * sizeof(int));
 249
 250         command_size = segment_count * segment_command_sz +
 251           thread_count*sizeof(struct thread_command) +
 252           tstate_size*thread_count;
 253
 254         header_size = command_size + mach_header_sz;
 255
 256         (void) kmem_alloc(kernel_map,
 257                                     (vm_offset_t *)&header,
 258                                     (vm_size_t)header_size);
 259
 260         /*
 261          *      Set up Mach-O header.
 262          */
 263         if (is_64) {
 264                 mh64 = (struct mach_header_64 *)header;
 265                 mh64->magic = MH_MAGIC_64;
 266                 mh64->cputype = cpu_type();
 267                 mh64->cpusubtype = cpu_subtype();
 268                 mh64->filetype = MH_CORE;
 269                 mh64->ncmds = segment_count + thread_count;
 270                 mh64->sizeofcmds = command_size;
 271                 mh64->reserved = 0;             /* 8 byte alignment */
 272         } else {
 273                 mh = (struct mach_header *)header;
 274                 mh->magic = MH_MAGIC;
 275                 mh->cputype = cpu_type();
 276                 mh->cpusubtype = cpu_subtype();
 277                 mh->filetype = MH_CORE;
 278                 mh->ncmds = segment_count + thread_count;
 279                 mh->sizeofcmds = command_size;
 280         }
 281
 282         hoffset = mach_header_sz;       /* offset into header */
 283         foffset = round_page(header_size);      /* offset into file */
 284         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 285
 286         /*
 287          * We use to check for an error, here, now we try and get
 288          * as much as we can
 289          */
 290         while (segment_count > 0) {
 291                 struct segment_command          *sc;
 292                 struct segment_command_64       *sc64;
 293
 294                 /*
 295                  *      Get region information for next region.
 296                  */
 297
 298                 while (1) {
 299                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 300                         if((kret = mach_vm_region_recurse(map,
 301                                         &vmoffset, &vmsize, &nesting_depth,
 302                                         (vm_region_recurse_info_t)&vbr,
 303                                         &vbrcount)) != KERN_SUCCESS) {
 304                                 break;
 305                         }
 306                         /*
 307                          * If we get a valid mapping back, but we're dumping
 308                          * a 32 bit process,  and it's over the allowable
 309                          * address space of a 32 bit process, it's the same
 310                          * as if mach_vm_region_recurse() failed.
 311                          */
 312                         if (!(is_64) &&
 313                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 314                                 kret = KERN_INVALID_ADDRESS;
 315                                 break;
 316                         }
 317                         if(vbr.is_submap) {
 318                                 nesting_depth++;
 319                                 continue;
 320                         } else {
 321                                 break;
 322                         }
 323                 }
 324                 if(kret != KERN_SUCCESS)
 325                         break;
 326
 327                 prot = vbr.protection;
 328                 maxprot = vbr.max_protection;
 329                 inherit = vbr.inheritance;
 330                 /*
 331                  *      Fill in segment command structure.
 332                  */
 333                 if (is_64) {
 334                         sc64 = (struct segment_command_64 *)(header + hoffset);
 335                         sc64->cmd = LC_SEGMENT_64;
 336                         sc64->cmdsize = sizeof(struct segment_command_64);
 337                         /* segment name is zeroed by kmem_alloc */
 338                         sc64->segname[0] = 0;
 339                         sc64->vmaddr = vmoffset;
 340                         sc64->vmsize = vmsize;
 341                         sc64->fileoff = foffset;
 342                         sc64->filesize = vmsize;
 343                         sc64->maxprot = maxprot;
 344                         sc64->initprot = prot;
 345                         sc64->nsects = 0;
 346                 } else  {
 347                         sc = (struct segment_command *) (header + hoffset);
 348                         sc->cmd = LC_SEGMENT;
 349                         sc->cmdsize = sizeof(struct segment_command);
 350                         /* segment name is zeroed by kmem_alloc */
 351                         sc->segname[0] = 0;
 352                         sc->vmaddr = CAST_DOWN(vm_offset_t,vmoffset);
 353                         sc->vmsize = CAST_DOWN(vm_size_t,vmsize);
 354                         sc->fileoff = CAST_DOWN(uint32_t,foffset);
 355                         sc->filesize = CAST_DOWN(uint32_t,vmsize);
 356                         sc->maxprot = maxprot;
 357                         sc->initprot = prot;
 358                         sc->nsects = 0;
 359                 }
 360
 361                 /*
 362                  *      Write segment out.  Try as hard as possible to
 363                  *      get read access to the data.
 364                  */
 365                 if ((prot & VM_PROT_READ) == 0) {
 366                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 367                                            prot|VM_PROT_READ);
 368                 }
 369                 /*
 370                  *      Only actually perform write if we can read.
 371                  *      Note: if we can't read, then we end up with
 372                  *      a hole in the file.
 373                  */
 374                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 375                         && vbr.user_tag != VM_MEMORY_IOKIT
 376                         && coredumpok(map,vmoffset)) {
 377                         vm_map_size_t   tmp_vmsize = vmsize;
 378                         off_t           xfer_foffset = foffset;
 379
 380                         //LP64todo - works around vn_rdwr_64() 2G limit
 381                         while (tmp_vmsize > 0) {
 382                                 vm_map_size_t   xfer_vmsize = tmp_vmsize;
 383                                 if (xfer_vmsize > INT_MAX)
 384                                         xfer_vmsize = INT_MAX;
 385                                 error = vn_rdwr_64(UIO_WRITE, vp,
 386                                                 vmoffset, xfer_vmsize, xfer_foffset,
 387                                         (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 388                                         IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 389                                 tmp_vmsize -= xfer_vmsize;
 390                                 xfer_foffset += xfer_vmsize;
 391                         }
 392                 }
 393
 394                 hoffset += segment_command_sz;
 395                 foffset += vmsize;
 396                 vmoffset += vmsize;
 397                 segment_count--;
 398         }
 399
 400         /*
 401          * If there are remaining segments which have not been written
 402          * out because break in the loop above, then they were not counted
 403          * because they exceed the real address space of the executable
 404          * type: remove them from the header's count.  This is OK, since
 405          * we are allowed to have a sparse area following the segments.
 406          */
 407         if (is_64) {
 408                 mh64->ncmds -= segment_count;
 409         } else {
 410                 mh->ncmds -= segment_count;
 411         }
 412
 413         tir1.header = header;
 414         tir1.hoffset = hoffset;
 415         tir1.flavors = flavors;
 416         tir1.tstate_size = tstate_size;
 417         task_act_iterate_wth_args(task, collectth_state,&tir1);
 418
 419         /*
 420          *      Write out the Mach header at the beginning of the
 421          *      file.  OK to use a 32 bit write for this.
 422          */
 423         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 424                         UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 425         kmem_free(kernel_map, header, header_size);
 426 out:
 427         error1 = vnode_close(vp, FWRITE, &context);
 428         if (error == 0)
 429                 error = error1;
 430
 431         return (error);
 432 }