bsd/kern/kern_core.c

   1 /*
   2  * Copyright (c) 2006 Apple Computer, Inc. All Rights Reserved.
   3  *
   4  * @APPLE_LICENSE_OSREFERENCE_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
  10  * License may not be used to create, or enable the creation or
  11  * redistribution of, unlawful or unlicensed copies of an Apple operating
  12  * system, or to circumvent, violate, or enable the circumvention or
  13  * violation of, any terms of an Apple operating system software license
  14  * agreement.
  15  *
  16  * Please obtain a copy of the License at
  17  * http://www.opensource.apple.com/apsl/ and read it before using this
  18  * file.
  19  *
  20  * The Original Code and all software distributed under the License are
  21  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  22  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  23  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  24  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  25  * Please see the License for the specific language governing rights and
  26  * limitations under the License.
  27  *
  28  * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
  29  */
  30 /* Copyright (c) 1991 NeXT Computer, Inc.  All rights reserved.
  31  *
  32  *      File:   bsd/kern/kern_core.c
  33  *
  34  *      This file contains machine independent code for performing core dumps.
  35  *
  36  */
  37
  38 #include <mach/vm_param.h>
  39 #include <mach/thread_status.h>
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/signalvar.h>
  44 #include <sys/resourcevar.h>
  45 #include <sys/namei.h>
  46 #include <sys/vnode_internal.h>
  47 #include <sys/proc_internal.h>
  48 #include <sys/kauth.h>
  49 #include <sys/timeb.h>
  50 #include <sys/times.h>
  51 #include <sys/acct.h>
  52 #include <sys/file_internal.h>
  53 #include <sys/uio.h>
  54 #include <sys/kernel.h>
  55 #include <sys/stat.h>
  56
  57 #include <mach-o/loader.h>
  58 #include <mach/vm_region.h>
  59 #include <mach/vm_statistics.h>
  60
  61 #include <vm/vm_kern.h>
  62 #include <vm/vm_protos.h> /* last */
  63 #include <vm/vm_map.h>          /* current_map() */
  64 #include <mach/mach_vm.h>       /* mach_vm_region_recurse() */
  65 #include <mach/task.h>          /* task_suspend() */
  66 #include <kern/task.h>          /* get_task_numacts() */
  67
  68 typedef struct {
  69         int     flavor;                 /* the number for this flavor */
  70         int     count;                  /* count of ints in this flavor */
  71 } mythread_state_flavor_t;
  72
  73 #if defined (__ppc__)
  74
  75 mythread_state_flavor_t thread_flavor_array[]={
  76                 {PPC_THREAD_STATE , PPC_THREAD_STATE_COUNT},
  77                 {PPC_FLOAT_STATE, PPC_FLOAT_STATE_COUNT},
  78                 {PPC_EXCEPTION_STATE, PPC_EXCEPTION_STATE_COUNT},
  79                 {PPC_VECTOR_STATE, PPC_VECTOR_STATE_COUNT}
  80                 };
  81 int mynum_flavors=4;
  82 #elif defined (__i386__)
  83 mythread_state_flavor_t thread_flavor_array [] = {
  84                 {i386_THREAD_STATE, i386_THREAD_STATE_COUNT},
  85                 {i386_THREAD_FPSTATE, i386_THREAD_FPSTATE_COUNT},
  86                 {i386_THREAD_EXCEPTSTATE, i386_THREAD_EXCEPTSTATE_COUNT},
  87                 {i386_THREAD_CTHREADSTATE, i386_THREAD_CTHREADSTATE_COUNT},
  88                 {i386_NEW_THREAD_STATE, i386_NEW_THREAD_STATE_COUNT},
  89                 {i386_FLOAT_STATE, i386_FLOAT_STATE_COUNT},
  90                 {i386_ISA_PORT_MAP_STATE, i386_ISA_PORT_MAP_STATE_COUNT},
  91                 {i386_V86_ASSIST_STATE, i386_V86_ASSIST_STATE_COUNT},
  92                 {THREAD_SYSCALL_STATE, i386_THREAD_SYSCALL_STATE_COUNT}
  93                 };
  94 int mynum_flavors=9;
  95
  96 #else
  97 #error architecture not supported
  98 #endif
  99
 100
 101 typedef struct {
 102         vm_offset_t header;
 103         int  hoffset;
 104         mythread_state_flavor_t *flavors;
 105         int tstate_size;
 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 < mynum_flavors; 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  * Create a core image on the file "core".
 161  */
 162 #define MAX_TSTATE_FLAVORS      10
 163 int
 164 coredump(struct proc *p)
 165 {
 166         int error=0;
 167         kauth_cred_t cred = kauth_cred_get();
 168         struct vnode_attr va;
 169         struct vfs_context context;
 170         vm_map_t        map;
 171         int             thread_count, segment_count;
 172         int             command_size, header_size, tstate_size;
 173         int             hoffset;
 174         off_t           foffset;
 175         vm_map_offset_t vmoffset;
 176         vm_offset_t     header;
 177         vm_map_size_t   vmsize;
 178         vm_prot_t       prot;
 179         vm_prot_t       maxprot;
 180         vm_inherit_t    inherit;
 181         int             error1;
 182         task_t          task;
 183         char            core_name[MAXCOMLEN+6];
 184         char            *name;
 185         mythread_state_flavor_t flavors[MAX_TSTATE_FLAVORS];
 186         vm_size_t       mapsize;
 187         int             i;
 188         int nesting_depth = 0;
 189         kern_return_t   kret;
 190         struct vm_region_submap_info_64 vbr;
 191         int vbrcount=0;
 192         tir_t tir1;
 193         struct vnode * vp;
 194         struct mach_header      *mh;
 195         struct mach_header_64   *mh64;
 196         int             is_64 = 0;
 197         size_t          mach_header_sz = sizeof(struct mach_header);
 198         size_t          segment_command_sz = sizeof(struct segment_command);
 199
 200         if (do_coredump == 0 ||         /* Not dumping at all */
 201             ( (sugid_coredump == 0) &&  /* Not dumping SUID/SGID binaries */
 202               ( (cred->cr_svuid != cred->cr_ruid) ||
 203                 (cred->cr_svgid != cred->cr_rgid)))) {
 204
 205                 return (EFAULT);
 206         }
 207
 208         if (IS_64BIT_PROCESS(p)) {
 209                 is_64 = 1;
 210                 mach_header_sz = sizeof(struct mach_header_64);
 211                 segment_command_sz = sizeof(struct segment_command_64);
 212         }
 213
 214         task = current_task();
 215         map = current_map();
 216         mapsize = get_vmmap_size(map);
 217
 218         if (mapsize >=  p->p_rlimit[RLIMIT_CORE].rlim_cur)
 219                 return (EFAULT);
 220         (void) task_suspend(task);
 221
 222         /* create name according to sysctl'able format string */
 223         name = proc_core_name(p->p_comm, kauth_cred_getuid(cred), p->p_pid);
 224
 225         /* if name creation fails, fall back to historical behaviour... */
 226         if (name == NULL) {
 227                 sprintf(core_name, "/cores/core.%d", p->p_pid);
 228                 name = core_name;
 229         }
 230         context.vc_proc = p;
 231         context.vc_ucred = cred;
 232
 233         if ((error = vnode_open(name, (O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, &context)))
 234                 return (error);
 235
 236         VATTR_INIT(&va);
 237         VATTR_WANTED(&va, va_nlink);
 238         /* Don't dump to non-regular files or files with links. */
 239         if (vp->v_type != VREG ||
 240             vnode_getattr(vp, &va, &context) || va.va_nlink != 1) {
 241                 error = EFAULT;
 242                 goto out;
 243         }
 244
 245         VATTR_INIT(&va);        /* better to do it here than waste more stack in vnode_setsize */
 246         VATTR_SET(&va, va_data_size, 0);
 247         vnode_setattr(vp, &va, &context);
 248         p->p_acflag |= ACORE;
 249
 250         /*
 251          *      If the task is modified while dumping the file
 252          *      (e.g., changes in threads or VM, the resulting
 253          *      file will not necessarily be correct.
 254          */
 255
 256         thread_count = get_task_numacts(task);
 257         segment_count = get_vmmap_entries(map); /* XXX */
 258         bcopy(thread_flavor_array,flavors,sizeof(thread_flavor_array));
 259         tstate_size = 0;
 260         for (i = 0; i < mynum_flavors; i++)
 261                 tstate_size += sizeof(mythread_state_flavor_t) +
 262                   (flavors[i].count * sizeof(int));
 263
 264         command_size = segment_count * segment_command_sz +
 265           thread_count*sizeof(struct thread_command) +
 266           tstate_size*thread_count;
 267
 268         header_size = command_size + mach_header_sz;
 269
 270         (void) kmem_alloc(kernel_map,
 271                                     (vm_offset_t *)&header,
 272                                     (vm_size_t)header_size);
 273
 274         /*
 275          *      Set up Mach-O header.
 276          */
 277         if (is_64) {
 278                 mh64 = (struct mach_header_64 *)header;
 279                 mh64->magic = MH_MAGIC_64;
 280                 mh64->cputype = cpu_type();
 281                 mh64->cpusubtype = cpu_subtype();
 282                 mh64->filetype = MH_CORE;
 283                 mh64->ncmds = segment_count + thread_count;
 284                 mh64->sizeofcmds = command_size;
 285                 mh64->reserved = 0;             /* 8 byte alignment */
 286         } else {
 287                 mh = (struct mach_header *)header;
 288                 mh->magic = MH_MAGIC;
 289                 mh->cputype = cpu_type();
 290                 mh->cpusubtype = cpu_subtype();
 291                 mh->filetype = MH_CORE;
 292                 mh->ncmds = segment_count + thread_count;
 293                 mh->sizeofcmds = command_size;
 294         }
 295
 296         hoffset = mach_header_sz;       /* offset into header */
 297         foffset = round_page(header_size);      /* offset into file */
 298         vmoffset = MACH_VM_MIN_ADDRESS;         /* offset into VM */
 299
 300         /*
 301          * We use to check for an error, here, now we try and get
 302          * as much as we can
 303          */
 304         while (segment_count > 0) {
 305                 struct segment_command          *sc;
 306                 struct segment_command_64       *sc64;
 307
 308                 /*
 309                  *      Get region information for next region.
 310                  */
 311
 312                 while (1) {
 313                         vbrcount = VM_REGION_SUBMAP_INFO_COUNT_64;
 314                         if((kret = mach_vm_region_recurse(map,
 315                                         &vmoffset, &vmsize, &nesting_depth,
 316                                         (vm_region_recurse_info_t)&vbr,
 317                                         &vbrcount)) != KERN_SUCCESS) {
 318                                 break;
 319                         }
 320                         /*
 321                          * If we get a valid mapping back, but we're dumping
 322                          * a 32 bit process,  and it's over the allowable
 323                          * address space of a 32 bit process, it's the same
 324                          * as if mach_vm_region_recurse() failed.
 325                          */
 326                         if (!(is_64) &&
 327                             (vmoffset + vmsize > VM_MAX_ADDRESS)) {
 328                                 kret = KERN_INVALID_ADDRESS;
 329                                 break;
 330                         }
 331                         if(vbr.is_submap) {
 332                                 nesting_depth++;
 333                                 continue;
 334                         } else {
 335                                 break;
 336                         }
 337                 }
 338                 if(kret != KERN_SUCCESS)
 339                         break;
 340
 341                 prot = vbr.protection;
 342                 maxprot = vbr.max_protection;
 343                 inherit = vbr.inheritance;
 344                 /*
 345                  *      Fill in segment command structure.
 346                  */
 347                 if (is_64) {
 348                         sc64 = (struct segment_command_64 *)(header + hoffset);
 349                         sc64->cmd = LC_SEGMENT_64;
 350                         sc64->cmdsize = sizeof(struct segment_command_64);
 351                         /* segment name is zeroed by kmem_alloc */
 352                         sc64->segname[0] = 0;
 353                         sc64->vmaddr = vmoffset;
 354                         sc64->vmsize = vmsize;
 355                         sc64->fileoff = foffset;
 356                         sc64->filesize = vmsize;
 357                         sc64->maxprot = maxprot;
 358                         sc64->initprot = prot;
 359                         sc64->nsects = 0;
 360                 } else  {
 361                         sc = (struct segment_command *) (header + hoffset);
 362                         sc->cmd = LC_SEGMENT;
 363                         sc->cmdsize = sizeof(struct segment_command);
 364                         /* segment name is zeroed by kmem_alloc */
 365                         sc->segname[0] = 0;
 366                         sc->vmaddr = CAST_DOWN(vm_offset_t,vmoffset);
 367                         sc->vmsize = CAST_DOWN(vm_size_t,vmsize);
 368                         sc->fileoff = CAST_DOWN(uint32_t,foffset);
 369                         sc->filesize = CAST_DOWN(uint32_t,vmsize);
 370                         sc->maxprot = maxprot;
 371                         sc->initprot = prot;
 372                         sc->nsects = 0;
 373                 }
 374
 375                 /*
 376                  *      Write segment out.  Try as hard as possible to
 377                  *      get read access to the data.
 378                  */
 379                 if ((prot & VM_PROT_READ) == 0) {
 380                         mach_vm_protect(map, vmoffset, vmsize, FALSE,
 381                                            prot|VM_PROT_READ);
 382                 }
 383                 /*
 384                  *      Only actually perform write if we can read.
 385                  *      Note: if we can't read, then we end up with
 386                  *      a hole in the file.
 387                  */
 388                 if ((maxprot & VM_PROT_READ) == VM_PROT_READ
 389                         && vbr.user_tag != VM_MEMORY_IOKIT
 390                         && coredumpok(map,vmoffset)) {
 391                         vm_map_size_t   tmp_vmsize = vmsize;
 392                         off_t           xfer_foffset = foffset;
 393
 394                         //LP64todo - works around vn_rdwr_64() 2G limit
 395                         while (tmp_vmsize > 0) {
 396                                 vm_map_size_t   xfer_vmsize = tmp_vmsize;
 397                                 if (xfer_vmsize > INT_MAX)
 398                                         xfer_vmsize = INT_MAX;
 399                                 error = vn_rdwr_64(UIO_WRITE, vp,
 400                                                 vmoffset, xfer_vmsize, xfer_foffset,
 401                                         (IS_64BIT_PROCESS(p) ? UIO_USERSPACE64 : UIO_USERSPACE32),
 402                                         IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 403                                 tmp_vmsize -= xfer_vmsize;
 404                                 xfer_foffset += xfer_vmsize;
 405                         }
 406                 }
 407
 408                 hoffset += segment_command_sz;
 409                 foffset += vmsize;
 410                 vmoffset += vmsize;
 411                 segment_count--;
 412         }
 413
 414         /*
 415          * If there are remaining segments which have not been written
 416          * out because break in the loop above, then they were not counted
 417          * because they exceed the real address space of the executable
 418          * type: remove them from the header's count.  This is OK, since
 419          * we are allowed to have a sparse area following the segments.
 420          */
 421         if (is_64) {
 422                 mh64->ncmds -= segment_count;
 423         } else {
 424                 mh->ncmds -= segment_count;
 425         }
 426
 427         tir1.header = header;
 428         tir1.hoffset = hoffset;
 429         tir1.flavors = flavors;
 430         tir1.tstate_size = tstate_size;
 431         task_act_iterate_wth_args(task, collectth_state,&tir1);
 432
 433         /*
 434          *      Write out the Mach header at the beginning of the
 435          *      file.  OK to use a 32 bit write for this.
 436          */
 437         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)header, header_size, (off_t)0,
 438                         UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p);
 439         kmem_free(kernel_map, header, header_size);
 440 out:
 441         error1 = vnode_close(vp, FWRITE, &context);
 442         if (error == 0)
 443                 error = error1;
 444
 445         return (error);
 446 }