sys/kern/kern_checkpoint.c

   1 /*-
   2  * Copyright (c) 2003 Kip Macy
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  */
  26
  27 #include <sys/types.h>
  28 #include <sys/param.h>
  29 #include <sys/proc.h>
  30 #include <sys/module.h>
  31 #include <sys/sysent.h>
  32 #include <sys/kernel.h>
  33 #include <sys/systm.h>
  34 #include <sys/nlookup.h>
  35
  36 #include <sys/file.h>
  37 #include <sys/fcntl.h>
  38 #include <sys/signal.h>
  39 #include <vm/vm_param.h>
  40 #include <vm/vm.h>
  41 #include <sys/imgact_elf.h>
  42 #include <sys/procfs.h>
  43
  44 #include <sys/lock.h>
  45 #include <vm/pmap.h>
  46 #include <vm/vm_map.h>
  47 #include <vm/vm_extern.h>
  48 #include <sys/mman.h>
  49 #include <sys/sysproto.h>
  50 #include <sys/resource.h>
  51 #include <sys/resourcevar.h>
  52 #include <sys/malloc.h>
  53 #include <sys/stat.h>
  54 #include <sys/uio.h>
  55 #include <sys/namei.h>
  56 #include <sys/vnode.h>
  57 #include <machine/inttypes.h>
  58 #include <machine/limits.h>
  59 #include <machine/frame.h>
  60 #include <sys/signalvar.h>
  61 #include <sys/syslog.h>
  62 #include <sys/sysctl.h>
  63 #include <machine/sigframe.h>
  64 #include <sys/exec.h>
  65 #include <sys/unistd.h>
  66 #include <sys/time.h>
  67 #include <sys/kern_syscall.h>
  68 #include <sys/checkpoint.h>
  69 #include <sys/mount.h>
  70 #include <sys/ckpt.h>
  71
  72 #include <sys/mplock2.h>
  73 #include <sys/file2.h>
  74
  75 static int elf_loadphdrs(struct file *fp,  Elf_Phdr *phdr, int numsegs);
  76 static int elf_getnotes(struct lwp *lp, struct file *fp, size_t notesz);
  77 static int elf_demarshalnotes(void *src, prpsinfo_t *psinfo,
  78                  prstatus_t *status, prfpregset_t *fpregset, int nthreads);
  79 static int elf_loadnotes(struct lwp *, prpsinfo_t *, prstatus_t *,
  80                  prfpregset_t *);
  81 static int elf_getsigs(struct lwp *lp, struct file *fp);
  82 static int elf_getfiles(struct lwp *lp, struct file *fp);
  83 static int elf_gettextvp(struct proc *p, struct file *fp);
  84 static char *ckpt_expand_name(const char *name, uid_t uid, pid_t pid);
  85
  86 static int ckptgroup = 0;       /* wheel only, -1 for any group */
  87 SYSCTL_INT(_kern, OID_AUTO, ckptgroup, CTLFLAG_RW, &ckptgroup, 0, "");
  88
  89 /* ref count to see how many processes that are being checkpointed */
  90 static int chptinuse = 0;
  91
  92 static __inline
  93 int
  94 read_check(struct file *fp, void *buf, size_t nbyte)
  95 {
  96         size_t nread;
  97         int error;
  98
  99         PRINTF(("reading %zd bytes\n", nbyte));
 100         error = fp_read(fp, buf, nbyte, &nread, 1, UIO_SYSSPACE);
 101         if (error) {
 102                 PRINTF(("read failed - %d", error));
 103         } else if (nread != nbyte) {
 104                 PRINTF(("wanted to read %zd - read %zd\n", nbyte, nread));
 105                 error = EINVAL;
 106         }
 107         return error;
 108 }
 109
 110 static int
 111 elf_gethdr(struct file *fp, Elf_Ehdr *ehdr)
 112 {
 113         size_t nbyte = sizeof(Elf_Ehdr);
 114         int error;
 115
 116         if ((error = read_check(fp, ehdr, nbyte)) != 0)
 117                 goto done;
 118         if (!(ehdr->e_ehsize == sizeof(Elf_Ehdr))) {
 119                 PRINTF(("wrong elf header size: %d\n"
 120                        "expected size        : %zd\n",
 121                        ehdr->e_ehsize, sizeof(Elf_Ehdr)));
 122                 return EINVAL;
 123         }
 124         if (!(ehdr->e_phentsize == sizeof(Elf_Phdr))) {
 125                 PRINTF(("wrong program header size: %d\n"
 126                        "expected size            : %zd\n",
 127                        ehdr->e_phentsize, sizeof(Elf_Phdr)));
 128                 return EINVAL;
 129         }
 130
 131         if (!(ehdr->e_ident[EI_MAG0] == ELFMAG0 &&
 132               ehdr->e_ident[EI_MAG1] == ELFMAG1 &&
 133               ehdr->e_ident[EI_MAG2] == ELFMAG2 &&
 134               ehdr->e_ident[EI_MAG3] == ELFMAG3 &&
 135               ehdr->e_ident[EI_CLASS] == ELF_CLASS &&
 136               ehdr->e_ident[EI_DATA] == ELF_DATA &&
 137               ehdr->e_ident[EI_VERSION] == EV_CURRENT &&
 138               ehdr->e_ident[EI_OSABI] == ELFOSABI_NONE &&
 139               ehdr->e_ident[EI_ABIVERSION] == 0)) {
 140                 PRINTF(("bad elf header\n there are %d segments\n",
 141                        ehdr->e_phnum));
 142                 return EINVAL;
 143
 144         }
 145         PRINTF(("Elf header size:           %d\n", ehdr->e_ehsize));
 146         PRINTF(("Program header size:       %d\n", ehdr->e_phentsize));
 147         PRINTF(("Number of Program headers: %d\n", ehdr->e_phnum));
 148  done:
 149         return error;
 150 }
 151
 152 static int
 153 elf_getphdrs(struct file *fp, Elf_Phdr *phdr, size_t nbyte)
 154 {
 155         int i;
 156         int error;
 157         int nheaders = nbyte/sizeof(Elf_Phdr);
 158
 159         PRINTF(("reading phdrs section\n"));
 160         if ((error = read_check(fp, phdr, nbyte)) != 0)
 161                 goto done;
 162         PRINTF(("headers section:\n"));
 163         for (i = 0; i < nheaders; i++) {
 164                 PRINTF(("entry type:   %d\n", phdr[i].p_type));
 165                 PRINTF(("file offset:  %jd\n", (intmax_t)phdr[i].p_offset));
 166                 PRINTF(("virt address: %p\n", (uint32_t *)phdr[i].p_vaddr));
 167                 PRINTF(("file size:    %jd\n", (intmax_t)phdr[i].p_filesz));
 168                 PRINTF(("memory size:  %jd\n", (intmax_t)phdr[i].p_memsz));
 169                 PRINTF(("\n"));
 170         }
 171  done:
 172         return error;
 173 }
 174
 175
 176 static int
 177 elf_getnotes(struct lwp *lp, struct file *fp, size_t notesz)
 178 {
 179         int error;
 180         int nthreads;
 181         char *note;
 182         prpsinfo_t *psinfo;
 183         prstatus_t *status;
 184         prfpregset_t *fpregset;
 185
 186         nthreads = (notesz - sizeof(prpsinfo_t))/(sizeof(prstatus_t) +
 187                                                   sizeof(prfpregset_t));
 188         PRINTF(("reading notes header nthreads=%d\n", nthreads));
 189         if (nthreads <= 0 || nthreads > CKPT_MAXTHREADS)
 190                 return EINVAL;
 191
 192         psinfo  = kmalloc(sizeof(prpsinfo_t), M_TEMP, M_ZERO | M_WAITOK);
 193         status  = kmalloc(nthreads*sizeof(prstatus_t), M_TEMP, M_WAITOK);
 194         fpregset  = kmalloc(nthreads*sizeof(prfpregset_t), M_TEMP, M_WAITOK);
 195         note = kmalloc(notesz, M_TEMP, M_WAITOK);
 196
 197
 198         PRINTF(("reading notes section\n"));
 199         if ((error = read_check(fp, note, notesz)) != 0)
 200                 goto done;
 201         error = elf_demarshalnotes(note, psinfo, status, fpregset, nthreads);
 202         if (error)
 203                 goto done;
 204         /* fetch register state from notes */
 205         error = elf_loadnotes(lp, psinfo, status, fpregset);
 206  done:
 207         if (psinfo)
 208                 kfree(psinfo, M_TEMP);
 209         if (status)
 210                 kfree(status, M_TEMP);
 211         if (fpregset)
 212                 kfree(fpregset, M_TEMP);
 213         if (note)
 214                 kfree(note, M_TEMP);
 215         return error;
 216 }
 217
 218 static int
 219 ckpt_thaw_proc(struct lwp *lp, struct file *fp)
 220 {
 221         struct proc *p = lp->lwp_proc;
 222         Elf_Phdr *phdr = NULL;
 223         Elf_Ehdr *ehdr = NULL;
 224         int error;
 225         size_t nbyte;
 226
 227         TRACE_ENTER;
 228
 229         ehdr = kmalloc(sizeof(Elf_Ehdr), M_TEMP, M_ZERO | M_WAITOK);
 230
 231         if ((error = elf_gethdr(fp, ehdr)) != 0)
 232                 goto done;
 233         nbyte = sizeof(Elf_Phdr) * ehdr->e_phnum;
 234         phdr = kmalloc(nbyte, M_TEMP, M_WAITOK);
 235
 236         /* fetch description of program writable mappings */
 237         if ((error = elf_getphdrs(fp, phdr, nbyte)) != 0)
 238                 goto done;
 239
 240         /* fetch notes section containing register state */
 241         if ((error = elf_getnotes(lp, fp, phdr->p_filesz)) != 0)
 242                 goto done;
 243
 244         /* fetch program text vnodes */
 245         if ((error = elf_gettextvp(p, fp)) != 0)
 246                 goto done;
 247
 248         /* fetch signal disposition */
 249         if ((error = elf_getsigs(lp, fp)) != 0) {
 250                 kprintf("failure in recovering signals\n");
 251                 goto done;
 252         }
 253
 254         /* fetch open files */
 255         if ((error = elf_getfiles(lp, fp)) != 0)
 256                 goto done;
 257
 258         /* handle mappings last in case we are reading from a socket */
 259         error = elf_loadphdrs(fp, phdr, ehdr->e_phnum);
 260
 261         /*
 262          * Set the textvp to the checkpoint file and mark the vnode so
 263          * a future checkpointing of this checkpoint-restored program
 264          * will copy out the contents of the mappings rather then trying
 265          * to record the vnode info related to the checkpoint file, which
 266          * is likely going to be destroyed when the program is re-checkpointed.
 267          */
 268         if (error == 0 && fp->f_data && fp->f_type == DTYPE_VNODE) {
 269                 if (p->p_textvp)
 270                         vrele(p->p_textvp);
 271                 p->p_textvp = (struct vnode *)fp->f_data;
 272                 vsetflags(p->p_textvp, VCKPT);
 273                 vref(p->p_textvp);
 274         }
 275 done:
 276         if (ehdr)
 277                 kfree(ehdr, M_TEMP);
 278         if (phdr)
 279                 kfree(phdr, M_TEMP);
 280         TRACE_EXIT;
 281         return error;
 282 }
 283
 284 static int
 285 elf_loadnotes(struct lwp *lp, prpsinfo_t *psinfo, prstatus_t *status,
 286            prfpregset_t *fpregset)
 287 {
 288         struct proc *p = lp->lwp_proc;
 289         int error;
 290
 291         /* validate status and psinfo */
 292         TRACE_ENTER;
 293         if (status->pr_version != PRSTATUS_VERSION ||
 294             status->pr_statussz != sizeof(prstatus_t) ||
 295             status->pr_gregsetsz != sizeof(gregset_t) ||
 296             status->pr_fpregsetsz != sizeof(fpregset_t) ||
 297             psinfo->pr_version != PRPSINFO_VERSION ||
 298             psinfo->pr_psinfosz != sizeof(prpsinfo_t)) {
 299                 PRINTF(("status check failed\n"));
 300                 error = EINVAL;
 301                 goto done;
 302         }
 303         /* XXX lwp handle more than one lwp*/
 304         if ((error = set_regs(lp, &status->pr_reg)) != 0)
 305                 goto done;
 306         error = set_fpregs(lp, fpregset);
 307         strlcpy(p->p_comm, psinfo->pr_fname, sizeof(p->p_comm));
 308         /* XXX psinfo->pr_psargs not yet implemented */
 309  done:
 310         TRACE_EXIT;
 311         return error;
 312 }
 313
 314 static int
 315 elf_getnote(void *src, size_t *off, const char *name, unsigned int type,
 316             void **desc, size_t descsz)
 317 {
 318         Elf_Note note;
 319         int error;
 320
 321         TRACE_ENTER;
 322         if (src == NULL) {
 323                 error = EFAULT;
 324                 goto done;
 325         }
 326         bcopy((char *)src + *off, &note, sizeof note);
 327
 328         PRINTF(("at offset: %zd expected note of type: %d - got: %d\n",
 329                *off, type, note.n_type));
 330         *off += sizeof note;
 331         if (type != note.n_type) {
 332                 TRACE_ERR;
 333                 error = EINVAL;
 334                 goto done;
 335         }
 336         if (strncmp(name, (char *) src + *off, note.n_namesz) != 0) {
 337                 error = EINVAL;
 338                 goto done;
 339         }
 340         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 341         if (note.n_descsz != descsz) {
 342                 TRACE_ERR;
 343                 error = EINVAL;
 344                 goto done;
 345         }
 346         if (desc)
 347                 bcopy((char *)src + *off, *desc, note.n_descsz);
 348         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 349         error = 0;
 350  done:
 351         TRACE_EXIT;
 352         return error;
 353 }
 354
 355 static int
 356 elf_demarshalnotes(void *src, prpsinfo_t *psinfo, prstatus_t *status,
 357                    prfpregset_t *fpregset, int nthreads)
 358 {
 359         int i;
 360         int error;
 361         size_t off = 0;
 362
 363         TRACE_ENTER;
 364         error = elf_getnote(src, &off, "CORE", NT_PRPSINFO,
 365                            (void **)&psinfo, sizeof(prpsinfo_t));
 366         if (error)
 367                 goto done;
 368         error = elf_getnote(src, &off, "CORE", NT_PRSTATUS,
 369                            (void **)&status, sizeof(prstatus_t));
 370         if (error)
 371                 goto done;
 372         error = elf_getnote(src, &off, "CORE", NT_FPREGSET,
 373                            (void **)&fpregset, sizeof(prfpregset_t));
 374         if (error)
 375                 goto done;
 376
 377         /*
 378          * The remaining portion needs to be an integer multiple
 379          * of prstatus_t and prfpregset_t
 380          */
 381         for (i = 0 ; i < nthreads - 1; i++) {
 382                 status++; fpregset++;
 383                 error = elf_getnote(src, &off, "CORE", NT_PRSTATUS,
 384                                    (void **)&status, sizeof (prstatus_t));
 385                 if (error)
 386                         goto done;
 387                 error = elf_getnote(src, &off, "CORE", NT_FPREGSET,
 388                                    (void **)&fpregset, sizeof(prfpregset_t));
 389                 if (error)
 390                         goto done;
 391         }
 392
 393  done:
 394         TRACE_EXIT;
 395         return error;
 396 }
 397
 398
 399 static int
 400 mmap_phdr(struct file *fp, Elf_Phdr *phdr)
 401 {
 402         int error;
 403         size_t len;
 404         int prot;
 405         void *addr;
 406         int flags;
 407         off_t pos;
 408
 409         TRACE_ENTER;
 410         pos = phdr->p_offset;
 411         len = phdr->p_filesz;
 412         addr = (void *)phdr->p_vaddr;
 413         flags = MAP_FIXED | MAP_NOSYNC | MAP_PRIVATE;
 414         prot = 0;
 415         if (phdr->p_flags & PF_R)
 416                 prot |= PROT_READ;
 417         if (phdr->p_flags & PF_W)
 418                 prot |= PROT_WRITE;
 419         if (phdr->p_flags & PF_X)
 420                 prot |= PROT_EXEC;
 421         if ((error = fp_mmap(addr, len, prot, flags, fp, pos, &addr)) != 0) {
 422                 PRINTF(("mmap failed: %d\n", error);       );
 423         }
 424         PRINTF(("map @%08"PRIxPTR"-%08"PRIxPTR" fileoff %08x-%08x\n", (uintptr_t)addr,
 425                    (uintptr_t)((char *)addr + len), (int)pos, (int)(pos + len)));
 426         TRACE_EXIT;
 427         return error;
 428 }
 429
 430 /*
 431  * Load memory mapped segments.  The segments are backed by the checkpoint
 432  * file.
 433  */
 434 static int
 435 elf_loadphdrs(struct file *fp, Elf_Phdr *phdr, int numsegs)
 436 {
 437         int i;
 438         int error = 0;
 439
 440         TRACE_ENTER;
 441         for (i = 1; i < numsegs; i++)  {
 442                 if ((error = mmap_phdr(fp, &phdr[i])) != 0)
 443                         break;
 444         }
 445         TRACE_EXIT;
 446         return error;
 447 }
 448
 449 static int
 450 elf_getsigs(struct lwp *lp, struct file *fp)
 451 {
 452         struct proc *p = lp->lwp_proc;
 453         int error;
 454         struct ckpt_siginfo *csi;
 455
 456         TRACE_ENTER;
 457         csi = kmalloc(sizeof(struct ckpt_siginfo), M_TEMP, M_ZERO | M_WAITOK);
 458         if ((error = read_check(fp, csi, sizeof(struct ckpt_siginfo))) != 0)
 459                 goto done;
 460
 461         if (csi->csi_ckptpisz != sizeof(struct ckpt_siginfo)) {
 462                 TRACE_ERR;
 463                 error = EINVAL;
 464                 goto done;
 465         }
 466         bcopy(&csi->csi_sigacts, p->p_sigacts, sizeof(struct sigacts));
 467         bcopy(&csi->csi_itimerval, &p->p_realtimer, sizeof(struct itimerval));
 468         SIG_CANTMASK(csi->csi_sigmask);
 469         /* XXX lwp handle more than one lwp */
 470         bcopy(&csi->csi_sigmask, &lp->lwp_sigmask, sizeof(sigset_t));
 471         p->p_sigparent = csi->csi_sigparent;
 472  done:
 473         if (csi)
 474                 kfree(csi, M_TEMP);
 475         TRACE_EXIT;
 476         return error;
 477 }
 478
 479 /*
 480  * Returns a locked, refd vnode
 481  */
 482 static int
 483 ckpt_fhtovp(fhandle_t *fh, struct vnode **vpp)
 484 {
 485         struct mount *mp;
 486         int error;
 487
 488         TRACE_ENTER;
 489         mp = vfs_getvfs(&fh->fh_fsid);
 490
 491         if (!mp) {
 492                 TRACE_ERR;
 493                 PRINTF(("failed to get mount - ESTALE\n"));
 494                 TRACE_EXIT;
 495                 return ESTALE;
 496         }
 497         error = VFS_FHTOVP(mp, NULL, &fh->fh_fid, vpp);
 498         mount_drop(mp);
 499         if (error) {
 500                 PRINTF(("failed with: %d\n", error));
 501                 TRACE_ERR;
 502         }
 503         TRACE_EXIT;
 504         return error;
 505 }
 506
 507 static int
 508 mmap_vp(struct vn_hdr *vnh)
 509 {
 510         struct vnode *vp;
 511         Elf_Phdr *phdr;
 512         struct file *fp;
 513         int error;
 514         TRACE_ENTER;
 515
 516         phdr = &vnh->vnh_phdr;
 517
 518         if ((error = ckpt_fhtovp(&vnh->vnh_fh, &vp)) != 0)
 519                 return error;
 520         /*
 521          * XXX O_RDONLY -> or O_RDWR if file is PROT_WRITE, MAP_SHARED
 522          */
 523         if ((error = fp_vpopen(vp, O_RDONLY, &fp)) != 0) {
 524                 vput(vp);
 525                 return error;
 526         }
 527         error = mmap_phdr(fp, phdr);
 528         fp_close(fp);
 529         TRACE_EXIT;
 530         return error;
 531 }
 532
 533
 534 static int
 535 elf_gettextvp(struct proc *p, struct file *fp)
 536 {
 537         int i;
 538         int error;
 539         int vpcount;
 540         struct ckpt_vminfo vminfo;
 541         struct vn_hdr *vnh = NULL;
 542
 543         TRACE_ENTER;
 544         if ((error = read_check(fp, &vminfo, sizeof(vminfo))) != 0)
 545                 goto done;
 546         if (vminfo.cvm_dsize < 0 ||
 547             vminfo.cvm_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur ||
 548             vminfo.cvm_tsize < 0 ||
 549             (u_quad_t)vminfo.cvm_tsize > maxtsiz ||
 550             vminfo.cvm_daddr >= (caddr_t)VM_MAX_USER_ADDRESS ||
 551             vminfo.cvm_taddr >= (caddr_t)VM_MAX_USER_ADDRESS
 552         ) {
 553             error = ERANGE;
 554             goto done;
 555         }
 556
 557         vmspace_exec(p, NULL);
 558         p->p_vmspace->vm_daddr = vminfo.cvm_daddr;
 559         p->p_vmspace->vm_dsize = vminfo.cvm_dsize;
 560         p->p_vmspace->vm_taddr = vminfo.cvm_taddr;
 561         p->p_vmspace->vm_tsize = vminfo.cvm_tsize;
 562         if ((error = read_check(fp, &vpcount, sizeof(int))) != 0)
 563                 goto done;
 564         vnh = kmalloc(sizeof(struct vn_hdr) * vpcount, M_TEMP, M_WAITOK);
 565         if ((error = read_check(fp, vnh, sizeof(struct vn_hdr)*vpcount)) != 0)
 566                 goto done;
 567         for (i = 0; i < vpcount; i++) {
 568                 if ((error = mmap_vp(&vnh[i])) != 0)
 569                         goto done;
 570         }
 571
 572  done:
 573         if (vnh)
 574                 kfree(vnh, M_TEMP);
 575         TRACE_EXIT;
 576         return error;
 577 }
 578
 579
 580
 581 /* place holder */
 582 static int
 583 elf_getfiles(struct lwp *lp, struct file *fp)
 584 {
 585         int error;
 586         int i;
 587         int filecount;
 588         int fd;
 589         struct ckpt_filehdr filehdr;
 590         struct ckpt_fileinfo *cfi_base = NULL;
 591         struct filedesc *fdp = lp->lwp_proc->p_fd;
 592         struct vnode *vp;
 593         struct file *tempfp;
 594         struct file *ofp;
 595
 596         TRACE_ENTER;
 597         if ((error = read_check(fp, &filehdr, sizeof(filehdr))) != 0)
 598                 goto done;
 599         filecount = filehdr.cfh_nfiles;
 600         cfi_base = kmalloc(filecount*sizeof(struct ckpt_fileinfo), M_TEMP, M_WAITOK);
 601         error = read_check(fp, cfi_base, filecount*sizeof(struct ckpt_fileinfo));
 602         if (error)
 603                 goto done;
 604
 605         /*
 606          * Close all file descriptors >= 3.  These descriptors are from the
 607          * checkpt(1) program itself and should not be retained.
 608          *
 609          * XXX we need a flag so a checkpoint restore can opt to supply the
 610          * descriptors, or the non-regular-file descripors.
 611          */
 612         for (i = 3; i < fdp->fd_nfiles; ++i)
 613                 kern_close(i);
 614
 615         /*
 616          * Scan files to load
 617          */
 618         for (i = 0; i < filecount; i++) {
 619                 struct ckpt_fileinfo *cfi= &cfi_base[i];
 620                 /*
 621                  * Ignore placeholder entries where cfi_index is less then
 622                  * zero.  This will occur if the elf core dump code thinks
 623                  * it can save a vnode but winds up not being able to.
 624                  */
 625                 if (cfi->cfi_index < 0)
 626                         continue;
 627
 628                 /*
 629                  * Restore a saved file descriptor.  If CKFIF_ISCKPTFD is
 630                  * set the descriptor represents the checkpoint file itself,
 631                  * probably due to the user calling sys_checkpoint().  We
 632                  * want to use the fp being used to restore the checkpoint
 633                  * instead of trying to restore the original filehandle.
 634                  */
 635                 if (cfi->cfi_ckflags & CKFIF_ISCKPTFD) {
 636                         fhold(fp);
 637                         tempfp = fp;
 638                         error = 0;
 639                 } else {
 640                         error = ckpt_fhtovp(&cfi->cfi_fh, &vp);
 641                         if (error == 0) {
 642                                 error = fp_vpopen(vp, OFLAGS(cfi->cfi_flags),
 643                                                   &tempfp);
 644                                 if (error)
 645                                         vput(vp);
 646                         }
 647                 }
 648                 if (error)
 649                         break;
 650                 tempfp->f_offset = cfi->cfi_offset;
 651
 652                 /*
 653                  * If overwriting a descriptor close the old descriptor.  This
 654                  * only occurs if the saved core saved descriptors that we
 655                  * have not already closed.
 656                  */
 657                 if (cfi->cfi_index < fdp->fd_nfiles &&
 658                     (ofp = fdp->fd_files[cfi->cfi_index].fp) != NULL) {
 659                         kern_close(cfi->cfi_index);
 660                 }
 661
 662                 /*
 663                  * Allocate the descriptor we want.
 664                  */
 665                 if (fdalloc(lp->lwp_proc, cfi->cfi_index, &fd) != 0) {
 666                         PRINTF(("can't currently restore fd: %d\n",
 667                                cfi->cfi_index));
 668                         fp_close(fp);
 669                         goto done;
 670                 }
 671                 KKASSERT(fd == cfi->cfi_index);
 672                 fsetfd(fdp, tempfp, fd);
 673                 fdrop(tempfp);
 674                 cfi++;
 675                 PRINTF(("restoring %d\n", cfi->cfi_index));
 676         }
 677
 678  done:
 679         if (cfi_base)
 680                 kfree(cfi_base, M_TEMP);
 681         TRACE_EXIT;
 682         return error;
 683 }
 684
 685 static int
 686 ckpt_freeze_proc(struct lwp *lp, struct file *fp)
 687 {
 688         struct proc *p = lp->lwp_proc;
 689         rlim_t limit;
 690         int error;
 691
 692         lwkt_gettoken(&p->p_token);     /* needed for proc_*() calls */
 693
 694         PRINTF(("calling generic_elf_coredump\n"));
 695         limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
 696         if (limit) {
 697                 if (p->p_stat != SCORE) {
 698                         proc_stop(p, SCORE);
 699                         while (p->p_nstopped < p->p_nthreads - 1)
 700                                 tsleep(&p->p_nstopped, 0, "freeze", 1);
 701                         error = generic_elf_coredump(lp, SIGCKPT, fp, limit);
 702                         proc_unstop(p, SCORE);
 703                 } else {
 704                         error = ERANGE;
 705                 }
 706         } else {
 707                 error = ERANGE;
 708         }
 709         lwkt_reltoken(&p->p_token);
 710         return error;
 711 }
 712
 713 /*
 714  * MPALMOSTSAFE
 715  */
 716 int
 717 sys_sys_checkpoint(struct sys_checkpoint_args *uap)
 718 {
 719         int error = 0;
 720         struct thread *td = curthread;
 721         struct proc *p = td->td_proc;
 722         struct filedesc *fdp = p->p_fd;
 723         struct file *fp;
 724
 725         /*
 726          * Only certain groups (to reduce our security exposure).  -1
 727          * allows any group.
 728          */
 729         if (ckptgroup >= 0 && groupmember(ckptgroup, td->td_ucred) == 0)
 730                 return (EPERM);
 731
 732         /*
 733          * For now we can only checkpoint the current process
 734          */
 735         if (uap->pid != -1 && uap->pid != p->p_pid)
 736                 return (EINVAL);
 737
 738         get_mplock();
 739
 740         switch (uap->type) {
 741         case CKPT_FREEZE:
 742                 fp = NULL;
 743                 if (uap->fd == -1 && uap->pid == (pid_t)-1)
 744                         error = checkpoint_signal_handler(td->td_lwp);
 745                 else if ((fp = holdfp(fdp, uap->fd, FWRITE)) == NULL)
 746                         error = EBADF;
 747                 else
 748                         error = ckpt_freeze_proc(td->td_lwp, fp);
 749                 if (fp)
 750                         fdrop(fp);
 751                 break;
 752         case CKPT_THAW:
 753                 if (uap->pid != -1) {
 754                         error = EINVAL;
 755                         break;
 756                 }
 757                 if ((fp = holdfp(fdp, uap->fd, FREAD)) == NULL) {
 758                         error = EBADF;
 759                         break;
 760                 }
 761                 uap->sysmsg_result = uap->retval;
 762                 error = ckpt_thaw_proc(td->td_lwp, fp);
 763                 fdrop(fp);
 764                 break;
 765         default:
 766                 error = EOPNOTSUPP;
 767                 break;
 768         }
 769         rel_mplock();
 770         return error;
 771 }
 772
 773 int
 774 checkpoint_signal_handler(struct lwp *lp)
 775 {
 776         struct thread *td = lp->lwp_thread;
 777         struct proc *p = lp->lwp_proc;
 778         char *buf;
 779         struct file *fp;
 780         struct nlookupdata nd;
 781         int error;
 782
 783         chptinuse++;
 784
 785         /*
 786          * Being able to checkpoint an suid or sgid program is not a good
 787          * idea.
 788          */
 789         if (sugid_coredump == 0 && (p->p_flags & P_SUGID)) {
 790                 chptinuse--;
 791                 return (EPERM);
 792         }
 793
 794         buf = ckpt_expand_name(p->p_comm, td->td_ucred->cr_uid, p->p_pid);
 795         if (buf == NULL) {
 796                 chptinuse--;
 797                 return (ENOMEM);
 798         }
 799
 800         log(LOG_INFO, "pid %d (%s), uid %d: checkpointing to %s\n",
 801                 p->p_pid, p->p_comm,
 802                 (td->td_ucred ? td->td_ucred->cr_uid : -1),
 803                 buf);
 804
 805         PRINTF(("ckpt handler called, using '%s'\n", buf));
 806
 807         /*
 808          * Use the same safety flags that the coredump code uses.  Remove
 809          * any previous checkpoint file before writing out the new one in
 810          * case we are re-checkpointing a program that had been checkpt
 811          * restored.  Otherwise we will corrupt the program space (which is
 812          * made up of mmap()ings of the previous checkpoint file) while we
 813          * write out the new one.
 814          */
 815         error = nlookup_init(&nd, buf, UIO_SYSSPACE, 0);
 816         if (error == 0)
 817                 error = kern_unlink(&nd);
 818         nlookup_done(&nd);
 819         error = fp_open(buf, O_WRONLY|O_CREAT|O_TRUNC|O_NOFOLLOW, 0600, &fp);
 820         if (error == 0) {
 821                 error = ckpt_freeze_proc(lp, fp);
 822                 fp_close(fp);
 823         } else {
 824                 kprintf("checkpoint failed with open - error: %d\n", error);
 825         }
 826         kfree(buf, M_TEMP);
 827         chptinuse--;
 828         return (error);
 829 }
 830
 831 static char ckptfilename[MAXPATHLEN] = {"%N.ckpt"};
 832 SYSCTL_STRING(_kern, OID_AUTO, ckptfile, CTLFLAG_RW, ckptfilename,
 833               sizeof(ckptfilename), "process checkpoint name format string");
 834
 835 /*
 836  * expand_name(name, uid, pid)
 837  * Expand the name described in corefilename, using name, uid, and pid.
 838  * corefilename is a kprintf-like string, with three format specifiers:
 839  *      %N      name of process ("name")
 840  *      %P      process id (pid)
 841  *      %U      user id (uid)
 842  * For example, "%N.core" is the default; they can be disabled completely
 843  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
 844  * This is controlled by the sysctl variable kern.corefile (see above).
 845  *
 846  * -- taken from the coredump code
 847  */
 848
 849 static
 850 char *
 851 ckpt_expand_name(const char *name, uid_t uid, pid_t pid)
 852 {
 853         char *temp;
 854         char *bp;
 855         char buf[11];           /* Buffer for pid/uid -- max 4B */
 856         int error;
 857         int i;
 858         int n;
 859         char *format = ckptfilename;
 860         size_t namelen;
 861
 862         temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
 863         if (temp == NULL)
 864                 return NULL;
 865         namelen = strlen(name);
 866         n = 0;
 867         if (ckptfilename[0] != '/') {
 868                 if ((bp = kern_getcwd(temp, MAXPATHLEN - 1, &error)) == NULL) {
 869                         kfree(temp, M_TEMP);
 870                         return NULL;
 871                 }
 872                 n = strlen(bp);
 873                 bcopy(bp, temp, n + 1); /* normalize location of the path */
 874                 temp[n++] = '/';
 875                 temp[n] = '\0';
 876         }
 877         for (i= 0; n < MAXPATHLEN && format[i]; i++) {
 878                 int l;
 879                 switch (format[i]) {
 880                 case '%':       /* Format character */
 881                         i++;
 882                         switch (format[i]) {
 883                         case '%':
 884                                 temp[n++] = '%';
 885                                 break;
 886                         case 'N':       /* process name */
 887                                 if ((n + namelen) > MAXPATHLEN) {
 888                                         log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
 889                                             pid, name, uid, temp, name);
 890                                         kfree(temp, M_TEMP);
 891                                         return NULL;
 892                                 }
 893                                 memcpy(temp+n, name, namelen);
 894                                 n += namelen;
 895                                 break;
 896                         case 'P':       /* process id */
 897                                 l = ksprintf(buf, "%u", pid);
 898                                 if ((n + l) > MAXPATHLEN) {
 899                                         log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
 900                                             pid, name, uid, temp, name);
 901                                         kfree(temp, M_TEMP);
 902                                         return NULL;
 903                                 }
 904                                 memcpy(temp+n, buf, l);
 905                                 n += l;
 906                                 break;
 907                         case 'U':       /* user id */
 908                                 l = ksprintf(buf, "%u", uid);
 909                                 if ((n + l) > MAXPATHLEN) {
 910                                         log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
 911                                             pid, name, uid, temp, name);
 912                                         kfree(temp, M_TEMP);
 913                                         return NULL;
 914                                 }
 915                                 memcpy(temp+n, buf, l);
 916                                 n += l;
 917                                 break;
 918                         default:
 919                                 log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
 920                         }
 921                         break;
 922                 default:
 923                         temp[n++] = format[i];
 924                 }
 925         }
 926         temp[n] = '\0';
 927         return temp;
 928 }
 929