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