1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2008-2009, Stacey Son <sson@freebsd.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/param.h> 32 #include <sys/systm.h> 33 #include <sys/kernel.h> 34 35 #include <sys/conf.h> 36 #include <sys/elf.h> 37 #include <sys/linker.h> 38 #include <sys/malloc.h> 39 #include <sys/mman.h> 40 #include <sys/module.h> 41 #include <sys/proc.h> 42 #include <sys/queue.h> 43 #include <sys/resourcevar.h> 44 #include <sys/rwlock.h> 45 #include <sys/stat.h> 46 #include <sys/sx.h> 47 #include <sys/uio.h> 48 49 #include <machine/elf.h> 50 51 #include <vm/pmap.h> 52 #include <vm/vm.h> 53 #include <vm/vm_extern.h> 54 #include <vm/vm_object.h> 55 #include <vm/vm_page.h> 56 #include <vm/vm_pager.h> 57 58 #include "linker_if.h" 59 60 #define SHDR_NULL 0 61 #define SHDR_SYMTAB 1 62 #define SHDR_STRTAB 2 63 #define SHDR_SHSTRTAB 3 64 65 #define SHDR_NUM 4 66 67 #define STR_SYMTAB ".symtab" 68 #define STR_STRTAB ".strtab" 69 #define STR_SHSTRTAB ".shstrtab" 70 71 #define KSYMS_DNAME "ksyms" 72 73 static d_open_t ksyms_open; 74 static d_read_t ksyms_read; 75 static d_mmap_single_t ksyms_mmap_single; 76 77 static struct cdevsw ksyms_cdevsw = { 78 .d_version = D_VERSION, 79 .d_flags = 0, 80 .d_open = ksyms_open, 81 .d_read = ksyms_read, 82 .d_mmap_single = ksyms_mmap_single, 83 .d_name = KSYMS_DNAME 84 }; 85 86 struct ksyms_softc { 87 LIST_ENTRY(ksyms_softc) sc_list; 88 vm_offset_t sc_uaddr; 89 size_t sc_usize; 90 vm_object_t sc_obj; 91 vm_size_t sc_objsz; 92 struct proc *sc_proc; 93 }; 94 95 static struct sx ksyms_mtx; 96 static struct cdev *ksyms_dev; 97 static LIST_HEAD(, ksyms_softc) ksyms_list = LIST_HEAD_INITIALIZER(ksyms_list); 98 99 static const char ksyms_shstrtab[] = 100 "\0" STR_SYMTAB "\0" STR_STRTAB "\0" STR_SHSTRTAB "\0"; 101 102 struct ksyms_hdr { 103 Elf_Ehdr kh_ehdr; 104 Elf_Phdr kh_txtphdr; 105 Elf_Phdr kh_datphdr; 106 Elf_Shdr kh_shdr[SHDR_NUM]; 107 char kh_shstrtab[sizeof(ksyms_shstrtab)]; 108 }; 109 110 struct tsizes { 111 size_t ts_symsz; 112 size_t ts_strsz; 113 }; 114 115 struct toffsets { 116 struct ksyms_softc *to_sc; 117 vm_offset_t to_symoff; 118 vm_offset_t to_stroff; 119 unsigned to_stridx; 120 size_t to_resid; 121 }; 122 123 static MALLOC_DEFINE(M_KSYMS, "KSYMS", "Kernel Symbol Table"); 124 125 /* 126 * Get the symbol and string table sizes for a kernel module. Add it to the 127 * running total. 128 */ 129 static int 130 ksyms_size_permod(linker_file_t lf, void *arg) 131 { 132 struct tsizes *ts; 133 const Elf_Sym *symtab; 134 caddr_t strtab; 135 long syms; 136 137 ts = arg; 138 139 syms = LINKER_SYMTAB_GET(lf, &symtab); 140 ts->ts_symsz += syms * sizeof(Elf_Sym); 141 ts->ts_strsz += LINKER_STRTAB_GET(lf, &strtab); 142 143 return (0); 144 } 145 146 /* 147 * For kernel module get the symbol and string table sizes, returning the 148 * totals in *ts. 149 */ 150 static void 151 ksyms_size_calc(struct tsizes *ts) 152 { 153 154 ts->ts_symsz = 0; 155 ts->ts_strsz = 0; 156 157 (void)linker_file_foreach(ksyms_size_permod, ts); 158 } 159 160 static int 161 ksyms_emit(struct ksyms_softc *sc, void *buf, off_t off, size_t sz) 162 { 163 struct iovec iov; 164 struct uio uio; 165 166 iov.iov_base = buf; 167 iov.iov_len = sz; 168 uio.uio_iov = &iov; 169 uio.uio_iovcnt = 1; 170 uio.uio_offset = off; 171 uio.uio_resid = (ssize_t)sz; 172 uio.uio_segflg = UIO_SYSSPACE; 173 uio.uio_rw = UIO_WRITE; 174 uio.uio_td = curthread; 175 176 return (uiomove_object(sc->sc_obj, sc->sc_objsz, &uio)); 177 } 178 179 #define SYMBLKSZ (256 * sizeof(Elf_Sym)) 180 181 /* 182 * For a kernel module, add the symbol and string tables into the 183 * snapshot buffer. Fix up the offsets in the tables. 184 */ 185 static int 186 ksyms_add(linker_file_t lf, void *arg) 187 { 188 char *buf; 189 struct ksyms_softc *sc; 190 struct toffsets *to; 191 const Elf_Sym *symtab; 192 Elf_Sym *symp; 193 caddr_t strtab; 194 size_t len, numsyms, strsz, symsz; 195 linker_symval_t symval; 196 int error, i, nsyms; 197 bool fixup; 198 199 buf = malloc(SYMBLKSZ, M_KSYMS, M_WAITOK); 200 to = arg; 201 sc = to->to_sc; 202 203 MOD_SLOCK; 204 numsyms = LINKER_SYMTAB_GET(lf, &symtab); 205 strsz = LINKER_STRTAB_GET(lf, &strtab); 206 symsz = numsyms * sizeof(Elf_Sym); 207 208 #ifdef RELOCATABLE_KERNEL 209 fixup = true; 210 #else 211 fixup = lf->id > 1; 212 #endif 213 214 while (symsz > 0) { 215 len = min(SYMBLKSZ, symsz); 216 bcopy(symtab, buf, len); 217 218 /* 219 * Fix up symbol table for kernel modules: 220 * string offsets need adjusted 221 * symbol values made absolute 222 */ 223 symp = (Elf_Sym *) buf; 224 nsyms = len / sizeof(Elf_Sym); 225 for (i = 0; i < nsyms; i++) { 226 symp[i].st_name += to->to_stridx; 227 if (fixup && LINKER_SYMBOL_VALUES(lf, 228 (c_linker_sym_t)&symtab[i], &symval) == 0) { 229 symp[i].st_value = (uintptr_t)symval.value; 230 } 231 } 232 233 if (len > to->to_resid) { 234 MOD_SUNLOCK; 235 free(buf, M_KSYMS); 236 return (ENXIO); 237 } 238 to->to_resid -= len; 239 error = ksyms_emit(sc, buf, to->to_symoff, len); 240 to->to_symoff += len; 241 if (error != 0) { 242 MOD_SUNLOCK; 243 free(buf, M_KSYMS); 244 return (error); 245 } 246 247 symtab += nsyms; 248 symsz -= len; 249 } 250 free(buf, M_KSYMS); 251 MOD_SUNLOCK; 252 253 if (strsz > to->to_resid) 254 return (ENXIO); 255 to->to_resid -= strsz; 256 error = ksyms_emit(sc, strtab, to->to_stroff, strsz); 257 to->to_stroff += strsz; 258 to->to_stridx += strsz; 259 260 return (error); 261 } 262 263 /* 264 * Create a single ELF symbol table for the kernel and kernel modules loaded 265 * at this time. Write this snapshot out in the process address space. Return 266 * 0 on success, otherwise error. 267 */ 268 static int 269 ksyms_snapshot(struct ksyms_softc *sc, struct tsizes *ts) 270 { 271 struct toffsets to; 272 struct ksyms_hdr *hdr; 273 int error; 274 275 hdr = malloc(sizeof(*hdr), M_KSYMS, M_WAITOK | M_ZERO); 276 277 /* 278 * Create the ELF header. 279 */ 280 hdr->kh_ehdr.e_ident[EI_PAD] = 0; 281 hdr->kh_ehdr.e_ident[EI_MAG0] = ELFMAG0; 282 hdr->kh_ehdr.e_ident[EI_MAG1] = ELFMAG1; 283 hdr->kh_ehdr.e_ident[EI_MAG2] = ELFMAG2; 284 hdr->kh_ehdr.e_ident[EI_MAG3] = ELFMAG3; 285 hdr->kh_ehdr.e_ident[EI_DATA] = ELF_DATA; 286 hdr->kh_ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 287 hdr->kh_ehdr.e_ident[EI_CLASS] = ELF_CLASS; 288 hdr->kh_ehdr.e_ident[EI_VERSION] = EV_CURRENT; 289 hdr->kh_ehdr.e_ident[EI_ABIVERSION] = 0; 290 hdr->kh_ehdr.e_type = ET_EXEC; 291 hdr->kh_ehdr.e_machine = ELF_ARCH; 292 hdr->kh_ehdr.e_version = EV_CURRENT; 293 hdr->kh_ehdr.e_entry = 0; 294 hdr->kh_ehdr.e_phoff = offsetof(struct ksyms_hdr, kh_txtphdr); 295 hdr->kh_ehdr.e_shoff = offsetof(struct ksyms_hdr, kh_shdr); 296 hdr->kh_ehdr.e_flags = 0; 297 hdr->kh_ehdr.e_ehsize = sizeof(Elf_Ehdr); 298 hdr->kh_ehdr.e_phentsize = sizeof(Elf_Phdr); 299 hdr->kh_ehdr.e_phnum = 2; /* Text and Data */ 300 hdr->kh_ehdr.e_shentsize = sizeof(Elf_Shdr); 301 hdr->kh_ehdr.e_shnum = SHDR_NUM; 302 hdr->kh_ehdr.e_shstrndx = SHDR_SHSTRTAB; 303 304 /* 305 * Add both the text and data program headers. 306 */ 307 hdr->kh_txtphdr.p_type = PT_LOAD; 308 /* XXX - is there a way to put the actual .text addr/size here? */ 309 hdr->kh_txtphdr.p_vaddr = 0; 310 hdr->kh_txtphdr.p_memsz = 0; 311 hdr->kh_txtphdr.p_flags = PF_R | PF_X; 312 313 hdr->kh_datphdr.p_type = PT_LOAD; 314 /* XXX - is there a way to put the actual .data addr/size here? */ 315 hdr->kh_datphdr.p_vaddr = 0; 316 hdr->kh_datphdr.p_memsz = 0; 317 hdr->kh_datphdr.p_flags = PF_R | PF_W | PF_X; 318 319 /* 320 * Add the section headers: null, symtab, strtab, shstrtab. 321 */ 322 323 /* First section header - null */ 324 325 /* Second section header - symtab */ 326 hdr->kh_shdr[SHDR_SYMTAB].sh_name = 1; /* String offset (skip null) */ 327 hdr->kh_shdr[SHDR_SYMTAB].sh_type = SHT_SYMTAB; 328 hdr->kh_shdr[SHDR_SYMTAB].sh_flags = 0; 329 hdr->kh_shdr[SHDR_SYMTAB].sh_addr = 0; 330 hdr->kh_shdr[SHDR_SYMTAB].sh_offset = sizeof(*hdr); 331 hdr->kh_shdr[SHDR_SYMTAB].sh_size = ts->ts_symsz; 332 hdr->kh_shdr[SHDR_SYMTAB].sh_link = SHDR_STRTAB; 333 hdr->kh_shdr[SHDR_SYMTAB].sh_info = ts->ts_symsz / sizeof(Elf_Sym); 334 hdr->kh_shdr[SHDR_SYMTAB].sh_addralign = sizeof(long); 335 hdr->kh_shdr[SHDR_SYMTAB].sh_entsize = sizeof(Elf_Sym); 336 337 /* Third section header - strtab */ 338 hdr->kh_shdr[SHDR_STRTAB].sh_name = 1 + sizeof(STR_SYMTAB); 339 hdr->kh_shdr[SHDR_STRTAB].sh_type = SHT_STRTAB; 340 hdr->kh_shdr[SHDR_STRTAB].sh_flags = 0; 341 hdr->kh_shdr[SHDR_STRTAB].sh_addr = 0; 342 hdr->kh_shdr[SHDR_STRTAB].sh_offset = 343 hdr->kh_shdr[SHDR_SYMTAB].sh_offset + ts->ts_symsz; 344 hdr->kh_shdr[SHDR_STRTAB].sh_size = ts->ts_strsz; 345 hdr->kh_shdr[SHDR_STRTAB].sh_link = 0; 346 hdr->kh_shdr[SHDR_STRTAB].sh_info = 0; 347 hdr->kh_shdr[SHDR_STRTAB].sh_addralign = sizeof(char); 348 hdr->kh_shdr[SHDR_STRTAB].sh_entsize = 0; 349 350 /* Fourth section - shstrtab */ 351 hdr->kh_shdr[SHDR_SHSTRTAB].sh_name = 1 + sizeof(STR_SYMTAB) + 352 sizeof(STR_STRTAB); 353 hdr->kh_shdr[SHDR_SHSTRTAB].sh_type = SHT_STRTAB; 354 hdr->kh_shdr[SHDR_SHSTRTAB].sh_flags = 0; 355 hdr->kh_shdr[SHDR_SHSTRTAB].sh_addr = 0; 356 hdr->kh_shdr[SHDR_SHSTRTAB].sh_offset = 357 offsetof(struct ksyms_hdr, kh_shstrtab); 358 hdr->kh_shdr[SHDR_SHSTRTAB].sh_size = sizeof(ksyms_shstrtab); 359 hdr->kh_shdr[SHDR_SHSTRTAB].sh_link = 0; 360 hdr->kh_shdr[SHDR_SHSTRTAB].sh_info = 0; 361 hdr->kh_shdr[SHDR_SHSTRTAB].sh_addralign = 0 /* sizeof(char) */; 362 hdr->kh_shdr[SHDR_SHSTRTAB].sh_entsize = 0; 363 364 /* Copy shstrtab into the header. */ 365 bcopy(ksyms_shstrtab, hdr->kh_shstrtab, sizeof(ksyms_shstrtab)); 366 367 to.to_sc = sc; 368 to.to_symoff = hdr->kh_shdr[SHDR_SYMTAB].sh_offset; 369 to.to_stroff = hdr->kh_shdr[SHDR_STRTAB].sh_offset; 370 to.to_stridx = 0; 371 to.to_resid = sc->sc_objsz - sizeof(struct ksyms_hdr); 372 373 /* emit header */ 374 error = ksyms_emit(sc, hdr, 0, sizeof(*hdr)); 375 free(hdr, M_KSYMS); 376 if (error != 0) 377 return (error); 378 379 /* Add symbol and string tables for each kernel module. */ 380 error = linker_file_foreach(ksyms_add, &to); 381 if (error != 0) 382 return (error); 383 if (to.to_resid != 0) 384 return (ENXIO); 385 return (0); 386 } 387 388 static void 389 ksyms_cdevpriv_dtr(void *data) 390 { 391 struct ksyms_softc *sc; 392 vm_object_t obj; 393 394 sc = (struct ksyms_softc *)data; 395 396 sx_xlock(&ksyms_mtx); 397 LIST_REMOVE(sc, sc_list); 398 sx_xunlock(&ksyms_mtx); 399 obj = sc->sc_obj; 400 if (obj != NULL) 401 vm_object_deallocate(obj); 402 free(sc, M_KSYMS); 403 } 404 405 static int 406 ksyms_open(struct cdev *dev, int flags, int fmt __unused, struct thread *td) 407 { 408 struct tsizes ts; 409 struct ksyms_softc *sc; 410 vm_object_t object; 411 vm_size_t elfsz; 412 int error, try; 413 414 /* 415 * Limit one open() per process. The process must close() 416 * before open()'ing again. 417 */ 418 sx_xlock(&ksyms_mtx); 419 LIST_FOREACH(sc, &ksyms_list, sc_list) { 420 if (sc->sc_proc == td->td_proc) { 421 sx_xunlock(&ksyms_mtx); 422 return (EBUSY); 423 } 424 } 425 426 sc = malloc(sizeof(*sc), M_KSYMS, M_WAITOK | M_ZERO); 427 sc->sc_proc = td->td_proc; 428 LIST_INSERT_HEAD(&ksyms_list, sc, sc_list); 429 sx_xunlock(&ksyms_mtx); 430 431 error = devfs_set_cdevpriv(sc, ksyms_cdevpriv_dtr); 432 if (error != 0) { 433 ksyms_cdevpriv_dtr(sc); 434 return (error); 435 } 436 437 /* 438 * MOD_SLOCK doesn't work here (because of a lock reversal with 439 * KLD_SLOCK). Therefore, simply try up to 3 times to get a "clean" 440 * snapshot of the kernel symbol table. This should work fine in the 441 * rare case of a kernel module being loaded/unloaded at the same 442 * time. 443 */ 444 for (try = 0; try < 3; try++) { 445 ksyms_size_calc(&ts); 446 elfsz = sizeof(struct ksyms_hdr) + ts.ts_symsz + ts.ts_strsz; 447 448 object = vm_pager_allocate(OBJT_PHYS, NULL, round_page(elfsz), 449 VM_PROT_ALL, 0, td->td_ucred); 450 sc->sc_obj = object; 451 sc->sc_objsz = elfsz; 452 453 error = ksyms_snapshot(sc, &ts); 454 if (error == 0) 455 break; 456 457 vm_object_deallocate(sc->sc_obj); 458 sc->sc_obj = NULL; 459 } 460 return (error); 461 } 462 463 static int 464 ksyms_read(struct cdev *dev, struct uio *uio, int flags __unused) 465 { 466 struct ksyms_softc *sc; 467 int error; 468 469 error = devfs_get_cdevpriv((void **)&sc); 470 if (error != 0) 471 return (error); 472 return (uiomove_object(sc->sc_obj, sc->sc_objsz, uio)); 473 } 474 475 static int 476 ksyms_mmap_single(struct cdev *dev, vm_ooffset_t *offset, vm_size_t size, 477 vm_object_t *objp, int nprot) 478 { 479 struct ksyms_softc *sc; 480 vm_object_t obj; 481 int error; 482 483 error = devfs_get_cdevpriv((void **)&sc); 484 if (error != 0) 485 return (error); 486 487 if (*offset < 0 || *offset >= round_page(sc->sc_objsz) || 488 size > round_page(sc->sc_objsz) - *offset || 489 (nprot & ~PROT_READ) != 0) 490 return (EINVAL); 491 492 obj = sc->sc_obj; 493 vm_object_reference(obj); 494 *objp = obj; 495 return (0); 496 } 497 498 static int 499 ksyms_modevent(module_t mod __unused, int type, void *data __unused) 500 { 501 int error; 502 503 error = 0; 504 switch (type) { 505 case MOD_LOAD: 506 sx_init(&ksyms_mtx, "KSyms mtx"); 507 ksyms_dev = make_dev(&ksyms_cdevsw, 0, UID_ROOT, GID_WHEEL, 508 0400, KSYMS_DNAME); 509 break; 510 case MOD_UNLOAD: 511 if (!LIST_EMPTY(&ksyms_list)) 512 return (EBUSY); 513 destroy_dev(ksyms_dev); 514 sx_destroy(&ksyms_mtx); 515 break; 516 case MOD_SHUTDOWN: 517 break; 518 default: 519 error = EOPNOTSUPP; 520 break; 521 } 522 return (error); 523 } 524 525 DEV_MODULE(ksyms, ksyms_modevent, NULL); 526 MODULE_VERSION(ksyms, 1); 527