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