1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <libzfs.h> 27 28 #undef verify /* both libzfs.h and zfs_context.h want to define this */ 29 30 #include <sys/zfs_context.h> 31 32 #include <errno.h> 33 #include <fcntl.h> 34 #include <stdarg.h> 35 #include <stddef.h> 36 #include <stdio.h> 37 #include <stdlib.h> 38 #include <strings.h> 39 #include <sys/file.h> 40 #include <sys/mntent.h> 41 #include <sys/mnttab.h> 42 #include <sys/param.h> 43 #include <sys/stat.h> 44 45 #include <sys/dmu.h> 46 #include <sys/dmu_objset.h> 47 #include <sys/dnode.h> 48 #include <sys/vdev_impl.h> 49 50 #include <sys/mkdev.h> 51 52 #include "zinject.h" 53 54 extern void kernel_init(int); 55 extern void kernel_fini(void); 56 57 static int debug; 58 59 static void 60 ziprintf(const char *fmt, ...) 61 { 62 va_list ap; 63 64 if (!debug) 65 return; 66 67 va_start(ap, fmt); 68 (void) vprintf(fmt, ap); 69 va_end(ap); 70 } 71 72 /* 73 * Given a full path to a file, translate into a dataset name and a relative 74 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters, 75 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64 76 * buffer, which we need later to get the object ID. 77 */ 78 static int 79 parse_pathname(const char *fullpath, char *dataset, char *relpath, 80 struct stat64 *statbuf) 81 { 82 struct extmnttab mp; 83 FILE *fp; 84 int match; 85 const char *rel; 86 87 if (fullpath[0] != '/') { 88 (void) fprintf(stderr, "invalid object '%s': must be full " 89 "path\n", fullpath); 90 usage(); 91 return (-1); 92 } 93 94 if (strlen(fullpath) >= MAXPATHLEN) { 95 (void) fprintf(stderr, "invalid object; pathname too long\n"); 96 return (-1); 97 } 98 99 if (stat64(fullpath, statbuf) != 0) { 100 (void) fprintf(stderr, "cannot open '%s': %s\n", 101 fullpath, strerror(errno)); 102 return (-1); 103 } 104 105 if ((fp = fopen(MNTTAB, "r")) == NULL) { 106 (void) fprintf(stderr, "cannot open /etc/mnttab\n"); 107 return (-1); 108 } 109 110 match = 0; 111 while (getextmntent(fp, &mp, sizeof (mp)) == 0) { 112 if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) { 113 match = 1; 114 break; 115 } 116 } 117 118 if (!match) { 119 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n", 120 fullpath); 121 return (-1); 122 } 123 124 if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) { 125 (void) fprintf(stderr, "invalid path '%s': not a ZFS " 126 "filesystem\n", fullpath); 127 return (-1); 128 } 129 130 if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) { 131 (void) fprintf(stderr, "invalid path '%s': mountpoint " 132 "doesn't match path\n", fullpath); 133 return (-1); 134 } 135 136 (void) strcpy(dataset, mp.mnt_special); 137 138 rel = fullpath + strlen(mp.mnt_mountp); 139 if (rel[0] == '/') 140 rel++; 141 (void) strcpy(relpath, rel); 142 143 return (0); 144 } 145 146 /* 147 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that 148 * we grab the object number from the inode number, since looking this up via 149 * libzpool is a real pain. 150 */ 151 /* ARGSUSED */ 152 static int 153 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf, 154 zinject_record_t *record) 155 { 156 objset_t *os; 157 int err; 158 159 /* 160 * Before doing any libzpool operations, call sync() to ensure that the 161 * on-disk state is consistent with the in-core state. 162 */ 163 sync(); 164 165 err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os); 166 if (err != 0) { 167 (void) fprintf(stderr, "cannot open dataset '%s': %s\n", 168 dataset, strerror(err)); 169 return (-1); 170 } 171 172 record->zi_objset = dmu_objset_id(os); 173 record->zi_object = statbuf->st_ino; 174 175 dmu_objset_disown(os, FTAG); 176 177 return (0); 178 } 179 180 /* 181 * Calculate the real range based on the type, level, and range given. 182 */ 183 static int 184 calculate_range(const char *dataset, err_type_t type, int level, char *range, 185 zinject_record_t *record) 186 { 187 objset_t *os = NULL; 188 dnode_t *dn = NULL; 189 int err; 190 int ret = -1; 191 192 /* 193 * Determine the numeric range from the string. 194 */ 195 if (range == NULL) { 196 /* 197 * If range is unspecified, set the range to [0,-1], which 198 * indicates that the whole object should be treated as an 199 * error. 200 */ 201 record->zi_start = 0; 202 record->zi_end = -1ULL; 203 } else { 204 char *end; 205 206 /* XXX add support for suffixes */ 207 record->zi_start = strtoull(range, &end, 10); 208 209 210 if (*end == '\0') 211 record->zi_end = record->zi_start + 1; 212 else if (*end == ',') 213 record->zi_end = strtoull(end + 1, &end, 10); 214 215 if (*end != '\0') { 216 (void) fprintf(stderr, "invalid range '%s': must be " 217 "a numeric range of the form 'start[,end]'\n", 218 range); 219 goto out; 220 } 221 } 222 223 switch (type) { 224 case TYPE_DATA: 225 break; 226 227 case TYPE_DNODE: 228 /* 229 * If this is a request to inject faults into the dnode, then we 230 * must translate the current (objset,object) pair into an 231 * offset within the metadnode for the objset. Specifying any 232 * kind of range with type 'dnode' is illegal. 233 */ 234 if (range != NULL) { 235 (void) fprintf(stderr, "range cannot be specified when " 236 "type is 'dnode'\n"); 237 goto out; 238 } 239 240 record->zi_start = record->zi_object * sizeof (dnode_phys_t); 241 record->zi_end = record->zi_start + sizeof (dnode_phys_t); 242 record->zi_object = 0; 243 break; 244 } 245 246 /* 247 * Get the dnode associated with object, so we can calculate the block 248 * size. 249 */ 250 if ((err = dmu_objset_own(dataset, DMU_OST_ANY, 251 B_TRUE, FTAG, &os)) != 0) { 252 (void) fprintf(stderr, "cannot open dataset '%s': %s\n", 253 dataset, strerror(err)); 254 goto out; 255 } 256 257 if (record->zi_object == 0) { 258 dn = os->os_meta_dnode; 259 } else { 260 err = dnode_hold(os, record->zi_object, FTAG, &dn); 261 if (err != 0) { 262 (void) fprintf(stderr, "failed to hold dnode " 263 "for object %llu\n", 264 (u_longlong_t)record->zi_object); 265 goto out; 266 } 267 } 268 269 270 ziprintf("data shift: %d\n", (int)dn->dn_datablkshift); 271 ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift); 272 273 /* 274 * Translate range into block IDs. 275 */ 276 if (record->zi_start != 0 || record->zi_end != -1ULL) { 277 record->zi_start >>= dn->dn_datablkshift; 278 record->zi_end >>= dn->dn_datablkshift; 279 } 280 281 /* 282 * Check level, and then translate level 0 blkids into ranges 283 * appropriate for level of indirection. 284 */ 285 record->zi_level = level; 286 if (level > 0) { 287 ziprintf("level 0 blkid range: [%llu, %llu]\n", 288 record->zi_start, record->zi_end); 289 290 if (level >= dn->dn_nlevels) { 291 (void) fprintf(stderr, "level %d exceeds max level " 292 "of object (%d)\n", level, dn->dn_nlevels - 1); 293 goto out; 294 } 295 296 if (record->zi_start != 0 || record->zi_end != 0) { 297 int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 298 299 for (; level > 0; level--) { 300 record->zi_start >>= shift; 301 record->zi_end >>= shift; 302 } 303 } 304 } 305 306 ret = 0; 307 out: 308 if (dn) { 309 if (dn != os->os_meta_dnode) 310 dnode_rele(dn, FTAG); 311 } 312 if (os) 313 dmu_objset_disown(os, FTAG); 314 315 return (ret); 316 } 317 318 int 319 translate_record(err_type_t type, const char *object, const char *range, 320 int level, zinject_record_t *record, char *poolname, char *dataset) 321 { 322 char path[MAXPATHLEN]; 323 char *slash; 324 struct stat64 statbuf; 325 int ret = -1; 326 327 kernel_init(FREAD); 328 329 debug = (getenv("ZINJECT_DEBUG") != NULL); 330 331 ziprintf("translating: %s\n", object); 332 333 if (MOS_TYPE(type)) { 334 /* 335 * MOS objects are treated specially. 336 */ 337 switch (type) { 338 case TYPE_MOS: 339 record->zi_type = 0; 340 break; 341 case TYPE_MOSDIR: 342 record->zi_type = DMU_OT_OBJECT_DIRECTORY; 343 break; 344 case TYPE_METASLAB: 345 record->zi_type = DMU_OT_OBJECT_ARRAY; 346 break; 347 case TYPE_CONFIG: 348 record->zi_type = DMU_OT_PACKED_NVLIST; 349 break; 350 case TYPE_BPLIST: 351 record->zi_type = DMU_OT_BPLIST; 352 break; 353 case TYPE_SPACEMAP: 354 record->zi_type = DMU_OT_SPACE_MAP; 355 break; 356 case TYPE_ERRLOG: 357 record->zi_type = DMU_OT_ERROR_LOG; 358 break; 359 } 360 361 dataset[0] = '\0'; 362 (void) strcpy(poolname, object); 363 return (0); 364 } 365 366 /* 367 * Convert a full path into a (dataset, file) pair. 368 */ 369 if (parse_pathname(object, dataset, path, &statbuf) != 0) 370 goto err; 371 372 ziprintf(" dataset: %s\n", dataset); 373 ziprintf(" path: %s\n", path); 374 375 /* 376 * Convert (dataset, file) into (objset, object) 377 */ 378 if (object_from_path(dataset, path, &statbuf, record) != 0) 379 goto err; 380 381 ziprintf("raw objset: %llu\n", record->zi_objset); 382 ziprintf("raw object: %llu\n", record->zi_object); 383 384 /* 385 * For the given object, calculate the real (type, level, range) 386 */ 387 if (calculate_range(dataset, type, level, (char *)range, record) != 0) 388 goto err; 389 390 ziprintf(" objset: %llu\n", record->zi_objset); 391 ziprintf(" object: %llu\n", record->zi_object); 392 if (record->zi_start == 0 && 393 record->zi_end == -1ULL) 394 ziprintf(" range: all\n"); 395 else 396 ziprintf(" range: [%llu, %llu]\n", record->zi_start, 397 record->zi_end); 398 399 /* 400 * Copy the pool name 401 */ 402 (void) strcpy(poolname, dataset); 403 if ((slash = strchr(poolname, '/')) != NULL) 404 *slash = '\0'; 405 406 ret = 0; 407 408 err: 409 kernel_fini(); 410 return (ret); 411 } 412 413 int 414 translate_raw(const char *str, zinject_record_t *record) 415 { 416 /* 417 * A raw bookmark of the form objset:object:level:blkid, where each 418 * number is a hexidecimal value. 419 */ 420 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset, 421 (u_longlong_t *)&record->zi_object, &record->zi_level, 422 (u_longlong_t *)&record->zi_start) != 4) { 423 (void) fprintf(stderr, "bad raw spec '%s': must be of the form " 424 "'objset:object:level:blkid'\n", str); 425 return (-1); 426 } 427 428 record->zi_end = record->zi_start; 429 430 return (0); 431 } 432 433 int 434 translate_device(const char *pool, const char *device, err_type_t label_type, 435 zinject_record_t *record) 436 { 437 char *end; 438 zpool_handle_t *zhp; 439 nvlist_t *tgt; 440 boolean_t isspare, iscache; 441 442 /* 443 * Given a device name or GUID, create an appropriate injection record 444 * with zi_guid set. 445 */ 446 if ((zhp = zpool_open(g_zfs, pool)) == NULL) 447 return (-1); 448 449 record->zi_guid = strtoull(device, &end, 16); 450 if (record->zi_guid == 0 || *end != '\0') { 451 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL); 452 453 if (tgt == NULL) { 454 (void) fprintf(stderr, "cannot find device '%s' in " 455 "pool '%s'\n", device, pool); 456 return (-1); 457 } 458 459 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, 460 &record->zi_guid) == 0); 461 } 462 463 switch (label_type) { 464 case TYPE_LABEL_UBERBLOCK: 465 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]); 466 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1; 467 break; 468 case TYPE_LABEL_NVLIST: 469 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys); 470 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1; 471 break; 472 } 473 return (0); 474 } 475