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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <libzfs.h> 29 30 #undef verify /* both libzfs.h and zfs_context.h want to define this */ 31 32 #include <sys/zfs_context.h> 33 34 #include <errno.h> 35 #include <fcntl.h> 36 #include <stdarg.h> 37 #include <stddef.h> 38 #include <stdio.h> 39 #include <stdlib.h> 40 #include <strings.h> 41 #include <sys/file.h> 42 #include <sys/mntent.h> 43 #include <sys/mnttab.h> 44 #include <sys/param.h> 45 #include <sys/stat.h> 46 47 #include <sys/dmu.h> 48 #include <sys/dmu_objset.h> 49 #include <sys/dnode.h> 50 51 #include <sys/mkdev.h> 52 53 #include "zinject.h" 54 55 extern void kernel_init(int); 56 extern void kernel_fini(void); 57 58 static int debug; 59 60 static void 61 ziprintf(const char *fmt, ...) 62 { 63 va_list ap; 64 65 if (!debug) 66 return; 67 68 va_start(ap, fmt); 69 (void) vprintf(fmt, ap); 70 va_end(ap); 71 } 72 73 /* 74 * Given a full path to a file, translate into a dataset name and a relative 75 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters, 76 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64 77 * buffer, which we need later to get the object ID. 78 */ 79 static int 80 parse_pathname(const char *fullpath, char *dataset, char *relpath, 81 struct stat64 *statbuf) 82 { 83 struct extmnttab mp; 84 FILE *fp; 85 int match; 86 const char *rel; 87 88 if (fullpath[0] != '/') { 89 (void) fprintf(stderr, "invalid object '%s': must be full " 90 "path\n", fullpath); 91 usage(); 92 return (-1); 93 } 94 95 if (strlen(fullpath) >= MAXPATHLEN) { 96 (void) fprintf(stderr, "invalid object; pathname too long\n"); 97 return (-1); 98 } 99 100 if (stat64(fullpath, statbuf) != 0) { 101 (void) fprintf(stderr, "cannot open '%s': %s\n", 102 fullpath, strerror(errno)); 103 return (-1); 104 } 105 106 if ((fp = fopen(MNTTAB, "r")) == NULL) { 107 (void) fprintf(stderr, "cannot open /etc/mnttab\n"); 108 return (-1); 109 } 110 111 match = 0; 112 while (getextmntent(fp, &mp, sizeof (mp)) == 0) { 113 if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) { 114 match = 1; 115 break; 116 } 117 } 118 119 if (!match) { 120 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n", 121 fullpath); 122 return (-1); 123 } 124 125 if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) { 126 (void) fprintf(stderr, "invalid path '%s': not a ZFS " 127 "filesystem\n", fullpath); 128 return (-1); 129 } 130 131 if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) { 132 (void) fprintf(stderr, "invalid path '%s': mountpoint " 133 "doesn't match path\n", fullpath); 134 return (-1); 135 } 136 137 (void) strcpy(dataset, mp.mnt_special); 138 139 rel = fullpath + strlen(mp.mnt_mountp); 140 if (rel[0] == '/') 141 rel++; 142 (void) strcpy(relpath, rel); 143 144 return (0); 145 } 146 147 /* 148 * Convert from a (dataset, path) pair into a (objset, object) pair. Note that 149 * we grab the object number from the inode number, since looking this up via 150 * libzpool is a real pain. 151 */ 152 /* ARGSUSED */ 153 static int 154 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf, 155 zinject_record_t *record) 156 { 157 objset_t *os; 158 int err; 159 160 /* 161 * Before doing any libzpool operations, call sync() to ensure that the 162 * on-disk state is consistent with the in-core state. 163 */ 164 sync(); 165 166 if ((err = dmu_objset_open(dataset, DMU_OST_ZFS, 167 DS_MODE_STANDARD | DS_MODE_READONLY, &os)) != 0) { 168 (void) fprintf(stderr, "cannot open dataset '%s': %s\n", 169 dataset, strerror(err)); 170 return (-1); 171 } 172 173 record->zi_objset = dmu_objset_id(os); 174 record->zi_object = statbuf->st_ino; 175 176 dmu_objset_close(os); 177 178 return (0); 179 } 180 181 /* 182 * Calculate the real range based on the type, level, and range given. 183 */ 184 static int 185 calculate_range(const char *dataset, err_type_t type, int level, char *range, 186 zinject_record_t *record) 187 { 188 objset_t *os = NULL; 189 dnode_t *dn = NULL; 190 int err; 191 int ret = -1; 192 193 /* 194 * Determine the numeric range from the string. 195 */ 196 if (range == NULL) { 197 /* 198 * If range is unspecified, set the range to [0,-1], which 199 * indicates that the whole object should be treated as an 200 * error. 201 */ 202 record->zi_start = 0; 203 record->zi_end = -1ULL; 204 } else { 205 char *end; 206 207 /* XXX add support for suffixes */ 208 record->zi_start = strtoull(range, &end, 10); 209 210 211 if (*end == '\0') 212 record->zi_end = record->zi_start + 1; 213 else if (*end == ',') 214 record->zi_end = strtoull(end + 1, &end, 10); 215 216 if (*end != '\0') { 217 (void) fprintf(stderr, "invalid range '%s': must be " 218 "a numeric range of the form 'start[,end]'\n", 219 range); 220 goto out; 221 } 222 } 223 224 switch (type) { 225 case TYPE_DATA: 226 break; 227 228 case TYPE_DNODE: 229 /* 230 * If this is a request to inject faults into the dnode, then we 231 * must translate the current (objset,object) pair into an 232 * offset within the metadnode for the objset. Specifying any 233 * kind of range with type 'dnode' is illegal. 234 */ 235 if (range != NULL) { 236 (void) fprintf(stderr, "range cannot be specified when " 237 "type is 'dnode'\n"); 238 goto out; 239 } 240 241 record->zi_start = record->zi_object * sizeof (dnode_phys_t); 242 record->zi_end = record->zi_start + sizeof (dnode_phys_t); 243 record->zi_object = 0; 244 break; 245 } 246 247 /* 248 * Get the dnode associated with object, so we can calculate the block 249 * size. 250 */ 251 if ((err = dmu_objset_open(dataset, DMU_OST_ANY, 252 DS_MODE_STANDARD | DS_MODE_READONLY, &os)) != 0) { 253 (void) fprintf(stderr, "cannot open dataset '%s': %s\n", 254 dataset, strerror(err)); 255 goto out; 256 } 257 258 if (record->zi_object == 0) { 259 dn = os->os->os_meta_dnode; 260 } else { 261 err = dnode_hold(os->os, record->zi_object, FTAG, &dn); 262 if (err != 0) { 263 (void) fprintf(stderr, "failed to hold dnode " 264 "for object %llu\n", 265 (u_longlong_t)record->zi_object); 266 goto out; 267 } 268 } 269 270 271 ziprintf("data shift: %d\n", (int)dn->dn_datablkshift); 272 ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift); 273 274 /* 275 * Translate range into block IDs. 276 */ 277 if (record->zi_start != 0 || record->zi_end != -1ULL) { 278 record->zi_start >>= dn->dn_datablkshift; 279 record->zi_end >>= dn->dn_datablkshift; 280 } 281 282 /* 283 * Check level, and then translate level 0 blkids into ranges 284 * appropriate for level of indirection. 285 */ 286 record->zi_level = level; 287 if (level > 0) { 288 ziprintf("level 0 blkid range: [%llu, %llu]\n", 289 record->zi_start, record->zi_end); 290 291 if (level >= dn->dn_nlevels) { 292 (void) fprintf(stderr, "level %d exceeds max level " 293 "of object (%d)\n", level, dn->dn_nlevels - 1); 294 goto out; 295 } 296 297 if (record->zi_start != 0 || record->zi_end != 0) { 298 int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT; 299 300 for (; level > 0; level--) { 301 record->zi_start >>= shift; 302 record->zi_end >>= shift; 303 } 304 } 305 } 306 307 ret = 0; 308 out: 309 if (dn) { 310 if (dn != os->os->os_meta_dnode) 311 dnode_rele(dn, FTAG); 312 } 313 if (os) 314 dmu_objset_close(os); 315 316 return (ret); 317 } 318 319 int 320 translate_record(err_type_t type, const char *object, const char *range, 321 int level, zinject_record_t *record, char *poolname, char *dataset) 322 { 323 char path[MAXPATHLEN]; 324 char *slash; 325 struct stat64 statbuf; 326 int ret = -1; 327 328 kernel_init(FREAD); 329 330 debug = (getenv("ZINJECT_DEBUG") != NULL); 331 332 ziprintf("translating: %s\n", object); 333 334 if (MOS_TYPE(type)) { 335 /* 336 * MOS objects are treated specially. 337 */ 338 switch (type) { 339 case TYPE_MOS: 340 record->zi_type = 0; 341 break; 342 case TYPE_MOSDIR: 343 record->zi_type = DMU_OT_OBJECT_DIRECTORY; 344 break; 345 case TYPE_METASLAB: 346 record->zi_type = DMU_OT_OBJECT_ARRAY; 347 break; 348 case TYPE_CONFIG: 349 record->zi_type = DMU_OT_PACKED_NVLIST; 350 break; 351 case TYPE_BPLIST: 352 record->zi_type = DMU_OT_BPLIST; 353 break; 354 case TYPE_SPACEMAP: 355 record->zi_type = DMU_OT_SPACE_MAP; 356 break; 357 case TYPE_ERRLOG: 358 record->zi_type = DMU_OT_ERROR_LOG; 359 break; 360 } 361 362 dataset[0] = '\0'; 363 (void) strcpy(poolname, object); 364 return (0); 365 } 366 367 /* 368 * Convert a full path into a (dataset, file) pair. 369 */ 370 if (parse_pathname(object, dataset, path, &statbuf) != 0) 371 goto err; 372 373 ziprintf(" dataset: %s\n", dataset); 374 ziprintf(" path: %s\n", path); 375 376 /* 377 * Convert (dataset, file) into (objset, object) 378 */ 379 if (object_from_path(dataset, path, &statbuf, record) != 0) 380 goto err; 381 382 ziprintf("raw objset: %llu\n", record->zi_objset); 383 ziprintf("raw object: %llu\n", record->zi_object); 384 385 /* 386 * For the given object, calculate the real (type, level, range) 387 */ 388 if (calculate_range(dataset, type, level, (char *)range, record) != 0) 389 goto err; 390 391 ziprintf(" objset: %llu\n", record->zi_objset); 392 ziprintf(" object: %llu\n", record->zi_object); 393 if (record->zi_start == 0 && 394 record->zi_end == -1ULL) 395 ziprintf(" range: all\n"); 396 else 397 ziprintf(" range: [%llu, %llu]\n", record->zi_start, 398 record->zi_end); 399 400 /* 401 * Copy the pool name 402 */ 403 (void) strcpy(poolname, dataset); 404 if ((slash = strchr(poolname, '/')) != NULL) 405 *slash = '\0'; 406 407 ret = 0; 408 409 err: 410 kernel_fini(); 411 return (ret); 412 } 413 414 int 415 translate_raw(const char *str, zinject_record_t *record) 416 { 417 /* 418 * A raw bookmark of the form objset:object:level:blkid, where each 419 * number is a hexidecimal value. 420 */ 421 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset, 422 (u_longlong_t *)&record->zi_object, &record->zi_level, 423 (u_longlong_t *)&record->zi_start) != 4) { 424 (void) fprintf(stderr, "bad raw spec '%s': must be of the form " 425 "'objset:object:level:blkid'\n", str); 426 return (-1); 427 } 428 429 record->zi_end = record->zi_start; 430 431 return (0); 432 } 433 434 int 435 translate_device(const char *pool, const char *device, 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); 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 return (0); 464 } 465