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 (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2020 by Delphix. All rights reserved. 24 */ 25 26 #include <libzfs.h> 27 28 #include <errno.h> 29 #include <fcntl.h> 30 #include <stdarg.h> 31 #include <stddef.h> 32 #include <stdio.h> 33 #include <stdlib.h> 34 #include <strings.h> 35 #include <sys/file.h> 36 #include <sys/mntent.h> 37 #include <sys/mnttab.h> 38 #include <sys/param.h> 39 #include <sys/stat.h> 40 41 #include <sys/dmu.h> 42 #include <sys/dmu_objset.h> 43 #include <sys/dnode.h> 44 #include <sys/vdev_impl.h> 45 46 #include <sys/mkdev.h> 47 48 #include "zinject.h" 49 50 static int debug; 51 52 static void 53 ziprintf(const char *fmt, ...) 54 { 55 va_list ap; 56 57 if (!debug) 58 return; 59 60 va_start(ap, fmt); 61 (void) vprintf(fmt, ap); 62 va_end(ap); 63 } 64 65 static void 66 compress_slashes(const char *src, char *dest) 67 { 68 while (*src != '\0') { 69 *dest = *src++; 70 while (*dest == '/' && *src == '/') 71 ++src; 72 ++dest; 73 } 74 *dest = '\0'; 75 } 76 77 /* 78 * Given a full path to a file, translate into a dataset name and a relative 79 * path within the dataset. 'dataset' must be at least MAXNAMELEN characters, 80 * and 'relpath' must be at least MAXPATHLEN characters. We also pass a stat64 81 * buffer, which we need later to get the object ID. 82 */ 83 static int 84 parse_pathname(const char *inpath, char *dataset, char *relpath, 85 struct stat64 *statbuf) 86 { 87 struct extmnttab mp; 88 const char *rel; 89 char fullpath[MAXPATHLEN]; 90 91 compress_slashes(inpath, fullpath); 92 93 if (fullpath[0] != '/') { 94 (void) fprintf(stderr, "invalid object '%s': must be full " 95 "path\n", fullpath); 96 usage(); 97 return (-1); 98 } 99 100 if (getextmntent(fullpath, &mp, statbuf) != 0) { 101 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n", 102 fullpath); 103 return (-1); 104 } 105 106 if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) { 107 (void) fprintf(stderr, "invalid path '%s': not a ZFS " 108 "filesystem\n", fullpath); 109 return (-1); 110 } 111 112 if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) { 113 (void) fprintf(stderr, "invalid path '%s': mountpoint " 114 "doesn't match path\n", fullpath); 115 return (-1); 116 } 117 118 (void) strcpy(dataset, mp.mnt_special); 119 120 rel = fullpath + strlen(mp.mnt_mountp); 121 if (rel[0] == '/') 122 rel++; 123 (void) strcpy(relpath, rel); 124 125 return (0); 126 } 127 128 /* 129 * Convert from a dataset to a objset id. Note that 130 * we grab the object number from the inode number. 131 */ 132 static int 133 object_from_path(const char *dataset, uint64_t object, zinject_record_t *record) 134 { 135 zfs_handle_t *zhp; 136 137 if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL) 138 return (-1); 139 140 record->zi_objset = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); 141 record->zi_object = object; 142 143 zfs_close(zhp); 144 145 return (0); 146 } 147 148 /* 149 * Initialize the range based on the type, level, and range given. 150 */ 151 static int 152 initialize_range(err_type_t type, int level, char *range, 153 zinject_record_t *record) 154 { 155 /* 156 * Determine the numeric range from the string. 157 */ 158 if (range == NULL) { 159 /* 160 * If range is unspecified, set the range to [0,-1], which 161 * indicates that the whole object should be treated as an 162 * error. 163 */ 164 record->zi_start = 0; 165 record->zi_end = -1ULL; 166 } else { 167 char *end; 168 169 /* XXX add support for suffixes */ 170 record->zi_start = strtoull(range, &end, 10); 171 172 173 if (*end == '\0') 174 record->zi_end = record->zi_start + 1; 175 else if (*end == ',') 176 record->zi_end = strtoull(end + 1, &end, 10); 177 178 if (*end != '\0') { 179 (void) fprintf(stderr, "invalid range '%s': must be " 180 "a numeric range of the form 'start[,end]'\n", 181 range); 182 return (-1); 183 } 184 } 185 186 switch (type) { 187 default: 188 break; 189 case TYPE_DATA: 190 break; 191 192 case TYPE_DNODE: 193 /* 194 * If this is a request to inject faults into the dnode, then we 195 * must translate the current (objset,object) pair into an 196 * offset within the metadnode for the objset. Specifying any 197 * kind of range with type 'dnode' is illegal. 198 */ 199 if (range != NULL) { 200 (void) fprintf(stderr, "range cannot be specified when " 201 "type is 'dnode'\n"); 202 return (-1); 203 } 204 205 record->zi_start = record->zi_object * sizeof (dnode_phys_t); 206 record->zi_end = record->zi_start + sizeof (dnode_phys_t); 207 record->zi_object = 0; 208 break; 209 } 210 211 record->zi_level = level; 212 213 return (0); 214 } 215 216 int 217 translate_record(err_type_t type, const char *object, const char *range, 218 int level, zinject_record_t *record, char *poolname, char *dataset) 219 { 220 char path[MAXPATHLEN]; 221 char *slash; 222 struct stat64 statbuf; 223 int ret = -1; 224 225 debug = (getenv("ZINJECT_DEBUG") != NULL); 226 227 ziprintf("translating: %s\n", object); 228 229 if (MOS_TYPE(type)) { 230 /* 231 * MOS objects are treated specially. 232 */ 233 switch (type) { 234 default: 235 break; 236 case TYPE_MOS: 237 record->zi_type = 0; 238 break; 239 case TYPE_MOSDIR: 240 record->zi_type = DMU_OT_OBJECT_DIRECTORY; 241 break; 242 case TYPE_METASLAB: 243 record->zi_type = DMU_OT_OBJECT_ARRAY; 244 break; 245 case TYPE_CONFIG: 246 record->zi_type = DMU_OT_PACKED_NVLIST; 247 break; 248 case TYPE_BPOBJ: 249 record->zi_type = DMU_OT_BPOBJ; 250 break; 251 case TYPE_SPACEMAP: 252 record->zi_type = DMU_OT_SPACE_MAP; 253 break; 254 case TYPE_ERRLOG: 255 record->zi_type = DMU_OT_ERROR_LOG; 256 break; 257 } 258 259 dataset[0] = '\0'; 260 (void) strcpy(poolname, object); 261 return (0); 262 } 263 264 /* 265 * Convert a full path into a (dataset, file) pair. 266 */ 267 if (parse_pathname(object, dataset, path, &statbuf) != 0) 268 goto err; 269 270 ziprintf(" dataset: %s\n", dataset); 271 ziprintf(" path: %s\n", path); 272 273 /* 274 * Convert (dataset, file) into (objset, object) 275 */ 276 if (object_from_path(dataset, statbuf.st_ino, record) != 0) 277 goto err; 278 279 ziprintf("raw objset: %llu\n", record->zi_objset); 280 ziprintf("raw object: %llu\n", record->zi_object); 281 282 /* 283 * For the given object, initialize the range in bytes 284 */ 285 if (initialize_range(type, level, (char *)range, record) != 0) 286 goto err; 287 288 ziprintf(" objset: %llu\n", record->zi_objset); 289 ziprintf(" object: %llu\n", record->zi_object); 290 if (record->zi_start == 0 && 291 record->zi_end == -1ULL) 292 ziprintf(" range: all\n"); 293 else 294 ziprintf(" range: [%llu, %llu]\n", record->zi_start, 295 record->zi_end); 296 297 /* 298 * Copy the pool name 299 */ 300 (void) strcpy(poolname, dataset); 301 if ((slash = strchr(poolname, '/')) != NULL) 302 *slash = '\0'; 303 304 ret = 0; 305 306 err: 307 return (ret); 308 } 309 310 int 311 translate_raw(const char *str, zinject_record_t *record) 312 { 313 /* 314 * A raw bookmark of the form objset:object:level:blkid, where each 315 * number is a hexadecimal value. 316 */ 317 if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset, 318 (u_longlong_t *)&record->zi_object, &record->zi_level, 319 (u_longlong_t *)&record->zi_start) != 4) { 320 (void) fprintf(stderr, "bad raw spec '%s': must be of the form " 321 "'objset:object:level:blkid'\n", str); 322 return (-1); 323 } 324 325 record->zi_end = record->zi_start; 326 327 return (0); 328 } 329 330 int 331 translate_device(const char *pool, const char *device, err_type_t label_type, 332 zinject_record_t *record) 333 { 334 char *end; 335 zpool_handle_t *zhp; 336 nvlist_t *tgt; 337 boolean_t isspare, iscache; 338 339 /* 340 * Given a device name or GUID, create an appropriate injection record 341 * with zi_guid set. 342 */ 343 if ((zhp = zpool_open(g_zfs, pool)) == NULL) 344 return (-1); 345 346 record->zi_guid = strtoull(device, &end, 0); 347 if (record->zi_guid == 0 || *end != '\0') { 348 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL); 349 350 if (tgt == NULL) { 351 (void) fprintf(stderr, "cannot find device '%s' in " 352 "pool '%s'\n", device, pool); 353 zpool_close(zhp); 354 return (-1); 355 } 356 357 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, 358 &record->zi_guid) == 0); 359 } 360 361 /* 362 * Device faults can take on three different forms: 363 * 1). delayed or hanging I/O 364 * 2). zfs label faults 365 * 3). generic disk faults 366 */ 367 if (record->zi_timer != 0) { 368 record->zi_cmd = ZINJECT_DELAY_IO; 369 } else if (label_type != TYPE_INVAL) { 370 record->zi_cmd = ZINJECT_LABEL_FAULT; 371 } else { 372 record->zi_cmd = ZINJECT_DEVICE_FAULT; 373 } 374 375 switch (label_type) { 376 default: 377 break; 378 case TYPE_LABEL_UBERBLOCK: 379 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]); 380 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1; 381 break; 382 case TYPE_LABEL_NVLIST: 383 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys); 384 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1; 385 break; 386 case TYPE_LABEL_PAD1: 387 record->zi_start = offsetof(vdev_label_t, vl_pad1); 388 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1; 389 break; 390 case TYPE_LABEL_PAD2: 391 record->zi_start = offsetof(vdev_label_t, vl_be); 392 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1; 393 break; 394 } 395 zpool_close(zhp); 396 return (0); 397 } 398