1ea8dc4b6Seschrock /* 2ea8dc4b6Seschrock * CDDL HEADER START 3ea8dc4b6Seschrock * 4ea8dc4b6Seschrock * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7ea8dc4b6Seschrock * 8ea8dc4b6Seschrock * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9ea8dc4b6Seschrock * or http://www.opensolaris.org/os/licensing. 10ea8dc4b6Seschrock * See the License for the specific language governing permissions 11ea8dc4b6Seschrock * and limitations under the License. 12ea8dc4b6Seschrock * 13ea8dc4b6Seschrock * When distributing Covered Code, include this CDDL HEADER in each 14ea8dc4b6Seschrock * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15ea8dc4b6Seschrock * If applicable, add the following below this CDDL HEADER, with the 16ea8dc4b6Seschrock * fields enclosed by brackets "[]" replaced with your own identifying 17ea8dc4b6Seschrock * information: Portions Copyright [yyyy] [name of copyright owner] 18ea8dc4b6Seschrock * 19ea8dc4b6Seschrock * CDDL HEADER END 20ea8dc4b6Seschrock */ 21ea8dc4b6Seschrock /* 22*21bf64a7Sgw25295 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23ea8dc4b6Seschrock * Use is subject to license terms. 24ea8dc4b6Seschrock */ 25ea8dc4b6Seschrock 26ea8dc4b6Seschrock #pragma ident "%Z%%M% %I% %E% SMI" 27ea8dc4b6Seschrock 28ea8dc4b6Seschrock /* 29ea8dc4b6Seschrock * ZFS Fault Injector 30ea8dc4b6Seschrock * 31ea8dc4b6Seschrock * This userland component takes a set of options and uses libzpool to translate 32ea8dc4b6Seschrock * from a user-visible object type and name to an internal representation. 33ea8dc4b6Seschrock * There are two basic types of faults: device faults and data faults. 34ea8dc4b6Seschrock * 35ea8dc4b6Seschrock * 36ea8dc4b6Seschrock * DEVICE FAULTS 37ea8dc4b6Seschrock * 38ea8dc4b6Seschrock * Errors can be injected into a particular vdev using the '-d' option. This 39ea8dc4b6Seschrock * option takes a path or vdev GUID to uniquely identify the device within a 40ea8dc4b6Seschrock * pool. There are two types of errors that can be injected, EIO and ENXIO, 41*21bf64a7Sgw25295 * that can be controlled through the '-e' option. The default is ENXIO. For 42ea8dc4b6Seschrock * EIO failures, any attempt to read data from the device will return EIO, but 43ea8dc4b6Seschrock * subsequent attempt to reopen the device will succeed. For ENXIO failures, 44ea8dc4b6Seschrock * any attempt to read from the device will return EIO, but any attempt to 45ea8dc4b6Seschrock * reopen the device will also return ENXIO. 46*21bf64a7Sgw25295 * For label faults, the -L option must be specified. This allows faults 47*21bf64a7Sgw25295 * to be injected into either the nvlist or uberblock region of all the labels 48*21bf64a7Sgw25295 * for the specified device. 49ea8dc4b6Seschrock * 50ea8dc4b6Seschrock * This form of the command looks like: 51ea8dc4b6Seschrock * 52*21bf64a7Sgw25295 * zinject -d device [-e errno] [-L <uber | nvlist>] pool 53ea8dc4b6Seschrock * 54ea8dc4b6Seschrock * 55ea8dc4b6Seschrock * DATA FAULTS 56ea8dc4b6Seschrock * 57ea8dc4b6Seschrock * We begin with a tuple of the form: 58ea8dc4b6Seschrock * 59ea8dc4b6Seschrock * <type,level,range,object> 60ea8dc4b6Seschrock * 61ea8dc4b6Seschrock * type A string describing the type of data to target. Each type 62ea8dc4b6Seschrock * implicitly describes how to interpret 'object'. Currently, 63ea8dc4b6Seschrock * the following values are supported: 64ea8dc4b6Seschrock * 65ea8dc4b6Seschrock * data User data for a file 66ea8dc4b6Seschrock * dnode Dnode for a file or directory 67ea8dc4b6Seschrock * 68ea8dc4b6Seschrock * The following MOS objects are special. Instead of injecting 69ea8dc4b6Seschrock * errors on a particular object or blkid, we inject errors across 70ea8dc4b6Seschrock * all objects of the given type. 71ea8dc4b6Seschrock * 72ea8dc4b6Seschrock * mos Any data in the MOS 73ea8dc4b6Seschrock * mosdir object directory 74ea8dc4b6Seschrock * config pool configuration 75ea8dc4b6Seschrock * bplist blkptr list 76ea8dc4b6Seschrock * spacemap spacemap 77ea8dc4b6Seschrock * metaslab metaslab 78ea8dc4b6Seschrock * errlog persistent error log 79ea8dc4b6Seschrock * 80ea8dc4b6Seschrock * level Object level. Defaults to '0', not applicable to all types. If 81ea8dc4b6Seschrock * a range is given, this corresponds to the indirect block 82ea8dc4b6Seschrock * corresponding to the specific range. 83ea8dc4b6Seschrock * 84ea8dc4b6Seschrock * range A numerical range [start,end) within the object. Defaults to 85ea8dc4b6Seschrock * the full size of the file. 86ea8dc4b6Seschrock * 87ea8dc4b6Seschrock * object A string describing the logical location of the object. For 88ea8dc4b6Seschrock * files and directories (currently the only supported types), 89ea8dc4b6Seschrock * this is the path of the object on disk. 90ea8dc4b6Seschrock * 91ea8dc4b6Seschrock * This is translated, via libzpool, into the following internal representation: 92ea8dc4b6Seschrock * 93ea8dc4b6Seschrock * <type,objset,object,level,range> 94ea8dc4b6Seschrock * 95ea8dc4b6Seschrock * These types should be self-explanatory. This tuple is then passed to the 96ea8dc4b6Seschrock * kernel via a special ioctl() to initiate fault injection for the given 97ea8dc4b6Seschrock * object. Note that 'type' is not strictly necessary for fault injection, but 98ea8dc4b6Seschrock * is used when translating existing faults into a human-readable string. 99ea8dc4b6Seschrock * 100ea8dc4b6Seschrock * 101ea8dc4b6Seschrock * The command itself takes one of the forms: 102ea8dc4b6Seschrock * 103ea8dc4b6Seschrock * zinject 104ea8dc4b6Seschrock * zinject <-a | -u pool> 105ea8dc4b6Seschrock * zinject -c <id|all> 106ea8dc4b6Seschrock * zinject [-q] <-t type> [-f freq] [-u] [-a] [-m] [-e errno] [-l level] 107ea8dc4b6Seschrock * [-r range] <object> 108ea8dc4b6Seschrock * zinject [-f freq] [-a] [-m] [-u] -b objset:object:level:start:end pool 109ea8dc4b6Seschrock * 110ea8dc4b6Seschrock * With no arguments, the command prints all currently registered injection 111ea8dc4b6Seschrock * handlers, with their numeric identifiers. 112ea8dc4b6Seschrock * 113ea8dc4b6Seschrock * The '-c' option will clear the given handler, or all handlers if 'all' is 114ea8dc4b6Seschrock * specified. 115ea8dc4b6Seschrock * 116ea8dc4b6Seschrock * The '-e' option takes a string describing the errno to simulate. This must 117ea8dc4b6Seschrock * be either 'io' or 'checksum'. In most cases this will result in the same 118ea8dc4b6Seschrock * behavior, but RAID-Z will produce a different set of ereports for this 119ea8dc4b6Seschrock * situation. 120ea8dc4b6Seschrock * 121ea8dc4b6Seschrock * The '-a', '-u', and '-m' flags toggle internal flush behavior. If '-a' is 122ea8dc4b6Seschrock * specified, then the ARC cache is flushed appropriately. If '-u' is 123ea8dc4b6Seschrock * specified, then the underlying SPA is unloaded. Either of these flags can be 124ea8dc4b6Seschrock * specified independently of any other handlers. The '-m' flag automatically 125ea8dc4b6Seschrock * does an unmount and remount of the underlying dataset to aid in flushing the 126ea8dc4b6Seschrock * cache. 127ea8dc4b6Seschrock * 128ea8dc4b6Seschrock * The '-f' flag controls the frequency of errors injected, expressed as a 129ea8dc4b6Seschrock * integer percentage between 1 and 100. The default is 100. 130ea8dc4b6Seschrock * 131ea8dc4b6Seschrock * The this form is responsible for actually injecting the handler into the 132ea8dc4b6Seschrock * framework. It takes the arguments described above, translates them to the 133ea8dc4b6Seschrock * internal tuple using libzpool, and then issues an ioctl() to register the 134ea8dc4b6Seschrock * handler. 135ea8dc4b6Seschrock * 136ea8dc4b6Seschrock * The final form can target a specific bookmark, regardless of whether a 137ea8dc4b6Seschrock * human-readable interface has been designed. It allows developers to specify 138ea8dc4b6Seschrock * a particular block by number. 139ea8dc4b6Seschrock */ 140ea8dc4b6Seschrock 141ea8dc4b6Seschrock #include <errno.h> 142ea8dc4b6Seschrock #include <fcntl.h> 143ea8dc4b6Seschrock #include <stdio.h> 144ea8dc4b6Seschrock #include <stdlib.h> 145ea8dc4b6Seschrock #include <strings.h> 146ea8dc4b6Seschrock #include <unistd.h> 147ea8dc4b6Seschrock 148ea8dc4b6Seschrock #include <sys/fs/zfs.h> 149ea8dc4b6Seschrock #include <sys/mount.h> 150ea8dc4b6Seschrock 151ea8dc4b6Seschrock #include <libzfs.h> 152ea8dc4b6Seschrock 153ea8dc4b6Seschrock #undef verify /* both libzfs.h and zfs_context.h want to define this */ 154ea8dc4b6Seschrock 155ea8dc4b6Seschrock #include "zinject.h" 156ea8dc4b6Seschrock 15799653d4eSeschrock libzfs_handle_t *g_zfs; 158ea8dc4b6Seschrock int zfs_fd; 159ea8dc4b6Seschrock 160ea8dc4b6Seschrock #define ECKSUM EBADE 161ea8dc4b6Seschrock 162ea8dc4b6Seschrock static const char *errtable[TYPE_INVAL] = { 163ea8dc4b6Seschrock "data", 164ea8dc4b6Seschrock "dnode", 165ea8dc4b6Seschrock "mos", 166ea8dc4b6Seschrock "mosdir", 167ea8dc4b6Seschrock "metaslab", 168ea8dc4b6Seschrock "config", 169ea8dc4b6Seschrock "bplist", 170ea8dc4b6Seschrock "spacemap", 171*21bf64a7Sgw25295 "errlog", 172*21bf64a7Sgw25295 "uber", 173*21bf64a7Sgw25295 "nvlist" 174ea8dc4b6Seschrock }; 175ea8dc4b6Seschrock 176ea8dc4b6Seschrock static err_type_t 177ea8dc4b6Seschrock name_to_type(const char *arg) 178ea8dc4b6Seschrock { 179ea8dc4b6Seschrock int i; 180ea8dc4b6Seschrock for (i = 0; i < TYPE_INVAL; i++) 181ea8dc4b6Seschrock if (strcmp(errtable[i], arg) == 0) 182ea8dc4b6Seschrock return (i); 183ea8dc4b6Seschrock 184ea8dc4b6Seschrock return (TYPE_INVAL); 185ea8dc4b6Seschrock } 186ea8dc4b6Seschrock 187ea8dc4b6Seschrock static const char * 188ea8dc4b6Seschrock type_to_name(uint64_t type) 189ea8dc4b6Seschrock { 190ea8dc4b6Seschrock switch (type) { 191ea8dc4b6Seschrock case DMU_OT_OBJECT_DIRECTORY: 192ea8dc4b6Seschrock return ("mosdir"); 193ea8dc4b6Seschrock case DMU_OT_OBJECT_ARRAY: 194ea8dc4b6Seschrock return ("metaslab"); 195ea8dc4b6Seschrock case DMU_OT_PACKED_NVLIST: 196ea8dc4b6Seschrock return ("config"); 197ea8dc4b6Seschrock case DMU_OT_BPLIST: 198ea8dc4b6Seschrock return ("bplist"); 199ea8dc4b6Seschrock case DMU_OT_SPACE_MAP: 200ea8dc4b6Seschrock return ("spacemap"); 201ea8dc4b6Seschrock case DMU_OT_ERROR_LOG: 202ea8dc4b6Seschrock return ("errlog"); 203ea8dc4b6Seschrock default: 204ea8dc4b6Seschrock return ("-"); 205ea8dc4b6Seschrock } 206ea8dc4b6Seschrock } 207ea8dc4b6Seschrock 208ea8dc4b6Seschrock 209ea8dc4b6Seschrock /* 210ea8dc4b6Seschrock * Print usage message. 211ea8dc4b6Seschrock */ 212ea8dc4b6Seschrock void 213ea8dc4b6Seschrock usage(void) 214ea8dc4b6Seschrock { 215ea8dc4b6Seschrock (void) printf( 216ea8dc4b6Seschrock "usage:\n" 217ea8dc4b6Seschrock "\n" 218ea8dc4b6Seschrock "\tzinject\n" 219ea8dc4b6Seschrock "\n" 220ea8dc4b6Seschrock "\t\tList all active injection records.\n" 221ea8dc4b6Seschrock "\n" 222ea8dc4b6Seschrock "\tzinject -c <id|all>\n" 223ea8dc4b6Seschrock "\n" 224ea8dc4b6Seschrock "\t\tClear the particular record (if given a numeric ID), or\n" 225ea8dc4b6Seschrock "\t\tall records if 'all' is specificed.\n" 226ea8dc4b6Seschrock "\n" 227*21bf64a7Sgw25295 "\tzinject -d device [-e errno] [-L <nvlist|uber>] pool\n" 228*21bf64a7Sgw25295 "\t\tInject a fault into a particular device or the device's\n" 229*21bf64a7Sgw25295 "\t\tlabel. Label injection can either be 'nvlist' or 'uber'.\n" 230*21bf64a7Sgw25295 "\t\t'errno' can either be 'nxio' (the default) or 'io'.\n" 231ea8dc4b6Seschrock "\n" 232ea8dc4b6Seschrock "\tzinject -b objset:object:level:blkid pool\n" 233ea8dc4b6Seschrock "\n" 234ea8dc4b6Seschrock "\t\tInject an error into pool 'pool' with the numeric bookmark\n" 235ea8dc4b6Seschrock "\t\tspecified by the remaining tuple. Each number is in\n" 236ea8dc4b6Seschrock "\t\thexidecimal, and only one block can be specified.\n" 237ea8dc4b6Seschrock "\n" 238ea8dc4b6Seschrock "\tzinject [-q] <-t type> [-e errno] [-l level] [-r range]\n" 239ea8dc4b6Seschrock "\t [-a] [-m] [-u] [-f freq] <object>\n" 240ea8dc4b6Seschrock "\n" 241ea8dc4b6Seschrock "\t\tInject an error into the object specified by the '-t' option\n" 242ea8dc4b6Seschrock "\t\tand the object descriptor. The 'object' parameter is\n" 243ea8dc4b6Seschrock "\t\tinterperted depending on the '-t' option.\n" 244ea8dc4b6Seschrock "\n" 245ea8dc4b6Seschrock "\t\t-q\tQuiet mode. Only print out the handler number added.\n" 246ea8dc4b6Seschrock "\t\t-e\tInject a specific error. Must be either 'io' or\n" 247ea8dc4b6Seschrock "\t\t\t'checksum'. Default is 'io'.\n" 248ea8dc4b6Seschrock "\t\t-l\tInject error at a particular block level. Default is " 249ea8dc4b6Seschrock "0.\n" 250ea8dc4b6Seschrock "\t\t-m\tAutomatically remount underlying filesystem.\n" 251ea8dc4b6Seschrock "\t\t-r\tInject error over a particular logical range of an\n" 252ea8dc4b6Seschrock "\t\t\tobject. Will be translated to the appropriate blkid\n" 253ea8dc4b6Seschrock "\t\t\trange according to the object's properties.\n" 254ea8dc4b6Seschrock "\t\t-a\tFlush the ARC cache. Can be specified without any\n" 255ea8dc4b6Seschrock "\t\t\tassociated object.\n" 256ea8dc4b6Seschrock "\t\t-u\tUnload the associated pool. Can be specified with only\n" 257ea8dc4b6Seschrock "\t\t\ta pool object.\n" 258ea8dc4b6Seschrock "\t\t-f\tOnly inject errors a fraction of the time. Expressed as\n" 259ea8dc4b6Seschrock "\t\t\ta percentage between 1 and 100.\n" 260ea8dc4b6Seschrock "\n" 261ea8dc4b6Seschrock "\t-t data\t\tInject an error into the plain file contents of a\n" 262ea8dc4b6Seschrock "\t\t\tfile. The object must be specified as a complete path\n" 263ea8dc4b6Seschrock "\t\t\tto a file on a ZFS filesystem.\n" 264ea8dc4b6Seschrock "\n" 265ea8dc4b6Seschrock "\t-t dnode\tInject an error into the metadnode in the block\n" 266ea8dc4b6Seschrock "\t\t\tcorresponding to the dnode for a file or directory. The\n" 267ea8dc4b6Seschrock "\t\t\t'-r' option is incompatible with this mode. The object\n" 268ea8dc4b6Seschrock "\t\t\tis specified as a complete path to a file or directory\n" 269ea8dc4b6Seschrock "\t\t\ton a ZFS filesystem.\n" 270ea8dc4b6Seschrock "\n" 271ea8dc4b6Seschrock "\t-t <mos>\tInject errors into the MOS for objects of the given\n" 272ea8dc4b6Seschrock "\t\t\ttype. Valid types are: mos, mosdir, config, bplist,\n" 27355434c77Sek110237 "\t\t\tspacemap, metaslab, errlog. The only valid <object> is\n" 27455434c77Sek110237 "\t\t\tthe poolname.\n"); 275ea8dc4b6Seschrock } 276ea8dc4b6Seschrock 277ea8dc4b6Seschrock static int 278ea8dc4b6Seschrock iter_handlers(int (*func)(int, const char *, zinject_record_t *, void *), 279ea8dc4b6Seschrock void *data) 280ea8dc4b6Seschrock { 281ea8dc4b6Seschrock zfs_cmd_t zc; 282ea8dc4b6Seschrock int ret; 283ea8dc4b6Seschrock 284ea8dc4b6Seschrock zc.zc_guid = 0; 285ea8dc4b6Seschrock 286ea8dc4b6Seschrock while (ioctl(zfs_fd, ZFS_IOC_INJECT_LIST_NEXT, &zc) == 0) 287ea8dc4b6Seschrock if ((ret = func((int)zc.zc_guid, zc.zc_name, 288ea8dc4b6Seschrock &zc.zc_inject_record, data)) != 0) 289ea8dc4b6Seschrock return (ret); 290ea8dc4b6Seschrock 291ea8dc4b6Seschrock return (0); 292ea8dc4b6Seschrock } 293ea8dc4b6Seschrock 294ea8dc4b6Seschrock static int 295ea8dc4b6Seschrock print_data_handler(int id, const char *pool, zinject_record_t *record, 296ea8dc4b6Seschrock void *data) 297ea8dc4b6Seschrock { 298ea8dc4b6Seschrock int *count = data; 299ea8dc4b6Seschrock 300ea8dc4b6Seschrock if (record->zi_guid != 0) 301ea8dc4b6Seschrock return (0); 302ea8dc4b6Seschrock 303ea8dc4b6Seschrock if (*count == 0) { 304ea8dc4b6Seschrock (void) printf("%3s %-15s %-6s %-6s %-8s %3s %-15s\n", 305ea8dc4b6Seschrock "ID", "POOL", "OBJSET", "OBJECT", "TYPE", "LVL", "RANGE"); 306ea8dc4b6Seschrock (void) printf("--- --------------- ------ " 307ea8dc4b6Seschrock "------ -------- --- ---------------\n"); 308ea8dc4b6Seschrock } 309ea8dc4b6Seschrock 310ea8dc4b6Seschrock *count += 1; 311ea8dc4b6Seschrock 312ea8dc4b6Seschrock (void) printf("%3d %-15s %-6llu %-6llu %-8s %3d ", id, pool, 313ea8dc4b6Seschrock (u_longlong_t)record->zi_objset, (u_longlong_t)record->zi_object, 314ea8dc4b6Seschrock type_to_name(record->zi_type), record->zi_level); 315ea8dc4b6Seschrock 316ea8dc4b6Seschrock if (record->zi_start == 0 && 317ea8dc4b6Seschrock record->zi_end == -1ULL) 318ea8dc4b6Seschrock (void) printf("all\n"); 319ea8dc4b6Seschrock else 320ea8dc4b6Seschrock (void) printf("[%llu, %llu]\n", (u_longlong_t)record->zi_start, 321ea8dc4b6Seschrock (u_longlong_t)record->zi_end); 322ea8dc4b6Seschrock 323ea8dc4b6Seschrock return (0); 324ea8dc4b6Seschrock } 325ea8dc4b6Seschrock 326ea8dc4b6Seschrock static int 327ea8dc4b6Seschrock print_device_handler(int id, const char *pool, zinject_record_t *record, 328ea8dc4b6Seschrock void *data) 329ea8dc4b6Seschrock { 330ea8dc4b6Seschrock int *count = data; 331ea8dc4b6Seschrock 332ea8dc4b6Seschrock if (record->zi_guid == 0) 333ea8dc4b6Seschrock return (0); 334ea8dc4b6Seschrock 335ea8dc4b6Seschrock if (*count == 0) { 336ea8dc4b6Seschrock (void) printf("%3s %-15s %s\n", "ID", "POOL", "GUID"); 337ea8dc4b6Seschrock (void) printf("--- --------------- ----------------\n"); 338ea8dc4b6Seschrock } 339ea8dc4b6Seschrock 340ea8dc4b6Seschrock *count += 1; 341ea8dc4b6Seschrock 342ea8dc4b6Seschrock (void) printf("%3d %-15s %llx\n", id, pool, 343ea8dc4b6Seschrock (u_longlong_t)record->zi_guid); 344ea8dc4b6Seschrock 345ea8dc4b6Seschrock return (0); 346ea8dc4b6Seschrock } 347ea8dc4b6Seschrock 348ea8dc4b6Seschrock /* 349ea8dc4b6Seschrock * Print all registered error handlers. Returns the number of handlers 350ea8dc4b6Seschrock * registered. 351ea8dc4b6Seschrock */ 352ea8dc4b6Seschrock static int 353ea8dc4b6Seschrock print_all_handlers(void) 354ea8dc4b6Seschrock { 355ea8dc4b6Seschrock int count = 0; 356ea8dc4b6Seschrock 357ea8dc4b6Seschrock (void) iter_handlers(print_device_handler, &count); 358ea8dc4b6Seschrock (void) printf("\n"); 359ea8dc4b6Seschrock count = 0; 360ea8dc4b6Seschrock (void) iter_handlers(print_data_handler, &count); 361ea8dc4b6Seschrock 362ea8dc4b6Seschrock return (count); 363ea8dc4b6Seschrock } 364ea8dc4b6Seschrock 365ea8dc4b6Seschrock /* ARGSUSED */ 366ea8dc4b6Seschrock static int 367ea8dc4b6Seschrock cancel_one_handler(int id, const char *pool, zinject_record_t *record, 368ea8dc4b6Seschrock void *data) 369ea8dc4b6Seschrock { 370ea8dc4b6Seschrock zfs_cmd_t zc; 371ea8dc4b6Seschrock 372ea8dc4b6Seschrock zc.zc_guid = (uint64_t)id; 373ea8dc4b6Seschrock 374ea8dc4b6Seschrock if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) { 375ea8dc4b6Seschrock (void) fprintf(stderr, "failed to remove handler %d: %s\n", 376ea8dc4b6Seschrock id, strerror(errno)); 377ea8dc4b6Seschrock return (1); 378ea8dc4b6Seschrock } 379ea8dc4b6Seschrock 380ea8dc4b6Seschrock return (0); 381ea8dc4b6Seschrock } 382ea8dc4b6Seschrock 383ea8dc4b6Seschrock /* 384ea8dc4b6Seschrock * Remove all fault injection handlers. 385ea8dc4b6Seschrock */ 386ea8dc4b6Seschrock static int 387ea8dc4b6Seschrock cancel_all_handlers(void) 388ea8dc4b6Seschrock { 389ea8dc4b6Seschrock int ret = iter_handlers(cancel_one_handler, NULL); 390ea8dc4b6Seschrock 391ea8dc4b6Seschrock (void) printf("removed all registered handlers\n"); 392ea8dc4b6Seschrock 393ea8dc4b6Seschrock return (ret); 394ea8dc4b6Seschrock } 395ea8dc4b6Seschrock 396ea8dc4b6Seschrock /* 397ea8dc4b6Seschrock * Remove a specific fault injection handler. 398ea8dc4b6Seschrock */ 399ea8dc4b6Seschrock static int 400ea8dc4b6Seschrock cancel_handler(int id) 401ea8dc4b6Seschrock { 402ea8dc4b6Seschrock zfs_cmd_t zc; 403ea8dc4b6Seschrock 404ea8dc4b6Seschrock zc.zc_guid = (uint64_t)id; 405ea8dc4b6Seschrock 406ea8dc4b6Seschrock if (ioctl(zfs_fd, ZFS_IOC_CLEAR_FAULT, &zc) != 0) { 407ea8dc4b6Seschrock (void) fprintf(stderr, "failed to remove handler %d: %s\n", 408ea8dc4b6Seschrock id, strerror(errno)); 409ea8dc4b6Seschrock return (1); 410ea8dc4b6Seschrock } 411ea8dc4b6Seschrock 412ea8dc4b6Seschrock (void) printf("removed handler %d\n", id); 413ea8dc4b6Seschrock 414ea8dc4b6Seschrock return (0); 415ea8dc4b6Seschrock } 416ea8dc4b6Seschrock 417ea8dc4b6Seschrock /* 418ea8dc4b6Seschrock * Register a new fault injection handler. 419ea8dc4b6Seschrock */ 420ea8dc4b6Seschrock static int 421ea8dc4b6Seschrock register_handler(const char *pool, int flags, zinject_record_t *record, 422ea8dc4b6Seschrock int quiet) 423ea8dc4b6Seschrock { 424ea8dc4b6Seschrock zfs_cmd_t zc; 425ea8dc4b6Seschrock 426ea8dc4b6Seschrock (void) strcpy(zc.zc_name, pool); 427ea8dc4b6Seschrock zc.zc_inject_record = *record; 428ea8dc4b6Seschrock zc.zc_guid = flags; 429ea8dc4b6Seschrock 430ea8dc4b6Seschrock if (ioctl(zfs_fd, ZFS_IOC_INJECT_FAULT, &zc) != 0) { 431ea8dc4b6Seschrock (void) fprintf(stderr, "failed to add handler: %s\n", 432ea8dc4b6Seschrock strerror(errno)); 433ea8dc4b6Seschrock return (1); 434ea8dc4b6Seschrock } 435ea8dc4b6Seschrock 436ea8dc4b6Seschrock if (flags & ZINJECT_NULL) 437ea8dc4b6Seschrock return (0); 438ea8dc4b6Seschrock 439ea8dc4b6Seschrock if (quiet) { 440ea8dc4b6Seschrock (void) printf("%llu\n", (u_longlong_t)zc.zc_guid); 441ea8dc4b6Seschrock } else { 442ea8dc4b6Seschrock (void) printf("Added handler %llu with the following " 443ea8dc4b6Seschrock "properties:\n", (u_longlong_t)zc.zc_guid); 444ea8dc4b6Seschrock (void) printf(" pool: %s\n", pool); 445ea8dc4b6Seschrock if (record->zi_guid) { 446ea8dc4b6Seschrock (void) printf(" vdev: %llx\n", 447ea8dc4b6Seschrock (u_longlong_t)record->zi_guid); 448ea8dc4b6Seschrock } else { 449ea8dc4b6Seschrock (void) printf("objset: %llu\n", 450ea8dc4b6Seschrock (u_longlong_t)record->zi_objset); 451ea8dc4b6Seschrock (void) printf("object: %llu\n", 452ea8dc4b6Seschrock (u_longlong_t)record->zi_object); 453ea8dc4b6Seschrock (void) printf(" type: %llu\n", 454ea8dc4b6Seschrock (u_longlong_t)record->zi_type); 455ea8dc4b6Seschrock (void) printf(" level: %d\n", record->zi_level); 456ea8dc4b6Seschrock if (record->zi_start == 0 && 457ea8dc4b6Seschrock record->zi_end == -1ULL) 458ea8dc4b6Seschrock (void) printf(" range: all\n"); 459ea8dc4b6Seschrock else 460ea8dc4b6Seschrock (void) printf(" range: [%llu, %llu)\n", 461ea8dc4b6Seschrock (u_longlong_t)record->zi_start, 462ea8dc4b6Seschrock (u_longlong_t)record->zi_end); 463ea8dc4b6Seschrock } 464ea8dc4b6Seschrock } 465ea8dc4b6Seschrock 466ea8dc4b6Seschrock return (0); 467ea8dc4b6Seschrock } 468ea8dc4b6Seschrock 469ea8dc4b6Seschrock int 470ea8dc4b6Seschrock main(int argc, char **argv) 471ea8dc4b6Seschrock { 472ea8dc4b6Seschrock int c; 473ea8dc4b6Seschrock char *range = NULL; 474ea8dc4b6Seschrock char *cancel = NULL; 475ea8dc4b6Seschrock char *end; 476ea8dc4b6Seschrock char *raw = NULL; 477ea8dc4b6Seschrock char *device = NULL; 478ea8dc4b6Seschrock int level = 0; 479ea8dc4b6Seschrock int quiet = 0; 480ea8dc4b6Seschrock int error = 0; 481ea8dc4b6Seschrock int domount = 0; 482ea8dc4b6Seschrock err_type_t type = TYPE_INVAL; 483*21bf64a7Sgw25295 err_type_t label = TYPE_INVAL; 484ea8dc4b6Seschrock zinject_record_t record = { 0 }; 485ea8dc4b6Seschrock char pool[MAXNAMELEN]; 486ea8dc4b6Seschrock char dataset[MAXNAMELEN]; 487ea8dc4b6Seschrock zfs_handle_t *zhp; 488ea8dc4b6Seschrock int ret; 489ea8dc4b6Seschrock int flags = 0; 490ea8dc4b6Seschrock 49199653d4eSeschrock if ((g_zfs = libzfs_init()) == NULL) { 49299653d4eSeschrock (void) fprintf(stderr, "internal error: failed to " 49399653d4eSeschrock "initialize ZFS library\n"); 49499653d4eSeschrock return (1); 49599653d4eSeschrock } 49699653d4eSeschrock 49799653d4eSeschrock libzfs_print_on_error(g_zfs, B_TRUE); 49899653d4eSeschrock 499ea8dc4b6Seschrock if ((zfs_fd = open(ZFS_DEV, O_RDWR)) < 0) { 500ea8dc4b6Seschrock (void) fprintf(stderr, "failed to open ZFS device\n"); 501ea8dc4b6Seschrock return (1); 502ea8dc4b6Seschrock } 503ea8dc4b6Seschrock 504ea8dc4b6Seschrock if (argc == 1) { 505ea8dc4b6Seschrock /* 506ea8dc4b6Seschrock * No arguments. Print the available handlers. If there are no 507ea8dc4b6Seschrock * available handlers, direct the user to '-h' for help 508ea8dc4b6Seschrock * information. 509ea8dc4b6Seschrock */ 510ea8dc4b6Seschrock if (print_all_handlers() == 0) { 511ea8dc4b6Seschrock (void) printf("No handlers registered.\n"); 512ea8dc4b6Seschrock (void) printf("Run 'zinject -h' for usage " 513ea8dc4b6Seschrock "information.\n"); 514ea8dc4b6Seschrock } 515ea8dc4b6Seschrock 516ea8dc4b6Seschrock return (0); 517ea8dc4b6Seschrock } 518ea8dc4b6Seschrock 519*21bf64a7Sgw25295 while ((c = getopt(argc, argv, ":ab:d:f:qhc:t:l:mr:e:uL:")) != -1) { 520ea8dc4b6Seschrock switch (c) { 521ea8dc4b6Seschrock case 'a': 522ea8dc4b6Seschrock flags |= ZINJECT_FLUSH_ARC; 523ea8dc4b6Seschrock break; 524ea8dc4b6Seschrock case 'b': 525ea8dc4b6Seschrock raw = optarg; 526ea8dc4b6Seschrock break; 527ea8dc4b6Seschrock case 'c': 528ea8dc4b6Seschrock cancel = optarg; 529ea8dc4b6Seschrock break; 530ea8dc4b6Seschrock case 'd': 531ea8dc4b6Seschrock device = optarg; 532ea8dc4b6Seschrock break; 533ea8dc4b6Seschrock case 'e': 534ea8dc4b6Seschrock if (strcasecmp(optarg, "io") == 0) { 535ea8dc4b6Seschrock error = EIO; 536ea8dc4b6Seschrock } else if (strcasecmp(optarg, "checksum") == 0) { 537ea8dc4b6Seschrock error = ECKSUM; 538ea8dc4b6Seschrock } else if (strcasecmp(optarg, "nxio") == 0) { 539ea8dc4b6Seschrock error = ENXIO; 540ea8dc4b6Seschrock } else { 541ea8dc4b6Seschrock (void) fprintf(stderr, "invalid error type " 542ea8dc4b6Seschrock "'%s': must be 'io', 'checksum' or " 543ea8dc4b6Seschrock "'nxio'\n", optarg); 544ea8dc4b6Seschrock usage(); 545ea8dc4b6Seschrock return (1); 546ea8dc4b6Seschrock } 547ea8dc4b6Seschrock break; 548ea8dc4b6Seschrock case 'f': 549ea8dc4b6Seschrock record.zi_freq = atoi(optarg); 550ea8dc4b6Seschrock if (record.zi_freq < 1 || record.zi_freq > 100) { 551ea8dc4b6Seschrock (void) fprintf(stderr, "frequency range must " 552ea8dc4b6Seschrock "be in the range (0, 100]\n"); 553ea8dc4b6Seschrock return (1); 554ea8dc4b6Seschrock } 555ea8dc4b6Seschrock break; 556ea8dc4b6Seschrock case 'h': 557ea8dc4b6Seschrock usage(); 558ea8dc4b6Seschrock return (0); 559ea8dc4b6Seschrock case 'l': 560ea8dc4b6Seschrock level = (int)strtol(optarg, &end, 10); 561ea8dc4b6Seschrock if (*end != '\0') { 562ea8dc4b6Seschrock (void) fprintf(stderr, "invalid level '%s': " 563ea8dc4b6Seschrock "must be an integer\n", optarg); 564ea8dc4b6Seschrock usage(); 565ea8dc4b6Seschrock return (1); 566ea8dc4b6Seschrock } 567ea8dc4b6Seschrock break; 568ea8dc4b6Seschrock case 'm': 569ea8dc4b6Seschrock domount = 1; 570ea8dc4b6Seschrock break; 571ea8dc4b6Seschrock case 'q': 572ea8dc4b6Seschrock quiet = 1; 573ea8dc4b6Seschrock break; 574ea8dc4b6Seschrock case 'r': 575ea8dc4b6Seschrock range = optarg; 576ea8dc4b6Seschrock break; 577ea8dc4b6Seschrock case 't': 578*21bf64a7Sgw25295 if ((type = name_to_type(optarg)) == TYPE_INVAL && 579*21bf64a7Sgw25295 !MOS_TYPE(type)) { 580ea8dc4b6Seschrock (void) fprintf(stderr, "invalid type '%s'\n", 581ea8dc4b6Seschrock optarg); 582ea8dc4b6Seschrock usage(); 583ea8dc4b6Seschrock return (1); 584ea8dc4b6Seschrock } 585ea8dc4b6Seschrock break; 586ea8dc4b6Seschrock case 'u': 587ea8dc4b6Seschrock flags |= ZINJECT_UNLOAD_SPA; 588ea8dc4b6Seschrock break; 589*21bf64a7Sgw25295 case 'L': 590*21bf64a7Sgw25295 if ((label = name_to_type(optarg)) == TYPE_INVAL && 591*21bf64a7Sgw25295 !LABEL_TYPE(type)) { 592*21bf64a7Sgw25295 (void) fprintf(stderr, "invalid label type " 593*21bf64a7Sgw25295 "'%s'\n", optarg); 594*21bf64a7Sgw25295 usage(); 595*21bf64a7Sgw25295 return (1); 596*21bf64a7Sgw25295 } 597*21bf64a7Sgw25295 break; 598ea8dc4b6Seschrock case ':': 599ea8dc4b6Seschrock (void) fprintf(stderr, "option -%c requires an " 600ea8dc4b6Seschrock "operand\n", optopt); 601ea8dc4b6Seschrock usage(); 602ea8dc4b6Seschrock return (1); 603ea8dc4b6Seschrock case '?': 604ea8dc4b6Seschrock (void) fprintf(stderr, "invalid option '%c'\n", 605ea8dc4b6Seschrock optopt); 606ea8dc4b6Seschrock usage(); 607ea8dc4b6Seschrock return (2); 608ea8dc4b6Seschrock } 609ea8dc4b6Seschrock } 610ea8dc4b6Seschrock 611ea8dc4b6Seschrock argc -= optind; 612ea8dc4b6Seschrock argv += optind; 613ea8dc4b6Seschrock 614ea8dc4b6Seschrock if (cancel != NULL) { 615ea8dc4b6Seschrock /* 616ea8dc4b6Seschrock * '-c' is invalid with any other options. 617ea8dc4b6Seschrock */ 618ea8dc4b6Seschrock if (raw != NULL || range != NULL || type != TYPE_INVAL || 619ea8dc4b6Seschrock level != 0) { 620ea8dc4b6Seschrock (void) fprintf(stderr, "cancel (-c) incompatible with " 621ea8dc4b6Seschrock "any other options\n"); 622ea8dc4b6Seschrock usage(); 623ea8dc4b6Seschrock return (2); 624ea8dc4b6Seschrock } 625ea8dc4b6Seschrock if (argc != 0) { 626ea8dc4b6Seschrock (void) fprintf(stderr, "extraneous argument to '-c'\n"); 627ea8dc4b6Seschrock usage(); 628ea8dc4b6Seschrock return (2); 629ea8dc4b6Seschrock } 630ea8dc4b6Seschrock 631ea8dc4b6Seschrock if (strcmp(cancel, "all") == 0) { 632ea8dc4b6Seschrock return (cancel_all_handlers()); 633ea8dc4b6Seschrock } else { 634ea8dc4b6Seschrock int id = (int)strtol(cancel, &end, 10); 635ea8dc4b6Seschrock if (*end != '\0') { 636ea8dc4b6Seschrock (void) fprintf(stderr, "invalid handle id '%s':" 637ea8dc4b6Seschrock " must be an integer or 'all'\n", cancel); 638ea8dc4b6Seschrock usage(); 639ea8dc4b6Seschrock return (1); 640ea8dc4b6Seschrock } 641ea8dc4b6Seschrock return (cancel_handler(id)); 642ea8dc4b6Seschrock } 643ea8dc4b6Seschrock } 644ea8dc4b6Seschrock 645ea8dc4b6Seschrock if (device != NULL) { 646ea8dc4b6Seschrock /* 647ea8dc4b6Seschrock * Device (-d) injection uses a completely different mechanism 648ea8dc4b6Seschrock * for doing injection, so handle it separately here. 649ea8dc4b6Seschrock */ 650ea8dc4b6Seschrock if (raw != NULL || range != NULL || type != TYPE_INVAL || 651ea8dc4b6Seschrock level != 0) { 652ea8dc4b6Seschrock (void) fprintf(stderr, "device (-d) incompatible with " 653ea8dc4b6Seschrock "data error injection\n"); 654ea8dc4b6Seschrock usage(); 655ea8dc4b6Seschrock return (2); 656ea8dc4b6Seschrock } 657ea8dc4b6Seschrock 658ea8dc4b6Seschrock if (argc != 1) { 659ea8dc4b6Seschrock (void) fprintf(stderr, "device (-d) injection requires " 660ea8dc4b6Seschrock "a single pool name\n"); 661ea8dc4b6Seschrock usage(); 662ea8dc4b6Seschrock return (2); 663ea8dc4b6Seschrock } 664ea8dc4b6Seschrock 665ea8dc4b6Seschrock (void) strcpy(pool, argv[0]); 666ea8dc4b6Seschrock dataset[0] = '\0'; 667ea8dc4b6Seschrock 668ea8dc4b6Seschrock if (error == ECKSUM) { 669ea8dc4b6Seschrock (void) fprintf(stderr, "device error type must be " 670ea8dc4b6Seschrock "'io' or 'nxio'\n"); 671ea8dc4b6Seschrock return (1); 672ea8dc4b6Seschrock } 673ea8dc4b6Seschrock 674*21bf64a7Sgw25295 if (translate_device(pool, device, label, &record) != 0) 675ea8dc4b6Seschrock return (1); 676ea8dc4b6Seschrock if (!error) 677ea8dc4b6Seschrock error = ENXIO; 678ea8dc4b6Seschrock } else if (raw != NULL) { 679ea8dc4b6Seschrock if (range != NULL || type != TYPE_INVAL || level != 0) { 680ea8dc4b6Seschrock (void) fprintf(stderr, "raw (-b) format with " 681ea8dc4b6Seschrock "any other options\n"); 682ea8dc4b6Seschrock usage(); 683ea8dc4b6Seschrock return (2); 684ea8dc4b6Seschrock } 685ea8dc4b6Seschrock 686ea8dc4b6Seschrock if (argc != 1) { 687ea8dc4b6Seschrock (void) fprintf(stderr, "raw (-b) format expects a " 688ea8dc4b6Seschrock "single pool name\n"); 689ea8dc4b6Seschrock usage(); 690ea8dc4b6Seschrock return (2); 691ea8dc4b6Seschrock } 692ea8dc4b6Seschrock 693ea8dc4b6Seschrock (void) strcpy(pool, argv[0]); 694ea8dc4b6Seschrock dataset[0] = '\0'; 695ea8dc4b6Seschrock 696ea8dc4b6Seschrock if (error == ENXIO) { 697ea8dc4b6Seschrock (void) fprintf(stderr, "data error type must be " 698ea8dc4b6Seschrock "'checksum' or 'io'\n"); 699ea8dc4b6Seschrock return (1); 700ea8dc4b6Seschrock } 701ea8dc4b6Seschrock 702ea8dc4b6Seschrock if (translate_raw(raw, &record) != 0) 703ea8dc4b6Seschrock return (1); 704ea8dc4b6Seschrock if (!error) 705ea8dc4b6Seschrock error = EIO; 706ea8dc4b6Seschrock } else if (type == TYPE_INVAL) { 707ea8dc4b6Seschrock if (flags == 0) { 708ea8dc4b6Seschrock (void) fprintf(stderr, "at least one of '-b', '-d', " 709ea8dc4b6Seschrock "'-t', '-a', or '-u' must be specified\n"); 710ea8dc4b6Seschrock usage(); 711ea8dc4b6Seschrock return (2); 712ea8dc4b6Seschrock } 713ea8dc4b6Seschrock 714ea8dc4b6Seschrock if (argc == 1 && (flags & ZINJECT_UNLOAD_SPA)) { 715ea8dc4b6Seschrock (void) strcpy(pool, argv[0]); 716ea8dc4b6Seschrock dataset[0] = '\0'; 717ea8dc4b6Seschrock } else if (argc != 0) { 718ea8dc4b6Seschrock (void) fprintf(stderr, "extraneous argument for " 719ea8dc4b6Seschrock "'-f'\n"); 720ea8dc4b6Seschrock usage(); 721ea8dc4b6Seschrock return (2); 722ea8dc4b6Seschrock } 723ea8dc4b6Seschrock 724ea8dc4b6Seschrock flags |= ZINJECT_NULL; 725ea8dc4b6Seschrock } else { 726ea8dc4b6Seschrock if (argc != 1) { 727ea8dc4b6Seschrock (void) fprintf(stderr, "missing object\n"); 728ea8dc4b6Seschrock usage(); 729ea8dc4b6Seschrock return (2); 730ea8dc4b6Seschrock } 731ea8dc4b6Seschrock 732ea8dc4b6Seschrock if (error == ENXIO) { 733ea8dc4b6Seschrock (void) fprintf(stderr, "data error type must be " 734ea8dc4b6Seschrock "'checksum' or 'io'\n"); 735ea8dc4b6Seschrock return (1); 736ea8dc4b6Seschrock } 737ea8dc4b6Seschrock 738ea8dc4b6Seschrock if (translate_record(type, argv[0], range, level, &record, pool, 739ea8dc4b6Seschrock dataset) != 0) 740ea8dc4b6Seschrock return (1); 741ea8dc4b6Seschrock if (!error) 742ea8dc4b6Seschrock error = EIO; 743ea8dc4b6Seschrock } 744ea8dc4b6Seschrock 745ea8dc4b6Seschrock /* 746ea8dc4b6Seschrock * If this is pool-wide metadata, unmount everything. The ioctl() will 747ea8dc4b6Seschrock * unload the pool, so that we trigger spa-wide reopen of metadata next 748ea8dc4b6Seschrock * time we access the pool. 749ea8dc4b6Seschrock */ 750ea8dc4b6Seschrock if (dataset[0] != '\0' && domount) { 751990b4856Slling if ((zhp = zfs_open(g_zfs, dataset, ZFS_TYPE_DATASET)) == NULL) 752ea8dc4b6Seschrock return (1); 753ea8dc4b6Seschrock 754ea8dc4b6Seschrock if (zfs_unmount(zhp, NULL, 0) != 0) 755ea8dc4b6Seschrock return (1); 756ea8dc4b6Seschrock } 757ea8dc4b6Seschrock 758ea8dc4b6Seschrock record.zi_error = error; 759ea8dc4b6Seschrock 760ea8dc4b6Seschrock ret = register_handler(pool, flags, &record, quiet); 761ea8dc4b6Seschrock 762ea8dc4b6Seschrock if (dataset[0] != '\0' && domount) 763ea8dc4b6Seschrock ret = (zfs_mount(zhp, NULL, 0) != 0); 764ea8dc4b6Seschrock 76599653d4eSeschrock libzfs_fini(g_zfs); 76699653d4eSeschrock 767ea8dc4b6Seschrock return (ret); 768ea8dc4b6Seschrock } 769