cmd/zinject/translate.c

   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  */
  24
  25 #include <libzfs.h>
  26
  27 #include <sys/zfs_context.h>
  28
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <stdarg.h>
  32 #include <stddef.h>
  33 #include <stdio.h>
  34 #include <stdlib.h>
  35 #include <strings.h>
  36 #include <sys/file.h>
  37 #include <sys/mntent.h>
  38 #include <sys/mnttab.h>
  39 #include <sys/param.h>
  40 #include <sys/stat.h>
  41
  42 #include <sys/dmu.h>
  43 #include <sys/dmu_objset.h>
  44 #include <sys/dnode.h>
  45 #include <sys/vdev_impl.h>
  46
  47 #include <sys/mkdev.h>
  48
  49 #include "zinject.h"
  50
  51 extern void kernel_init(int);
  52 extern void kernel_fini(void);
  53
  54 static int debug;
  55
  56 static void
  57 ziprintf(const char *fmt, ...)
  58 {
  59         va_list ap;
  60
  61         if (!debug)
  62                 return;
  63
  64         va_start(ap, fmt);
  65         (void) vprintf(fmt, ap);
  66         va_end(ap);
  67 }
  68
  69 static void
  70 compress_slashes(const char *src, char *dest)
  71 {
  72         while (*src != '\0') {
  73                 *dest = *src++;
  74                 while (*dest == '/' && *src == '/')
  75                         ++src;
  76                 ++dest;
  77         }
  78         *dest = '\0';
  79 }
  80
  81 /*
  82  * Given a full path to a file, translate into a dataset name and a relative
  83  * path within the dataset.  'dataset' must be at least MAXNAMELEN characters,
  84  * and 'relpath' must be at least MAXPATHLEN characters.  We also pass a stat64
  85  * buffer, which we need later to get the object ID.
  86  */
  87 static int
  88 parse_pathname(const char *inpath, char *dataset, char *relpath,
  89     struct stat64 *statbuf)
  90 {
  91         struct extmnttab mp;
  92         FILE *fp;
  93         int match;
  94         const char *rel;
  95         char fullpath[MAXPATHLEN];
  96
  97         compress_slashes(inpath, fullpath);
  98
  99         if (fullpath[0] != '/') {
 100                 (void) fprintf(stderr, "invalid object '%s': must be full "
 101                     "path\n", fullpath);
 102                 usage();
 103                 return (-1);
 104         }
 105
 106         if (strlen(fullpath) >= MAXPATHLEN) {
 107                 (void) fprintf(stderr, "invalid object; pathname too long\n");
 108                 return (-1);
 109         }
 110
 111         if (stat64(fullpath, statbuf) != 0) {
 112                 (void) fprintf(stderr, "cannot open '%s': %s\n",
 113                     fullpath, strerror(errno));
 114                 return (-1);
 115         }
 116
 117         if ((fp = fopen(MNTTAB, "r")) == NULL) {
 118                 (void) fprintf(stderr, "cannot open /etc/mtab\n");
 119                 return (-1);
 120         }
 121
 122         match = 0;
 123         while (getextmntent(fp, &mp, sizeof (mp)) == 0) {
 124                 if (makedev(mp.mnt_major, mp.mnt_minor) == statbuf->st_dev) {
 125                         match = 1;
 126                         break;
 127                 }
 128         }
 129
 130         if (!match) {
 131                 (void) fprintf(stderr, "cannot find mountpoint for '%s'\n",
 132                     fullpath);
 133                 return (-1);
 134         }
 135
 136         if (strcmp(mp.mnt_fstype, MNTTYPE_ZFS) != 0) {
 137                 (void) fprintf(stderr, "invalid path '%s': not a ZFS "
 138                     "filesystem\n", fullpath);
 139                 return (-1);
 140         }
 141
 142         if (strncmp(fullpath, mp.mnt_mountp, strlen(mp.mnt_mountp)) != 0) {
 143                 (void) fprintf(stderr, "invalid path '%s': mountpoint "
 144                     "doesn't match path\n", fullpath);
 145                 return (-1);
 146         }
 147
 148         (void) strcpy(dataset, mp.mnt_special);
 149
 150         rel = fullpath + strlen(mp.mnt_mountp);
 151         if (rel[0] == '/')
 152                 rel++;
 153         (void) strcpy(relpath, rel);
 154
 155         return (0);
 156 }
 157
 158 /*
 159  * Convert from a (dataset, path) pair into a (objset, object) pair.  Note that
 160  * we grab the object number from the inode number, since looking this up via
 161  * libzpool is a real pain.
 162  */
 163 /* ARGSUSED */
 164 static int
 165 object_from_path(const char *dataset, const char *path, struct stat64 *statbuf,
 166     zinject_record_t *record)
 167 {
 168         objset_t *os;
 169         int err;
 170
 171         /*
 172          * Before doing any libzpool operations, call sync() to ensure that the
 173          * on-disk state is consistent with the in-core state.
 174          */
 175         sync();
 176
 177         err = dmu_objset_own(dataset, DMU_OST_ZFS, B_TRUE, FTAG, &os);
 178         if (err != 0) {
 179                 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
 180                     dataset, strerror(err));
 181                 return (-1);
 182         }
 183
 184         record->zi_objset = dmu_objset_id(os);
 185         record->zi_object = statbuf->st_ino;
 186
 187         dmu_objset_disown(os, FTAG);
 188
 189         return (0);
 190 }
 191
 192 /*
 193  * Calculate the real range based on the type, level, and range given.
 194  */
 195 static int
 196 calculate_range(const char *dataset, err_type_t type, int level, char *range,
 197     zinject_record_t *record)
 198 {
 199         objset_t *os = NULL;
 200         dnode_t *dn = NULL;
 201         int err;
 202         int ret = -1;
 203
 204         /*
 205          * Determine the numeric range from the string.
 206          */
 207         if (range == NULL) {
 208                 /*
 209                  * If range is unspecified, set the range to [0,-1], which
 210                  * indicates that the whole object should be treated as an
 211                  * error.
 212                  */
 213                 record->zi_start = 0;
 214                 record->zi_end = -1ULL;
 215         } else {
 216                 char *end;
 217
 218                 /* XXX add support for suffixes */
 219                 record->zi_start = strtoull(range, &end, 10);
 220
 221
 222                 if (*end == '\0')
 223                         record->zi_end = record->zi_start + 1;
 224                 else if (*end == ',')
 225                         record->zi_end = strtoull(end + 1, &end, 10);
 226
 227                 if (*end != '\0') {
 228                         (void) fprintf(stderr, "invalid range '%s': must be "
 229                             "a numeric range of the form 'start[,end]'\n",
 230                             range);
 231                         goto out;
 232                 }
 233         }
 234
 235         switch (type) {
 236         default:
 237                 break;
 238         case TYPE_DATA:
 239                 break;
 240
 241         case TYPE_DNODE:
 242                 /*
 243                  * If this is a request to inject faults into the dnode, then we
 244                  * must translate the current (objset,object) pair into an
 245                  * offset within the metadnode for the objset.  Specifying any
 246                  * kind of range with type 'dnode' is illegal.
 247                  */
 248                 if (range != NULL) {
 249                         (void) fprintf(stderr, "range cannot be specified when "
 250                             "type is 'dnode'\n");
 251                         goto out;
 252                 }
 253
 254                 record->zi_start = record->zi_object * sizeof (dnode_phys_t);
 255                 record->zi_end = record->zi_start + sizeof (dnode_phys_t);
 256                 record->zi_object = 0;
 257                 break;
 258         }
 259
 260         /*
 261          * Get the dnode associated with object, so we can calculate the block
 262          * size.
 263          */
 264         if ((err = dmu_objset_own(dataset, DMU_OST_ANY,
 265             B_TRUE, FTAG, &os)) != 0) {
 266                 (void) fprintf(stderr, "cannot open dataset '%s': %s\n",
 267                     dataset, strerror(err));
 268                 goto out;
 269         }
 270
 271         if (record->zi_object == 0) {
 272                 dn = DMU_META_DNODE(os);
 273         } else {
 274                 err = dnode_hold(os, record->zi_object, FTAG, &dn);
 275                 if (err != 0) {
 276                         (void) fprintf(stderr, "failed to hold dnode "
 277                             "for object %llu\n",
 278                             (u_longlong_t)record->zi_object);
 279                         goto out;
 280                 }
 281         }
 282
 283
 284         ziprintf("data shift: %d\n", (int)dn->dn_datablkshift);
 285         ziprintf(" ind shift: %d\n", (int)dn->dn_indblkshift);
 286
 287         /*
 288          * Translate range into block IDs.
 289          */
 290         if (record->zi_start != 0 || record->zi_end != -1ULL) {
 291                 record->zi_start >>= dn->dn_datablkshift;
 292                 record->zi_end >>= dn->dn_datablkshift;
 293         }
 294
 295         /*
 296          * Check level, and then translate level 0 blkids into ranges
 297          * appropriate for level of indirection.
 298          */
 299         record->zi_level = level;
 300         if (level > 0) {
 301                 ziprintf("level 0 blkid range: [%llu, %llu]\n",
 302                     record->zi_start, record->zi_end);
 303
 304                 if (level >= dn->dn_nlevels) {
 305                         (void) fprintf(stderr, "level %d exceeds max level "
 306                             "of object (%d)\n", level, dn->dn_nlevels - 1);
 307                         goto out;
 308                 }
 309
 310                 if (record->zi_start != 0 || record->zi_end != 0) {
 311                         int shift = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
 312
 313                         for (; level > 0; level--) {
 314                                 record->zi_start >>= shift;
 315                                 record->zi_end >>= shift;
 316                         }
 317                 }
 318         }
 319
 320         ret = 0;
 321 out:
 322         if (dn) {
 323                 if (dn != DMU_META_DNODE(os))
 324                         dnode_rele(dn, FTAG);
 325         }
 326         if (os)
 327                 dmu_objset_disown(os, FTAG);
 328
 329         return (ret);
 330 }
 331
 332 int
 333 translate_record(err_type_t type, const char *object, const char *range,
 334     int level, zinject_record_t *record, char *poolname, char *dataset)
 335 {
 336         char path[MAXPATHLEN];
 337         char *slash;
 338         struct stat64 statbuf;
 339         int ret = -1;
 340
 341         kernel_init(FREAD);
 342
 343         debug = (getenv("ZINJECT_DEBUG") != NULL);
 344
 345         ziprintf("translating: %s\n", object);
 346
 347         if (MOS_TYPE(type)) {
 348                 /*
 349                  * MOS objects are treated specially.
 350                  */
 351                 switch (type) {
 352                 default:
 353                         break;
 354                 case TYPE_MOS:
 355                         record->zi_type = 0;
 356                         break;
 357                 case TYPE_MOSDIR:
 358                         record->zi_type = DMU_OT_OBJECT_DIRECTORY;
 359                         break;
 360                 case TYPE_METASLAB:
 361                         record->zi_type = DMU_OT_OBJECT_ARRAY;
 362                         break;
 363                 case TYPE_CONFIG:
 364                         record->zi_type = DMU_OT_PACKED_NVLIST;
 365                         break;
 366                 case TYPE_BPOBJ:
 367                         record->zi_type = DMU_OT_BPOBJ;
 368                         break;
 369                 case TYPE_SPACEMAP:
 370                         record->zi_type = DMU_OT_SPACE_MAP;
 371                         break;
 372                 case TYPE_ERRLOG:
 373                         record->zi_type = DMU_OT_ERROR_LOG;
 374                         break;
 375                 }
 376
 377                 dataset[0] = '\0';
 378                 (void) strcpy(poolname, object);
 379                 return (0);
 380         }
 381
 382         /*
 383          * Convert a full path into a (dataset, file) pair.
 384          */
 385         if (parse_pathname(object, dataset, path, &statbuf) != 0)
 386                 goto err;
 387
 388         ziprintf("   dataset: %s\n", dataset);
 389         ziprintf("      path: %s\n", path);
 390
 391         /*
 392          * Convert (dataset, file) into (objset, object)
 393          */
 394         if (object_from_path(dataset, path, &statbuf, record) != 0)
 395                 goto err;
 396
 397         ziprintf("raw objset: %llu\n", record->zi_objset);
 398         ziprintf("raw object: %llu\n", record->zi_object);
 399
 400         /*
 401          * For the given object, calculate the real (type, level, range)
 402          */
 403         if (calculate_range(dataset, type, level, (char *)range, record) != 0)
 404                 goto err;
 405
 406         ziprintf("    objset: %llu\n", record->zi_objset);
 407         ziprintf("    object: %llu\n", record->zi_object);
 408         if (record->zi_start == 0 &&
 409             record->zi_end == -1ULL)
 410                 ziprintf("     range: all\n");
 411         else
 412                 ziprintf("     range: [%llu, %llu]\n", record->zi_start,
 413                     record->zi_end);
 414
 415         /*
 416          * Copy the pool name
 417          */
 418         (void) strcpy(poolname, dataset);
 419         if ((slash = strchr(poolname, '/')) != NULL)
 420                 *slash = '\0';
 421
 422         ret = 0;
 423
 424 err:
 425         kernel_fini();
 426         return (ret);
 427 }
 428
 429 int
 430 translate_raw(const char *str, zinject_record_t *record)
 431 {
 432         /*
 433          * A raw bookmark of the form objset:object:level:blkid, where each
 434          * number is a hexidecimal value.
 435          */
 436         if (sscanf(str, "%llx:%llx:%x:%llx", (u_longlong_t *)&record->zi_objset,
 437             (u_longlong_t *)&record->zi_object, &record->zi_level,
 438             (u_longlong_t *)&record->zi_start) != 4) {
 439                 (void) fprintf(stderr, "bad raw spec '%s': must be of the form "
 440                     "'objset:object:level:blkid'\n", str);
 441                 return (-1);
 442         }
 443
 444         record->zi_end = record->zi_start;
 445
 446         return (0);
 447 }
 448
 449 int
 450 translate_device(const char *pool, const char *device, err_type_t label_type,
 451     zinject_record_t *record)
 452 {
 453         char *end;
 454         zpool_handle_t *zhp;
 455         nvlist_t *tgt;
 456         boolean_t isspare, iscache;
 457
 458         /*
 459          * Given a device name or GUID, create an appropriate injection record
 460          * with zi_guid set.
 461          */
 462         if ((zhp = zpool_open(g_zfs, pool)) == NULL)
 463                 return (-1);
 464
 465         record->zi_guid = strtoull(device, &end, 16);
 466         if (record->zi_guid == 0 || *end != '\0') {
 467                 tgt = zpool_find_vdev(zhp, device, &isspare, &iscache, NULL);
 468
 469                 if (tgt == NULL) {
 470                         (void) fprintf(stderr, "cannot find device '%s' in "
 471                             "pool '%s'\n", device, pool);
 472                         return (-1);
 473                 }
 474
 475                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
 476                     &record->zi_guid) == 0);
 477         }
 478
 479         switch (label_type) {
 480         default:
 481                 break;
 482         case TYPE_LABEL_UBERBLOCK:
 483                 record->zi_start = offsetof(vdev_label_t, vl_uberblock[0]);
 484                 record->zi_end = record->zi_start + VDEV_UBERBLOCK_RING - 1;
 485                 break;
 486         case TYPE_LABEL_NVLIST:
 487                 record->zi_start = offsetof(vdev_label_t, vl_vdev_phys);
 488                 record->zi_end = record->zi_start + VDEV_PHYS_SIZE - 1;
 489                 break;
 490         case TYPE_LABEL_PAD1:
 491                 record->zi_start = offsetof(vdev_label_t, vl_pad1);
 492                 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
 493                 break;
 494         case TYPE_LABEL_PAD2:
 495                 record->zi_start = offsetof(vdev_label_t, vl_pad2);
 496                 record->zi_end = record->zi_start + VDEV_PAD_SIZE - 1;
 497                 break;
 498         }
 499         return (0);
 500 }