src/vol/salvager.c

   1 /*
   2  * Copyright 2000, International Business Machines Corporation and others.
   3  * All Rights Reserved.
   4  *
   5  * This software has been released under the terms of the IBM Public
   6  * License.  For details, see the LICENSE file in the top-level source
   7  * directory or online at http://www.openafs.org/dl/license10.html
   8  */
   9
  10 /*
  11  *      System:         VICE-TWO
  12  *      Module:         salvager.c
  13  *      Institution:    The Information Technology Center, Carnegie-Mellon University
  14  */
  15
  16
  17 /* Main program file. Define globals. */
  18 #define MAIN 1
  19
  20 #include <afsconfig.h>
  21 #include <afs/param.h>
  22
  23 #include <afs/procmgmt.h>
  24 #include <roken.h>
  25
  26 #ifdef HAVE_SYS_FILE_H
  27 #include <sys/file.h>
  28 #endif
  29
  30 #ifdef AFS_NT40_ENV
  31 #include <WINNT/afsevent.h>
  32 #endif
  33
  34 #ifndef WCOREDUMP
  35 #define WCOREDUMP(x)    ((x) & 0200)
  36 #endif
  37
  38 #include <rx/xdr.h>
  39 #include <afs/afsint.h>
  40 #if !defined(AFS_SGI_ENV) && !defined(AFS_NT40_ENV)
  41 #if defined(AFS_VFSINCL_ENV)
  42 #include <sys/vnode.h>
  43 #ifdef  AFS_SUN5_ENV
  44 #include <sys/fs/ufs_inode.h>
  45 #else
  46 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
  47 #include <ufs/ufs/dinode.h>
  48 #include <ufs/ffs/fs.h>
  49 #else
  50 #include <ufs/inode.h>
  51 #endif
  52 #endif
  53 #else /* AFS_VFSINCL_ENV */
  54 #ifdef  AFS_OSF_ENV
  55 #include <ufs/inode.h>
  56 #else /* AFS_OSF_ENV */
  57 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
  58 #include <sys/inode.h>
  59 #endif
  60 #endif
  61 #endif /* AFS_VFSINCL_ENV */
  62 #endif /* AFS_SGI_ENV */
  63 #ifdef  AFS_AIX_ENV
  64 #include <sys/vfs.h>
  65 #include <sys/lockf.h>
  66 #else
  67 #ifdef  AFS_HPUX_ENV
  68 #include <checklist.h>
  69 #else
  70 #if defined(AFS_SGI_ENV)
  71 #include <mntent.h>
  72 #else
  73 #if     defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
  74 #ifdef    AFS_SUN5_ENV
  75 #include <sys/mnttab.h>
  76 #include <sys/mntent.h>
  77 #else
  78 #include <mntent.h>
  79 #endif
  80 #else
  81 #endif /* AFS_SGI_ENV */
  82 #endif /* AFS_HPUX_ENV */
  83 #endif
  84 #endif
  85 #ifndef AFS_NT40_ENV
  86 #include <afs/osi_inode.h>
  87 #endif
  88 #include <afs/cmd.h>
  89 #include <afs/dir.h>
  90 #include <afs/afsutil.h>
  91 #include <afs/fileutil.h>
  92 #include <rx/rx_queue.h>
  93
  94 #include "nfs.h"
  95 #include "lwp.h"
  96 #include "lock.h"
  97 #include <afs/afssyscalls.h>
  98 #include "ihandle.h"
  99 #include "vnode.h"
 100 #include "volume.h"
 101 #include "partition.h"
 102 #include "daemon_com.h"
 103 #include "fssync.h"
 104 #include "salvsync.h"
 105 #include "viceinode.h"
 106 #include "salvage.h"
 107 #include "vol-salvage.h"
 108 #include "common.h"
 109 #ifdef AFS_NT40_ENV
 110 #include <pthread.h>
 111 pthread_t main_thread;
 112 #endif
 113
 114 extern char cml_version_number[];
 115 static int get_salvage_lock = 0;
 116
 117 struct CmdLine {
 118    int argc;
 119    char **argv;
 120 };
 121
 122 static int
 123 TimeStampLogFile(char **logfile)
 124 {
 125     char *stampSlvgLog;
 126     struct tm *lt;
 127     time_t now;
 128
 129     now = time(0);
 130     lt = localtime(&now);
 131     if (asprintf(&stampSlvgLog,
 132                  "%s.%04d-%02d-%02d.%02d:%02d:%02d",
 133                  AFSDIR_SERVER_SLVGLOG_FILEPATH,
 134                  lt->tm_year + 1900, lt->tm_mon + 1, lt->tm_mday, lt->tm_hour,
 135                  lt->tm_min, lt->tm_sec) < 0) {
 136         return ENOMEM;
 137     }
 138     *logfile = stampSlvgLog;
 139     return 0;
 140 }
 141
 142 static int
 143 handleit(struct cmd_syndesc *as, void *arock)
 144 {
 145     struct CmdLine *cmdline = (struct CmdLine*)arock;
 146     struct cmd_item *ti;
 147     char pname[100], *temp;
 148     afs_int32 seenpart = 0, seenvol = 0;
 149     VolumeId vid = 0;
 150     ProgramType pt;
 151
 152 #ifdef FAST_RESTART
 153     afs_int32  seenany = 0;
 154 #endif
 155
 156     char *filename = NULL;
 157     struct logOptions logopts;
 158     VolumePackageOptions opts;
 159     struct DiskPartition64 *partP;
 160
 161     memset(&logopts, 0, sizeof(logopts));
 162
 163 #ifdef AFS_SGI_VNODE_GLUE
 164     if (afs_init_kernel_config(-1) < 0) {
 165         printf
 166             ("Can't determine NUMA configuration, not starting salvager.\n");
 167         exit(1);
 168     }
 169 #endif
 170
 171 #ifdef FAST_RESTART
 172     {
 173         afs_int32 i;
 174         for (i = 0; i < CMD_MAXPARMS; i++) {
 175             if (as->parms[i].items) {
 176                 seenany = 1;
 177                 break;
 178             }
 179         }
 180     }
 181     if (!seenany) {
 182         printf
 183             ("Exiting immediately without salvage. "
 184              "Look into the FileLog to find volumes which really need to be salvaged!\n");
 185         Exit(0);
 186     }
 187 #endif /* FAST_RESTART */
 188     if ((ti = as->parms[0].items)) {    /* -partition */
 189         seenpart = 1;
 190         strncpy(pname, ti->data, 100);
 191     }
 192     if ((ti = as->parms[1].items)) {    /* -volumeid */
 193         char *end;
 194         unsigned long vid_l;
 195         if (!seenpart) {
 196             printf
 197                 ("You must also specify '-partition' option with the '-volumeid' option\n");
 198             exit(-1);
 199         }
 200         seenvol = 1;
 201         vid_l = strtoul(ti->data, &end, 10);
 202         if (vid_l >= MAX_AFS_UINT32 || vid_l == ULONG_MAX || *end != '\0') {
 203             fprintf(stderr, "salvage: invalid volume id specified; salvage aborted\n");
 204             Exit(1);
 205         }
 206         vid = (VolumeId)vid_l;
 207     }
 208     if (as->parms[2].items)     /* -debug */
 209         debug = 1;
 210     if (as->parms[3].items)     /* -nowrite */
 211         Testing = 1;
 212     if (as->parms[4].items)     /* -inodes */
 213         ListInodeOption = 1;
 214     if (as->parms[5].items || as->parms[21].items)      /* -force, -f */
 215         ForceSalvage = 1;
 216     if (as->parms[6].items)     /* -oktozap */
 217         OKToZap = 1;
 218     if (as->parms[7].items)     /* -rootinodes */
 219         ShowRootFiles = 1;
 220     if (as->parms[8].items)     /* -RebuildDirs */
 221         RebuildDirs = 1;
 222     if (as->parms[9].items)     /* -ForceReads */
 223         forceR = 1;
 224     if ((ti = as->parms[10].items)) {   /* -Parallel # */
 225         temp = ti->data;
 226         if (strncmp(temp, "all", 3) == 0) {
 227             PartsPerDisk = 1;
 228             temp += 3;
 229         }
 230         if (strlen(temp) != 0) {
 231             Parallel = atoi(temp);
 232             if (Parallel < 1)
 233                 Parallel = 1;
 234             if (Parallel > MAXPARALLEL) {
 235                 printf("Setting parallel salvages to maximum of %d \n",
 236                        MAXPARALLEL);
 237                 Parallel = MAXPARALLEL;
 238             }
 239         }
 240     }
 241     if ((ti = as->parms[11].items)) {   /* -tmpdir */
 242         DIR *dirp;
 243
 244         tmpdir = ti->data;
 245         dirp = opendir(tmpdir);
 246         if (!dirp) {
 247             printf
 248                 ("Can't open temporary placeholder dir %s; using current partition \n",
 249                  tmpdir);
 250             tmpdir = NULL;
 251         } else
 252             closedir(dirp);
 253     }
 254     if ((ti = as->parms[12].items))     /* -showlog */
 255         ShowLog = 1;
 256     if ((ti = as->parms[13].items)) {   /* -showsuid */
 257         Testing = 1;
 258         ShowSuid = 1;
 259         Showmode = 1;
 260     }
 261     if ((ti = as->parms[14].items)) {   /* -showmounts */
 262         Testing = 1;
 263         Showmode = 1;
 264         ShowMounts = 1;
 265     }
 266     if ((ti = as->parms[15].items)) {   /* -orphans */
 267         if (Testing)
 268             orphans = ORPH_IGNORE;
 269         else if (strcmp(ti->data, "remove") == 0
 270                  || strcmp(ti->data, "r") == 0)
 271             orphans = ORPH_REMOVE;
 272         else if (strcmp(ti->data, "attach") == 0
 273                  || strcmp(ti->data, "a") == 0)
 274             orphans = ORPH_ATTACH;
 275     }
 276
 277     if ((ti = as->parms[16].items)) {   /* -syslog */
 278         if (ShowLog) {
 279             fprintf(stderr, "Invalid options: -syslog and -showlog are exclusive.\n");
 280             Exit(1);
 281         }
 282         if ((ti = as->parms[18].items)) {       /* -datelogs */
 283             fprintf(stderr, "Invalid option: -syslog and -datelogs are exclusive.\n");
 284             Exit(1);
 285         }
 286 #ifndef HAVE_SYSLOG
 287         /* Do not silently ignore. */
 288         fprintf(stderr, "Invalid option: -syslog is not available on this platform.\n");
 289         Exit(1);
 290 #else
 291         logopts.lopt_dest = logDest_syslog;
 292         logopts.lopt_tag = "salvager";
 293
 294         if ((ti = as->parms[17].items))  /* -syslogfacility */
 295             logopts.lopt_facility = atoi(ti->data);
 296         else
 297             logopts.lopt_facility = LOG_DAEMON; /* default value */
 298 #endif
 299     } else {
 300         logopts.lopt_dest = logDest_file;
 301
 302         if ((ti = as->parms[18].items)) {       /* -datelogs */
 303             int code = TimeStampLogFile(&filename);
 304             if (code != 0) {
 305                 fprintf(stderr, "Failed to format log file name for -datelogs; code=%d\n", code);
 306                 Exit(code);
 307             }
 308             logopts.lopt_filename = filename;
 309         } else {
 310             logopts.lopt_filename = AFSDIR_SERVER_SLVGLOG_FILEPATH;
 311         }
 312     }
 313
 314     OpenLog(&logopts);
 315     SetupLogSignals();
 316     free(filename); /* Free string created by -datelogs, if one. */
 317
 318     Log("%s\n", cml_version_number);
 319     LogCommandLine(cmdline->argc, cmdline->argv, "SALVAGER", SalvageVersion, "STARTING AFS", Log);
 320
 321 #ifdef FAST_RESTART
 322     if (ti = as->parms[19].items) {     /* -DontSalvage */
 323         char *msg =
 324             "Exiting immediately without salvage. Look into the FileLog to find volumes which really need to be salvaged!";
 325         Log("%s\n", msg);
 326         printf("%s\n", msg);
 327         Exit(0);
 328     }
 329 #endif
 330
 331     /* Note:  if seenvol we initialize this as a standard volume utility:  this has the
 332      * implication that the file server may be running; negotations have to be made with
 333      * the file server in this case to take the read write volume and associated read-only
 334      * volumes off line before salvaging */
 335 #ifdef AFS_NT40_ENV
 336     if (seenvol) {
 337         if (afs_winsockInit() < 0) {
 338             ReportErrorEventAlt(AFSEVT_SVR_WINSOCK_INIT_FAILED, 0,
 339                                 AFSDIR_SALVAGER_FILE, 0);
 340             Log("Failed to initailize winsock, exiting.\n");
 341             Exit(1);
 342         }
 343     }
 344 #endif
 345
 346     if (seenvol) {
 347         pt = volumeSalvager;
 348     } else {
 349         pt = salvager;
 350     }
 351
 352     VOptDefaults(pt, &opts);
 353     if (VInitVolumePackage2(pt, &opts)) {
 354         Log("errors encountered initializing volume package; salvage aborted\n");
 355         Exit(1);
 356     }
 357
 358     /* defer lock until we init volume package */
 359     if (get_salvage_lock) {
 360         if (seenvol && AskDAFS()) /* support forceDAFS */
 361             ObtainSharedSalvageLock();
 362         else
 363             ObtainSalvageLock();
 364     }
 365
 366     /*
 367      * Ok to defer this as Exit will clean up and no real work is done
 368      * init'ing volume package
 369      */
 370     if (seenvol) {
 371         char *msg = NULL;
 372 #ifdef AFS_DEMAND_ATTACH_FS
 373         if (!AskDAFS()) {
 374             msg =
 375                 "The DAFS dasalvager cannot be run with a non-DAFS fileserver.  Please use 'salvager'.";
 376         }
 377         if (!msg && !as->parms[20].items) {
 378             msg =
 379                 "The standalone salvager cannot be run concurrently with a Demand Attach Fileserver.  Please use 'salvageserver -client <partition> <volume id>' to manually schedule volume salvages with the salvageserver (new versions of 'bos salvage' automatically do this for you).  Or, if you insist on using the standalone salvager, add the -forceDAFS flag to your salvager command line.";
 380         }
 381 #else
 382         if (AskDAFS()) {
 383             msg =
 384                 "The non-DAFS salvager cannot be run with a Demand Attach Fileserver.  Please use 'salvageserver -client <partition> <volume id>' to manually schedule volume salvages with the salvageserver (new versions of 'bos salvage' automatically do this for you).  Or, if you insist on using the standalone salvager, run dasalvager with the -forceDAFS flag.";
 385         }
 386 #endif
 387
 388         if (msg) {
 389             Log("%s\n", msg);
 390             printf("%s\n", msg);
 391             Exit(1);
 392         }
 393     }
 394
 395     DInit(10);
 396 #ifdef AFS_NT40_ENV
 397     if (myjob.cj_number != NOT_CHILD) {
 398         if (!seenpart) {
 399             seenpart = 1;
 400             (void)strcpy(pname, myjob.cj_part);
 401         }
 402     }
 403 #endif
 404     if (seenpart == 0) {
 405         for (partP = DiskPartitionList; partP; partP = partP->next) {
 406             SalvageFileSysParallel(partP);
 407         }
 408         SalvageFileSysParallel(0);
 409     } else {
 410         partP = VGetPartition(pname, 0);
 411         if (!partP) {
 412             Log("salvage: Unknown or unmounted partition %s; salvage aborted\n", pname);
 413             Exit(1);
 414         }
 415         if (!seenvol)
 416             SalvageFileSys(partP, 0);
 417         else {
 418             /* Salvage individual volume */
 419             SalvageFileSys(partP, vid);
 420         }
 421     }
 422     return (0);
 423 }
 424
 425
 426 #ifndef AFS_NT40_ENV
 427 #include "AFS_component_version_number.c"
 428 #endif
 429
 430 int
 431 main(int argc, char **argv)
 432 {
 433     struct CmdLine cmdline;
 434     struct cmd_syndesc *ts;
 435     int err = 0;
 436
 437 #ifdef  AFS_AIX32_ENV
 438     /*
 439      * The following signal action for AIX is necessary so that in case of a
 440      * crash (i.e. core is generated) we can include the user's data section
 441      * in the core dump. Unfortunately, by default, only a partial core is
 442      * generated which, in many cases, isn't too useful.
 443      */
 444     struct sigaction nsa;
 445
 446     sigemptyset(&nsa.sa_mask);
 447     nsa.sa_handler = SIG_DFL;
 448     nsa.sa_flags = SA_FULLDUMP;
 449     sigaction(SIGABRT, &nsa, NULL);
 450     sigaction(SIGSEGV, &nsa, NULL);
 451 #endif
 452
 453     /* Initialize directory paths */
 454     if (!(initAFSDirPath() & AFSDIR_SERVER_PATHS_OK)) {
 455 #ifdef AFS_NT40_ENV
 456         ReportErrorEventAlt(AFSEVT_SVR_NO_INSTALL_DIR, 0, argv[0], 0);
 457 #endif
 458         fprintf(stderr, "%s: Unable to obtain AFS server directory.\n",
 459                 argv[0]);
 460         exit(2);
 461     }
 462 #ifdef AFS_NT40_ENV
 463     /* Default to binary mode for fopen() */
 464     _set_fmode(_O_BINARY);
 465
 466     main_thread = pthread_self();
 467     if (spawnDatap && spawnDataLen) {
 468         /* This is a child per partition salvager. Don't setup log or
 469          * try to lock the salvager lock.
 470          */
 471         if (nt_SetupPartitionSalvage(spawnDatap, spawnDataLen) < 0)
 472             exit(3);
 473     } else {
 474 #endif
 475
 476 #ifndef AFS_NT40_ENV
 477         if (geteuid() != 0) {
 478             printf("Salvager must be run as root.\n");
 479             fflush(stdout);
 480             Exit(0);
 481         }
 482 #endif
 483
 484         /* Get and hold a lock for the duration of the salvage to make sure
 485          * that no other salvage runs at the same time.  The routine
 486          * VInitVolumePackage2 (called below) makes sure that a file server or
 487          * other volume utilities don't interfere with the salvage.
 488          */
 489         get_salvage_lock = 1;
 490 #ifdef AFS_NT40_ENV
 491     }
 492 #endif
 493
 494     cmdline.argc = argc;
 495     cmdline.argv = argv;
 496     ts = cmd_CreateSyntax("initcmd", handleit, &cmdline, 0, "initialize the program");
 497     cmd_AddParm(ts, "-partition", CMD_SINGLE, CMD_OPTIONAL,
 498                 "Name of partition to salvage");
 499     cmd_AddParm(ts, "-volumeid", CMD_SINGLE, CMD_OPTIONAL,
 500                 "Volume Id to salvage");
 501     cmd_AddParm(ts, "-debug", CMD_FLAG, CMD_OPTIONAL,
 502                 "Run in Debugging mode");
 503     cmd_AddParm(ts, "-nowrite", CMD_FLAG, CMD_OPTIONAL,
 504                 "Run readonly/test mode");
 505     cmd_AddParm(ts, "-inodes", CMD_FLAG, CMD_OPTIONAL,
 506                 "Just list affected afs inodes - debugging flag");
 507     cmd_AddParm(ts, "-force", CMD_FLAG, CMD_OPTIONAL, "Force full salvaging");
 508     cmd_AddParm(ts, "-oktozap", CMD_FLAG, CMD_OPTIONAL,
 509                 "Give permission to destroy bogus inodes/volumes - debugging flag");
 510     cmd_AddParm(ts, "-rootinodes", CMD_FLAG, CMD_OPTIONAL,
 511                 "Show inodes owned by root - debugging flag");
 512     cmd_AddParm(ts, "-salvagedirs", CMD_FLAG, CMD_OPTIONAL,
 513                 "Force rebuild/salvage of all directories");
 514     cmd_AddParm(ts, "-blockreads", CMD_FLAG, CMD_OPTIONAL,
 515                 "Read smaller blocks to handle IO/bad blocks");
 516     cmd_AddParm(ts, "-parallel", CMD_SINGLE, CMD_OPTIONAL,
 517                 "# of max parallel partition salvaging");
 518     cmd_AddParm(ts, "-tmpdir", CMD_SINGLE, CMD_OPTIONAL,
 519                 "Name of dir to place tmp files ");
 520     cmd_AddParm(ts, "-showlog", CMD_FLAG, CMD_OPTIONAL,
 521                 "Show log file upon completion");
 522     cmd_AddParm(ts, "-showsuid", CMD_FLAG, CMD_OPTIONAL,
 523                 "Report on suid/sgid files");
 524     cmd_AddParm(ts, "-showmounts", CMD_FLAG, CMD_OPTIONAL,
 525                 "Report on mountpoints");
 526     cmd_AddParm(ts, "-orphans", CMD_SINGLE, CMD_OPTIONAL,
 527                 "ignore | remove | attach");
 528
 529     /* note - syslog isn't avail on NT, but if we make it conditional, have
 530      * to deal with screwy offsets for cmd params */
 531     cmd_AddParm(ts, "-syslog", CMD_FLAG, CMD_OPTIONAL,
 532                 "Write salvage log to syslogs");
 533     cmd_AddParm(ts, "-syslogfacility", CMD_SINGLE, CMD_OPTIONAL,
 534                 "Syslog facility number to use");
 535     cmd_AddParm(ts, "-datelogs", CMD_FLAG, CMD_OPTIONAL,
 536                 "Include timestamp in logfile filename");
 537 #ifdef FAST_RESTART
 538     cmd_AddParm(ts, "-DontSalvage", CMD_FLAG, CMD_OPTIONAL,
 539                 "Don't salvage. This my be set in BosConfig to let the fileserver restart immediately after a crash. Bad volumes will be taken offline");
 540 #elif defined(AFS_DEMAND_ATTACH_FS)
 541     cmd_Seek(ts, 20); /* skip DontSalvage */
 542     cmd_AddParm(ts, "-forceDAFS", CMD_FLAG, CMD_OPTIONAL,
 543                 "For Demand Attach Fileserver, permit a manual volume salvage outside of the salvageserver");
 544 #endif /* FAST_RESTART */
 545     cmd_Seek(ts, 21); /* skip DontSalvage and forceDAFS if needed */
 546     cmd_AddParm(ts, "-f", CMD_FLAG, CMD_OPTIONAL, "Alias for -force");
 547     err = cmd_Dispatch(argc, argv);
 548     Exit(err);
 549     return 0; /* not reached */
 550 }
 551