lustre/utils/lustre_rsync.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/utils/lustre_rsync.c
  37  *
  38  * Author: Kalpak Shah <Kalpak.Shah@Sun.COM>
  39  * Author: Manoj Joseph <Manoj.Joseph@Sun.COM>
  40  */
  41
  42 /*
  43  * - lustre_rsync is a tool for replicating a lustre filesystem.
  44  *
  45  * - The source-fs is a live lustre filesystem. It is not a
  46  * snapshot. It is mounted and undergoing changes
  47  *
  48  * - The target-fs is a copy of the source-fs from the past. Let's
  49  * call this point, the 'sync point'.
  50  *
  51  * - There is a changelog of all metadata operations that happened on
  52  * the filesystem since the 'sync point'.
  53  *
  54  * - lustre_rsync replicates all the operations saved in the changelog
  55  * on to the target filesystem to make it identical to the source.
  56  *
  57  * To facilitate replication, the lustre filesystem provides
  58  *    a) a way to get the current filesystem path of a given FID
  59  *    b) a way to open files by specifying its FID
  60  *
  61  * The changelog only has a limited amount of information.
  62  *  tfid - The FID of the target file
  63  *  pfid - The FID of the parent of the target file (at the time of
  64  *         the operation)
  65  *  sfid - The FID of the source file
  66  *  spfid - The FID of the parent of the source file
  67  *  name - The name of the target file (at the time of the operation), the name
  68  *         of the source file is appended (delimited with '\0') if this
  69  *         operation involves a source
  70  *
  71  * With just this information, it is not alwasy possible to determine
  72  * the file paths for each operation. For instance, if pfid does not
  73  * exist on the source-fs (due to a subsequent deletion), its path
  74  * cannot be queried. In such cases, lustre_rsync keeps the files in a
  75  * special directory ("/.lustrerepl"). Once all the operations in a
  76  * changelog are replayed, all the files in this special directory
  77  * will get moved to the location as in the source-fs.
  78  *
  79  * Shorthand used: f2p(fid) = fid2path(fid)
  80  *
  81  * The following are the metadata operations of interest.
  82  * 1. creat
  83  *    If tfid is absent on the source-fs, ignore this operation
  84  *    If pfid is absent on the source-fs [or]
  85  *    if f2p(pfid) is not present on target-fs [or]
  86  *    if f2p(pfid)+name != f2p(tfid)
  87  *      creat .lustrerepl/tfid
  88  *      track [pfid,tfid,name]
  89  *    Else
  90  *      creat f2p[tfid]
  91  *
  92  * 2. remove
  93  *    If .lustrerepl/[tfid] is present on the target
  94  *      rm .lustrerepl/[tfid]
  95  *    Else if pfid is present on the source-fs,
  96  *      if f2p(pfid)+name is present,
  97  *        rm f2p(pfid)+name
  98  *
  99  * 3. move (spfid,sname) to (pfid,name)
 100  *    If pfid is present
 101  *      if spfid is also present, mv (spfid,sname) to (pfid,name)
 102  *      else mv .lustrerepl/[sfid] to (pfid,name)
 103  *    Else if pfid is not present,
 104  *      if spfid is present, mv (spfid,sname) .lustrerepl/[sfid]
 105  *    If moving out of .lustrerepl
 106  *      move out all its children in .lustrerepl.
 107  *      [pfid,tfid,name] tracked from (1) is used for this.
 108  */
 109
 110 #include <stdio.h>
 111 #include <stdlib.h>
 112 #include <string.h>
 113 #include <unistd.h>
 114 #include <getopt.h>
 115 #include <stdarg.h>
 116 #include <fcntl.h>
 117 #include <sys/stat.h>
 118 #include <sys/types.h>
 119 #include <errno.h>
 120 #include <limits.h>
 121 #include <utime.h>
 122 #include <sys/xattr.h>
 123
 124 #include <libcfs/libcfsutil.h>
 125 #include <lustre/lustreapi.h>
 126 #include <lustre/lustre_idl.h>
 127 #include "lustre_rsync.h"
 128
 129 #define REPLICATE_STATUS_VER 1
 130 #define CLEAR_INTERVAL 100
 131 #define DEFAULT_RSYNC_THRESHOLD 0xA00000 /* 10 MB */
 132
 133 #define TYPE_STR_LEN 16
 134
 135 #define DEFAULT_MDT "-MDT0000"
 136 #define SPECIAL_DIR ".lustrerepl"
 137 #define RSYNC "rsync"
 138 #define TYPE "type"
 139
 140 /* Debug flags */
 141 #define DINFO 1
 142 #define DTRACE 2
 143
 144 /* Not used; declared for fulfilling obd.c's dependency. */
 145 command_t cmdlist[0];
 146 extern int obd_initialize(int argc, char **argv);
 147
 148 /* Information for processing a changelog record. This structure is
 149    allocated on the heap instead of allocating large variables on the
 150    stack. */
 151 struct lr_info {
 152         long long recno;
 153         int target_no;
 154         unsigned int is_extended:1;
 155         enum changelog_rec_type type;
 156         char tfid[LR_FID_STR_LEN];
 157         char pfid[LR_FID_STR_LEN];
 158         char sfid[LR_FID_STR_LEN];
 159         char spfid[LR_FID_STR_LEN];
 160         char sname[NAME_MAX + 1];
 161         char name[NAME_MAX + 1];
 162         char src[PATH_MAX + 1];
 163         char dest[PATH_MAX + 1];
 164         char path[PATH_MAX + 1];
 165         char savedpath[PATH_MAX + 1];
 166         char link[PATH_MAX + 1];
 167         char linktmp[PATH_MAX + 1];
 168         char cmd[PATH_MAX];
 169         int bufsize;
 170         char *buf;
 171
 172         /* Variables for querying the xattributes */
 173         char *xlist;
 174         size_t xsize;
 175         char *xvalue;
 176         size_t xvsize;
 177 };
 178
 179 struct lr_parent_child_list {
 180         struct lr_parent_child_log pc_log;
 181         struct lr_parent_child_list *pc_next;
 182 };
 183
 184 struct lustre_rsync_status *status;
 185 char *statuslog;  /* Name of the status log file */
 186 int logbackedup;
 187 int noxattr;    /* Flag to turn off replicating xattrs */
 188 int noclear;    /* Flag to turn off clearing changelogs */
 189 int debug;      /* Flag to turn debugging information on and off */
 190 int verbose;    /* Verbose output */
 191 long long rec_count; /* No of changelog records that were processed */
 192 int errors;
 193 int dryrun;
 194 int use_rsync;  /* Flag to turn on use of rsync to copy data */
 195 long long rsync_threshold = DEFAULT_RSYNC_THRESHOLD;
 196 int quit;       /* Flag to stop processing the changelog; set on the
 197                    receipt of a signal */
 198 int abort_on_err = 0;
 199
 200 char rsync[PATH_MAX];
 201 char rsync_ver[PATH_MAX];
 202 struct lr_parent_child_list *parents;
 203
 204 FILE *debug_log;
 205
 206 /* Command line options */
 207 struct option long_opts[] = {
 208         {"source",      required_argument, 0, 's'},
 209         {"target",      required_argument, 0, 't'},
 210         {"mdt",         required_argument, 0, 'm'},
 211         {"user",        required_argument, 0, 'u'},
 212         {"statuslog",   required_argument, 0, 'l'},
 213         {"verbose",     no_argument,       0, 'v'},
 214         {"xattr",       required_argument, 0, 'x'},
 215         {"dry-run",     no_argument,       0, 'z'},
 216         /* Undocumented options follow */
 217         {"cl-clear",    required_argument, 0, 'c'},
 218         {"use-rsync",   no_argument,       0, 'r'},
 219         {"rsync-threshold", required_argument, 0, 'y'},
 220         {"start-recno", required_argument, 0, 'n'},
 221         {"abort-on-err",no_argument,       0, 'a'},
 222         {"debug",       required_argument, 0, 'd'},
 223         {"debuglog",    required_argument, 0, 'D'},
 224         {0, 0, 0, 0}
 225 };
 226
 227 /* Command line usage */
 228 void lr_usage()
 229 {
 230         fprintf(stderr, "\tlustre_rsync -s <lustre_root_path> -t <target_path> "
 231                 "-m <mdt> -r <user id> -l <status log>\n"
 232                 "lustre_rsync can also pick up parameters from a "
 233                 "status log created earlier.\n"
 234                 "\tlustre_rsync -l <log_file>\n"
 235                 "options:\n"
 236                 "\t--xattr <yes|no> replicate EAs\n"
 237                 "\t--abort-on-err   abort at first err\n"
 238                 "\t--verbose\n"
 239                 "\t--dry-run        don't write anything\n");
 240 }
 241
 242 #define DEBUG_ENTRY(info)                                                      \
 243         lr_debug(D_TRACE, "***** Start %lld %s (%d) %s %s %s *****\n",         \
 244                  (info)->recno, changelog_type2str((info)->type),              \
 245                  (info)->type, (info)->tfid, (info)->pfid, (info)->name);
 246
 247 #define DEBUG_EXIT(info, rc)                                                   \
 248         lr_debug(D_TRACE, "##### End %lld %s (%d) %s %s %s rc=%d #####\n",     \
 249                  (info)->recno, changelog_type2str((info)->type),              \
 250                  (info)->type, (info)->tfid, (info)->pfid, (info)->name, rc);
 251
 252 /* Print debug information. This is controlled by the value of the
 253    global variable 'debug' */
 254 void lr_debug(int level, const char *fmt, ...)
 255 {
 256         va_list ap;
 257
 258         if (level > debug)
 259                 return;
 260
 261         va_start(ap, fmt);
 262         if (debug_log != NULL)
 263                 vfprintf(debug_log, fmt, ap);
 264         else
 265                 vfprintf(stdout, fmt, ap);
 266         va_end(ap);
 267 }
 268
 269
 270 void * lr_grow_buf(void *buf, int size)
 271 {
 272         void *ptr;
 273
 274         ptr = realloc(buf, size);
 275         if (ptr == NULL)
 276                 free(buf);
 277         return ptr;
 278 }
 279
 280
 281 /* Use rsync to replicate file data */
 282 int lr_rsync_data(struct lr_info *info)
 283 {
 284         int rc;
 285         struct stat st_src, st_dest;
 286         char cmd[PATH_MAX];
 287
 288         lr_debug(DTRACE, "Syncing data%s\n", info->tfid);
 289
 290         rc = stat(info->src, &st_src);
 291         if (rc == -1) {
 292                 fprintf(stderr, "Error: Unable to stat src=%s %s\n",
 293                         info->src, info->name);
 294                 if (errno == ENOENT)
 295                         return 0;
 296                 else
 297                         return -errno;
 298         }
 299         rc = stat(info->dest, &st_dest);
 300         if (rc == -1) {
 301                 fprintf(stderr, "Error: Unable to stat dest=%s\n",
 302                         info->dest);
 303                 return -errno;
 304         }
 305
 306         if (st_src.st_mtime != st_dest.st_mtime ||
 307             st_src.st_size != st_dest.st_size) {
 308                 /* XXX spawning off an rsync for every data sync and
 309                  * waiting synchronously is bad for performance.
 310                  * librsync could possibly used here. But it does not
 311                  * seem to be of production grade. Multi-threaded
 312                  * replication is also to be considered.
 313                  */
 314                 int status;
 315                 snprintf(cmd, PATH_MAX, "%s --inplace %s %s", rsync, info->src,
 316                         info->dest);
 317                 lr_debug(DTRACE, "\t%s %s\n", cmd, info->tfid);
 318                 status = system(cmd);
 319                 if (status == -1) {
 320                         rc = -errno;
 321                 } else if (WIFEXITED(status)) {
 322                         status = WEXITSTATUS(status);
 323                         if (!status)
 324                                 rc = 0;
 325                         else if (status == 23 || status == 24)
 326                                 /* Error due to vanished source files;
 327                                    Ignore this error*/
 328                                 rc = 0;
 329                         else
 330                                 rc = -EINVAL;
 331                         if (status)
 332                                 lr_debug(DINFO, "rsync %s exited with %d %d\n",
 333                                          info->src, status, rc);
 334                 } else {
 335                         rc = -EINTR;
 336                 }
 337         } else {
 338                 lr_debug(DTRACE, "Not syncing %s and %s %s\n", info->src,
 339                          info->dest, info->tfid);
 340         }
 341
 342         return rc;
 343 }
 344
 345 int lr_copy_data(struct lr_info *info)
 346 {
 347         int fd_src = -1;
 348         int fd_dest = -1;
 349         int bufsize;
 350         int rsize;
 351         int rc = 0;
 352         struct stat st_src;
 353         struct stat st_dest;
 354
 355         fd_src = open(info->src, O_RDONLY);
 356         if (fd_src == -1)
 357                 return -errno;
 358         if (fstat(fd_src, &st_src) == -1 ||
 359             stat(info->dest, &st_dest) == -1)
 360                 goto out;
 361
 362         if (st_src.st_mtime == st_dest.st_mtime &&
 363             st_src.st_size == st_dest.st_size)
 364                 goto out;
 365
 366         if (st_src.st_size > rsync_threshold && rsync[0] != '\0') {
 367                 /* It is more efficient to use rsync to replicate
 368                    large files. Any file larger than rsync_threshold
 369                    is handed off to rsync. */
 370                 lr_debug(DTRACE, "Using rsync to replicate %s\n", info->tfid);
 371                 rc = lr_rsync_data(info);
 372                 goto out;
 373         }
 374
 375         fd_dest = open(info->dest, O_WRONLY | O_TRUNC, st_src.st_mode);
 376         if (fd_dest == -1) {
 377                 rc = -errno;
 378                 goto out;
 379         }
 380         bufsize = st_dest.st_blksize;
 381
 382         if (info->bufsize < bufsize) {
 383                 /* Grow buffer */
 384                 info->buf = lr_grow_buf(info->buf, bufsize);
 385                 if (info->buf == NULL) {
 386                         rc = -ENOMEM;
 387                         goto out;
 388                 }
 389                 info->bufsize = bufsize;
 390         }
 391
 392         while (1) {
 393                 rsize = read(fd_src, info->buf, bufsize);
 394                 if (rsize == 0) {
 395                         break;
 396                 } else if (rsize < 0) {
 397                         rc = -errno;
 398                         goto out;
 399                 }
 400                 errno = 0;
 401                 if (write(fd_dest, info->buf, rsize) != rsize) {
 402                         if (errno != 0)
 403                                 rc = -errno;
 404                         else
 405                                 rc = -EINTR;
 406                 }
 407         }
 408         fsync(fd_dest);
 409
 410 out:
 411         if (fd_src != -1)
 412                 close(fd_src);
 413         if (fd_dest != -1)
 414                 close(fd_dest);
 415
 416         return rc;
 417 }
 418
 419 /* Copy data from source to destination */
 420 int lr_sync_data(struct lr_info *info)
 421 {
 422         if (use_rsync)
 423                 return lr_rsync_data(info);
 424         else
 425                 return lr_copy_data(info);
 426 }
 427
 428 /* Copy all attributes from file src to file dest */
 429 int lr_copy_attr(char *src, char *dest)
 430 {
 431         struct stat st;
 432         struct utimbuf time;
 433
 434         if (stat(src, &st) == -1 ||
 435             chmod(dest, st.st_mode) == -1 ||
 436             chown(dest, st.st_uid, st.st_gid) == -1)
 437                 return -errno;
 438
 439         time.actime = st.st_atime;
 440         time.modtime = st.st_mtime;
 441         if (utime(dest, &time) == -1)
 442                 return -errno;
 443         return 0;
 444 }
 445
 446 /* Copy all xattrs from file info->src to info->dest */
 447 int lr_copy_xattr(struct lr_info *info)
 448 {
 449         size_t size = info->xsize;
 450         int start;
 451         int len;
 452         int rc;
 453
 454         if (noxattr)
 455                 return 0;
 456
 457         errno = 0;
 458         rc = llistxattr(info->src, info->xlist, size);
 459         lr_debug(DTRACE, "llistxattr(%s,%p) returned %d, errno=%d\n",
 460                  info->src, info->xlist, rc, errno);
 461         if ((rc > 0 && info->xlist == NULL) || errno == ERANGE) {
 462                 size = rc > PATH_MAX ? rc : PATH_MAX;
 463                 info->xlist = lr_grow_buf(info->xlist, size);
 464                 if (info->xlist == NULL)
 465                         return -ENOMEM;
 466                 info->xsize = size;
 467                 rc = llistxattr(info->src, info->xlist, size);
 468                 lr_debug(DTRACE, "llistxattr %s returned %d, errno=%d\n",
 469                          info->src, rc, errno);
 470         }
 471         if (rc < 0)
 472                 return rc;
 473
 474         len = rc;
 475         start = 0;
 476         while (start < len) {
 477                 size = info->xvsize;
 478                 rc = lgetxattr(info->src, info->xlist + start,
 479                                info->xvalue, size);
 480                 if (info->xvalue == NULL || errno == ERANGE) {
 481                         size = rc > PATH_MAX ? rc : PATH_MAX;
 482                         info->xvalue = lr_grow_buf(info->xvalue, size);
 483                         if (info->xvalue == NULL)
 484                                 return -ENOMEM;
 485                         info->xvsize = size;
 486                         rc = lgetxattr(info->src, info->xlist + start,
 487                                        info->xvalue, size);
 488                 }
 489                 lr_debug(DTRACE, "\t(%s,%d) rc=%p\n", info->xlist + start,
 490                          info->xvalue, rc);
 491                 if (rc > 0) {
 492                         size = rc;
 493                         rc = lsetxattr(info->dest, info->xlist + start,
 494                                        info->xvalue, size, 0);
 495                         lr_debug(DTRACE, "\tlsetxattr(), rc=%d, errno=%d\n",
 496                                  rc, errno);
 497                         if (rc == -1) {
 498                                 if (errno != ENOTSUP) {
 499                                         fprintf(stderr, "Error replicating "
 500                                                 " xattr for %s: %d\n",
 501                                                 info->dest, errno);
 502                                         errors++;
 503                                 }
 504                                 rc = 0;
 505                         }
 506                 }
 507                 start += strlen(info->xlist + start) + 1;
 508         }
 509
 510         lr_debug(DINFO, "setxattr: %s %s\n", info->src, info->dest);
 511
 512         return rc;
 513 }
 514
 515 /* Retrieve the filesystem path for a given FID and a given
 516    linkno. The path is returned in info->path */
 517 int lr_get_path_ln(struct lr_info *info, char *fidstr, int linkno)
 518 {
 519         long long recno = -1;
 520         int rc;
 521
 522         rc = llapi_fid2path(status->ls_source, fidstr, info->path,
 523                             PATH_MAX, &recno, &linkno);
 524         if (rc < 0 && rc != -ENOENT) {
 525                 fprintf(stderr, "fid2path error: (%s, %s) %d %s\n",
 526                         status->ls_source, fidstr, -rc, strerror(errno = -rc));
 527         }
 528
 529         return rc;
 530 }
 531
 532 /* Retrieve the filesystem path for a given FID. The path is returned
 533    in info->path */
 534 int lr_get_path(struct lr_info *info, char *fidstr)
 535 {
 536         return lr_get_path_ln(info, fidstr, 0);
 537 }
 538
 539 /* Generate the path for opening by FID */
 540 void lr_get_FID_PATH(char *mntpt, char *fidstr, char *buf, int bufsize)
 541 {
 542         /* Open-by-FID path is <mntpt>/.lustre/fid/[SEQ:OID:VER] */
 543         snprintf(buf, bufsize, "%s/%s/fid/%s", mntpt, dot_lustre_name,
 544                  fidstr);
 545         return;
 546 }
 547
 548 /* Read the symlink information into 'info->link' */
 549 int lr_get_symlink(struct lr_info *info)
 550 {
 551         int rc;
 552         char *link;
 553
 554         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
 555         rc = readlink(info->src, info->linktmp, PATH_MAX);
 556         if (rc > 0)
 557                 info->linktmp[rc] = '\0';
 558         else
 559                 return rc;
 560         lr_debug(DTRACE, "symlink: readlink returned %s\n", info->linktmp);
 561
 562         if (strncmp(info->linktmp, status->ls_source,
 563                     strlen(status->ls_source)) == 0) {
 564                 /* Strip source fs path and replace with target fs path. */
 565                 link = info->linktmp + strlen(status->ls_source);
 566                 snprintf(info->src, PATH_MAX, "%s%s",
 567                          status->ls_targets[info->target_no], link);
 568                 link = info->src;
 569         } else {
 570                 link = info->linktmp;
 571         }
 572         strlcpy(info->link, link, sizeof(info->link));
 573
 574         return rc;
 575 }
 576
 577 /* Create file/directory/device file/symlink. */
 578 int lr_mkfile(struct lr_info *info)
 579 {
 580         struct stat st;
 581         int rc = 0;
 582
 583         errno = 0;
 584         lr_debug(DINFO, "mkfile(%d) %s \n", info->type, info->dest);
 585         if (info->type == CL_MKDIR) {
 586                 rc = mkdir(info->dest, 0777);
 587         } else if (info->type == CL_SOFTLINK) {
 588                 lr_get_symlink(info);
 589                 rc = symlink(info->link, info->dest);
 590         } else if (info->type == CL_MKNOD) {
 591                 lr_get_FID_PATH(status->ls_source, info->tfid,
 592                                     info->src, PATH_MAX);
 593                 rc = stat(info->src, &st);
 594                 if (rc == -1) {
 595                         if (errno == ENOENT)
 596                                 return 0;
 597                         else
 598                                 return -errno;
 599                 }
 600                 rc = mknod(info->dest, st.st_mode, st.st_rdev);
 601         } else {
 602                 rc = mknod(info->dest, S_IFREG | 0777, 0);
 603         }
 604         if (rc)
 605                 return -errno;
 606
 607         /* Sync data and attributes */
 608         if (info->type == CL_CREATE || info->type == CL_MKDIR) {
 609                 lr_debug(DTRACE, "Syncing data and attributes %s\n",
 610                          info->tfid);
 611                 (void) lr_copy_xattr(info);
 612                 if (info->type == CL_CREATE)
 613                         rc = lr_sync_data(info);
 614                 if (!rc)
 615                         rc = lr_copy_attr(info->src, info->dest);
 616
 617                 if (rc == -ENOENT)
 618                         /* Source file has disappeared. Not an error. */
 619                         rc = 0;
 620         } else {
 621                 lr_debug(DTRACE, "Not syncing data and attributes %s\n",
 622                          info->tfid);
 623         }
 624
 625         return rc;
 626 }
 627
 628 int lr_add_pc(const char *pfid, const char *tfid, const char *name)
 629 {
 630         struct lr_parent_child_list *p;
 631         size_t len;
 632
 633         p = calloc(1, sizeof(*p));
 634         if (p == NULL)
 635                 return -ENOMEM;
 636         len = strlcpy(p->pc_log.pcl_pfid, pfid, sizeof(p->pc_log.pcl_pfid));
 637         if (len >= sizeof(p->pc_log.pcl_pfid))
 638                 goto out_err;
 639         len = strlcpy(p->pc_log.pcl_tfid, tfid, sizeof(p->pc_log.pcl_tfid));
 640         if (len >= sizeof(p->pc_log.pcl_tfid))
 641                 goto out_err;
 642         len = strlcpy(p->pc_log.pcl_name, name, sizeof(p->pc_log.pcl_name));
 643         if (len >= sizeof(p->pc_log.pcl_name))
 644                 goto out_err;
 645
 646         p->pc_next = parents;
 647         parents = p;
 648         return 0;
 649
 650 out_err:
 651         free(p);
 652         return -E2BIG;
 653 }
 654
 655 void lr_cascade_move(const char *fid, const char *dest, struct lr_info *info)
 656 {
 657         struct lr_parent_child_list *curr, *prev;
 658         char *d;
 659         int rc;
 660
 661         d = calloc(1, PATH_MAX + 1);
 662         prev = curr = parents;
 663         while (curr) {
 664                 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
 665                         snprintf(d, PATH_MAX, "%s/%s", dest,
 666                                  curr->pc_log.pcl_name);
 667                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 668                                 status->ls_targets[info->target_no],
 669                                 SPECIAL_DIR, curr->pc_log.pcl_tfid);
 670                         rc = rename(info->src, d);
 671                         if (rc == -1) {
 672                                 fprintf(stderr, "Error renaming file "
 673                                         " %s to %s: %d\n",
 674                                         info->src, d, errno);
 675                                 errors++;
 676                         }
 677                         lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
 678                         if (curr == parents)
 679                                 parents = curr->pc_next;
 680                         else
 681                                 prev->pc_next = curr->pc_next;
 682                         free(curr);
 683                         prev = curr = parents;
 684
 685                 } else {
 686                         prev = curr;
 687                         curr = curr->pc_next;
 688                 }
 689         }
 690
 691         free(d);
 692 }
 693
 694 /* remove [info->spfid, info->sfid] from parents */
 695 int lr_remove_pc(const char *pfid, const char *tfid)
 696 {
 697         struct lr_parent_child_list *curr, *prev;
 698
 699         for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
 700                 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
 701                     strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
 702                         if (curr == parents)
 703                                 parents = curr->pc_next;
 704                         else
 705                                 prev->pc_next = curr->pc_next;
 706                         free(curr);
 707                         break;
 708                 }
 709         }
 710         return 0;
 711 }
 712
 713 /* Create file under SPECIAL_DIR with its tfid as its name. */
 714 int lr_mk_special(struct lr_info *info)
 715 {
 716         int rc;
 717
 718         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 719                 status->ls_targets[info->target_no], SPECIAL_DIR,
 720                 info->tfid);
 721
 722         rc = lr_mkfile(info);
 723         if (rc)
 724                 return rc;
 725
 726         rc = lr_add_pc(info->pfid, info->tfid, info->name);
 727         return rc;
 728 }
 729
 730 /* Remove a file or directory */
 731 int lr_rmfile(struct lr_info *info)
 732 {
 733         int rc;
 734
 735         if (info->type == CL_RMDIR)
 736                 rc = rmdir(info->dest);
 737         else
 738                 rc = unlink(info->dest);
 739         if (rc == -1)
 740                 rc = -errno;
 741         return rc;
 742 }
 743
 744 /* Recursively remove directory and its contents */
 745 int lr_rm_recursive(struct lr_info *info)
 746 {
 747         int rc;
 748
 749         snprintf(info->cmd, PATH_MAX, "rm -rf %s", info->dest);
 750         rc = system(info->cmd);
 751         if (rc == -1)
 752                 rc = -errno;
 753
 754         return rc;
 755 }
 756
 757 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
 758 int lr_rm_special(struct lr_info *info)
 759 {
 760         int rc;
 761
 762         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 763                  status->ls_targets[info->target_no], SPECIAL_DIR,
 764                  info->tfid);
 765         rc = lr_rmfile(info);
 766
 767         if (rc)
 768                 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
 769                          info->dest, rc, errno);
 770         return rc;
 771 }
 772
 773 /* Replicate file and directory create events */
 774 int lr_create(struct lr_info *info)
 775 {
 776         int len;
 777         int rc1 = 0;
 778         int rc;
 779         int mkspecial = 0;
 780
 781         /* Is target FID present on the source? */
 782         rc = lr_get_path(info, info->tfid);
 783         if (rc == -ENOENT) {
 784                 /* Source file has disappeared. Not an error. */
 785                 lr_debug(DINFO, "create: tfid %s not found on"
 786                          "source-fs\n", info->tfid);
 787                 return 0;
 788         } else if (rc) {
 789                 return rc;
 790         }
 791         strcpy(info->savedpath, info->path);
 792
 793         /* Is parent FID present on the source */
 794         rc = lr_get_path(info, info->pfid);
 795         if (rc == -ENOENT) {
 796                 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
 797                          info->tfid);
 798                 mkspecial = 1;
 799         } else if (rc < 0) {
 800                 return rc;
 801         }
 802
 803         /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
 804         len = strlen(info->path);
 805         if (len == 1 && info->path[0] == '/')
 806                 snprintf(info->dest, PATH_MAX, "%s", info->name);
 807         else if (len - 1 > 0 && info->path[len - 1] == '/')
 808                 snprintf(info->dest, PATH_MAX, "%s%s", info->path, info->name);
 809         else
 810                 snprintf(info->dest, PATH_MAX, "%s/%s", info->path, info->name);
 811
 812         lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
 813                  info->savedpath);
 814         if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
 815                 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
 816                          info->tfid, info->dest, info->savedpath);
 817                 mkspecial = 1;
 818         }
 819
 820         /* Is f2p(pfid) present on the target? If not, the parent has
 821            moved */
 822         if (!mkspecial) {
 823                 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[0],
 824                         info->path);
 825                 if (access(info->dest, F_OK) != 0) {
 826                         lr_debug(DTRACE, "create: parent %s not found\n",
 827                                 info->dest);
 828                         mkspecial = 1;
 829                 }
 830         }
 831         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 832              info->target_no++) {
 833                 snprintf(info->dest, PATH_MAX, "%s/%s",
 834                         status->ls_targets[info->target_no], info->savedpath);
 835                 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
 836                                     PATH_MAX);
 837
 838                 if (!mkspecial)
 839                         rc1 = lr_mkfile(info);
 840                 if (mkspecial || rc1 == -ENOENT) {
 841                         rc1 = lr_mk_special(info);
 842                 }
 843                 if (rc1)
 844                         rc = rc1;
 845         }
 846         return rc;
 847 }
 848
 849 /* Replicate a file remove (rmdir/unlink) operation */
 850 int lr_remove(struct lr_info *info)
 851 {
 852         int rc = 0;
 853         int rc1;
 854
 855         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 856              info->target_no++) {
 857
 858                 rc1 = lr_rm_special(info);
 859                 if (!rc1)
 860                         continue;
 861
 862                 rc1 = lr_get_path(info, info->pfid);
 863                 if (rc1 == -ENOENT) {
 864                         lr_debug(DINFO, "remove: pfid %s not found\n",
 865                                  info->pfid);
 866                         continue;
 867                 }
 868                 if (rc1) {
 869                         rc = rc1;
 870                         continue;
 871                 }
 872                 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 873                         status->ls_targets[info->target_no], info->path,
 874                         info->name);
 875
 876                 rc1 = lr_rmfile(info);
 877                 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
 878                          info->dest, rc1, errno);
 879                 if (rc1 == -ENOTEMPTY)
 880                         rc1 = lr_rm_recursive(info);
 881
 882                 if (rc1) {
 883                         rc = rc1;
 884                         continue;
 885                 }
 886         }
 887         return rc;
 888 }
 889
 890 /* Replicate a rename/move operation. */
 891 int lr_move(struct lr_info *info)
 892 {
 893         int rc = 0;
 894         int rc1;
 895         int rc_dest, rc_src;
 896         int special_src = 0;
 897         int special_dest = 0;
 898         char srcpath[PATH_MAX + 1] = "";
 899
 900         LASSERT(info->is_extended);
 901
 902         rc_src = lr_get_path(info, info->spfid);
 903         if (rc_src < 0 && rc_src != -ENOENT)
 904                 return rc_src;
 905         memcpy(srcpath, info->path, strlen(info->path));
 906
 907         rc_dest = lr_get_path(info, info->pfid);
 908         if (rc_dest < 0 && rc_dest != -ENOENT)
 909                 return rc_dest;
 910
 911         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 912              info->target_no++) {
 913
 914                 if (!rc_dest) {
 915                         snprintf(info->dest, PATH_MAX, "%s/%s",
 916                                 status->ls_targets[info->target_no],
 917                                 info->path);
 918                         if (access(info->dest, F_OK) != 0) {
 919                                 rc_dest = -errno;
 920                         } else {
 921                                 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 922                                         status->ls_targets[info->target_no],
 923                                         info->path, info->name);
 924                         }
 925                         lr_debug(DINFO, "dest path %s rc_dest=%d\n", info->dest,
 926                                  rc_dest);
 927                 }
 928                 if (rc_dest == -ENOENT) {
 929                         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 930                                 status->ls_targets[info->target_no],
 931                                 SPECIAL_DIR, info->sfid);
 932                         special_dest = 1;
 933                         lr_debug(DINFO, "special dest %s\n", info->dest);
 934                 }
 935
 936                 if (!rc_src) {
 937                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 938                                 status->ls_targets[info->target_no],
 939                                 srcpath, info->sname);
 940                         lr_debug(DINFO, "src path %s rc_src=%d\n", info->src,
 941                                  rc_src);
 942                 }
 943                 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
 944                                           errno == ENOENT)) {
 945                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 946                                 status->ls_targets[info->target_no],
 947                                 SPECIAL_DIR, info->sfid);
 948                         special_src = 1;
 949                         lr_debug(DINFO, "special src %s\n", info->src);
 950                 }
 951
 952                 rc1 = 0;
 953                 errno = 0;
 954                 if (strcmp(info->src, info->dest) != 0) {
 955                         rc1 = rename(info->src, info->dest);
 956                         if (rc1 == -1)
 957                                 rc1 = -errno;
 958                         lr_debug(DINFO, "rename returns %d\n", rc1);
 959                 }
 960
 961                 if (special_src) {
 962                         rc1 = lr_remove_pc(info->spfid, info->sfid);
 963                         if (!special_dest)
 964                                 lr_cascade_move(info->sfid, info->dest, info);
 965                 }
 966                 if (special_dest)
 967                         rc1 = lr_add_pc(info->pfid, info->sfid, info->name);
 968
 969                 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
 970                          info->src, info->dest, rc1, errno);
 971                 if (rc1)
 972                         rc = rc1;
 973         }
 974         return rc;
 975 }
 976
 977 /* Replicate a hard link */
 978 int lr_link(struct lr_info *info)
 979 {
 980         int i;
 981         int len;
 982         int rc;
 983         int rc1;
 984         struct stat st;
 985
 986         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
 987         rc = stat(info->src, &st);
 988         if (rc == -1)
 989                 return -errno;
 990
 991         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 992              info->target_no++) {
 993
 994                 info->src[0] = 0;
 995                 info->dest[0] = 0;
 996                 rc1 = 0;
 997
 998                 /* Search through the hardlinks to get the src and dest */
 999                 for (i = 0; i < st.st_nlink && (info->src[0] == 0 ||
1000                                                 info->dest[0] == 0); i++) {
1001                         rc1 = lr_get_path_ln(info, info->tfid, i);
1002                         lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
1003                                  info->path, info->name, i, rc1);
1004                         if (rc1)
1005                                 break;
1006
1007                         len = strlen(info->path) - strlen(info->name);
1008                         if (len > 0 && strcmp(info->path + len,
1009                                               info->name) == 0)
1010                                 snprintf(info->dest, PATH_MAX, "%s/%s",
1011                                         status->ls_targets[info->target_no],
1012                                         info->path);
1013                         else if (info->src[0] == 0)
1014                                 snprintf(info->src, PATH_MAX, "%s/%s",
1015                                         status->ls_targets[info->target_no],
1016                                         info->path);
1017                 }
1018
1019                 if (rc1) {
1020                         rc = rc1;
1021                         continue;
1022                 }
1023
1024                 if (info->src[0] == 0 || info->dest[0] == 0)
1025                         /* Could not find the source or destination.
1026                          This can happen when some links don't exist
1027                          anymore. */
1028                         return -EINVAL;
1029
1030                 if (info->src[0] == 0)
1031                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
1032                                 status->ls_targets[info->target_no],
1033                                 SPECIAL_DIR, info->tfid);
1034                 else if (info->dest[0] == 0)
1035                         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
1036                                 status->ls_targets[info->target_no],
1037                                 SPECIAL_DIR, info->tfid);
1038
1039                 rc1 = link(info->src, info->dest);
1040                 lr_debug(rc1?0:DINFO, "link: %s [to] %s; rc1=%d %s\n",
1041                          info->src, info->dest, rc1, strerror(errno));
1042
1043                 if (rc1)
1044                         rc = rc1;
1045         }
1046         return rc;
1047 }
1048
1049 /* Replicate file attributes */
1050 int lr_setattr(struct lr_info *info)
1051 {
1052         int rc1;
1053         int rc;
1054
1055         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1056
1057         rc = lr_get_path(info, info->tfid);
1058         if (rc == -ENOENT)
1059                 lr_debug(DINFO, "setattr: %s not present on source-fs\n",
1060                          info->src);
1061         if (rc)
1062                 return rc;
1063
1064         for (info->target_no = 0; info->target_no < status->ls_num_targets;
1065              info->target_no++) {
1066
1067                 snprintf(info->dest, PATH_MAX, "%s/%s",
1068                          status->ls_targets[info->target_no], info->path);
1069                 lr_debug(DINFO, "setattr: %s %s %s", info->src, info->dest,
1070                          info->tfid);
1071
1072                 rc1 = lr_sync_data(info);
1073                 if (!rc1)
1074                         rc1 = lr_copy_attr(info->src, info->dest);
1075                 if (rc1)
1076                         rc = rc1;
1077         }
1078         return rc;
1079 }
1080
1081 /* Replicate xattrs */
1082 int lr_setxattr(struct lr_info *info)
1083 {
1084         int rc, rc1;
1085
1086         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1087
1088         rc = lr_get_path(info, info->tfid);
1089         if (rc == -ENOENT)
1090                 lr_debug(DINFO, "setxattr: %s not present on source-fs\n",
1091                          info->src);
1092         if (rc)
1093                 return rc;
1094
1095         for (info->target_no = 0; info->target_no < status->ls_num_targets;
1096              info->target_no++) {
1097
1098                 snprintf(info->dest, PATH_MAX, "%s/%s",
1099                         status->ls_targets[info->target_no], info->path);
1100                 lr_debug(DINFO, "setxattr: %s %s %s\n", info->src, info->dest,
1101                          info->tfid);
1102
1103                 rc1 = lr_copy_xattr(info);
1104                 if (rc1)
1105                         rc = rc1;
1106         }
1107
1108         return rc;
1109 }
1110
1111 /* Parse a line of changelog entry */
1112 int lr_parse_line(void *priv, struct lr_info *info)
1113 {
1114         struct changelog_ext_rec *rec;
1115
1116         if (llapi_changelog_recv(priv, &rec) != 0)
1117                 return -1;
1118
1119         info->is_extended = CHANGELOG_REC_EXTENDED(rec);
1120         info->recno = rec->cr_index;
1121         info->type = rec->cr_type;
1122         sprintf(info->tfid, DFID, PFID(&rec->cr_tfid));
1123         sprintf(info->pfid, DFID, PFID(&rec->cr_pfid));
1124         strncpy(info->name, rec->cr_name, rec->cr_namelen);
1125         info->name[rec->cr_namelen] = '\0';
1126
1127         if (fid_is_sane(&rec->cr_sfid)) {
1128                 sprintf(info->sfid, DFID, PFID(&rec->cr_sfid));
1129                 sprintf(info->spfid, DFID, PFID(&rec->cr_spfid));
1130                 strncpy(info->sname, changelog_rec_sname(rec),
1131                         changelog_rec_snamelen(rec));
1132                 info->sname[changelog_rec_snamelen(rec)] = '\0';
1133
1134                 if (verbose > 1)
1135                         printf("Rec %lld: %d %s %s\n", info->recno, info->type,
1136                                 info->name, info->sname);
1137         } else {
1138                 if (verbose > 1)
1139                         printf("Rec %lld: %d %s\n", info->recno, info->type,
1140                                 info->name);
1141         }
1142
1143         llapi_changelog_free(&rec);
1144
1145         rec_count++;
1146         return 0;
1147 }
1148
1149 /* Initialize the replication parameters */
1150 int lr_init_status()
1151 {
1152         size_t size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1153
1154         if (status != NULL)
1155                 return 0;
1156         status = calloc(size, 1);
1157         if (status == NULL)
1158                 return -ENOMEM;
1159         status->ls_version = REPLICATE_STATUS_VER;
1160         status->ls_size = size;
1161         status->ls_last_recno = -1;
1162         return 0;
1163 }
1164
1165 /* Make a backup of the statuslog */
1166 void lr_backup_log()
1167 {
1168         char backupfile[PATH_MAX];
1169
1170         if (logbackedup)
1171                 return;
1172         snprintf(backupfile, PATH_MAX, "%s.old", statuslog);
1173         (void) rename(statuslog, backupfile);
1174         logbackedup = 1;
1175
1176         return;
1177 }
1178
1179 /* Save replication parameters to a statuslog. */
1180 int lr_write_log()
1181 {
1182         int fd;
1183         size_t size;
1184         size_t write_size = status->ls_size;
1185         struct lr_parent_child_list *curr;
1186         int rc = 0;
1187
1188         if (statuslog == NULL)
1189                 return 0;
1190
1191         lr_backup_log();
1192
1193         fd = open(statuslog, O_WRONLY | O_CREAT | O_SYNC,
1194                              S_IRUSR | S_IWUSR);
1195         if (fd == -1) {
1196                 fprintf(stderr, "Error opening log file for writing (%s)\n",
1197                         statuslog);
1198                 return -1;
1199         }
1200         errno = 0;
1201         size = write(fd, status, write_size);
1202         if (size != write_size) {
1203                 fprintf(stderr, "Error writing to log file (%s) %d\n",
1204                         statuslog, errno);
1205                 close(fd);
1206                 return -1;
1207         }
1208
1209         for (curr = parents; curr; curr = curr->pc_next) {
1210                 size = write(fd, &curr->pc_log, sizeof(curr->pc_log));
1211                 if (size != sizeof(curr->pc_log)) {
1212                         fprintf(stderr, "Error writing to log file (%s) %d\n",
1213                                 statuslog, errno);
1214                         rc = -1;
1215                         break;
1216                 }
1217         }
1218         close(fd);
1219         return rc;
1220 }
1221
1222 /* Read statuslog and populate the replication parameters.  Command
1223  * line parameters take precedence over parameters in the log file.*/
1224 int lr_read_log()
1225 {
1226         struct lr_parent_child_list *tmp;
1227         struct lr_parent_child_log rec;
1228         struct lustre_rsync_status *s;
1229         int fd = -1;
1230         size_t size;
1231         size_t read_size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1232         int rc = 0;
1233
1234         if (statuslog == NULL)
1235                 return 0;
1236
1237         s = calloc(1, read_size);
1238         if (s == NULL) {
1239                 rc = -ENOMEM;
1240                 goto out;
1241         }
1242
1243         fd = open(statuslog, O_RDONLY);
1244         if (fd == -1) {
1245                 rc = -errno;
1246                 goto out;
1247         }
1248
1249         size = read(fd, s, read_size);
1250         if (size != read_size) {
1251                 rc = -EINVAL;
1252                 goto out;
1253         }
1254
1255         if (read_size < s->ls_size) {
1256                 read_size = s->ls_size;
1257                 s = lr_grow_buf(s, read_size);
1258                 if (s == NULL) {
1259                         rc = -ENOMEM;
1260                         goto out;
1261                 }
1262
1263                 if (lseek(fd, 0, SEEK_SET) == -1) {
1264                         rc = -ENOMEM;
1265                         goto out;
1266                 }
1267
1268                 size = read(fd, s, read_size);
1269                 if (size != read_size) {
1270                         rc = -EINVAL;
1271                         goto out;
1272                 }
1273         }
1274
1275         while (read(fd, &rec, sizeof(rec)) != 0) {
1276                 tmp = calloc(1, sizeof(*tmp));
1277                 if (!tmp) {
1278                         rc = -ENOMEM;
1279                         goto out;
1280                 }
1281
1282                 tmp->pc_log = rec;
1283                 tmp->pc_next = parents;
1284                 parents = tmp;
1285         }
1286
1287         /* copy uninitialized fields to status */
1288         if (status->ls_num_targets == 0) {
1289                 if (status->ls_size != s->ls_size) {
1290                         status = lr_grow_buf(status, s->ls_size);
1291                         if (status == NULL) {
1292                                 rc = -ENOMEM;
1293                                 goto out;
1294                         }
1295
1296                         status->ls_size = s->ls_size;
1297                 }
1298                 status->ls_num_targets = s->ls_num_targets;
1299                 memcpy(status->ls_targets, s->ls_targets,
1300                        (PATH_MAX + 1) * s->ls_num_targets);
1301         }
1302         if (status->ls_last_recno == -1)
1303                 status->ls_last_recno = s->ls_last_recno;
1304
1305         if (status->ls_registration[0] == '\0')
1306                 strlcpy(status->ls_registration, s->ls_registration,
1307                         sizeof(status->ls_registration));
1308
1309         if (status->ls_mdt_device[0] == '\0')
1310                 strlcpy(status->ls_mdt_device, s->ls_mdt_device,
1311                         sizeof(status->ls_mdt_device));
1312
1313         if (status->ls_source_fs[0] == '\0')
1314                 strlcpy(status->ls_source_fs, s->ls_source_fs,
1315                         sizeof(status->ls_source_fs));
1316
1317         if (status->ls_source[0] == '\0')
1318                 strlcpy(status->ls_source, s->ls_source,
1319                         sizeof(status->ls_source));
1320
1321  out:
1322         if (fd != -1)
1323                 close(fd);
1324         if (s)
1325                 free(s);
1326         return rc;
1327 }
1328
1329 /* Clear changelogs every CLEAR_INTERVAL records or at the end of
1330    processing. */
1331 int lr_clear_cl(struct lr_info *info, int force)
1332 {
1333         char            mdt_device[LR_NAME_MAXLEN + 1];
1334         long long       rec;
1335         int             rc = 0;
1336
1337         if (force || info->recno > status->ls_last_recno + CLEAR_INTERVAL) {
1338                 if (info->type == CL_RENAME)
1339                         rec = info->recno + 1;
1340                 else
1341                         rec = info->recno;
1342                 if (!noclear && !dryrun) {
1343                         /* llapi_changelog_clear modifies the mdt
1344                          * device name so make a copy of it until this
1345                          * is fixed.
1346                         */
1347                         strlcpy(mdt_device, status->ls_mdt_device,
1348                                 sizeof(mdt_device));
1349                         rc = llapi_changelog_clear(mdt_device,
1350                                                    status->ls_registration,
1351                                                    rec);
1352                         if (rc)
1353                                 printf("Changelog clear (%s, %s, %lld) "
1354                                        "returned %d\n", status->ls_mdt_device,
1355                                        status->ls_registration, rec, rc);
1356                 }
1357                 if (!rc && !dryrun) {
1358                         status->ls_last_recno = rec;
1359                         lr_write_log();
1360
1361                 }
1362         }
1363
1364         return rc;
1365 }
1366
1367 /* Locate a usable version of rsync. At this point we'll use any
1368    version. */
1369 int lr_locate_rsync()
1370 {
1371         FILE *fp;
1372         int len;
1373
1374         /* Locate rsync */
1375         snprintf(rsync, PATH_MAX, "%s -p %s", TYPE, RSYNC);
1376         fp = popen(rsync, "r");
1377         if (fp == NULL)
1378                 return -1;
1379
1380         if (fgets(rsync, PATH_MAX, fp) == NULL) {
1381                 fclose(fp);
1382                 return -1;
1383         }
1384
1385         len = strlen(rsync);
1386         if (len > 0 && rsync[len - 1] == '\n')
1387                 rsync[len - 1] = '\0';
1388         fclose(fp);
1389
1390         /* Determine the version of rsync */
1391         snprintf(rsync_ver, PATH_MAX, "%s --version", rsync);
1392         fp = popen(rsync_ver, "r");
1393         if (fp == NULL)
1394                 return -1;
1395
1396         if (fgets(rsync_ver, PATH_MAX, fp) == NULL) {
1397                 fclose(fp);
1398                 return -1;
1399         }
1400         len = strlen(rsync_ver);
1401         if (len > 0 && rsync_ver[len - 1] == '\n')
1402                 rsync_ver[len - 1] = '\0';
1403         fclose(fp);
1404
1405         return 0;
1406
1407 }
1408
1409 /* Print the replication parameters */
1410 void lr_print_status(struct lr_info *info)
1411 {
1412         int i;
1413
1414         if (!verbose)
1415                 return;
1416
1417         printf("Lustre filesystem: %s\n", status->ls_source_fs);
1418         printf("MDT device: %s\n", status->ls_mdt_device);
1419         printf("Source: %s\n", status->ls_source);
1420         for (i = 0; i < status->ls_num_targets; i++)
1421                 printf("Target: %s\n", status->ls_targets[i]);
1422         if (statuslog != NULL)
1423                 printf("Statuslog: %s\n", statuslog);
1424         printf("Changelog registration: %s\n", status->ls_registration);
1425         printf("Starting changelog record: "LPD64"\n", status->ls_last_recno);
1426         if (noxattr)
1427                 printf("Replicate xattrs: no\n");
1428         if (noclear)
1429                 printf("Clear changelog after use: no\n");
1430         if (use_rsync)
1431                 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
1432 }
1433
1434 void lr_print_failure(struct lr_info *info, int rc)
1435 {
1436         fprintf(stderr, "Replication of operation failed(%d):"
1437                 " %lld %s (%d) %s %s %s\n", rc, info->recno,
1438                 changelog_type2str(info->type), info->type, info->tfid,
1439                 info->pfid, info->name);
1440 }
1441
1442 /* Replicate filesystem operations from src_path to target_path */
1443 int lr_replicate()
1444 {
1445         void *changelog_priv;
1446         struct lr_info *info;
1447         struct lr_info *ext = NULL;
1448         time_t start;
1449         int xattr_not_supp;
1450         int i;
1451         int rc;
1452
1453         start = time(NULL);
1454
1455         info = calloc(1, sizeof(struct lr_info));
1456         if (info == NULL)
1457                 return -ENOMEM;
1458
1459         rc = llapi_search_fsname(status->ls_source, status->ls_source_fs);
1460         if (rc) {
1461                 fprintf(stderr, "Source path is not a valid Lustre client "
1462                         "mountpoint.\n");
1463                 goto out;
1464         }
1465         if (status->ls_mdt_device[0] == '\0')
1466                 snprintf(status->ls_mdt_device, LR_NAME_MAXLEN, "%s%s",
1467                         status->ls_source_fs, DEFAULT_MDT);
1468
1469         ext = calloc(1, sizeof(struct lr_info));
1470         if (ext == NULL) {
1471                 rc = -ENOMEM;
1472                 goto out;
1473         }
1474
1475         for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1476                 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[i],
1477                         SPECIAL_DIR);
1478                 rc = mkdir(info->dest, 0777);
1479                 if (rc == -1 && errno != EEXIST) {
1480                         fprintf(stderr, "Error writing to target path %s.\n",
1481                                 status->ls_targets[i]);
1482                         rc = -errno;
1483                         goto out;
1484                 }
1485                 rc = llistxattr(info->src, info->xlist, info->xsize);
1486                 if (rc == -1 && errno == ENOTSUP) {
1487                         fprintf(stderr, "xattrs not supported on %s\n",
1488                                 status->ls_targets[i]);
1489                         xattr_not_supp++;
1490                 }
1491         }
1492         if (xattr_not_supp == status->ls_num_targets)
1493                 /* None of the targets support xattrs. */
1494                 noxattr = 1;
1495
1496         lr_print_status(info);
1497
1498         /* Open changelogs for consumption*/
1499         rc = llapi_changelog_start(&changelog_priv, CHANGELOG_FLAG_BLOCK,
1500                                    status->ls_source_fs, status->ls_last_recno);
1501         if (rc < 0) {
1502                 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1503                         status->ls_source_fs);
1504                 goto out;
1505         }
1506
1507         while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1508                 rc = 0;
1509                 if (info->type == CL_RENAME && !info->is_extended) {
1510                         /* Newer rename operations extends changelog to store
1511                          * source file information, but old changelog has
1512                          * another record.
1513                          */
1514                         if (lr_parse_line(changelog_priv, ext) != 0)
1515                                 break;
1516                         memcpy(info->sfid, info->tfid, sizeof(info->sfid));
1517                         memcpy(info->spfid, info->pfid, sizeof(info->spfid));
1518                         memcpy(info->tfid, ext->tfid, sizeof(info->tfid));
1519                         memcpy(info->pfid, ext->pfid, sizeof(info->pfid));
1520                         strlcpy(info->sname, info->name, sizeof(info->sname));
1521                         strlcpy(info->name, ext->name, sizeof(info->name));
1522                         info->is_extended = 1;
1523                 }
1524
1525                 if (dryrun)
1526                         continue;
1527
1528                 DEBUG_ENTRY(info);
1529
1530                 switch(info->type) {
1531                 case CL_CREATE:
1532                 case CL_MKDIR:
1533                 case CL_MKNOD:
1534                 case CL_SOFTLINK:
1535                         rc = lr_create(info);
1536                         break;
1537                 case CL_RMDIR:
1538                 case CL_UNLINK:
1539                         rc = lr_remove(info);
1540                         break;
1541                 case CL_RENAME:
1542                         rc = lr_move(info);
1543                         break;
1544                 case CL_HARDLINK:
1545                         rc = lr_link(info);
1546                         break;
1547                 case CL_TRUNC:
1548                 case CL_SETATTR:
1549                         rc = lr_setattr(info);
1550                         break;
1551                 case CL_XATTR:
1552                         rc = lr_setxattr(info);
1553                         break;
1554                 case CL_CLOSE:
1555                 case CL_EXT:
1556                 case CL_OPEN:
1557                 case CL_LAYOUT:
1558                 case CL_MARK:
1559                         /* Nothing needs to be done for these entries */
1560                         /* fallthrough */
1561                 default:
1562                         break;
1563                 }
1564                 DEBUG_EXIT(info, rc);
1565                 if (rc && rc != -ENOENT) {
1566                         lr_print_failure(info, rc);
1567                         errors++;
1568                         if (abort_on_err)
1569                                 break;
1570                 }
1571                 lr_clear_cl(info, 0);
1572                 if (debug) {
1573                         bzero(info, sizeof(struct lr_info));
1574                         bzero(ext, sizeof(struct lr_info));
1575                 }
1576         }
1577
1578         llapi_changelog_fini(&changelog_priv);
1579
1580         if (errors || verbose)
1581                 printf("Errors: %d\n", errors);
1582
1583         /* Clear changelog records used so far */
1584         lr_clear_cl(info, 1);
1585
1586         if (verbose) {
1587                 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1588                 printf("Changelog records consumed: %lld\n", rec_count);
1589         }
1590
1591         rc = 0;
1592
1593 out:
1594         if (info != NULL)
1595                 free(info);
1596         if (ext != NULL)
1597                 free(ext);
1598
1599         return rc;
1600 }
1601
1602 void
1603 termination_handler (int signum)
1604 {
1605         /* Set a flag for the replicator to gracefully shutdown */
1606         quit = 1;
1607         printf("lustre_rsync halting.\n");
1608 }
1609
1610 int main(int argc, char *argv[])
1611 {
1612         int newsize;
1613         int numtargets = 0;
1614         int rc = 0;
1615
1616         if ((rc = lr_init_status()) != 0)
1617                 return rc;
1618
1619         while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:D:",
1620                                  long_opts, NULL)) >= 0) {
1621                 switch (rc) {
1622                 case 'a':
1623                         /* Assume absolute paths */
1624                         abort_on_err++;
1625                         break;
1626                 case 's':
1627                         /* Assume absolute paths */
1628                         strlcpy(status->ls_source, optarg,
1629                                 sizeof(status->ls_source));
1630                         break;
1631                 case 't':
1632                         status->ls_num_targets++;
1633                         numtargets++;
1634                         if (numtargets != status->ls_num_targets) {
1635                                 /* Targets were read from a log
1636                                    file. The ones specified on the
1637                                    command line take precedence. The
1638                                    ones from the log file will be
1639                                    ignored. */
1640                                 status->ls_num_targets = numtargets;
1641                         }
1642                         newsize = sizeof (struct lustre_rsync_status) +
1643                                 (status->ls_num_targets * (PATH_MAX + 1));
1644                         if (status->ls_size != newsize) {
1645                                 status->ls_size = newsize;
1646                                 status = lr_grow_buf(status, newsize);
1647                                 if (status == NULL)
1648                                         return -ENOMEM;
1649                         }
1650                         strlcpy(status->ls_targets[status->ls_num_targets - 1],
1651                                 optarg, sizeof(status->ls_targets[0]));
1652                         break;
1653                 case 'm':
1654                         strlcpy(status->ls_mdt_device, optarg,
1655                                 sizeof(status->ls_mdt_device));
1656                         break;
1657                 case 'u':
1658                         strlcpy(status->ls_registration, optarg,
1659                                 sizeof(status->ls_registration));
1660                         break;
1661                 case 'l':
1662                         statuslog = optarg;
1663                         (void) lr_read_log();
1664                         break;
1665                 case 'v':
1666                         verbose++;
1667                         break;
1668                 case 'x':
1669                         if (strcmp("no", optarg) == 0) {
1670                                 noxattr = 1;
1671                         } else if (strcmp("yes", optarg) != 0) {
1672                                 printf("Invalid parameter %s. "
1673                                        "Specify --xattr=no or --xattr=yes\n",
1674                                        optarg);
1675                                 return -1;
1676                         }
1677                         break;
1678                 case 'z':
1679                         dryrun = 1;
1680                         break;
1681                 case 'c':
1682                         /* Undocumented option cl-clear */
1683                         if (strcmp("no", optarg) == 0) {
1684                                 noclear = 1;
1685                         } else if (strcmp("yes", optarg) != 0) {
1686                                 printf("Invalid parameter %s. "
1687                                        "Specify --cl-clear=no "
1688                                        "or --cl-clear=yes\n",
1689                                        optarg);
1690                                 return -1;
1691                         }
1692                         break;
1693                 case 'r':
1694                         /* Undocumented option use-rsync */
1695                         use_rsync = 1;
1696                         break;
1697                 case 'y':
1698                         /* Undocumented option rsync-threshold */
1699                         rsync_threshold = atol(optarg);
1700                         break;
1701                 case 'n':
1702                         /* Undocumented option start-recno */
1703                         status->ls_last_recno = atol(optarg);
1704                         break;
1705                 case 'd':
1706                         /* Undocumented option debug */
1707                         debug = atoi(optarg);
1708                         if (debug < 0 || debug > 2)
1709                                 debug = 0;
1710                         break;
1711                 case 'D':
1712                         /* Undocumented option debug log file */
1713                         if (debug_log != NULL)
1714                                 fclose(debug_log);
1715                         debug_log = fopen(optarg, "a");
1716                         if (debug_log == NULL) {
1717                                 printf("Cannot open %s for debug log\n",
1718                                        optarg);
1719                                 return -1;
1720                         }
1721                         break;
1722                 default:
1723                         fprintf(stderr, "error: %s: option '%s' "
1724                                 "unrecognized.\n", argv[0], argv[optind - 1]);
1725                         lr_usage();
1726                         return -1;
1727                 }
1728         }
1729
1730         if (status->ls_last_recno == -1)
1731                 status->ls_last_recno = 0;
1732         if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1733                 /* No registration ID was passed in. */
1734                 printf("Please specify changelog consumer registration id.\n");
1735                 lr_usage();
1736                 return -1;
1737         }
1738         if (strnlen(status->ls_source, PATH_MAX) == 0) {
1739                 fprintf(stderr, "Please specify the source path.\n");
1740                 lr_usage();
1741                 return -1;
1742         }
1743         if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1744                 fprintf(stderr, "Please specify the target path.\n");
1745                 lr_usage();
1746                 return -1;
1747         }
1748
1749         /* This plumbing is needed for some of the ioctls behind
1750            llapi calls to work. */
1751         if (obd_initialize(argc, argv) < 0) {
1752                 fprintf(stderr, "obd_initialize failed.\n");
1753                 exit(-1);
1754         }
1755
1756         rc = lr_locate_rsync();
1757         if (use_rsync && rc != 0) {
1758                 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1759                 exit(-1);
1760         }
1761
1762         signal(SIGINT, termination_handler);
1763         signal(SIGHUP, termination_handler);
1764         signal(SIGTERM, termination_handler);
1765
1766         rc = lr_replicate();
1767
1768         if (debug_log != NULL)
1769                 fclose(debug_log);
1770         return rc;
1771 }