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