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  *  name - The name of the target file (at the time of the operation)
  64  *
  65  * With just this information, it is not alwasy possible to determine
  66  * the file paths for each operation. For instance, if pfid does not
  67  * exist on the source-fs (due to a subsequent deletion), its path
  68  * cannot be queried. In such cases, lustre_rsync keeps the files in a
  69  * special directory ("/.lustrerepl"). Once all the operations in a
  70  * changelog are replayed, all the files in this special directory
  71  * will get moved to the location as in the source-fs.
  72  *
  73  * Shorthand used: f2p(tfid) = fid2path(tfid)
  74  *
  75  * The following are the metadata operations of interest.
  76  * 1. creat
  77  *    If tfid is absent on the source-fs, ignore this operation
  78  *    If pfid is absent on the source-fs [or]
  79  *    if f2p(pfid) is not present on target-fs [or]
  80  *    if f2p(pfid)+name != f2p(tfid)
  81  *      creat .lustrerepl/tfid
  82  *      track [pfid,tfid,name]
  83  *    Else
  84  *      creat f2p[tfid]
  85  *
  86  * 2. remove
  87  *    If .lustrerepl/[tfid] is present on the target
  88  *      rm .lustrerepl/[tfid]
  89  *    Else if pfid is present on the source-fs,
  90  *      if f2p(pfid)+name is present,
  91  *        rm f2p(pfid)+name(pfid,name)
  92  *
  93  * 3. move (pfid1,name1) to (pfid2,name2)
  94  *    If pfid2 is present
  95  *      if pfid1 is also present, mv (pfid1,name1) to (pfid2,name2)
  96  *      else mv .lustrerepl/[tfid] to (pfid2,name2)
  97  *    If pfid2 is not present,
  98  *      if pfid1 is present, mv (pfid1,name1) .lustrerepl/[tfid]
  99  *    If moving out of .lustrerepl
 100  *      move out all its children in .lustrerepl.
 101  *      [pfid,tfid,name] tracked from (1) is used for this.
 102  */
 103
 104 #include <stdio.h>
 105 #include <stdlib.h>
 106 #include <string.h>
 107 #include <unistd.h>
 108 #include <getopt.h>
 109 #include <stdarg.h>
 110 #include <fcntl.h>
 111 #include <sys/stat.h>
 112 #include <sys/types.h>
 113 #include <errno.h>
 114 #include <limits.h>
 115 #include <utime.h>
 116 #include <sys/xattr.h>
 117
 118 #include <libcfs/libcfs.h>
 119 #include <libcfs/libcfsutil.h>
 120 #include <lustre/liblustreapi.h>
 121 #include <lustre/lustre_idl.h>
 122 #include "lustre_rsync.h"
 123
 124 #define REPLICATE_STATUS_VER 1
 125 #define CLEAR_INTERVAL 100
 126 #define DEFAULT_RSYNC_THRESHOLD 0xA00000 /* 10 MB */
 127
 128 #define TYPE_STR_LEN 16
 129
 130 #define DEFAULT_MDT "-MDT0000"
 131 #define SPECIAL_DIR ".lustrerepl"
 132 #define RSYNC "rsync"
 133 #define TYPE "type"
 134
 135 /* Debug flags */
 136 #define DINFO 1
 137 #define DTRACE 2
 138
 139 /* Not used; declared for fulfilling obd.c's dependency. */
 140 command_t cmdlist[0];
 141 extern int obd_initialize(int argc, char **argv);
 142
 143 /* Information for processing a changelog record. This structure is
 144    allocated on the heap instead of allocating large variables on the
 145    stack. */
 146 struct lr_info {
 147         long long recno;
 148         int target_no;
 149         enum changelog_rec_type type;
 150         char pfid[LR_FID_STR_LEN];
 151         char tfid[LR_FID_STR_LEN];
 152         char name[PATH_MAX + 1];
 153         char src[PATH_MAX + 1];
 154         char dest[PATH_MAX + 1];
 155         char path[PATH_MAX + 1];
 156         char savedpath[PATH_MAX + 1];
 157         char link[PATH_MAX + 1];
 158         char linktmp[PATH_MAX + 1];
 159         char cmd[PATH_MAX];
 160         int bufsize;
 161         char *buf;
 162
 163         /* Variables for querying the xattributes */
 164         char *xlist;
 165         size_t xsize;
 166         char *xvalue;
 167         size_t xvsize;
 168 };
 169
 170 struct lr_parent_child_list {
 171         struct lr_parent_child_log pc_log;
 172         struct lr_parent_child_list *pc_next;
 173 };
 174
 175 struct lustre_rsync_status *status;
 176 char *statuslog;  /* Name of the status log file */
 177 int logbackedup;
 178 int noxattr;    /* Flag to turn off replicating xattrs */
 179 int noclear;    /* Flag to turn off clearing changelogs */
 180 int debug;      /* Flag to turn debugging information on and off */
 181 int verbose;    /* Verbose output */
 182 long long rec_count; /* No of changelog records that were processed */
 183 int errors;
 184 int dryrun;
 185 int use_rsync;  /* Flag to turn on use of rsync to copy data */
 186 long long rsync_threshold = DEFAULT_RSYNC_THRESHOLD;
 187 int quit;       /* Flag to stop processing the changelog; set on the
 188                    receipt of a signal */
 189 int abort_on_err = 0;
 190
 191 char rsync[PATH_MAX];
 192 char rsync_ver[PATH_MAX];
 193 struct lr_parent_child_list *parents;
 194
 195 /* Command line options */
 196 struct option long_opts[] = {
 197         {"source",      required_argument, 0, 's'},
 198         {"target",      required_argument, 0, 't'},
 199         {"mdt",         required_argument, 0, 'm'},
 200         {"user",        required_argument, 0, 'u'},
 201         {"statuslog",   required_argument, 0, 'l'},
 202         {"verbose",     no_argument,       0, 'v'},
 203         {"xattr",       required_argument, 0, 'x'},
 204         {"dry-run",     no_argument,       0, 'z'},
 205         /* Undocumented options follow */
 206         {"cl-clear",    required_argument, 0, 'c'},
 207         {"use-rsync",   no_argument,       0, 'r'},
 208         {"rsync-threshold", required_argument, 0, 'y'},
 209         {"start-recno", required_argument, 0, 'n'},
 210         {"abort-on-err",no_argument,       0, 'a'},
 211         {"debug",       required_argument, 0, 'd'},
 212         {0, 0, 0, 0}
 213 };
 214
 215 /* Command line usage */
 216 void lr_usage()
 217 {
 218         fprintf(stderr, "\tlustre_rsync -s <lustre_root_path> -t <target_path> "
 219                 "-m <mdt> -r <user id> -l <status log>\n"
 220                 "lustre_rsync can also pick up parameters from a "
 221                 "status log created earlier.\n"
 222                 "\tlustre_rsync -l <log_file>\n"
 223                 "options:\n"
 224                 "\t--xattr <yes|no> replicate EAs\n"
 225                 "\t--abort-on-err   abort at first err\n"
 226                 "\t--verbose\n"
 227                 "\t--dry-run        don't write anything\n");
 228 }
 229
 230 /* Print debug information. This is controlled by the value of the
 231    global variable 'debug' */
 232 void lr_debug(int level, const char *fmt, ...)
 233 {
 234         va_list ap;
 235
 236         if (level > debug)
 237                 return;
 238
 239         va_start(ap, fmt);
 240         vprintf(fmt, ap);
 241         va_end(ap);
 242 }
 243
 244
 245 void * lr_grow_buf(void *buf, int size)
 246 {
 247         void *ptr;
 248
 249         ptr = realloc(buf, size);
 250         if (ptr == NULL)
 251                 free(buf);
 252         return ptr;
 253 }
 254
 255
 256 /* Use rsync to replicate file data */
 257 int lr_rsync_data(struct lr_info *info)
 258 {
 259         int rc;
 260         struct stat st_src, st_dest;
 261         char cmd[PATH_MAX];
 262
 263         lr_debug(DTRACE, "Syncing data%s\n", info->tfid);
 264
 265         rc = stat(info->src, &st_src);
 266         if (rc == -1) {
 267                 fprintf(stderr, "Error: Unable to stat src=%s %s\n",
 268                         info->src, info->name);
 269                 if (errno == ENOENT)
 270                         return 0;
 271                 else
 272                         return -errno;
 273         }
 274         rc = stat(info->dest, &st_dest);
 275         if (rc == -1) {
 276                 fprintf(stderr, "Error: Unable to stat dest=%s\n",
 277                         info->dest);
 278                 return -errno;
 279         }
 280
 281         if (st_src.st_mtime != st_dest.st_mtime ||
 282             st_src.st_size != st_dest.st_size) {
 283                 /* XXX spawning off an rsync for every data sync and
 284                  * waiting synchronously is bad for performance.
 285                  * librsync could possibly used here. But it does not
 286                  * seem to be of production grade. Multi-threaded
 287                  * replication is also to be considered.
 288                  */
 289                 int status;
 290                 snprintf(cmd, PATH_MAX, "%s --inplace %s %s", rsync, info->src,
 291                         info->dest);
 292                 lr_debug(DTRACE, "\t%s %s\n", cmd, info->tfid);
 293                 status = system(cmd);
 294                 if (status == -1) {
 295                         rc = -errno;
 296                 } else if (WIFEXITED(status)) {
 297                         status = WEXITSTATUS(status);
 298                         if (!status)
 299                                 rc = 0;
 300                         else if (status == 23 || status == 24)
 301                                 /* Error due to vanished source files;
 302                                    Ignore this error*/
 303                                 rc = 0;
 304                         else
 305                                 rc = -EINVAL;
 306                         if (status)
 307                                 lr_debug(DINFO, "rsync %s exited with %d %d\n",
 308                                          info->src, status, rc);
 309                 } else {
 310                         rc = -EINTR;
 311                 }
 312         } else {
 313                 lr_debug(DTRACE, "Not syncing %s and %s %s\n", info->src,
 314                          info->dest, info->tfid);
 315         }
 316
 317         return rc;
 318 }
 319
 320 int lr_copy_data(struct lr_info *info)
 321 {
 322         int fd_src = -1;
 323         int fd_dest = -1;
 324         int bufsize;
 325         int rsize;
 326         int rc = 0;
 327         struct stat st_src;
 328         struct stat st_dest;
 329
 330         fd_src = open(info->src, O_RDONLY);
 331         if (fd_src == -1)
 332                 return -errno;
 333         if (fstat(fd_src, &st_src) == -1 ||
 334             stat(info->dest, &st_dest) == -1)
 335                 goto out;
 336
 337         if (st_src.st_mtime == st_dest.st_mtime &&
 338             st_src.st_size == st_dest.st_size)
 339                 goto out;
 340
 341         if (st_src.st_size > rsync_threshold && rsync[0] != '\0') {
 342                 /* It is more efficient to use rsync to replicate
 343                    large files. Any file larger than rsync_threshold
 344                    is handed off to rsync. */
 345                 lr_debug(DTRACE, "Using rsync to replicate %s\n", info->tfid);
 346                 rc = lr_rsync_data(info);
 347                 goto out;
 348         }
 349
 350         fd_dest = open(info->dest, O_WRONLY | O_TRUNC, st_src.st_mode);
 351         if (fd_dest == -1) {
 352                 rc = -errno;
 353                 goto out;
 354         }
 355         bufsize = st_dest.st_blksize;
 356
 357         if (info->bufsize < bufsize) {
 358                 /* Grow buffer */
 359                 info->buf = lr_grow_buf(info->buf, bufsize);
 360                 if (info->buf == NULL) {
 361                         rc = -ENOMEM;
 362                         goto out;
 363                 }
 364                 info->bufsize = bufsize;
 365         }
 366
 367         while (1) {
 368                 rsize = read(fd_src, info->buf, bufsize);
 369                 if (rsize == 0) {
 370                         break;
 371                 } else if (rsize < 0) {
 372                         rc = -errno;
 373                         goto out;
 374                 }
 375                 errno = 0;
 376                 if (write(fd_dest, info->buf, rsize) != rsize) {
 377                         if (errno != 0)
 378                                 rc = -errno;
 379                         else
 380                                 rc = -EINTR;
 381                 }
 382         }
 383         fsync(fd_dest);
 384
 385 out:
 386         if (fd_src != -1)
 387                 close(fd_src);
 388         if (fd_dest != -1)
 389                 close(fd_dest);
 390
 391         return rc;
 392 }
 393
 394 /* Copy data from source to destination */
 395 int lr_sync_data(struct lr_info *info)
 396 {
 397         if (use_rsync)
 398                 return lr_rsync_data(info);
 399         else
 400                 return lr_copy_data(info);
 401 }
 402
 403 /* Copy all attributes from file src to file dest */
 404 int lr_copy_attr(char *src, char *dest)
 405 {
 406         struct stat st;
 407         struct utimbuf time;
 408
 409         if (stat(src, &st) == -1 ||
 410             chmod(dest, st.st_mode) == -1 ||
 411             chown(dest, st.st_uid, st.st_gid) == -1)
 412                 return -errno;
 413
 414         time.actime = st.st_atime;
 415         time.modtime = st.st_mtime;
 416         if (utime(dest, &time) == -1)
 417                 return -errno;
 418         return 0;
 419 }
 420
 421 /* Copy all xattrs from file info->src to info->dest */
 422 int lr_copy_xattr(struct lr_info *info)
 423 {
 424         size_t size = info->xsize;
 425         int start;
 426         int len;
 427         int rc;
 428
 429         if (noxattr)
 430                 return 0;
 431
 432         errno = 0;
 433         rc = llistxattr(info->src, info->xlist, size);
 434         lr_debug(DTRACE, "llistxattr(%s,%p) returned %d, errno=%d\n",
 435                  info->src, info->xlist, rc, errno);
 436         if ((rc > 0 && info->xlist == NULL) || errno == ERANGE) {
 437                 size = rc > PATH_MAX ? rc : PATH_MAX;
 438                 info->xlist = lr_grow_buf(info->xlist, size);
 439                 if (info->xlist == NULL)
 440                         return -ENOMEM;
 441                 info->xsize = size;
 442                 rc = llistxattr(info->src, info->xlist, size);
 443                 lr_debug(DTRACE, "llistxattr %s returned %d, errno=%d\n",
 444                          info->src, rc, errno);
 445         }
 446         if (rc < 0)
 447                 return rc;
 448
 449         len = rc;
 450         start = 0;
 451         while (start < len) {
 452                 size = info->xvsize;
 453                 rc = lgetxattr(info->src, info->xlist + start,
 454                                info->xvalue, size);
 455                 if (info->xvalue == NULL || errno == ERANGE) {
 456                         size = rc > PATH_MAX ? rc : PATH_MAX;
 457                         info->xvalue = lr_grow_buf(info->xvalue, size);
 458                         if (info->xvalue == NULL)
 459                                 return -ENOMEM;
 460                         info->xvsize = size;
 461                         rc = lgetxattr(info->src, info->xlist + start,
 462                                        info->xvalue, size);
 463                 }
 464                 lr_debug(DTRACE, "\t(%s,%d) rc=%p\n", info->xlist + start,
 465                          info->xvalue, rc);
 466                 if (rc > 0) {
 467                         size = rc;
 468                         rc = lsetxattr(info->dest, info->xlist + start,
 469                                        info->xvalue, size, 0);
 470                         lr_debug(DTRACE, "\tlsetxattr(), rc=%d, errno=%d\n",
 471                                  rc, errno);
 472                         if (rc == -1) {
 473                                 if (errno != ENOTSUP) {
 474                                         fprintf(stderr, "Error replicating "
 475                                                 " xattr for %s: %d\n",
 476                                                 info->dest, errno);
 477                                         errors++;
 478                                 }
 479                                 rc = 0;
 480                         }
 481                 }
 482                 start += strlen(info->xlist + start) + 1;
 483         }
 484
 485         lr_debug(DINFO, "setxattr: %s %s\n", info->src, info->dest);
 486
 487         return rc;
 488 }
 489
 490 /* Retrieve the filesystem path for a given FID and a given
 491    linkno. The path is returned in info->path */
 492 int lr_get_path_ln(struct lr_info *info, char *fidstr, int linkno)
 493 {
 494         long long recno = -1;
 495         int rc;
 496
 497         rc = llapi_fid2path(status->ls_source, fidstr, info->path,
 498                             PATH_MAX, &recno, &linkno);
 499         if (rc < 0 && rc != -ENOENT) {
 500                 fprintf(stderr, "fid2path error: (%s, %s) %d %s\n",
 501                         status->ls_source, fidstr, -rc, strerror(errno = -rc));
 502         }
 503
 504         return rc;
 505 }
 506
 507 /* Retrieve the filesystem path for a given FID. The path is returned
 508    in info->path */
 509 int lr_get_path(struct lr_info *info, char *fidstr)
 510 {
 511         return lr_get_path_ln(info, fidstr, 0);
 512 }
 513
 514 /* Generate the path for opening by FID */
 515 void lr_get_FID_PATH(char *mntpt, char *fidstr, char *buf, int bufsize)
 516 {
 517         /* Open-by-FID path is <mntpt>/.lustre/fid/[SEQ:OID:VER] */
 518         snprintf(buf, bufsize, "%s/%s/fid/%s", mntpt, dot_lustre_name,
 519                  fidstr);
 520         return;
 521 }
 522
 523 /* Read the symlink information into 'info->link' */
 524 int lr_get_symlink(struct lr_info *info)
 525 {
 526         int rc;
 527         char *link;
 528
 529         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
 530         rc = readlink(info->src, info->linktmp, PATH_MAX);
 531         if (rc > 0)
 532                 info->linktmp[rc] = '\0';
 533         else
 534                 return rc;
 535         lr_debug(DTRACE, "symlink: readlink returned %s\n", info->linktmp);
 536
 537         if (strncmp(info->linktmp, status->ls_source,
 538                     strlen(status->ls_source)) == 0) {
 539                 /* Strip source fs path and replace with target fs path. */
 540                 link = info->linktmp + strlen(status->ls_source);
 541                 snprintf(info->src, PATH_MAX, "%s%s",
 542                          status->ls_targets[info->target_no], link);
 543                 link = info->src;
 544         } else {
 545                 link = info->linktmp;
 546         }
 547         strncpy(info->link, link, PATH_MAX);
 548         info->link[PATH_MAX] = '\0';
 549
 550         return rc;
 551 }
 552
 553 /* Create file/directory/device file/symlink. */
 554 int lr_mkfile(struct lr_info *info)
 555 {
 556         struct stat st;
 557         int rc = 0;
 558
 559         errno = 0;
 560         lr_debug(DINFO, "mkfile(%d) %s \n", info->type, info->dest);
 561         if (info->type == CL_MKDIR) {
 562                 rc = mkdir(info->dest, 0777);
 563         } else if (info->type == CL_SOFTLINK) {
 564                 lr_get_symlink(info);
 565                 rc = symlink(info->link, info->dest);
 566         } else if (info->type == CL_MKNOD) {
 567                 lr_get_FID_PATH(status->ls_source, info->tfid,
 568                                     info->src, PATH_MAX);
 569                 rc = stat(info->src, &st);
 570                 if (rc == -1) {
 571                         if (errno == ENOENT)
 572                                 return 0;
 573                         else
 574                                 return -errno;
 575                 }
 576                 rc = mknod(info->dest, st.st_mode, st.st_rdev);
 577         } else {
 578                 rc = mknod(info->dest, S_IFREG | 0777, 0);
 579         }
 580         if (rc)
 581                 return -errno;
 582
 583         /* Sync data and attributes */
 584         if (info->type == CL_CREATE || info->type == CL_MKDIR) {
 585                 lr_debug(DTRACE, "Syncing data and attributes %s\n",
 586                          info->tfid);
 587                 (void) lr_copy_xattr(info);
 588                 if (info->type == CL_CREATE)
 589                         rc = lr_sync_data(info);
 590                 if (!rc)
 591                         rc = lr_copy_attr(info->src, info->dest);
 592
 593                 if (rc == -ENOENT)
 594                         /* Source file has disappeared. Not an error. */
 595                         rc = 0;
 596         } else {
 597                 lr_debug(DTRACE, "Not syncing data and attributes %s\n",
 598                          info->tfid);
 599         }
 600
 601         return rc;
 602 }
 603
 604 int lr_add_pc(const char *pfid, const char *tfid, const char *name)
 605 {
 606         struct lr_parent_child_list *p;
 607
 608         p = calloc(1, sizeof(*p));
 609         if (!p)
 610                 return -ENOMEM;
 611         strcpy(p->pc_log.pcl_pfid, pfid);
 612         strcpy(p->pc_log.pcl_tfid, tfid);
 613         strcpy(p->pc_log.pcl_name, name);
 614
 615         p->pc_next = parents;
 616         parents = p;
 617         return 0;
 618 }
 619
 620 void lr_cascade_move(const char *fid, const char *dest, struct lr_info *info)
 621 {
 622         struct lr_parent_child_list *curr, *prev;
 623         char *d;
 624         int rc;
 625
 626         d = calloc(1, PATH_MAX + 1);
 627         prev = curr = parents;
 628         while (curr) {
 629                 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
 630                         snprintf(d, PATH_MAX, "%s/%s", dest,
 631                                  curr->pc_log.pcl_name);
 632                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 633                                 status->ls_targets[info->target_no],
 634                                 SPECIAL_DIR, curr->pc_log.pcl_tfid);
 635                         rc = rename(info->src, d);
 636                         if (rc == -1) {
 637                                 fprintf(stderr, "Error renaming file "
 638                                         " %s to %s: %d\n",
 639                                         info->src, d, errno);
 640                                 errors++;
 641                         }
 642                         lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
 643                         if (curr == parents)
 644                                 parents = curr->pc_next;
 645                         else
 646                                 prev->pc_next = curr->pc_next;
 647                         free(curr);
 648                         prev = curr = parents;
 649
 650                 } else {
 651                         prev = curr;
 652                         curr = curr->pc_next;
 653                 }
 654         }
 655
 656         free(d);
 657 }
 658
 659 /* remove [info->pfid, ext->tfid] from parents */
 660 int lr_remove_pc(const char *pfid, const char *tfid)
 661 {
 662         struct lr_parent_child_list *curr, *prev;
 663
 664         for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
 665                 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
 666                     strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
 667                         if (curr == parents)
 668                                 parents = curr->pc_next;
 669                         else
 670                                 prev->pc_next = curr->pc_next;
 671                         free(curr);
 672                         break;
 673                 }
 674         }
 675         return 0;
 676 }
 677
 678 /* Create file under SPECIAL_DIR with its tfid as its name. */
 679 int lr_mk_special(struct lr_info *info)
 680 {
 681         int rc;
 682
 683         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 684                 status->ls_targets[info->target_no], SPECIAL_DIR,
 685                 info->tfid);
 686
 687         rc = lr_mkfile(info);
 688         if (rc)
 689                 return rc;
 690
 691         rc = lr_add_pc(info->pfid, info->tfid, info->name);
 692         return rc;
 693 }
 694
 695 /* Remove a file or directory */
 696 int lr_rmfile(struct lr_info *info)
 697 {
 698         int rc;
 699
 700         if (info->type == CL_RMDIR)
 701                 rc = rmdir(info->dest);
 702         else
 703                 rc = unlink(info->dest);
 704         if (rc == -1)
 705                 rc = -errno;
 706         return rc;
 707 }
 708
 709 /* Recursively remove directory and its contents */
 710 int lr_rm_recursive(struct lr_info *info)
 711 {
 712         int rc;
 713
 714         snprintf(info->cmd, PATH_MAX, "rm -rf %s", info->dest);
 715         rc = system(info->cmd);
 716         if (rc == -1)
 717                 rc = -errno;
 718
 719         return rc;
 720 }
 721
 722 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
 723 int lr_rm_special(struct lr_info *info)
 724 {
 725         int rc;
 726
 727         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 728                  status->ls_targets[info->target_no], SPECIAL_DIR,
 729                  info->tfid);
 730         rc = lr_rmfile(info);
 731
 732         if (rc)
 733                 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
 734                          info->dest, rc, errno);
 735         return rc;
 736 }
 737
 738 /* Replicate file and directory create events */
 739 int lr_create(struct lr_info *info)
 740 {
 741         int len;
 742         int rc1 = 0;
 743         int rc;
 744         int mkspecial = 0;
 745
 746         /* Is target FID present on the source? */
 747         rc = lr_get_path(info, info->tfid);
 748         if (rc == -ENOENT) {
 749                 /* Source file has disappeared. Not an error. */
 750                 lr_debug(DINFO, "create: tfid %s not found on"
 751                          "source-fs\n", info->tfid);
 752                 return 0;
 753         } else if (rc) {
 754                 return rc;
 755         }
 756         strcpy(info->savedpath, info->path);
 757
 758         /* Is parent FID present on the source */
 759         rc = lr_get_path(info, info->pfid);
 760         if (rc == -ENOENT) {
 761                 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
 762                          info->tfid);
 763                 mkspecial = 1;
 764         } else if (rc < 0) {
 765                 return rc;
 766         }
 767
 768         /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
 769         len = strlen(info->path);
 770         if (len - 1 >= 0 && info->path[len - 1] == '/')
 771                 snprintf(info->dest, PATH_MAX, "%s%s", info->path, info->name);
 772         else
 773                 snprintf(info->dest, PATH_MAX, "%s/%s", info->path, info->name);
 774
 775         lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
 776                  info->savedpath);
 777         if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
 778                 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
 779                          info->tfid, info->dest, info->savedpath);
 780                 mkspecial = 1;
 781         }
 782
 783         /* Is f2p(pfid) present on the target? If not, the parent has
 784            moved */
 785         if (!mkspecial) {
 786                 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[0],
 787                         info->path);
 788                 if (access(info->dest, F_OK) != 0)
 789                         mkspecial = 1;
 790         }
 791         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 792              info->target_no++) {
 793                 snprintf(info->dest, PATH_MAX, "%s/%s",
 794                         status->ls_targets[info->target_no], info->savedpath);
 795                 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
 796                                     PATH_MAX);
 797
 798                 if (!mkspecial)
 799                         rc1 = lr_mkfile(info);
 800                 if (mkspecial || rc1 == -ENOENT) {
 801                         rc1 = lr_mk_special(info);
 802                 }
 803                 if (rc1)
 804                         rc = rc1;
 805         }
 806         return rc;
 807 }
 808
 809 /* Replicate a file remove (rmdir/unlink) operation */
 810 int lr_remove(struct lr_info *info)
 811 {
 812         int rc = 0;
 813         int rc1;
 814
 815         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 816              info->target_no++) {
 817
 818                 rc1 = lr_rm_special(info);
 819                 if (!rc1)
 820                         continue;
 821
 822                 rc1 = lr_get_path(info, info->pfid);
 823                 if (rc1 == -ENOENT) {
 824                         lr_debug(DINFO, "remove: pfid %s not found\n",
 825                                  info->pfid);
 826                         continue;
 827                 }
 828                 if (rc1) {
 829                         rc = rc1;
 830                         continue;
 831                 }
 832                 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 833                         status->ls_targets[info->target_no], info->path,
 834                         info->name);
 835
 836                 rc1 = lr_rmfile(info);
 837                 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
 838                          info->dest, rc1, errno);
 839                 if (rc1 == -ENOTEMPTY)
 840                         rc1 = lr_rm_recursive(info);
 841
 842                 if (rc1) {
 843                         rc = rc1;
 844                         continue;
 845                 }
 846         }
 847         return rc;
 848 }
 849
 850 /* Replicate a rename/move operation. This operations are tracked by
 851    two changelog records. */
 852 int lr_move(struct lr_info *info, struct lr_info *ext)
 853 {
 854         int rc = 0;
 855         int rc1;
 856         int rc_dest, rc_src;
 857         int special_src = 0;
 858         int special_dest = 0;
 859
 860         rc_dest = lr_get_path(ext, ext->pfid);
 861         if (rc_dest < 0 && rc_dest != -ENOENT)
 862                 return rc_dest;
 863
 864         rc_src = lr_get_path(info, info->pfid);
 865         if (rc_src < 0 && rc_src != -ENOENT)
 866                 return rc_src;
 867
 868         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 869              info->target_no++) {
 870
 871                 if (!rc_dest) {
 872                         snprintf(info->dest, PATH_MAX, "%s/%s",
 873                                 status->ls_targets[info->target_no],
 874                                 ext->path);
 875                         if (access(info->dest, F_OK) != 0) {
 876                                 rc_dest = -errno;
 877                         } else {
 878                                 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 879                                         status->ls_targets[info->target_no],
 880                                         ext->path, ext->name);
 881                         }
 882                 }
 883                 if (rc_dest == -ENOENT) {
 884                         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 885                                 status->ls_targets[info->target_no],
 886                                 SPECIAL_DIR, info->tfid);
 887                         special_dest = 1;
 888                 }
 889
 890                 if (!rc_src)
 891                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 892                                 status->ls_targets[info->target_no],
 893                                 info->path, info->name);
 894                 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
 895                                           errno == ENOENT)) {
 896                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 897                                 status->ls_targets[info->target_no],
 898                                 SPECIAL_DIR, info->tfid);
 899                         special_src = 1;
 900                 }
 901
 902                 rc1 = 0;
 903                 if (strcmp(info->src, info->dest) != 0) {
 904                         rc1 = rename(info->src, info->dest);
 905                         if (rc1 == -1)
 906                                 rc1 = -errno;
 907                 }
 908
 909                 if (special_src) {
 910                         lr_remove_pc(info->pfid, info->tfid);
 911                         if (!special_dest)
 912                                 lr_cascade_move(info->tfid, info->dest, info);
 913                 }
 914                 if (special_dest)
 915                         lr_add_pc(ext->pfid, info->tfid, ext->name);
 916
 917                 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
 918                          info->src, info->dest, rc1, errno);
 919                 if (rc1)
 920                         rc = rc1;
 921         }
 922         return rc;
 923 }
 924
 925 /* Replicate a hard link */
 926 int lr_link(struct lr_info *info)
 927 {
 928         int i;
 929         int len;
 930         int rc;
 931         int rc1;
 932         struct stat st;
 933
 934         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
 935         rc = stat(info->src, &st);
 936         if (rc == -1)
 937                 return -errno;
 938
 939         for (info->target_no = 0; info->target_no < status->ls_num_targets;
 940              info->target_no++) {
 941
 942                 info->src[0] = 0;
 943                 info->dest[0] = 0;
 944                 rc1 = 0;
 945
 946                 /* Search through the hardlinks to get the src and dest */
 947                 for (i = 0; i < st.st_nlink && (info->src[0] == 0 ||
 948                                                 info->dest[0] == 0); i++) {
 949                         rc1 = lr_get_path_ln(info, info->tfid, i);
 950                         lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
 951                                  info->path, info->name, i, rc1);
 952                         if (rc1)
 953                                 break;
 954
 955                         len = strlen(info->path) - strlen(info->name);
 956                         if (len > 0 && strcmp(info->path + len,
 957                                               info->name) == 0)
 958                                 snprintf(info->dest, PATH_MAX, "%s/%s",
 959                                         status->ls_targets[info->target_no],
 960                                         info->path);
 961                         else if (info->src[0] == 0)
 962                                 snprintf(info->src, PATH_MAX, "%s/%s",
 963                                         status->ls_targets[info->target_no],
 964                                         info->path);
 965                 }
 966
 967                 if (rc1) {
 968                         rc = rc1;
 969                         continue;
 970                 }
 971
 972                 if (info->src[0] == 0 || info->dest[0] == 0)
 973                         /* Could not find the source or destination.
 974                          This can happen when some links don't exist
 975                          anymore. */
 976                         return -EINVAL;
 977
 978                 if (info->src[0] == 0)
 979                         snprintf(info->src, PATH_MAX, "%s/%s/%s",
 980                                 status->ls_targets[info->target_no],
 981                                 SPECIAL_DIR, info->tfid);
 982                 else if (info->dest[0] == 0)
 983                         snprintf(info->dest, PATH_MAX, "%s/%s/%s",
 984                                 status->ls_targets[info->target_no],
 985                                 SPECIAL_DIR, info->tfid);
 986
 987                 rc1 = link(info->src, info->dest);
 988                 lr_debug(rc1?0:DINFO, "link: %s [to] %s; rc1=%d %s\n",
 989                          info->src, info->dest, rc1, strerror(errno));
 990
 991                 if (rc1)
 992                         rc = rc1;
 993         }
 994         return rc;
 995 }
 996
 997 /* Replicate file attributes */
 998 int lr_setattr(struct lr_info *info)
 999 {
1000         int rc1;
1001         int rc;
1002
1003         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1004
1005         rc = lr_get_path(info, info->tfid);
1006         if (rc == -ENOENT)
1007                 lr_debug(DINFO, "setattr: %s not present on source-fs\n",
1008                          info->src);
1009         if (rc)
1010                 return rc;
1011
1012         for (info->target_no = 0; info->target_no < status->ls_num_targets;
1013              info->target_no++) {
1014
1015                 snprintf(info->dest, PATH_MAX, "%s/%s",
1016                          status->ls_targets[info->target_no], info->path);
1017                 lr_debug(DINFO, "setattr: %s %s %s", info->src, info->dest,
1018                          info->tfid);
1019
1020                 rc1 = lr_sync_data(info);
1021                 if (!rc1)
1022                         rc1 = lr_copy_attr(info->src, info->dest);
1023                 if (rc1)
1024                         rc = rc1;
1025         }
1026         return rc;
1027 }
1028
1029 /* Replicate xattrs */
1030 int lr_setxattr(struct lr_info *info)
1031 {
1032         int rc, rc1;
1033
1034         lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1035
1036         rc = lr_get_path(info, info->tfid);
1037         if (rc == -ENOENT)
1038                 lr_debug(DINFO, "setxattr: %s not present on source-fs\n",
1039                          info->src);
1040         if (rc)
1041                 return rc;
1042
1043         for (info->target_no = 0; info->target_no < status->ls_num_targets;
1044              info->target_no++) {
1045
1046                 snprintf(info->dest, PATH_MAX, "%s/%s",
1047                         status->ls_targets[info->target_no], info->path);
1048                 lr_debug(DINFO, "setxattr: %s %s %s\n", info->src, info->dest,
1049                          info->tfid);
1050
1051                 rc1 = lr_copy_xattr(info);
1052                 if (rc1)
1053                         rc = rc1;
1054         }
1055
1056         return rc;
1057 }
1058
1059 /* Parse a line of changelog entry */
1060 int lr_parse_line(void *priv, struct lr_info *info)
1061 {
1062         struct changelog_rec *rec;
1063
1064         if (llapi_changelog_recv(priv, &rec) != 0)
1065                 return -1;
1066
1067         info->recno = rec->cr_index;
1068         info->type = rec->cr_type;
1069         sprintf(info->tfid, DFID, PFID(&rec->cr_tfid));
1070         sprintf(info->pfid, DFID, PFID(&rec->cr_pfid));
1071         strncpy(info->name, rec->cr_name, rec->cr_namelen);
1072         info->name[rec->cr_namelen] = '\0';
1073
1074         if (verbose > 1)
1075                 printf("Rec %lld: %d %s\n", info->recno, info->type,info->name);
1076
1077         llapi_changelog_free(&rec);
1078
1079         rec_count++;
1080         return 0;
1081 }
1082
1083 /* Initialize the replication parameters */
1084 int lr_init_status()
1085 {
1086         size_t size = sizeof(struct lustre_rsync_status) + PATH_MAX;
1087
1088         if (status != NULL)
1089                 return 0;
1090         status = calloc(size, 1);
1091         if (status == NULL)
1092                 return -ENOMEM;
1093         status->ls_version = REPLICATE_STATUS_VER;
1094         status->ls_size = size;
1095         status->ls_last_recno = -1;
1096         return 0;
1097 }
1098
1099 /* Make a backup of the statuslog */
1100 void lr_backup_log()
1101 {
1102         char backupfile[PATH_MAX];
1103
1104         if (logbackedup)
1105                 return;
1106         snprintf(backupfile, PATH_MAX, "%s.old", statuslog);
1107         (void) rename(statuslog, backupfile);
1108         logbackedup = 1;
1109
1110         return;
1111 }
1112
1113 /* Save replication parameters to a statuslog. */
1114 int lr_write_log()
1115 {
1116         int fd;
1117         size_t size;
1118         size_t write_size = status->ls_size;
1119         struct lr_parent_child_list *curr;
1120         int rc = 0;
1121
1122         if (statuslog == NULL)
1123                 return 0;
1124
1125         lr_backup_log();
1126
1127         fd = open(statuslog, O_WRONLY | O_CREAT | O_SYNC,
1128                              S_IRUSR | S_IWUSR);
1129         if (fd == -1) {
1130                 fprintf(stderr, "Error opening log file for writing (%s)\n",
1131                         statuslog);
1132                 return -1;
1133         }
1134         errno = 0;
1135         size = write(fd, status, write_size);
1136         if (size != write_size) {
1137                 fprintf(stderr, "Error writing to log file (%s) %d\n",
1138                         statuslog, errno);
1139                 close(fd);
1140                 return -1;
1141         }
1142
1143         for (curr = parents; curr; curr = curr->pc_next) {
1144                 size = write(fd, &curr->pc_log, sizeof(curr->pc_log));
1145                 if (size != sizeof(curr->pc_log)) {
1146                         fprintf(stderr, "Error writing to log file (%s) %d\n",
1147                                 statuslog, errno);
1148                         rc = -1;
1149                         break;
1150                 }
1151         }
1152         close(fd);
1153         return rc;
1154 }
1155
1156 /* Read statuslog and populate the replication parameters.  Command
1157  * line parameters take precedence over parameters in the log file.*/
1158 int lr_read_log()
1159 {
1160         struct lr_parent_child_list *tmp;
1161         struct lr_parent_child_log rec;
1162         struct lustre_rsync_status *s;
1163         int fd = -1;
1164         size_t size;
1165         size_t read_size = sizeof(struct lustre_rsync_status) + PATH_MAX;
1166         int rc = 0;
1167
1168         if (statuslog == NULL)
1169                 return 0;
1170
1171         s = calloc(1, read_size);
1172         if (s == NULL)
1173                 GOTO(out, rc = -ENOMEM);
1174
1175         fd = open(statuslog, O_RDONLY);
1176         if (fd == -1)
1177                 GOTO(out, rc = -errno);
1178         size = read(fd, s, read_size);
1179         if (size != read_size)
1180                 GOTO(out, rc = -EINVAL);
1181         if (read_size < s->ls_size) {
1182                 read_size = s->ls_size;
1183                 s = lr_grow_buf(s, read_size);
1184                 if (s == NULL)
1185                         GOTO(out, rc = -ENOMEM);
1186                 if (lseek(fd, 0, SEEK_SET) == -1)
1187                         GOTO(out, rc = -errno);
1188                 size = read(fd, s, read_size);
1189                 if (size != read_size)
1190                         GOTO(out, rc = -EINVAL);
1191         }
1192
1193         while (read(fd, &rec, sizeof(rec)) != 0) {
1194                 tmp = calloc(1, sizeof(*tmp));
1195                 if (!tmp)
1196                         GOTO(out, rc = -ENOMEM);
1197                 tmp->pc_log = rec;
1198                 tmp->pc_next = parents;
1199                 parents = tmp;
1200         }
1201
1202         /* copy uninitialized fields to status */
1203         if (status->ls_num_targets == 0) {
1204                 if (status->ls_size != s->ls_size) {
1205                         status = lr_grow_buf(status, s->ls_size);
1206                         if (status == NULL)
1207                                 GOTO(out, rc = -ENOMEM);
1208                         status->ls_size = s->ls_size;
1209                 }
1210                 status->ls_num_targets = s->ls_num_targets;
1211                 memcpy(status->ls_targets, s->ls_targets,
1212                        PATH_MAX * s->ls_num_targets);
1213         }
1214         if (status->ls_last_recno == -1)
1215                 status->ls_last_recno = s->ls_last_recno;
1216
1217         if (status->ls_registration[0] == '\0')
1218                 strncpy(status->ls_registration, s->ls_registration,
1219                         LR_NAME_MAXLEN);
1220
1221         if (status->ls_mdt_device[0] == '\0')
1222                 strncpy(status->ls_mdt_device, s->ls_mdt_device,
1223                         LR_NAME_MAXLEN);
1224
1225         if (status->ls_source_fs[0] == '\0')
1226                 strncpy(status->ls_source_fs, s->ls_source_fs,
1227                         LR_NAME_MAXLEN);
1228
1229         if (status->ls_source[0] == '\0')
1230                 strncpy(status->ls_source, s->ls_source, PATH_MAX);
1231
1232  out:
1233         if (fd != -1)
1234                 close(fd);
1235         if (s)
1236                 free(s);
1237         return rc;
1238 }
1239
1240 /* Clear changelogs every CLEAR_INTERVAL records or at the end of
1241    processing. */
1242 int lr_clear_cl(struct lr_info *info, int force)
1243 {
1244         char    mdt_device[LR_NAME_MAXLEN + 1];
1245         long long rec;
1246         int rc = 0;
1247
1248         if (force || info->recno > status->ls_last_recno + CLEAR_INTERVAL) {
1249                 if (info->type == CL_RENAME)
1250                         rec = info->recno + 1;
1251                 else
1252                         rec = info->recno;
1253                 if (!noclear && !dryrun) {
1254                         /* llapi_changelog_clear modifies the mdt
1255                          * device name so make a copy of it until this
1256                          * is fixed.
1257                         */
1258                         strncpy(mdt_device, status->ls_mdt_device,
1259                                 LR_NAME_MAXLEN);
1260                         rc = llapi_changelog_clear(mdt_device,
1261                                                    status->ls_registration,
1262                                                    rec);
1263                         if (rc)
1264                                 printf("Changelog clear (%s, %s, %lld) "
1265                                        "returned %d\n", status->ls_mdt_device,
1266                                        status->ls_registration, rec, rc);
1267                 }
1268                 if (!rc && !dryrun) {
1269                         status->ls_last_recno = rec;
1270                         lr_write_log();
1271
1272                 }
1273         }
1274
1275         return rc;
1276 }
1277
1278 /* Locate a usable version of rsync. At this point we'll use any
1279    version. */
1280 int lr_locate_rsync()
1281 {
1282         FILE *fp;
1283         int len;
1284
1285         /* Locate rsync */
1286         snprintf(rsync, PATH_MAX, "%s -p %s", TYPE, RSYNC);
1287         fp = popen(rsync, "r");
1288         if (fp == NULL)
1289                 return -1;
1290
1291         if (fgets(rsync, PATH_MAX, fp) == NULL) {
1292                 fclose(fp);
1293                 return -1;
1294         }
1295
1296         len = strlen(rsync);
1297         if (len > 0 && rsync[len - 1] == '\n')
1298                 rsync[len - 1] = '\0';
1299         fclose(fp);
1300
1301         /* Determine the version of rsync */
1302         snprintf(rsync_ver, PATH_MAX, "%s --version", rsync);
1303         fp = popen(rsync_ver, "r");
1304         if (fp == NULL)
1305                 return -1;
1306
1307         if (fgets(rsync_ver, PATH_MAX, fp) == NULL) {
1308                 fclose(fp);
1309                 return -1;
1310         }
1311         len = strlen(rsync_ver);
1312         if (len > 0 && rsync_ver[len - 1] == '\n')
1313                 rsync_ver[len - 1] = '\0';
1314         fclose(fp);
1315
1316         return 0;
1317
1318 }
1319
1320 /* Print the replication parameters */
1321 void lr_print_status(struct lr_info *info)
1322 {
1323         int i;
1324
1325         if (!verbose)
1326                 return;
1327
1328         printf("Lustre filesystem: %s\n", status->ls_source_fs);
1329         printf("MDT device: %s\n", status->ls_mdt_device);
1330         printf("Source: %s\n", status->ls_source);
1331         for (i = 0; i < status->ls_num_targets; i++)
1332                 printf("Target: %s\n", status->ls_targets[i]);
1333         if (statuslog != NULL)
1334                 printf("Statuslog: %s\n", statuslog);
1335         printf("Changelog registration: %s\n", status->ls_registration);
1336         printf("Starting changelog record: "LPD64"\n", status->ls_last_recno);
1337         if (noxattr)
1338                 printf("Replicate xattrs: no\n");
1339         if (noclear)
1340                 printf("Clear changelog after use: no\n");
1341         if (use_rsync)
1342                 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
1343 }
1344
1345 void lr_print_failure(struct lr_info *info, int rc)
1346 {
1347         fprintf(stderr, "Replication of operation failed(%d):"
1348                 " %lld %s (%d) %s %s %s\n", rc, info->recno,
1349                 changelog_type2str(info->type), info->type, info->tfid,
1350                 info->pfid, info->name);
1351 }
1352
1353 /* Replicate filesystem operations from src_path to target_path */
1354 int lr_replicate()
1355 {
1356         void *changelog_priv;
1357         struct lr_info *info;
1358         struct lr_info *ext;
1359         time_t start;
1360         int xattr_not_supp;
1361         int i;
1362         int rc;
1363
1364         start = time(NULL);
1365
1366         info = calloc(1, sizeof(struct lr_info));
1367         if (info == NULL)
1368                 return -ENOMEM;
1369
1370         rc = llapi_search_fsname(status->ls_source, status->ls_source_fs);
1371         if (rc) {
1372                 fprintf(stderr, "Source path is not a valid Lustre client "
1373                         "mountpoint.\n");
1374                 return rc;
1375         }
1376         if (status->ls_mdt_device[0] == '\0')
1377                 snprintf(status->ls_mdt_device, LR_NAME_MAXLEN, "%s%s",
1378                         status->ls_source_fs, DEFAULT_MDT);
1379
1380         ext = calloc(1, sizeof(struct lr_info));
1381         if (ext == NULL)
1382                 return -ENOMEM;
1383         memcpy(ext, info, sizeof(struct lr_info));
1384
1385         for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1386                 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[i],
1387                         SPECIAL_DIR);
1388                 rc = mkdir(info->dest, 0777);
1389                 if (rc == -1 && errno != EEXIST) {
1390                         fprintf(stderr, "Error writing to target path %s.\n",
1391                                 status->ls_targets[i]);
1392                         return -errno;
1393                 }
1394                 rc = llistxattr(info->src, info->xlist, info->xsize);
1395                 if (rc == -1 && errno == ENOTSUP) {
1396                         fprintf(stderr, "xattrs not supported on %s\n",
1397                                 status->ls_targets[i]);
1398                         xattr_not_supp++;
1399                 }
1400         }
1401         if (xattr_not_supp == status->ls_num_targets)
1402                 /* None of the targets support xattrs. */
1403                 noxattr = 1;
1404
1405         lr_print_status(info);
1406
1407         /* Open changelogs for consumption*/
1408         rc = llapi_changelog_start(&changelog_priv, CHANGELOG_FLAG_BLOCK,
1409                                    status->ls_source_fs, status->ls_last_recno);
1410         if (rc < 0) {
1411                 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1412                         status->ls_source_fs);
1413                 return rc;
1414         }
1415
1416         while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1417                 rc = 0;
1418                 if (info->type == CL_RENAME)
1419                         /* Rename operations have an additional changelog
1420                            record of information. */
1421                         lr_parse_line(changelog_priv, ext);
1422
1423                 if (dryrun)
1424                         continue;
1425
1426                 switch(info->type) {
1427                 case CL_CREATE:
1428                 case CL_MKDIR:
1429                 case CL_MKNOD:
1430                 case CL_SOFTLINK:
1431                         rc = lr_create(info);
1432                         break;
1433                 case CL_RMDIR:
1434                 case CL_UNLINK:
1435                         rc = lr_remove(info);
1436                         break;
1437                 case CL_RENAME:
1438                         rc = lr_move(info, ext);
1439                         break;
1440                 case CL_HARDLINK:
1441                         rc = lr_link(info);
1442                         break;
1443                 case CL_TRUNC:
1444                 case CL_SETATTR:
1445                         rc = lr_setattr(info);
1446                         break;
1447                 case CL_XATTR:
1448                         rc = lr_setxattr(info);
1449                         break;
1450                 case CL_CLOSE:
1451                 case CL_EXT:
1452                 case CL_OPEN:
1453                 case CL_IOCTL:
1454                 case CL_MARK:
1455                         /* Nothing needs to be done for these entries */
1456                 default:
1457                         break;
1458                 }
1459                 if (rc && rc != -ENOENT) {
1460                         lr_print_failure(info, rc);
1461                         errors++;
1462                         if (abort_on_err)
1463                                 break;
1464                 }
1465                 lr_clear_cl(info, 0);
1466                 if (debug) {
1467                         bzero(info, sizeof(struct lr_info));
1468                         bzero(ext, sizeof(struct lr_info));
1469                 }
1470         }
1471
1472         llapi_changelog_fini(&changelog_priv);
1473
1474         if (errors || verbose)
1475                 printf("Errors: %d\n", errors);
1476
1477         /* Clear changelog records used so far */
1478         lr_clear_cl(info, 1);
1479
1480         if (verbose) {
1481                 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1482                 printf("Changelog records consumed: %lld\n", rec_count);
1483         }
1484
1485         return 0;
1486 }
1487
1488 void
1489 termination_handler (int signum)
1490 {
1491         /* Set a flag for the replicator to gracefully shutdown */
1492         quit = 1;
1493         printf("lustre_rsync halting.\n");
1494 }
1495
1496 int main(int argc, char *argv[])
1497 {
1498         int newsize;
1499         int numtargets = 0;
1500         int rc = 0;
1501
1502         if ((rc = lr_init_status()) != 0)
1503                 return rc;
1504
1505         while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:",
1506                                 long_opts, NULL)) >= 0) {
1507                 switch (rc) {
1508                 case 'a':
1509                         /* Assume absolute paths */
1510                         abort_on_err++;
1511                         break;
1512                 case 's':
1513                         /* Assume absolute paths */
1514                         strncpy(status->ls_source, optarg, PATH_MAX);
1515                         break;
1516                 case 't':
1517                         status->ls_num_targets++;
1518                         numtargets++;
1519                         if (numtargets != status->ls_num_targets) {
1520                                 /* Targets were read from a log
1521                                    file. The ones specified on the
1522                                    command line take precedence. The
1523                                    ones from the log file will be
1524                                    ignored. */
1525                                 status->ls_num_targets = numtargets;
1526                         }
1527                         newsize = sizeof (struct lustre_rsync_status) +
1528                                 (status->ls_num_targets * PATH_MAX);
1529                         if (status->ls_size != newsize) {
1530                                 status->ls_size = newsize;
1531                                 status = lr_grow_buf(status, newsize);
1532                                 if (status == NULL)
1533                                         return -ENOMEM;
1534                         }
1535                         strncpy(status->ls_targets[status->ls_num_targets - 1],
1536                                 optarg,
1537                                 PATH_MAX);
1538                         break;
1539                 case 'm':
1540                         strncpy(status->ls_mdt_device, optarg, LR_NAME_MAXLEN);
1541                         break;
1542                 case 'u':
1543                         strncpy(status->ls_registration, optarg,
1544                                 LR_NAME_MAXLEN);
1545                         break;
1546                 case 'l':
1547                         statuslog = optarg;
1548                         (void) lr_read_log();
1549                         break;
1550                 case 'v':
1551                         verbose++;
1552                         break;
1553                 case 'x':
1554                         if (strcmp("no", optarg) == 0) {
1555                                 noxattr = 1;
1556                         } else if (strcmp("yes", optarg) != 0) {
1557                                 printf("Invalid parameter %s. "
1558                                        "Specify --xattr=no or --xattr=yes\n",
1559                                        optarg);
1560                                 return -1;
1561                         }
1562                         break;
1563                 case 'z':
1564                         dryrun = 1;
1565                         break;
1566                 case 'c':
1567                         /* Undocumented option cl-clear */
1568                         if (strcmp("no", optarg) == 0) {
1569                                 noclear = 1;
1570                         } else if (strcmp("yes", optarg) != 0) {
1571                                 printf("Invalid parameter %s. "
1572                                        "Specify --cl-clear=no "
1573                                        "or --cl-clear=yes\n",
1574                                        optarg);
1575                                 return -1;
1576                         }
1577                         break;
1578                 case 'r':
1579                         /* Undocumented option use-rsync */
1580                         use_rsync = 1;
1581                         break;
1582                 case 'y':
1583                         /* Undocumented option rsync-threshold */
1584                         rsync_threshold = atol(optarg);
1585                         break;
1586                 case 'n':
1587                         /* Undocumented option start-recno */
1588                         status->ls_last_recno = atol(optarg);
1589                         break;
1590                 case 'd':
1591                         /* Undocumented option debug */
1592                         debug = atoi(optarg);
1593                         if (debug < 0 || debug > 2)
1594                                 debug = 0;
1595                         break;
1596                 default:
1597                         fprintf(stderr, "error: %s: option '%s' "
1598                                 "unrecognized.\n", argv[0], argv[optind - 1]);
1599                         lr_usage();
1600                         return -1;
1601                 }
1602         }
1603
1604         if (status->ls_last_recno == -1)
1605                 status->ls_last_recno = 0;
1606         if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1607                 /* No registration ID was passed in. */
1608                 printf("Please specify changelog consumer registration id.\n");
1609                 lr_usage();
1610                 return -1;
1611         }
1612         if (strnlen(status->ls_source, PATH_MAX) == 0) {
1613                 fprintf(stderr, "Please specify the source path.\n");
1614                 lr_usage();
1615                 return -1;
1616         }
1617         if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1618                 fprintf(stderr, "Please specify the target path.\n");
1619                 lr_usage();
1620                 return -1;
1621         }
1622
1623         /* This plumbing is needed for some of the ioctls behind
1624            llapi calls to work. */
1625         if (obd_initialize(argc, argv) < 0) {
1626                 fprintf(stderr, "obd_initialize failed.\n");
1627                 exit(-1);
1628         }
1629
1630         rc = lr_locate_rsync();
1631         if (use_rsync && rc != 0) {
1632                 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1633                 exit(-1);
1634         }
1635
1636         signal(SIGINT, termination_handler);
1637         signal(SIGHUP, termination_handler);
1638         signal(SIGTERM, termination_handler);
1639
1640         rc = lr_replicate();
1641
1642         return rc;
1643 }