Whamcloud - gitweb
LU-4545 hsm: Always report errors to coordinator.
[fs/lustre-release.git] / lustre / utils / lhsmtool_posix.c
index f182dac..d90b725 100644 (file)
@@ -78,6 +78,7 @@ struct options {
        unsigned long long       o_bandwidth;
        size_t                   o_chunk_size;
        enum ct_action           o_action;
+       char                    *o_event_fifo;
        char                    *o_mnt;
        char                    *o_hsm_root;
        char                    *o_src; /* for import, or rebind */
@@ -164,16 +165,19 @@ static void usage(const char *name, int rc)
        "       each line of <list_file> consists of <old_FID> <new_FID>\n"
        "   %s [options] --max-sequence <fsname>\n"
        "       return the max fid sequence of archived files\n"
-       "   -A, --archive <#>        Archive number (repeatable)\n"
-       "   -p, --hsm-root <path>    Target HSM mount point\n"
-       "   -q, --quiet              Produce less verbose output\n"
-       "   -v, --verbose            Produce more verbose output\n"
-       "   -c, --chunk-size <sz>    I/O size used during data copy\n"
-       "                            (unit can be used, default is MB)\n"
-       "   --abort-on-error         Abort operation on major error\n"
-       "   --dry-run                Don't run, just show what would be done\n"
-       "   --bandwidth <bw>         Limit I/O bandwidth (unit can be used\n,"
-       "                            default is MB)\n",
+       "   --abort-on-error          Abort operation on major error\n"
+       "   -A, --archive <#>         Archive number (repeatable)\n"
+       "   -b, --bandwidth <bw>      Limit I/O bandwidth (unit can be used\n,"
+       "                             default is MB)\n"
+       "   --dry-run                 Don't run, just show what would be done\n"
+       "   -c, --chunk-size <sz>     I/O size used during data copy\n"
+       "                             (unit can be used, default is MB)\n"
+       "   -f, --event-fifo <path>   Write events stream to fifo\n"
+       "   -p, --hsm-root <path>     Target HSM mount point\n"
+       "   -q, --quiet               Produce less verbose output\n"
+       "   -u, --update-interval <s> Interval between progress reports sent\n"
+       "                             to Coordinator\n"
+       "   -v, --verbose             Produce more verbose output\n",
        cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
 
        exit(rc);
@@ -189,6 +193,8 @@ static int ct_parseopts(int argc, char * const *argv)
                {"chunk-size",     required_argument, NULL,                'c'},
                {"chunk_size",     required_argument, NULL,                'c'},
                {"daemon",         no_argument,       &opt.o_daemonize,     1},
+               {"event-fifo",     required_argument, NULL,                'f'},
+               {"event_fifo",     required_argument, NULL,                'f'},
                {"dry-run",        no_argument,       &opt.o_dry_run,       1},
                {"help",           no_argument,       NULL,                'h'},
                {"hsm-root",       required_argument, NULL,                'p'},
@@ -204,7 +210,8 @@ static int ct_parseopts(int argc, char * const *argv)
                {"no_xattr",       no_argument,       &opt.o_copy_xattrs,   0},
                {"quiet",          no_argument,       NULL,                'q'},
                {"rebind",         no_argument,       NULL,                'r'},
-               {"report",         required_argument, &opt.o_report_int,    0},
+               {"update-interval", required_argument,  NULL,              'u'},
+               {"update_interval", required_argument,  NULL,              'u'},
                {"verbose",        no_argument,       NULL,                'v'},
                {0, 0, 0, 0}
        };
@@ -213,7 +220,7 @@ static int ct_parseopts(int argc, char * const *argv)
        unsigned long long       unit;
 
        optind = 0;
-       while ((c = getopt_long(argc, argv, "A:b:c:hiMp:qrv",
+       while ((c = getopt_long(argc, argv, "A:b:c:f:hiMp:qru:v",
                                long_opts, NULL)) != -1) {
                switch (c) {
                case 'A':
@@ -241,6 +248,9 @@ static int ct_parseopts(int argc, char * const *argv)
                        else
                                opt.o_bandwidth = value;
                        break;
+               case 'f':
+                       opt.o_event_fifo = optarg;
+                       break;
                case 'h':
                        usage(argv[0], 0);
                case 'i':
@@ -258,6 +268,15 @@ static int ct_parseopts(int argc, char * const *argv)
                case 'r':
                        opt.o_action = CA_REBIND;
                        break;
+               case 'u':
+                       opt.o_report_int = atoi(optarg);
+                       if (opt.o_report_int < 0) {
+                               rc = -EINVAL;
+                               CT_ERROR(rc, "bad value for -%c '%s'", c,
+                                        optarg);
+                               return rc;
+                       }
+                       break;
                case 'v':
                        opt.o_verbose++;
                        break;
@@ -548,9 +567,12 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                return rc;
        }
 
+       /* Don't read beyond a given extent */
+       length = min(hai->hai_extent.length, src_st.st_size);
+
        he.offset = offset;
        he.length = 0;
-       rc = llapi_hsm_action_progress(hcp, &he, 0);
+       rc = llapi_hsm_action_progress(hcp, &he, length, 0);
        if (rc < 0) {
                /* Action has been canceled or something wrong
                 * is happening. Stop copying data. */
@@ -560,8 +582,6 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
        }
 
        errno = 0;
-       /* Don't read beyond a given extent */
-       length = min(hai->hai_extent.length, src_st.st_size);
 
        buf = malloc(opt.o_chunk_size);
        if (buf == NULL) {
@@ -606,7 +626,7 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                        last_print_time = time(0);
                        CT_TRACE("%%"LPU64" ", 100 * write_total / length);
                        he.length = write_total;
-                       rc = llapi_hsm_action_progress(hcp, &he, 0);
+                       rc = llapi_hsm_action_progress(hcp, &he, length, 0);
                        if (rc < 0) {
                                /* Action has been canceled or something wrong
                                 * is happening. Stop copying data. */
@@ -769,8 +789,9 @@ static int ct_begin(struct hsm_copyaction_private **phcp,
 static int ct_fini(struct hsm_copyaction_private **phcp,
                   const struct hsm_action_item *hai, int hp_flags, int ct_rc)
 {
-       char    lstr[PATH_MAX];
-       int     rc;
+       struct hsm_copyaction_private   *hcp;
+       char                             lstr[PATH_MAX];
+       int                              rc;
 
        CT_TRACE("Action completed, notifying coordinator "
                 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
@@ -778,6 +799,17 @@ static int ct_fini(struct hsm_copyaction_private **phcp,
                 hp_flags, -ct_rc);
 
        ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
+
+       if (phcp == NULL || *phcp == NULL) {
+               rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
+               if (rc < 0) {
+                       CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed",
+                                lstr);
+                       return rc;
+               }
+               phcp = &hcp;
+       }
+
        rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
        if (rc == -ECANCELED)
                CT_ERROR(rc, "completed action on '%s' has been canceled: "
@@ -1027,8 +1059,7 @@ out:
        if (!(dst_fd < 0))
                close(dst_fd);
 
-       if (hcp != NULL)
-               rc = ct_fini(&hcp, hai, hp_flags, rcf);
+       rc = ct_fini(&hcp, hai, hp_flags, rcf);
 
        return rc;
 }
@@ -1125,8 +1156,7 @@ static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
        CT_TRACE("data restore from '%s' to '%s' done", src, dst);
 
 fini:
-       if (hcp != NULL)
-               rc = ct_fini(&hcp, hai, hp_flags, rc);
+       rc = ct_fini(&hcp, hai, hp_flags, rc);
 
        /* object swaping is done by cdt at copy end, so close of volatile file
         * cannot be done before */
@@ -1176,23 +1206,7 @@ static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
        }
 
 fini:
-       if (hcp != NULL)
-               rc = ct_fini(&hcp, hai, 0, rc);
-
-       return rc;
-}
-
-static int ct_report_error(const struct hsm_action_item *hai, int flags,
-                          int errval)
-{
-       struct hsm_copyaction_private   *hcp;
-       int                              rc;
-
-       rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
-       if (rc < 0)
-               return rc;
-
-       rc = llapi_hsm_action_end(&hcp, &hai->hai_extent, flags, abs(errval));
+       rc = ct_fini(&hcp, hai, 0, rc);
 
        return rc;
 }
@@ -1245,7 +1259,7 @@ static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
                CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
                         opt.o_mnt);
                err_minor++;
-               ct_report_error(hai, 0, rc);
+               ct_fini(NULL, hai, 0, rc);
        }
 
        return 0;
@@ -1695,6 +1709,15 @@ static int ct_run(void)
 
        setbuf(stdout, NULL);
 
+       if (opt.o_event_fifo != NULL) {
+               rc = llapi_hsm_register_event_fifo(opt.o_event_fifo);
+               if (rc < 0) {
+                       CT_ERROR(rc, "failed to register event fifo");
+                       return rc;
+               }
+               llapi_error_callback_set(llapi_hsm_log_error);
+       }
+
        rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
                                         opt.o_archive_cnt, opt.o_archive_id);
        if (rc < 0) {
@@ -1768,6 +1791,8 @@ static int ct_run(void)
        }
 
        llapi_hsm_copytool_unregister(&ctdata);
+       if (opt.o_event_fifo != NULL)
+               llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
 
        return rc;
 }