Whamcloud - gitweb
LU-8080 utils: Replace calls to signal with sigaction
[fs/lustre-release.git] / lustre / utils / lhsmtool_posix.c
index 60077d8..d99fce3 100644 (file)
@@ -23,6 +23,7 @@
  * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
  *     alternatives
  *
+ * Copyright (c) 2013, 2015, Intel Corporation.
  */
 /* HSM copytool program for POSIX filesystem-based HSM's.
  *
  *
  * This particular tool can also import an existing HSM archive.
  */
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <ctype.h>
+#include <fcntl.h>
+#include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
+#include <string.h>
 #include <dirent.h>
 #include <errno.h>
+#include <getopt.h>
+#include <pthread.h>
+#include <time.h>
+#include <unistd.h>
 #include <utime.h>
+#include <signal.h>
+#include <sys/time.h>
 #include <sys/xattr.h>
 #include <sys/syscall.h>
 #include <sys/types.h>
+
+#include <libcfs/util/string.h>
 #include <lustre/lustre_idl.h>
 #include <lustre/lustreapi.h>
 
 
 #define ONE_MB 0x100000
 
-/* copytool uses a 32b bitmask field to register with kuc
- * archive num = 0 => all
- * archive num from 1 to 32
- */
-#define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
+#ifndef NSEC_PER_SEC
+# define NSEC_PER_SEC 1000000000UL
+#endif
 
 enum ct_action {
        CA_IMPORT = 1,
@@ -73,12 +88,14 @@ struct options {
        int                      o_verbose;
        int                      o_copy_xattrs;
        int                      o_archive_cnt;
-       int                      o_archive_id[MAX_ARCHIVE_CNT];
+       int                      o_archive_id[LL_HSM_MAX_ARCHIVE];
        int                      o_report_int;
        unsigned long long       o_bandwidth;
        size_t                   o_chunk_size;
        enum ct_action           o_action;
+       char                    *o_event_fifo;
        char                    *o_mnt;
+       int                      o_mnt_fd;
        char                    *o_hsm_root;
        char                    *o_src; /* for import, or rebind */
        char                    *o_dst; /* for import, or rebind */
@@ -109,27 +126,34 @@ static char fs_name[MAX_OBD_NAME + 1];
 
 static struct hsm_copytool_private *ctdata;
 
+static inline double ct_now(void)
+{
+       struct timeval tv;
+
+       gettimeofday(&tv, NULL);
 
-#define CT_ERROR(_rc, _format, ...) \
+       return tv.tv_sec + 0.000001 * tv.tv_usec;
+}
+
+#define CT_ERROR(_rc, _format, ...)                                    \
        llapi_error(LLAPI_MSG_ERROR, _rc,                               \
-                   "%s[%ld]: "_format,                                 \
-                   cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_DEBUG(_format, ...) \
+                   "%f %s[%ld]: "_format,                              \
+                   ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
+#define CT_DEBUG(_format, ...)                                         \
        llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0,            \
-                   "%s[%ld]: "_format,                                 \
-                   cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+                   "%f %s[%ld]: "_format,                              \
+                   ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
 #define CT_WARN(_format, ...) \
        llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0,             \
-                   "%s[%ld]: "_format,                                 \
-                   cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_TRACE(_format, ...) \
+                   "%f %s[%ld]: "_format,                              \
+                   ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
+#define CT_TRACE(_format, ...)                                         \
        llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0,             \
-                   "%s[%ld]: "_format,                                 \
-                   cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_PRINTF(_format, ...) \
-       llapi_printf(LLAPI_MSG_NORMAL,                                  \
-                    "%s[%ld]: "_format,                                \
-                    cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+                   "%f %s[%ld]: "_format,                              \
+                   ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
 
 static void usage(const char *name, int rc)
 {
@@ -152,9 +176,10 @@ static void usage(const char *name, int rc)
        "into a Lustre filesystem.\n"
        " Usage:\n"
        "   %s [options] --import <src> <dst> <lustre_mount_point>\n"
-       "      import an archived subtree at\n"
-       "       <src> (relative to hsm_root) into the Lustre filesystem at\n"
-       "       <dst> (absolute)\n"
+       "      import an archived subtree from\n"
+       "       <src> (FID or relative path to hsm_root) into the Lustre\n"
+       "             filesystem at\n"
+       "       <dst> (absolute path)\n"
        "   %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
        "      rebind an entry in the HSM to a new FID\n"
        "       <old_FID> old FID the HSM entry is bound to\n"
@@ -164,16 +189,19 @@ 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 +217,8 @@ 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 +234,8 @@ 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}
        };
@@ -213,15 +244,15 @@ 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':
-                       if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
-                           (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
+                       if ((opt.o_archive_cnt >= LL_HSM_MAX_ARCHIVE) ||
+                           (atoi(optarg) >= LL_HSM_MAX_ARCHIVE)) {
                                rc = -E2BIG;
                                CT_ERROR(rc, "archive number must be less"
-                                        "than %zu", MAX_ARCHIVE_CNT);
+                                        "than %zu", LL_HSM_MAX_ARCHIVE);
                                return rc;
                        }
                        opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
@@ -241,6 +272,9 @@ 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 +292,15 @@ 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;
@@ -306,11 +349,12 @@ static int ct_parseopts(int argc, char * const *argv)
        }
 
        opt.o_mnt = argv[optind];
+       opt.o_mnt_fd = -1;
 
        CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
                 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
 
-       if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
+       if (opt.o_hsm_root == NULL) {
                rc = -EINVAL;
                CT_ERROR(rc, "must specify a root directory for the backend");
                return rc;
@@ -341,6 +385,9 @@ static int ct_mkdir_p(const char *path)
        int      rc;
 
        ptr = strdup(path);
+       if (ptr == NULL)
+               return -errno;
+
        saved = ptr;
        while (*ptr == '/')
                ptr++;
@@ -460,80 +507,36 @@ static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
        rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
                       XATTR_CREATE);
        if (rc < 0) {
-               CT_ERROR(errno, "cannot set lov EA on '%s'", dst);
                rc = -errno;
+               CT_ERROR(rc, "cannot set lov EA on '%s'", dst);
        }
 
        return rc;
 }
 
-/* non-blocking read or write */
-static int nonblock_rw(bool wr, int fd, char *buf, int size)
-{
-       int rc;
-
-       if (wr)
-               rc = write(fd, buf, size);
-       else
-               rc = read(fd, buf, size);
-
-       if ((rc < 0) && (errno == -EAGAIN)) {
-               fd_set set;
-               struct timeval timeout;
-
-               timeout.tv_sec = opt.o_report_int;
-
-               FD_ZERO(&set);
-               FD_SET(fd, &set);
-               if (wr)
-                       rc = select(FD_SETSIZE, NULL, &set, NULL, &timeout);
-               else
-                       rc = select(FD_SETSIZE, &set, NULL, NULL, &timeout);
-               if (rc < 0)
-                       return -errno;
-               if (rc == 0)
-                       /* Timed out, we read nothing */
-                       return -EAGAIN;
-
-               /* Should be available now */
-               if (wr)
-                       rc = write(fd, buf, size);
-               else
-                       rc = read(fd, buf, size);
-       }
-
-       if (rc < 0)
-               rc = -errno;
-
-       return rc;
-}
-
 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                        const char *dst, int src_fd, int dst_fd,
                        const struct hsm_action_item *hai, long hal_flags)
 {
        struct hsm_extent        he;
+       __u64                    offset = hai->hai_extent.offset;
        struct stat              src_st;
        struct stat              dst_st;
-       char                    *buf;
-       __u64                    wpos = 0;
-       __u64                    rpos = 0;
-       __u64                    rlen;
-       time_t                   last_print_time = time(0);
-       int                      rsize;
-       int                      wsize;
-       int                      bufoff = 0;
+       char                    *buf = NULL;
+       __u64                    write_total = 0;
+       __u64                    length = hai->hai_extent.length;
+       time_t                   last_report_time;
        int                      rc = 0;
-
-       CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
-
-       buf = malloc(opt.o_chunk_size);
-       if (buf == NULL)
-               return -ENOMEM;
+       double                   start_ct_now = ct_now();
+       /* Bandwidth Control */
+       time_t                  start_time;
+       time_t                  now;
+       time_t                  last_bw_print;
 
        if (fstat(src_fd, &src_st) < 0) {
-               CT_ERROR(errno, "cannot stat '%s'", src);
-               return -errno;
+               rc = -errno;
+               CT_ERROR(rc, "cannot stat '%s'", src);
+               return rc;
        }
 
        if (!S_ISREG(src_st.st_mode)) {
@@ -542,18 +545,19 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                return rc;
        }
 
-       rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
-       if (rc < 0) {
-               CT_ERROR(errno,
-                        "cannot seek for read to "LPU64" (len %jd) in '%s'",
-                        hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
-               rc = -errno;
-               goto out;
+       if (hai->hai_extent.offset > (__u64)src_st.st_size) {
+               rc = -EINVAL;
+               CT_ERROR(rc, "Trying to start reading past end (%ju > "
+                        "%jd) of '%s' source file",
+                        (uintmax_t)hai->hai_extent.offset,
+                        (intmax_t)src_st.st_size, src);
+               return rc;
        }
 
        if (fstat(dst_fd, &dst_st) < 0) {
-               CT_ERROR(errno, "cannot stat '%s'", dst);
-               return -errno;
+               rc = -errno;
+               CT_ERROR(rc, "cannot stat '%s'", dst);
+               return rc;
        }
 
        if (!S_ISREG(dst_st.st_mode)) {
@@ -562,17 +566,15 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                return rc;
        }
 
-       rc = lseek(dst_fd, hai->hai_extent.offset, SEEK_SET);
-       if (rc < 0) {
-               rc = -errno;
-               CT_ERROR(rc, "cannot seek for write to "LPU64" on '%s'",
-                        hai->hai_extent.offset, src);
-               goto out;
-       }
+       /* Don't read beyond a given extent */
+       if (length > src_st.st_size - hai->hai_extent.offset)
+               length = src_st.st_size - hai->hai_extent.offset;
+
+       start_time = last_bw_print = last_report_time = time(NULL);
 
-       he.offset = hai->hai_extent.offset;
+       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. */
@@ -582,93 +584,92 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
        }
 
        errno = 0;
-       /* Don't read beyond a given extent */
-       rlen = min(hai->hai_extent.length, src_st.st_size);
-
-       CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", rlen, src, dst);
 
-       while (wpos < rlen) {
-               int chunk = (rlen - wpos > opt.o_chunk_size) ?
-                           opt.o_chunk_size : rlen - wpos;
+       buf = malloc(opt.o_chunk_size);
+       if (buf == NULL) {
+               rc = -ENOMEM;
+               goto out;
+       }
 
-               /* Only read more if we wrote everything in the buffer */
-               if (wpos == rpos) {
-                       rsize = nonblock_rw(0, src_fd, buf, chunk);
-                       if (rsize == 0)
-                               /* EOF */
-                               break;
+       CT_TRACE("start copy of %ju bytes from '%s' to '%s'",
+                (uintmax_t)length, src, dst);
 
-                       if (rsize == -EAGAIN) {
-                               /* Timed out */
-                               rsize = 0;
-                               if (rpos == 0) {
-                                       /* Haven't read anything yet, let's
-                                        * give it back to the coordinator
-                                        * for rescheduling */
-                                       rc = -EAGAIN;
-                                       break;
-                               }
-                       }
+       while (write_total < length) {
+               ssize_t rsize;
+               ssize_t wsize;
+               int     chunk = (length - write_total > opt.o_chunk_size) ?
+                                opt.o_chunk_size : length - write_total;
 
-                       if (rsize < 0) {
-                               rc = rsize;
-                               CT_ERROR(rc, "cannot read from '%s'", src);
-                               break;
-                       }
+               rsize = pread(src_fd, buf, chunk, offset);
+               if (rsize == 0)
+                       /* EOF */
+                       break;
 
-                       rpos += rsize;
-                       bufoff = 0;
+               if (rsize < 0) {
+                       rc = -errno;
+                       CT_ERROR(rc, "cannot read from '%s'", src);
+                       break;
                }
 
-               wsize = nonblock_rw(1, dst_fd, buf + bufoff, rpos - wpos);
-               if (wsize == -EAGAIN)
-                       /* Timed out */
-                       wsize = 0;
-
+               wsize = pwrite(dst_fd, buf, rsize, offset);
                if (wsize < 0) {
-                       rc = wsize;
+                       rc = -errno;
                        CT_ERROR(rc, "cannot write to '%s'", dst);
                        break;
                }
 
-               wpos += wsize;
-               bufoff += wsize;
+               write_total += wsize;
+               offset += wsize;
 
+               now = time(NULL);
+               /* sleep if needed, to honor bandwidth limits */
                if (opt.o_bandwidth != 0) {
-                       static unsigned long long       tot_bytes;
-                       static time_t                   start_time, last_time;
-                       time_t                          now = time(0);
-                       double                          tot_time, excess;
-                       unsigned int                    sleep_time;
-
-                       if (now > last_time + 5) {
-                               tot_bytes = 0;
-                               start_time = last_time = now;
-                       }
+                       unsigned long long write_theory;
+
+                       write_theory = (now - start_time) * opt.o_bandwidth;
 
-                       tot_bytes += wsize;
-                       tot_time = now - start_time;
-                       if (tot_time < 1)
-                               tot_time = 1;
+                       if (write_theory < write_total) {
+                               unsigned long long      excess;
+                               struct timespec         delay;
 
-                       excess = tot_bytes - tot_time * opt.o_bandwidth;
-                       sleep_time = excess * 1000000 / opt.o_bandwidth;
-                       if ((now - start_time) % 10 == 1)
-                               CT_TRACE("bandwith control: excess=%E"
-                                        " sleep for %dus",
-                                        excess, sleep_time);
+                               excess = write_total - write_theory;
 
-                       if (excess > 0)
-                               usleep(sleep_time);
+                               delay.tv_sec = excess / opt.o_bandwidth;
+                               delay.tv_nsec = (excess % opt.o_bandwidth) *
+                                       NSEC_PER_SEC / opt.o_bandwidth;
 
-                       last_time = now;
+                               if (now >= last_bw_print + opt.o_report_int) {
+                                       CT_TRACE("bandwith control: %lluB/s "
+                                                "excess=%llu sleep for "
+                                                "%lld.%09lds",
+                                                opt.o_bandwidth, excess,
+                                                (long long)delay.tv_sec,
+                                                delay.tv_nsec);
+                                       last_bw_print = now;
+                               }
+
+                               do {
+                                       rc = nanosleep(&delay, &delay);
+                               } while (rc < 0 && errno == EINTR);
+                               if (rc < 0) {
+                                       CT_ERROR(errno, "delay for bandwidth "
+                                                "control failed to sleep: "
+                                                "residual=%lld.%09lds",
+                                                (long long)delay.tv_sec,
+                                                delay.tv_nsec);
+                                       rc = 0;
+                               }
+                       }
                }
 
-               if (time(0) >= last_print_time + opt.o_report_int) {
-                       last_print_time = time(0);
-                       CT_TRACE("%%"LPU64" ", 100 * wpos / rlen);
-                       he.length = wpos;
-                       rc = llapi_hsm_action_progress(hcp, &he, 0);
+               now = time(NULL);
+               if (now >= last_report_time + opt.o_report_int) {
+                       last_report_time = now;
+                       CT_TRACE("%%%ju ", (uintmax_t)(100 * write_total / length));
+                       /* only give the length of the write since the last
+                        * progress report */
+                       he.length = offset - he.offset;
+                       rc = llapi_hsm_action_progress(hcp, &he, length, 0);
                        if (rc < 0) {
                                /* Action has been canceled or something wrong
                                 * is happening. Stop copying data. */
@@ -676,6 +677,7 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                                         " '%s'->'%s' failed", src, dst);
                                goto out;
                        }
+                       he.offset = offset;
                }
                rc = 0;
        }
@@ -695,14 +697,17 @@ out:
                rc = ftruncate(dst_fd, src_st.st_size);
                if (rc < 0) {
                        rc = -errno;
-                       CT_ERROR(rc,
-                                "cannot truncate '%s' to size %jd",
+                       CT_ERROR(rc, "cannot truncate '%s' to size %jd",
                                 dst, (intmax_t)src_st.st_size);
                        err_major++;
                }
        }
 
-       free(buf);
+       if (buf != NULL)
+               free(buf);
+
+       CT_TRACE("copied %ju bytes in %f seconds",
+                (uintmax_t)length, ct_now() - start_ct_now);
 
        return rc;
 }
@@ -831,20 +836,32 @@ 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",
-                hai->hai_cookie, PFID(&hai->hai_fid),
+                "cookie=%#jx, FID="DFID", hp_flags=%d err=%d",
+                (uintmax_t)hai->hai_cookie, PFID(&hai->hai_fid),
                 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: "
-                        "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
-                        PFID(&hai->hai_fid));
+                        "cookie=%#jx, FID="DFID, lstr,
+                        (uintmax_t)hai->hai_cookie, PFID(&hai->hai_fid));
        else if (rc < 0)
                CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
        else
@@ -858,7 +875,7 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
 {
        struct hsm_copyaction_private   *hcp = NULL;
        char                             src[PATH_MAX];
-       char                             dst[PATH_MAX];
+       char                             dst[PATH_MAX] = "";
        int                              rc;
        int                              rcf = 0;
        bool                             rename_needed = false;
@@ -880,7 +897,7 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
        if (hai->hai_extent.length == -1) {
                /* whole file, write it to tmp location and atomically
                 * replace old archived file */
-               strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
+               strlcat(dst, "_tmp", sizeof(dst));
                /* we cannot rely on the same test because ct_copy_data()
                 * updates hai_extent.length */
                rename_needed = true;
@@ -901,12 +918,12 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
 
        src_fd = llapi_hsm_action_get_fd(hcp);
        if (src_fd < 0) {
-               rc = -errno;
+               rc = src_fd;
                CT_ERROR(rc, "cannot open '%s' for read", src);
                goto fini_major;
        }
 
-       open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
+       open_flags = O_WRONLY | O_NOFOLLOW;
        /* If extent is specified, don't truncate an old archived copy */
        open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
 
@@ -998,7 +1015,7 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
                int              linkno = 0;
                char            *ptr;
                int              depth = 0;
-               int              sz;
+               ssize_t          sz;
 
                sprintf(buf, DFID, PFID(&hai->hai_fid));
                sprintf(src, "%s/shadow/", opt.o_hsm_root);
@@ -1032,6 +1049,12 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
                }
                /* symlink already exists ? */
                sz = readlink(src, buf, sizeof(buf));
+               /* detect truncation */
+               if (sz == sizeof(buf)) {
+                       rcf = rcf ? rcf : -E2BIG;
+                       CT_ERROR(rcf, "readlink '%s' truncated", src);
+                       goto fini_minor;
+               }
                if (sz >= 0) {
                        buf[sz] = '\0';
                        if (sz == 0 || strncmp(buf, dst, sz) != 0) {
@@ -1083,8 +1106,7 @@ 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;
 }
@@ -1100,7 +1122,7 @@ static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
        int                              hp_flags = 0;
        int                              src_fd = -1;
        int                              dst_fd = -1;
-       int                              mdt_index = -1; /* Not implemented */
+       int                              mdt_index = -1;
        int                              open_flags = 0;
        bool                             set_lovea;
        struct lu_fid                    dfid;
@@ -1112,6 +1134,13 @@ static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
        /* build backend file name from released file FID */
        ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
 
+       rc = llapi_get_mdt_index_by_fid(opt.o_mnt_fd, &hai->hai_fid,
+                                       &mdt_index);
+       if (rc < 0) {
+               CT_ERROR(rc, "cannot get mdt index "DFID"",
+                        PFID(&hai->hai_fid));
+               return rc;
+       }
        /* restore loads and sets the LOVEA w/o interpreting it to avoid
         * dependency on the structure format. */
        rc = ct_load_stripe(src, lov_buf, &lov_size);
@@ -1147,7 +1176,7 @@ static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
                goto fini;
        }
 
-       src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
+       src_fd = open(src, O_RDONLY | O_NOATIME | O_NOFOLLOW);
        if (src_fd < 0) {
                rc = -errno;
                CT_ERROR(rc, "cannot open '%s' for read", src);
@@ -1155,6 +1184,11 @@ static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
        }
 
        dst_fd = llapi_hsm_action_get_fd(hcp);
+       if (dst_fd < 0) {
+               rc = dst_fd;
+               CT_ERROR(rc, "cannot open '%s' for write", dst);
+               goto fini;
+       }
 
        if (set_lovea) {
                /* the layout cannot be allocated through .fid so we have to
@@ -1181,8 +1215,7 @@ 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 */
@@ -1222,7 +1255,7 @@ static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
                goto fini;
        }
 
-       strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
+       strlcat(dst, ".lov", sizeof(dst));
        rc = unlink(dst);
        if (rc < 0) {
                rc = -errno;
@@ -1232,23 +1265,7 @@ 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;
 }
@@ -1265,9 +1282,9 @@ static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
                int             linkno = 0;
 
                sprintf(fid, DFID, PFID(&hai->hai_fid));
-               CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
+               CT_TRACE("'%s' action %s reclen %d, cookie=%#jx",
                         fid, hsm_copytool_action2name(hai->hai_action),
-                        hai->hai_len, hai->hai_cookie);
+                        hai->hai_len, (uintmax_t)hai->hai_cookie);
                rc = llapi_fid2path(opt.o_mnt, fid, path,
                                    sizeof(path), &recno, &linkno);
                if (rc < 0)
@@ -1301,7 +1318,7 @@ 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;
@@ -1417,29 +1434,53 @@ static int ct_import_one(const char *src, const char *dst)
 static char *path_concat(const char *dirname, const char *basename)
 {
        char    *result;
-       int      dirlen = strlen(dirname);
+       int      rc;
 
-       result = malloc(dirlen + strlen(basename) + 2);
-       if (result == NULL)
+       rc = asprintf(&result, "%s/%s", dirname, basename);
+       if (rc < 0)
                return NULL;
 
-       memcpy(result, dirname, dirlen);
-       result[dirlen] = '/';
-       strcpy(result + dirlen + 1, basename);
-
        return result;
 }
 
+static int ct_import_fid(const lustre_fid *import_fid)
+{
+       char    fid_path[PATH_MAX];
+       int     rc;
+
+       ct_path_lustre(fid_path, sizeof(fid_path), opt.o_mnt, import_fid);
+       rc = access(fid_path, F_OK);
+       if (rc == 0 || errno != ENOENT) {
+               rc = (errno == 0) ? -EEXIST : -errno;
+               CT_ERROR(rc, "cannot import '"DFID"'", PFID(import_fid));
+               return rc;
+       }
+
+       ct_path_archive(fid_path, sizeof(fid_path), opt.o_hsm_root,
+                       import_fid);
+
+       CT_TRACE("Resolving "DFID" to %s", PFID(import_fid), fid_path);
+
+       return ct_import_one(fid_path, opt.o_dst);
+}
+
 static int ct_import_recurse(const char *relpath)
 {
        DIR             *dir;
        struct dirent    ent, *cookie = NULL;
        char            *srcpath, *newpath;
+       lustre_fid       import_fid;
        int              rc;
 
        if (relpath == NULL)
                return -EINVAL;
 
+       /* Is relpath a FID? In which case SFID should expand to three
+        * elements. */
+       rc = sscanf(relpath, SFID, RFID(&import_fid));
+       if (rc == 3)
+               return ct_import_fid(&import_fid);
+
        srcpath = path_concat(opt.o_hsm_root, relpath);
        if (srcpath == NULL) {
                err_major++;
@@ -1541,10 +1582,10 @@ static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
                        return -errno;
                }
                /* rename lov file */
-               strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
-               strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
+               strlcat(src, ".lov", sizeof(src));
+               strlcat(dst, ".lov", sizeof(dst));
                if (rename(src, dst))
-                       CT_ERROR(errno, "cannot '%s' rename to '%s'", src, dst);
+                       CT_ERROR(errno, "cannot rename '%s' to '%s'", src, dst);
 
        }
        return 0;
@@ -1620,7 +1661,7 @@ static int ct_rebind_list(const char *list)
        if (line)
                free(line);
 
-       /* return 0 if all rebinds were sucessful */
+       /* return 0 if all rebinds were successful */
        CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
 
        return ok == nl ? 0 : -1;
@@ -1702,24 +1743,31 @@ out:
 
 static int ct_max_sequence(void)
 {
-       int   rc, i;
-       char  path[PATH_MAX];
-       __u64 seq = 0;
-       __u16 subseq;
+       int     rc, i;
+       char    path[PATH_MAX];
+       __u64   seq = 0;
+       __u16   subseq;
 
-       strncpy(path, opt.o_hsm_root, sizeof(path));
+       strlcpy(path, opt.o_hsm_root, sizeof(path));
        /* FID sequence is stored in top-level directory names:
         * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
         */
        for (i = 0; i < 4; i++) {
+               size_t path_len;
+
                rc = ct_dir_level_max(path, &subseq);
                if (rc != 0)
                        return rc;
                seq |= ((__u64)subseq << ((3 - i) * 16));
-               sprintf(path + strlen(path), "/%04x", subseq);
+               path_len = strlen(path);
+               rc = snprintf(path + path_len, sizeof(path) - path_len,
+                             "/%04x", subseq);
+               if (rc >= (sizeof(path) - path_len))
+                       return -E2BIG;
+               path[sizeof(path) - 1] = '\0';
        }
 
-       printf("max_sequence: %016Lx\n", seq);
+       printf("max_sequence: %#jx\n", (uintmax_t)seq);
 
        return 0;
 }
@@ -1732,13 +1780,18 @@ static void handler(int signal)
         * does successfully unmount and the mount is actually gone, but the
         * mtab entry remains. So this just makes mtab happier. */
        llapi_hsm_copytool_unregister(&ctdata);
+
+       /* Also remove fifo upon signal as during normal/error exit */
+       if (opt.o_event_fifo != NULL)
+               llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
        _exit(1);
 }
 
 /* Daemon waits for messages from the kernel; run it in the background. */
 static int ct_run(void)
 {
-       int     rc;
+       struct sigaction cleanup_sigaction;
+       int rc;
 
        if (opt.o_daemonize) {
                rc = daemon(1, 1);
@@ -1751,21 +1804,34 @@ static int ct_run(void)
 
        setbuf(stdout, NULL);
 
-       rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
-                                        opt.o_archive_cnt, opt.o_archive_id);
+       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,
+                                        opt.o_archive_cnt,
+                                        opt.o_archive_id, 0);
        if (rc < 0) {
                CT_ERROR(rc, "cannot start copytool interface");
                return rc;
        }
 
-       signal(SIGINT, handler);
-       signal(SIGTERM, handler);
+       memset(&cleanup_sigaction, 0, sizeof(cleanup_sigaction));
+       cleanup_sigaction.sa_handler = handler;
+       sigemptyset(&cleanup_sigaction.sa_mask);
+       sigaction(SIGINT, &cleanup_sigaction, NULL);
+       sigaction(SIGTERM, &cleanup_sigaction, NULL);
 
        while (1) {
-               struct hsm_action_list  *hal;
-               struct hsm_action_item  *hai;
-               int                      msgsize;
-               int                      i = 0;
+               struct hsm_action_list *hal;
+               struct hsm_action_item *hai;
+               int msgsize;
+               int i = 0;
 
                CT_TRACE("waiting for message from kernel");
 
@@ -1773,8 +1839,6 @@ static int ct_run(void)
                if (rc == -ESHUTDOWN) {
                        CT_TRACE("shutting down");
                        break;
-               } else if (rc == -EAGAIN) {
-                       continue; /* msg not for us */
                } else if (rc < 0) {
                        CT_WARN("cannot receive action list: %s",
                                strerror(-rc));
@@ -1789,7 +1853,8 @@ static int ct_run(void)
                         hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
 
                if (strcmp(hal->hal_fsname, fs_name) != 0) {
-                       CT_ERROR(EINVAL, "'%s' invalid fs name, expecting: %s",
+                       rc = -EINVAL;
+                       CT_ERROR(rc, "'%s' invalid fs name, expecting: %s",
                                 hal->hal_fsname, fs_name);
                        err_major++;
                        if (opt.o_abort_on_error)
@@ -1817,13 +1882,13 @@ static int ct_run(void)
                        hai = hai_next(hai);
                }
 
-               llapi_hsm_action_list_free(&hal);
-
                if (opt.o_abort_on_error && err_major)
                        break;
        }
 
        llapi_hsm_copytool_unregister(&ctdata);
+       if (opt.o_event_fifo != NULL)
+               llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
 
        return rc;
 }
@@ -1846,7 +1911,15 @@ static int ct_setup(void)
        if (rc < 0) {
                CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
                         opt.o_mnt);
-               return -rc;
+               return rc;
+       }
+
+       opt.o_mnt_fd = open(opt.o_mnt, O_RDONLY);
+       if (opt.o_mnt_fd < 0) {
+               rc = -errno;
+               CT_ERROR(rc, "cannot open mount point at '%s'",
+                        opt.o_mnt);
+               return rc;
        }
 
        return rc;
@@ -1856,13 +1929,22 @@ static int ct_cleanup(void)
 {
        int     rc;
 
-       if (arc_fd < 0)
-               return 0;
+       if (opt.o_mnt_fd >= 0) {
+               rc = close(opt.o_mnt_fd);
+               if (rc < 0) {
+                       rc = -errno;
+                       CT_ERROR(rc, "cannot close mount point");
+                       return rc;
+               }
+       }
 
-       if (close(arc_fd) < 0) {
-               rc = -errno;
-               CT_ERROR(rc, "cannot close archive root directory");
-               return rc;
+       if (arc_fd >= 0) {
+               rc = close(arc_fd);
+               if (rc < 0) {
+                       rc = -errno;
+                       CT_ERROR(rc, "cannot close archive root directory");
+                       return rc;
+               }
        }
 
        return 0;
@@ -1872,14 +1954,16 @@ int main(int argc, char **argv)
 {
        int     rc;
 
-       strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
+       strlcpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
        rc = ct_parseopts(argc, argv);
        if (rc < 0) {
                CT_WARN("try '%s --help' for more information", cmd_name);
                return -rc;
        }
 
-       ct_setup();
+       rc = ct_setup();
+       if (rc < 0)
+               goto error_cleanup;
 
        switch (opt.o_action) {
        case CA_IMPORT:
@@ -1901,6 +1985,7 @@ int main(int argc, char **argv)
                         " rc=%d (%s)", err_major, err_minor, rc,
                         strerror(-rc));
 
+error_cleanup:
        ct_cleanup();
 
        return -rc;