Whamcloud - gitweb
LU-4839 utils: fix bandwidth ctl in lhsmtool
[fs/lustre-release.git] / lustre / utils / lhsmtool_posix.c
index d90b725..183d910 100644 (file)
  *
  * This particular tool can also import an existing HSM archive.
  */
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <dirent.h>
 #include <errno.h>
+#include <time.h>
 #include <utime.h>
+#include <sys/time.h>
 #include <sys/xattr.h>
 #include <sys/syscall.h>
 #include <sys/types.h>
 
 #define ONE_MB 0x100000
 
+#ifndef NSEC_PER_SEC
+# define NSEC_PER_SEC 1000000000UL
+#endif
+
 /* copytool uses a 32b bitmask field to register with kuc
  * archive num = 0 => all
  * archive num from 1 to 32
@@ -80,6 +91,7 @@ struct options {
        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 */
@@ -110,27 +122,34 @@ static char fs_name[MAX_OBD_NAME + 1];
 
 static struct hsm_copytool_private *ctdata;
 
+static inline double ct_now(void)
+{
+       struct timeval tv;
 
-#define CT_ERROR(_rc, _format, ...) \
+       gettimeofday(&tv, NULL);
+
+       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)
 {
@@ -153,9 +172,10 @@ 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"
@@ -325,6 +345,7 @@ 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);
@@ -360,6 +381,9 @@ static int ct_mkdir_p(const char *path)
        int      rc;
 
        ptr = strdup(path);
+       if (ptr == NULL)
+               return -errno;
+
        saved = ptr;
        while (*ptr == '/')
                ptr++;
@@ -479,45 +503,13 @@ 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;
 }
 
-static void bandwidth_ctl_delay(int wsize)
-{
-       static unsigned long long       tot_bytes;
-       static time_t                   start_time;
-       static time_t                   last_time;
-       time_t                          now = time(0);
-       double                          tot_time;
-       double                          excess;
-       unsigned int                    sleep_time;
-
-       if (now > last_time + 5) {
-               tot_bytes = 0;
-               start_time = last_time = now;
-       }
-
-       tot_bytes += wsize;
-       tot_time = now - start_time;
-       if (tot_time < 1)
-               tot_time = 1;
-
-       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);
-
-       if (excess > 0)
-               usleep(sleep_time);
-
-       last_time = now;
-}
-
 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)
@@ -529,10 +521,13 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
        char                    *buf = NULL;
        __u64                    write_total = 0;
        __u64                    length;
-       time_t                   last_print_time = time(NULL);
+       time_t                   last_report_time;
        int                      rc = 0;
-
-       CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
+       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) {
                rc = -errno;
@@ -570,6 +565,8 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
        /* Don't read beyond a given extent */
        length = min(hai->hai_extent.length, src_st.st_size);
 
+       start_time = last_bw_print = last_report_time = time(NULL);
+
        he.offset = offset;
        he.length = 0;
        rc = llapi_hsm_action_progress(hcp, &he, length, 0);
@@ -589,7 +586,8 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                goto out;
        }
 
-       CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", length, src, dst);
+       CT_TRACE("start copy of "LPU64" bytes from '%s' to '%s'",
+                length, src, dst);
 
        while (write_total < length) {
                ssize_t rsize;
@@ -618,12 +616,50 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                write_total += wsize;
                offset += wsize;
 
-               if (opt.o_bandwidth != 0)
-                       /* sleep if needed, to honor bandwidth limits */
-                       bandwidth_ctl_delay(wsize);
+               now = time(NULL);
+               /* sleep if needed, to honor bandwidth limits */
+               if (opt.o_bandwidth != 0) {
+                       unsigned long long write_theory;
+
+                       write_theory = (now - start_time) * opt.o_bandwidth;
+
+                       if (write_theory < write_total) {
+                               unsigned long long      excess;
+                               struct timespec         delay;
 
-               if (time(0) >= last_print_time + opt.o_report_int) {
-                       last_print_time = time(0);
+                               excess = write_total - write_theory;
+
+                               delay.tv_sec = excess / opt.o_bandwidth;
+                               delay.tv_nsec = (excess % opt.o_bandwidth) *
+                                       NSEC_PER_SEC / opt.o_bandwidth;
+
+                               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;
+                               }
+                       }
+               }
+
+               now = time(NULL);
+               if (now >= last_report_time + opt.o_report_int) {
+                       last_report_time = now;
                        CT_TRACE("%%"LPU64" ", 100 * write_total / length);
                        he.length = write_total;
                        rc = llapi_hsm_action_progress(hcp, &he, length, 0);
@@ -662,6 +698,9 @@ out:
        if (buf != NULL)
                free(buf);
 
+       CT_TRACE("copied "LPU64" bytes in %f seconds",
+                length, ct_now() - start_ct_now);
+
        return rc;
 }
 
@@ -850,7 +889,7 @@ 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;
@@ -871,7 +910,7 @@ 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;
        }
@@ -1075,7 +1114,7 @@ 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;
@@ -1087,6 +1126,13 @@ 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);
@@ -1130,6 +1176,11 @@ 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
@@ -1196,7 +1247,7 @@ 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;
@@ -1375,29 +1426,53 @@ 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++;
@@ -1499,10 +1574,10 @@ 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;
@@ -1660,24 +1735,31 @@ 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: "LPX64"\n", seq);
 
        return 0;
 }
@@ -1718,8 +1800,9 @@ static int ct_run(void)
                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);
+       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;
@@ -1740,8 +1823,6 @@ 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));
@@ -1756,7 +1837,8 @@ 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)
@@ -1784,8 +1866,6 @@ static int ct_run(void)
                        hai = hai_next(hai);
                }
 
-               llapi_hsm_action_list_free(&hal);
-
                if (opt.o_abort_on_error && err_major)
                        break;
        }
@@ -1815,7 +1895,15 @@ 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;
@@ -1825,13 +1913,22 @@ 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;
@@ -1841,14 +1938,16 @@ 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:
@@ -1870,6 +1969,7 @@ int main(int argc, char **argv)
                         " rc=%d (%s)", err_major, err_minor, rc,
                         strerror(-rc));
 
+error_cleanup:
        ct_cleanup();
 
        return -rc;