Whamcloud - gitweb
LU-5710 all: second batch of corrected typos and grammar errors
[fs/lustre-release.git] / lustre / utils / lhsmtool_posix.c
index ff157dc..8eef8e6 100644 (file)
@@ -23,6 +23,7 @@
  * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
  *     alternatives
  *
+ * Copyright (c) 2013, 2014, Intel Corporation.
  */
 /* HSM copytool program for POSIX filesystem-based HSM's.
  *
 #include <stdlib.h>
 #include <dirent.h>
 #include <errno.h>
+#include <getopt.h>
+#include <pthread.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
 
-/* 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,
@@ -78,13 +81,14 @@ 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 */
@@ -115,27 +119,34 @@ 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)
 {
@@ -230,11 +241,11 @@ static int ct_parseopts(int argc, char * const *argv)
                                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);
@@ -331,11 +342,12 @@ 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;
@@ -495,38 +507,6 @@ static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
        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)
@@ -538,10 +518,13 @@ 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;
@@ -555,6 +538,14 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
                return rc;
        }
 
+       if (hai->hai_extent.offset > (__u64)src_st.st_size) {
+               rc = -EINVAL;
+               CT_ERROR(rc, "Trying to start reading past end ("LPU64" > "
+                        "%jd) of '%s' source file", hai->hai_extent.offset,
+                        (intmax_t)src_st.st_size, src);
+               return rc;
+       }
+
        if (fstat(dst_fd, &dst_st) < 0) {
                rc = -errno;
                CT_ERROR(rc, "cannot stat '%s'", dst);
@@ -567,17 +558,11 @@ 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) {
-               rc = -errno;
-               CT_ERROR(rc,
-                        "cannot seek for read to "LPU64" (len %jd) in '%s'",
-                        hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
-               return rc;
-       }
-
        /* Don't read beyond a given extent */
-       length = min(hai->hai_extent.length, src_st.st_size);
+       length = min(hai->hai_extent.length,
+                    src_st.st_size - hai->hai_extent.offset);
+
+       start_time = last_bw_print = last_report_time = time(NULL);
 
        he.offset = offset;
        he.length = 0;
@@ -598,7 +583,8 @@ 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;
@@ -627,12 +613,50 @@ 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;
+
+                               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;
+                               }
+                       }
+               }
 
-               if (time(0) >= last_print_time + opt.o_report_int) {
-                       last_print_time = time(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);
@@ -671,6 +695,9 @@ out:
        if (buf != NULL)
                free(buf);
 
+       CT_TRACE("copied "LPU64" bytes in %f seconds",
+                length, ct_now() - start_ct_now);
+
        return rc;
 }
 
@@ -885,7 +912,7 @@ static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
                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;
 
@@ -1084,7 +1111,7 @@ 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;
@@ -1096,6 +1123,13 @@ 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);
@@ -1131,7 +1165,7 @@ 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);
@@ -1616,7 +1650,7 @@ 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;
@@ -1735,6 +1769,10 @@ 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);
 }
 
@@ -1861,6 +1899,14 @@ static int ct_setup(void)
                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;
 }
 
@@ -1868,13 +1914,22 @@ 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;