X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Futils%2Flhsmtool_posix.c;h=d99fce3c87512f9fc6fac8bb902dc711a9342744;hp=dd85559bc51c7fb4551970f3c99a15a608a4957a;hb=5fbb786b96b7db3a977dd8a8fca9a40f52f66b82;hpb=062275772df6ba9027710048f2c1ee26e5574a51 diff --git a/lustre/utils/lhsmtool_posix.c b/lustre/utils/lhsmtool_posix.c index dd85559..d99fce3 100644 --- a/lustre/utils/lhsmtool_posix.c +++ b/lustre/utils/lhsmtool_posix.c @@ -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. * @@ -36,15 +37,26 @@ #ifndef _GNU_SOURCE #define _GNU_SOURCE #endif +#include +#include #include #include #include +#include #include #include +#include +#include +#include +#include #include +#include +#include #include #include #include + +#include #include #include @@ -57,11 +69,9 @@ #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, @@ -78,7 +88,7 @@ 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; @@ -116,27 +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) { @@ -231,11 +248,11 @@ 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); @@ -337,7 +354,7 @@ static int ct_parseopts(int argc, char * const *argv) 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; @@ -497,38 +514,6 @@ 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) @@ -539,11 +524,14 @@ static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src, struct stat dst_st; char *buf = NULL; __u64 write_total = 0; - __u64 length; - time_t last_print_time = time(NULL); + __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); + 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; @@ -557,6 +545,15 @@ 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 (%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) { rc = -errno; CT_ERROR(rc, "cannot stat '%s'", dst); @@ -569,17 +566,11 @@ 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); + 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 = offset; he.length = 0; @@ -600,7 +591,8 @@ 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 %ju bytes from '%s' to '%s'", + (uintmax_t)length, src, dst); while (write_total < length) { ssize_t rsize; @@ -629,14 +621,54 @@ 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); - CT_TRACE("%%"LPU64" ", 100 * write_total / length); - he.length = write_total; + 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 @@ -645,6 +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; } @@ -673,6 +706,9 @@ out: if (buf != NULL) free(buf); + CT_TRACE("copied %ju bytes in %f seconds", + (uintmax_t)length, ct_now() - start_ct_now); + return rc; } @@ -805,8 +841,8 @@ static int ct_fini(struct hsm_copyaction_private **phcp, 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); @@ -824,8 +860,8 @@ static int ct_fini(struct hsm_copyaction_private **phcp, 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 @@ -887,7 +923,7 @@ 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; @@ -1140,7 +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); @@ -1246,9 +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) @@ -1625,7 +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; @@ -1731,7 +1767,7 @@ static int ct_max_sequence(void) path[sizeof(path) - 1] = '\0'; } - printf("max_sequence: "LPX64"\n", seq); + printf("max_sequence: %#jx\n", (uintmax_t)seq); return 0; } @@ -1744,13 +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); @@ -1780,14 +1821,17 @@ static int ct_run(void) 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");