* (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
* alternatives
*
+ * Copyright (c) 2013, 2016, Intel Corporation.
+ *
* All rights reserved. This program and the accompanying materials
* are made available under the terms of the GNU Lesser General Public License
* (LGPL) version 2.1 or (at your discretion) any later version.
* Author: Henri Doreau <henri.doreau@cea.fr>
*/
+#include <fcntl.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <dirent.h>
#include <stdarg.h>
#include <sys/stat.h>
+#include <sys/time.h>
#include <sys/types.h>
+#include <time.h>
#include <utime.h>
#include <sys/syscall.h>
#include <fnmatch.h>
-#include <glob.h>
#include <signal.h>
#ifdef HAVE_LINUX_UNISTD_H
#include <linux/unistd.h>
#include <unistd.h>
#endif
-#include <libcfs/libcfs.h>
#include <lnet/lnetctl.h>
#include <lustre/lustre_idl.h>
#include <lustre/lustreapi.h>
struct kuc_hdr *kuch;
int mnt_fd;
int open_by_fid_fd;
- lustre_kernelcomm kuc;
+ struct lustre_kernelcomm kuc;
__u32 archives;
};
__s32 data_fd;
const struct hsm_copytool_private *ct_priv;
struct hsm_copy copy;
- struct stat stat;
+ lstat_t stat;
};
-#include <libcfs/libcfs.h>
-
enum ct_progress_type {
CT_START = 0,
CT_RUNNING = 50,
};
/* initialized in llapi_hsm_register_event_fifo() */
-int llapi_hsm_event_fd = -1;
+static int llapi_hsm_event_fd = -1;
+static bool created_hsm_event_fifo;
static inline const char *llapi_hsm_ct_ev2str(int type)
{
* \retval 0 on success.
* \retval -errno on error.
*/
-int llapi_hsm_write_json_event(struct llapi_json_item_list **event)
+static int llapi_hsm_write_json_event(struct llapi_json_item_list **event)
{
int rc;
char time_string[40];
* \retval 0 on success.
* \retval -errno on error.
*/
-int llapi_hsm_log_ct_registration(struct hsm_copytool_private **priv,
- __u32 event_type)
+static int llapi_hsm_log_ct_registration(struct hsm_copytool_private **priv,
+ __u32 event_type)
{
int rc;
char agent_uuid[UUID_MAX];
struct hsm_copytool_private *ct;
struct llapi_json_item_list *json_items;
+ /* Noop unless the event fd was initialized */
+ if (llapi_hsm_event_fd < 0)
+ return 0;
+
if (priv == NULL || *priv == NULL)
return -EINVAL;
* \retval 0 on success.
* \retval -errno on error.
*/
-int llapi_hsm_log_ct_progress(struct hsm_copyaction_private **phcp,
- const struct hsm_action_item *hai, __u32 progress_type,
- __u64 total, __u64 current)
+static int llapi_hsm_log_ct_progress(struct hsm_copyaction_private **phcp,
+ const struct hsm_action_item *hai,
+ __u32 progress_type,
+ __u64 total, __u64 current)
{
int rc;
int linkno = 0;
struct hsm_copyaction_private *hcp;
struct llapi_json_item_list *json_items;
+ /* Noop unless the event fd was initialized */
+ if (llapi_hsm_event_fd < 0)
+ return 0;
+
if (phcp == NULL || *phcp == NULL)
return -EINVAL;
goto err;
}
- if (progress_type == CT_RUNNING)
+ if (progress_type == CT_RUNNING) {
rc = llapi_json_add_item(&json_items, "current_bytes",
LLAPI_JSON_BIGNUM, ¤t);
if (rc < 0)
goto err;
+ }
cancel:
rc = llapi_json_add_item(&json_items, "event_type", LLAPI_JSON_STRING,
* \retval 0 on success.
* \retval -errno on error.
*/
-int llapi_hsm_register_event_fifo(char *path)
+int llapi_hsm_register_event_fifo(const char *path)
{
int read_fd;
struct stat statbuf;
+ struct sigaction ignore_action;
+ int rc;
/* Create the FIFO if necessary. */
if ((mkfifo(path, 0644) < 0) && (errno != EEXIST)) {
"not a pipe or has a wrong mode", path);
return -errno;
}
+ } else {
+ created_hsm_event_fifo = true;
}
/* Open the FIFO for read so that the subsequent open for write
/* Open the FIFO for writes, but don't block on waiting
* for a reader. */
llapi_hsm_event_fd = open(path, O_WRONLY | O_NONBLOCK);
- if (llapi_hsm_event_fd < 0) {
- llapi_error(LLAPI_MSG_ERROR, errno,
- "cannot open(%s) for write", path);
- return -errno;
- }
+ rc = -errno;
/* Now close the reader. An external monitoring process can
* now open the FIFO for reads. If no reader comes along the
* events are lost. NOTE: Only one reader at a time! */
close(read_fd);
+ if (llapi_hsm_event_fd < 0) {
+ llapi_error(LLAPI_MSG_ERROR, -rc,
+ "cannot open(%s) for write", path);
+ return rc;
+ }
+
/* Ignore SIGPIPEs -- can occur if the reader goes away. */
- signal(SIGPIPE, SIG_IGN);
+ memset(&ignore_action, 0, sizeof(ignore_action));
+ ignore_action.sa_handler = SIG_IGN;
+ sigemptyset(&ignore_action.sa_mask);
+ sigaction(SIGPIPE, &ignore_action, NULL);
return 0;
}
* \retval 0 on success.
* \retval -errno on error.
*/
-int llapi_hsm_unregister_event_fifo(char *path)
+int llapi_hsm_unregister_event_fifo(const char *path)
{
/* Noop unless the event fd was initialized */
if (llapi_hsm_event_fd < 0)
if (close(llapi_hsm_event_fd) < 0)
return -errno;
- unlink(path);
+ if (created_hsm_event_fifo) {
+ unlink(path);
+ created_hsm_event_fifo = false;
+ }
llapi_hsm_event_fd = -1;
}
/** Register a copytool
- * \param[out] priv Opaque private control structure
- * \param mnt Lustre filesystem mount point
- * \param flags Open flags, currently unused (e.g. O_NONBLOCK)
- * \param archive_count
- * \param archives Which archive numbers this copytool is responsible for
+ * \param[out] priv Opaque private control structure
+ * \param mnt Lustre filesystem mount point
+ * \param archive_count Number of valid archive IDs in \a archives
+ * \param archives Which archive numbers this copytool is
+ * responsible for
+ * \param rfd_flags flags applied to read fd of pipe
+ * (e.g. O_NONBLOCK)
+ *
+ * \retval 0 on success.
+ * \retval -errno on error.
*/
int llapi_hsm_copytool_register(struct hsm_copytool_private **priv,
- const char *mnt, int flags, int archive_count,
- int *archives)
+ const char *mnt, int archive_count,
+ int *archives, int rfd_flags)
{
struct hsm_copytool_private *ct;
int rc;
return -EINVAL;
}
+ if (archive_count > LL_HSM_MAX_ARCHIVE) {
+ llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when maximum "
+ "of %zu archives supported", archive_count,
+ LL_HSM_MAX_ARCHIVE);
+ return -EINVAL;
+ }
+
ct = calloc(1, sizeof(*ct));
if (ct == NULL)
return -ENOMEM;
/* no archives specified means "match all". */
ct->archives = 0;
for (rc = 0; rc < archive_count; rc++) {
- if (archives[rc] > 8 * sizeof(ct->archives)) {
- llapi_err_noerrno(LLAPI_MSG_ERROR,
- "maximum of %zu archives supported",
- 8 * sizeof(ct->archives));
+ if ((archives[rc] > LL_HSM_MAX_ARCHIVE) || (archives[rc] < 0)) {
+ llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when "
+ "archive id [0 - %zu] is supported",
+ archives[rc], LL_HSM_MAX_ARCHIVE);
+ rc = -EINVAL;
goto out_err;
}
- /* in the list we have a all archive wildcard
+ /* in the list we have an all archive wildcard
* so move to all archives mode
*/
if (archives[rc] == 0) {
ct->archives |= (1 << (archives[rc] - 1));
}
- rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM);
+ rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM, rfd_flags);
if (rc < 0)
goto out_err;
rc = -errno;
llapi_error(LLAPI_MSG_ERROR, rc,
"cannot start copytool on '%s'", mnt);
- goto out_err;
- } else {
- rc = 0;
+ goto out_kuc;
}
llapi_hsm_log_ct_registration(&ct, CT_REGISTER);
/* Only the kernel reference keeps the write side open */
close(ct->kuc.lk_wfd);
ct->kuc.lk_wfd = LK_NOFD;
- if (rc < 0)
- goto out_kuc;
-
*priv = ct;
+
return 0;
out_kuc:
if (ct->magic != CT_PRIV_MAGIC)
return -EINVAL;
+ /* Close the read side of the KUC pipe. This should be done
+ * before unregistering to avoid deadlock: a ldlm_cb thread
+ * enters libcfs_kkuc_group_put() acquires kg_sem and blocks
+ * in pipe_write() due to full pipe; then we attempt to
+ * unregister and block on kg_sem. */
+ libcfs_ukuc_stop(&ct->kuc);
+
/* Tell the kernel to stop sending us messages */
ct->kuc.lk_flags = LK_FLG_STOP;
ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, &ct->kuc);
- /* Shut down the kernelcomms */
- libcfs_ukuc_stop(&ct->kuc);
-
llapi_hsm_log_ct_registration(&ct, CT_UNREGISTER);
close(ct->open_by_fid_fd);
return 0;
}
+/** Returns a file descriptor to poll/select on.
+ * \param ct Opaque private control structure
+ * \retval -EINVAL on error
+ * \retval the file descriptor for reading HSM events from the kernel
+ */
+int llapi_hsm_copytool_get_fd(struct hsm_copytool_private *ct)
+{
+ if (ct == NULL || ct->magic != CT_PRIV_MAGIC)
+ return -EINVAL;
+
+ return libcfs_ukuc_get_rfd(&ct->kuc);
+}
+
/** Wait for the next hsm_action_list
* \param ct Opaque private control structure
* \param halh Action list handle, will be allocated here
kuch = ct->kuch;
+repeat:
rc = libcfs_ukuc_msg_get(&ct->kuc, (char *)kuch,
HAL_MAXSIZE + sizeof(*kuch),
KUC_TRANSPORT_HSM);
" ignoring this request."
" Mask of served archive is 0x%.8X",
hal->hal_archive_id, ct->archives);
- rc = -EAGAIN;
- goto out_err;
+ goto repeat;
}
*halh = hal;
const struct lu_fid *fid, int open_flags)
{
char fid_name[FID_NOBRACE_LEN + 1];
+ int fd;
snprintf(fid_name, sizeof(fid_name), DFID_NOBRACE, PFID(fid));
- return openat(ct->open_by_fid_fd, fid_name, open_flags);
+ fd = openat(ct->open_by_fid_fd, fid_name, open_flags);
+ return fd < 0 ? -errno : fd;
}
-static int ct_stat_by_fid(const struct hsm_copytool_private *ct,
- const struct lu_fid *fid,
- struct stat *buf)
+/**
+ * Get metadata attributes of file by FID.
+ *
+ * Use the IOC_MDC_GETFILEINFO ioctl (to send a MDS_GETATTR_NAME RPC)
+ * to get the attributes of the file identified by \a fid. This
+ * returns only the attributes stored on the MDT and avoids taking
+ * layout locks or accessing OST objects. It also bypasses the inode
+ * cache. Attributes are returned in \a st.
+ */
+static int ct_md_getattr(const struct hsm_copytool_private *ct,
+ const struct lu_fid *fid,
+ lstat_t *st)
{
- char fid_name[FID_NOBRACE_LEN + 1];
+ struct lov_user_mds_data *lmd;
+ size_t lmd_size;
+ int rc;
- snprintf(fid_name, sizeof(fid_name), DFID_NOBRACE, PFID(fid));
+ lmd_size = sizeof(lmd->lmd_st) +
+ lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3);
+
+ if (lmd_size < sizeof(lmd->lmd_st) + XATTR_SIZE_MAX)
+ lmd_size = sizeof(lmd->lmd_st) + XATTR_SIZE_MAX;
+
+ if (lmd_size < FID_NOBRACE_LEN + 1)
+ lmd_size = FID_NOBRACE_LEN + 1;
+
+ lmd = malloc(lmd_size);
+ if (lmd == NULL)
+ return -ENOMEM;
+
+ snprintf((char *)lmd, lmd_size, DFID_NOBRACE, PFID(fid));
+
+ rc = ioctl(ct->open_by_fid_fd, IOC_MDC_GETFILEINFO, lmd);
+ if (rc != 0) {
+ rc = -errno;
+ llapi_error(LLAPI_MSG_ERROR, rc,
+ "cannot get metadata attributes of "DFID" in '%s'",
+ PFID(fid), ct->mnt);
+ goto out;
+ }
+
+ *st = lmd->lmd_st;
+out:
+ free(lmd);
- return fstatat(ct->open_by_fid_fd, fid_name, buf, 0);
+ return rc;
}
/** Create the destination volatile file for a restore operation.
goto ok_out;
if (hai->hai_action == HSMA_RESTORE) {
- rc = ct_stat_by_fid(hcp->ct_priv, &hai->hai_fid, &hcp->stat);
+ rc = ct_md_getattr(hcp->ct_priv, &hai->hai_fid, &hcp->stat);
if (rc < 0)
goto err_out;
hai = &hcp->copy.hc_hai;
if (hai->hai_action == HSMA_RESTORE && errval == 0) {
- struct timeval tv[2];
-
- /* Set {a,m}time of volatile file to that of original. */
- tv[0].tv_sec = hcp->stat.st_atime;
- tv[0].tv_usec = 0;
- tv[1].tv_sec = hcp->stat.st_mtime;
- tv[1].tv_usec = 0;
- if (futimes(hcp->data_fd, tv) < 0) {
+ struct ll_futimes_3 lfu = {
+ .lfu_atime_sec = hcp->stat.st_atim.tv_sec,
+ .lfu_atime_nsec = hcp->stat.st_atim.tv_nsec,
+ .lfu_mtime_sec = hcp->stat.st_mtim.tv_sec,
+ .lfu_mtime_nsec = hcp->stat.st_mtim.tv_nsec,
+ .lfu_ctime_sec = hcp->stat.st_ctim.tv_sec,
+ .lfu_ctime_nsec = hcp->stat.st_ctim.tv_nsec,
+ };
+
+ rc = fsync(hcp->data_fd);
+ if (rc < 0) {
errval = -errno;
goto end;
}
- rc = fsync(hcp->data_fd);
+ /* Set {a,m,c}time of volatile file to that of original. */
+ rc = ioctl(hcp->data_fd, LL_IOC_FUTIMES_3, &lfu);
if (rc < 0) {
errval = -errno;
goto end;
int llapi_hsm_action_get_fd(const struct hsm_copyaction_private *hcp)
{
const struct hsm_action_item *hai = &hcp->copy.hc_hai;
+ int fd;
if (hcp->magic != CP_PRIV_MAGIC)
return -EINVAL;
- if (hai->hai_action == HSMA_ARCHIVE)
+ if (hai->hai_action == HSMA_ARCHIVE) {
return ct_open_by_fid(hcp->ct_priv, &hai->hai_dfid,
O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NONBLOCK);
- else if (hai->hai_action == HSMA_RESTORE)
- return dup(hcp->data_fd);
- else
+ } else if (hai->hai_action == HSMA_RESTORE) {
+ fd = dup(hcp->data_fd);
+ return fd < 0 ? -errno : fd;
+ } else {
return -EINVAL;
+ }
}
/**
stripe_pattern | LOV_PATTERN_F_RELEASED,
pool_name);
if (fd < 0) {
- llapi_error(LLAPI_MSG_ERROR, -errno,
+ llapi_error(LLAPI_MSG_ERROR, fd,
"cannot create '%s' for import", dst);
- return -errno;
+ return fd;
}
/* Get the new fid in Lustre. Caller needs to use this fid
hui.hui_mtime_ns = st->st_mtim.tv_nsec;
rc = ioctl(fd, LL_IOC_HSM_IMPORT, &hui);
if (rc != 0) {
- llapi_error(LLAPI_MSG_ERROR, rc, "cannot import '%s'", dst);
rc = -errno;
+ llapi_error(LLAPI_MSG_ERROR, rc, "cannot import '%s'", dst);
goto out_unlink;
}