4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
9 * Copyright (c) 2013, 2016, Intel Corporation.
11 * All rights reserved. This program and the accompanying materials
12 * are made available under the terms of the GNU Lesser General Public License
13 * (LGPL) version 2.1 or (at your discretion) any later version.
14 * (LGPL) version 2.1 accompanies this distribution, and is available at
15 * http://www.gnu.org/licenses/lgpl-2.1.html
17 * This library is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
25 * lustre/utils/liblustreapi_hsm.c
27 * lustreapi library for hsm calls
29 * Author: Aurelien Degremont <aurelien.degremont@cea.fr>
30 * Author: JC Lafoucriere <jacques-charles.lafoucriere@cea.fr>
31 * Author: Thomas Leibovici <thomas.leibovici@cea.fr>
32 * Author: Henri Doreau <henri.doreau@cea.fr>
40 #include <sys/ioctl.h>
48 #include <sys/types.h>
51 #include <sys/syscall.h>
54 #ifdef HAVE_LINUX_UNISTD_H
55 #include <linux/unistd.h>
60 #include <lnet/lnetctl.h>
61 #include <lustre/lustre_idl.h>
62 #include <lustre/lustreapi.h>
63 #include "lustreapi_internal.h"
65 #define OPEN_BY_FID_PATH dot_lustre_name"/fid"
67 /****** HSM Copytool API ********/
68 #define CT_PRIV_MAGIC 0xC0BE2001
69 struct hsm_copytool_private {
75 struct lustre_kernelcomm kuc;
79 #define CP_PRIV_MAGIC 0x19880429
80 struct hsm_copyaction_private {
84 const struct hsm_copytool_private *ct_priv;
89 enum ct_progress_type {
100 CT_ARCHIVE_START = HSMA_ARCHIVE,
101 CT_ARCHIVE_RUNNING = HSMA_ARCHIVE + CT_RUNNING,
102 CT_ARCHIVE_FINISH = HSMA_ARCHIVE + CT_FINISH,
103 CT_ARCHIVE_CANCEL = HSMA_ARCHIVE + CT_CANCEL,
104 CT_ARCHIVE_ERROR = HSMA_ARCHIVE + CT_ERROR,
105 CT_RESTORE_START = HSMA_RESTORE,
106 CT_RESTORE_RUNNING = HSMA_RESTORE + CT_RUNNING,
107 CT_RESTORE_FINISH = HSMA_RESTORE + CT_FINISH,
108 CT_RESTORE_CANCEL = HSMA_RESTORE + CT_CANCEL,
109 CT_RESTORE_ERROR = HSMA_RESTORE + CT_ERROR,
110 CT_REMOVE_START = HSMA_REMOVE,
111 CT_REMOVE_RUNNING = HSMA_REMOVE + CT_RUNNING,
112 CT_REMOVE_FINISH = HSMA_REMOVE + CT_FINISH,
113 CT_REMOVE_CANCEL = HSMA_REMOVE + CT_CANCEL,
114 CT_REMOVE_ERROR = HSMA_REMOVE + CT_ERROR,
118 /* initialized in llapi_hsm_register_event_fifo() */
119 static int llapi_hsm_event_fd = -1;
120 static bool created_hsm_event_fifo;
122 static inline const char *llapi_hsm_ct_ev2str(int type)
129 case CT_ARCHIVE_START:
130 return "ARCHIVE_START";
131 case CT_ARCHIVE_RUNNING:
132 return "ARCHIVE_RUNNING";
133 case CT_ARCHIVE_FINISH:
134 return "ARCHIVE_FINISH";
135 case CT_ARCHIVE_CANCEL:
136 return "ARCHIVE_CANCEL";
137 case CT_ARCHIVE_ERROR:
138 return "ARCHIVE_ERROR";
139 case CT_RESTORE_START:
140 return "RESTORE_START";
141 case CT_RESTORE_RUNNING:
142 return "RESTORE_RUNNING";
143 case CT_RESTORE_FINISH:
144 return "RESTORE_FINISH";
145 case CT_RESTORE_CANCEL:
146 return "RESTORE_CANCEL";
147 case CT_RESTORE_ERROR:
148 return "RESTORE_ERROR";
149 case CT_REMOVE_START:
150 return "REMOVE_START";
151 case CT_REMOVE_RUNNING:
152 return "REMOVE_RUNNING";
153 case CT_REMOVE_FINISH:
154 return "REMOVE_FINISH";
155 case CT_REMOVE_CANCEL:
156 return "REMOVE_CANCEL";
157 case CT_REMOVE_ERROR:
158 return "REMOVE_ERROR";
160 llapi_err_noerrno(LLAPI_MSG_ERROR,
161 "Unknown event type: %d", type);
167 * Writes a JSON event to the monitor FIFO. Noop if no FIFO has been
170 * \param event A list of llapi_json_items comprising a
171 * single JSON-formatted event.
173 * \retval 0 on success.
174 * \retval -errno on error.
176 static int llapi_hsm_write_json_event(struct llapi_json_item_list **event)
179 char time_string[40];
180 char json_buf[PIPE_BUF];
182 time_t event_time = time(0);
183 struct tm time_components;
184 struct llapi_json_item_list *json_items;
186 /* Noop unless the event fd was initialized */
187 if (llapi_hsm_event_fd < 0)
190 if (event == NULL || *event == NULL)
195 localtime_r(&event_time, &time_components);
197 if (strftime(time_string, sizeof(time_string), "%Y-%m-%d %T %z",
198 &time_components) == 0) {
200 llapi_error(LLAPI_MSG_ERROR, rc, "strftime() failed");
204 rc = llapi_json_add_item(&json_items, "event_time", LLAPI_JSON_STRING,
207 llapi_error(LLAPI_MSG_ERROR, -rc, "error in "
208 "llapi_json_add_item()");
212 buf_file = fmemopen(json_buf, sizeof(json_buf), "w");
213 if (buf_file == NULL)
216 rc = llapi_json_write_list(event, buf_file);
224 if (write(llapi_hsm_event_fd, json_buf, strlen(json_buf)) < 0) {
225 /* Ignore write failures due to missing reader. */
234 * Hook for llapi_hsm_copytool_register and llapi_hsm_copytool_unregister
235 * to generate JSON events suitable for consumption by a copytool
236 * monitoring process.
238 * \param priv Opaque private control structure.
239 * \param event_type The type of event (register or unregister).
241 * \retval 0 on success.
242 * \retval -errno on error.
244 static int llapi_hsm_log_ct_registration(struct hsm_copytool_private **priv,
248 char agent_uuid[UUID_MAX];
249 struct hsm_copytool_private *ct;
250 struct llapi_json_item_list *json_items;
252 /* Noop unless the event fd was initialized */
253 if (llapi_hsm_event_fd < 0)
256 if (priv == NULL || *priv == NULL)
260 if (ct->magic != CT_PRIV_MAGIC)
263 if (event_type != CT_REGISTER && event_type != CT_UNREGISTER)
266 rc = llapi_json_init_list(&json_items);
270 rc = llapi_get_agent_uuid(ct->mnt, agent_uuid, sizeof(agent_uuid));
273 llapi_chomp_string(agent_uuid);
275 rc = llapi_json_add_item(&json_items, "uuid", LLAPI_JSON_STRING,
280 rc = llapi_json_add_item(&json_items, "mount_point", LLAPI_JSON_STRING,
285 rc = llapi_json_add_item(&json_items, "archive", LLAPI_JSON_INTEGER,
290 rc = llapi_json_add_item(&json_items, "event_type", LLAPI_JSON_STRING,
291 (char *)llapi_hsm_ct_ev2str(event_type));
295 rc = llapi_hsm_write_json_event(&json_items);
302 llapi_error(LLAPI_MSG_ERROR, rc, "error in "
303 "llapi_hsm_log_ct_registration()");
306 if (json_items != NULL)
307 llapi_json_destroy_list(&json_items);
313 * Given a copytool progress update, construct a JSON event suitable for
314 * consumption by a copytool monitoring process.
316 * Examples of various events generated here and written by
317 * llapi_hsm_write_json_event:
319 * Copytool registration and deregistration:
320 * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "REGISTER",
321 * "archive": 0, "mount_point": "/mnt/lustre",
322 * "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
323 * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "UNREGISTER",
324 * "archive": 0, "mount_point": "/mnt/lustre",
325 * "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
327 * An archive action, start to completion:
328 * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "ARCHIVE_START",
329 * "total_bytes": 0, "lustre_path": "d71.sanity-hsm/f71.sanity-hsm",
330 * "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
331 * {"event_time": "2014-02-26 14:50:18 -0500", "event_type": "ARCHIVE_RUNNING",
332 * "current_bytes": 5242880, "total_bytes": 39000000,
333 * "lustre_path": "d71.sanity-hsm/f71.sanity-hsm",
334 * "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
335 * {"event_time": "2014-02-26 14:50:50 -0500", "event_type": "ARCHIVE_FINISH",
336 * "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
339 * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "LOGGED_MESSAGE",
341 * "message": "lhsmtool_posix[42]: copytool fs=lustre archive#=2 item_count=1"}
343 * \param hcp Opaque action handle returned by
344 * llapi_hsm_action_start.
345 * \param hai The hsm_action_item describing the request.
346 * \param progress_type The ct_progress_type describing the update.
347 * \param total The total expected bytes for the request.
348 * \param current The current copied byte count for the request.
350 * \retval 0 on success.
351 * \retval -errno on error.
353 static int llapi_hsm_log_ct_progress(struct hsm_copyaction_private **phcp,
354 const struct hsm_action_item *hai,
356 __u64 total, __u64 current)
360 long long recno = -1;
361 char lustre_path[PATH_MAX];
362 char strfid[FID_NOBRACE_LEN + 1];
363 struct hsm_copyaction_private *hcp;
364 struct llapi_json_item_list *json_items;
366 /* Noop unless the event fd was initialized */
367 if (llapi_hsm_event_fd < 0)
370 if (phcp == NULL || *phcp == NULL)
375 rc = llapi_json_init_list(&json_items);
379 snprintf(strfid, sizeof(strfid), DFID_NOBRACE, PFID(&hai->hai_dfid));
380 rc = llapi_json_add_item(&json_items, "data_fid",
381 LLAPI_JSON_STRING, strfid);
385 snprintf(strfid, sizeof(strfid), DFID_NOBRACE, PFID(&hai->hai_fid));
386 rc = llapi_json_add_item(&json_items, "source_fid",
387 LLAPI_JSON_STRING, strfid);
391 if (hcp->copy.hc_errval == ECANCELED) {
392 progress_type = CT_CANCEL;
396 if (hcp->copy.hc_errval != 0) {
397 progress_type = CT_ERROR;
399 rc = llapi_json_add_item(&json_items, "errno",
401 &hcp->copy.hc_errval);
405 rc = llapi_json_add_item(&json_items, "error",
407 strerror(hcp->copy.hc_errval));
414 /* lustre_path isn't available after a restore completes */
415 /* total_bytes isn't available after a restore or archive completes */
416 if (progress_type != CT_FINISH) {
417 rc = llapi_fid2path(hcp->ct_priv->mnt, strfid, lustre_path,
418 sizeof(lustre_path), &recno, &linkno);
422 rc = llapi_json_add_item(&json_items, "lustre_path",
423 LLAPI_JSON_STRING, lustre_path);
427 rc = llapi_json_add_item(&json_items, "total_bytes",
428 LLAPI_JSON_BIGNUM, &total);
433 if (progress_type == CT_RUNNING) {
434 rc = llapi_json_add_item(&json_items, "current_bytes",
435 LLAPI_JSON_BIGNUM, ¤t);
441 rc = llapi_json_add_item(&json_items, "event_type", LLAPI_JSON_STRING,
442 (char *)llapi_hsm_ct_ev2str(hai->hai_action +
447 rc = llapi_hsm_write_json_event(&json_items);
454 llapi_error(LLAPI_MSG_ERROR, rc, "error in "
455 "llapi_hsm_log_ct_progress()");
458 if (json_items != NULL)
459 llapi_json_destroy_list(&json_items);
465 * Given a path to a FIFO, create a filehandle for nonblocking writes to it.
466 * Intended to be used for copytool monitoring processes that read an
467 * event stream from the FIFO. Events written in the absence of a reader
470 * \param path Path to monitor FIFO.
472 * \retval 0 on success.
473 * \retval -errno on error.
475 int llapi_hsm_register_event_fifo(const char *path)
479 struct sigaction ignore_action;
482 /* Create the FIFO if necessary. */
483 if ((mkfifo(path, 0644) < 0) && (errno != EEXIST)) {
484 llapi_error(LLAPI_MSG_ERROR, errno, "mkfifo(%s) failed", path);
487 if (errno == EEXIST) {
488 if (stat(path, &statbuf) < 0) {
489 llapi_error(LLAPI_MSG_ERROR, errno, "mkfifo(%s) failed",
493 if (!S_ISFIFO(statbuf.st_mode) ||
494 ((statbuf.st_mode & 0777) != 0644)) {
495 llapi_error(LLAPI_MSG_ERROR, errno, "%s exists but is "
496 "not a pipe or has a wrong mode", path);
500 created_hsm_event_fifo = true;
503 /* Open the FIFO for read so that the subsequent open for write
504 * doesn't immediately fail. */
505 read_fd = open(path, O_RDONLY | O_NONBLOCK);
507 llapi_error(LLAPI_MSG_ERROR, errno,
508 "cannot open(%s) for read", path);
512 /* Open the FIFO for writes, but don't block on waiting
514 llapi_hsm_event_fd = open(path, O_WRONLY | O_NONBLOCK);
517 /* Now close the reader. An external monitoring process can
518 * now open the FIFO for reads. If no reader comes along the
519 * events are lost. NOTE: Only one reader at a time! */
522 if (llapi_hsm_event_fd < 0) {
523 llapi_error(LLAPI_MSG_ERROR, -rc,
524 "cannot open(%s) for write", path);
528 /* Ignore SIGPIPEs -- can occur if the reader goes away. */
529 memset(&ignore_action, 0, sizeof(ignore_action));
530 ignore_action.sa_handler = SIG_IGN;
531 sigemptyset(&ignore_action.sa_mask);
532 sigaction(SIGPIPE, &ignore_action, NULL);
538 * Given a path to a FIFO, close its filehandle and delete the FIFO.
540 * \param path Path to monitor FIFO.
542 * \retval 0 on success.
543 * \retval -errno on error.
545 int llapi_hsm_unregister_event_fifo(const char *path)
547 /* Noop unless the event fd was initialized */
548 if (llapi_hsm_event_fd < 0)
551 if (close(llapi_hsm_event_fd) < 0)
554 if (created_hsm_event_fifo) {
556 created_hsm_event_fifo = false;
559 llapi_hsm_event_fd = -1;
565 * Custom logging callback to be used when a monitoring FIFO has been
566 * registered. Formats log entries as JSON events suitable for
567 * consumption by a copytool monitoring process.
569 * \param level The message loglevel.
570 * \param _rc The returncode associated with the message.
571 * \param fmt The message format string.
572 * \param args Arguments to be formatted by the format string.
576 void llapi_hsm_log_error(enum llapi_message_level level, int _rc,
577 const char *fmt, va_list args)
584 struct llapi_json_item_list *json_items;
586 /* Noop unless the event fd was initialized */
587 if (llapi_hsm_event_fd < 0)
590 rc = llapi_json_init_list(&json_items);
594 if ((level & LLAPI_MSG_NO_ERRNO) == 0) {
595 rc = llapi_json_add_item(&json_items, "errno",
601 rc = llapi_json_add_item(&json_items, "error",
608 va_copy(args2, args);
609 msg_len = vsnprintf(NULL, 0, fmt, args2) + 1;
612 msg = (char *) alloca(msg_len);
618 rc = vsnprintf(msg, msg_len, fmt, args);
622 rc = llapi_json_add_item(&json_items, "message",
628 rc = llapi_json_add_item(&json_items, "message",
630 "INTERNAL ERROR: message failed");
635 real_level = level & LLAPI_MSG_NO_ERRNO;
636 real_level = real_level > 0 ? level - LLAPI_MSG_NO_ERRNO : level;
638 rc = llapi_json_add_item(&json_items, "level", LLAPI_JSON_STRING,
639 (void *)llapi_msg_level2str(real_level));
643 rc = llapi_json_add_item(&json_items, "event_type", LLAPI_JSON_STRING,
648 rc = llapi_hsm_write_json_event(&json_items);
655 /* Write directly to stderr to avoid llapi_error, which now
656 * emits JSON event messages. */
657 fprintf(stderr, "\nFATAL ERROR IN llapi_hsm_log_error(): rc %d,", rc);
660 if (json_items != NULL)
661 llapi_json_destroy_list(&json_items);
666 /** Register a copytool
667 * \param[out] priv Opaque private control structure
668 * \param mnt Lustre filesystem mount point
669 * \param archive_count Number of valid archive IDs in \a archives
670 * \param archives Which archive numbers this copytool is
672 * \param rfd_flags flags applied to read fd of pipe
675 * \retval 0 on success.
676 * \retval -errno on error.
678 int llapi_hsm_copytool_register(struct hsm_copytool_private **priv,
679 const char *mnt, int archive_count,
680 int *archives, int rfd_flags)
682 struct hsm_copytool_private *ct;
685 if (archive_count > 0 && archives == NULL) {
686 llapi_err_noerrno(LLAPI_MSG_ERROR,
687 "NULL archive numbers");
691 if (archive_count > LL_HSM_MAX_ARCHIVE) {
692 llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when maximum "
693 "of %zu archives supported", archive_count,
698 ct = calloc(1, sizeof(*ct));
702 ct->magic = CT_PRIV_MAGIC;
704 ct->open_by_fid_fd = -1;
705 ct->kuc.lk_rfd = LK_NOFD;
706 ct->kuc.lk_wfd = LK_NOFD;
708 ct->mnt = strdup(mnt);
709 if (ct->mnt == NULL) {
714 ct->kuch = malloc(HAL_MAXSIZE + sizeof(*ct->kuch));
715 if (ct->kuch == NULL) {
720 ct->mnt_fd = open(ct->mnt, O_RDONLY);
721 if (ct->mnt_fd < 0) {
726 ct->open_by_fid_fd = openat(ct->mnt_fd, OPEN_BY_FID_PATH, O_RDONLY);
727 if (ct->open_by_fid_fd < 0) {
732 /* no archives specified means "match all". */
734 for (rc = 0; rc < archive_count; rc++) {
735 if ((archives[rc] > LL_HSM_MAX_ARCHIVE) || (archives[rc] < 0)) {
736 llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when "
737 "archive id [0 - %zu] is supported",
738 archives[rc], LL_HSM_MAX_ARCHIVE);
742 /* in the list we have an all archive wildcard
743 * so move to all archives mode
745 if (archives[rc] == 0) {
750 ct->archives |= (1 << (archives[rc] - 1));
753 rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM, rfd_flags);
757 /* Storing archive(s) in lk_data; see mdc_ioc_hsm_ct_start */
758 ct->kuc.lk_data = ct->archives;
759 rc = ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, &ct->kuc);
762 llapi_error(LLAPI_MSG_ERROR, rc,
763 "cannot start copytool on '%s'", mnt);
767 llapi_hsm_log_ct_registration(&ct, CT_REGISTER);
769 /* Only the kernel reference keeps the write side open */
770 close(ct->kuc.lk_wfd);
771 ct->kuc.lk_wfd = LK_NOFD;
777 /* cleanup the kuc channel */
778 libcfs_ukuc_stop(&ct->kuc);
781 if (!(ct->mnt_fd < 0))
784 if (!(ct->open_by_fid_fd < 0))
785 close(ct->open_by_fid_fd);
796 /** Deregister a copytool
797 * Note: under Linux, until llapi_hsm_copytool_unregister is called
798 * (or the program is killed), the libcfs module will be referenced
799 * and unremovable, even after Lustre services stop.
801 int llapi_hsm_copytool_unregister(struct hsm_copytool_private **priv)
803 struct hsm_copytool_private *ct;
805 if (priv == NULL || *priv == NULL)
809 if (ct->magic != CT_PRIV_MAGIC)
812 /* Close the read side of the KUC pipe. This should be done
813 * before unregistering to avoid deadlock: a ldlm_cb thread
814 * enters libcfs_kkuc_group_put() acquires kg_sem and blocks
815 * in pipe_write() due to full pipe; then we attempt to
816 * unregister and block on kg_sem. */
817 libcfs_ukuc_stop(&ct->kuc);
819 /* Tell the kernel to stop sending us messages */
820 ct->kuc.lk_flags = LK_FLG_STOP;
821 ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, &ct->kuc);
823 llapi_hsm_log_ct_registration(&ct, CT_UNREGISTER);
825 close(ct->open_by_fid_fd);
835 /** Returns a file descriptor to poll/select on.
836 * \param ct Opaque private control structure
837 * \retval -EINVAL on error
838 * \retval the file descriptor for reading HSM events from the kernel
840 int llapi_hsm_copytool_get_fd(struct hsm_copytool_private *ct)
842 if (ct == NULL || ct->magic != CT_PRIV_MAGIC)
845 return libcfs_ukuc_get_rfd(&ct->kuc);
848 /** Wait for the next hsm_action_list
849 * \param ct Opaque private control structure
850 * \param halh Action list handle, will be allocated here
851 * \param msgsize Number of bytes in the message, will be set here
852 * \return 0 valid message received; halh and msgsize are set
854 * Note: The application must not call llapi_hsm_copytool_recv until it has
855 * cleared the data in ct->kuch from the previous call.
857 int llapi_hsm_copytool_recv(struct hsm_copytool_private *ct,
858 struct hsm_action_list **halh, int *msgsize)
860 struct kuc_hdr *kuch;
861 struct hsm_action_list *hal;
864 if (ct == NULL || ct->magic != CT_PRIV_MAGIC)
867 if (halh == NULL || msgsize == NULL)
873 rc = libcfs_ukuc_msg_get(&ct->kuc, (char *)kuch,
874 HAL_MAXSIZE + sizeof(*kuch),
879 /* Handle generic messages */
880 if (kuch->kuc_transport == KUC_TRANSPORT_GENERIC &&
881 kuch->kuc_msgtype == KUC_MSG_SHUTDOWN) {
886 if (kuch->kuc_transport != KUC_TRANSPORT_HSM ||
887 kuch->kuc_msgtype != HMT_ACTION_LIST) {
888 llapi_err_noerrno(LLAPI_MSG_ERROR,
889 "Unknown HSM message type %d:%d\n",
890 kuch->kuc_transport, kuch->kuc_msgtype);
895 if (kuch->kuc_msglen < sizeof(*kuch) + sizeof(*hal)) {
896 llapi_err_noerrno(LLAPI_MSG_ERROR, "Short HSM message %d",
902 /* Our message is a hsm_action_list. Use pointer math to skip
903 * kuch_hdr and point directly to the message payload.
905 hal = (struct hsm_action_list *)(kuch + 1);
907 /* Check that we have registered for this archive #
908 * if 0 registered, we serve any archive */
910 ((1 << (hal->hal_archive_id - 1)) & ct->archives) == 0) {
911 llapi_err_noerrno(LLAPI_MSG_INFO,
912 "This copytool does not service archive #%d,"
913 " ignoring this request."
914 " Mask of served archive is 0x%.8X",
915 hal->hal_archive_id, ct->archives);
921 *msgsize = kuch->kuc_msglen - sizeof(*kuch);
930 /** Get parent path from mount point and fid.
932 * \param mnt Filesystem root path.
933 * \param fid Object FID.
934 * \param parent Destination buffer.
935 * \param parent_len Destination buffer size.
936 * \return 0 on success.
938 static int fid_parent(const char *mnt, const lustre_fid *fid, char *parent,
943 long long recno = -1;
945 char strfid[FID_NOBRACE_LEN + 1];
948 snprintf(strfid, sizeof(strfid), DFID_NOBRACE, PFID(fid));
950 rc = llapi_fid2path(mnt, strfid, file, sizeof(file),
955 /* fid2path returns a relative path */
956 rc = snprintf(parent, parent_len, "%s/%s", mnt, file);
957 if (rc >= parent_len)
958 return -ENAMETOOLONG;
960 /* remove file name */
961 ptr = strrchr(parent, '/');
962 if (ptr == NULL || ptr == parent) {
972 static int ct_open_by_fid(const struct hsm_copytool_private *ct,
973 const struct lu_fid *fid, int open_flags)
975 char fid_name[FID_NOBRACE_LEN + 1];
978 snprintf(fid_name, sizeof(fid_name), DFID_NOBRACE, PFID(fid));
980 fd = openat(ct->open_by_fid_fd, fid_name, open_flags);
981 return fd < 0 ? -errno : fd;
985 * Get metadata attributes of file by FID.
987 * Use the IOC_MDC_GETFILEINFO ioctl (to send a MDS_GETATTR_NAME RPC)
988 * to get the attributes of the file identified by \a fid. This
989 * returns only the attributes stored on the MDT and avoids taking
990 * layout locks or accessing OST objects. It also bypasses the inode
991 * cache. Attributes are returned in \a st.
993 static int ct_md_getattr(const struct hsm_copytool_private *ct,
994 const struct lu_fid *fid,
997 struct lov_user_mds_data *lmd;
1001 lmd_size = sizeof(lmd->lmd_st) +
1002 lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3);
1004 if (lmd_size < sizeof(lmd->lmd_st) + XATTR_SIZE_MAX)
1005 lmd_size = sizeof(lmd->lmd_st) + XATTR_SIZE_MAX;
1007 if (lmd_size < FID_NOBRACE_LEN + 1)
1008 lmd_size = FID_NOBRACE_LEN + 1;
1010 lmd = malloc(lmd_size);
1014 snprintf((char *)lmd, lmd_size, DFID_NOBRACE, PFID(fid));
1016 rc = ioctl(ct->open_by_fid_fd, IOC_MDC_GETFILEINFO, lmd);
1019 llapi_error(LLAPI_MSG_ERROR, rc,
1020 "cannot get metadata attributes of "DFID" in '%s'",
1021 PFID(fid), ct->mnt);
1032 /** Create the destination volatile file for a restore operation.
1034 * \param hcp Private copyaction handle.
1035 * \param mdt_index MDT index where to create the volatile file.
1036 * \param flags Volatile file creation flags.
1037 * \return 0 on success.
1039 static int create_restore_volatile(struct hsm_copyaction_private *hcp,
1040 int mdt_index, int open_flags)
1044 char parent[PATH_MAX + 1];
1045 const char *mnt = hcp->ct_priv->mnt;
1046 struct hsm_action_item *hai = &hcp->copy.hc_hai;
1048 rc = fid_parent(mnt, &hai->hai_fid, parent, sizeof(parent));
1050 /* fid_parent() failed, try to keep on going */
1051 llapi_error(LLAPI_MSG_ERROR, rc,
1052 "cannot get parent path to restore "DFID" "
1053 "using '%s'", PFID(&hai->hai_fid), mnt);
1054 snprintf(parent, sizeof(parent), "%s", mnt);
1057 fd = llapi_create_volatile_idx(parent, mdt_index, open_flags);
1061 rc = fchown(fd, hcp->stat.st_uid, hcp->stat.st_gid);
1065 rc = llapi_fd2fid(fd, &hai->hai_dfid);
1080 /** Start processing an HSM action.
1081 * Should be called by copytools just before starting handling a request.
1082 * It could be skipped if copytool only want to directly report an error,
1083 * \see llapi_hsm_action_end().
1085 * \param hcp Opaque action handle to be passed to
1086 * llapi_hsm_action_progress and llapi_hsm_action_end.
1087 * \param ct Copytool handle acquired at registration.
1088 * \param hai The hsm_action_item describing the request.
1089 * \param restore_mdt_index On restore: MDT index where to create the volatile
1090 * file. Use -1 for default.
1091 * \param restore_open_flags On restore: volatile file creation mode. Use
1092 * O_LOV_DELAY_CREATE to manually set the LOVEA
1094 * \param is_error Whether this call is just to report an error.
1096 * \return 0 on success.
1098 int llapi_hsm_action_begin(struct hsm_copyaction_private **phcp,
1099 const struct hsm_copytool_private *ct,
1100 const struct hsm_action_item *hai,
1101 int restore_mdt_index, int restore_open_flags,
1104 struct hsm_copyaction_private *hcp;
1108 hcp = calloc(1, sizeof(*hcp));
1112 hcp->source_fd = -1;
1115 hcp->copy.hc_hai = *hai;
1116 hcp->copy.hc_hai.hai_len = sizeof(*hai);
1121 if (hai->hai_action == HSMA_ARCHIVE) {
1122 fd = ct_open_by_fid(hcp->ct_priv, &hai->hai_dfid,
1123 O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NONBLOCK);
1129 hcp->source_fd = fd;
1130 } else if (hai->hai_action == HSMA_RESTORE) {
1131 rc = ct_md_getattr(hcp->ct_priv, &hai->hai_fid, &hcp->stat);
1135 rc = create_restore_volatile(hcp, restore_mdt_index,
1136 restore_open_flags);
1139 } else if (hai->hai_action == HSMA_REMOVE) {
1140 /* Since remove is atomic there is no need to send an
1141 * initial MDS_HSM_PROGRESS RPC. */
1145 rc = ioctl(ct->mnt_fd, LL_IOC_HSM_COPY_START, &hcp->copy);
1152 llapi_hsm_log_ct_progress(&hcp, hai, CT_START, 0, 0);
1155 hcp->magic = CP_PRIV_MAGIC;
1160 if (!(hcp->source_fd < 0))
1161 close(hcp->source_fd);
1163 if (!(hcp->data_fd < 0))
1164 close(hcp->data_fd);
1171 /** Terminate an HSM action processing.
1172 * Should be called by copytools just having finished handling the request.
1173 * \param hdl[in,out] Handle returned by llapi_hsm_action_start.
1174 * \param he[in] The final range of copied data (for copy actions).
1175 * \param errval[in] The status code of the operation.
1176 * \param flags[in] The flags about the termination status (HP_FLAG_RETRY if
1177 * the error is retryable).
1179 * \return 0 on success.
1181 int llapi_hsm_action_end(struct hsm_copyaction_private **phcp,
1182 const struct hsm_extent *he, int hp_flags, int errval)
1184 struct hsm_copyaction_private *hcp;
1185 struct hsm_action_item *hai;
1188 if (phcp == NULL || *phcp == NULL || he == NULL)
1193 if (hcp->magic != CP_PRIV_MAGIC)
1196 hai = &hcp->copy.hc_hai;
1198 if (hai->hai_action == HSMA_RESTORE && errval == 0) {
1199 struct ll_futimes_3 lfu = {
1200 .lfu_atime_sec = hcp->stat.st_atim.tv_sec,
1201 .lfu_atime_nsec = hcp->stat.st_atim.tv_nsec,
1202 .lfu_mtime_sec = hcp->stat.st_mtim.tv_sec,
1203 .lfu_mtime_nsec = hcp->stat.st_mtim.tv_nsec,
1204 .lfu_ctime_sec = hcp->stat.st_ctim.tv_sec,
1205 .lfu_ctime_nsec = hcp->stat.st_ctim.tv_nsec,
1208 rc = fsync(hcp->data_fd);
1214 /* Set {a,m,c}time of volatile file to that of original. */
1215 rc = ioctl(hcp->data_fd, LL_IOC_FUTIMES_3, &lfu);
1223 /* In some cases, like restore, 2 FIDs are used.
1224 * Set the right FID to use here. */
1225 if (hai->hai_action == HSMA_ARCHIVE || hai->hai_action == HSMA_RESTORE)
1226 hai->hai_fid = hai->hai_dfid;
1228 /* Fill the last missing data that will be needed by
1229 * kernel to send a hsm_progress. */
1230 hcp->copy.hc_flags = hp_flags;
1231 hcp->copy.hc_errval = abs(errval);
1233 hcp->copy.hc_hai.hai_extent = *he;
1235 rc = ioctl(hcp->ct_priv->mnt_fd, LL_IOC_HSM_COPY_END, &hcp->copy);
1241 llapi_hsm_log_ct_progress(&hcp, hai, CT_FINISH, 0, 0);
1244 if (!(hcp->source_fd < 0))
1245 close(hcp->source_fd);
1247 if (!(hcp->data_fd < 0))
1248 close(hcp->data_fd);
1256 /** Notify a progress in processing an HSM action.
1257 * \param hdl[in,out] handle returned by llapi_hsm_action_start.
1258 * \param he[in] the range of copied data (for copy actions).
1259 * \param total[in] the expected total of copied data (for copy actions).
1260 * \param hp_flags[in] HSM progress flags.
1261 * \return 0 on success.
1263 int llapi_hsm_action_progress(struct hsm_copyaction_private *hcp,
1264 const struct hsm_extent *he, __u64 total,
1268 struct hsm_progress hp;
1269 struct hsm_action_item *hai;
1271 if (hcp == NULL || he == NULL)
1274 if (hcp->magic != CP_PRIV_MAGIC)
1277 hai = &hcp->copy.hc_hai;
1279 memset(&hp, 0, sizeof(hp));
1281 hp.hp_cookie = hai->hai_cookie;
1282 hp.hp_flags = hp_flags;
1284 /* Progress is made on the data fid */
1285 hp.hp_fid = hai->hai_dfid;
1288 rc = ioctl(hcp->ct_priv->mnt_fd, LL_IOC_HSM_PROGRESS, &hp);
1292 llapi_hsm_log_ct_progress(&hcp, hai, CT_RUNNING, total, he->length);
1297 /** Get the fid of object to be used for copying data.
1298 * @return error code if the action is not a copy operation.
1300 int llapi_hsm_action_get_dfid(const struct hsm_copyaction_private *hcp,
1303 const struct hsm_action_item *hai = &hcp->copy.hc_hai;
1305 if (hcp->magic != CP_PRIV_MAGIC)
1308 if (hai->hai_action != HSMA_RESTORE && hai->hai_action != HSMA_ARCHIVE)
1311 *fid = hai->hai_dfid;
1317 * Get a file descriptor to be used for copying data. It's up to the
1318 * caller to close the FDs obtained from this function.
1320 * @retval a file descriptor on success.
1321 * @retval a negative error code on failure.
1323 int llapi_hsm_action_get_fd(const struct hsm_copyaction_private *hcp)
1325 const struct hsm_action_item *hai = &hcp->copy.hc_hai;
1328 if (hcp->magic != CP_PRIV_MAGIC)
1331 if (hai->hai_action == HSMA_ARCHIVE) {
1332 fd = dup(hcp->source_fd);
1333 return fd < 0 ? -errno : fd;
1334 } else if (hai->hai_action == HSMA_RESTORE) {
1335 fd = dup(hcp->data_fd);
1336 return fd < 0 ? -errno : fd;
1343 * Import an existing hsm-archived file into Lustre.
1345 * Caller must access file by (returned) newfid value from now on.
1347 * \param dst path to Lustre destination (e.g. /mnt/lustre/my/file).
1348 * \param archive archive number.
1349 * \param st struct stat buffer containing file ownership, perm, etc.
1350 * \param stripe_* Striping options. Currently ignored, since the restore
1351 * operation will set the striping. In V2, this striping might
1353 * \param newfid[out] Filled with new Lustre fid.
1355 int llapi_hsm_import(const char *dst, int archive, const struct stat *st,
1356 unsigned long long stripe_size, int stripe_offset,
1357 int stripe_count, int stripe_pattern, char *pool_name,
1360 struct hsm_user_import hui;
1364 if (stripe_pattern == 0)
1365 stripe_pattern = LOV_PATTERN_RAID0;
1367 /* Create a non-striped file */
1368 fd = llapi_file_open_pool(dst, O_CREAT | O_WRONLY, st->st_mode,
1369 stripe_size, stripe_offset, stripe_count,
1370 stripe_pattern | LOV_PATTERN_F_RELEASED,
1373 llapi_error(LLAPI_MSG_ERROR, fd,
1374 "cannot create '%s' for import", dst);
1378 /* Get the new fid in Lustre. Caller needs to use this fid
1380 rc = llapi_fd2fid(fd, newfid);
1382 llapi_error(LLAPI_MSG_ERROR, rc,
1383 "cannot get fid of '%s' for import", dst);
1387 hui.hui_uid = st->st_uid;
1388 hui.hui_gid = st->st_gid;
1389 hui.hui_mode = st->st_mode;
1390 hui.hui_size = st->st_size;
1391 hui.hui_archive_id = archive;
1392 hui.hui_atime = st->st_atime;
1393 hui.hui_atime_ns = st->st_atim.tv_nsec;
1394 hui.hui_mtime = st->st_mtime;
1395 hui.hui_mtime_ns = st->st_mtim.tv_nsec;
1396 rc = ioctl(fd, LL_IOC_HSM_IMPORT, &hui);
1399 llapi_error(LLAPI_MSG_ERROR, rc, "cannot import '%s'", dst);
1412 * Return the current HSM states and HSM requests related to file pointed by \a
1415 * \param hus Should be allocated by caller. Will be filled with current file
1418 * \retval 0 on success.
1419 * \retval -errno on error.
1421 int llapi_hsm_state_get_fd(int fd, struct hsm_user_state *hus)
1425 rc = ioctl(fd, LL_IOC_HSM_STATE_GET, hus);
1426 /* If error, save errno value */
1427 rc = rc ? -errno : 0;
1433 * Return the current HSM states and HSM requests related to file pointed by \a
1436 * see llapi_hsm_state_get_fd() for args use and return
1438 int llapi_hsm_state_get(const char *path, struct hsm_user_state *hus)
1443 fd = open(path, O_RDONLY | O_NONBLOCK);
1447 rc = llapi_hsm_state_get_fd(fd, hus);
1454 * Set HSM states of file pointed by \a fd
1456 * Using the provided bitmasks, the current HSM states for this file will be
1457 * changed. \a archive_id could be used to change the archive number also. Set
1458 * it to 0 if you do not want to change it.
1460 * \param setmask Bitmask for flag to be set.
1461 * \param clearmask Bitmask for flag to be cleared.
1462 * \param archive_id Archive number identifier to use. 0 means no change.
1464 * \retval 0 on success.
1465 * \retval -errno on error.
1467 int llapi_hsm_state_set_fd(int fd, __u64 setmask, __u64 clearmask,
1470 struct hsm_state_set hss;
1473 hss.hss_valid = HSS_SETMASK|HSS_CLEARMASK;
1474 hss.hss_setmask = setmask;
1475 hss.hss_clearmask = clearmask;
1476 /* Change archive_id if provided. We can only change
1477 * to set something different than 0. */
1478 if (archive_id > 0) {
1479 hss.hss_valid |= HSS_ARCHIVE_ID;
1480 hss.hss_archive_id = archive_id;
1482 rc = ioctl(fd, LL_IOC_HSM_STATE_SET, &hss);
1483 /* If error, save errno value */
1484 rc = rc ? -errno : 0;
1490 * Set HSM states of file pointed by \a path.
1492 * see llapi_hsm_state_set_fd() for args use and return
1494 int llapi_hsm_state_set(const char *path, __u64 setmask, __u64 clearmask,
1500 fd = open(path, O_WRONLY | O_LOV_DELAY_CREATE | O_NONBLOCK);
1504 rc = llapi_hsm_state_set_fd(fd, setmask, clearmask, archive_id);
1511 * Return the current HSM request related to file pointed by \a path.
1513 * \param hca Should be allocated by caller. Will be filled with current file
1516 * \retval 0 on success.
1517 * \retval -errno on error.
1519 int llapi_hsm_current_action(const char *path, struct hsm_current_action *hca)
1524 fd = open(path, O_RDONLY | O_NONBLOCK);
1528 rc = ioctl(fd, LL_IOC_HSM_ACTION, hca);
1529 /* If error, save errno value */
1530 rc = rc ? -errno : 0;
1537 * Allocate a hsm_user_request with the specified carateristics.
1538 * This structure should be freed with free().
1540 * \return an allocated structure on success, NULL otherwise.
1542 struct hsm_user_request *llapi_hsm_user_request_alloc(int itemcount,
1547 len += sizeof(struct hsm_user_request);
1548 len += sizeof(struct hsm_user_item) * itemcount;
1551 return (struct hsm_user_request *)malloc(len);
1555 * Send a HSM request to Lustre, described in \param request.
1557 * \param path Fullpath to the file to operate on.
1558 * \param request The request, allocated with llapi_hsm_user_request_alloc().
1560 * \return 0 on success, an error code otherwise.
1562 int llapi_hsm_request(const char *path, const struct hsm_user_request *request)
1567 rc = get_root_path(WANT_FD, NULL, &fd, (char *)path, -1);
1571 rc = ioctl(fd, LL_IOC_HSM_REQUEST, request);
1572 /* If error, save errno value */
1573 rc = rc ? -errno : 0;