4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lod/lod_dev.c
34 * Lustre Logical Object Device
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 * The Logical Object Device (LOD) layer manages access to striped
41 * objects (both regular files and directories). It implements the DT
42 * device and object APIs and is responsible for creating, storing,
43 * and loading striping information as an extended attribute of the
44 * underlying OSD object. LOD is the server side analog of the LOV and
45 * LMV layers on the client side.
47 * Metadata LU object stack (layers of the same compound LU object,
48 * all have the same FID):
58 * During LOD object initialization the localness or remoteness of the
59 * object FID dictates the choice between OSD and OSP.
61 * An LOD object (file or directory) with N stripes (each has a
68 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
70 * When upper layers must access an object's stripes (which are
71 * themselves OST or MDT LU objects) LOD finds these objects by their
72 * FIDs and stores them as an array of DT object pointers on the
73 * object. Declarations and operations on LOD objects are received by
74 * LOD (as DT object operations) and performed on the underlying
75 * OS[DP] object and (as needed) on the stripes. From the perspective
76 * of LOD, a stripe-less file (created by mknod() or open with
77 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
78 * while a non-striped directory (created by mkdir()) is an object
79 * which will never have stripes.
81 * The LOD layer also implements a small subset of the OBD device API
82 * to support MDT stack initialization and finalization (an MDD device
83 * connects and disconnects itself to and from the underlying LOD
84 * device), and pool management. In turn LOD uses the OBD device API
85 * to connect it self to the underlying OSD, and to connect itself to
86 * OSP devices representing the MDTs and OSTs that bear the stripes of
90 #define DEBUG_SUBSYSTEM S_MDS
92 #include <linux/kthread.h>
93 #include <obd_class.h>
94 #include <md_object.h>
95 #include <lustre_fid.h>
96 #include <lustre_param.h>
97 #include <lustre_update.h>
98 #include <lustre_log.h>
100 #include "lod_internal.h"
102 static const char lod_update_log_name[] = "update_log";
103 static const char lod_update_log_dir_name[] = "update_log_dir";
106 * Lookup target by FID.
108 * Lookup MDT/OST target index by FID. Type of the target can be
111 * \param[in] env LU environment provided by the caller
112 * \param[in] lod lod device
114 * \param[out] tgt result target index
115 * \param[in] type expected type of the target:
116 * LU_SEQ_RANGE_{MDT,OST,ANY}
118 * \retval 0 on success
119 * \retval negative negated errno on error
121 int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
122 const struct lu_fid *fid, __u32 *tgt, int *type)
124 struct lu_seq_range range = { 0 };
125 struct lu_server_fld *server_fld;
129 if (!fid_is_sane(fid)) {
130 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
135 if (fid_is_idif(fid)) {
136 *tgt = fid_idif_ost_idx(fid);
137 *type = LU_SEQ_RANGE_OST;
141 if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
143 *type = LU_SEQ_RANGE_MDT;
147 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
148 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
150 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
151 *type = LU_SEQ_RANGE_MDT;
155 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
156 if (server_fld == NULL)
159 fld_range_set_type(&range, *type);
160 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
164 *tgt = range.lsr_index;
165 *type = range.lsr_flags;
167 CDEBUG(D_INFO, "%s: got tgt %x for sequence: "LPX64"\n",
168 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
173 /* Slab for OSD object allocation */
174 struct kmem_cache *lod_object_kmem;
176 /* Slab for dt_txn_callback */
177 struct kmem_cache *lod_txn_callback_kmem;
178 static struct lu_kmem_descr lod_caches[] = {
180 .ckd_cache = &lod_object_kmem,
181 .ckd_name = "lod_obj",
182 .ckd_size = sizeof(struct lod_object)
185 .ckd_cache = &lod_txn_callback_kmem,
186 .ckd_name = "lod_txn_callback",
187 .ckd_size = sizeof(struct dt_txn_callback)
194 static struct lu_device *lod_device_fini(const struct lu_env *env,
195 struct lu_device *d);
198 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
200 * Allocates and initializes LOD's slice in the given object.
202 * see include/lu_object.h for the details.
204 static struct lu_object *lod_object_alloc(const struct lu_env *env,
205 const struct lu_object_header *hdr,
206 struct lu_device *dev)
208 struct lod_object *lod_obj;
209 struct lu_object *lu_obj;
212 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
214 RETURN(ERR_PTR(-ENOMEM));
216 lu_obj = lod2lu_obj(lod_obj);
217 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
218 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
219 lu_obj->lo_ops = &lod_lu_obj_ops;
225 * Process the config log for all sub device.
227 * The function goes through all the targets in the given table
228 * and apply given configuration command on to the targets.
229 * Used to cleanup the targets at unmount.
231 * \param[in] env LU environment provided by the caller
232 * \param[in] lod lod device
233 * \param[in] ltd target's table to go through
234 * \param[in] lcfg configuration command to apply
236 * \retval 0 on success
237 * \retval negative negated errno on error
239 static int lod_sub_process_config(const struct lu_env *env,
240 struct lod_device *lod,
241 struct lod_tgt_descs *ltd,
242 struct lustre_cfg *lcfg)
244 struct lu_device *next;
249 if (ltd->ltd_tgts_size <= 0) {
250 lod_putref(lod, ltd);
253 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
254 struct lod_tgt_desc *tgt;
257 tgt = LTD_TGT(ltd, i);
258 LASSERT(tgt && tgt->ltd_tgt);
259 next = &tgt->ltd_tgt->dd_lu_dev;
260 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
262 CERROR("%s: error cleaning up LOD index %u: cmd %#x"
263 ": rc = %d\n", lod2obd(lod)->obd_name, i,
264 lcfg->lcfg_command, rc1);
268 lod_putref(lod, ltd);
272 struct lod_recovery_data {
273 struct lod_device *lrd_lod;
274 struct lod_tgt_desc *lrd_ltd;
275 struct ptlrpc_thread *lrd_thread;
281 * process update recovery record
283 * Add the update recovery recode to the update recovery list in
284 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
285 * will redo these updates.
287 * \param[in]env execution environment
288 * \param[in]llh log handle of update record
289 * \param[in]rec update record to be replayed
290 * \param[in]data update recovery data which holds the necessary
291 * arguments for recovery (see struct lod_recovery_data)
293 * \retval 0 if the record is processed successfully.
294 * \retval negative errno if the record processing fails.
296 static int lod_process_recovery_updates(const struct lu_env *env,
297 struct llog_handle *llh,
298 struct llog_rec_hdr *rec,
301 struct lod_recovery_data *lrd = data;
302 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
303 struct lu_target *lut;
307 if (lrd->lrd_ltd == NULL) {
310 rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
314 index = lrd->lrd_ltd->ltd_index;
318 llog_update_record_size((struct llog_update_record *)rec)) {
319 CERROR("%s broken update record! index %u "DOSTID":%u :"
320 " rc = %d\n", lod2obd(lrd->lrd_lod)->obd_name, index,
321 POSTID(&llh->lgh_id.lgl_oi), rec->lrh_index, -EIO);
325 cookie->lgc_lgl = llh->lgh_id;
326 cookie->lgc_index = rec->lrh_index;
327 cookie->lgc_subsys = LLOG_UPDATELOG_ORIG_CTXT;
329 CDEBUG(D_HA, "%s: process recovery updates "DOSTID":%u\n",
330 lod2obd(lrd->lrd_lod)->obd_name,
331 POSTID(&llh->lgh_id.lgl_oi), rec->lrh_index);
332 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
334 return insert_update_records_to_replay_list(lut->lut_tdtd,
335 (struct llog_update_record *)rec,
340 * recovery thread for update log
342 * Start recovery thread and prepare the sub llog, then it will retrieve
343 * the update records from the correpondent MDT and do recovery.
345 * \param[in] arg pointer to the recovery data
347 * \retval 0 if recovery succeeds
348 * \retval negative errno if recovery failed.
350 static int lod_sub_recovery_thread(void *arg)
352 struct lod_recovery_data *lrd = arg;
353 struct lod_device *lod = lrd->lrd_lod;
354 struct dt_device *dt;
355 struct ptlrpc_thread *thread = lrd->lrd_thread;
356 struct llog_ctxt *ctxt;
361 thread->t_flags = SVC_RUNNING;
362 wake_up(&thread->t_ctl_waitq);
364 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
367 CERROR("%s: can't initialize env: rc = %d\n",
368 lod2obd(lod)->obd_name, rc);
372 if (lrd->lrd_ltd == NULL)
375 dt = lrd->lrd_ltd->ltd_tgt;
378 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
382 /* Process the recovery record */
383 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd, LLOG_UPDATELOG_ORIG_CTXT);
384 LASSERT(ctxt != NULL);
385 LASSERT(ctxt->loc_handle != NULL);
387 rc = llog_cat_process(&env, ctxt->loc_handle,
388 lod_process_recovery_updates, lrd, 0, 0);
392 struct lu_device *top_device;
394 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
395 /* Because the remote target might failover at the same time,
396 * let's retry here */
397 if (rc == -ETIMEDOUT && dt != lod->lod_child &&
398 !top_device->ld_obd->obd_force_abort_recovery)
401 CERROR("%s getting update log failed: rc = %d\n",
402 dt->dd_lu_dev.ld_obd->obd_name, rc);
406 CDEBUG(D_HA, "%s retrieve update log: rc = %d\n",
407 dt->dd_lu_dev.ld_obd->obd_name, rc);
409 if (lrd->lrd_ltd == NULL)
410 lod->lod_child_got_update_log = 1;
412 lrd->lrd_ltd->ltd_got_update_log = 1;
414 if (lod->lod_child_got_update_log) {
415 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
416 struct lod_tgt_desc *tgt = NULL;
417 bool all_got_log = true;
420 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
421 tgt = LTD_TGT(ltd, i);
422 if (!tgt->ltd_got_update_log) {
429 struct lu_target *lut;
431 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
432 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
433 lut->lut_obd->obd_name);
434 lut->lut_tdtd->tdtd_replay_ready = 1;
435 wake_up(&lut->lut_obd->obd_next_transno_waitq);
441 thread->t_flags = SVC_STOPPED;
442 wake_up(&thread->t_ctl_waitq);
448 * finish sub llog context
450 * Stop update recovery thread for the sub device, then cleanup the
451 * correspondent llog ctxt.
453 * \param[in] env execution environment
454 * \param[in] lod lod device to do update recovery
455 * \param[in] thread recovery thread on this sub device
457 void lod_sub_fini_llog(const struct lu_env *env,
458 struct dt_device *dt, struct ptlrpc_thread *thread)
460 struct obd_device *obd;
461 struct llog_ctxt *ctxt;
464 obd = dt->dd_lu_dev.ld_obd;
465 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
466 /* Stop recovery thread first */
467 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
468 thread->t_flags = SVC_STOPPING;
469 wake_up(&thread->t_ctl_waitq);
470 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
473 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
477 if (ctxt->loc_handle != NULL)
478 llog_cat_close(env, ctxt->loc_handle);
480 llog_cleanup(env, ctxt);
486 * Extract MDT target index from a device name.
488 * a helper function to extract index from the given device name
489 * like "fsname-MDTxxxx-mdtlov"
491 * \param[in] lodname device name
492 * \param[out] mdt_index extracted index
494 * \retval 0 on success
495 * \retval -EINVAL if the name is invalid
497 int lodname2mdt_index(char *lodname, __u32 *mdt_index)
502 /* 1.8 configs don't have "-MDT0000" at the end */
503 ptr = strstr(lodname, "-MDT");
509 ptr = strrchr(lodname, '-');
511 CERROR("invalid MDT index in '%s'\n", lodname);
515 if (strncmp(ptr, "-mdtlov", 7) != 0) {
516 CERROR("invalid MDT index in '%s'\n", lodname);
520 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
521 CERROR("invalid MDT index in '%s'\n", lodname);
525 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
526 CERROR("invalid MDT index in '%s'\n", lodname);
530 index = simple_strtol(ptr - 4, &tmp, 16);
531 if (*tmp != '-' || index > INT_MAX) {
532 CERROR("invalid MDT index in '%s'\n", lodname);
540 * Init sub llog context
542 * Setup update llog ctxt for update recovery threads, then start the
543 * recovery thread (lod_sub_recovery_thread) to read update llog from
544 * the correspondent MDT to do update recovery.
546 * \param[in] env execution environment
547 * \param[in] lod lod device to do update recovery
548 * \param[in] dt sub dt device for which the recovery thread is
550 * \retval 0 if initialization succeeds.
551 * \retval negative errno if initialization fails.
553 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
554 struct dt_device *dt)
556 struct obd_device *obd;
557 struct lod_recovery_data *lrd = NULL;
558 struct ptlrpc_thread *thread;
559 struct task_struct *task;
560 struct l_wait_info lwi = { 0 };
561 struct lod_tgt_desc *sub_ltd = NULL;
567 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
575 if (lod->lod_child == dt) {
576 thread = &lod->lod_child_recovery_thread;
577 index = master_index;
579 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
580 struct lod_tgt_desc *tgt = NULL;
583 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
584 tgt = LTD_TGT(ltd, i);
585 if (tgt->ltd_tgt == dt) {
586 index = tgt->ltd_index;
591 LASSERT(sub_ltd != NULL);
592 OBD_ALLOC_PTR(sub_ltd->ltd_recovery_thread);
593 if (sub_ltd->ltd_recovery_thread == NULL)
594 GOTO(free_lrd, rc = -ENOMEM);
596 thread = sub_ltd->ltd_recovery_thread;
599 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
600 dt->dd_lu_dev.ld_obd->obd_name);
602 lrd->lrd_ltd = sub_ltd;
603 lrd->lrd_thread = thread;
604 lrd->lrd_idx = index;
605 init_waitqueue_head(&thread->t_ctl_waitq);
607 obd = dt->dd_lu_dev.ld_obd;
608 obd->obd_lvfs_ctxt.dt = dt;
609 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
610 NULL, &llog_common_cat_ops);
612 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
614 GOTO(free_thread, rc);
617 /* Start the recovery thread */
618 task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
619 master_index, index);
622 CERROR("%s: cannot start recovery thread: rc = %d\n",
627 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
628 thread->t_flags & SVC_STOPPED, &lwi);
632 lod_sub_fini_llog(env, dt, thread);
634 if (lod->lod_child != dt) {
635 OBD_FREE_PTR(sub_ltd->ltd_recovery_thread);
636 sub_ltd->ltd_recovery_thread = NULL;
644 * Stop sub recovery thread
646 * Stop sub recovery thread on all subs.
648 * \param[in] env execution environment
649 * \param[in] lod lod device to do update recovery
651 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
652 struct lod_device *lod)
654 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
655 struct ptlrpc_thread *thread;
658 /* Stop the update log commit cancel threads and finish master
660 thread = &lod->lod_child_recovery_thread;
661 /* Stop recovery thread first */
662 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
663 thread->t_flags = SVC_STOPPING;
664 wake_up(&thread->t_ctl_waitq);
665 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
669 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
670 struct lod_tgt_desc *tgt;
672 tgt = LTD_TGT(ltd, i);
673 thread = tgt->ltd_recovery_thread;
674 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
675 thread->t_flags = SVC_STOPPING;
676 wake_up(&thread->t_ctl_waitq);
677 wait_event(thread->t_ctl_waitq,
678 thread->t_flags & SVC_STOPPED);
679 OBD_FREE_PTR(tgt->ltd_recovery_thread);
680 tgt->ltd_recovery_thread = NULL;
684 lod_putref(lod, ltd);
688 * finish all sub llog
690 * cleanup all of sub llog ctxt on the LOD.
692 * \param[in] env execution environment
693 * \param[in] lod lod device to do update recovery
695 static void lod_sub_fini_all_llogs(const struct lu_env *env,
696 struct lod_device *lod)
698 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
701 /* Stop the update log commit cancel threads and finish master
703 lod_sub_fini_llog(env, lod->lod_child,
704 &lod->lod_child_recovery_thread);
706 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
707 struct lod_tgt_desc *tgt;
709 tgt = LTD_TGT(ltd, i);
710 lod_sub_fini_llog(env, tgt->ltd_tgt,
711 tgt->ltd_recovery_thread);
714 lod_putref(lod, ltd);
718 * Prepare distribute txn
720 * Prepare distribute txn structure for LOD
722 * \param[in] env execution environment
723 * \param[in] lod_device LOD device
725 * \retval 0 if preparation succeeds.
726 * \retval negative errno if preparation fails.
728 static int lod_prepare_distribute_txn(const struct lu_env *env,
729 struct lod_device *lod)
731 struct target_distribute_txn_data *tdtd;
732 struct lu_target *lut;
736 /* Init update recovery data */
741 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
742 tdtd->tdtd_dt = &lod->lod_dt_dev;
743 rc = distribute_txn_init(env, lut, tdtd,
744 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
747 CERROR("%s: cannot init distribute txn: rc = %d\n",
748 lod2obd(lod)->obd_name, rc);
753 lut->lut_tdtd = tdtd;
759 * Finish distribute txn
761 * Release the resource holding by distribute txn, i.e. stop distribute
764 * \param[in] env execution environment
765 * \param[in] lod lod device
767 static void lod_fini_distribute_txn(const struct lu_env *env,
768 struct lod_device *lod)
770 struct lu_target *lut;
772 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
773 if (lut->lut_tdtd == NULL)
776 distribute_txn_fini(env, lut->lut_tdtd);
778 OBD_FREE_PTR(lut->lut_tdtd);
779 lut->lut_tdtd = NULL;
783 * Implementation of lu_device_operations::ldo_process_config() for LOD
785 * The method is called by the configuration subsystem during setup,
786 * cleanup and when the configuration changes. The method processes
787 * few specific commands like adding/removing the targets, changing
788 * the runtime parameters.
790 * \param[in] env LU environment provided by the caller
791 * \param[in] dev lod device
792 * \param[in] lcfg configuration command to apply
794 * \retval 0 on success
795 * \retval negative negated errno on error
797 * The examples are below.
799 * Add osc config log:
800 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
801 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
802 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
803 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
804 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
805 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
807 * Add mdc config log:
808 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
809 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
810 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
811 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
812 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
813 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
815 static int lod_process_config(const struct lu_env *env,
816 struct lu_device *dev,
817 struct lustre_cfg *lcfg)
819 struct lod_device *lod = lu2lod_dev(dev);
820 struct lu_device *next = &lod->lod_child->dd_lu_dev;
825 switch(lcfg->lcfg_command) {
826 case LCFG_LOV_DEL_OBD:
827 case LCFG_LOV_ADD_INA:
828 case LCFG_LOV_ADD_OBD:
833 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
834 * modify_mdc_tgts add 0:lustre-MDT0001
835 * 1:lustre-MDT0001-mdc0002
837 arg1 = lustre_cfg_string(lcfg, 1);
839 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
840 GOTO(out, rc = -EINVAL);
841 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
842 GOTO(out, rc = -EINVAL);
844 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
847 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
852 rc = lod_add_device(env, lod, arg1, index, gen,
853 mdt_index, LUSTRE_OSC_NAME, 1);
854 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
856 rc = lod_add_device(env, lod, arg1, index, gen,
857 mdt_index, LUSTRE_MDC_NAME, 1);
858 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
859 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
861 rc = lod_add_device(env, lod, arg1, index, gen,
862 mdt_index, LUSTRE_OSC_NAME, 0);
864 rc = lod_del_device(env, lod,
866 arg1, index, gen, true);
873 struct obd_device *obd;
876 /* Check if it is activate/deactivate mdc
877 * lustre-MDTXXXX-osp-MDTXXXX.active=1 */
878 param = lustre_cfg_buf(lcfg, 1);
879 if (strstr(param, "osp") != NULL &&
880 strstr(param, ".active=") != NULL) {
881 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
882 struct lod_tgt_desc *sub_tgt = NULL;
887 ptr = strstr(param, ".");
889 obd = class_name2obd(param);
891 CERROR("%s: can not find %s: rc = %d\n",
892 lod2obd(lod)->obd_name, param, -EINVAL);
897 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
898 struct lod_tgt_desc *tgt;
900 tgt = LTD_TGT(ltd, i);
901 if (tgt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
907 if (sub_tgt == NULL) {
908 CERROR("%s: can not find %s: rc = %d\n",
909 lod2obd(lod)->obd_name, param, -EINVAL);
915 tmp = strstr(param, "=");
918 struct llog_ctxt *ctxt;
920 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
921 ctxt = llog_get_context(obd,
922 LLOG_UPDATELOG_ORIG_CTXT);
924 rc = llog_setup(env, obd, &obd->obd_olg,
925 LLOG_UPDATELOG_ORIG_CTXT,
926 NULL, &llog_common_cat_ops);
932 rc = lod_sub_prep_llog(env, lod,
936 sub_tgt->ltd_active = 1;
938 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
940 sub_tgt->ltd_active = 0;
946 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
952 case LCFG_PRE_CLEANUP: {
953 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
954 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
955 next = &lod->lod_child->dd_lu_dev;
956 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
958 CDEBUG(D_HA, "%s: can't process %u: %d\n",
959 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
961 lod_sub_stop_recovery_threads(env, lod);
962 lod_fini_distribute_txn(env, lod);
963 lod_sub_fini_all_llogs(env, lod);
968 * do cleanup on underlying storage only when
969 * all OSPs are cleaned up, as they use that OSD as well
971 lu_dev_del_linkage(dev->ld_site, dev);
972 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
973 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
974 next = &lod->lod_child->dd_lu_dev;
975 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
977 CERROR("%s: can't process %u: %d\n",
978 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
980 rc = obd_disconnect(lod->lod_child_exp);
982 CERROR("error in disconnect from storage: %d\n", rc);
986 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
997 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
999 * The method is called once the recovery is complete. This implementation
1000 * distributes the notification to all the known targets.
1002 * see include/lu_object.h for the details
1004 static int lod_recovery_complete(const struct lu_env *env,
1005 struct lu_device *dev)
1007 struct lod_device *lod = lu2lod_dev(dev);
1008 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1013 LASSERT(lod->lod_recovery_completed == 0);
1014 lod->lod_recovery_completed = 1;
1016 rc = next->ld_ops->ldo_recovery_complete(env, next);
1018 lod_getref(&lod->lod_ost_descs);
1019 if (lod->lod_osts_size > 0) {
1020 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
1021 struct lod_tgt_desc *tgt;
1022 tgt = OST_TGT(lod, i);
1023 LASSERT(tgt && tgt->ltd_tgt);
1024 next = &tgt->ltd_ost->dd_lu_dev;
1025 rc = next->ld_ops->ldo_recovery_complete(env, next);
1027 CERROR("%s: can't complete recovery on #%d:"
1028 "%d\n", lod2obd(lod)->obd_name, i, rc);
1031 lod_putref(lod, &lod->lod_ost_descs);
1036 * Init update logs on all sub device
1038 * LOD initialize update logs on all of sub devices. Because the initialization
1039 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1040 * lod_object_init(), this API has to be called after LOD is initialized.
1041 * \param[in] env execution environment
1042 * \param[in] lod lod device
1044 * \retval 0 if update log is initialized successfully.
1045 * \retval negative errno if initialization fails.
1047 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1049 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1054 /* llog must be setup after LOD is initialized, because llog
1055 * initialization include FLD lookup */
1056 LASSERT(lod->lod_initialized);
1058 /* Init the llog in its own stack */
1059 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1063 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1064 struct lod_tgt_desc *tgt;
1066 tgt = LTD_TGT(ltd, i);
1067 rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
1076 * Implementation of lu_device_operations::ldo_prepare() for LOD
1078 * see include/lu_object.h for the details.
1080 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1081 struct lu_device *cdev)
1083 struct lod_device *lod = lu2lod_dev(cdev);
1084 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1085 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1087 struct dt_object *root;
1088 struct dt_object *dto;
1092 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1094 CERROR("%s: prepare bottom error: rc = %d\n",
1095 lod2obd(lod)->obd_name, rc);
1099 lod->lod_initialized = 1;
1101 rc = dt_root_get(env, lod->lod_child, fid);
1105 root = dt_locate(env, lod->lod_child, fid);
1107 RETURN(PTR_ERR(root));
1109 /* Create update log object */
1110 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1111 lu_update_log_fid(fid, index);
1113 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1115 lod_update_log_name,
1116 S_IFREG | S_IRUGO | S_IWUSR);
1118 GOTO(out_put, rc = PTR_ERR(dto));
1120 lu_object_put(env, &dto->do_lu);
1122 /* Create update log dir */
1123 lu_update_log_dir_fid(fid, index);
1124 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1126 lod_update_log_dir_name,
1127 S_IFDIR | S_IRUGO | S_IWUSR);
1129 GOTO(out_put, rc = PTR_ERR(dto));
1131 lu_object_put(env, &dto->do_lu);
1133 rc = lod_prepare_distribute_txn(env, lod);
1137 rc = lod_sub_init_llogs(env, lod);
1142 lu_object_put(env, &root->do_lu);
1147 const struct lu_device_operations lod_lu_ops = {
1148 .ldo_object_alloc = lod_object_alloc,
1149 .ldo_process_config = lod_process_config,
1150 .ldo_recovery_complete = lod_recovery_complete,
1151 .ldo_prepare = lod_prepare,
1155 * Implementation of dt_device_operations::dt_root_get() for LOD
1157 * see include/dt_object.h for the details.
1159 static int lod_root_get(const struct lu_env *env,
1160 struct dt_device *dev, struct lu_fid *f)
1162 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1166 * Implementation of dt_device_operations::dt_statfs() for LOD
1168 * see include/dt_object.h for the details.
1170 static int lod_statfs(const struct lu_env *env,
1171 struct dt_device *dev, struct obd_statfs *sfs)
1173 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1177 * Implementation of dt_device_operations::dt_trans_create() for LOD
1179 * Creates a transaction using local (to this node) OSD.
1181 * see include/dt_object.h for the details.
1183 static struct thandle *lod_trans_create(const struct lu_env *env,
1184 struct dt_device *dt)
1188 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1198 * Implementation of dt_device_operations::dt_trans_start() for LOD
1200 * Starts the set of local transactions using the targets involved
1201 * in declare phase. Initial support for the distributed transactions.
1203 * see include/dt_object.h for the details.
1205 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1208 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1211 static int lod_trans_cb_add(struct thandle *th,
1212 struct dt_txn_commit_cb *dcb)
1214 struct top_thandle *top_th = container_of(th, struct top_thandle,
1216 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1220 * add noop update to the update records
1222 * Add noop updates to the update records, which is only used in
1225 * \param[in] env execution environment
1226 * \param[in] dt dt device of lod
1227 * \param[in] th thandle
1228 * \param[in] count the count of update records to be added.
1230 * \retval 0 if adding succeeds.
1231 * \retval negative errno if adding fails.
1233 static int lod_add_noop_records(const struct lu_env *env,
1234 struct dt_device *dt, struct thandle *th,
1237 struct top_thandle *top_th;
1238 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1242 top_th = container_of(th, struct top_thandle, tt_super);
1243 if (top_th->tt_multiple_thandle == NULL)
1247 for (i = 0; i < count; i++) {
1248 rc = update_record_pack(noop, th, fid);
1256 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1258 * Stops the set of local transactions using the targets involved
1259 * in declare phase. Initial support for the distributed transactions.
1261 * see include/dt_object.h for the details.
1263 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1266 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1269 rc = lod_add_noop_records(env, dt, th, 5000);
1273 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1277 * Implementation of dt_device_operations::dt_conf_get() for LOD
1279 * Currently returns the configuration provided by the local OSD.
1281 * see include/dt_object.h for the details.
1283 static void lod_conf_get(const struct lu_env *env,
1284 const struct dt_device *dev,
1285 struct dt_device_param *param)
1287 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1291 * Implementation of dt_device_operations::dt_sync() for LOD
1293 * Syncs all known OST targets. Very very expensive and used
1294 * rarely by LFSCK now. Should not be used in general.
1296 * see include/dt_object.h for the details.
1298 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1300 struct lod_device *lod = dt2lod_dev(dev);
1301 struct lod_ost_desc *ost;
1306 lod_getref(&lod->lod_ost_descs);
1307 lod_foreach_ost(lod, i) {
1308 ost = OST_TGT(lod, i);
1309 LASSERT(ost && ost->ltd_ost);
1310 rc = dt_sync(env, ost->ltd_ost);
1312 CERROR("%s: can't sync %u: %d\n",
1313 lod2obd(lod)->obd_name, i, rc);
1317 lod_putref(lod, &lod->lod_ost_descs);
1319 rc = dt_sync(env, lod->lod_child);
1325 * Implementation of dt_device_operations::dt_ro() for LOD
1327 * Turns local OSD read-only, used for the testing only.
1329 * see include/dt_object.h for the details.
1331 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1333 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1337 * Implementation of dt_device_operations::dt_commit_async() for LOD
1339 * Asks local OSD to commit sooner.
1341 * see include/dt_object.h for the details.
1343 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1345 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1348 static const struct dt_device_operations lod_dt_ops = {
1349 .dt_root_get = lod_root_get,
1350 .dt_statfs = lod_statfs,
1351 .dt_trans_create = lod_trans_create,
1352 .dt_trans_start = lod_trans_start,
1353 .dt_trans_stop = lod_trans_stop,
1354 .dt_conf_get = lod_conf_get,
1355 .dt_sync = lod_sync,
1357 .dt_commit_async = lod_commit_async,
1358 .dt_trans_cb_add = lod_trans_cb_add,
1362 * Connect to a local OSD.
1364 * Used to connect to the local OSD at mount. OSD name is taken from the
1365 * configuration command passed. This connection is used to identify LU
1366 * site and pin the OSD from early removal.
1368 * \param[in] env LU environment provided by the caller
1369 * \param[in] lod lod device
1370 * \param[in] cfg configuration command to apply
1372 * \retval 0 on success
1373 * \retval negative negated errno on error
1375 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1376 struct lustre_cfg *cfg)
1378 struct obd_connect_data *data = NULL;
1379 struct obd_device *obd;
1380 char *nextdev = NULL, *p, *s;
1385 LASSERT(lod->lod_child_exp == NULL);
1387 /* compatibility hack: we still use old config logs
1388 * which specify LOV, but we need to learn underlying
1389 * OSD device, which is supposed to be:
1390 * <fsname>-MDTxxxx-osd
1392 * 2.x MGS generates lines like the following:
1393 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1394 * 1.8 MGS generates lines like the following:
1395 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1397 * we use "-MDT" to differentiate 2.x from 1.8 */
1399 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
1400 len = strlen(p) + 6;
1401 OBD_ALLOC(nextdev, len);
1402 if (nextdev == NULL)
1403 GOTO(out, rc = -ENOMEM);
1406 s = strstr(nextdev, "-mdtlov");
1407 if (unlikely(s == NULL)) {
1408 CERROR("unable to parse device name %s\n",
1409 lustre_cfg_string(cfg, 0));
1410 GOTO(out, rc = -EINVAL);
1413 if (strstr(nextdev, "-MDT")) {
1418 strcpy(s, "-MDT0000-osd");
1421 CERROR("unable to parse device name %s\n",
1422 lustre_cfg_string(cfg, 0));
1423 GOTO(out, rc = -EINVAL);
1426 OBD_ALLOC_PTR(data);
1428 GOTO(out, rc = -ENOMEM);
1430 obd = class_name2obd(nextdev);
1432 CERROR("can not locate next device: %s\n", nextdev);
1433 GOTO(out, rc = -ENOTCONN);
1436 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1437 data->ocd_version = LUSTRE_VERSION_CODE;
1439 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1442 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1446 lod->lod_dt_dev.dd_lu_dev.ld_site =
1447 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1448 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1449 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1455 OBD_FREE(nextdev, len);
1460 * Allocate and initialize target table.
1462 * A helper function to initialize the target table and allocate
1463 * a bitmap of the available targets.
1465 * \param[in] ltd target's table to initialize
1467 * \retval 0 on success
1468 * \retval negative negated errno on error
1470 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
1472 mutex_init(<d->ltd_mutex);
1473 init_rwsem(<d->ltd_rw_sem);
1475 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
1476 * added to the LOD, see lod_add_device() */
1477 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
1478 if (ltd->ltd_tgt_bitmap == NULL)
1481 ltd->ltd_tgts_size = 32;
1484 ltd->ltd_death_row = 0;
1485 ltd->ltd_refcount = 0;
1490 * Initialize LOD device at setup.
1492 * Initializes the given LOD device using the original configuration command.
1493 * The function initiates a connection to the local OSD and initializes few
1494 * internal structures like pools, target tables, etc.
1496 * \param[in] env LU environment provided by the caller
1497 * \param[in] lod lod device
1498 * \param[in] ldt not used
1499 * \param[in] cfg configuration command
1501 * \retval 0 on success
1502 * \retval negative negated errno on error
1504 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1505 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1507 struct dt_device_param ddp;
1508 struct obd_device *obd;
1512 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1514 CERROR("Cannot find obd with name %s\n",
1515 lustre_cfg_string(cfg, 0));
1519 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1520 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1521 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1522 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1524 rc = lod_connect_to_osd(env, lod, cfg);
1528 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1529 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1531 /* setup obd to be used with old lov code */
1532 rc = lod_pools_init(lod, cfg);
1534 GOTO(out_disconnect, rc);
1536 rc = lod_procfs_init(lod);
1538 GOTO(out_pools, rc);
1540 spin_lock_init(&lod->lod_desc_lock);
1541 spin_lock_init(&lod->lod_connects_lock);
1542 lod_tgt_desc_init(&lod->lod_mdt_descs);
1543 lod_tgt_desc_init(&lod->lod_ost_descs);
1548 lod_pools_fini(lod);
1550 obd_disconnect(lod->lod_child_exp);
1555 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1557 * Releases the memory allocated for LOD device.
1559 * see include/lu_object.h for the details.
1561 static struct lu_device *lod_device_free(const struct lu_env *env,
1562 struct lu_device *lu)
1564 struct lod_device *lod = lu2lod_dev(lu);
1565 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1568 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1569 dt_device_fini(&lod->lod_dt_dev);
1575 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1577 * Allocates LOD device and calls the helpers to initialize it.
1579 * see include/lu_object.h for the details.
1581 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1582 struct lu_device_type *type,
1583 struct lustre_cfg *lcfg)
1585 struct lod_device *lod;
1586 struct lu_device *lu_dev;
1590 lu_dev = ERR_PTR(-ENOMEM);
1594 lu_dev = lod2lu_dev(lod);
1595 dt_device_init(&lod->lod_dt_dev, type);
1596 rc = lod_init0(env, lod, type, lcfg);
1598 lod_device_free(env, lu_dev);
1599 lu_dev = ERR_PTR(rc);
1607 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1609 * Releases the internal resources used by LOD device.
1611 * see include/lu_object.h for the details.
1613 static struct lu_device *lod_device_fini(const struct lu_env *env,
1614 struct lu_device *d)
1616 struct lod_device *lod = lu2lod_dev(d);
1620 lod_pools_fini(lod);
1622 lod_procfs_fini(lod);
1624 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
1626 CERROR("%s:can not fini ost descs %d\n",
1627 lod2obd(lod)->obd_name, rc);
1629 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
1631 CERROR("%s:can not fini mdt descs %d\n",
1632 lod2obd(lod)->obd_name, rc);
1638 * Implementation of obd_ops::o_connect() for LOD
1640 * Used to track all the users of this specific LOD device,
1641 * so the device stays up until the last user disconnected.
1643 * \param[in] env LU environment provided by the caller
1644 * \param[out] exp export the caller will be using to access LOD
1645 * \param[in] obd OBD device representing LOD device
1646 * \param[in] cluuid unique identifier of the caller
1647 * \param[in] data not used
1648 * \param[in] localdata not used
1650 * \retval 0 on success
1651 * \retval negative negated errno on error
1653 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1654 struct obd_device *obd, struct obd_uuid *cluuid,
1655 struct obd_connect_data *data, void *localdata)
1657 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1658 struct lustre_handle conn;
1662 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1664 rc = class_connect(&conn, obd, cluuid);
1668 *exp = class_conn2export(&conn);
1670 spin_lock(&lod->lod_connects_lock);
1671 lod->lod_connects++;
1672 /* at the moment we expect the only user */
1673 LASSERT(lod->lod_connects == 1);
1674 spin_unlock(&lod->lod_connects_lock);
1681 * Implementation of obd_ops::o_disconnect() for LOD
1683 * When the caller doesn't need to use this LOD instance, it calls
1684 * obd_disconnect() and LOD releases corresponding export/reference count.
1685 * Once all the users gone, LOD device is released.
1687 * \param[in] exp export provided to the caller in obd_connect()
1689 * \retval 0 on success
1690 * \retval negative negated errno on error
1692 static int lod_obd_disconnect(struct obd_export *exp)
1694 struct obd_device *obd = exp->exp_obd;
1695 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1696 int rc, release = 0;
1699 /* Only disconnect the underlying layers on the final disconnect. */
1700 spin_lock(&lod->lod_connects_lock);
1701 lod->lod_connects--;
1702 if (lod->lod_connects != 0) {
1703 /* why should there be more than 1 connect? */
1704 spin_unlock(&lod->lod_connects_lock);
1705 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1709 spin_unlock(&lod->lod_connects_lock);
1711 /* the last user of lod has gone, let's release the device */
1715 rc = class_disconnect(exp); /* bz 9811 */
1717 if (rc == 0 && release)
1718 class_manual_cleanup(obd);
1722 LU_KEY_INIT(lod, struct lod_thread_info);
1724 static void lod_key_fini(const struct lu_context *ctx,
1725 struct lu_context_key *key, void *data)
1727 struct lod_thread_info *info = data;
1728 /* allocated in lod_get_lov_ea
1729 * XXX: this is overload, a tread may have such store but used only
1730 * once. Probably better would be pool of such stores per LOD.
1732 if (info->lti_ea_store) {
1733 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1734 info->lti_ea_store = NULL;
1735 info->lti_ea_store_size = 0;
1737 lu_buf_free(&info->lti_linkea_buf);
1741 /* context key: lod_thread_key */
1742 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1744 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1746 static struct lu_device_type_operations lod_device_type_ops = {
1747 .ldto_init = lod_type_init,
1748 .ldto_fini = lod_type_fini,
1750 .ldto_start = lod_type_start,
1751 .ldto_stop = lod_type_stop,
1753 .ldto_device_alloc = lod_device_alloc,
1754 .ldto_device_free = lod_device_free,
1756 .ldto_device_fini = lod_device_fini
1759 static struct lu_device_type lod_device_type = {
1760 .ldt_tags = LU_DEVICE_DT,
1761 .ldt_name = LUSTRE_LOD_NAME,
1762 .ldt_ops = &lod_device_type_ops,
1763 .ldt_ctx_tags = LCT_MD_THREAD,
1767 * Implementation of obd_ops::o_get_info() for LOD
1769 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1770 * the caller binary status whether LOD has seen connection to any OST target.
1771 * It will also check if the MDT update log context being initialized (if
1774 * \param[in] env LU environment provided by the caller
1775 * \param[in] exp export of the caller
1776 * \param[in] keylen len of the key
1777 * \param[in] key the key
1778 * \param[in] vallen not used
1779 * \param[in] val not used
1781 * \retval 0 if a connection was seen
1782 * \retval -EAGAIN if LOD isn't running yet or no
1783 * connection has been seen yet
1784 * \retval -EINVAL if not supported key is requested
1786 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1787 __u32 keylen, void *key, __u32 *vallen, void *val)
1791 if (KEY_IS(KEY_OSP_CONNECTED)) {
1792 struct obd_device *obd = exp->exp_obd;
1793 struct lod_device *d;
1794 struct lod_tgt_desc *tgt;
1798 if (!obd->obd_set_up || obd->obd_stopping)
1801 d = lu2lod_dev(obd->obd_lu_dev);
1802 lod_getref(&d->lod_ost_descs);
1803 lod_foreach_ost(d, i) {
1804 tgt = OST_TGT(d, i);
1805 LASSERT(tgt && tgt->ltd_tgt);
1806 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
1808 /* one healthy device is enough */
1812 lod_putref(d, &d->lod_ost_descs);
1814 lod_getref(&d->lod_mdt_descs);
1815 lod_foreach_mdt(d, i) {
1816 struct llog_ctxt *ctxt;
1818 tgt = MDT_TGT(d, i);
1819 LASSERT(tgt != NULL);
1820 LASSERT(tgt->ltd_tgt != NULL);
1821 if (!tgt->ltd_active)
1824 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
1825 LLOG_UPDATELOG_ORIG_CTXT);
1827 CDEBUG(D_INFO, "%s: %s is not ready.\n",
1829 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
1833 if (ctxt->loc_handle == NULL) {
1834 CDEBUG(D_INFO, "%s: %s is not ready.\n",
1836 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
1838 llog_ctxt_put(ctxt);
1841 llog_ctxt_put(ctxt);
1843 lod_putref(d, &d->lod_mdt_descs);
1851 static int lod_obd_set_info_async(const struct lu_env *env,
1852 struct obd_export *exp,
1853 __u32 keylen, void *key,
1854 __u32 vallen, void *val,
1855 struct ptlrpc_request_set *set)
1857 struct obd_device *obd = class_exp2obd(exp);
1858 struct lod_device *d;
1859 struct lod_tgt_desc *tgt;
1866 set = ptlrpc_prep_set();
1871 d = lu2lod_dev(obd->obd_lu_dev);
1872 lod_getref(&d->lod_ost_descs);
1873 lod_foreach_ost(d, i) {
1874 tgt = OST_TGT(d, i);
1875 LASSERT(tgt && tgt->ltd_tgt);
1876 if (!tgt->ltd_active)
1879 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1881 if (rc2 != 0 && rc == 0)
1884 lod_putref(d, &d->lod_ost_descs);
1886 lod_getref(&d->lod_mdt_descs);
1887 lod_foreach_mdt(d, i) {
1888 tgt = MDT_TGT(d, i);
1889 LASSERT(tgt && tgt->ltd_tgt);
1890 if (!tgt->ltd_active)
1892 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1894 if (rc2 != 0 && rc == 0)
1897 lod_putref(d, &d->lod_mdt_descs);
1901 rc2 = ptlrpc_set_wait(set);
1902 if (rc2 == 0 && rc == 0)
1904 ptlrpc_set_destroy(set);
1909 static struct obd_ops lod_obd_device_ops = {
1910 .o_owner = THIS_MODULE,
1911 .o_connect = lod_obd_connect,
1912 .o_disconnect = lod_obd_disconnect,
1913 .o_get_info = lod_obd_get_info,
1914 .o_set_info_async = lod_obd_set_info_async,
1915 .o_pool_new = lod_pool_new,
1916 .o_pool_rem = lod_pool_remove,
1917 .o_pool_add = lod_pool_add,
1918 .o_pool_del = lod_pool_del,
1921 static int __init lod_init(void)
1923 struct obd_type *type;
1926 rc = lu_kmem_init(lod_caches);
1930 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
1931 LUSTRE_LOD_NAME, &lod_device_type);
1933 lu_kmem_fini(lod_caches);
1937 /* create "lov" entry in procfs for compatibility purposes */
1938 type = class_search_type(LUSTRE_LOV_NAME);
1939 if (type != NULL && type->typ_procroot != NULL)
1942 type = class_search_type(LUSTRE_LOD_NAME);
1943 type->typ_procsym = lprocfs_register("lov", proc_lustre_root,
1945 if (IS_ERR(type->typ_procsym)) {
1946 CERROR("lod: can't create compat entry \"lov\": %d\n",
1947 (int)PTR_ERR(type->typ_procsym));
1948 type->typ_procsym = NULL;
1953 static void __exit lod_exit(void)
1955 class_unregister_type(LUSTRE_LOD_NAME);
1956 lu_kmem_fini(lod_caches);
1959 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1960 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1961 MODULE_VERSION(LUSTRE_VERSION_STRING);
1962 MODULE_LICENSE("GPL");
1964 module_init(lod_init);
1965 module_exit(lod_exit);