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, 2014, 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;
280 * recovery thread for update log
282 * Start recovery thread and prepare the sub llog, then it will retrieve
283 * the update records from the correpondent MDT and do recovery.
285 * \param[in] arg pointer to the recovery data
287 * \retval 0 if recovery succeeds
288 * \retval negative errno if recovery failed.
290 static int lod_sub_recovery_thread(void *arg)
292 struct lod_recovery_data *lrd = arg;
293 struct lod_device *lod = lrd->lrd_lod;
294 struct dt_device *dt;
295 struct ptlrpc_thread *thread = lrd->lrd_thread;
300 thread->t_flags = SVC_RUNNING;
301 wake_up(&thread->t_ctl_waitq);
303 rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
306 CERROR("%s: can't initialize env: rc = %d\n",
307 lod2obd(lod)->obd_name, rc);
311 if (lrd->lrd_ltd == NULL)
314 dt = lrd->lrd_ltd->ltd_tgt;
316 rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
320 /* XXX do recovery in the following patches */
324 thread->t_flags = SVC_STOPPED;
325 wake_up(&thread->t_ctl_waitq);
331 * finish sub llog context
333 * Stop update recovery thread for the sub device, then cleanup the
334 * correspondent llog ctxt.
336 * \param[in] env execution environment
337 * \param[in] lod lod device to do update recovery
338 * \param[in] thread recovery thread on this sub device
340 void lod_sub_fini_llog(const struct lu_env *env,
341 struct dt_device *dt, struct ptlrpc_thread *thread)
343 struct obd_device *obd;
344 struct llog_ctxt *ctxt;
347 obd = dt->dd_lu_dev.ld_obd;
348 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
349 /* Stop recovery thread first */
350 if (thread != NULL && thread->t_flags & SVC_RUNNING) {
351 thread->t_flags = SVC_STOPPING;
352 wake_up(&thread->t_ctl_waitq);
353 wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
356 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
360 if (ctxt->loc_handle != NULL)
361 llog_cat_close(env, ctxt->loc_handle);
363 llog_cleanup(env, ctxt);
369 * Extract MDT target index from a device name.
371 * a helper function to extract index from the given device name
372 * like "fsname-MDTxxxx-mdtlov"
374 * \param[in] lodname device name
375 * \param[out] mdt_index extracted index
377 * \retval 0 on success
378 * \retval -EINVAL if the name is invalid
380 int lodname2mdt_index(char *lodname, __u32 *mdt_index)
385 /* 1.8 configs don't have "-MDT0000" at the end */
386 ptr = strstr(lodname, "-MDT");
392 ptr = strrchr(lodname, '-');
394 CERROR("invalid MDT index in '%s'\n", lodname);
398 if (strncmp(ptr, "-mdtlov", 7) != 0) {
399 CERROR("invalid MDT index in '%s'\n", lodname);
403 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
404 CERROR("invalid MDT index in '%s'\n", lodname);
408 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
409 CERROR("invalid MDT index in '%s'\n", lodname);
413 index = simple_strtol(ptr - 4, &tmp, 16);
414 if (*tmp != '-' || index > INT_MAX) {
415 CERROR("invalid MDT index in '%s'\n", lodname);
423 * Init sub llog context
425 * Setup update llog ctxt for update recovery threads, then start the
426 * recovery thread (lod_sub_recovery_thread) to read update llog from
427 * the correspondent MDT to do update recovery.
429 * \param[in] env execution environment
430 * \param[in] lod lod device to do update recovery
431 * \param[in] dt sub dt device for which the recovery thread is
433 * \retval 0 if initialization succeeds.
434 * \retval negative errno if initialization fails.
436 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
437 struct dt_device *dt)
439 struct obd_device *obd;
440 struct lod_recovery_data *lrd = NULL;
441 struct ptlrpc_thread *thread;
442 struct task_struct *task;
443 struct l_wait_info lwi = { 0 };
444 struct lod_tgt_desc *sub_ltd = NULL;
453 if (lod->lod_child == dt) {
454 thread = &lod->lod_child_recovery_thread;
455 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &index);
459 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
460 struct lod_tgt_desc *tgt = NULL;
463 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
464 tgt = LTD_TGT(ltd, i);
465 if (tgt->ltd_tgt == dt) {
466 index = tgt->ltd_index;
471 LASSERT(sub_ltd != NULL);
472 OBD_ALLOC_PTR(sub_ltd->ltd_recovery_thread);
473 if (sub_ltd->ltd_recovery_thread == NULL)
474 GOTO(free_lrd, rc = -ENOMEM);
476 thread = sub_ltd->ltd_recovery_thread;
480 lrd->lrd_ltd = sub_ltd;
481 lrd->lrd_thread = thread;
482 lrd->lrd_idx = index;
483 init_waitqueue_head(&thread->t_ctl_waitq);
485 obd = dt->dd_lu_dev.ld_obd;
486 obd->obd_lvfs_ctxt.dt = dt;
487 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
488 NULL, &llog_common_cat_ops);
490 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
492 GOTO(free_thread, rc);
495 /* Start the recovery thread */
496 task = kthread_run(lod_sub_recovery_thread, lrd, "lod_recov_%04x",
500 CERROR("%s: cannot start recovery thread: rc = %d\n",
505 l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
506 thread->t_flags & SVC_STOPPED, &lwi);
510 lod_sub_fini_llog(env, dt, thread);
512 if (lod->lod_child != dt) {
513 OBD_FREE_PTR(sub_ltd->ltd_recovery_thread);
514 sub_ltd->ltd_recovery_thread = NULL;
522 * finish all sub llog
524 * cleanup all of sub llog ctxt on the LOD.
526 * \param[in] env execution environment
527 * \param[in] lod lod device to do update recovery
529 static void lod_sub_fini_all_llogs(const struct lu_env *env,
530 struct lod_device *lod)
532 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
535 /* Stop the update log commit cancel threads and finish master
537 lod_sub_fini_llog(env, lod->lod_child,
538 &lod->lod_child_recovery_thread);
540 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
541 struct lod_tgt_desc *tgt;
543 tgt = LTD_TGT(ltd, i);
544 if (tgt->ltd_recovery_thread != NULL) {
545 lod_sub_fini_llog(env, tgt->ltd_tgt,
546 tgt->ltd_recovery_thread);
547 OBD_FREE_PTR(tgt->ltd_recovery_thread);
548 tgt->ltd_recovery_thread = NULL;
552 lod_putref(lod, ltd);
556 * Prepare distribute txn
558 * Prepare distribute txn structure for LOD
560 * \param[in] env execution environment
561 * \param[in] lod_device LOD device
563 * \retval 0 if preparation succeeds.
564 * \retval negative errno if preparation fails.
566 static int lod_prepare_distribute_txn(const struct lu_env *env,
567 struct lod_device *lod)
569 struct target_distribute_txn_data *tdtd;
570 struct lu_target *lut;
574 /* Init update recovery data */
579 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
581 rc = distribute_txn_init(env, lut, tdtd,
582 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
585 CERROR("%s: cannot init distribute txn: rc = %d\n",
586 lod2obd(lod)->obd_name, rc);
591 lut->lut_tdtd = tdtd;
597 * Finish distribute txn
599 * Release the resource holding by distribute txn, i.e. stop distribute
602 * \param[in] env execution environment
603 * \param[in] lod lod device
605 static void lod_fini_distribute_txn(const struct lu_env *env,
606 struct lod_device *lod)
608 struct lu_target *lut;
610 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
611 if (lut->lut_tdtd == NULL)
614 distribute_txn_fini(env, lut->lut_tdtd);
616 OBD_FREE_PTR(lut->lut_tdtd);
617 lut->lut_tdtd = NULL;
621 * Implementation of lu_device_operations::ldo_process_config() for LOD
623 * The method is called by the configuration subsystem during setup,
624 * cleanup and when the configuration changes. The method processes
625 * few specific commands like adding/removing the targets, changing
626 * the runtime parameters.
628 * \param[in] env LU environment provided by the caller
629 * \param[in] dev lod device
630 * \param[in] lcfg configuration command to apply
632 * \retval 0 on success
633 * \retval negative negated errno on error
635 * The examples are below.
637 * Add osc config log:
638 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
639 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
640 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
641 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
642 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
643 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
645 * Add mdc config log:
646 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
647 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
648 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
649 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
650 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
651 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
653 static int lod_process_config(const struct lu_env *env,
654 struct lu_device *dev,
655 struct lustre_cfg *lcfg)
657 struct lod_device *lod = lu2lod_dev(dev);
658 struct lu_device *next = &lod->lod_child->dd_lu_dev;
663 switch(lcfg->lcfg_command) {
664 case LCFG_LOV_DEL_OBD:
665 case LCFG_LOV_ADD_INA:
666 case LCFG_LOV_ADD_OBD:
671 /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
672 * modify_mdc_tgts add 0:lustre-MDT0001
673 * 1:lustre-MDT0001-mdc0002
675 arg1 = lustre_cfg_string(lcfg, 1);
677 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
678 GOTO(out, rc = -EINVAL);
679 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
680 GOTO(out, rc = -EINVAL);
682 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
685 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
690 rc = lod_add_device(env, lod, arg1, index, gen,
691 mdt_index, LUSTRE_OSC_NAME, 1);
692 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
694 rc = lod_add_device(env, lod, arg1, index, gen,
695 mdt_index, LUSTRE_MDC_NAME, 1);
696 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
697 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
699 rc = lod_add_device(env, lod, arg1, index, gen,
700 mdt_index, LUSTRE_OSC_NAME, 0);
702 rc = lod_del_device(env, lod,
704 arg1, index, gen, true);
711 struct obd_device *obd = lod2obd(lod);
713 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
719 case LCFG_PRE_CLEANUP: {
720 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
721 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
722 next = &lod->lod_child->dd_lu_dev;
723 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
725 CDEBUG(D_HA, "%s: can't process %u: %d\n",
726 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
728 lod_fini_distribute_txn(env, lod);
729 lod_sub_fini_all_llogs(env, lod);
734 * do cleanup on underlying storage only when
735 * all OSPs are cleaned up, as they use that OSD as well
737 lu_dev_del_linkage(dev->ld_site, dev);
738 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
739 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
740 next = &lod->lod_child->dd_lu_dev;
741 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
743 CERROR("%s: can't process %u: %d\n",
744 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
746 rc = obd_disconnect(lod->lod_child_exp);
748 CERROR("error in disconnect from storage: %d\n", rc);
752 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
763 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
765 * The method is called once the recovery is complete. This implementation
766 * distributes the notification to all the known targets.
768 * see include/lu_object.h for the details
770 static int lod_recovery_complete(const struct lu_env *env,
771 struct lu_device *dev)
773 struct lod_device *lod = lu2lod_dev(dev);
774 struct lu_device *next = &lod->lod_child->dd_lu_dev;
779 LASSERT(lod->lod_recovery_completed == 0);
780 lod->lod_recovery_completed = 1;
782 rc = next->ld_ops->ldo_recovery_complete(env, next);
784 lod_getref(&lod->lod_ost_descs);
785 if (lod->lod_osts_size > 0) {
786 cfs_foreach_bit(lod->lod_ost_bitmap, i) {
787 struct lod_tgt_desc *tgt;
788 tgt = OST_TGT(lod, i);
789 LASSERT(tgt && tgt->ltd_tgt);
790 next = &tgt->ltd_ost->dd_lu_dev;
791 rc = next->ld_ops->ldo_recovery_complete(env, next);
793 CERROR("%s: can't complete recovery on #%d:"
794 "%d\n", lod2obd(lod)->obd_name, i, rc);
797 lod_putref(lod, &lod->lod_ost_descs);
802 * Init update logs on all sub device
804 * LOD initialize update logs on all of sub devices. Because the initialization
805 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
806 * lod_object_init(), this API has to be called after LOD is initialized.
807 * \param[in] env execution environment
808 * \param[in] lod lod device
810 * \retval 0 if update log is initialized successfully.
811 * \retval negative errno if initialization fails.
813 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
815 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
820 /* llog must be setup after LOD is initialized, because llog
821 * initialization include FLD lookup */
822 LASSERT(lod->lod_initialized);
824 /* Init the llog in its own stack */
825 rc = lod_sub_init_llog(env, lod, lod->lod_child);
829 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
830 struct lod_tgt_desc *tgt;
832 tgt = LTD_TGT(ltd, i);
833 rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
842 * Implementation of lu_device_operations::ldo_prepare() for LOD
844 * see include/lu_object.h for the details.
846 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
847 struct lu_device *cdev)
849 struct lod_device *lod = lu2lod_dev(cdev);
850 struct lu_device *next = &lod->lod_child->dd_lu_dev;
851 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
853 struct dt_object *root;
854 struct dt_object *dto;
858 rc = next->ld_ops->ldo_prepare(env, pdev, next);
860 CERROR("%s: prepare bottom error: rc = %d\n",
861 lod2obd(lod)->obd_name, rc);
865 lod->lod_initialized = 1;
867 rc = dt_root_get(env, lod->lod_child, fid);
871 root = dt_locate(env, lod->lod_child, fid);
873 RETURN(PTR_ERR(root));
875 /* Create update log object */
876 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
877 lu_update_log_fid(fid, index);
879 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
882 S_IFREG | S_IRUGO | S_IWUSR);
884 GOTO(out_put, rc = PTR_ERR(dto));
886 lu_object_put(env, &dto->do_lu);
888 /* Create update log dir */
889 lu_update_log_dir_fid(fid, index);
890 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
892 lod_update_log_dir_name,
893 S_IFDIR | S_IRUGO | S_IWUSR);
895 GOTO(out_put, rc = PTR_ERR(dto));
897 lu_object_put(env, &dto->do_lu);
899 rc = lod_prepare_distribute_txn(env, lod);
903 rc = lod_sub_init_llogs(env, lod);
908 lu_object_put(env, &root->do_lu);
913 const struct lu_device_operations lod_lu_ops = {
914 .ldo_object_alloc = lod_object_alloc,
915 .ldo_process_config = lod_process_config,
916 .ldo_recovery_complete = lod_recovery_complete,
917 .ldo_prepare = lod_prepare,
921 * Implementation of dt_device_operations::dt_root_get() for LOD
923 * see include/dt_object.h for the details.
925 static int lod_root_get(const struct lu_env *env,
926 struct dt_device *dev, struct lu_fid *f)
928 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
932 * Implementation of dt_device_operations::dt_statfs() for LOD
934 * see include/dt_object.h for the details.
936 static int lod_statfs(const struct lu_env *env,
937 struct dt_device *dev, struct obd_statfs *sfs)
939 return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
943 * Implementation of dt_device_operations::dt_trans_create() for LOD
945 * Creates a transaction using local (to this node) OSD.
947 * see include/dt_object.h for the details.
949 static struct thandle *lod_trans_create(const struct lu_env *env,
950 struct dt_device *dt)
954 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
964 * Implementation of dt_device_operations::dt_trans_start() for LOD
966 * Starts the set of local transactions using the targets involved
967 * in declare phase. Initial support for the distributed transactions.
969 * see include/dt_object.h for the details.
971 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
974 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
977 static int lod_trans_cb_add(struct thandle *th,
978 struct dt_txn_commit_cb *dcb)
980 struct top_thandle *top_th = container_of(th, struct top_thandle,
982 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
986 * Implementation of dt_device_operations::dt_trans_stop() for LOD
988 * Stops the set of local transactions using the targets involved
989 * in declare phase. Initial support for the distributed transactions.
991 * see include/dt_object.h for the details.
993 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
996 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1000 * Implementation of dt_device_operations::dt_conf_get() for LOD
1002 * Currently returns the configuration provided by the local OSD.
1004 * see include/dt_object.h for the details.
1006 static void lod_conf_get(const struct lu_env *env,
1007 const struct dt_device *dev,
1008 struct dt_device_param *param)
1010 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1014 * Implementation of dt_device_operations::dt_sync() for LOD
1016 * Syncs all known OST targets. Very very expensive and used
1017 * rarely by LFSCK now. Should not be used in general.
1019 * see include/dt_object.h for the details.
1021 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1023 struct lod_device *lod = dt2lod_dev(dev);
1024 struct lod_ost_desc *ost;
1029 lod_getref(&lod->lod_ost_descs);
1030 lod_foreach_ost(lod, i) {
1031 ost = OST_TGT(lod, i);
1032 LASSERT(ost && ost->ltd_ost);
1033 rc = dt_sync(env, ost->ltd_ost);
1035 CERROR("%s: can't sync %u: %d\n",
1036 lod2obd(lod)->obd_name, i, rc);
1040 lod_putref(lod, &lod->lod_ost_descs);
1042 rc = dt_sync(env, lod->lod_child);
1048 * Implementation of dt_device_operations::dt_ro() for LOD
1050 * Turns local OSD read-only, used for the testing only.
1052 * see include/dt_object.h for the details.
1054 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1056 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1060 * Implementation of dt_device_operations::dt_commit_async() for LOD
1062 * Asks local OSD to commit sooner.
1064 * see include/dt_object.h for the details.
1066 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1068 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1071 static const struct dt_device_operations lod_dt_ops = {
1072 .dt_root_get = lod_root_get,
1073 .dt_statfs = lod_statfs,
1074 .dt_trans_create = lod_trans_create,
1075 .dt_trans_start = lod_trans_start,
1076 .dt_trans_stop = lod_trans_stop,
1077 .dt_conf_get = lod_conf_get,
1078 .dt_sync = lod_sync,
1080 .dt_commit_async = lod_commit_async,
1081 .dt_trans_cb_add = lod_trans_cb_add,
1085 * Connect to a local OSD.
1087 * Used to connect to the local OSD at mount. OSD name is taken from the
1088 * configuration command passed. This connection is used to identify LU
1089 * site and pin the OSD from early removal.
1091 * \param[in] env LU environment provided by the caller
1092 * \param[in] lod lod device
1093 * \param[in] cfg configuration command to apply
1095 * \retval 0 on success
1096 * \retval negative negated errno on error
1098 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1099 struct lustre_cfg *cfg)
1101 struct obd_connect_data *data = NULL;
1102 struct obd_device *obd;
1103 char *nextdev = NULL, *p, *s;
1108 LASSERT(lod->lod_child_exp == NULL);
1110 /* compatibility hack: we still use old config logs
1111 * which specify LOV, but we need to learn underlying
1112 * OSD device, which is supposed to be:
1113 * <fsname>-MDTxxxx-osd
1115 * 2.x MGS generates lines like the following:
1116 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1117 * 1.8 MGS generates lines like the following:
1118 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1120 * we use "-MDT" to differentiate 2.x from 1.8 */
1122 if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
1123 len = strlen(p) + 1;
1124 OBD_ALLOC(nextdev, len);
1125 if (nextdev == NULL)
1126 GOTO(out, rc = -ENOMEM);
1129 s = strstr(nextdev, "-mdtlov");
1130 if (unlikely(s == NULL)) {
1131 CERROR("unable to parse device name %s\n",
1132 lustre_cfg_string(cfg, 0));
1133 GOTO(out, rc = -EINVAL);
1136 if (strstr(nextdev, "-MDT")) {
1141 strcpy(s, "-MDT0000-osd");
1144 CERROR("unable to parse device name %s\n",
1145 lustre_cfg_string(cfg, 0));
1146 GOTO(out, rc = -EINVAL);
1149 OBD_ALLOC_PTR(data);
1151 GOTO(out, rc = -ENOMEM);
1153 obd = class_name2obd(nextdev);
1155 CERROR("can not locate next device: %s\n", nextdev);
1156 GOTO(out, rc = -ENOTCONN);
1159 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1160 data->ocd_version = LUSTRE_VERSION_CODE;
1162 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1165 CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
1169 lod->lod_dt_dev.dd_lu_dev.ld_site =
1170 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1171 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1172 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1178 OBD_FREE(nextdev, len);
1183 * Allocate and initialize target table.
1185 * A helper function to initialize the target table and allocate
1186 * a bitmap of the available targets.
1188 * \param[in] ltd target's table to initialize
1190 * \retval 0 on success
1191 * \retval negative negated errno on error
1193 static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
1195 mutex_init(<d->ltd_mutex);
1196 init_rwsem(<d->ltd_rw_sem);
1198 /* the OST array and bitmap are allocated/grown dynamically as OSTs are
1199 * added to the LOD, see lod_add_device() */
1200 ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
1201 if (ltd->ltd_tgt_bitmap == NULL)
1204 ltd->ltd_tgts_size = 32;
1207 ltd->ltd_death_row = 0;
1208 ltd->ltd_refcount = 0;
1213 * Initialize LOD device at setup.
1215 * Initializes the given LOD device using the original configuration command.
1216 * The function initiates a connection to the local OSD and initializes few
1217 * internal structures like pools, target tables, etc.
1219 * \param[in] env LU environment provided by the caller
1220 * \param[in] lod lod device
1221 * \param[in] ldt not used
1222 * \param[in] cfg configuration command
1224 * \retval 0 on success
1225 * \retval negative negated errno on error
1227 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1228 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1230 struct dt_device_param ddp;
1231 struct obd_device *obd;
1235 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1237 CERROR("Cannot find obd with name %s\n",
1238 lustre_cfg_string(cfg, 0));
1242 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1243 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1244 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1245 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1247 rc = lod_connect_to_osd(env, lod, cfg);
1251 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1252 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1254 /* setup obd to be used with old lov code */
1255 rc = lod_pools_init(lod, cfg);
1257 GOTO(out_disconnect, rc);
1259 rc = lod_procfs_init(lod);
1261 GOTO(out_pools, rc);
1263 spin_lock_init(&lod->lod_desc_lock);
1264 spin_lock_init(&lod->lod_connects_lock);
1265 lod_tgt_desc_init(&lod->lod_mdt_descs);
1266 lod_tgt_desc_init(&lod->lod_ost_descs);
1271 lod_pools_fini(lod);
1273 obd_disconnect(lod->lod_child_exp);
1278 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1280 * Releases the memory allocated for LOD device.
1282 * see include/lu_object.h for the details.
1284 static struct lu_device *lod_device_free(const struct lu_env *env,
1285 struct lu_device *lu)
1287 struct lod_device *lod = lu2lod_dev(lu);
1288 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1291 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1292 dt_device_fini(&lod->lod_dt_dev);
1298 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1300 * Allocates LOD device and calls the helpers to initialize it.
1302 * see include/lu_object.h for the details.
1304 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1305 struct lu_device_type *type,
1306 struct lustre_cfg *lcfg)
1308 struct lod_device *lod;
1309 struct lu_device *lu_dev;
1313 lu_dev = ERR_PTR(-ENOMEM);
1317 lu_dev = lod2lu_dev(lod);
1318 dt_device_init(&lod->lod_dt_dev, type);
1319 rc = lod_init0(env, lod, type, lcfg);
1321 lod_device_free(env, lu_dev);
1322 lu_dev = ERR_PTR(rc);
1330 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1332 * Releases the internal resources used by LOD device.
1334 * see include/lu_object.h for the details.
1336 static struct lu_device *lod_device_fini(const struct lu_env *env,
1337 struct lu_device *d)
1339 struct lod_device *lod = lu2lod_dev(d);
1343 lod_pools_fini(lod);
1345 lod_procfs_fini(lod);
1347 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
1349 CERROR("%s:can not fini ost descs %d\n",
1350 lod2obd(lod)->obd_name, rc);
1352 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
1354 CERROR("%s:can not fini mdt descs %d\n",
1355 lod2obd(lod)->obd_name, rc);
1361 * Implementation of obd_ops::o_connect() for LOD
1363 * Used to track all the users of this specific LOD device,
1364 * so the device stays up until the last user disconnected.
1366 * \param[in] env LU environment provided by the caller
1367 * \param[out] exp export the caller will be using to access LOD
1368 * \param[in] obd OBD device representing LOD device
1369 * \param[in] cluuid unique identifier of the caller
1370 * \param[in] data not used
1371 * \param[in] localdata not used
1373 * \retval 0 on success
1374 * \retval negative negated errno on error
1376 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1377 struct obd_device *obd, struct obd_uuid *cluuid,
1378 struct obd_connect_data *data, void *localdata)
1380 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1381 struct lustre_handle conn;
1385 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1387 rc = class_connect(&conn, obd, cluuid);
1391 *exp = class_conn2export(&conn);
1393 spin_lock(&lod->lod_connects_lock);
1394 lod->lod_connects++;
1395 /* at the moment we expect the only user */
1396 LASSERT(lod->lod_connects == 1);
1397 spin_unlock(&lod->lod_connects_lock);
1404 * Implementation of obd_ops::o_disconnect() for LOD
1406 * When the caller doesn't need to use this LOD instance, it calls
1407 * obd_disconnect() and LOD releases corresponding export/reference count.
1408 * Once all the users gone, LOD device is released.
1410 * \param[in] exp export provided to the caller in obd_connect()
1412 * \retval 0 on success
1413 * \retval negative negated errno on error
1415 static int lod_obd_disconnect(struct obd_export *exp)
1417 struct obd_device *obd = exp->exp_obd;
1418 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1419 int rc, release = 0;
1422 /* Only disconnect the underlying layers on the final disconnect. */
1423 spin_lock(&lod->lod_connects_lock);
1424 lod->lod_connects--;
1425 if (lod->lod_connects != 0) {
1426 /* why should there be more than 1 connect? */
1427 spin_unlock(&lod->lod_connects_lock);
1428 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
1432 spin_unlock(&lod->lod_connects_lock);
1434 /* the last user of lod has gone, let's release the device */
1438 rc = class_disconnect(exp); /* bz 9811 */
1440 if (rc == 0 && release)
1441 class_manual_cleanup(obd);
1445 LU_KEY_INIT(lod, struct lod_thread_info);
1447 static void lod_key_fini(const struct lu_context *ctx,
1448 struct lu_context_key *key, void *data)
1450 struct lod_thread_info *info = data;
1451 /* allocated in lod_get_lov_ea
1452 * XXX: this is overload, a tread may have such store but used only
1453 * once. Probably better would be pool of such stores per LOD.
1455 if (info->lti_ea_store) {
1456 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
1457 info->lti_ea_store = NULL;
1458 info->lti_ea_store_size = 0;
1460 lu_buf_free(&info->lti_linkea_buf);
1464 /* context key: lod_thread_key */
1465 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
1467 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
1469 static struct lu_device_type_operations lod_device_type_ops = {
1470 .ldto_init = lod_type_init,
1471 .ldto_fini = lod_type_fini,
1473 .ldto_start = lod_type_start,
1474 .ldto_stop = lod_type_stop,
1476 .ldto_device_alloc = lod_device_alloc,
1477 .ldto_device_free = lod_device_free,
1479 .ldto_device_fini = lod_device_fini
1482 static struct lu_device_type lod_device_type = {
1483 .ldt_tags = LU_DEVICE_DT,
1484 .ldt_name = LUSTRE_LOD_NAME,
1485 .ldt_ops = &lod_device_type_ops,
1486 .ldt_ctx_tags = LCT_MD_THREAD,
1490 * Implementation of obd_ops::o_get_info() for LOD
1492 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
1493 * the caller binary status whether LOD has seen connection to any OST target.
1494 * It will also check if the MDT update log context being initialized (if
1497 * \param[in] env LU environment provided by the caller
1498 * \param[in] exp export of the caller
1499 * \param[in] keylen len of the key
1500 * \param[in] key the key
1501 * \param[in] vallen not used
1502 * \param[in] val not used
1504 * \retval 0 if a connection was seen
1505 * \retval -EAGAIN if LOD isn't running yet or no
1506 * connection has been seen yet
1507 * \retval -EINVAL if not supported key is requested
1509 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
1510 __u32 keylen, void *key, __u32 *vallen, void *val)
1514 if (KEY_IS(KEY_OSP_CONNECTED)) {
1515 struct obd_device *obd = exp->exp_obd;
1516 struct lod_device *d;
1517 struct lod_tgt_desc *tgt;
1521 if (!obd->obd_set_up || obd->obd_stopping)
1524 d = lu2lod_dev(obd->obd_lu_dev);
1525 lod_getref(&d->lod_ost_descs);
1526 lod_foreach_ost(d, i) {
1527 tgt = OST_TGT(d, i);
1528 LASSERT(tgt && tgt->ltd_tgt);
1529 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
1531 /* one healthy device is enough */
1535 lod_putref(d, &d->lod_ost_descs);
1537 lod_getref(&d->lod_mdt_descs);
1538 lod_foreach_mdt(d, i) {
1539 struct llog_ctxt *ctxt;
1541 tgt = MDT_TGT(d, i);
1542 LASSERT(tgt != NULL);
1543 LASSERT(tgt->ltd_tgt != NULL);
1544 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
1545 LLOG_UPDATELOG_ORIG_CTXT);
1550 if (ctxt->loc_handle == NULL) {
1552 llog_ctxt_put(ctxt);
1555 llog_ctxt_put(ctxt);
1557 lod_putref(d, &d->lod_mdt_descs);
1565 static struct obd_ops lod_obd_device_ops = {
1566 .o_owner = THIS_MODULE,
1567 .o_connect = lod_obd_connect,
1568 .o_disconnect = lod_obd_disconnect,
1569 .o_get_info = lod_obd_get_info,
1570 .o_pool_new = lod_pool_new,
1571 .o_pool_rem = lod_pool_remove,
1572 .o_pool_add = lod_pool_add,
1573 .o_pool_del = lod_pool_del,
1576 static int __init lod_mod_init(void)
1578 struct obd_type *type;
1581 rc = lu_kmem_init(lod_caches);
1585 rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
1586 LUSTRE_LOD_NAME, &lod_device_type);
1588 lu_kmem_fini(lod_caches);
1592 /* create "lov" entry in procfs for compatibility purposes */
1593 type = class_search_type(LUSTRE_LOV_NAME);
1594 if (type != NULL && type->typ_procroot != NULL)
1597 type = class_search_type(LUSTRE_LOD_NAME);
1598 type->typ_procsym = lprocfs_register("lov", proc_lustre_root,
1600 if (IS_ERR(type->typ_procsym)) {
1601 CERROR("lod: can't create compat entry \"lov\": %d\n",
1602 (int)PTR_ERR(type->typ_procsym));
1603 type->typ_procsym = NULL;
1608 static void __exit lod_mod_exit(void)
1610 class_unregister_type(LUSTRE_LOD_NAME);
1611 lu_kmem_fini(lod_caches);
1614 MODULE_AUTHOR("Intel Corporation. <https://wiki.hpdd.intel.com/>");
1615 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
1616 MODULE_LICENSE("GPL");
1618 module_init(lod_mod_init);
1619 module_exit(lod_mod_exit);