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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/lod/lod_dev.c
33 * Lustre Logical Object Device
35 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
36 * Author: Mikhail Pershin <mike.pershin@intel.com>
39 * The Logical Object Device (LOD) layer manages access to striped
40 * objects (both regular files and directories). It implements the DT
41 * device and object APIs and is responsible for creating, storing,
42 * and loading striping information as an extended attribute of the
43 * underlying OSD object. LOD is the server side analog of the LOV and
44 * LMV layers on the client side.
46 * Metadata LU object stack (layers of the same compound LU object,
47 * all have the same FID):
57 * During LOD object initialization the localness or remoteness of the
58 * object FID dictates the choice between OSD and OSP.
60 * An LOD object (file or directory) with N stripes (each has a
67 * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
69 * When upper layers must access an object's stripes (which are
70 * themselves OST or MDT LU objects) LOD finds these objects by their
71 * FIDs and stores them as an array of DT object pointers on the
72 * object. Declarations and operations on LOD objects are received by
73 * LOD (as DT object operations) and performed on the underlying
74 * OS[DP] object and (as needed) on the stripes. From the perspective
75 * of LOD, a stripe-less file (created by mknod() or open with
76 * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
77 * while a non-striped directory (created by mkdir()) is an object
78 * which will never have stripes.
80 * The LOD layer also implements a small subset of the OBD device API
81 * to support MDT stack initialization and finalization (an MDD device
82 * connects and disconnects itself to and from the underlying LOD
83 * device), and pool management. In turn LOD uses the OBD device API
84 * to connect it self to the underlying OSD, and to connect itself to
85 * OSP devices representing the MDTs and OSTs that bear the stripes of
89 #define DEBUG_SUBSYSTEM S_MDS
91 #include <linux/kthread.h>
92 #include <obd_class.h>
93 #include <md_object.h>
94 #include <lustre_fid.h>
95 #include <uapi/linux/lustre/lustre_param.h>
96 #include <lustre_update.h>
97 #include <lustre_log.h>
98 #include <lustre_lmv.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;
130 if (!fid_is_sane(fid)) {
131 CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
136 if (fid_is_idif(fid)) {
137 *tgt = fid_idif_ost_idx(fid);
138 *type = LU_SEQ_RANGE_OST;
142 if (fid_is_update_log(fid) || fid_is_update_log_dir(fid)) {
144 *type = LU_SEQ_RANGE_MDT;
148 if (!lod->lod_initialized || (!fid_seq_in_fldb(fid_seq(fid)))) {
149 LASSERT(lu_site2seq(lod2lu_dev(lod)->ld_site) != NULL);
151 *tgt = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
152 *type = LU_SEQ_RANGE_MDT;
156 server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
160 fld_range_set_type(&range, *type);
161 rc = fld_server_lookup(env, server_fld, fid_seq(fid), &range);
165 *tgt = range.lsr_index;
166 *type = range.lsr_flags;
168 CDEBUG(D_INFO, "%s: got tgt %x for sequence: %#llx\n",
169 lod2obd(lod)->obd_name, *tgt, fid_seq(fid));
174 /* Slab for OSD object allocation */
175 struct kmem_cache *lod_object_kmem;
177 /* Slab for dt_txn_callback */
178 struct kmem_cache *lod_txn_callback_kmem;
179 static struct lu_kmem_descr lod_caches[] = {
181 .ckd_cache = &lod_object_kmem,
182 .ckd_name = "lod_obj",
183 .ckd_size = sizeof(struct lod_object)
186 .ckd_cache = &lod_txn_callback_kmem,
187 .ckd_name = "lod_txn_callback",
188 .ckd_size = sizeof(struct dt_txn_callback)
195 static struct lu_device *lod_device_fini(const struct lu_env *env,
196 struct lu_device *d);
199 * Implementation of lu_device_operations::ldo_object_alloc() for LOD
201 * Allocates and initializes LOD's slice in the given object.
203 * see include/lu_object.h for the details.
205 static struct lu_object *lod_object_alloc(const struct lu_env *env,
206 const struct lu_object_header *hdr,
207 struct lu_device *dev)
209 struct lod_object *lod_obj;
210 struct lu_object *lu_obj;
214 OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
216 RETURN(ERR_PTR(-ENOMEM));
218 mutex_init(&lod_obj->ldo_layout_mutex);
219 lu_obj = lod2lu_obj(lod_obj);
220 dt_object_init(&lod_obj->ldo_obj, NULL, dev);
221 lod_obj->ldo_obj.do_ops = &lod_obj_ops;
222 lu_obj->lo_ops = &lod_lu_obj_ops;
228 * Process the config log for all sub device.
230 * The function goes through all the targets in the given table
231 * and apply given configuration command on to the targets.
232 * Used to cleanup the targets at unmount.
234 * \param[in] env LU environment provided by the caller
235 * \param[in] lod lod device
236 * \param[in] ltd target's table to go through
237 * \param[in] lcfg configuration command to apply
239 * \retval 0 on success
240 * \retval negative negated errno on error
242 static int lod_sub_process_config(const struct lu_env *env,
243 struct lod_device *lod,
244 struct lod_tgt_descs *ltd,
245 struct lustre_cfg *lcfg)
247 struct lu_device *next;
248 struct lu_tgt_desc *tgt;
252 ltd_foreach_tgt(ltd, tgt) {
255 LASSERT(tgt && tgt->ltd_tgt);
256 next = &tgt->ltd_tgt->dd_lu_dev;
257 rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
259 CERROR("%s: error cleaning up LOD index %u: cmd %#x : rc = %d\n",
260 lod2obd(lod)->obd_name, tgt->ltd_index,
261 lcfg->lcfg_command, rc1);
265 lod_putref(lod, ltd);
269 struct lod_recovery_data {
270 struct lod_device *lrd_lod;
271 struct lod_tgt_desc *lrd_ltd;
272 struct task_struct **lrd_task;
274 struct lu_env lrd_env;
275 struct completion *lrd_started;
280 * process update recovery record
282 * Add the update recovery recode to the update recovery list in
283 * lod_recovery_data. Then the recovery thread (target_recovery_thread)
284 * will redo these updates.
286 * \param[in]env execution environment
287 * \param[in]llh log handle of update record
288 * \param[in]rec update record to be replayed
289 * \param[in]data update recovery data which holds the necessary
290 * arguments for recovery (see struct lod_recovery_data)
292 * \retval 0 if the record is processed successfully.
293 * \retval negative errno if the record processing fails.
295 static int lod_process_recovery_updates(const struct lu_env *env,
296 struct llog_handle *llh,
297 struct llog_rec_hdr *rec,
300 struct lod_recovery_data *lrd = data;
301 struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
302 struct lu_target *lut;
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 "DFID".%u: rc = %d\n",
320 lod2obd(lrd->lrd_lod)->obd_name, index,
321 PFID(&llh->lgh_id.lgl_oi.oi_fid), 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 "DFID".%u\n",
330 lod2obd(lrd->lrd_lod)->obd_name,
331 PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index);
332 lut = lod2lu_dev(lrd->lrd_lod)->ld_site->ls_tgt;
334 if (lut->lut_obd->obd_stopping ||
335 lut->lut_obd->obd_abort_recovery)
338 return insert_update_records_to_replay_list(lut->lut_tdtd,
339 (struct llog_update_record *)rec,
344 * recovery thread for update log
346 * Start recovery thread and prepare the sub llog, then it will retrieve
347 * the update records from the correpondent MDT and do recovery.
349 * \param[in] arg pointer to the recovery data
351 * \retval 0 if recovery succeeds
352 * \retval negative errno if recovery failed.
354 static int lod_sub_recovery_thread(void *arg)
356 struct lod_recovery_data *lrd = arg;
357 struct lod_device *lod = lrd->lrd_lod;
358 struct dt_device *dt;
359 struct llog_ctxt *ctxt = NULL;
360 struct lu_env *env = &lrd->lrd_env;
361 struct lu_target *lut;
362 struct lu_tgt_desc *mdt = NULL;
369 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
370 atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
374 dt = lrd->lrd_ltd->ltd_tgt;
376 start = ktime_get_real_seconds();
377 complete(lrd->lrd_started);
381 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT)) &&
383 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val);
386 rc = lod_sub_prep_llog(env, lod, dt, lrd->lrd_idx);
388 if (!rc && !lod->lod_child->dd_rdonly) {
389 /* Process the recovery record */
390 ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
391 LLOG_UPDATELOG_ORIG_CTXT);
392 LASSERT(ctxt != NULL);
393 LASSERT(ctxt->loc_handle != NULL);
395 rc = llog_cat_process(env, ctxt->loc_handle,
396 lod_process_recovery_updates, lrd, 0, 0);
400 struct lu_device *top_device;
402 top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
404 * Because the remote target might failover at the same time,
407 if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
408 dt != lod->lod_child &&
409 !top_device->ld_obd->obd_abort_recovery &&
410 !top_device->ld_obd->obd_stopping) {
412 if (ctxt->loc_handle)
418 CDEBUG(D_HA, "%s get update log failed %d, retry\n",
419 dt->dd_lu_dev.ld_obd->obd_name, rc);
423 CERROR("%s get update log failed: rc = %d\n",
424 dt->dd_lu_dev.ld_obd->obd_name, rc);
427 spin_lock(&top_device->ld_obd->obd_dev_lock);
428 if (!top_device->ld_obd->obd_abort_recovery &&
429 !top_device->ld_obd->obd_stopping)
430 top_device->ld_obd->obd_abort_recovery = 1;
431 spin_unlock(&top_device->ld_obd->obd_dev_lock);
437 CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
438 dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
441 spin_lock(&lod->lod_lock);
443 lod->lod_child_got_update_log = 1;
445 lrd->lrd_ltd->ltd_got_update_log = 1;
447 if (!lod->lod_child_got_update_log) {
448 spin_unlock(&lod->lod_lock);
452 lod_foreach_mdt(lod, mdt) {
453 if (!mdt->ltd_got_update_log) {
454 spin_unlock(&lod->lod_lock);
458 lut->lut_tdtd->tdtd_replay_ready = 1;
459 spin_unlock(&lod->lod_lock);
461 CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
462 lut->lut_obd->obd_name);
463 wake_up(&lut->lut_obd->obd_next_transno_waitq);
467 atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
468 wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
469 if (xchg(lrd->lrd_task, NULL) == NULL)
470 /* Someone is waiting for us to finish, need
471 * to synchronize cleanly.
473 wait_var_event(lrd, kthread_should_stop());
480 * finish sub llog context
482 * Stop update recovery thread for the sub device, then cleanup the
483 * correspondent llog ctxt.
485 * \param[in] env execution environment
486 * \param[in] lod lod device to do update recovery
487 * \param[in] thread recovery thread on this sub device
489 void lod_sub_fini_llog(const struct lu_env *env,
490 struct dt_device *dt, struct task_struct **thread)
492 struct obd_device *obd;
493 struct llog_ctxt *ctxt;
494 struct task_struct *task = NULL;
498 obd = dt->dd_lu_dev.ld_obd;
499 CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
500 /* Wait for recovery thread to complete */
502 task = xchg(thread, NULL);
506 ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
510 if (ctxt->loc_handle)
511 llog_cat_close(env, ctxt->loc_handle);
513 llog_cleanup(env, ctxt);
519 * Extract MDT target index from a device name.
521 * a helper function to extract index from the given device name
522 * like "fsname-MDTxxxx-mdtlov"
524 * \param[in] lodname device name
525 * \param[out] mdt_index extracted index
527 * \retval 0 on success
528 * \retval -EINVAL if the name is invalid
530 int lodname2mdt_index(char *lodname, u32 *mdt_index)
533 const char *ptr, *tmp;
536 /* 1.8 configs don't have "-MDT0000" at the end */
537 ptr = strstr(lodname, "-MDT");
543 ptr = strrchr(lodname, '-');
546 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
550 if (strncmp(ptr, "-mdtlov", 7) != 0) {
552 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
556 if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
558 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
562 if (strncmp(ptr - 8, "-MDT", 4) != 0) {
564 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
568 rc = target_name2index(ptr - 7, &index, &tmp);
569 if (rc < 0 || rc & LDD_F_SV_ALL || *tmp != '-') {
571 CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
579 * Init sub llog context
581 * Setup update llog ctxt for update recovery threads, then start the
582 * recovery thread (lod_sub_recovery_thread) to read update llog from
583 * the correspondent MDT to do update recovery.
585 * \param[in] env execution environment
586 * \param[in] lod lod device to do update recovery
587 * \param[in] dt sub dt device for which the recovery thread is
589 * \retval 0 if initialization succeeds.
590 * \retval negative errno if initialization fails.
592 int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
593 struct dt_device *dt)
595 struct obd_device *obd;
596 struct lod_recovery_data *lrd = NULL;
597 DECLARE_COMPLETION_ONSTACK(started);
598 struct task_struct **taskp;
599 struct task_struct *task;
600 struct lod_tgt_desc *subtgt = NULL;
607 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
615 if (lod->lod_child == dt) {
616 taskp = &lod->lod_child_recovery_task;
617 index = master_index;
619 struct lu_tgt_desc *mdt;
621 lod_foreach_mdt(lod, mdt) {
622 if (mdt->ltd_tgt == dt) {
623 index = mdt->ltd_index;
628 LASSERT(subtgt != NULL);
629 taskp = &subtgt->ltd_recovery_task;
632 CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
633 dt->dd_lu_dev.ld_obd->obd_name);
635 lrd->lrd_ltd = subtgt;
636 lrd->lrd_task = taskp;
637 lrd->lrd_idx = index;
638 lrd->lrd_started = &started;
640 obd = dt->dd_lu_dev.ld_obd;
641 obd->obd_lvfs_ctxt.dt = dt;
642 rc = llog_setup(env, obd, &obd->obd_olg, LLOG_UPDATELOG_ORIG_CTXT,
643 NULL, &llog_common_cat_ops);
645 CERROR("%s: cannot setup updatelog llog: rc = %d\n",
650 rc = lu_env_init(&lrd->lrd_env, LCT_LOCAL | LCT_MD_THREAD);
652 CERROR("%s: can't initialize env: rc = %d\n",
653 lod2obd(lod)->obd_name, rc);
657 /* Start the recovery thread */
658 task = kthread_create(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
659 master_index, index);
662 CERROR("%s: cannot start recovery thread: rc = %d\n",
664 lu_env_fini(&lrd->lrd_env);
668 wake_up_process(task);
669 wait_for_completion(&started);
673 lod_sub_fini_llog(env, dt, taskp);
680 * Stop sub recovery thread
682 * Stop sub recovery thread on all subs.
684 * \param[in] env execution environment
685 * \param[in] lod lod device to do update recovery
687 static void lod_sub_stop_recovery_threads(const struct lu_env *env,
688 struct lod_device *lod)
690 struct task_struct *task;
691 struct lu_tgt_desc *mdt;
694 * Stop the update log commit cancel threads and finish master
697 task = xchg(&lod->lod_child_recovery_task, NULL);
701 lod_getref(&lod->lod_mdt_descs);
702 lod_foreach_mdt(lod, mdt) {
703 task = xchg(&mdt->ltd_recovery_task, NULL);
707 lod_putref(lod, &lod->lod_mdt_descs);
711 * finish all sub llog
713 * cleanup all of sub llog ctxt on the LOD.
715 * \param[in] env execution environment
716 * \param[in] lod lod device to do update recovery
718 static void lod_sub_fini_all_llogs(const struct lu_env *env,
719 struct lod_device *lod)
721 struct lu_tgt_desc *mdt;
724 * Stop the update log commit cancel threads and finish master
727 lod_sub_fini_llog(env, lod->lod_child,
728 &lod->lod_child_recovery_task);
729 lod_getref(&lod->lod_mdt_descs);
730 lod_foreach_mdt(lod, mdt)
731 lod_sub_fini_llog(env, mdt->ltd_tgt,
732 &mdt->ltd_recovery_task);
733 lod_putref(lod, &lod->lod_mdt_descs);
736 static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
738 struct lod_device *lod = (struct lod_device *)data;
739 struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
740 struct lu_tgt_desc *mdt = NULL;
746 *count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
752 *size = 5 * *count + 1;
753 OBD_ALLOC(buf, *size);
758 memset(buf, 0, *size);
760 if (!lod->lod_child_got_update_log) {
761 rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
762 LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
764 rc = scnprintf(buf + len, *size - len, " %04x", i);
771 lod_foreach_mdt(lod, mdt) {
772 if (!mdt->ltd_got_update_log) {
773 rc = scnprintf(buf + len, *size - len, " %04x",
775 if (unlikely(rc <= 0))
787 * Prepare distribute txn
789 * Prepare distribute txn structure for LOD
791 * \param[in] env execution environment
792 * \param[in] lod_device LOD device
794 * \retval 0 if preparation succeeds.
795 * \retval negative errno if preparation fails.
797 static int lod_prepare_distribute_txn(const struct lu_env *env,
798 struct lod_device *lod)
800 struct target_distribute_txn_data *tdtd;
801 struct lu_target *lut;
806 /* Init update recovery data */
811 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
812 tdtd->tdtd_dt = &lod->lod_dt_dev;
813 rc = distribute_txn_init(env, lut, tdtd,
814 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id);
817 CERROR("%s: cannot init distribute txn: rc = %d\n",
818 lod2obd(lod)->obd_name, rc);
823 tdtd->tdtd_show_update_logs_retrievers =
824 lod_show_update_logs_retrievers;
825 tdtd->tdtd_show_retrievers_cbdata = lod;
827 lut->lut_tdtd = tdtd;
833 * Finish distribute txn
835 * Release the resource holding by distribute txn, i.e. stop distribute
838 * \param[in] env execution environment
839 * \param[in] lod lod device
841 static void lod_fini_distribute_txn(const struct lu_env *env,
842 struct lod_device *lod)
844 struct lu_target *lut;
846 lut = lod2lu_dev(lod)->ld_site->ls_tgt;
847 target_recovery_fini(lut->lut_obd);
851 distribute_txn_fini(env, lut->lut_tdtd);
853 OBD_FREE_PTR(lut->lut_tdtd);
854 lut->lut_tdtd = NULL;
858 * Implementation of lu_device_operations::ldo_process_config() for LOD
860 * The method is called by the configuration subsystem during setup,
861 * cleanup and when the configuration changes. The method processes
862 * few specific commands like adding/removing the targets, changing
863 * the runtime parameters.
865 * \param[in] env LU environment provided by the caller
866 * \param[in] dev lod device
867 * \param[in] lcfg configuration command to apply
869 * \retval 0 on success
870 * \retval negative negated errno on error
872 * The examples are below.
874 * Add osc config log:
875 * marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
876 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
877 * attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
878 * setup 0:lustre-OST0001-osc-MDT0001 1:lustre-OST0001_UUID 2:nid
879 * lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
880 * marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
882 * Add mdc config log:
883 * marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
884 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
885 * attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
886 * setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
887 * modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
888 * marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
890 static int lod_process_config(const struct lu_env *env,
891 struct lu_device *dev,
892 struct lustre_cfg *lcfg)
894 struct lod_device *lod = lu2lod_dev(dev);
895 struct lu_device *next = &lod->lod_child->dd_lu_dev;
901 switch (lcfg->lcfg_command) {
902 case LCFG_LOV_DEL_OBD:
903 case LCFG_LOV_ADD_INA:
904 case LCFG_LOV_ADD_OBD:
910 * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
911 * modify_mdc_tgts add 0:lustre-MDT0001
912 * 1:lustre-MDT0001-mdc0002
915 arg1 = lustre_cfg_string(lcfg, 1);
917 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
918 GOTO(out, rc = -EINVAL);
919 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
920 GOTO(out, rc = -EINVAL);
922 if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
925 rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
930 rc = lod_add_device(env, lod, arg1, index, gen,
931 mdt_index, LUSTRE_OSC_NAME, 1);
932 } else if (lcfg->lcfg_command == LCFG_ADD_MDC) {
934 rc = lod_add_device(env, lod, arg1, index, gen,
935 mdt_index, LUSTRE_MDC_NAME, 1);
936 } else if (lcfg->lcfg_command == LCFG_LOV_ADD_INA) {
937 /*FIXME: Add mdt_index for LCFG_LOV_ADD_INA*/
939 rc = lod_add_device(env, lod, arg1, index, gen,
940 mdt_index, LUSTRE_OSC_NAME, 0);
942 rc = lod_del_device(env, lod, &lod->lod_ost_descs,
950 struct obd_device *obd;
955 * Check if it is activate/deactivate mdc
956 * lustre-MDTXXXX-osp-MDTXXXX.active=1
958 param = lustre_cfg_buf(lcfg, 1);
959 if (strstr(param, "osp") && strstr(param, ".active=")) {
960 struct lod_tgt_desc *sub_tgt = NULL;
961 struct lu_tgt_desc *mdt;
965 ptr = strstr(param, ".");
967 obd = class_name2obd(param);
969 CERROR("%s: can not find %s: rc = %d\n",
970 lod2obd(lod)->obd_name, param, -EINVAL);
975 lod_foreach_mdt(lod, mdt) {
976 if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
983 CERROR("%s: can not find %s: rc = %d\n",
984 lod2obd(lod)->obd_name, param, -EINVAL);
990 tmp = strstr(param, "=");
993 struct llog_ctxt *ctxt;
995 obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
996 ctxt = llog_get_context(obd,
997 LLOG_UPDATELOG_ORIG_CTXT);
999 rc = llog_setup(env, obd, &obd->obd_olg,
1000 LLOG_UPDATELOG_ORIG_CTXT,
1001 NULL, &llog_common_cat_ops);
1005 llog_ctxt_put(ctxt);
1007 rc = lod_sub_prep_llog(env, lod,
1009 sub_tgt->ltd_index);
1011 sub_tgt->ltd_active = 1;
1013 lod_sub_fini_llog(env, sub_tgt->ltd_tgt,
1015 sub_tgt->ltd_active = 0;
1021 if (strstr(param, PARAM_LOD) != NULL)
1022 count = class_modify_config(lcfg, PARAM_LOD,
1023 &lod->lod_dt_dev.dd_kobj);
1025 count = class_modify_config(lcfg, PARAM_LOV,
1026 &lod->lod_dt_dev.dd_kobj);
1027 rc = count > 0 ? 0 : count;
1030 case LCFG_PRE_CLEANUP: {
1031 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1032 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1033 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val * 2);
1034 next = &lod->lod_child->dd_lu_dev;
1035 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1037 CDEBUG(D_HA, "%s: can't process %u: %d\n",
1038 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1040 lod_sub_stop_recovery_threads(env, lod);
1041 lod_fini_distribute_txn(env, lod);
1042 lod_sub_fini_all_llogs(env, lod);
1045 case LCFG_CLEANUP: {
1046 if (lod->lod_md_root) {
1047 dt_object_put(env, &lod->lod_md_root->ldo_obj);
1048 lod->lod_md_root = NULL;
1052 * do cleanup on underlying storage only when
1053 * all OSPs are cleaned up, as they use that OSD as well
1055 lu_dev_del_linkage(dev->ld_site, dev);
1056 lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
1057 lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
1058 next = &lod->lod_child->dd_lu_dev;
1059 rc = next->ld_ops->ldo_process_config(env, next, lcfg);
1061 CERROR("%s: can't process %u: rc = %d\n",
1062 lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
1064 rc = obd_disconnect(lod->lod_child_exp);
1066 CERROR("error in disconnect from storage: rc = %d\n",
1071 CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
1072 lcfg->lcfg_command);
1082 * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
1084 * The method is called once the recovery is complete. This implementation
1085 * distributes the notification to all the known targets.
1087 * see include/lu_object.h for the details
1089 static int lod_recovery_complete(const struct lu_env *env,
1090 struct lu_device *dev)
1092 struct lod_device *lod = lu2lod_dev(dev);
1093 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1094 struct lod_tgt_desc *tgt;
1099 LASSERT(lod->lod_recovery_completed == 0);
1100 lod->lod_recovery_completed = 1;
1102 rc = next->ld_ops->ldo_recovery_complete(env, next);
1104 lod_getref(&lod->lod_ost_descs);
1105 if (lod->lod_ost_descs.ltd_tgts_size > 0) {
1106 lod_foreach_ost(lod, tgt) {
1107 LASSERT(tgt && tgt->ltd_tgt);
1108 next = &tgt->ltd_tgt->dd_lu_dev;
1109 rc = next->ld_ops->ldo_recovery_complete(env, next);
1111 CERROR("%s: can't complete recovery on #%d: rc = %d\n",
1112 lod2obd(lod)->obd_name, tgt->ltd_index,
1116 lod_putref(lod, &lod->lod_ost_descs);
1121 * Init update logs on all sub device
1123 * LOD initialize update logs on all of sub devices. Because the initialization
1124 * process might need FLD lookup, see llog_osd_open()->dt_locate()->...->
1125 * lod_object_init(), this API has to be called after LOD is initialized.
1126 * \param[in] env execution environment
1127 * \param[in] lod lod device
1129 * \retval 0 if update log is initialized successfully.
1130 * \retval negative errno if initialization fails.
1132 static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
1134 struct lu_tgt_desc *mdt;
1140 * llog must be setup after LOD is initialized, because llog
1141 * initialization include FLD lookup
1143 LASSERT(lod->lod_initialized);
1145 /* Init the llog in its own stack */
1146 rc = lod_sub_init_llog(env, lod, lod->lod_child);
1150 lod_foreach_mdt(lod, mdt) {
1151 rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
1160 * Implementation of lu_device_operations::ldo_prepare() for LOD
1162 * see include/lu_object.h for the details.
1164 static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
1165 struct lu_device *cdev)
1167 struct lod_device *lod = lu2lod_dev(cdev);
1168 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1169 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1171 struct dt_object *root;
1172 struct dt_object *dto;
1177 rc = next->ld_ops->ldo_prepare(env, pdev, next);
1179 CERROR("%s: prepare bottom error: rc = %d\n",
1180 lod2obd(lod)->obd_name, rc);
1184 lod->lod_initialized = 1;
1186 rc = dt_root_get(env, lod->lod_child, fid);
1190 root = dt_locate(env, lod->lod_child, fid);
1192 RETURN(PTR_ERR(root));
1194 /* Create update log object */
1195 index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1196 lu_update_log_fid(fid, index);
1198 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1200 lod_update_log_name,
1203 GOTO(out_put, rc = PTR_ERR(dto));
1205 dt_object_put(env, dto);
1207 /* Create update log dir */
1208 lu_update_log_dir_fid(fid, index);
1209 dto = local_file_find_or_create_with_fid(env, lod->lod_child,
1211 lod_update_log_dir_name,
1214 GOTO(out_put, rc = PTR_ERR(dto));
1216 dt_object_put(env, dto);
1218 rc = lod_prepare_distribute_txn(env, lod);
1222 rc = lod_sub_init_llogs(env, lod);
1227 dt_object_put(env, root);
1233 * Implementation of lu_device_operations::ldo_fid_alloc() for LOD
1235 * Find corresponding device by passed parent and name, and allocate FID from
1238 * see include/lu_object.h for the details.
1240 static int lod_fid_alloc(const struct lu_env *env, struct lu_device *d,
1241 struct lu_fid *fid, struct lu_object *parent,
1242 const struct lu_name *name)
1244 struct lod_device *lod = lu2lod_dev(d);
1245 struct lod_object *lo = lu2lod_obj(parent);
1246 struct dt_device *next;
1251 /* if @parent is remote, we don't know whether its layout was changed,
1252 * always reload layout.
1254 if (lu_object_remote(parent))
1255 lod_striping_free(env, lo);
1257 rc = lod_striping_load(env, lo);
1261 if (lo->ldo_dir_stripe_count > 0 && name) {
1262 struct dt_object *stripe;
1265 idx = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
1266 lo->ldo_dir_stripe_count,
1267 lo->ldo_dir_migrate_hash,
1268 lo->ldo_dir_migrate_offset,
1270 name->ln_namelen, true);
1274 stripe = lo->ldo_stripe[idx];
1275 if (!stripe || !dt_object_exists(stripe))
1278 next = lu2dt_dev(stripe->do_lu.lo_dev);
1280 next = lod->lod_child;
1283 rc = dt_fid_alloc(env, next, fid, parent, name);
1288 const struct lu_device_operations lod_lu_ops = {
1289 .ldo_object_alloc = lod_object_alloc,
1290 .ldo_process_config = lod_process_config,
1291 .ldo_recovery_complete = lod_recovery_complete,
1292 .ldo_prepare = lod_prepare,
1293 .ldo_fid_alloc = lod_fid_alloc,
1297 * Implementation of dt_device_operations::dt_root_get() for LOD
1299 * see include/dt_object.h for the details.
1301 static int lod_root_get(const struct lu_env *env,
1302 struct dt_device *dev, struct lu_fid *f)
1304 return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
1307 static void lod_statfs_sum(struct obd_statfs *sfs,
1308 struct obd_statfs *ost_sfs, int *bs)
1310 while (ost_sfs->os_bsize < *bs) {
1312 sfs->os_bsize >>= 1;
1313 sfs->os_bavail <<= 1;
1314 sfs->os_blocks <<= 1;
1315 sfs->os_bfree <<= 1;
1316 sfs->os_granted <<= 1;
1318 while (ost_sfs->os_bsize > *bs) {
1319 ost_sfs->os_bsize >>= 1;
1320 ost_sfs->os_bavail <<= 1;
1321 ost_sfs->os_blocks <<= 1;
1322 ost_sfs->os_bfree <<= 1;
1323 ost_sfs->os_granted <<= 1;
1325 sfs->os_bavail += ost_sfs->os_bavail;
1326 sfs->os_blocks += ost_sfs->os_blocks;
1327 sfs->os_bfree += ost_sfs->os_bfree;
1328 sfs->os_granted += ost_sfs->os_granted;
1332 * Implementation of dt_device_operations::dt_statfs() for LOD
1334 * see include/dt_object.h for the details.
1336 static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
1337 struct obd_statfs *sfs, struct obd_statfs_info *info)
1339 struct lod_device *lod = dt2lod_dev(dev);
1340 struct lu_tgt_desc *tgt;
1341 struct obd_statfs ost_sfs;
1346 rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
1355 sfs->os_granted = 0;
1357 lod_getref(&lod->lod_mdt_descs);
1358 lod_foreach_mdt(lod, tgt) {
1359 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1363 sfs->os_files += ost_sfs.os_files;
1364 sfs->os_ffree += ost_sfs.os_ffree;
1365 lod_statfs_sum(sfs, &ost_sfs, &bs);
1367 lod_putref(lod, &lod->lod_mdt_descs);
1370 * at some point we can check whether DoM is enabled and
1371 * decide how to account MDT space. for simplicity let's
1372 * just fallback to pre-DoM policy if any OST is alive
1374 lod_getref(&lod->lod_ost_descs);
1375 lod_foreach_ost(lod, tgt) {
1376 rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
1378 if (rc || ost_sfs.os_bsize == 0)
1382 * if only MDTs with DoM then report only MDT blocks,
1383 * otherwise show only OST blocks, and DoM is "free"
1388 sfs->os_granted = 0;
1390 ost_files += ost_sfs.os_files;
1391 ost_ffree += ost_sfs.os_ffree;
1392 ost_sfs.os_bavail += ost_sfs.os_granted;
1393 lod_statfs_sum(sfs, &ost_sfs, &bs);
1394 LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
1395 (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
1397 lod_putref(lod, &lod->lod_ost_descs);
1398 sfs->os_state |= OS_STATFS_SUM;
1400 /* If we have _some_ OSTs, but don't have as many free objects on the
1401 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1402 * to compensate, so that the "inodes in use" number is correct.
1403 * This should be kept in sync with ll_statfs_internal().
1405 if (ost_files && ost_ffree < sfs->os_ffree) {
1406 sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
1407 sfs->os_ffree = ost_ffree;
1410 /* a single successful statfs should be enough */
1418 * Implementation of dt_device_operations::dt_trans_create() for LOD
1420 * Creates a transaction using local (to this node) OSD.
1422 * see include/dt_object.h for the details.
1424 static struct thandle *lod_trans_create(const struct lu_env *env,
1425 struct dt_device *dt)
1429 th = top_trans_create(env, dt2lod_dev(dt)->lod_child);
1439 * Implementation of dt_device_operations::dt_trans_start() for LOD
1441 * Starts the set of local transactions using the targets involved
1442 * in declare phase. Initial support for the distributed transactions.
1444 * see include/dt_object.h for the details.
1446 static int lod_trans_start(const struct lu_env *env, struct dt_device *dt,
1449 return top_trans_start(env, dt2lod_dev(dt)->lod_child, th);
1452 static int lod_trans_cb_add(struct thandle *th,
1453 struct dt_txn_commit_cb *dcb)
1455 struct top_thandle *top_th = container_of(th, struct top_thandle,
1457 return dt_trans_cb_add(top_th->tt_master_sub_thandle, dcb);
1461 * add noop update to the update records
1463 * Add noop updates to the update records, which is only used in
1466 * \param[in] env execution environment
1467 * \param[in] dt dt device of lod
1468 * \param[in] th thandle
1469 * \param[in] count the count of update records to be added.
1471 * \retval 0 if adding succeeds.
1472 * \retval negative errno if adding fails.
1474 static int lod_add_noop_records(const struct lu_env *env,
1475 struct dt_device *dt, struct thandle *th,
1478 struct top_thandle *top_th;
1479 struct lu_fid *fid = &lod_env_info(env)->lti_fid;
1483 top_th = container_of(th, struct top_thandle, tt_super);
1484 if (!top_th->tt_multiple_thandle)
1488 for (i = 0; i < count; i++) {
1489 rc = update_record_pack(noop, th, fid);
1497 * Implementation of dt_device_operations::dt_trans_stop() for LOD
1499 * Stops the set of local transactions using the targets involved
1500 * in declare phase. Initial support for the distributed transactions.
1502 * see include/dt_object.h for the details.
1504 static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
1507 if (OBD_FAIL_CHECK(OBD_FAIL_SPLIT_UPDATE_REC)) {
1510 rc = lod_add_noop_records(env, dt, th, 5000);
1514 return top_trans_stop(env, dt2lod_dev(dt)->lod_child, th);
1518 * Implementation of dt_device_operations::dt_conf_get() for LOD
1520 * Currently returns the configuration provided by the local OSD.
1522 * see include/dt_object.h for the details.
1524 static void lod_conf_get(const struct lu_env *env,
1525 const struct dt_device *dev,
1526 struct dt_device_param *param)
1528 dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
1532 * Implementation of dt_device_operations::dt_sync() for LOD
1534 * Syncs all known OST targets. Very very expensive and used
1535 * rarely by LFSCK now. Should not be used in general.
1537 * see include/dt_object.h for the details.
1539 static int lod_sync(const struct lu_env *env, struct dt_device *dev)
1541 struct lod_device *lod = dt2lod_dev(dev);
1542 struct lu_tgt_desc *tgt;
1547 lod_getref(&lod->lod_ost_descs);
1548 lod_foreach_ost(lod, tgt) {
1549 if (!tgt->ltd_active)
1551 rc = dt_sync(env, tgt->ltd_tgt);
1553 if (rc != -ENOTCONN) {
1554 CERROR("%s: can't sync ost %u: rc = %d\n",
1555 lod2obd(lod)->obd_name, tgt->ltd_index,
1562 lod_putref(lod, &lod->lod_ost_descs);
1567 lod_getref(&lod->lod_mdt_descs);
1568 lod_foreach_mdt(lod, tgt) {
1569 if (!tgt->ltd_active)
1571 rc = dt_sync(env, tgt->ltd_tgt);
1573 if (rc != -ENOTCONN) {
1574 CERROR("%s: can't sync mdt %u: rc = %d\n",
1575 lod2obd(lod)->obd_name, tgt->ltd_index,
1582 lod_putref(lod, &lod->lod_mdt_descs);
1585 rc = dt_sync(env, lod->lod_child);
1591 * Implementation of dt_device_operations::dt_ro() for LOD
1593 * Turns local OSD read-only, used for the testing only.
1595 * see include/dt_object.h for the details.
1597 static int lod_ro(const struct lu_env *env, struct dt_device *dev)
1599 return dt_ro(env, dt2lod_dev(dev)->lod_child);
1603 * Implementation of dt_device_operations::dt_commit_async() for LOD
1605 * Asks local OSD to commit sooner.
1607 * see include/dt_object.h for the details.
1609 static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
1611 return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
1614 static const struct dt_device_operations lod_dt_ops = {
1615 .dt_root_get = lod_root_get,
1616 .dt_statfs = lod_statfs,
1617 .dt_trans_create = lod_trans_create,
1618 .dt_trans_start = lod_trans_start,
1619 .dt_trans_stop = lod_trans_stop,
1620 .dt_conf_get = lod_conf_get,
1621 .dt_sync = lod_sync,
1623 .dt_commit_async = lod_commit_async,
1624 .dt_trans_cb_add = lod_trans_cb_add,
1628 * Connect to a local OSD.
1630 * Used to connect to the local OSD at mount. OSD name is taken from the
1631 * configuration command passed. This connection is used to identify LU
1632 * site and pin the OSD from early removal.
1634 * \param[in] env LU environment provided by the caller
1635 * \param[in] lod lod device
1636 * \param[in] cfg configuration command to apply
1638 * \retval 0 on success
1639 * \retval negative negated errno on error
1641 static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
1642 struct lustre_cfg *cfg)
1644 struct obd_connect_data *data = NULL;
1645 struct obd_device *obd;
1646 char *nextdev = NULL, *p, *s;
1652 LASSERT(lod->lod_child_exp == NULL);
1655 * compatibility hack: we still use old config logs
1656 * which specify LOV, but we need to learn underlying
1657 * OSD device, which is supposed to be:
1658 * <fsname>-MDTxxxx-osd
1660 * 2.x MGS generates lines like the following:
1661 * #03 (176)lov_setup 0:lustre-MDT0000-mdtlov 1:(struct lov_desc)
1662 * 1.8 MGS generates lines like the following:
1663 * #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
1665 * we use "-MDT" to differentiate 2.x from 1.8
1667 p = lustre_cfg_string(cfg, 0);
1668 if (p && strstr(p, "-mdtlov")) {
1669 len = strlen(p) + 6;
1670 OBD_ALLOC(nextdev, len);
1672 GOTO(out, rc = -ENOMEM);
1675 s = strstr(nextdev, "-mdtlov");
1677 CERROR("%s: unable to parse device name: rc = %d\n",
1678 lustre_cfg_string(cfg, 0), -EINVAL);
1679 GOTO(out, rc = -EINVAL);
1682 if (strstr(nextdev, "-MDT")) {
1687 strcpy(s, "-MDT0000-osd");
1690 CERROR("%s: unable to parse device name: rc = %d\n",
1691 lustre_cfg_string(cfg, 0), -EINVAL);
1692 GOTO(out, rc = -EINVAL);
1695 OBD_ALLOC_PTR(data);
1697 GOTO(out, rc = -ENOMEM);
1699 obd = class_name2obd(nextdev);
1701 CERROR("%s: can not locate next device: rc = %d\n",
1702 nextdev, -ENOTCONN);
1703 GOTO(out, rc = -ENOTCONN);
1706 data->ocd_connect_flags = OBD_CONNECT_VERSION;
1707 data->ocd_version = LUSTRE_VERSION_CODE;
1709 rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
1712 CERROR("%s: cannot connect to next dev: rc = %d\n",
1717 lod->lod_dt_dev.dd_lu_dev.ld_site =
1718 lod->lod_child_exp->exp_obd->obd_lu_dev->ld_site;
1719 LASSERT(lod->lod_dt_dev.dd_lu_dev.ld_site);
1720 lod->lod_child = lu2dt_dev(lod->lod_child_exp->exp_obd->obd_lu_dev);
1726 OBD_FREE(nextdev, len);
1730 static int lod_lsfs_init(const struct lu_env *env, struct lod_device *d)
1732 struct obd_statfs sfs;
1735 rc = dt_statfs(env, d->lod_child, &sfs);
1737 CDEBUG(D_LAYOUT, "%s: failed to get OSD statfs, rc = %d\n",
1738 lod2obd(d)->obd_name, rc);
1742 /* udpate local OSD cached statfs data */
1743 spin_lock_init(&d->lod_lsfs_lock);
1744 d->lod_lsfs_age = ktime_get_seconds();
1745 d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
1746 d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
1751 * Initialize LOD device at setup.
1753 * Initializes the given LOD device using the original configuration command.
1754 * The function initiates a connection to the local OSD and initializes few
1755 * internal structures like pools, target tables, etc.
1757 * \param[in] env LU environment provided by the caller
1758 * \param[in] lod lod device
1759 * \param[in] ldt not used
1760 * \param[in] cfg configuration command
1762 * \retval 0 on success
1763 * \retval negative negated errno on error
1765 static int lod_init0(const struct lu_env *env, struct lod_device *lod,
1766 struct lu_device_type *ldt, struct lustre_cfg *cfg)
1768 struct dt_device_param ddp;
1769 struct obd_device *obd;
1774 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1777 CERROR("Cannot find obd with name '%s': rc = %d\n",
1778 lustre_cfg_string(cfg, 0), rc);
1782 obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
1783 lod->lod_dt_dev.dd_lu_dev.ld_obd = obd;
1784 lod->lod_dt_dev.dd_lu_dev.ld_ops = &lod_lu_ops;
1785 lod->lod_dt_dev.dd_ops = &lod_dt_ops;
1787 rc = lod_connect_to_osd(env, lod, cfg);
1791 dt_conf_get(env, &lod->lod_dt_dev, &ddp);
1792 lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
1793 lod->lod_dom_stripesize_max_kb = (1ULL << 10); /* 1Mb is default */
1795 /* initialize local statfs cached values */
1796 rc = lod_lsfs_init(env, lod);
1798 GOTO(out_disconnect, rc);
1800 /* default threshold as half of total space, in MiB */
1801 lod->lod_dom_threshold_free_mb = lod->lod_lsfs_total_mb / 2;
1802 /* set default DoM stripe size based on free space amount */
1803 lod_dom_stripesize_recalc(lod);
1805 /* setup obd to be used with old lov code */
1806 rc = lod_pools_init(lod, cfg);
1808 GOTO(out_disconnect, rc);
1810 rc = lod_procfs_init(lod);
1812 GOTO(out_pools, rc);
1814 spin_lock_init(&lod->lod_lock);
1815 spin_lock_init(&lod->lod_connects_lock);
1816 lu_tgt_descs_init(&lod->lod_mdt_descs, true);
1817 lu_tgt_descs_init(&lod->lod_ost_descs, false);
1818 lu_qos_rr_init(&lod->lod_mdt_descs.ltd_qos.lq_rr);
1819 lu_qos_rr_init(&lod->lod_ost_descs.ltd_qos.lq_rr);
1824 lod_pools_fini(lod);
1826 obd_disconnect(lod->lod_child_exp);
1831 * Implementation of lu_device_type_operations::ldto_device_free() for LOD
1833 * Releases the memory allocated for LOD device.
1835 * see include/lu_object.h for the details.
1837 static struct lu_device *lod_device_free(const struct lu_env *env,
1838 struct lu_device *lu)
1840 struct lod_device *lod = lu2lod_dev(lu);
1841 struct lu_device *next = &lod->lod_child->dd_lu_dev;
1845 if (atomic_read(&lu->ld_site->ls_obj_hash.nelems)) {
1846 lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR,
1849 LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
1850 dt_device_fini(&lod->lod_dt_dev);
1856 * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
1858 * Allocates LOD device and calls the helpers to initialize it.
1860 * see include/lu_object.h for the details.
1862 static struct lu_device *lod_device_alloc(const struct lu_env *env,
1863 struct lu_device_type *type,
1864 struct lustre_cfg *lcfg)
1866 struct lod_device *lod;
1867 struct lu_device *lu_dev;
1871 lu_dev = ERR_PTR(-ENOMEM);
1875 lu_dev = lod2lu_dev(lod);
1876 dt_device_init(&lod->lod_dt_dev, type);
1877 rc = lod_init0(env, lod, type, lcfg);
1879 lod_device_free(env, lu_dev);
1880 lu_dev = ERR_PTR(rc);
1887 static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
1889 if (lag->lag_oss_avoid_array)
1890 OBD_FREE_PTR_ARRAY(lag->lag_oss_avoid_array,
1892 bitmap_free(lag->lag_ost_avoid_bitmap);
1896 * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
1898 * Releases the internal resources used by LOD device.
1900 * see include/lu_object.h for the details.
1902 static struct lu_device *lod_device_fini(const struct lu_env *env,
1903 struct lu_device *d)
1905 struct lod_device *lod = lu2lod_dev(d);
1910 lod_pools_fini(lod);
1912 lod_procfs_fini(lod);
1914 rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
1916 CERROR("%s: can not fini ost descriptors: rc = %d\n",
1917 lod2obd(lod)->obd_name, rc);
1919 rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
1921 CERROR("%s: can not fini mdt descriptors: rc = %d\n",
1922 lod2obd(lod)->obd_name, rc);
1928 * Implementation of obd_ops::o_connect() for LOD
1930 * Used to track all the users of this specific LOD device,
1931 * so the device stays up until the last user disconnected.
1933 * \param[in] env LU environment provided by the caller
1934 * \param[out] exp export the caller will be using to access LOD
1935 * \param[in] obd OBD device representing LOD device
1936 * \param[in] cluuid unique identifier of the caller
1937 * \param[in] data not used
1938 * \param[in] localdata not used
1940 * \retval 0 on success
1941 * \retval negative negated errno on error
1943 static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
1944 struct obd_device *obd, struct obd_uuid *cluuid,
1945 struct obd_connect_data *data, void *localdata)
1947 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1948 struct lustre_handle conn;
1953 CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
1955 rc = class_connect(&conn, obd, cluuid);
1959 *exp = class_conn2export(&conn);
1961 spin_lock(&lod->lod_connects_lock);
1962 lod->lod_connects++;
1963 /* at the moment we expect the only user */
1964 LASSERT(lod->lod_connects == 1);
1965 spin_unlock(&lod->lod_connects_lock);
1972 * Implementation of obd_ops::o_disconnect() for LOD
1974 * When the caller doesn't need to use this LOD instance, it calls
1975 * obd_disconnect() and LOD releases corresponding export/reference count.
1976 * Once all the users gone, LOD device is released.
1978 * \param[in] exp export provided to the caller in obd_connect()
1980 * \retval 0 on success
1981 * \retval negative negated errno on error
1983 static int lod_obd_disconnect(struct obd_export *exp)
1985 struct obd_device *obd = exp->exp_obd;
1986 struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
1987 int rc, release = 0;
1991 /* Only disconnect the underlying layers on the final disconnect. */
1992 spin_lock(&lod->lod_connects_lock);
1993 lod->lod_connects--;
1994 if (lod->lod_connects != 0) {
1995 /* why should there be more than 1 connect? */
1996 spin_unlock(&lod->lod_connects_lock);
1997 CERROR("%s: disconnect #%d\n", exp->exp_obd->obd_name,
2001 spin_unlock(&lod->lod_connects_lock);
2003 /* the last user of lod has gone, let's release the device */
2007 rc = class_disconnect(exp); /* bz 9811 */
2009 if (rc == 0 && release)
2010 class_manual_cleanup(obd);
2014 LU_KEY_INIT(lod, struct lod_thread_info);
2016 static void lod_key_fini(const struct lu_context *ctx,
2017 struct lu_context_key *key, void *data)
2019 struct lod_thread_info *info = data;
2020 struct lod_layout_component *lds =
2021 info->lti_def_striping.lds_def_comp_entries;
2024 * allocated in lod_get_lov_ea
2025 * XXX: this is overload, a tread may have such store but used only
2026 * once. Probably better would be pool of such stores per LOD.
2028 if (info->lti_ea_store) {
2029 OBD_FREE_LARGE(info->lti_ea_store, info->lti_ea_store_size);
2030 info->lti_ea_store = NULL;
2031 info->lti_ea_store_size = 0;
2033 lu_buf_free(&info->lti_linkea_buf);
2036 lod_free_def_comp_entries(&info->lti_def_striping);
2038 if (info->lti_comp_size > 0)
2039 OBD_FREE_PTR_ARRAY(info->lti_comp_idx,
2040 info->lti_comp_size);
2042 lod_avoid_guide_fini(&info->lti_avoid);
2047 /* context key: lod_thread_key */
2048 LU_CONTEXT_KEY_DEFINE(lod, LCT_MD_THREAD);
2050 LU_TYPE_INIT_FINI(lod, &lod_thread_key);
2052 static const struct lu_device_type_operations lod_device_type_ops = {
2053 .ldto_init = lod_type_init,
2054 .ldto_fini = lod_type_fini,
2056 .ldto_start = lod_type_start,
2057 .ldto_stop = lod_type_stop,
2059 .ldto_device_alloc = lod_device_alloc,
2060 .ldto_device_free = lod_device_free,
2062 .ldto_device_fini = lod_device_fini
2065 static struct lu_device_type lod_device_type = {
2066 .ldt_tags = LU_DEVICE_DT,
2067 .ldt_name = LUSTRE_LOD_NAME,
2068 .ldt_ops = &lod_device_type_ops,
2069 .ldt_ctx_tags = LCT_MD_THREAD,
2073 * Implementation of obd_ops::o_get_info() for LOD
2075 * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
2076 * the caller binary status whether LOD has seen connection to any OST target.
2077 * It will also check if the MDT update log context being initialized (if
2080 * \param[in] env LU environment provided by the caller
2081 * \param[in] exp export of the caller
2082 * \param[in] keylen len of the key
2083 * \param[in] key the key
2084 * \param[in] vallen not used
2085 * \param[in] val not used
2087 * \retval 0 if a connection was seen
2088 * \retval -EAGAIN if LOD isn't running yet or no
2089 * connection has been seen yet
2090 * \retval -EINVAL if not supported key is requested
2092 static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
2093 u32 keylen, void *key, u32 *vallen, void *val)
2097 if (KEY_IS(KEY_OSP_CONNECTED)) {
2098 struct obd_device *obd = exp->exp_obd;
2099 struct lod_device *d;
2100 struct lod_tgt_desc *tgt;
2103 if (!obd->obd_set_up || obd->obd_stopping)
2106 d = lu2lod_dev(obd->obd_lu_dev);
2107 lod_getref(&d->lod_ost_descs);
2108 lod_foreach_ost(d, tgt) {
2109 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2111 /* one healthy device is enough */
2115 lod_putref(d, &d->lod_ost_descs);
2117 lod_getref(&d->lod_mdt_descs);
2118 lod_foreach_mdt(d, tgt) {
2119 struct llog_ctxt *ctxt;
2121 if (!tgt->ltd_active)
2124 ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
2125 LLOG_UPDATELOG_ORIG_CTXT);
2127 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2129 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2133 if (!ctxt->loc_handle) {
2134 CDEBUG(D_INFO, "%s: %s is not ready.\n",
2136 tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
2138 llog_ctxt_put(ctxt);
2141 llog_ctxt_put(ctxt);
2143 lod_putref(d, &d->lod_mdt_descs);
2151 static int lod_obd_set_info_async(const struct lu_env *env,
2152 struct obd_export *exp,
2153 u32 keylen, void *key,
2154 u32 vallen, void *val,
2155 struct ptlrpc_request_set *set)
2157 struct obd_device *obd = class_exp2obd(exp);
2158 struct lod_device *d;
2159 struct lod_tgt_desc *tgt;
2167 set = ptlrpc_prep_set();
2172 d = lu2lod_dev(obd->obd_lu_dev);
2173 lod_getref(&d->lod_ost_descs);
2174 lod_foreach_ost(d, tgt) {
2175 if (!tgt->ltd_active)
2178 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2180 if (rc2 != 0 && rc == 0)
2183 lod_putref(d, &d->lod_ost_descs);
2185 lod_getref(&d->lod_mdt_descs);
2186 lod_foreach_mdt(d, tgt) {
2187 if (!tgt->ltd_active)
2189 rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
2191 if (rc2 != 0 && rc == 0)
2194 lod_putref(d, &d->lod_mdt_descs);
2198 rc2 = ptlrpc_set_wait(env, set);
2199 if (rc2 == 0 && rc == 0)
2201 ptlrpc_set_destroy(set);
2207 #define QMT0_DEV_NAME_LEN (LUSTRE_MAXFSNAME + sizeof("-QMT0000"))
2208 static struct obd_device *obd_find_qmt0(char *obd_name)
2210 char qmt_name[QMT0_DEV_NAME_LEN];
2211 struct obd_device *qmt = NULL;
2213 if (!server_name2fsname(obd_name, qmt_name, NULL)) {
2214 strlcat(qmt_name, "-QMT0000", QMT0_DEV_NAME_LEN);
2215 qmt = class_name2obd(qmt_name);
2221 static int lod_pool_new_q(struct obd_device *obd, char *poolname)
2223 int err = lod_pool_new(obd, poolname);
2226 obd = obd_find_qmt0(obd->obd_name);
2228 obd_pool_new(obd, poolname);
2234 static int lod_pool_remove_q(struct obd_device *obd, char *poolname,
2237 int err = lod_pool_remove(obd, poolname, ostname);
2240 obd = obd_find_qmt0(obd->obd_name);
2242 obd_pool_rem(obd, poolname, ostname);
2248 static int lod_pool_add_q(struct obd_device *obd, char *poolname, char *ostname)
2250 int err = lod_pool_add(obd, poolname, ostname);
2253 obd = obd_find_qmt0(obd->obd_name);
2255 obd_pool_add(obd, poolname, ostname);
2261 static int lod_pool_del_q(struct obd_device *obd, char *poolname)
2263 int err = lod_pool_del(obd, poolname);
2266 obd = obd_find_qmt0(obd->obd_name);
2268 obd_pool_del(obd, poolname);
2274 static const struct obd_ops lod_obd_device_ops = {
2275 .o_owner = THIS_MODULE,
2276 .o_connect = lod_obd_connect,
2277 .o_disconnect = lod_obd_disconnect,
2278 .o_get_info = lod_obd_get_info,
2279 .o_set_info_async = lod_obd_set_info_async,
2280 .o_pool_new = lod_pool_new_q,
2281 .o_pool_rem = lod_pool_remove_q,
2282 .o_pool_add = lod_pool_add_q,
2283 .o_pool_del = lod_pool_del_q,
2286 static int __init lod_init(void)
2288 struct obd_type *sym;
2291 rc = lu_kmem_init(lod_caches);
2295 rc = class_register_type(&lod_obd_device_ops, NULL, true,
2296 LUSTRE_LOD_NAME, &lod_device_type);
2298 lu_kmem_fini(lod_caches);
2302 /* create "lov" entry for compatibility purposes */
2303 sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
2306 /* does real "lov" already exist ? */
2314 static void __exit lod_exit(void)
2316 struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
2318 /* if this was never fully initialized by the lov layer
2319 * then we are responsible for freeing this obd_type
2322 /* final put if we manage this obd type */
2323 if (sym->typ_sym_filter)
2324 kobject_put(&sym->typ_kobj);
2325 /* put reference taken by class_search_type */
2326 kobject_put(&sym->typ_kobj);
2329 class_unregister_type(LUSTRE_LOD_NAME);
2330 lu_kmem_fini(lod_caches);
2333 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2334 MODULE_DESCRIPTION("Lustre Logical Object Device ("LUSTRE_LOD_NAME")");
2335 MODULE_VERSION(LUSTRE_VERSION_STRING);
2336 MODULE_LICENSE("GPL");
2338 module_init(lod_init);
2339 module_exit(lod_exit);