X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosp%2Fosp_dev.c;h=04ad620ba8a38aa97ce65d742b82790ca9e1853b;hp=db105c49ba8fb7288c49976c9568f65b7b708059;hb=68825cdf81688faa8d090a3a5f2321b66c2cec63;hpb=26b8238659974959780cd49de92595b4b0bdf89f diff --git a/lustre/osp/osp_dev.c b/lustre/osp/osp_dev.c index db105c4..04ad620 100644 --- a/lustre/osp/osp_dev.c +++ b/lustre/osp/osp_dev.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2014, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -74,10 +70,13 @@ #define DEBUG_SUBSYSTEM S_MDS -#include -#include -#include +#include + +#include #include +#include +#include +#include #include "osp_internal.h" @@ -184,7 +183,7 @@ static struct dt_object if (rc) { CERROR("%s: can't be initialized: rc = %d\n", osp->opd_obd->obd_name, rc); - lu_object_put(env, &dto->do_lu); + dt_object_put(env, dto); RETURN(ERR_PTR(rc)); } RETURN(dto); @@ -212,6 +211,9 @@ static int osp_write_local_file(const struct lu_env *env, struct thandle *th; int rc; + if (osp->opd_storage->dd_rdonly) + RETURN(0); + th = dt_trans_create(env, osp->opd_storage); if (IS_ERR(th)) RETURN(PTR_ERR(th)); @@ -256,20 +258,23 @@ static int osp_init_last_objid(const struct lu_env *env, struct osp_device *osp) if (IS_ERR(dto)) RETURN(PTR_ERR(dto)); + osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &osp->opd_last_id, + osp->opd_index); + /* object will be released in device cleanup path */ - if (osi->osi_attr.la_size >= - sizeof(osi->osi_id) * (osp->opd_index + 1)) { - osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid, - osp->opd_index); + if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) { rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off); if (rc != 0 && rc != -EFAULT) GOTO(out, rc); + /* In case of idif bits 32-48 go to f_seq + * (see osp_init_last_seq). So don't care + * about u64->u32 convertion. */ + fid->f_oid = osp->opd_last_id; } if (rc == -EFAULT) { /* fresh LAST_ID */ + osp->opd_last_id = 0; fid->f_oid = 0; - osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_oid, - osp->opd_index); rc = osp_write_local_file(env, osp, dto, &osi->osi_lb, osi->osi_off); if (rc != 0) @@ -281,7 +286,7 @@ out: /* object will be released in device cleanup path */ CERROR("%s: can't initialize lov_objid: rc = %d\n", osp->opd_obd->obd_name, rc); - lu_object_put(env, &dto->do_lu); + dt_object_put(env, dto); osp->opd_last_used_oid_file = NULL; RETURN(rc); } @@ -313,20 +318,21 @@ static int osp_init_last_seq(const struct lu_env *env, struct osp_device *osp) if (IS_ERR(dto)) RETURN(PTR_ERR(dto)); + osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq, + osp->opd_index); + /* object will be released in device cleanup path */ - if (osi->osi_attr.la_size >= - sizeof(osi->osi_id) * (osp->opd_index + 1)) { - osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq, - osp->opd_index); + if (osi->osi_attr.la_size >= (osi->osi_off + osi->osi_lb.lb_len)) { rc = dt_record_read(env, dto, &osi->osi_lb, &osi->osi_off); if (rc != 0 && rc != -EFAULT) GOTO(out, rc); + if (fid_is_idif(fid)) + fid->f_seq = fid_idif_seq(osp->opd_last_id, + osp->opd_index); } if (rc == -EFAULT) { /* fresh OSP */ fid->f_seq = 0; - osp_objseq_buf_prep(&osi->osi_lb, &osi->osi_off, &fid->f_seq, - osp->opd_index); rc = osp_write_local_file(env, osp, dto, &osi->osi_lb, osi->osi_off); if (rc != 0) @@ -338,7 +344,7 @@ out: /* object will be released in device cleanup path */ CERROR("%s: can't initialize lov_seq: rc = %d\n", osp->opd_obd->obd_name, rc); - lu_object_put(env, &dto->do_lu); + dt_object_put(env, dto); osp->opd_last_used_seq_file = NULL; RETURN(rc); } @@ -372,7 +378,7 @@ static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp) rc = osp_init_last_seq(env, osp); if (rc < 0) { - CERROR("%s: Can not get ids %d from old objid!\n", + CERROR("%s: Can not get sequence %d from old objseq!\n", osp->opd_obd->obd_name, rc); GOTO(out, rc); } @@ -403,16 +409,17 @@ static int osp_last_used_init(const struct lu_env *env, struct osp_device *osp) GOTO(out, rc = -EINVAL); } + osp_fid_to_obdid(&osp->opd_last_used_fid, &osp->opd_last_id); CDEBUG(D_INFO, "%s: Init last used fid "DFID"\n", osp->opd_obd->obd_name, PFID(&osp->opd_last_used_fid)); out: if (rc != 0) { if (osp->opd_last_used_oid_file != NULL) { - lu_object_put(env, &osp->opd_last_used_oid_file->do_lu); + dt_object_put(env, osp->opd_last_used_oid_file); osp->opd_last_used_oid_file = NULL; } if (osp->opd_last_used_seq_file != NULL) { - lu_object_put(env, &osp->opd_last_used_seq_file->do_lu); + dt_object_put(env, osp->opd_last_used_seq_file); osp->opd_last_used_seq_file = NULL; } } @@ -430,12 +437,12 @@ static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp) { /* release last_used file */ if (osp->opd_last_used_oid_file != NULL) { - lu_object_put(env, &osp->opd_last_used_oid_file->do_lu); + dt_object_put(env, osp->opd_last_used_oid_file); osp->opd_last_used_oid_file = NULL; } if (osp->opd_last_used_seq_file != NULL) { - lu_object_put(env, &osp->opd_last_used_seq_file->do_lu); + dt_object_put(env, osp->opd_last_used_seq_file); osp->opd_last_used_seq_file = NULL; } } @@ -452,10 +459,11 @@ static void osp_last_used_fini(const struct lu_env *env, struct osp_device *osp) */ static int osp_disconnect(struct osp_device *d) { + struct obd_device *obd = d->opd_obd; struct obd_import *imp; int rc = 0; - imp = d->opd_obd->u.cli.cl_import; + imp = obd->u.cli.cl_import; /* Mark import deactivated now, so we don't try to reconnect if any * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't @@ -474,8 +482,7 @@ static int osp_disconnect(struct osp_device *d) rc = ptlrpc_disconnect_import(imp, 0); if (rc != 0) - CERROR("%s: can't disconnect: rc = %d\n", - d->opd_obd->obd_name, rc); + CERROR("%s: can't disconnect: rc = %d\n", obd->obd_name, rc); ptlrpc_invalidate_import(imp); @@ -483,6 +490,104 @@ static int osp_disconnect(struct osp_device *d) } /** + * Initialize the osp_update structure in OSP device + * + * Allocate osp update structure and start update thread. + * + * \param[in] osp OSP device + * + * \retval 0 if initialization succeeds. + * \retval negative errno if initialization fails. + */ +static int osp_update_init(struct osp_device *osp) +{ + struct task_struct *task; + int rc; + + ENTRY; + + LASSERT(osp->opd_connect_mdt); + + if (osp->opd_storage->dd_rdonly) + RETURN(0); + + OBD_ALLOC_PTR(osp->opd_update); + if (osp->opd_update == NULL) + RETURN(-ENOMEM); + + init_waitqueue_head(&osp->opd_update->ou_waitq); + spin_lock_init(&osp->opd_update->ou_lock); + INIT_LIST_HEAD(&osp->opd_update->ou_list); + osp->opd_update->ou_rpc_version = 1; + osp->opd_update->ou_version = 1; + osp->opd_update->ou_generation = 0; + + rc = lu_env_init(&osp->opd_update->ou_env, + osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags); + if (rc < 0) { + CERROR("%s: init env error: rc = %d\n", osp->opd_obd->obd_name, + rc); + OBD_FREE_PTR(osp->opd_update); + osp->opd_update = NULL; + RETURN(rc); + } + /* start thread handling sending updates to the remote MDT */ + task = kthread_create(osp_send_update_thread, osp, + "osp_up%u-%u", osp->opd_index, osp->opd_group); + if (IS_ERR(task)) { + int rc = PTR_ERR(task); + + lu_env_fini(&osp->opd_update->ou_env); + OBD_FREE_PTR(osp->opd_update); + osp->opd_update = NULL; + CERROR("%s: can't start precreate thread: rc = %d\n", + osp->opd_obd->obd_name, rc); + RETURN(rc); + } + + osp->opd_update->ou_update_task = task; + wake_up_process(task); + + RETURN(0); +} + +/** + * Finialize osp_update structure in OSP device + * + * Stop the OSP update sending thread, then delete the left + * osp thandle in the sending list. + * + * \param [in] osp OSP device. + */ +static void osp_update_fini(const struct lu_env *env, struct osp_device *osp) +{ + struct osp_update_request *our; + struct osp_update_request *tmp; + struct osp_updates *ou = osp->opd_update; + + if (ou == NULL) + return; + + kthread_stop(ou->ou_update_task); + lu_env_fini(&ou->ou_env); + + /* Remove the left osp thandle from the list */ + spin_lock(&ou->ou_lock); + list_for_each_entry_safe(our, tmp, &ou->ou_list, + our_list) { + list_del_init(&our->our_list); + LASSERT(our->our_th != NULL); + osp_trans_callback(env, our->our_th, -EIO); + /* our will be destroyed in osp_thandle_put() */ + osp_thandle_put(env, our->our_th); + } + spin_unlock(&ou->ou_lock); + + OBD_FREE_PTR(ou); + osp->opd_update = NULL; +} + +/** * Cleanup OSP, which includes disconnect import, cleanup unlink log, stop * precreate threads etc. * @@ -501,9 +606,12 @@ static int osp_shutdown(const struct lu_env *env, struct osp_device *d) rc = osp_disconnect(d); - osp_sync_fini(d); + osp_statfs_fini(d); if (!d->opd_connect_mdt) { + /* stop sync thread */ + osp_sync_fini(d); + /* stop precreate thread */ osp_precreate_fini(d); @@ -533,36 +641,40 @@ static int osp_shutdown(const struct lu_env *env, struct osp_device *d) static int osp_process_config(const struct lu_env *env, struct lu_device *dev, struct lustre_cfg *lcfg) { - struct osp_device *d = lu2osp_dev(dev); - struct obd_device *obd = d->opd_obd; - int rc; + struct osp_device *d = lu2osp_dev(dev); + struct dt_device *dt = lu2dt_dev(dev); + struct obd_device *obd = d->opd_obd; + ssize_t count; + int rc; ENTRY; switch (lcfg->lcfg_command) { case LCFG_PRE_CLEANUP: rc = osp_disconnect(d); + osp_update_fini(env, d); + if (obd->obd_namespace != NULL) + ldlm_namespace_free_prior(obd->obd_namespace, NULL, 1); break; case LCFG_CLEANUP: lu_dev_del_linkage(dev->ld_site, dev); rc = osp_shutdown(env, d); break; case LCFG_PARAM: - LASSERT(obd); - rc = class_process_proc_param(PARAM_OSC, obd->obd_vars, - lcfg, obd); - if (rc > 0) - rc = 0; - if (rc == -ENOSYS) { - /* class_process_proc_param() haven't found matching - * parameter and returned ENOSYS so that layer(s) + count = class_modify_config(lcfg, d->opd_connect_mdt ? + PARAM_OSP : PARAM_OSC, + &dt->dd_kobj); + if (count < 0) { + /* class_modify_config() haven't found matching + * parameter and returned an error so that layer(s) * below could use that. But OSP is the bottom, so - * just ignore it */ + * just ignore it + */ CERROR("%s: unknown param %s\n", (char *)lustre_cfg_string(lcfg, 0), (char *)lustre_cfg_string(lcfg, 1)); - rc = 0; } + rc = 0; break; default: CERROR("%s: unknown command %u\n", @@ -599,10 +711,41 @@ static int osp_recovery_complete(const struct lu_env *env, RETURN(0); } +/** + * Implementation of lu_device_operations::ldo_fid_alloc() for OSP + * + * Allocate FID from remote MDT. + * + * see include/lu_object.h for the details. + */ +static int osp_fid_alloc(const struct lu_env *env, struct lu_device *d, + struct lu_fid *fid, struct lu_object *parent, + const struct lu_name *name) +{ + struct osp_device *osp = lu2osp_dev(d); + struct client_obd *cli = &osp->opd_obd->u.cli; + struct lu_client_seq *seq = cli->cl_seq; + int rc; + + ENTRY; + + /* Sigh, fid client is not ready yet */ + if (!osp->opd_obd->u.cli.cl_seq) + RETURN(-ENOTCONN); + + if (!osp->opd_obd->u.cli.cl_seq->lcs_exp) + RETURN(-ENOTCONN); + + rc = seq_client_alloc_fid(env, seq, fid); + + RETURN(rc); +} + const struct lu_device_operations osp_lu_ops = { .ldo_object_alloc = osp_object_alloc, .ldo_process_config = osp_process_config, .ldo_recovery_complete = osp_recovery_complete, + .ldo_fid_alloc = osp_fid_alloc, }; /** @@ -623,50 +766,56 @@ const struct lu_device_operations osp_lu_ops = { * \retval negative negative errno if get statfs failed. */ static int osp_statfs(const struct lu_env *env, struct dt_device *dev, - struct obd_statfs *sfs) + struct obd_statfs *sfs, struct obd_statfs_info *info) { struct osp_device *d = dt2osp_dev(dev); + struct obd_import *imp = d->opd_obd->u.cli.cl_import; ENTRY; + if (imp->imp_state == LUSTRE_IMP_CLOSED) + RETURN(-ESHUTDOWN); + if (unlikely(d->opd_imp_active == 0)) RETURN(-ENOTCONN); + /* return recently updated data */ + *sfs = d->opd_statfs; + if (info) { + info->os_reserved_mb_low = d->opd_reserved_mb_low; + info->os_reserved_mb_high = d->opd_reserved_mb_high; + } + if (d->opd_pre == NULL) RETURN(0); - /* return recently updated data */ - *sfs = d->opd_statfs; + CDEBUG(D_OTHER, "%s: %llu blocks, %llu free, %llu avail, " + "%u bsize, %u reserved mb low, %u reserved mb high, " + "%llu files, %llu free files\n", d->opd_obd->obd_name, + sfs->os_blocks, sfs->os_bfree, sfs->os_bavail, sfs->os_bsize, + d->opd_reserved_mb_low, d->opd_reserved_mb_high, + sfs->os_files, sfs->os_ffree); + + + if (info && !info->os_enable_pre) + RETURN(0); /* - * layer above osp (usually lod) can use ffree to estimate - * how many objects are available for immediate creation + * The layer above osp (usually lod) can use f_precreated to + * estimate how many objects are available for immediate usage. */ spin_lock(&d->opd_pre_lock); - LASSERTF(fid_seq(&d->opd_pre_last_created_fid) == - fid_seq(&d->opd_pre_used_fid), - "last_created "DFID", next_fid "DFID"\n", - PFID(&d->opd_pre_last_created_fid), - PFID(&d->opd_pre_used_fid)); - sfs->os_fprecreated = fid_oid(&d->opd_pre_last_created_fid) - - fid_oid(&d->opd_pre_used_fid); + sfs->os_fprecreated = osp_fid_diff(&d->opd_pre_last_created_fid, + &d->opd_pre_used_fid); sfs->os_fprecreated -= d->opd_pre_reserved; + LASSERTF(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2, + "last_created "DFID", next_fid "DFID", reserved %llu\n", + PFID(&d->opd_pre_last_created_fid), PFID(&d->opd_pre_used_fid), + d->opd_pre_reserved); spin_unlock(&d->opd_pre_lock); - - LASSERT(sfs->os_fprecreated <= OST_MAX_PRECREATE * 2); - - CDEBUG(D_OTHER, "%s: "LPU64" blocks, "LPU64" free, "LPU64" avail, " - LPU64" files, "LPU64" free files\n", d->opd_obd->obd_name, - sfs->os_blocks, sfs->os_bfree, sfs->os_bavail, - sfs->os_files, sfs->os_ffree); RETURN(0); } -static int osp_sync_timeout(void *data) -{ - return 1; -} - /** * Implementation of dt_device_operations::dt_sync * @@ -682,52 +831,59 @@ static int osp_sync_timeout(void *data) static int osp_sync(const struct lu_env *env, struct dt_device *dev) { struct osp_device *d = dt2osp_dev(dev); - cfs_time_t expire; - struct l_wait_info lwi = { 0 }; - unsigned long id, old; - int rc = 0; - unsigned long start = cfs_time_current(); + time64_t start = ktime_get_seconds(); + int recs, rc = 0; + u64 old; + ENTRY; + /* No Sync between MDTs yet. */ + if (d->opd_connect_mdt) + RETURN(0); + + recs = atomic_read(&d->opd_sync_changes); + old = atomic64_read(&d->opd_sync_processed_recs); + + osp_sync_force(env, dt2osp_dev(dev)); + if (unlikely(d->opd_imp_active == 0)) RETURN(-ENOTCONN); - id = d->opd_syn_last_used_id; down_write(&d->opd_async_updates_rwsem); CDEBUG(D_OTHER, "%s: async updates %d\n", d->opd_obd->obd_name, atomic_read(&d->opd_async_updates_count)); /* make sure the connection is fine */ - expire = cfs_time_shift(obd_timeout); - lwi = LWI_TIMEOUT(expire - cfs_time_current(), osp_sync_timeout, d); - rc = l_wait_event(d->opd_syn_barrier_waitq, - atomic_read(&d->opd_async_updates_count) == 0, - &lwi); + rc = wait_event_idle_timeout( + d->opd_sync_barrier_waitq, + atomic_read(&d->opd_async_updates_count) == 0, + cfs_time_seconds(obd_timeout)); + if (rc > 0) + rc = 0; + else if (rc == 0) + rc = -ETIMEDOUT; + up_write(&d->opd_async_updates_rwsem); if (rc != 0) GOTO(out, rc); - CDEBUG(D_CACHE, "%s: id: used %lu, processed %lu\n", - d->opd_obd->obd_name, id, d->opd_syn_last_processed_id); - - /* wait till all-in-line are processed */ - while (d->opd_syn_last_processed_id < id) { - - old = d->opd_syn_last_processed_id; + CDEBUG(D_CACHE, "%s: processed %llu\n", d->opd_obd->obd_name, + (unsigned long long)atomic64_read(&d->opd_sync_processed_recs)); + while (atomic64_read(&d->opd_sync_processed_recs) < old + recs) { + __u64 last = atomic64_read(&d->opd_sync_processed_recs); /* make sure the connection is fine */ - expire = cfs_time_shift(obd_timeout); - lwi = LWI_TIMEOUT(expire - cfs_time_current(), - osp_sync_timeout, d); - l_wait_event(d->opd_syn_barrier_waitq, - d->opd_syn_last_processed_id >= id, - &lwi); - - if (d->opd_syn_last_processed_id >= id) + wait_event_idle_timeout( + d->opd_sync_barrier_waitq, + atomic64_read(&d->opd_sync_processed_recs) + >= old + recs, + cfs_time_seconds(obd_timeout)); + + if (atomic64_read(&d->opd_sync_processed_recs) >= old + recs) break; - if (d->opd_syn_last_processed_id != old) { + if (atomic64_read(&d->opd_sync_processed_recs) != last) { /* some progress have been made, * keep trying... */ continue; @@ -738,29 +894,27 @@ static int osp_sync(const struct lu_env *env, struct dt_device *dev) } /* block new processing (barrier>0 - few callers are possible */ - atomic_inc(&d->opd_syn_barrier); + atomic_inc(&d->opd_sync_barrier); CDEBUG(D_CACHE, "%s: %u in flight\n", d->opd_obd->obd_name, - d->opd_syn_rpc_in_flight); + atomic_read(&d->opd_sync_rpcs_in_flight)); /* wait till all-in-flight are replied, so executed by the target */ /* XXX: this is used by LFSCK at the moment, which doesn't require * all the changes to be committed, but in general it'd be * better to wait till commit */ - while (d->opd_syn_rpc_in_flight > 0) { + while (atomic_read(&d->opd_sync_rpcs_in_flight) > 0) { + old = atomic_read(&d->opd_sync_rpcs_in_flight); - old = d->opd_syn_rpc_in_flight; + wait_event_idle_timeout( + d->opd_sync_barrier_waitq, + atomic_read(&d->opd_sync_rpcs_in_flight) == 0, + cfs_time_seconds(obd_timeout)); - expire = cfs_time_shift(obd_timeout); - lwi = LWI_TIMEOUT(expire - cfs_time_current(), - osp_sync_timeout, d); - l_wait_event(d->opd_syn_barrier_waitq, - d->opd_syn_rpc_in_flight == 0, &lwi); - - if (d->opd_syn_rpc_in_flight == 0) + if (atomic_read(&d->opd_sync_rpcs_in_flight) == 0) break; - if (d->opd_syn_rpc_in_flight != old) { + if (atomic_read(&d->opd_sync_rpcs_in_flight) != old) { /* some progress have been made */ continue; } @@ -771,11 +925,11 @@ static int osp_sync(const struct lu_env *env, struct dt_device *dev) out: /* resume normal processing (barrier=0) */ - atomic_dec(&d->opd_syn_barrier); - __osp_sync_check_for_work(d); + atomic_dec(&d->opd_sync_barrier); + osp_sync_check_for_work(d); - CDEBUG(D_CACHE, "%s: done in %lu: rc = %d\n", d->opd_obd->obd_name, - cfs_time_current() - start, rc); + CDEBUG(D_CACHE, "%s: done in %lld: rc = %d\n", d->opd_obd->obd_name, + ktime_get_seconds() - start, rc); RETURN(rc); } @@ -786,6 +940,7 @@ const struct dt_device_operations osp_dt_ops = { .dt_trans_create = osp_trans_create, .dt_trans_start = osp_trans_start, .dt_trans_stop = osp_trans_stop, + .dt_trans_cb_add = osp_trans_cb_add, }; /** @@ -844,6 +999,27 @@ out: } /** + * Determine if the lock needs to be cancelled + * + * Determine if the unused lock should be cancelled before replay, see + * (ldlm_cancel_no_wait_policy()). Currently, only inode bits lock exists + * between MDTs. + * + * \param[in] lock lock to be checked. + * + * \retval 1 if the lock needs to be cancelled before replay. + * \retval 0 if the lock does not need to be cancelled before + * replay. + */ +static int osp_cancel_weight(struct ldlm_lock *lock) +{ + if (lock->l_resource->lr_type != LDLM_IBITS) + RETURN(0); + + RETURN(1); +} + +/** * Initialize OSP device according to the parameters in the configuration * log \a cfg. * @@ -873,10 +1049,10 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, { struct obd_device *obd; struct obd_import *imp; - class_uuid_t uuid; - char *src, *tgt, *mdt, *osdname = NULL; + char *src, *tgt, *osdname = NULL; + const char *mdt; int rc; - long idx; + u32 idx; ENTRY; @@ -923,8 +1099,8 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, RETURN(-EINVAL); } - idx = simple_strtol(tgt + 4, &mdt, 16); - if (mdt[0] != '-' || idx > INT_MAX || idx < 0) { + rc = target_name2index(tgt + 1, &idx, &mdt); + if (rc < 0 || rc & LDD_F_SV_ALL || mdt[0] != '-') { CERROR("%s: invalid OST index in '%s': rc = %d\n", osp->opd_obd->obd_name, src, -EINVAL); RETURN(-EINVAL); @@ -942,8 +1118,8 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, RETURN(-EINVAL); } - idx = simple_strtol(tgt + 4, &mdt, 16); - if (*mdt != '\0' || idx > INT_MAX || idx < 0) { + rc = target_name2index(tgt + 1, &idx, &mdt); + if (rc < 0 || rc & LDD_F_SV_ALL || *mdt != '\0') { CERROR("%s: invalid OST index in '%s': rc = %d\n", osp->opd_obd->obd_name, src, -EINVAL); RETURN(-EINVAL); @@ -962,8 +1138,8 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, if (strncmp(tgt - 12, "-MDT", 4) == 0) osp->opd_connect_mdt = 1; - idx = simple_strtol(tgt - 8, &mdt, 16); - if (mdt[0] != '-' || idx > INT_MAX || idx < 0) { + rc = target_name2index(tgt - 11, &idx, &mdt); + if (rc < 0 || rc & LDD_F_SV_ALL || mdt[0] != '-') { CERROR("%s: invalid OST index in '%s': rc =%d\n", osp->opd_obd->obd_name, src, -EINVAL); RETURN(-EINVAL); @@ -994,14 +1170,7 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, strcat(osdname, "-osd"); CDEBUG(D_HA, "%s: connect to %s (%s)\n", obd->obd_name, osdname, src); - if (osp->opd_connect_mdt) { - struct client_obd *cli = &osp->opd_obd->u.cli; - - OBD_ALLOC(cli->cl_rpc_lock, sizeof(*cli->cl_rpc_lock)); - if (!cli->cl_rpc_lock) - GOTO(out_fini, rc = -ENOMEM); - osp_init_rpc_lock(cli->cl_rpc_lock); - } + osp_init_rpc_lock(osp); osp->opd_dt_dev.dd_lu_dev.ld_ops = &osp_lu_ops; osp->opd_dt_dev.dd_ops = &osp_dt_ops; @@ -1023,7 +1192,7 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, GOTO(out_ref, rc); } - osp_lprocfs_init(osp); + osp_tunables_init(osp); rc = obd_fid_init(osp->opd_obd, NULL, osp->opd_connect_mdt ? LUSTRE_SEQ_METADATA : LUSTRE_SEQ_DATA); @@ -1036,33 +1205,42 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, if (!osp->opd_connect_mdt) { /* Initialize last id from the storage - will be * used in orphan cleanup. */ - rc = osp_last_used_init(env, osp); - if (rc) - GOTO(out_fid, rc); - + if (!osp->opd_storage->dd_rdonly) { + rc = osp_last_used_init(env, osp); + if (rc) + GOTO(out_fid, rc); + } /* Initialize precreation thread, it handles new * connections as well. */ rc = osp_init_precreate(osp); if (rc) GOTO(out_last_used, rc); + + /* + * Initialize synhronization mechanism taking + * care of propogating changes to OST in near + * transactional manner. + */ + rc = osp_sync_init(env, osp); + if (rc < 0) + GOTO(out_precreat, rc); + } else { + osp->opd_got_disconnected = 1; + rc = osp_update_init(osp); + if (rc != 0) + GOTO(out_fid, rc); } - /* - * Initialize synhronization mechanism taking - * care of propogating changes to OST in near - * transactional manner. - */ - rc = osp_sync_init(env, osp); + rc = osp_init_statfs(osp); if (rc) GOTO(out_precreat, rc); + ns_register_cancel(obd->obd_namespace, osp_cancel_weight); + /* * Initiate connect to OST */ - ll_generate_random_uuid(uuid); - class_uuid_unparse(uuid, &osp->opd_cluuid); - imp = obd->u.cli.cl_import; rc = ptlrpc_init_import(imp); @@ -1073,33 +1251,26 @@ static int osp_init0(const struct lu_env *env, struct osp_device *osp, RETURN(0); out: - /* stop sync thread */ - osp_sync_fini(osp); + if (!osp->opd_connect_mdt) + /* stop sync thread */ + osp_sync_fini(osp); out_precreat: /* stop precreate thread */ if (!osp->opd_connect_mdt) osp_precreate_fini(osp); + else + osp_update_fini(env, osp); out_last_used: if (!osp->opd_connect_mdt) osp_last_used_fini(env, osp); out_fid: obd_fid_fini(osp->opd_obd); out_proc: - ptlrpc_lprocfs_unregister_obd(obd); - lprocfs_obd_cleanup(obd); - if (osp->opd_symlink) - lprocfs_remove(&osp->opd_symlink); + osp_tunables_fini(osp); client_obd_cleanup(obd); out_ref: ptlrpcd_decref(); out_disconnect: - if (osp->opd_connect_mdt) { - struct client_obd *cli = &osp->opd_obd->u.cli; - if (cli->cl_rpc_lock != NULL) { - OBD_FREE_PTR(cli->cl_rpc_lock); - cli->cl_rpc_lock = NULL; - } - } obd_disconnect(osp->opd_storage_exp); out_fini: if (osdname) @@ -1123,10 +1294,8 @@ static struct lu_device *osp_device_free(const struct lu_env *env, { struct osp_device *osp = lu2osp_dev(lu); - if (atomic_read(&lu->ld_ref) && lu->ld_site) { - LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL); - lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer); - } + lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR, + lu_cdebug_printer); dt_device_fini(&osp->opd_dt_dev); OBD_FREE_PTR(osp); @@ -1186,40 +1355,23 @@ static struct lu_device *osp_device_fini(const struct lu_env *env, struct lu_device *ld) { struct osp_device *osp = lu2osp_dev(ld); - struct obd_import *imp; int rc; ENTRY; if (osp->opd_async_requests != NULL) { - dt_update_request_destroy(osp->opd_async_requests); + osp_update_request_destroy(env, osp->opd_async_requests); osp->opd_async_requests = NULL; } - if (osp->opd_storage_exp) + if (osp->opd_storage_exp) { + /* wait for the commit callbacks to complete */ + wait_event(osp->opd_sync_waitq, + atomic_read(&osp->opd_commits_registered) == 0); obd_disconnect(osp->opd_storage_exp); - - imp = osp->opd_obd->u.cli.cl_import; - - if (imp->imp_rq_pool) { - ptlrpc_free_rq_pool(imp->imp_rq_pool); - imp->imp_rq_pool = NULL; } - if (osp->opd_symlink) - lprocfs_remove(&osp->opd_symlink); - LASSERT(osp->opd_obd); - ptlrpc_lprocfs_unregister_obd(osp->opd_obd); - lprocfs_obd_cleanup(osp->opd_obd); - - if (osp->opd_connect_mdt) { - struct client_obd *cli = &osp->opd_obd->u.cli; - if (cli->cl_rpc_lock != NULL) { - OBD_FREE_PTR(cli->cl_rpc_lock); - cli->cl_rpc_lock = NULL; - } - } rc = client_obd_cleanup(osp->opd_obd); if (rc != 0) { @@ -1227,6 +1379,8 @@ static struct lu_device *osp_device_fini(const struct lu_env *env, RETURN(ERR_PTR(rc)); } + osp_tunables_fini(osp); + ptlrpcd_decref(); RETURN(NULL); @@ -1300,15 +1454,18 @@ static int osp_obd_connect(const struct lu_env *env, struct obd_export **exp, *ocd = *data; imp->imp_connect_flags_orig = ocd->ocd_connect_flags; + imp->imp_connect_flags2_orig = ocd->ocd_connect_flags2; ocd->ocd_version = LUSTRE_VERSION_CODE; ocd->ocd_index = data->ocd_index; - imp->imp_connect_flags_orig = ocd->ocd_connect_flags; rc = ptlrpc_connect_import(imp); if (rc) { CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc); GOTO(out, rc); + } else { + osp->opd_obd->u.cli.cl_seq->lcs_exp = + class_export_get(osp->opd_exp); } ptlrpc_pinger_add_import(imp); @@ -1369,7 +1526,7 @@ static int osp_obd_disconnect(struct obd_export *exp) * \retval negative negative errno if statfs failed. */ static int osp_obd_statfs(const struct lu_env *env, struct obd_export *exp, - struct obd_statfs *osfs, __u64 unused, __u32 flags) + struct obd_statfs *osfs, time64_t unused, __u32 flags) { struct obd_statfs *msfs; struct ptlrpc_request *req; @@ -1426,26 +1583,6 @@ out: } /** - * Prepare fid client. - * - * This function prepares the FID client for the OSP. It will check and assign - * the export (to MDT0) for its FID client, so OSP can allocate super sequence - * or lookup sequence in FLDB of MDT0. - * - * \param[in] osp OSP device - */ -static void osp_prepare_fid_client(struct osp_device *osp) -{ - LASSERT(osp->opd_obd->u.cli.cl_seq != NULL); - if (osp->opd_obd->u.cli.cl_seq->lcs_exp != NULL) - return; - - LASSERT(osp->opd_exp != NULL); - osp->opd_obd->u.cli.cl_seq->lcs_exp = - class_export_get(osp->opd_exp); -} - -/** * Implementation of obd_ops::o_import_event * * This function is called when some related import event happens. It will @@ -1463,6 +1600,7 @@ static int osp_import_event(struct obd_device *obd, struct obd_import *imp, enum obd_import_event event) { struct osp_device *d = lu2osp_dev(obd->obd_lu_dev); + int rc; switch (event) { case IMP_EVENT_DISCON: @@ -1480,11 +1618,15 @@ static int osp_import_event(struct obd_device *obd, struct obd_import *imp, break; case IMP_EVENT_INACTIVE: d->opd_imp_active = 0; + d->opd_imp_connected = 0; + d->opd_obd->obd_inactive = 1; if (d->opd_connect_mdt) break; - if (d->opd_pre != NULL) { - osp_pre_update_status(d, -ENODEV); + /* Import is invalid, we can`t get stripes so + * wakeup waiters */ + rc = imp->imp_deactive ? -ESHUTDOWN : -ENODEV; + osp_pre_update_status(d, rc); wake_up(&d->opd_pre_waitq); } @@ -1493,21 +1635,22 @@ static int osp_import_event(struct obd_device *obd, struct obd_import *imp, case IMP_EVENT_ACTIVE: d->opd_imp_active = 1; - osp_prepare_fid_client(d); if (d->opd_got_disconnected) d->opd_new_connection = 1; d->opd_imp_connected = 1; d->opd_imp_seen_connected = 1; + d->opd_obd->obd_inactive = 0; + wake_up(&d->opd_pre_waitq); if (d->opd_connect_mdt) break; - if (d->opd_pre != NULL) - wake_up(&d->opd_pre_waitq); - - __osp_sync_check_for_work(d); + osp_sync_check_for_work(d); CDEBUG(D_HA, "got connected\n"); break; case IMP_EVENT_INVALIDATE: + if (d->opd_connect_mdt) + osp_invalidate_request(d); + if (obd->obd_namespace == NULL) break; ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY); @@ -1541,7 +1684,7 @@ static int osp_import_event(struct obd_device *obd, struct obd_import *imp, * \retval negative negative errno if the ioctl handling failed. */ static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len, - void *karg, void *uarg) + void *karg, void __user *uarg) { struct obd_device *obd = exp->exp_obd; struct osp_device *d; @@ -1571,18 +1714,16 @@ static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = ptlrpc_set_import_active(obd->u.cli.cl_import, data->ioc_offset); break; - case OBD_IOC_PING_TARGET: - rc = ptlrpc_obd_ping(obd); - break; default: CERROR("%s: unrecognized ioctl %#x by %s\n", obd->obd_name, - cmd, current_comm()); + cmd, current->comm); rc = -ENOTTY; } module_put(THIS_MODULE); return rc; } + /** * Implementation of obd_ops::o_get_info * @@ -1596,14 +1737,12 @@ static int osp_iocontrol(unsigned int cmd, struct obd_export *exp, int len, * \param[in] key the key * \param[out] vallen length of \a val * \param[out] val holds the value returned by the key - * \param[in] unused necessary for the interface but unused * * \retval 0 0 if getting information succeeded. * \retval negative negative errno if getting information failed. */ static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp, - __u32 keylen, void *key, __u32 *vallen, void *val, - struct lov_stripe_md *unused) + __u32 keylen, void *key, __u32 *vallen, void *val) { int rc = -EINVAL; @@ -1627,43 +1766,56 @@ static int osp_obd_get_info(const struct lu_env *env, struct obd_export *exp, RETURN(rc); } -/** - * Implementation of obd_ops: o_fid_alloc - * - * Allocate a FID. There are two cases in which OSP performs - * FID allocation. - * - * 1. FID precreation for data objects, which is done in - * osp_precreate_fids() instead of this function. - * 2. FID allocation for each sub-stripe of a striped directory. - * Similar to other FID clients, OSP requests the sequence - * from its corresponding remote MDT, which in turn requests - * sequences from the sequence controller (MDT0). - * - * \param[in] env execution environment - * \param[in] exp export of the OSP - * \param[out] fid FID being allocated - * \param[in] unused necessary for the interface but unused. - * - * \retval 0 0 FID allocated successfully. - * \retval 1 1 FID allocated successfully and new sequence - * requested from seq meta server - * \retval negative negative errno if FID allocation failed. - */ -static int osp_fid_alloc(const struct lu_env *env, struct obd_export *exp, - struct lu_fid *fid, struct md_op_data *unused) +static int osp_obd_set_info_async(const struct lu_env *env, + struct obd_export *exp, + u32 keylen, void *key, + u32 vallen, void *val, + struct ptlrpc_request_set *set) { - struct client_obd *cli = &exp->exp_obd->u.cli; - struct osp_device *osp = lu2osp_dev(exp->exp_obd->obd_lu_dev); - struct lu_client_seq *seq = cli->cl_seq; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct obd_import *imp = obd->u.cli.cl_import; + struct osp_device *osp; + struct ptlrpc_request *req; + char *tmp; + int rc; - LASSERT(osp->opd_obd->u.cli.cl_seq != NULL); - /* Sigh, fid client is not ready yet */ - if (osp->opd_obd->u.cli.cl_seq->lcs_exp == NULL) - RETURN(-ENOTCONN); + if (KEY_IS(KEY_SPTLRPC_CONF)) { + sptlrpc_conf_client_adapt(exp->exp_obd); + RETURN(0); + } - RETURN(seq_client_alloc_fid(env, seq, fid)); + LASSERT(set != NULL); + if (!obd->obd_set_up || obd->obd_stopping) + RETURN(-EAGAIN); + osp = lu2osp_dev(obd->obd_lu_dev); + + req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO); + if (req == NULL) + RETURN(-ENOMEM); + + req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY, + RCL_CLIENT, keylen); + req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL, + RCL_CLIENT, vallen); + if (osp->opd_connect_mdt) + rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SET_INFO); + else + rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO); + if (rc) { + ptlrpc_request_free(req); + RETURN(rc); + } + + tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY); + memcpy(tmp, key, keylen); + tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL); + memcpy(tmp, val, vallen); + + ptlrpc_request_set_replen(req); + ptlrpc_set_add_req(set, req); + ptlrpc_check_set(NULL, set); + + RETURN(0); } /* context key constructor/destructor: mdt_key_init, mdt_key_fini */ @@ -1687,7 +1839,7 @@ struct lu_context_key osp_thread_key = { LU_KEY_INIT_FINI(osp_txn, struct osp_txn_info); struct lu_context_key osp_txn_key = { - .lct_tags = LCT_OSP_THREAD | LCT_TX_HANDLE, + .lct_tags = LCT_OSP_THREAD, .lct_init = osp_txn_key_init, .lct_fini = osp_txn_key_fini }; @@ -1713,7 +1865,7 @@ static struct lu_device_type osp_device_type = { .ldt_ctx_tags = LCT_MD_THREAD | LCT_DT_THREAD, }; -static struct obd_ops osp_obd_device_ops = { +static const struct obd_ops osp_obd_device_ops = { .o_owner = THIS_MODULE, .o_add_conn = client_import_add_conn, .o_del_conn = client_import_del_conn, @@ -1721,16 +1873,14 @@ static struct obd_ops osp_obd_device_ops = { .o_connect = osp_obd_connect, .o_disconnect = osp_obd_disconnect, .o_get_info = osp_obd_get_info, + .o_set_info_async = osp_obd_set_info_async, .o_import_event = osp_import_event, .o_iocontrol = osp_iocontrol, .o_statfs = osp_obd_statfs, .o_fid_init = client_fid_init, .o_fid_fini = client_fid_fini, - .o_fid_alloc = osp_fid_alloc, }; -struct llog_operations osp_mds_ost_orig_logops; - /** * Initialize OSP module. * @@ -1743,24 +1893,23 @@ struct llog_operations osp_mds_ost_orig_logops; * \retval 0 0 if initialization succeeds. * \retval negative negative errno if initialization failed. */ -static int __init osp_mod_init(void) +static int __init osp_init(void) { - struct obd_type *type; + struct obd_type *sym; int rc; rc = lu_kmem_init(osp_caches); if (rc) return rc; - - rc = class_register_type(&osp_obd_device_ops, NULL, true, NULL, + rc = class_register_type(&osp_obd_device_ops, NULL, false, NULL, LUSTRE_OSP_NAME, &osp_device_type); if (rc != 0) { lu_kmem_fini(osp_caches); return rc; } - rc = class_register_type(&lwp_obd_device_ops, NULL, true, NULL, + rc = class_register_type(&lwp_obd_device_ops, NULL, false, NULL, LUSTRE_LWP_NAME, &lwp_device_type); if (rc != 0) { class_unregister_type(LUSTRE_OSP_NAME); @@ -1768,24 +1917,15 @@ static int __init osp_mod_init(void) return rc; } - /* Note: add_rec/delcare_add_rec will be only used by catalogs */ - osp_mds_ost_orig_logops = llog_osd_ops; - osp_mds_ost_orig_logops.lop_add = llog_cat_add_rec; - osp_mds_ost_orig_logops.lop_declare_add = llog_cat_declare_add_rec; - - /* create "osc" entry in procfs for compatibility purposes */ - type = class_search_type(LUSTRE_OSC_NAME); - if (type != NULL && type->typ_procroot != NULL) - return rc; - - type = class_search_type(LUSTRE_OSP_NAME); - type->typ_procsym = lprocfs_register("osc", proc_lustre_root, - NULL, NULL); - if (IS_ERR(type->typ_procsym)) { - CERROR("osp: can't create compat entry \"osc\": %d\n", - (int) PTR_ERR(type->typ_procsym)); - type->typ_procsym = NULL; + /* create "osc" entry for compatibility purposes */ + sym = class_add_symlinks(LUSTRE_OSC_NAME, true); + if (IS_ERR(sym)) { + rc = PTR_ERR(sym); + /* does real "osc" already exist ? */ + if (rc == -EEXIST) + rc = 0; } + return rc; } @@ -1795,15 +1935,30 @@ static int __init osp_mod_init(void) * This callback is called when kernel unloads OSP module from memory, and * it will deregister OSP and LWP device type from obd_types (\see class_obd.c). */ -static void __exit osp_mod_exit(void) +static void __exit osp_exit(void) { + struct obd_type *sym = class_search_type(LUSTRE_OSC_NAME); + + /* if this was never fully initialized by the osc layer + * then we are responsible for freeing this obd_type + */ + if (sym) { + /* final put if we manage this obd type */ + if (sym->typ_sym_filter) + kobject_put(&sym->typ_kobj); + /* put reference taken by class_search_type */ + kobject_put(&sym->typ_kobj); + } + class_unregister_type(LUSTRE_LWP_NAME); class_unregister_type(LUSTRE_OSP_NAME); lu_kmem_fini(osp_caches); } -MODULE_AUTHOR("Intel, Inc. "); -MODULE_DESCRIPTION("Lustre OST Proxy Device ("LUSTRE_OSP_NAME")"); +MODULE_AUTHOR("OpenSFS, Inc. "); +MODULE_DESCRIPTION("Lustre OSD Storage Proxy ("LUSTRE_OSP_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(osp, LUSTRE_VERSION_STRING, osp_mod_init, osp_mod_exit); +module_init(osp_init); +module_exit(osp_exit);