X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmgs%2Fmgs_nids.c;h=abeff6363bf4482db4803c8e8f267e39ba3fbae1;hb=ad1f3518fa418b75d83809ccfcf6dd1489493726;hp=359e321e3e0f4f243b13950e6571815d00f51230;hpb=d190c63f22c5726c303393fce41e13d2dfccaee4;p=fs%2Flustre-release.git diff --git a/lustre/mgs/mgs_nids.c b/lustre/mgs/mgs_nids.c index 359e321..abeff63 100644 --- a/lustre/mgs/mgs_nids.c +++ b/lustre/mgs/mgs_nids.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,28 +40,17 @@ * Author: Jinshan Xiong */ -#ifndef EXPORT_SYMTAB -#define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MGS #define D_MGS D_CONFIG #ifdef __KERNEL__ -#include #include -#include #endif #include -#include #include -#include -#include -#include -#include -#include #include -#include + #include "mgs_internal.h" static unsigned int ir_timeout; @@ -71,7 +60,7 @@ static int nidtbl_is_sane(struct mgs_nidtbl *tbl) struct mgs_nidtbl_target *tgt; int version = 0; - LASSERT(cfs_mutex_is_locked(&tbl->mn_lock)); + LASSERT(mutex_is_locked(&tbl->mn_lock)); cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) { if (!tgt->mnt_version) continue; @@ -89,8 +78,8 @@ static int nidtbl_is_sane(struct mgs_nidtbl *tbl) * nidtbl entries will be packed in @pages by @unit_size units - entries * shouldn't cross unit boundaries. */ -static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl, - struct mgs_config_res *res, cfs_page_t **pages, +static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl, + struct mgs_config_res *res, struct page **pages, int nrpages, int units_total, int unit_size) { struct mgs_nidtbl_target *tgt; @@ -108,9 +97,9 @@ static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl, /* make sure unit_size is power 2 */ LASSERT((unit_size & (unit_size - 1)) == 0); - LASSERT(nrpages << CFS_PAGE_SHIFT >= units_total * unit_size); + LASSERT(nrpages << PAGE_CACHE_SHIFT >= units_total * unit_size); - cfs_mutex_lock(&tbl->mn_lock); + mutex_lock(&tbl->mn_lock); LASSERT(nidtbl_is_sane(tbl)); /* no more entries ? */ @@ -149,35 +138,41 @@ static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl, /* check if we need to consume remaining bytes. */ if (last_in_unit != NULL && bytes_in_unit) { - /* entry has been swapped. */ - __swab32s(&last_in_unit->mne_length); - last_in_unit->mne_length += bytes_in_unit; - __swab32s(&last_in_unit->mne_length); +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0) + /* May need to swab back to update the length.*/ + if (exp->exp_need_mne_swab) + lustre_swab_mgs_nidtbl_entry(last_in_unit); +#endif + last_in_unit->mne_length += bytes_in_unit; +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0) + if (exp->exp_need_mne_swab) + lustre_swab_mgs_nidtbl_entry(last_in_unit); +#endif rc += bytes_in_unit; buf += bytes_in_unit; last_in_unit = NULL; } LASSERT((rc & (unit_size - 1)) == 0); - if (units_in_page == 0) { - /* allocate a new page */ - pages[index] = cfs_alloc_page(CFS_ALLOC_STD); - if (pages[index] == NULL) { - rc = -ENOMEM; - break; - } + if (units_in_page == 0) { + /* allocate a new page */ + pages[index] = alloc_page(GFP_IOFS); + if (pages[index] == NULL) { + rc = -ENOMEM; + break; + } - /* destroy previous map */ - if (index > 0) - cfs_kunmap(pages[index - 1]); + /* destroy previous map */ + if (index > 0) + kunmap(pages[index - 1]); - /* reassign buffer */ - buf = cfs_kmap(pages[index]); - ++index; + /* reassign buffer */ + buf = kmap(pages[index]); + ++index; - units_in_page = CFS_PAGE_SIZE / unit_size; - LASSERT(units_in_page > 0); - } + units_in_page = PAGE_CACHE_SIZE / unit_size; + LASSERT(units_in_page > 0); + } /* allocate an unit */ LASSERT(((long)buf & (unit_size - 1)) == 0); @@ -198,7 +193,12 @@ static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl, entry->mne_nid_count = mti->mti_nid_count; memcpy(entry->u.nids, mti->mti_nids, mti->mti_nid_count * sizeof(lnet_nid_t)); - lustre_swab_mgs_nidtbl_entry(entry); + +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0) + /* For LU-1644, swab entry for 2.2 clients. */ + if (exp->exp_need_mne_swab) + lustre_swab_mgs_nidtbl_entry(entry); +#endif version = tgt->mnt_version; rc += entry_len; @@ -212,12 +212,12 @@ static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl, bytes_in_unit, index, nrpages, units_total); } if (index > 0) - cfs_kunmap(pages[index - 1]); + kunmap(pages[index - 1]); out: LASSERT(version <= tbl->mn_version); res->mcr_size = tbl->mn_version; res->mcr_offset = nobuf ? version : tbl->mn_version; - cfs_mutex_unlock(&tbl->mn_lock); + mutex_unlock(&tbl->mn_lock); LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */ CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n", @@ -225,81 +225,102 @@ out: RETURN(rc); } -static int nidtbl_update_version(struct obd_device *obd, struct mgs_nidtbl *tbl) +static int nidtbl_update_version(const struct lu_env *env, + struct mgs_device *mgs, + struct mgs_nidtbl *tbl) { - struct lvfs_run_ctxt saved; - struct file *file = NULL; - char filename[sizeof(MGS_NIDTBL_DIR) + 9]; - u64 version; - loff_t off = 0; - int rc; + struct dt_object *fsdb; + struct thandle *th; + u64 version; + struct lu_buf buf = { + .lb_buf = &version, + .lb_len = sizeof(version) + }; + loff_t off = 0; + int rc; ENTRY; - LASSERT(cfs_mutex_is_locked(&tbl->mn_lock)); - LASSERT(sizeof(filename) < 32); + LASSERT(mutex_is_locked(&tbl->mn_lock)); - sprintf(filename, "%s/%s", - MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name); + fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir, + tbl->mn_fsdb->fsdb_name, + S_IFREG | S_IRUGO | S_IWUSR); + if (IS_ERR(fsdb)) + RETURN(PTR_ERR(fsdb)); - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + th = dt_trans_create(env, mgs->mgs_bottom); + if (IS_ERR(th)) + GOTO(out_put, rc = PTR_ERR(th)); - file = l_filp_open(filename, O_RDWR|O_CREAT, 0660); - if (!IS_ERR(file)) { - version = cpu_to_le64(tbl->mn_version); - rc = lustre_fwrite(file, &version, sizeof(version), &off); - if (rc == sizeof(version)) - rc = 0; - filp_close(file, 0); - fsfilt_sync(obd, obd->u.mgs.mgs_sb); - } else { - rc = PTR_ERR(file); - } + th->th_sync = 1; /* update table synchronously */ + rc = dt_declare_record_write(env, fsdb, &buf, off, th); + if (rc) + GOTO(out, rc); - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - RETURN(rc); + rc = dt_trans_start_local(env, mgs->mgs_bottom, th); + if (rc) + GOTO(out, rc); + + version = cpu_to_le64(tbl->mn_version); + rc = dt_record_write(env, fsdb, &buf, &off, th); + +out: + dt_trans_stop(env, mgs->mgs_bottom, th); +out_put: + lu_object_put(env, &fsdb->do_lu); + RETURN(rc); } #define MGS_NIDTBL_VERSION_INIT 2 -static int nidtbl_read_version(struct obd_device *obd, struct mgs_nidtbl *tbl) +static int nidtbl_read_version(const struct lu_env *env, + struct mgs_device *mgs, struct mgs_nidtbl *tbl, + u64 *version) { - struct lvfs_run_ctxt saved; - struct file *file = NULL; - char filename[sizeof(MGS_NIDTBL_DIR) + 9]; - u64 version; + struct dt_object *fsdb; + struct lu_fid fid; + u64 tmpver; + struct lu_buf buf = { + .lb_buf = &tmpver, + .lb_len = sizeof(tmpver) + }; loff_t off = 0; int rc; ENTRY; - LASSERT(cfs_mutex_is_locked(&tbl->mn_lock)); - LASSERT(sizeof(filename) < 32); - - sprintf(filename, "%s/%s", - MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name); - - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - - file = l_filp_open(filename, O_RDONLY, 0); - if (!IS_ERR(file)) { - rc = lustre_fread(file, &version, sizeof(version), &off); - if (rc == sizeof(version)) - rc = cpu_to_le64(version); - else if (rc == 0) - rc = MGS_NIDTBL_VERSION_INIT; - else - CERROR("read version file %s error %d\n", filename, rc); - filp_close(file, 0); - } else { - rc = PTR_ERR(file); - if (rc == -ENOENT) - rc = MGS_NIDTBL_VERSION_INIT; - } - - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - RETURN(rc); + LASSERT(mutex_is_locked(&tbl->mn_lock)); + + LASSERT(mgs->mgs_nidtbl_dir); + rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name, + &fid); + if (rc == -ENOENT) { + *version = MGS_NIDTBL_VERSION_INIT; + RETURN(0); + } else if (rc < 0) { + RETURN(rc); + } + + fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid, + &mgs->mgs_dt_dev.dd_lu_dev, NULL); + if (IS_ERR(fsdb)) + RETURN(PTR_ERR(fsdb)); + + rc = dt_read(env, fsdb, &buf, &off); + if (rc == buf.lb_len) { + *version = le64_to_cpu(tmpver); + rc = 0; + } else if (rc == 0) { + *version = MGS_NIDTBL_VERSION_INIT; + } else { + CERROR("%s: read version file %s error %d\n", + mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc); + } + lu_object_put(env, &fsdb->do_lu); + RETURN(rc); } -static int mgs_nidtbl_write(struct fs_db *fsdb, struct mgs_target_info *mti) +static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb, + struct mgs_target_info *mti) { struct mgs_nidtbl *tbl; struct mgs_nidtbl_target *tgt; @@ -312,7 +333,7 @@ static int mgs_nidtbl_write(struct fs_db *fsdb, struct mgs_target_info *mti) LASSERT(type != 0); tbl = &fsdb->fsdb_nidtbl; - cfs_mutex_lock(&tbl->mn_lock); + mutex_lock(&tbl->mn_lock); cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) { struct mgs_target_info *info = &tgt->mnt_mti; if (type == tgt->mnt_type && @@ -339,11 +360,11 @@ static int mgs_nidtbl_write(struct fs_db *fsdb, struct mgs_target_info *mti) cfs_list_move_tail(&tgt->mnt_list, &tbl->mn_targets); - rc = nidtbl_update_version(fsdb->fsdb_obd, tbl); + rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl); EXIT; out: - cfs_mutex_unlock(&tbl->mn_lock); + mutex_unlock(&tbl->mn_lock); if (rc) CERROR("Write NID table version for file system %s error %d\n", fsdb->fsdb_name, rc); @@ -355,10 +376,10 @@ static void mgs_nidtbl_fini_fs(struct fs_db *fsdb) struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl; CFS_LIST_HEAD(head); - cfs_mutex_lock(&tbl->mn_lock); + mutex_lock(&tbl->mn_lock); tbl->mn_nr_targets = 0; cfs_list_splice_init(&tbl->mn_targets, &head); - cfs_mutex_unlock(&tbl->mn_lock); + mutex_unlock(&tbl->mn_lock); while (!cfs_list_empty(&head)) { struct mgs_nidtbl_target *tgt; @@ -368,23 +389,48 @@ static void mgs_nidtbl_fini_fs(struct fs_db *fsdb) } } -static int mgs_nidtbl_init_fs(struct fs_db *fsdb) +static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb) { struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl; + int rc; CFS_INIT_LIST_HEAD(&tbl->mn_targets); - cfs_mutex_init(&tbl->mn_lock); + mutex_init(&tbl->mn_lock); tbl->mn_nr_targets = 0; tbl->mn_fsdb = fsdb; - cfs_mutex_lock(&tbl->mn_lock); - tbl->mn_version = nidtbl_read_version(fsdb->fsdb_obd, tbl); - cfs_mutex_unlock(&tbl->mn_lock); - CDEBUG(D_MGS, "IR: current version is %llu\n", tbl->mn_version); - - return 0; + mutex_lock(&tbl->mn_lock); + rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version); + mutex_unlock(&tbl->mn_lock); + if (rc < 0) + CERROR("%s: IR: failed to read current version, rc = %d\n", + fsdb->fsdb_mgs->mgs_obd->obd_name, rc); + else + CDEBUG(D_MGS, "IR: current version is %llu\n", + tbl->mn_version); + + return rc; } /* --------- Imperative Recovery relies on nidtbl stuff ------- */ +void mgs_ir_notify_complete(struct fs_db *fsdb) +{ + struct timeval tv; + cfs_duration_t delta; + + atomic_set(&fsdb->fsdb_notify_phase, 0); + + /* do statistic */ + fsdb->fsdb_notify_count++; + delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start); + fsdb->fsdb_notify_total += delta; + if (delta > fsdb->fsdb_notify_max) + fsdb->fsdb_notify_max = delta; + + cfs_duration_usec(delta, &tv); + CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n", + fsdb->fsdb_name, tv.tv_sec, tv.tv_usec); +} + static int mgs_ir_notify(void *arg) { struct fs_db *fsdb = arg; @@ -394,106 +440,96 @@ static int mgs_ir_notify(void *arg) LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n"); sprintf(name, "mgs_%s_notify", fsdb->fsdb_name); - cfs_daemonize(name); - - cfs_complete(&fsdb->fsdb_notify_comp); - - mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER); - while (1) { - struct l_wait_info lwi = { 0 }; - struct lustre_handle lockh; - cfs_time_t curtime; - int lockrc; - int delta; - - l_wait_event(fsdb->fsdb_notify_waitq, - fsdb->fsdb_notify_stop || - cfs_atomic_read(&fsdb->fsdb_notify_phase), - &lwi); - if (fsdb->fsdb_notify_stop) - break; - CDEBUG(D_MGS, "%s woken up, phase is %d\n", - name, cfs_atomic_read(&fsdb->fsdb_notify_phase)); - - curtime = cfs_time_current(); - lockrc = mgs_get_lock(fsdb->fsdb_obd, &resid, &lockh); - if (lockrc == ELDLM_OK) { - cfs_atomic_set(&fsdb->fsdb_notify_phase, 0); - mgs_put_lock(&lockh); - - /* do statistic */ - fsdb->fsdb_notify_count++; - delta = (cfs_time_current() - curtime) / NSEC_PER_USEC; - fsdb->fsdb_notify_total += delta; - if (delta > fsdb->fsdb_notify_max) - fsdb->fsdb_notify_max = delta; - CDEBUG(D_MGS, "Revoke recover lock of %s spent %dus\n", - fsdb->fsdb_name, delta); - } else { - CERROR("Fatal error %d for fs %s\n", - lockrc, fsdb->fsdb_name); - } - } + complete(&fsdb->fsdb_notify_comp); - cfs_complete(&fsdb->fsdb_notify_comp); - return 0; -} + set_user_nice(current, -2); -int mgs_ir_init_fs(struct obd_device *obd, struct fs_db *fsdb) -{ - struct mgs_obd *mgs = &obd->u.mgs; - int rc; + mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER); + while (1) { + struct l_wait_info lwi = { 0 }; - if (!ir_timeout) - ir_timeout = OBD_IR_MGS_TIMEOUT; - - fsdb->fsdb_ir_state = IR_FULL; - if (cfs_time_before(cfs_time_current_sec(), - mgs->mgs_start_time + ir_timeout)) - fsdb->fsdb_ir_state = IR_STARTUP; - - /* start notify thread */ - fsdb->fsdb_obd = obd; - cfs_atomic_set(&fsdb->fsdb_notify_phase, 0); - cfs_waitq_init(&fsdb->fsdb_notify_waitq); - cfs_init_completion(&fsdb->fsdb_notify_comp); - rc = cfs_create_thread(mgs_ir_notify, fsdb, CFS_DAEMON_FLAGS); - if (rc > 0) - cfs_wait_for_completion(&fsdb->fsdb_notify_comp); - else - CERROR("Start notify thread error %d\n", rc); + l_wait_event(fsdb->fsdb_notify_waitq, + fsdb->fsdb_notify_stop || + atomic_read(&fsdb->fsdb_notify_phase), + &lwi); + if (fsdb->fsdb_notify_stop) + break; - mgs_nidtbl_init_fs(fsdb); - return 0; + CDEBUG(D_MGS, "%s woken up, phase is %d\n", + name, atomic_read(&fsdb->fsdb_notify_phase)); + + fsdb->fsdb_notify_start = cfs_time_current(); + mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER); + } + + complete(&fsdb->fsdb_notify_comp); + return 0; } -void mgs_ir_fini_fs(struct obd_device *obd, struct fs_db *fsdb) +int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs, + struct fs_db *fsdb) { - if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) - return; + struct task_struct *task; + + if (!ir_timeout) + ir_timeout = OBD_IR_MGS_TIMEOUT; + + fsdb->fsdb_ir_state = IR_FULL; + if (cfs_time_before(cfs_time_current_sec(), + mgs->mgs_start_time + ir_timeout)) + fsdb->fsdb_ir_state = IR_STARTUP; + fsdb->fsdb_nonir_clients = 0; + CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients); + + /* start notify thread */ + fsdb->fsdb_mgs = mgs; + atomic_set(&fsdb->fsdb_notify_phase, 0); + init_waitqueue_head(&fsdb->fsdb_notify_waitq); + init_completion(&fsdb->fsdb_notify_comp); + + task = kthread_run(mgs_ir_notify, fsdb, + "mgs_%s_notify", fsdb->fsdb_name); + if (!IS_ERR(task)) + wait_for_completion(&fsdb->fsdb_notify_comp); + else + CERROR("Start notify thread error %ld\n", PTR_ERR(task)); + + mgs_nidtbl_init_fs(env, fsdb); + return 0; +} - mgs_nidtbl_fini_fs(fsdb); +void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb) +{ + if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) + return; + + mgs_fsc_cleanup_by_fsdb(fsdb); + + mgs_nidtbl_fini_fs(fsdb); - fsdb->fsdb_notify_stop = 1; - cfs_waitq_signal(&fsdb->fsdb_notify_waitq); - cfs_wait_for_completion(&fsdb->fsdb_notify_comp); + LASSERT(cfs_list_empty(&fsdb->fsdb_clients)); + + fsdb->fsdb_notify_stop = 1; + wake_up(&fsdb->fsdb_notify_waitq); + wait_for_completion(&fsdb->fsdb_notify_comp); } -/* caller must have held fsdb_sem */ +/* caller must have held fsdb_mutex */ static inline void ir_state_graduate(struct fs_db *fsdb) { - struct mgs_obd *mgs = &fsdb->fsdb_obd->u.mgs; - if (fsdb->fsdb_ir_state == IR_STARTUP) { - if (cfs_time_before(mgs->mgs_start_time + ir_timeout, + if (cfs_time_before(fsdb->fsdb_mgs->mgs_start_time + ir_timeout, cfs_time_current_sec())) { fsdb->fsdb_ir_state = IR_FULL; + if (fsdb->fsdb_nonir_clients) + fsdb->fsdb_ir_state = IR_PARTIAL; } } } -int mgs_ir_update(struct obd_device *obd, struct mgs_target_info *mti) +int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs, + struct mgs_target_info *mti) { struct fs_db *fsdb; bool notify = true; @@ -502,16 +538,16 @@ int mgs_ir_update(struct obd_device *obd, struct mgs_target_info *mti) if (mti->mti_instance == 0) return -EINVAL; - rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb); + rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb); if (rc) return rc; - rc = mgs_nidtbl_write(fsdb, mti); + rc = mgs_nidtbl_write(env, fsdb, mti); if (rc) return rc; /* check ir state */ - cfs_down(&fsdb->fsdb_sem); + mutex_lock(&fsdb->fsdb_mutex); ir_state_graduate(fsdb); switch (fsdb->fsdb_ir_state) { case IR_FULL: @@ -525,16 +561,16 @@ int mgs_ir_update(struct obd_device *obd, struct mgs_target_info *mti) default: LBUG(); } - cfs_up(&fsdb->fsdb_sem); - - LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify)); - if (notify) { - CDEBUG(D_MGS, "Try to revoke recover lock of %s\n", - fsdb->fsdb_name); - cfs_atomic_inc(&fsdb->fsdb_notify_phase); - cfs_waitq_signal(&fsdb->fsdb_notify_waitq); - } - return 0; + mutex_unlock(&fsdb->fsdb_mutex); + + LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify)); + if (notify) { + CDEBUG(D_MGS, "Try to revoke recover lock of %s\n", + fsdb->fsdb_name); + atomic_inc(&fsdb->fsdb_notify_phase); + wake_up(&fsdb->fsdb_notify_waitq); + } + return 0; } /* NID table can be cached by two entities: Clients and MDTs */ @@ -576,7 +612,8 @@ static int delogname(char *logname, char *fsname, int *typ) int mgs_get_ir_logs(struct ptlrpc_request *req) { - struct obd_device *obd = req->rq_export->exp_obd; + struct lu_env *env = req->rq_svc_thread->t_env; + struct mgs_device *mgs = exp2mgs_dev(req->rq_export); struct fs_db *fsdb; struct mgs_config_body *body; struct mgs_config_res *res; @@ -585,14 +622,13 @@ int mgs_get_ir_logs(struct ptlrpc_request *req) char fsname[16]; long bufsize; int unit_size; - int type; int rc = 0; int i; int bytes; int page_count; int nrpages; - cfs_page_t **pages = NULL; + struct page **pages = NULL; ENTRY; body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY); @@ -606,12 +642,12 @@ int mgs_get_ir_logs(struct ptlrpc_request *req) if (rc) RETURN(rc); - rc = mgs_find_or_make_fsdb(obd, fsname, &fsdb); + rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb); if (rc) - GOTO(out, rc); + RETURN(rc); bufsize = body->mcb_units << body->mcb_bits; - nrpages = (bufsize + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; + nrpages = (bufsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; if (nrpages > PTLRPC_MAX_BRW_PAGES) RETURN(-EINVAL); @@ -622,48 +658,42 @@ int mgs_get_ir_logs(struct ptlrpc_request *req) if (pages == NULL) RETURN(-ENOMEM); - rc = req_capsule_server_pack(&req->rq_pill); - if (rc) - GOTO(out, rc); - res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES); if (res == NULL) GOTO(out, rc = -EINVAL); res->mcr_offset = body->mcb_offset; - unit_size = min_t(int, 1 << body->mcb_bits, CFS_PAGE_SIZE); - bytes = mgs_nidtbl_read(obd, &fsdb->fsdb_nidtbl, res, pages, nrpages, - bufsize / unit_size, unit_size); - if (bytes < 0) - GOTO(out, rc = bytes); - - /* start bulk transfer */ - page_count = (bytes + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; - LASSERT(page_count <= nrpages); - desc = ptlrpc_prep_bulk_exp(req, page_count, - BULK_PUT_SOURCE, MGS_BULK_PORTAL); - if (desc == NULL) - GOTO(out, rc = -ENOMEM); - - for (i = 0; i < page_count && bytes > 0; i++) { - ptlrpc_prep_bulk_page(desc, pages[i], 0, - min_t(int, bytes, CFS_PAGE_SIZE)); - bytes -= CFS_PAGE_SIZE; + unit_size = min_t(int, 1 << body->mcb_bits, PAGE_CACHE_SIZE); + bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res, + pages, nrpages, bufsize / unit_size, unit_size); + if (bytes < 0) + GOTO(out, rc = bytes); + + /* start bulk transfer */ + page_count = (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; + LASSERT(page_count <= nrpages); + desc = ptlrpc_prep_bulk_exp(req, page_count, 1, + BULK_PUT_SOURCE, MGS_BULK_PORTAL); + if (desc == NULL) + GOTO(out, rc = -ENOMEM); + + for (i = 0; i < page_count && bytes > 0; i++) { + ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, + min_t(int, bytes, PAGE_CACHE_SIZE)); + bytes -= PAGE_CACHE_SIZE; } rc = target_bulk_io(req->rq_export, desc, &lwi); - ptlrpc_free_bulk(desc); + ptlrpc_free_bulk_pin(desc); out: - if (pages) { - for (i = 0; i < nrpages; i++) { - if (pages[i] == NULL) - break; - cfs_free_page(pages[i]); - } - OBD_FREE(pages, sizeof(*pages) * nrpages); - } - return rc; + for (i = 0; i < nrpages; i++) { + if (pages[i] == NULL) + break; + __free_page(pages[i]); + } + OBD_FREE(pages, sizeof(*pages) * nrpages); + return rc; } static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf) @@ -683,9 +713,11 @@ static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf) CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n", fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]); - cfs_down(&fsdb->fsdb_sem); + mutex_lock(&fsdb->fsdb_mutex); + if (state == IR_FULL && fsdb->fsdb_nonir_clients) + state = IR_PARTIAL; fsdb->fsdb_ir_state = state; - cfs_up(&fsdb->fsdb_sem); + mutex_unlock(&fsdb->fsdb_mutex); return 0; } @@ -716,25 +748,25 @@ static struct lproc_ir_cmd { { "0", 1, lprocfs_ir_clear_stats } }; -int lprocfs_wr_ir_state(struct file *file, const char *buffer, - unsigned long count, void *data) +int lprocfs_wr_ir_state(struct file *file, const char __user *buffer, + size_t count, void *data) { struct fs_db *fsdb = data; char *kbuf; char *ptr; int rc = 0; - if (count > CFS_PAGE_SIZE) - return -EINVAL; + if (count == 0 || count >= PAGE_CACHE_SIZE) + return -EINVAL; - OBD_ALLOC(kbuf, count + 1); - if (kbuf == NULL) - return -ENOMEM; + OBD_ALLOC(kbuf, count + 1); + if (kbuf == NULL) + return -ENOMEM; - if (copy_from_user(kbuf, buffer, count)) { - OBD_FREE(kbuf, count); - return -EFAULT; - } + if (copy_from_user(kbuf, buffer, count)) { + OBD_FREE(kbuf, count + 1); + return -EFAULT; + } kbuf[count] = 0; /* buffer is supposed to end with 0 */ if (kbuf[count - 1] == '\n') @@ -779,29 +811,163 @@ int lprocfs_rd_ir_state(struct seq_file *seq, void *data) struct fs_db *fsdb = data; struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl; const char *ir_strings[] = IR_STRINGS; + struct timeval tv_max; + struct timeval tv; - /* mgs_live_seq_show() already holds fsdb_sem. */ + /* mgs_live_seq_show() already holds fsdb_mutex. */ ir_state_graduate(fsdb); + seq_printf(seq, "\nimperative_recovery_state:\n"); seq_printf(seq, - "\tstate: %s, nidtbl version: %lld\n", - ir_strings[fsdb->fsdb_ir_state], tbl->mn_version); - seq_printf(seq, "\tnotify total/max/count: %u/%u/%u\n", - fsdb->fsdb_notify_total, fsdb->fsdb_notify_max, + " state: %s\n" + " nonir_clients: %d\n" + " nidtbl_version: %lld\n", + ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients, + tbl->mn_version); + + cfs_duration_usec(fsdb->fsdb_notify_total, &tv); + cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max); + + seq_printf(seq, " notify_duration_total: %lu.%06lu\n" + " notify_duation_max: %lu.%06lu\n" + " notify_count: %u\n", + tv.tv_sec, tv.tv_usec, + tv_max.tv_sec, tv_max.tv_usec, fsdb->fsdb_notify_count); + return 0; } -int lprocfs_rd_ir_timeout(char *page, char **start, off_t off, int count, - int *eof, void *data) +int lprocfs_ir_timeout_seq_show(struct seq_file *m, void *data) { - *eof = 1; - return snprintf(page, count, "%d\n", ir_timeout); + return lprocfs_uint_seq_show(m, &ir_timeout); } -int lprocfs_wr_ir_timeout(struct file *file, const char *buffer, - unsigned long count, void *data) +ssize_t lprocfs_ir_timeout_seq_write(struct file *file, + const char __user *buffer, + size_t count, loff_t *off) { return lprocfs_wr_uint(file, buffer, count, &ir_timeout); } +/* --------------- Handle non IR support clients --------------- */ +/* attach a lustre file system to an export */ +int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp, + char *fsname) +{ + struct mgs_export_data *data = &exp->u.eu_mgs_data; + struct mgs_device *mgs = exp2mgs_dev(exp); + struct fs_db *fsdb; + struct mgs_fsc *fsc = NULL; + struct mgs_fsc *new_fsc = NULL; + bool found = false; + int rc; + ENTRY; + + rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb); + if (rc) + RETURN(rc); + + /* allocate a new fsc in case we need it in spinlock. */ + OBD_ALLOC_PTR(new_fsc); + if (new_fsc == NULL) + RETURN(-ENOMEM); + + CFS_INIT_LIST_HEAD(&new_fsc->mfc_export_list); + CFS_INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list); + new_fsc->mfc_fsdb = fsdb; + new_fsc->mfc_export = class_export_get(exp); + new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) & + OBD_CONNECT_IMP_RECOV); + + rc = -EEXIST; + mutex_lock(&fsdb->fsdb_mutex); + + /* tend to find it in export list because this list is shorter. */ + spin_lock(&data->med_lock); + cfs_list_for_each_entry(fsc, &data->med_clients, mfc_export_list) { + if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) { + found = true; + break; + } + } + if (!found) { + fsc = new_fsc; + new_fsc = NULL; + + /* add it into export list. */ + cfs_list_add(&fsc->mfc_export_list, &data->med_clients); + + /* add into fsdb list. */ + cfs_list_add(&fsc->mfc_fsdb_list, &fsdb->fsdb_clients); + if (!fsc->mfc_ir_capable) { + ++fsdb->fsdb_nonir_clients; + if (fsdb->fsdb_ir_state == IR_FULL) + fsdb->fsdb_ir_state = IR_PARTIAL; + } + rc = 0; + } + spin_unlock(&data->med_lock); + mutex_unlock(&fsdb->fsdb_mutex); + + if (new_fsc) { + class_export_put(new_fsc->mfc_export); + OBD_FREE_PTR(new_fsc); + } + RETURN(rc); +} + +void mgs_fsc_cleanup(struct obd_export *exp) +{ + struct mgs_export_data *data = &exp->u.eu_mgs_data; + struct mgs_fsc *fsc, *tmp; + CFS_LIST_HEAD(head); + + spin_lock(&data->med_lock); + cfs_list_splice_init(&data->med_clients, &head); + spin_unlock(&data->med_lock); + + cfs_list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) { + struct fs_db *fsdb = fsc->mfc_fsdb; + + LASSERT(fsc->mfc_export == exp); + + mutex_lock(&fsdb->fsdb_mutex); + cfs_list_del_init(&fsc->mfc_fsdb_list); + if (fsc->mfc_ir_capable == 0) { + --fsdb->fsdb_nonir_clients; + LASSERT(fsdb->fsdb_ir_state != IR_FULL); + if (fsdb->fsdb_nonir_clients == 0 && + fsdb->fsdb_ir_state == IR_PARTIAL) + fsdb->fsdb_ir_state = IR_FULL; + } + mutex_unlock(&fsdb->fsdb_mutex); + cfs_list_del_init(&fsc->mfc_export_list); + class_export_put(fsc->mfc_export); + OBD_FREE_PTR(fsc); + } +} + +/* must be called with fsdb->fsdb_mutex held */ +void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb) +{ + struct mgs_fsc *fsc, *tmp; + + cfs_list_for_each_entry_safe(fsc, tmp, &fsdb->fsdb_clients, + mfc_fsdb_list) { + struct mgs_export_data *data = &fsc->mfc_export->u.eu_mgs_data; + + LASSERT(fsdb == fsc->mfc_fsdb); + cfs_list_del_init(&fsc->mfc_fsdb_list); + + spin_lock(&data->med_lock); + cfs_list_del_init(&fsc->mfc_export_list); + spin_unlock(&data->med_lock); + class_export_put(fsc->mfc_export); + OBD_FREE_PTR(fsc); + } + + fsdb->fsdb_nonir_clients = 0; + if (fsdb->fsdb_ir_state == IR_PARTIAL) + fsdb->fsdb_ir_state = IR_FULL; +}