/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License version 2 for more details (a copy is included * in the LICENSE file that accompanied this code). * * 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. * * GPL HEADER END */ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * * Copyright (c) 2010, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/ldlm/ldlm_pool.c * * Author: Yury Umanets */ /* * Idea of this code is rather simple. Each second, for each server namespace * we have SLV - server lock volume which is calculated on current number of * granted locks, grant speed for past period, etc - that is, locking load. * This SLV number may be thought as a flow definition for simplicity. It is * sent to clients with each occasion to let them know what is current load * situation on the server. By default, at the beginning, SLV on server is * set max value which is calculated as the following: allow to one client * have all locks of limit ->pl_limit for 10h. * * Next, on clients, number of cached locks is not limited artificially in any * way as it was before. Instead, client calculates CLV, that is, client lock * volume for each lock and compares it with last SLV from the server. CLV is * calculated as the number of locks in LRU * lock live time in seconds. If * CLV > SLV - lock is canceled. * * Client has LVF, that is, lock volume factor which regulates how much * sensitive client should be about last SLV from server. The higher LVF is the * more locks will be canceled on client. Default value for it is 1. Setting LVF * to 2 means that client will cancel locks 2 times faster. * * Locks on a client will be canceled more intensively in these cases: * (1) if SLV is smaller, that is, load is higher on the server; * (2) client has a lot of locks (the more locks are held by client, the bigger * chances that some of them should be canceled); * (3) client has old locks (taken some time ago); * * Thus, according to flow paradigm that we use for better understanding SLV, * CLV is the volume of particle in flow described by SLV. According to this, * if flow is getting thinner, more and more particles become outside of it and * as particles are locks, they should be canceled. * * General idea of this belongs to Vitaly Fertman (vitaly@clusterfs.com). * Andreas Dilger (adilger@clusterfs.com) proposed few nice ideas like using * LVF and many cleanups. Flow definition to allow more easy understanding of * the logic belongs to Nikita Danilov (nikita@clusterfs.com) as well as many * cleanups and fixes. And design and implementation are done by Yury Umanets * (umka@clusterfs.com). * * Glossary for terms used: * * pl_limit - Number of allowed locks in pool. Applies to server and client * side (tunable); * * pl_granted - Number of granted locks (calculated); * pl_grant_rate - Number of granted locks for last T (calculated); * pl_cancel_rate - Number of canceled locks for last T (calculated); * pl_grant_speed - Grant speed (GR - CR) for last T (calculated); * pl_grant_plan - Planned number of granted locks for next T (calculated); * pl_server_lock_volume - Current server lock volume (calculated); * * As it may be seen from list above, we have few possible tunables which may * affect behavior much. They all may be modified via proc. However, they also * give a possibility for constructing few pre-defined behavior policies. If * none of predefines is suitable for a working pattern being used, new one may * be "constructed" via proc tunables. */ #define DEBUG_SUBSYSTEM S_LDLM #include "../include/lustre_dlm.h" #include "../include/cl_object.h" #include "../include/obd_class.h" #include "../include/obd_support.h" #include "ldlm_internal.h" /* * 50 ldlm locks for 1MB of RAM. */ #define LDLM_POOL_HOST_L ((NUM_CACHEPAGES >> (20 - PAGE_CACHE_SHIFT)) * 50) /* * Maximal possible grant step plan in %. */ #define LDLM_POOL_MAX_GSP (30) /* * Minimal possible grant step plan in %. */ #define LDLM_POOL_MIN_GSP (1) /* * This controls the speed of reaching LDLM_POOL_MAX_GSP * with increasing thread period. */ #define LDLM_POOL_GSP_STEP_SHIFT (2) /* * LDLM_POOL_GSP% of all locks is default GP. */ #define LDLM_POOL_GP(L) (((L) * LDLM_POOL_MAX_GSP) / 100) /* * Max age for locks on clients. */ #define LDLM_POOL_MAX_AGE (36000) /* * The granularity of SLV calculation. */ #define LDLM_POOL_SLV_SHIFT (10) static inline __u64 dru(__u64 val, __u32 shift, int round_up) { return (val + (round_up ? (1 << shift) - 1 : 0)) >> shift; } static inline __u64 ldlm_pool_slv_max(__u32 L) { /* * Allow to have all locks for 1 client for 10 hrs. * Formula is the following: limit * 10h / 1 client. */ __u64 lim = (__u64)L * LDLM_POOL_MAX_AGE / 1; return lim; } static inline __u64 ldlm_pool_slv_min(__u32 L) { return 1; } enum { LDLM_POOL_FIRST_STAT = 0, LDLM_POOL_GRANTED_STAT = LDLM_POOL_FIRST_STAT, LDLM_POOL_GRANT_STAT, LDLM_POOL_CANCEL_STAT, LDLM_POOL_GRANT_RATE_STAT, LDLM_POOL_CANCEL_RATE_STAT, LDLM_POOL_GRANT_PLAN_STAT, LDLM_POOL_SLV_STAT, LDLM_POOL_SHRINK_REQTD_STAT, LDLM_POOL_SHRINK_FREED_STAT, LDLM_POOL_RECALC_STAT, LDLM_POOL_TIMING_STAT, LDLM_POOL_LAST_STAT }; static inline struct ldlm_namespace *ldlm_pl2ns(struct ldlm_pool *pl) { return container_of(pl, struct ldlm_namespace, ns_pool); } /** * Calculates suggested grant_step in % of available locks for passed * \a period. This is later used in grant_plan calculations. */ static inline int ldlm_pool_t2gsp(unsigned int t) { /* * This yields 1% grant step for anything below LDLM_POOL_GSP_STEP * and up to 30% for anything higher than LDLM_POOL_GSP_STEP. * * How this will affect execution is the following: * * - for thread period 1s we will have grant_step 1% which good from * pov of taking some load off from server and push it out to clients. * This is like that because 1% for grant_step means that server will * not allow clients to get lots of locks in short period of time and * keep all old locks in their caches. Clients will always have to * get some locks back if they want to take some new; * * - for thread period 10s (which is default) we will have 23% which * means that clients will have enough of room to take some new locks * without getting some back. All locks from this 23% which were not * taken by clients in current period will contribute in SLV growing. * SLV growing means more locks cached on clients until limit or grant * plan is reached. */ return LDLM_POOL_MAX_GSP - ((LDLM_POOL_MAX_GSP - LDLM_POOL_MIN_GSP) >> (t >> LDLM_POOL_GSP_STEP_SHIFT)); } /** * Recalculates next grant limit on passed \a pl. * * \pre ->pl_lock is locked. */ static void ldlm_pool_recalc_grant_plan(struct ldlm_pool *pl) { int granted, grant_step, limit; limit = ldlm_pool_get_limit(pl); granted = atomic_read(&pl->pl_granted); grant_step = ldlm_pool_t2gsp(pl->pl_recalc_period); grant_step = ((limit - granted) * grant_step) / 100; pl->pl_grant_plan = granted + grant_step; limit = (limit * 5) >> 2; if (pl->pl_grant_plan > limit) pl->pl_grant_plan = limit; } /** * Recalculates next SLV on passed \a pl. * * \pre ->pl_lock is locked. */ static void ldlm_pool_recalc_slv(struct ldlm_pool *pl) { int granted; int grant_plan; int round_up; __u64 slv; __u64 slv_factor; __u64 grant_usage; __u32 limit; slv = pl->pl_server_lock_volume; grant_plan = pl->pl_grant_plan; limit = ldlm_pool_get_limit(pl); granted = atomic_read(&pl->pl_granted); round_up = granted < limit; grant_usage = max_t(int, limit - (granted - grant_plan), 1); /* * Find out SLV change factor which is the ratio of grant usage * from limit. SLV changes as fast as the ratio of grant plan * consumption. The more locks from grant plan are not consumed * by clients in last interval (idle time), the faster grows * SLV. And the opposite, the more grant plan is over-consumed * (load time) the faster drops SLV. */ slv_factor = grant_usage << LDLM_POOL_SLV_SHIFT; do_div(slv_factor, limit); slv = slv * slv_factor; slv = dru(slv, LDLM_POOL_SLV_SHIFT, round_up); if (slv > ldlm_pool_slv_max(limit)) slv = ldlm_pool_slv_max(limit); else if (slv < ldlm_pool_slv_min(limit)) slv = ldlm_pool_slv_min(limit); pl->pl_server_lock_volume = slv; } /** * Recalculates next stats on passed \a pl. * * \pre ->pl_lock is locked. */ static void ldlm_pool_recalc_stats(struct ldlm_pool *pl) { int grant_plan = pl->pl_grant_plan; __u64 slv = pl->pl_server_lock_volume; int granted = atomic_read(&pl->pl_granted); int grant_rate = atomic_read(&pl->pl_grant_rate); int cancel_rate = atomic_read(&pl->pl_cancel_rate); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_SLV_STAT, slv); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANTED_STAT, granted); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANT_RATE_STAT, grant_rate); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANT_PLAN_STAT, grant_plan); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_CANCEL_RATE_STAT, cancel_rate); } /** * Sets current SLV into obd accessible via ldlm_pl2ns(pl)->ns_obd. */ static void ldlm_srv_pool_push_slv(struct ldlm_pool *pl) { struct obd_device *obd; /* * Set new SLV in obd field for using it later without accessing the * pool. This is required to avoid race between sending reply to client * with new SLV and cleanup server stack in which we can't guarantee * that namespace is still alive. We know only that obd is alive as * long as valid export is alive. */ obd = ldlm_pl2ns(pl)->ns_obd; LASSERT(obd != NULL); write_lock(&obd->obd_pool_lock); obd->obd_pool_slv = pl->pl_server_lock_volume; write_unlock(&obd->obd_pool_lock); } /** * Recalculates all pool fields on passed \a pl. * * \pre ->pl_lock is not locked. */ static int ldlm_srv_pool_recalc(struct ldlm_pool *pl) { time_t recalc_interval_sec; recalc_interval_sec = get_seconds() - pl->pl_recalc_time; if (recalc_interval_sec < pl->pl_recalc_period) return 0; spin_lock(&pl->pl_lock); recalc_interval_sec = get_seconds() - pl->pl_recalc_time; if (recalc_interval_sec < pl->pl_recalc_period) { spin_unlock(&pl->pl_lock); return 0; } /* * Recalc SLV after last period. This should be done * _before_ recalculating new grant plan. */ ldlm_pool_recalc_slv(pl); /* * Make sure that pool informed obd of last SLV changes. */ ldlm_srv_pool_push_slv(pl); /* * Update grant_plan for new period. */ ldlm_pool_recalc_grant_plan(pl); pl->pl_recalc_time = get_seconds(); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_TIMING_STAT, recalc_interval_sec); spin_unlock(&pl->pl_lock); return 0; } /** * This function is used on server side as main entry point for memory * pressure handling. It decreases SLV on \a pl according to passed * \a nr and \a gfp_mask. * * Our goal here is to decrease SLV such a way that clients hold \a nr * locks smaller in next 10h. */ static int ldlm_srv_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask) { __u32 limit; /* * VM is asking how many entries may be potentially freed. */ if (nr == 0) return atomic_read(&pl->pl_granted); /* * Client already canceled locks but server is already in shrinker * and can't cancel anything. Let's catch this race. */ if (atomic_read(&pl->pl_granted) == 0) return 0; spin_lock(&pl->pl_lock); /* * We want shrinker to possibly cause cancellation of @nr locks from * clients or grant approximately @nr locks smaller next intervals. * * This is why we decreased SLV by @nr. This effect will only be as * long as one re-calc interval (1s these days) and this should be * enough to pass this decreased SLV to all clients. On next recalc * interval pool will either increase SLV if locks load is not high * or will keep on same level or even decrease again, thus, shrinker * decreased SLV will affect next recalc intervals and this way will * make locking load lower. */ if (nr < pl->pl_server_lock_volume) { pl->pl_server_lock_volume = pl->pl_server_lock_volume - nr; } else { limit = ldlm_pool_get_limit(pl); pl->pl_server_lock_volume = ldlm_pool_slv_min(limit); } /* * Make sure that pool informed obd of last SLV changes. */ ldlm_srv_pool_push_slv(pl); spin_unlock(&pl->pl_lock); /* * We did not really free any memory here so far, it only will be * freed later may be, so that we return 0 to not confuse VM. */ return 0; } /** * Setup server side pool \a pl with passed \a limit. */ static int ldlm_srv_pool_setup(struct ldlm_pool *pl, int limit) { struct obd_device *obd; obd = ldlm_pl2ns(pl)->ns_obd; LASSERT(obd != NULL && obd != LP_POISON); LASSERT(obd->obd_type != LP_POISON); write_lock(&obd->obd_pool_lock); obd->obd_pool_limit = limit; write_unlock(&obd->obd_pool_lock); ldlm_pool_set_limit(pl, limit); return 0; } /** * Sets SLV and Limit from ldlm_pl2ns(pl)->ns_obd tp passed \a pl. */ static void ldlm_cli_pool_pop_slv(struct ldlm_pool *pl) { struct obd_device *obd; /* * Get new SLV and Limit from obd which is updated with coming * RPCs. */ obd = ldlm_pl2ns(pl)->ns_obd; LASSERT(obd != NULL); read_lock(&obd->obd_pool_lock); pl->pl_server_lock_volume = obd->obd_pool_slv; ldlm_pool_set_limit(pl, obd->obd_pool_limit); read_unlock(&obd->obd_pool_lock); } /** * Recalculates client size pool \a pl according to current SLV and Limit. */ static int ldlm_cli_pool_recalc(struct ldlm_pool *pl) { time_t recalc_interval_sec; int ret; recalc_interval_sec = get_seconds() - pl->pl_recalc_time; if (recalc_interval_sec < pl->pl_recalc_period) return 0; spin_lock(&pl->pl_lock); /* * Check if we need to recalc lists now. */ recalc_interval_sec = get_seconds() - pl->pl_recalc_time; if (recalc_interval_sec < pl->pl_recalc_period) { spin_unlock(&pl->pl_lock); return 0; } /* * Make sure that pool knows last SLV and Limit from obd. */ ldlm_cli_pool_pop_slv(pl); spin_unlock(&pl->pl_lock); /* * Do not cancel locks in case lru resize is disabled for this ns. */ if (!ns_connect_lru_resize(ldlm_pl2ns(pl))) { ret = 0; goto out; } /* * In the time of canceling locks on client we do not need to maintain * sharp timing, we only want to cancel locks asap according to new SLV. * It may be called when SLV has changed much, this is why we do not * take into account pl->pl_recalc_time here. */ ret = ldlm_cancel_lru(ldlm_pl2ns(pl), 0, LCF_ASYNC, LDLM_CANCEL_LRUR); out: spin_lock(&pl->pl_lock); /* * Time of LRU resizing might be longer than period, * so update after LRU resizing rather than before it. */ pl->pl_recalc_time = get_seconds(); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_TIMING_STAT, recalc_interval_sec); spin_unlock(&pl->pl_lock); return ret; } /** * This function is main entry point for memory pressure handling on client * side. Main goal of this function is to cancel some number of locks on * passed \a pl according to \a nr and \a gfp_mask. */ static int ldlm_cli_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask) { struct ldlm_namespace *ns; int unused; ns = ldlm_pl2ns(pl); /* * Do not cancel locks in case lru resize is disabled for this ns. */ if (!ns_connect_lru_resize(ns)) return 0; /* * Make sure that pool knows last SLV and Limit from obd. */ ldlm_cli_pool_pop_slv(pl); spin_lock(&ns->ns_lock); unused = ns->ns_nr_unused; spin_unlock(&ns->ns_lock); if (nr == 0) return (unused / 100) * sysctl_vfs_cache_pressure; else return ldlm_cancel_lru(ns, nr, LCF_ASYNC, LDLM_CANCEL_SHRINK); } static const struct ldlm_pool_ops ldlm_srv_pool_ops = { .po_recalc = ldlm_srv_pool_recalc, .po_shrink = ldlm_srv_pool_shrink, .po_setup = ldlm_srv_pool_setup }; static const struct ldlm_pool_ops ldlm_cli_pool_ops = { .po_recalc = ldlm_cli_pool_recalc, .po_shrink = ldlm_cli_pool_shrink }; /** * Pool recalc wrapper. Will call either client or server pool recalc callback * depending what pool \a pl is used. */ int ldlm_pool_recalc(struct ldlm_pool *pl) { time_t recalc_interval_sec; int count; recalc_interval_sec = get_seconds() - pl->pl_recalc_time; if (recalc_interval_sec <= 0) goto recalc; spin_lock(&pl->pl_lock); if (recalc_interval_sec > 0) { /* * Update pool statistics every 1s. */ ldlm_pool_recalc_stats(pl); /* * Zero out all rates and speed for the last period. */ atomic_set(&pl->pl_grant_rate, 0); atomic_set(&pl->pl_cancel_rate, 0); } spin_unlock(&pl->pl_lock); recalc: if (pl->pl_ops->po_recalc != NULL) { count = pl->pl_ops->po_recalc(pl); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_RECALC_STAT, count); } recalc_interval_sec = pl->pl_recalc_time - get_seconds() + pl->pl_recalc_period; if (recalc_interval_sec <= 0) { /* Prevent too frequent recalculation. */ CDEBUG(D_DLMTRACE, "Negative interval(%ld), " "too short period(%ld)", recalc_interval_sec, pl->pl_recalc_period); recalc_interval_sec = 1; } return recalc_interval_sec; } /* * Pool shrink wrapper. Will call either client or server pool recalc callback * depending what pool pl is used. When nr == 0, just return the number of * freeable locks. Otherwise, return the number of canceled locks. */ int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask) { int cancel = 0; if (pl->pl_ops->po_shrink != NULL) { cancel = pl->pl_ops->po_shrink(pl, nr, gfp_mask); if (nr > 0) { lprocfs_counter_add(pl->pl_stats, LDLM_POOL_SHRINK_REQTD_STAT, nr); lprocfs_counter_add(pl->pl_stats, LDLM_POOL_SHRINK_FREED_STAT, cancel); CDEBUG(D_DLMTRACE, "%s: request to shrink %d locks, shrunk %d\n", pl->pl_name, nr, cancel); } } return cancel; } EXPORT_SYMBOL(ldlm_pool_shrink); /** * Pool setup wrapper. Will call either client or server pool recalc callback * depending what pool \a pl is used. * * Sets passed \a limit into pool \a pl. */ int ldlm_pool_setup(struct ldlm_pool *pl, int limit) { if (pl->pl_ops->po_setup != NULL) return pl->pl_ops->po_setup(pl, limit); return 0; } EXPORT_SYMBOL(ldlm_pool_setup); #if defined(CONFIG_PROC_FS) static int lprocfs_pool_state_seq_show(struct seq_file *m, void *unused) { int granted, grant_rate, cancel_rate, grant_step; int grant_speed, grant_plan, lvf; struct ldlm_pool *pl = m->private; __u64 slv, clv; __u32 limit; spin_lock(&pl->pl_lock); slv = pl->pl_server_lock_volume; clv = pl->pl_client_lock_volume; limit = ldlm_pool_get_limit(pl); grant_plan = pl->pl_grant_plan; granted = atomic_read(&pl->pl_granted); grant_rate = atomic_read(&pl->pl_grant_rate); cancel_rate = atomic_read(&pl->pl_cancel_rate); grant_speed = grant_rate - cancel_rate; lvf = atomic_read(&pl->pl_lock_volume_factor); grant_step = ldlm_pool_t2gsp(pl->pl_recalc_period); spin_unlock(&pl->pl_lock); seq_printf(m, "LDLM pool state (%s):\n" " SLV: %llu\n" " CLV: %llu\n" " LVF: %d\n", pl->pl_name, slv, clv, lvf); if (ns_is_server(ldlm_pl2ns(pl))) { seq_printf(m, " GSP: %d%%\n" " GP: %d\n", grant_step, grant_plan); } seq_printf(m, " GR: %d\n CR: %d\n GS: %d\n" " G: %d\n L: %d\n", grant_rate, cancel_rate, grant_speed, granted, limit); return 0; } LPROC_SEQ_FOPS_RO(lprocfs_pool_state); static int lprocfs_grant_speed_seq_show(struct seq_file *m, void *unused) { struct ldlm_pool *pl = m->private; int grant_speed; spin_lock(&pl->pl_lock); /* serialize with ldlm_pool_recalc */ grant_speed = atomic_read(&pl->pl_grant_rate) - atomic_read(&pl->pl_cancel_rate); spin_unlock(&pl->pl_lock); return lprocfs_rd_uint(m, &grant_speed); } LDLM_POOL_PROC_READER_SEQ_SHOW(grant_plan, int); LPROC_SEQ_FOPS_RO(lprocfs_grant_plan); LDLM_POOL_PROC_READER_SEQ_SHOW(recalc_period, int); LDLM_POOL_PROC_WRITER(recalc_period, int); static ssize_t lprocfs_recalc_period_seq_write(struct file *file, const char __user *buf, size_t len, loff_t *off) { struct seq_file *seq = file->private_data; return lprocfs_wr_recalc_period(file, buf, len, seq->private); } LPROC_SEQ_FOPS(lprocfs_recalc_period); LPROC_SEQ_FOPS_RO_TYPE(ldlm_pool, u64); LPROC_SEQ_FOPS_RO_TYPE(ldlm_pool, atomic); LPROC_SEQ_FOPS_RW_TYPE(ldlm_pool_rw, atomic); LPROC_SEQ_FOPS_RO(lprocfs_grant_speed); #define LDLM_POOL_ADD_VAR(name, var, ops) \ do { \ snprintf(var_name, MAX_STRING_SIZE, #name); \ pool_vars[0].data = var; \ pool_vars[0].fops = ops; \ lprocfs_add_vars(pl->pl_proc_dir, pool_vars, NULL);\ } while (0) static int ldlm_pool_proc_init(struct ldlm_pool *pl) { struct ldlm_namespace *ns = ldlm_pl2ns(pl); struct proc_dir_entry *parent_ns_proc; struct lprocfs_vars pool_vars[2]; char *var_name = NULL; int rc = 0; OBD_ALLOC(var_name, MAX_STRING_SIZE + 1); if (!var_name) return -ENOMEM; parent_ns_proc = ns->ns_proc_dir_entry; if (parent_ns_proc == NULL) { CERROR("%s: proc entry is not initialized\n", ldlm_ns_name(ns)); rc = -EINVAL; goto out_free_name; } pl->pl_proc_dir = lprocfs_register("pool", parent_ns_proc, NULL, NULL); if (IS_ERR(pl->pl_proc_dir)) { CERROR("LProcFS failed in ldlm-pool-init\n"); rc = PTR_ERR(pl->pl_proc_dir); pl->pl_proc_dir = NULL; goto out_free_name; } var_name[MAX_STRING_SIZE] = '\0'; memset(pool_vars, 0, sizeof(pool_vars)); pool_vars[0].name = var_name; LDLM_POOL_ADD_VAR("server_lock_volume", &pl->pl_server_lock_volume, &ldlm_pool_u64_fops); LDLM_POOL_ADD_VAR("limit", &pl->pl_limit, &ldlm_pool_rw_atomic_fops); LDLM_POOL_ADD_VAR("granted", &pl->pl_granted, &ldlm_pool_atomic_fops); LDLM_POOL_ADD_VAR("grant_speed", pl, &lprocfs_grant_speed_fops); LDLM_POOL_ADD_VAR("cancel_rate", &pl->pl_cancel_rate, &ldlm_pool_atomic_fops); LDLM_POOL_ADD_VAR("grant_rate", &pl->pl_grant_rate, &ldlm_pool_atomic_fops); LDLM_POOL_ADD_VAR("grant_plan", pl, &lprocfs_grant_plan_fops); LDLM_POOL_ADD_VAR("recalc_period", pl, &lprocfs_recalc_period_fops); LDLM_POOL_ADD_VAR("lock_volume_factor", &pl->pl_lock_volume_factor, &ldlm_pool_rw_atomic_fops); LDLM_POOL_ADD_VAR("state", pl, &lprocfs_pool_state_fops); pl->pl_stats = lprocfs_alloc_stats(LDLM_POOL_LAST_STAT - LDLM_POOL_FIRST_STAT, 0); if (!pl->pl_stats) { rc = -ENOMEM; goto out_free_name; } lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANTED_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "granted", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "grant", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_CANCEL_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "cancel", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_RATE_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "grant_rate", "locks/s"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_CANCEL_RATE_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "cancel_rate", "locks/s"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_PLAN_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "grant_plan", "locks/s"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SLV_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "slv", "slv"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SHRINK_REQTD_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "shrink_request", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SHRINK_FREED_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "shrink_freed", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_RECALC_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "recalc_freed", "locks"); lprocfs_counter_init(pl->pl_stats, LDLM_POOL_TIMING_STAT, LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV, "recalc_timing", "sec"); rc = lprocfs_register_stats(pl->pl_proc_dir, "stats", pl->pl_stats); out_free_name: OBD_FREE(var_name, MAX_STRING_SIZE + 1); return rc; } static void ldlm_pool_proc_fini(struct ldlm_pool *pl) { if (pl->pl_stats != NULL) { lprocfs_free_stats(&pl->pl_stats); pl->pl_stats = NULL; } if (pl->pl_proc_dir != NULL) { lprocfs_remove(&pl->pl_proc_dir); pl->pl_proc_dir = NULL; } } #else /* !CONFIG_PROC_FS */ static int ldlm_pool_proc_init(struct ldlm_pool *pl) { return 0; } static void ldlm_pool_proc_fini(struct ldlm_pool *pl) {} #endif /* CONFIG_PROC_FS */ int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns, int idx, ldlm_side_t client) { int rc; spin_lock_init(&pl->pl_lock); atomic_set(&pl->pl_granted, 0); pl->pl_recalc_time = get_seconds(); atomic_set(&pl->pl_lock_volume_factor, 1); atomic_set(&pl->pl_grant_rate, 0); atomic_set(&pl->pl_cancel_rate, 0); pl->pl_grant_plan = LDLM_POOL_GP(LDLM_POOL_HOST_L); snprintf(pl->pl_name, sizeof(pl->pl_name), "ldlm-pool-%s-%d", ldlm_ns_name(ns), idx); if (client == LDLM_NAMESPACE_SERVER) { pl->pl_ops = &ldlm_srv_pool_ops; ldlm_pool_set_limit(pl, LDLM_POOL_HOST_L); pl->pl_recalc_period = LDLM_POOL_SRV_DEF_RECALC_PERIOD; pl->pl_server_lock_volume = ldlm_pool_slv_max(LDLM_POOL_HOST_L); } else { ldlm_pool_set_limit(pl, 1); pl->pl_server_lock_volume = 0; pl->pl_ops = &ldlm_cli_pool_ops; pl->pl_recalc_period = LDLM_POOL_CLI_DEF_RECALC_PERIOD; } pl->pl_client_lock_volume = 0; rc = ldlm_pool_proc_init(pl); if (rc) return rc; CDEBUG(D_DLMTRACE, "Lock pool %s is initialized\n", pl->pl_name); return rc; } EXPORT_SYMBOL(ldlm_pool_init); void ldlm_pool_fini(struct ldlm_pool *pl) { ldlm_pool_proc_fini(pl); /* * Pool should not be used after this point. We can't free it here as * it lives in struct ldlm_namespace, but still interested in catching * any abnormal using cases. */ POISON(pl, 0x5a, sizeof(*pl)); } EXPORT_SYMBOL(ldlm_pool_fini); /** * Add new taken ldlm lock \a lock into pool \a pl accounting. */ void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock) { /* * FLOCK locks are special in a sense that they are almost never * cancelled, instead special kind of lock is used to drop them. * also there is no LRU for flock locks, so no point in tracking * them anyway. */ if (lock->l_resource->lr_type == LDLM_FLOCK) return; atomic_inc(&pl->pl_granted); atomic_inc(&pl->pl_grant_rate); lprocfs_counter_incr(pl->pl_stats, LDLM_POOL_GRANT_STAT); /* * Do not do pool recalc for client side as all locks which * potentially may be canceled has already been packed into * enqueue/cancel rpc. Also we do not want to run out of stack * with too long call paths. */ if (ns_is_server(ldlm_pl2ns(pl))) ldlm_pool_recalc(pl); } EXPORT_SYMBOL(ldlm_pool_add); /** * Remove ldlm lock \a lock from pool \a pl accounting. */ void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock) { /* * Filter out FLOCK locks. Read above comment in ldlm_pool_add(). */ if (lock->l_resource->lr_type == LDLM_FLOCK) return; LASSERT(atomic_read(&pl->pl_granted) > 0); atomic_dec(&pl->pl_granted); atomic_inc(&pl->pl_cancel_rate); lprocfs_counter_incr(pl->pl_stats, LDLM_POOL_CANCEL_STAT); if (ns_is_server(ldlm_pl2ns(pl))) ldlm_pool_recalc(pl); } EXPORT_SYMBOL(ldlm_pool_del); /** * Returns current \a pl SLV. * * \pre ->pl_lock is not locked. */ __u64 ldlm_pool_get_slv(struct ldlm_pool *pl) { __u64 slv; spin_lock(&pl->pl_lock); slv = pl->pl_server_lock_volume; spin_unlock(&pl->pl_lock); return slv; } EXPORT_SYMBOL(ldlm_pool_get_slv); /** * Sets passed \a slv to \a pl. * * \pre ->pl_lock is not locked. */ void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv) { spin_lock(&pl->pl_lock); pl->pl_server_lock_volume = slv; spin_unlock(&pl->pl_lock); } EXPORT_SYMBOL(ldlm_pool_set_slv); /** * Returns current \a pl CLV. * * \pre ->pl_lock is not locked. */ __u64 ldlm_pool_get_clv(struct ldlm_pool *pl) { __u64 slv; spin_lock(&pl->pl_lock); slv = pl->pl_client_lock_volume; spin_unlock(&pl->pl_lock); return slv; } EXPORT_SYMBOL(ldlm_pool_get_clv); /** * Sets passed \a clv to \a pl. * * \pre ->pl_lock is not locked. */ void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv) { spin_lock(&pl->pl_lock); pl->pl_client_lock_volume = clv; spin_unlock(&pl->pl_lock); } EXPORT_SYMBOL(ldlm_pool_set_clv); /** * Returns current \a pl limit. */ __u32 ldlm_pool_get_limit(struct ldlm_pool *pl) { return atomic_read(&pl->pl_limit); } EXPORT_SYMBOL(ldlm_pool_get_limit); /** * Sets passed \a limit to \a pl. */ void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit) { atomic_set(&pl->pl_limit, limit); } EXPORT_SYMBOL(ldlm_pool_set_limit); /** * Returns current LVF from \a pl. */ __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl) { return atomic_read(&pl->pl_lock_volume_factor); } EXPORT_SYMBOL(ldlm_pool_get_lvf); static int ldlm_pool_granted(struct ldlm_pool *pl) { return atomic_read(&pl->pl_granted); } static struct ptlrpc_thread *ldlm_pools_thread; static struct completion ldlm_pools_comp; /* * count locks from all namespaces (if possible). Returns number of * cached locks. */ static unsigned long ldlm_pools_count(ldlm_side_t client, gfp_t gfp_mask) { int total = 0, nr_ns; struct ldlm_namespace *ns; struct ldlm_namespace *ns_old = NULL; /* loop detection */ void *cookie; if (client == LDLM_NAMESPACE_CLIENT && !(gfp_mask & __GFP_FS)) return 0; CDEBUG(D_DLMTRACE, "Request to count %s locks from all pools\n", client == LDLM_NAMESPACE_CLIENT ? "client" : "server"); cookie = cl_env_reenter(); /* * Find out how many resources we may release. */ for (nr_ns = ldlm_namespace_nr_read(client); nr_ns > 0; nr_ns--) { mutex_lock(ldlm_namespace_lock(client)); if (list_empty(ldlm_namespace_list(client))) { mutex_unlock(ldlm_namespace_lock(client)); cl_env_reexit(cookie); return 0; } ns = ldlm_namespace_first_locked(client); if (ns == ns_old) { mutex_unlock(ldlm_namespace_lock(client)); break; } if (ldlm_ns_empty(ns)) { ldlm_namespace_move_to_inactive_locked(ns, client); mutex_unlock(ldlm_namespace_lock(client)); continue; } if (ns_old == NULL) ns_old = ns; ldlm_namespace_get(ns); ldlm_namespace_move_to_active_locked(ns, client); mutex_unlock(ldlm_namespace_lock(client)); total += ldlm_pool_shrink(&ns->ns_pool, 0, gfp_mask); ldlm_namespace_put(ns); } cl_env_reexit(cookie); return total; } static unsigned long ldlm_pools_scan(ldlm_side_t client, int nr, gfp_t gfp_mask) { unsigned long freed = 0; int tmp, nr_ns; struct ldlm_namespace *ns; void *cookie; if (client == LDLM_NAMESPACE_CLIENT && !(gfp_mask & __GFP_FS)) return -1; cookie = cl_env_reenter(); /* * Shrink at least ldlm_namespace_nr_read(client) namespaces. */ for (tmp = nr_ns = ldlm_namespace_nr_read(client); tmp > 0; tmp--) { int cancel, nr_locks; /* * Do not call shrink under ldlm_namespace_lock(client) */ mutex_lock(ldlm_namespace_lock(client)); if (list_empty(ldlm_namespace_list(client))) { mutex_unlock(ldlm_namespace_lock(client)); break; } ns = ldlm_namespace_first_locked(client); ldlm_namespace_get(ns); ldlm_namespace_move_to_active_locked(ns, client); mutex_unlock(ldlm_namespace_lock(client)); nr_locks = ldlm_pool_granted(&ns->ns_pool); /* * We use to shrink propotionally but with new shrinker API, * we lost the total number of freeable locks. */ cancel = 1 + min_t(int, nr_locks, nr / nr_ns); freed += ldlm_pool_shrink(&ns->ns_pool, cancel, gfp_mask); ldlm_namespace_put(ns); } cl_env_reexit(cookie); /* * we only decrease the SLV in server pools shrinker, return * SHRINK_STOP to kernel to avoid needless loop. LU-1128 */ return (client == LDLM_NAMESPACE_SERVER) ? SHRINK_STOP : freed; } static unsigned long ldlm_pools_srv_count(struct shrinker *s, struct shrink_control *sc) { return ldlm_pools_count(LDLM_NAMESPACE_SERVER, sc->gfp_mask); } static unsigned long ldlm_pools_srv_scan(struct shrinker *s, struct shrink_control *sc) { return ldlm_pools_scan(LDLM_NAMESPACE_SERVER, sc->nr_to_scan, sc->gfp_mask); } static unsigned long ldlm_pools_cli_count(struct shrinker *s, struct shrink_control *sc) { return ldlm_pools_count(LDLM_NAMESPACE_CLIENT, sc->gfp_mask); } static unsigned long ldlm_pools_cli_scan(struct shrinker *s, struct shrink_control *sc) { return ldlm_pools_scan(LDLM_NAMESPACE_CLIENT, sc->nr_to_scan, sc->gfp_mask); } int ldlm_pools_recalc(ldlm_side_t client) { __u32 nr_l = 0, nr_p = 0, l; struct ldlm_namespace *ns; struct ldlm_namespace *ns_old = NULL; int nr, equal = 0; int time = 50; /* seconds of sleep if no active namespaces */ /* * No need to setup pool limit for client pools. */ if (client == LDLM_NAMESPACE_SERVER) { /* * Check all modest namespaces first. */ mutex_lock(ldlm_namespace_lock(client)); list_for_each_entry(ns, ldlm_namespace_list(client), ns_list_chain) { if (ns->ns_appetite != LDLM_NAMESPACE_MODEST) continue; l = ldlm_pool_granted(&ns->ns_pool); if (l == 0) l = 1; /* * Set the modest pools limit equal to their avg granted * locks + ~6%. */ l += dru(l, LDLM_POOLS_MODEST_MARGIN_SHIFT, 0); ldlm_pool_setup(&ns->ns_pool, l); nr_l += l; nr_p++; } /* * Make sure that modest namespaces did not eat more that 2/3 * of limit. */ if (nr_l >= 2 * (LDLM_POOL_HOST_L / 3)) { CWARN("\"Modest\" pools eat out 2/3 of server locks limit (%d of %lu). This means that you have too many clients for this amount of server RAM. Upgrade server!\n", nr_l, LDLM_POOL_HOST_L); equal = 1; } /* * The rest is given to greedy namespaces. */ list_for_each_entry(ns, ldlm_namespace_list(client), ns_list_chain) { if (!equal && ns->ns_appetite != LDLM_NAMESPACE_GREEDY) continue; if (equal) { /* * In the case 2/3 locks are eaten out by * modest pools, we re-setup equal limit * for _all_ pools. */ l = LDLM_POOL_HOST_L / ldlm_namespace_nr_read(client); } else { /* * All the rest of greedy pools will have * all locks in equal parts. */ l = (LDLM_POOL_HOST_L - nr_l) / (ldlm_namespace_nr_read(client) - nr_p); } ldlm_pool_setup(&ns->ns_pool, l); } mutex_unlock(ldlm_namespace_lock(client)); } /* * Recalc at least ldlm_namespace_nr_read(client) namespaces. */ for (nr = ldlm_namespace_nr_read(client); nr > 0; nr--) { int skip; /* * Lock the list, get first @ns in the list, getref, move it * to the tail, unlock and call pool recalc. This way we avoid * calling recalc under @ns lock what is really good as we get * rid of potential deadlock on client nodes when canceling * locks synchronously. */ mutex_lock(ldlm_namespace_lock(client)); if (list_empty(ldlm_namespace_list(client))) { mutex_unlock(ldlm_namespace_lock(client)); break; } ns = ldlm_namespace_first_locked(client); if (ns_old == ns) { /* Full pass complete */ mutex_unlock(ldlm_namespace_lock(client)); break; } /* We got an empty namespace, need to move it back to inactive * list. * The race with parallel resource creation is fine: * - If they do namespace_get before our check, we fail the * check and they move this item to the end of the list anyway * - If we do the check and then they do namespace_get, then * we move the namespace to inactive and they will move * it back to active (synchronised by the lock, so no clash * there). */ if (ldlm_ns_empty(ns)) { ldlm_namespace_move_to_inactive_locked(ns, client); mutex_unlock(ldlm_namespace_lock(client)); continue; } if (ns_old == NULL) ns_old = ns; spin_lock(&ns->ns_lock); /* * skip ns which is being freed, and we don't want to increase * its refcount again, not even temporarily. bz21519 & LU-499. */ if (ns->ns_stopping) { skip = 1; } else { skip = 0; ldlm_namespace_get(ns); } spin_unlock(&ns->ns_lock); ldlm_namespace_move_to_active_locked(ns, client); mutex_unlock(ldlm_namespace_lock(client)); /* * After setup is done - recalc the pool. */ if (!skip) { int ttime = ldlm_pool_recalc(&ns->ns_pool); if (ttime < time) time = ttime; ldlm_namespace_put(ns); } } return time; } EXPORT_SYMBOL(ldlm_pools_recalc); static int ldlm_pools_thread_main(void *arg) { struct ptlrpc_thread *thread = (struct ptlrpc_thread *)arg; int s_time, c_time; thread_set_flags(thread, SVC_RUNNING); wake_up(&thread->t_ctl_waitq); CDEBUG(D_DLMTRACE, "%s: pool thread starting, process %d\n", "ldlm_poold", current_pid()); while (1) { struct l_wait_info lwi; /* * Recal all pools on this tick. */ s_time = ldlm_pools_recalc(LDLM_NAMESPACE_SERVER); c_time = ldlm_pools_recalc(LDLM_NAMESPACE_CLIENT); /* * Wait until the next check time, or until we're * stopped. */ lwi = LWI_TIMEOUT(cfs_time_seconds(min(s_time, c_time)), NULL, NULL); l_wait_event(thread->t_ctl_waitq, thread_is_stopping(thread) || thread_is_event(thread), &lwi); if (thread_test_and_clear_flags(thread, SVC_STOPPING)) break; else thread_test_and_clear_flags(thread, SVC_EVENT); } thread_set_flags(thread, SVC_STOPPED); wake_up(&thread->t_ctl_waitq); CDEBUG(D_DLMTRACE, "%s: pool thread exiting, process %d\n", "ldlm_poold", current_pid()); complete_and_exit(&ldlm_pools_comp, 0); } static int ldlm_pools_thread_start(void) { struct l_wait_info lwi = { 0 }; struct task_struct *task; if (ldlm_pools_thread != NULL) return -EALREADY; OBD_ALLOC_PTR(ldlm_pools_thread); if (ldlm_pools_thread == NULL) return -ENOMEM; init_completion(&ldlm_pools_comp); init_waitqueue_head(&ldlm_pools_thread->t_ctl_waitq); task = kthread_run(ldlm_pools_thread_main, ldlm_pools_thread, "ldlm_poold"); if (IS_ERR(task)) { CERROR("Can't start pool thread, error %ld\n", PTR_ERR(task)); OBD_FREE(ldlm_pools_thread, sizeof(*ldlm_pools_thread)); ldlm_pools_thread = NULL; return PTR_ERR(task); } l_wait_event(ldlm_pools_thread->t_ctl_waitq, thread_is_running(ldlm_pools_thread), &lwi); return 0; } static void ldlm_pools_thread_stop(void) { if (ldlm_pools_thread == NULL) return; thread_set_flags(ldlm_pools_thread, SVC_STOPPING); wake_up(&ldlm_pools_thread->t_ctl_waitq); /* * Make sure that pools thread is finished before freeing @thread. * This fixes possible race and oops due to accessing freed memory * in pools thread. */ wait_for_completion(&ldlm_pools_comp); OBD_FREE_PTR(ldlm_pools_thread); ldlm_pools_thread = NULL; } static struct shrinker ldlm_pools_srv_shrinker = { .count_objects = ldlm_pools_srv_count, .scan_objects = ldlm_pools_srv_scan, .seeks = DEFAULT_SEEKS, }; static struct shrinker ldlm_pools_cli_shrinker = { .count_objects = ldlm_pools_cli_count, .scan_objects = ldlm_pools_cli_scan, .seeks = DEFAULT_SEEKS, }; int ldlm_pools_init(void) { int rc; rc = ldlm_pools_thread_start(); if (rc == 0) { register_shrinker(&ldlm_pools_srv_shrinker); register_shrinker(&ldlm_pools_cli_shrinker); } return rc; } EXPORT_SYMBOL(ldlm_pools_init); void ldlm_pools_fini(void) { unregister_shrinker(&ldlm_pools_srv_shrinker); unregister_shrinker(&ldlm_pools_cli_shrinker); ldlm_pools_thread_stop(); } EXPORT_SYMBOL(ldlm_pools_fini);