X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_lproc.c;h=7f281a6562f242c5ae6e7c50e274bdc8b566f74f;hp=a95bbffecf475b1c3e8a9e4867f255abc34c025f;hb=f5ecf580adc319fd9abe41bd79389b50015d5c13;hpb=4b7cbec396fcd7afb81d601a2facb70ee8c7ad28 diff --git a/lustre/mdt/mdt_lproc.c b/lustre/mdt/mdt_lproc.c index a95bbff..7f281a6 100644 --- a/lustre/mdt/mdt_lproc.c +++ b/lustre/mdt/mdt_lproc.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/ @@ -39,9 +39,6 @@ * Author: Fan Yong */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include @@ -74,6 +71,160 @@ enum { static const char *mdt_proc_names[LPROC_MDT_NR] = { }; +/** + * The rename stats output would be YAML formats, like + * rename_stats: + * - snapshot_time: 1234567890.123456 + * - same_dir: + * 4kB: { samples: 1230, pct: 33, cum_pct: 45 } + * 8kB: { samples: 1242, pct: 33, cum_pct: 78 } + * 16kB: { samples: 132, pct: 3, cum_pct: 81 } + * - crossdir_src: + * 4kB: { samples: 123, pct: 33, cum_pct: 45 } + * 8kB: { samples: 124, pct: 33, cum_pct: 78 } + * 16kB: { samples: 12, pct: 3, cum_pct: 81 } + * - crossdir_tgt: + * 4kB: { samples: 123, pct: 33, cum_pct: 45 } + * 8kB: { samples: 124, pct: 33, cum_pct: 78 } + * 16kB: { samples: 12, pct: 3, cum_pct: 81 } + **/ + +#define pct(a, b) (b ? a * 100 / b : 0) + +static void display_rename_stats(struct seq_file *seq, char *name, + struct obd_histogram *hist) +{ + unsigned long tot, t, cum = 0; + int i; + + tot = lprocfs_oh_sum(hist); + if (tot > 0) + seq_printf(seq, "- %-15s\n", name); + /* dir size start from 4K, start i from 10(2^10) here */ + for (i = 0; i < OBD_HIST_MAX; i++) { + t = hist->oh_buckets[i]; + cum += t; + if (cum == 0) + continue; + + if (i < 10) + seq_printf(seq, "%6s%d%s", " ", 1<< i, "bytes:"); + else if (i < 20) + seq_printf(seq, "%6s%d%s", " ", 1<<(i-10), "KB:"); + else + seq_printf(seq, "%6s%d%s", " ", 1<<(i-20), "MB:"); + + seq_printf(seq, " { sample: %3lu, pct: %3lu, cum_pct: %3lu }\n", + t, pct(t, tot), pct(cum, tot)); + + if (cum == tot) + break; + } +} + +static void rename_stats_show(struct seq_file *seq, + struct rename_stats *rename_stats) +{ + struct timeval now; + + /* this sampling races with updates */ + do_gettimeofday(&now); + seq_printf(seq, "rename_stats:\n"); + seq_printf(seq, "- %-15s %lu.%lu\n", "snapshot_time:", + now.tv_sec, now.tv_usec); + + display_rename_stats(seq, "same_dir", + &rename_stats->hist[RENAME_SAMEDIR_SIZE]); + display_rename_stats(seq, "crossdir_src", + &rename_stats->hist[RENAME_CROSSDIR_SRC_SIZE]); + display_rename_stats(seq, "crossdir_tgt", + &rename_stats->hist[RENAME_CROSSDIR_TGT_SIZE]); +} + +#undef pct + +static int mdt_rename_stats_seq_show(struct seq_file *seq, void *v) +{ + struct mdt_device *mdt = seq->private; + + rename_stats_show(seq, &mdt->mdt_rename_stats); + + return 0; +} + +static ssize_t mdt_rename_stats_seq_write(struct file *file, const char *buf, + size_t len, loff_t *off) +{ + struct seq_file *seq = file->private_data; + struct mdt_device *mdt = seq->private; + int i; + + for (i = 0; i < RENAME_LAST; i++) + lprocfs_oh_clear(&mdt->mdt_rename_stats.hist[i]); + + return len; +} + +LPROC_SEQ_FOPS(mdt_rename_stats); + +static int lproc_mdt_attach_rename_seqstat(struct mdt_device *mdt) +{ + struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev; + struct obd_device *obd = ld->ld_obd; + int i; + + for (i = 0; i < RENAME_LAST; i++) + spin_lock_init(&mdt->mdt_rename_stats.hist[i].oh_lock); + + return lprocfs_obd_seq_create(obd, "rename_stats", 0444, + &mdt_rename_stats_fops, mdt); +} + +void mdt_rename_counter_tally(struct mdt_thread_info *info, + struct mdt_device *mdt, + struct ptlrpc_request *req, + struct mdt_object *src, + struct mdt_object *tgt) +{ + struct md_attr *ma = &info->mti_attr; + struct rename_stats *rstats = &mdt->mdt_rename_stats; + int rc; + + ma->ma_need = MA_INODE; + ma->ma_valid = 0; + rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma); + if (rc) { + CERROR("%s: "DFID" attr_get, rc = %d\n", + req->rq_export->exp_obd->obd_name, + PFID(mdt_object_fid(src)), rc); + return; + } + + if (src == tgt) { + mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME); + lprocfs_oh_tally_log2(&rstats->hist[RENAME_SAMEDIR_SIZE], + (unsigned int)ma->ma_attr.la_size); + return; + } + + mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME); + lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_SRC_SIZE], + (unsigned int)ma->ma_attr.la_size); + + ma->ma_need = MA_INODE; + ma->ma_valid = 0; + rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma); + if (rc) { + CERROR("%s: "DFID" attr_get, rc = %d\n", + req->rq_export->exp_obd->obd_name, + PFID(mdt_object_fid(tgt)), rc); + return; + } + + lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_TGT_SIZE], + (unsigned int)ma->ma_attr.la_size); +} + int mdt_procfs_init(struct mdt_device *mdt, const char *name) { struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev; @@ -112,8 +263,17 @@ int mdt_procfs_init(struct mdt_device *mdt, const char *name) "clear", lprocfs_nid_stats_clear_read, lprocfs_nid_stats_clear_write, obd, NULL); rc = lprocfs_alloc_md_stats(obd, LPROC_MDT_LAST); - if (rc == 0) - mdt_stats_counter_init(obd->md_stats); + if (rc) + return rc; + mdt_stats_counter_init(obd->md_stats); + + rc = lprocfs_job_stats_init(obd, LPROC_MDT_LAST, + mdt_stats_counter_init); + + rc = lproc_mdt_attach_rename_seqstat(mdt); + if (rc) + CERROR("%s: MDT can not create rename stats rc = %d\n", + obd->obd_name, rc); RETURN(rc); } @@ -123,18 +283,22 @@ int mdt_procfs_fini(struct mdt_device *mdt) struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev; struct obd_device *obd = ld->ld_obd; - if (mdt->mdt_proc_entry) { - lu_time_fini(&ld->ld_site->ls_time_stats); - lu_time_fini(&mdt->mdt_stats); - mdt->mdt_proc_entry = NULL; - } + lprocfs_job_stats_fini(obd); + if (obd->obd_proc_exports_entry) { lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry); obd->obd_proc_exports_entry = NULL; } + lprocfs_free_per_client_stats(obd); + lprocfs_obd_cleanup(obd); ptlrpc_lprocfs_unregister_obd(obd); + if (mdt->mdt_proc_entry) { + lu_time_fini(&ld->ld_site->ls_time_stats); + lu_time_fini(&mdt->mdt_stats); + mdt->mdt_proc_entry = NULL; + } lprocfs_free_md_stats(obd); - lprocfs_obd_cleanup(obd); + lprocfs_free_obd_stats(obd); RETURN(0); } @@ -212,11 +376,11 @@ static int lprocfs_rd_identity_upcall(char *page, char **start, off_t off, struct upcall_cache *hash = mdt->mdt_identity_cache; int len; - *eof = 1; - cfs_read_lock(&hash->uc_upcall_rwlock); - len = snprintf(page, count, "%s\n", hash->uc_upcall); - cfs_read_unlock(&hash->uc_upcall_rwlock); - return len; + *eof = 1; + read_lock(&hash->uc_upcall_rwlock); + len = snprintf(page, count, "%s\n", hash->uc_upcall); + read_unlock(&hash->uc_upcall_rwlock); + return len; } static int lprocfs_wr_identity_upcall(struct file *file, const char *buffer, @@ -239,9 +403,9 @@ static int lprocfs_wr_identity_upcall(struct file *file, const char *buffer, GOTO(failed, rc = -EFAULT); /* Remove any extraneous bits from the upcall (e.g. linefeeds) */ - cfs_write_lock(&hash->uc_upcall_rwlock); - sscanf(kernbuf, "%s", hash->uc_upcall); - cfs_write_unlock(&hash->uc_upcall_rwlock); + write_lock(&hash->uc_upcall_rwlock); + sscanf(kernbuf, "%s", hash->uc_upcall); + write_unlock(&hash->uc_upcall_rwlock); if (strcmp(hash->uc_name, obd->obd_name) != 0) CWARN("%s: write to upcall name %s\n", @@ -457,21 +621,44 @@ static int lprocfs_wr_ck_timeout(struct file *file, const char *buffer, return count; } +#define BUFLEN (UUID_MAX + 4) + static int lprocfs_mdt_wr_evict_client(struct file *file, const char *buffer, unsigned long count, void *data) { - char tmpbuf[sizeof(struct obd_uuid)]; + char *kbuf; + char *tmpbuf; - sscanf(buffer, "%40s", tmpbuf); + OBD_ALLOC(kbuf, BUFLEN); + if (kbuf == NULL) + return -ENOMEM; - if (strncmp(tmpbuf, "nid:", 4) != 0) - return lprocfs_wr_evict_client(file, buffer, count, data); + /* + * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1 + * bytes into kbuf, to ensure that the string is NUL-terminated. + * UUID_MAX should include a trailing NUL already. + */ + if (cfs_copy_from_user(kbuf, buffer, + min_t(unsigned long, BUFLEN - 1, count))) { + count = -EFAULT; + goto out; + } + tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count)); + + if (strncmp(tmpbuf, "nid:", 4) != 0) { + count = lprocfs_wr_evict_client(file, buffer, count, data); + goto out; + } CERROR("NOT implement evict client by nid %s\n", tmpbuf); +out: + OBD_FREE(kbuf, BUFLEN); return count; } +#undef BUFLEN + static int lprocfs_rd_sec_level(char *page, char **start, off_t off, int count, int *eof, void *data) { @@ -533,7 +720,6 @@ static int lprocfs_rd_root_squash(char *page, char **start, off_t off, { struct obd_device *obd = data; struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev); - ENTRY; return snprintf(page, count, "%u:%u\n", mdt->mdt_squash_uid, mdt->mdt_squash_gid); @@ -646,7 +832,7 @@ static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer, if (!strcmp(kernbuf, "NONE") || !strcmp(kernbuf, "clear")) { /* empty string is special case */ - cfs_down_write(&mdt->mdt_squash_sem); + down_write(&mdt->mdt_squash_sem); if (!cfs_list_empty(&mdt->mdt_nosquash_nids)) { cfs_free_nidlist(&mdt->mdt_nosquash_nids); OBD_FREE(mdt->mdt_nosquash_str, @@ -654,7 +840,7 @@ static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer, mdt->mdt_nosquash_str = NULL; mdt->mdt_nosquash_strlen = 0; } - cfs_up_write(&mdt->mdt_squash_sem); + up_write(&mdt->mdt_squash_sem); LCONSOLE_INFO("%s: nosquash_nids is cleared\n", obd->obd_name); OBD_FREE(kernbuf, count + 1); @@ -667,7 +853,7 @@ static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer, GOTO(failed, rc = -EINVAL); } - cfs_down_write(&mdt->mdt_squash_sem); + down_write(&mdt->mdt_squash_sem); if (!cfs_list_empty(&mdt->mdt_nosquash_nids)) { cfs_free_nidlist(&mdt->mdt_nosquash_nids); OBD_FREE(mdt->mdt_nosquash_str, mdt->mdt_nosquash_strlen); @@ -678,7 +864,7 @@ static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer, LCONSOLE_INFO("%s: nosquash_nids is set to %s\n", obd->obd_name, kernbuf); - cfs_up_write(&mdt->mdt_squash_sem); + up_write(&mdt->mdt_squash_sem); RETURN(count); failed: @@ -755,21 +941,45 @@ static int lprocfs_mdt_wr_mdc(struct file *file, const char *buffer, { struct obd_device *obd = data; struct obd_export *exp = NULL; - struct obd_uuid uuid; - char tmpbuf[sizeof(struct obd_uuid)]; + struct obd_uuid *uuid; + char *kbuf; + char *tmpbuf; - sscanf(buffer, "%40s", tmpbuf); + OBD_ALLOC(kbuf, UUID_MAX); + if (kbuf == NULL) + return -ENOMEM; - obd_str2uuid(&uuid, tmpbuf); - exp = cfs_hash_lookup(obd->obd_uuid_hash, &uuid); + /* + * OBD_ALLOC() will zero kbuf, but we only copy UUID_MAX - 1 + * bytes into kbuf, to ensure that the string is NUL-terminated. + * UUID_MAX should include a trailing NUL already. + */ + if (cfs_copy_from_user(kbuf, buffer, + min_t(unsigned long, UUID_MAX - 1, count))) { + count = -EFAULT; + goto out; + } + tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, UUID_MAX - 1, count)); + + OBD_ALLOC(uuid, UUID_MAX); + if (uuid == NULL) { + count = -ENOMEM; + goto out; + } + + obd_str2uuid(uuid, tmpbuf); + exp = cfs_hash_lookup(obd->obd_uuid_hash, uuid); if (exp == NULL) { CERROR("%s: no export %s found\n", - obd->obd_name, obd_uuid2str(&uuid)); + obd->obd_name, obd_uuid2str(uuid)); } else { mdt_hsm_copytool_send(exp); class_export_put(exp); } + OBD_FREE(uuid, UUID_MAX); +out: + OBD_FREE(kbuf, UUID_MAX); return count; } @@ -805,6 +1015,11 @@ static struct lprocfs_vars lprocfs_mdt_obd_vars[] = { { "som", lprocfs_rd_mdt_som, lprocfs_wr_mdt_som, 0 }, { "mdccomm", 0, lprocfs_mdt_wr_mdc, 0 }, + { "instance", lprocfs_target_rd_instance, 0 }, + { "ir_factor", lprocfs_obd_rd_ir_factor, + lprocfs_obd_wr_ir_factor, 0 }, + { "job_cleanup_interval", lprocfs_rd_job_interval, + lprocfs_wr_job_interval, 0 }, { 0 } }; @@ -815,17 +1030,33 @@ static struct lprocfs_vars lprocfs_mdt_module_vars[] = { void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars) { - lvars->module_vars = lprocfs_mdt_module_vars; - lvars->obd_vars = lprocfs_mdt_obd_vars; + lvars->module_vars = lprocfs_mdt_module_vars; + lvars->obd_vars = lprocfs_mdt_obd_vars; } -void mdt_counter_incr(struct obd_export *exp, int opcode) -{ - if (exp->exp_obd && exp->exp_obd->md_stats) - lprocfs_counter_incr(exp->exp_obd->md_stats, opcode); - if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL) - lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode); +struct lprocfs_vars lprocfs_mds_obd_vars[] = { + { "uuid", lprocfs_rd_uuid, 0, 0 }, + { 0 } +}; +struct lprocfs_vars lprocfs_mds_module_vars[] = { + { "num_refs", lprocfs_rd_numrefs, 0, 0 }, + { 0 } +}; + +void mdt_counter_incr(struct ptlrpc_request *req, int opcode) +{ + struct obd_export *exp = req->rq_export; + + if (exp->exp_obd && exp->exp_obd->md_stats) + lprocfs_counter_incr(exp->exp_obd->md_stats, opcode); + if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL) + lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode); + if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash && + (exp->exp_connect_flags & OBD_CONNECT_JOBSTATS)) + lprocfs_job_stats_log(exp->exp_obd, + lustre_msg_get_jobid(req->rq_reqmsg), + opcode, 1); } void mdt_stats_counter_init(struct lprocfs_stats *stats) @@ -844,4 +1075,8 @@ void mdt_stats_counter_init(struct lprocfs_stats *stats) lprocfs_counter_init(stats, LPROC_MDT_SETXATTR, 0, "setxattr", "reqs"); lprocfs_counter_init(stats, LPROC_MDT_STATFS, 0, "statfs", "reqs"); lprocfs_counter_init(stats, LPROC_MDT_SYNC, 0, "sync", "reqs"); + lprocfs_counter_init(stats, LPROC_MDT_SAMEDIR_RENAME, 0, + "samedir_rename", "reqs"); + lprocfs_counter_init(stats, LPROC_MDT_CROSSDIR_RENAME, 0, + "crossdir_rename", "reqs"); }