X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Flproc_ofd.c;h=c8f6d5d445cd6872a798de0d894b214f87b12337;hp=e11a81b6a50e28e50d7030a192d894da0cedd1b6;hb=8701b62863eb71a67b80da873726903920d1bada;hpb=a926e593b0c769d20aef191e2b0b6b1d881b6637 diff --git a/lustre/ofd/lproc_ofd.c b/lustre/ofd/lproc_ofd.c index e11a81b..c8f6d5d 100644 --- a/lustre/ofd/lproc_ofd.c +++ b/lustre/ofd/lproc_ofd.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 */ @@ -27,13 +23,20 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/ofd/lproc_ofd.c + * + * This file provides functions of procfs interface for OBD Filter Device (OFD). + * + * Author: Andreas Dilger + * Author: Mikhail Pershin + * Author: Johann Lombardi + * Author: Fan Yong */ #define DEBUG_SUBSYSTEM S_CLASS @@ -41,134 +44,99 @@ #include #include #include +#include #include "ofd_internal.h" -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS -static int lprocfs_ofd_rd_seqs(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show number of FID allocation sequences. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t seqs_allocated_show(struct kobject *kobj, struct attribute *attr, + char *buf) { - struct obd_device *obd = (struct obd_device *)data; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, "%u\n", ofd->ofd_seq_count); -} - -static int lprocfs_ofd_rd_tot_dirty(char *page, char **start, off_t off, - int count, int *eof, void *data) -{ - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd; - - LASSERT(obd != NULL); - ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, LPU64"\n", ofd->ofd_tot_dirty); + return sprintf(buf, "%u\n", ofd->ofd_seq_count); } +LUSTRE_RO_ATTR(seqs_allocated); -static int lprocfs_ofd_rd_tot_granted(char *page, char **start, off_t off, - int count, int *eof, void *data) -{ - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd; - - LASSERT(obd != NULL); - ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, LPU64"\n", ofd->ofd_tot_granted); -} - -static int lprocfs_ofd_rd_tot_pending(char *page, char **start, off_t off, - int count, int *eof, void *data) -{ - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd; - - LASSERT(obd != NULL); - ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, LPU64"\n", ofd->ofd_tot_pending); -} - -static int lprocfs_ofd_rd_grant_precreate(char *page, char **start, off_t off, - int count, int *eof, void *data) -{ - struct obd_device *obd = (struct obd_device *)data; - - LASSERT(obd != NULL); - *eof = 1; - return snprintf(page, count, "%ld\n", - obd->obd_self_export->exp_filter_data.fed_grant); -} - -static int lprocfs_ofd_rd_grant_ratio(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show total number of grants for precreate. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t grant_precreate_show(struct kobject *kobj, + struct attribute *attr, + char *buf) { - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); - LASSERT(obd != NULL); - ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, "%d%%\n", - (int) ofd_grant_reserved(ofd, 100)); + return sprintf(buf, "%ld\n", + obd->obd_self_export->exp_target_data.ted_grant); } +LUSTRE_RO_ATTR(grant_precreate); -static int lprocfs_ofd_wr_grant_ratio(struct file *file, const char *buffer, - unsigned long count, void *data) -{ - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; - int rc; - - rc = lprocfs_write_helper(buffer, count, &val); - if (rc) - return rc; - - if (val > 100 || val < 0) - return -EINVAL; - - if (val == 0) - CWARN("%s: disabling grant error margin\n", obd->obd_name); - if (val > 50) - CWARN("%s: setting grant error margin >50%%, be warned that " - "a huge part of the free space is now reserved for " - "grants\n", obd->obd_name); - - spin_lock(&ofd->ofd_grant_lock); - ofd->ofd_grant_ratio = ofd_grant_ratio_conv(val); - spin_unlock(&ofd->ofd_grant_lock); - return count; -} - -static int lprocfs_ofd_rd_precreate_batch(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show number of precreates allowed in a single transaction. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t precreate_batch_show(struct kobject *kobj, + struct attribute *attr, + char *buf) { - struct obd_device *obd = (struct obd_device *)data; - struct ofd_device *ofd; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - LASSERT(obd != NULL); - ofd = ofd_dev(obd->obd_lu_dev); - *eof = 1; - return snprintf(page, count, "%d\n", ofd->ofd_precreate_batch); + return sprintf(buf, "%d\n", ofd->ofd_precreate_batch); } -static int lprocfs_ofd_wr_precreate_batch(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Change number of precreates allowed in a single transaction. + * + * \param[in] file proc file + * \param[in] buffer string which represents maximum number + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t precreate_batch_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = (struct obd_device *)data; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; + unsigned int val; int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = kstrtouint(buffer, 0, &val); if (rc) return rc; - if (val < 1) + if (val < 1 || val > 65536) return -EINVAL; spin_lock(&ofd->ofd_batch_lock); @@ -176,14 +144,22 @@ static int lprocfs_ofd_wr_precreate_batch(struct file *file, const char *buffer, spin_unlock(&ofd->ofd_batch_lock); return count; } +LUSTRE_RW_ATTR(precreate_batch); -static int lprocfs_ofd_rd_last_id(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show the last used ID for each FID sequence used by OFD. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static int ofd_last_id_seq_show(struct seq_file *m, void *data) { - struct obd_device *obd = data; + struct obd_device *obd = m->private; struct ofd_device *ofd; struct ofd_seq *oseq = NULL; - int retval = 0, rc; if (obd == NULL) return 0; @@ -191,238 +167,389 @@ static int lprocfs_ofd_rd_last_id(char *page, char **start, off_t off, ofd = ofd_dev(obd->obd_lu_dev); read_lock(&ofd->ofd_seq_list_lock); - cfs_list_for_each_entry(oseq, &ofd->ofd_seq_list, os_list) { + list_for_each_entry(oseq, &ofd->ofd_seq_list, os_list) { __u64 seq; seq = ostid_seq(&oseq->os_oi) == 0 ? fid_idif_seq(ostid_id(&oseq->os_oi), ofd->ofd_lut.lut_lsd.lsd_osd_index) : ostid_seq(&oseq->os_oi); - rc = snprintf(page, count, DOSTID"\n", seq, - ostid_id(&oseq->os_oi)); - if (rc < 0) { - retval = rc; - break; - } - page += rc; - count -= rc; - retval += rc; + seq_printf(m, DOSTID"\n", seq, ostid_id(&oseq->os_oi)); } read_unlock(&ofd->ofd_seq_list_lock); - return retval; + return 0; } -int lprocfs_ofd_rd_fmd_max_num(char *page, char **start, off_t off, - int count, int *eof, void *data) +LPROC_SEQ_FOPS_RO(ofd_last_id); + +/** + * Show maximum number of Filter Modification Data (FMD) maintained by OFD. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t client_cache_count_show(struct kobject *kobj, + struct attribute *attr, + char *buf) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - rc = snprintf(page, count, "%u\n", ofd->ofd_fmd_max_num); - return rc; + return sprintf(buf, "%u\n", ofd->ofd_fmd_max_num); } -int lprocfs_ofd_wr_fmd_max_num(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Change number of FMDs maintained by OFD. + * + * This defines how large the list of FMDs can be. + * + * \param[in] file proc file + * \param[in] buffer string which represents maximum number + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t client_cache_count_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + int val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = kstrtoint(buffer, 0, &val); if (rc) return rc; - if (val > 65536 || val < 1) + if (val < 1 || val > 65536) return -EINVAL; ofd->ofd_fmd_max_num = val; return count; } +LUSTRE_RW_ATTR(client_cache_count); -int lprocfs_ofd_rd_fmd_max_age(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show the maximum age of FMD data in seconds. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t client_cache_seconds_show(struct kobject *kobj, + struct attribute *attr, + char *buf) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - rc = snprintf(page, count, "%ld\n", ofd->ofd_fmd_max_age / HZ); - return rc; + return sprintf(buf, "%lld\n", ofd->ofd_fmd_max_age); } -int lprocfs_ofd_wr_fmd_max_age(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Set the maximum age of FMD data in seconds. + * + * This defines how long FMD data stays in the FMD list. + * + * \param[in] file proc file + * \param[in] buffer string which represents maximum number + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t client_cache_seconds_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + time64_t val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = kstrtoll(buffer, 0, &val); if (rc) return rc; - if (val > 65536 || val < 1) + if (val < 1 || val > 65536) /* ~ 18 hour max */ return -EINVAL; - ofd->ofd_fmd_max_age = val * HZ; + ofd->ofd_fmd_max_age = val; return count; } +LUSTRE_RW_ATTR(client_cache_seconds); -static int lprocfs_ofd_rd_capa(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show if the OFD is in degraded mode. + * + * Degraded means OFD has a failed drive or is undergoing RAID rebuild. + * The MDS will try to avoid using this OST for new object allocations + * to reduce the impact to global IO performance when clients writing to + * this OST are slowed down. It also reduces the contention on the OST + * RAID device, allowing it to rebuild more quickly. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t degraded_show(struct kobject *kobj, struct attribute *attr, + char *buf) { - struct obd_device *obd = data; - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - rc = snprintf(page, count, "capability on: %s\n", - obd->u.filter.fo_fl_oss_capa ? "oss" : ""); - return rc; + return sprintf(buf, "%u\n", ofd->ofd_raid_degraded); } -static int lprocfs_ofd_wr_capa(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Set OFD to degraded mode. + * + * This is used to interface to userspace administrative tools for + * the underlying RAID storage, so that they can mark an OST + * as having degraded performance. + * + * \param[in] file proc file + * \param[in] buffer string which represents mode + * 1: set degraded mode + * 0: unset degraded mode + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t degraded_store(struct kobject *kobj, struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - int val, rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + bool val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = kstrtobool(buffer, &val); if (rc) return rc; - if (val & ~0x1) { - CERROR("invalid capability mode, only 0/1 are accepted.\n" - " 1: enable oss fid capability\n" - " 0: disable oss fid capability\n"); - return -EINVAL; - } - - obd->u.filter.fo_fl_oss_capa = val; - LCONSOLE_INFO("OSS %s %s fid capability.\n", obd->obd_name, - val ? "enabled" : "disabled"); + spin_lock(&ofd->ofd_flags_lock); + ofd->ofd_raid_degraded = val; + spin_unlock(&ofd->ofd_flags_lock); return count; } +LUSTRE_RW_ATTR(degraded); -static int lprocfs_ofd_rd_capa_count(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show OFD filesystem type. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr, + char *buf) { - return snprintf(page, count, "%d %d\n", - capa_count[CAPA_SITE_CLIENT], - capa_count[CAPA_SITE_SERVER]); + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + struct lu_device *d; + + LASSERT(ofd->ofd_osd); + d = &ofd->ofd_osd->dd_lu_dev; + LASSERT(d->ld_type); + return sprintf(buf, "%s\n", d->ld_type->ldt_name); } +LUSTRE_RO_ATTR(fstype); -int lprocfs_ofd_rd_degraded(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show journal handling mode: synchronous or asynchronous. + * + * When running in asynchronous mode the journal transactions are not + * committed to disk before the RPC is replied back to the client. + * This will typically improve client performance when only a small number + * of clients are writing, since the client(s) can have more write RPCs + * in flight. However, it also means that the client has to handle recovery + * on bulk RPCs, and will have to keep more dirty pages in cache before they + * are committed on the OST. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t sync_journal_show(struct kobject *kobj, struct attribute *attr, + char *buf) { - struct obd_device *obd = data; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - return snprintf(page, count, "%u\n", ofd->ofd_raid_degraded); + return sprintf(buf, "%u\n", ofd->ofd_sync_journal); } -int lprocfs_ofd_wr_degraded(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Set journal mode to synchronous or asynchronous. + * + * \param[in] file proc file + * \param[in] buffer string which represents mode + * 1: synchronous mode + * 0: asynchronous mode + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t sync_journal_store(struct kobject *kobj, struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val, rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + bool val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = kstrtobool(buffer, &val); if (rc) return rc; spin_lock(&ofd->ofd_flags_lock); - ofd->ofd_raid_degraded = !!val; + ofd->ofd_sync_journal = val; + ofd_slc_set(ofd); spin_unlock(&ofd->ofd_flags_lock); return count; } +LUSTRE_RW_ATTR(sync_journal); -int lprocfs_ofd_rd_fstype(char *page, char **start, off_t off, int count, - int *eof, void *data) -{ - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - struct lu_device *d; - - LASSERT(ofd->ofd_osd); - d = &ofd->ofd_osd->dd_lu_dev; - LASSERT(d->ld_type); - return snprintf(page, count, "%s\n", d->ld_type->ldt_name); -} +/* This must be longer than the longest string below */ +#define SYNC_STATES_MAXLEN 16 -int lprocfs_ofd_rd_syncjournal(char *page, char **start, off_t off, - int count, int *eof, void *data) +static int ofd_brw_size_seq_show(struct seq_file *m, void *data) { - struct obd_device *obd = data; + struct obd_device *obd = m->private; struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int rc; - rc = snprintf(page, count, "%u\n", ofd->ofd_syncjournal); - return rc; + seq_printf(m, "%u\n", ofd->ofd_brw_size / ONE_MB_BRW_SIZE); + return 0; } -int lprocfs_ofd_wr_syncjournal(struct file *file, const char *buffer, - unsigned long count, void *data) +static ssize_t +ofd_brw_size_seq_write(struct file *file, const char __user *buffer, + size_t count, loff_t *off) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; - int rc; + struct seq_file *m = file->private_data; + struct obd_device *obd = m->private; + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + __s64 val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); + rc = lprocfs_str_with_units_to_s64(buffer, count, &val, 'M'); if (rc) return rc; - if (val < 0) + if (val <= 0) return -EINVAL; + if (val > DT_MAX_BRW_SIZE || + val < (1 << ofd->ofd_lut.lut_tgd.tgd_blockbits)) + return -ERANGE; + spin_lock(&ofd->ofd_flags_lock); - ofd->ofd_syncjournal = !!val; - ofd_slc_set(ofd); + ofd->ofd_brw_size = val; spin_unlock(&ofd->ofd_flags_lock); return count; } +LPROC_SEQ_FOPS(ofd_brw_size); + static char *sync_on_cancel_states[] = {"never", "blocking", "always" }; -int lprocfs_ofd_rd_sync_lock_cancel(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show OFD policy for handling dirty data under a lock being cancelled. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t sync_lock_cancel_show(struct kobject *kobj, + struct attribute *attr, char *buf) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct lu_target *tgt = obd->u.obt.obt_lut; - rc = snprintf(page, count, "%s\n", - sync_on_cancel_states[ofd->ofd_sync_lock_cancel]); - return rc; + return sprintf(buf, "%s\n", + sync_on_cancel_states[tgt->lut_sync_lock_cancel]); } -int lprocfs_ofd_wr_sync_lock_cancel(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Change OFD policy for handling dirty data under a lock being cancelled. + * + * This variable defines what action OFD takes upon lock cancel + * There are three possible modes: + * 1) never - never do sync upon lock cancel. This can lead to data + * inconsistencies if both the OST and client crash while writing a file + * that is also concurrently being read by another client. In these cases, + * this may allow the file data to "rewind" to an earlier state. + * 2) blocking - do sync only if there is blocking lock, e.g. if another + * client is trying to access this same object + * 3) always - do sync always + * + * \param[in] file proc file + * \param[in] buffer string which represents policy + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t sync_lock_cancel_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val = -1; - int i; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct lu_target *tgt = obd->u.obt.obt_lut; + int val = -1; + int i; + + if (count == 0 || count >= SYNC_STATES_MAXLEN) + return -EINVAL; for (i = 0 ; i < NUM_SYNC_ON_CANCEL_STATES; i++) { - if (memcmp(buffer, sync_on_cancel_states[i], - strlen(sync_on_cancel_states[i])) == 0) { + if (strcmp(buffer, sync_on_cancel_states[i]) == 0) { val = i; break; } } - if (val == -1) { - int rc; - rc = lprocfs_write_helper(buffer, count, &val); + /* Legacy numeric codes */ + if (val == -1) { + int rc = kstrtoint(buffer, 0, &val); if (rc) return rc; } @@ -430,118 +557,450 @@ int lprocfs_ofd_wr_sync_lock_cancel(struct file *file, const char *buffer, if (val < 0 || val > 2) return -EINVAL; - spin_lock(&ofd->ofd_flags_lock); - ofd->ofd_sync_lock_cancel = val; - spin_unlock(&ofd->ofd_flags_lock); + spin_lock(&tgt->lut_flags_lock); + tgt->lut_sync_lock_cancel = val; + spin_unlock(&tgt->lut_flags_lock); return count; } +LUSTRE_RW_ATTR(sync_lock_cancel); -int lprocfs_ofd_rd_grant_compat_disable(char *page, char **start, off_t off, - int count, int *eof, void *data) +/** + * Show the limit of soft sync RPCs. + * + * This value defines how many IO RPCs with OBD_BRW_SOFT_SYNC flag + * are allowed before sync update will be triggered. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t soft_sync_limit_show(struct kobject *kobj, + struct attribute *attr, char *buf) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - rc = snprintf(page, count, "%u\n", ofd->ofd_grant_compat_disable); - return rc; + return sprintf(buf, "%u\n", ofd->ofd_soft_sync_limit); } -int lprocfs_ofd_wr_grant_compat_disable(struct file *file, const char *buffer, - unsigned long count, void *data) +/** + * Change the limit of soft sync RPCs. + * + * Define how many IO RPCs with OBD_BRW_SOFT_SYNC flag + * allowed before sync update will be done. + * + * This limit is global across all exports. + * + * \param[in] file proc file + * \param[in] buffer string which represents limit + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t soft_sync_limit_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - struct obd_device *obd = data; - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - int val; - int rc; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + unsigned int val; + int rc; - rc = lprocfs_write_helper(buffer, count, &val); - if (rc) + rc = kstrtouint(buffer, 0, &val); + if (rc < 0) return rc; - if (val < 0) - return -EINVAL; + ofd->ofd_soft_sync_limit = val; + return 0; +} +LUSTRE_RW_ATTR(soft_sync_limit); - spin_lock(&ofd->ofd_flags_lock); - ofd->ofd_grant_compat_disable = !!val; - spin_unlock(&ofd->ofd_flags_lock); +/** + * Show the LFSCK speed limit. + * + * The maximum number of items scanned per second. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t lfsck_speed_limit_show(struct kobject *kobj, + struct attribute *attr, char *buf) +{ + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + + return lfsck_get_speed(NULL, buf, ofd->ofd_osd); +} + +/** + * Change the LFSCK speed limit. + * + * Limit number of items that may be scanned per second. + * + * \param[in] file proc file + * \param[in] buffer string which represents limit + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t lfsck_speed_limit_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) +{ + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + unsigned int val; + int rc; + + rc = kstrtouint(buffer, 0, &val); + if (rc != 0) + return rc; + + rc = lfsck_set_speed(ofd->ofd_osd, val); + + return rc != 0 ? rc : count; +} +LUSTRE_RW_ATTR(lfsck_speed_limit); + +/** + * Show LFSCK layout verification stats from the most recent LFSCK run. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static int ofd_lfsck_layout_seq_show(struct seq_file *m, void *data) +{ + struct obd_device *obd = m->private; + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + + return lfsck_dump(m, ofd->ofd_osd, LFSCK_TYPE_LAYOUT); +} + +LPROC_SEQ_FOPS_RO(ofd_lfsck_layout); + +/** + * Show if LFSCK performed parent FID verification. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static int ofd_lfsck_verify_pfid_seq_show(struct seq_file *m, void *data) +{ + struct obd_device *obd = m->private; + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + + seq_printf(m, "switch: %s\ndetected: %llu\nrepaired: %llu\n", + ofd->ofd_lfsck_verify_pfid ? "on" : "off", + ofd->ofd_inconsistency_self_detected, + ofd->ofd_inconsistency_self_repaired); + return 0; +} + +/** + * Set the LFSCK behavior to verify parent FID correctness. + * + * If flag ofd_lfsck_verify_pfid is set then LFSCK does parent FID + * verification during read/write operations. + * + * \param[in] file proc file + * \param[in] buffer string which represents behavior + * 1: verify parent FID + * 0: don't verify parent FID + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t +ofd_lfsck_verify_pfid_seq_write(struct file *file, const char __user *buffer, + size_t count, loff_t *off) +{ + struct seq_file *m = file->private_data; + struct obd_device *obd = m->private; + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + bool val; + int rc; + + rc = kstrtobool_from_user(buffer, count, &val); + if (rc) + return rc; + + ofd->ofd_lfsck_verify_pfid = val; + if (!ofd->ofd_lfsck_verify_pfid) { + ofd->ofd_inconsistency_self_detected = 0; + ofd->ofd_inconsistency_self_repaired = 0; + } return count; } -static struct lprocfs_vars lprocfs_ofd_obd_vars[] = { - { "uuid", lprocfs_rd_uuid, 0, 0 }, - { "blocksize", lprocfs_rd_blksize, 0, 0 }, - { "kbytestotal", lprocfs_rd_kbytestotal, 0, 0 }, - { "kbytesfree", lprocfs_rd_kbytesfree, 0, 0 }, - { "kbytesavail", lprocfs_rd_kbytesavail, 0, 0 }, - { "filestotal", lprocfs_rd_filestotal, 0, 0 }, - { "filesfree", lprocfs_rd_filesfree, 0, 0 }, - { "seqs_allocated", lprocfs_ofd_rd_seqs, 0, 0 }, - { "fstype", lprocfs_ofd_rd_fstype, 0, 0 }, - { "last_id", lprocfs_ofd_rd_last_id, 0, 0 }, - { "tot_dirty", lprocfs_ofd_rd_tot_dirty, 0, 0 }, - { "tot_pending", lprocfs_ofd_rd_tot_pending, 0, 0 }, - { "tot_granted", lprocfs_ofd_rd_tot_granted, 0, 0 }, - { "grant_precreate", lprocfs_ofd_rd_grant_precreate, 0, 0 }, - { "grant_ratio", lprocfs_ofd_rd_grant_ratio, - lprocfs_ofd_wr_grant_ratio, 0, 0 }, - { "precreate_batch", lprocfs_ofd_rd_precreate_batch, - lprocfs_ofd_wr_precreate_batch, 0 }, - { "recovery_status", lprocfs_obd_rd_recovery_status, 0, 0 }, - { "recovery_time_soft", lprocfs_obd_rd_recovery_time_soft, - lprocfs_obd_wr_recovery_time_soft, 0}, - { "recovery_time_hard", lprocfs_obd_rd_recovery_time_hard, - lprocfs_obd_wr_recovery_time_hard, 0}, - { "evict_client", 0, lprocfs_wr_evict_client, 0, - &lprocfs_evict_client_fops}, - { "num_exports", lprocfs_rd_num_exports, 0, 0 }, - { "degraded", lprocfs_ofd_rd_degraded, - lprocfs_ofd_wr_degraded, 0}, - { "sync_journal", lprocfs_ofd_rd_syncjournal, - lprocfs_ofd_wr_syncjournal, 0 }, - { "sync_on_lock_cancel", lprocfs_ofd_rd_sync_lock_cancel, - lprocfs_ofd_wr_sync_lock_cancel, 0 }, - { "instance", lprocfs_target_rd_instance, 0 }, - { "ir_factor", lprocfs_obd_rd_ir_factor, - lprocfs_obd_wr_ir_factor, 0}, - { "grant_compat_disable", lprocfs_ofd_rd_grant_compat_disable, - lprocfs_ofd_wr_grant_compat_disable, 0 }, - { "client_cache_count", lprocfs_ofd_rd_fmd_max_num, - lprocfs_ofd_wr_fmd_max_num, 0 }, - { "client_cache_seconds", lprocfs_ofd_rd_fmd_max_age, - lprocfs_ofd_wr_fmd_max_age, 0 }, - { "capa", lprocfs_ofd_rd_capa, - lprocfs_ofd_wr_capa, 0 }, - { "capa_count", lprocfs_ofd_rd_capa_count, 0, 0 }, - { "job_cleanup_interval", lprocfs_rd_job_interval, - lprocfs_wr_job_interval, 0}, - { 0 } -}; +LPROC_SEQ_FOPS(ofd_lfsck_verify_pfid); -static struct lprocfs_vars lprocfs_ofd_module_vars[] = { - { "num_refs", lprocfs_rd_numrefs, 0, 0 }, - { 0 } -}; +static int ofd_site_stats_seq_show(struct seq_file *m, void *data) +{ + struct obd_device *obd = m->private; + + return lu_site_stats_seq_print(obd->obd_lu_dev->ld_site, m); +} -void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars) +LPROC_SEQ_FOPS_RO(ofd_site_stats); + +/** + * Show if the OFD enforces T10PI checksum. + * + * \param[in] m seq_file handle + * \param[in] data unused for single entry + * + * \retval 0 on success + * \retval negative value on error + */ +static ssize_t checksum_t10pi_enforce_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + + return sprintf(buf, "%u\n", ofd->ofd_checksum_t10pi_enforce); +} + +/** + * Force specific T10PI checksum modes to be enabled + * + * If T10PI *is* supported in hardware, allow only the supported T10PI type + * to be used. If T10PI is *not* supported by the OSD, setting the enforce + * parameter forces all T10PI types to be enabled (even if slower) for + * testing. + * + * The final determination of which algorithm to be used depends whether + * the client supports T10PI or not, and is handled at client connect time. + * + * \param[in] file proc file + * \param[in] buffer string which represents mode + * 1: set T10PI checksums enforced + * 0: unset T10PI checksums enforced + * \param[in] count \a buffer length + * \param[in] off unused for single entry + * + * \retval \a count on success + * \retval negative number on error + */ +static ssize_t checksum_t10pi_enforce_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) { - lvars->module_vars = lprocfs_ofd_module_vars; - lvars->obd_vars = lprocfs_ofd_obd_vars; + struct obd_device *obd = container_of(kobj, struct obd_device, + obd_kset.kobj); + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + bool enforce; + int rc; + + rc = kstrtobool(buffer, &enforce); + if (rc) + return rc; + + spin_lock(&ofd->ofd_flags_lock); + ofd->ofd_checksum_t10pi_enforce = enforce; + spin_unlock(&ofd->ofd_flags_lock); + return count; } +LUSTRE_RW_ATTR(checksum_t10pi_enforce); + +LPROC_SEQ_FOPS_RO_TYPE(ofd, recovery_status); +LPROC_SEQ_FOPS_RW_TYPE(ofd, recovery_time_soft); +LPROC_SEQ_FOPS_RW_TYPE(ofd, recovery_time_hard); +LPROC_SEQ_FOPS_WR_ONLY(ofd, evict_client); +LPROC_SEQ_FOPS_RO_TYPE(ofd, num_exports); +LPROC_SEQ_FOPS_RO_TYPE(ofd, target_instance); +LPROC_SEQ_FOPS_RW_TYPE(ofd, ir_factor); +LPROC_SEQ_FOPS_RW_TYPE(ofd, checksum_dump); +LPROC_SEQ_FOPS_RW_TYPE(ofd, job_interval); + +LPROC_SEQ_FOPS_RO(tgt_tot_dirty); +LPROC_SEQ_FOPS_RO(tgt_tot_granted); +LPROC_SEQ_FOPS_RO(tgt_tot_pending); +LPROC_SEQ_FOPS(tgt_grant_compat_disable); + +struct lprocfs_vars lprocfs_ofd_obd_vars[] = { + { .name = "last_id", + .fops = &ofd_last_id_fops }, + { .name = "tot_dirty", + .fops = &tgt_tot_dirty_fops }, + { .name = "tot_pending", + .fops = &tgt_tot_pending_fops }, + { .name = "tot_granted", + .fops = &tgt_tot_granted_fops }, + { .name = "recovery_status", + .fops = &ofd_recovery_status_fops }, + { .name = "recovery_time_soft", + .fops = &ofd_recovery_time_soft_fops }, + { .name = "recovery_time_hard", + .fops = &ofd_recovery_time_hard_fops }, + { .name = "evict_client", + .fops = &ofd_evict_client_fops }, + { .name = "num_exports", + .fops = &ofd_num_exports_fops }, + { .name = "brw_size", + .fops = &ofd_brw_size_fops }, + { .name = "instance", + .fops = &ofd_target_instance_fops }, + { .name = "ir_factor", + .fops = &ofd_ir_factor_fops }, + { .name = "checksum_dump", + .fops = &ofd_checksum_dump_fops }, + { .name = "grant_compat_disable", + .fops = &tgt_grant_compat_disable_fops }, + { .name = "job_cleanup_interval", + .fops = &ofd_job_interval_fops }, + { .name = "lfsck_layout", + .fops = &ofd_lfsck_layout_fops }, + { .name = "lfsck_verify_pfid", + .fops = &ofd_lfsck_verify_pfid_fops }, + { .name = "site_stats", + .fops = &ofd_site_stats_fops }, + { NULL } +}; +/** + * Initialize OFD statistics counters + * + * param[in] stats statistics counters + */ void ofd_stats_counter_init(struct lprocfs_stats *stats) { - LASSERT(stats && stats->ls_num == LPROC_OFD_STATS_LAST); + LASSERT(stats && stats->ls_num >= LPROC_OFD_STATS_LAST); + lprocfs_counter_init(stats, LPROC_OFD_STATS_READ, - LPROCFS_CNTR_AVGMINMAX, "read", "bytes"); + LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes"); lprocfs_counter_init(stats, LPROC_OFD_STATS_WRITE, - LPROCFS_CNTR_AVGMINMAX, "write", "bytes"); + LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_GETATTR, + 0, "getattr", "reqs"); lprocfs_counter_init(stats, LPROC_OFD_STATS_SETATTR, 0, "setattr", "reqs"); lprocfs_counter_init(stats, LPROC_OFD_STATS_PUNCH, 0, "punch", "reqs"); lprocfs_counter_init(stats, LPROC_OFD_STATS_SYNC, 0, "sync", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_DESTROY, + 0, "destroy", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_CREATE, + 0, "create", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_STATFS, + 0, "statfs", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_GET_INFO, + 0, "get_info", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_SET_INFO, + 0, "set_info", "reqs"); + lprocfs_counter_init(stats, LPROC_OFD_STATS_QUOTACTL, + 0, "quotactl", "reqs"); +} + +LPROC_SEQ_FOPS(lprocfs_nid_stats_clear); + +static struct attribute *ofd_attrs[] = { + &lustre_attr_seqs_allocated.attr, + &lustre_attr_grant_precreate.attr, + &lustre_attr_precreate_batch.attr, + &lustre_attr_client_cache_count.attr, + &lustre_attr_client_cache_seconds.attr, + &lustre_attr_degraded.attr, + &lustre_attr_fstype.attr, + &lustre_attr_sync_journal.attr, + &lustre_attr_sync_lock_cancel.attr, + &lustre_attr_soft_sync_limit.attr, + &lustre_attr_lfsck_speed_limit.attr, + &lustre_attr_checksum_t10pi_enforce.attr, + NULL, +}; + +/** + * Initialize all needed procfs entries for OFD device. + * + * \param[in] ofd OFD device + * + * \retval 0 if successful + * \retval negative value on error + */ +int ofd_tunables_init(struct ofd_device *ofd) +{ + struct obd_device *obd = ofd_obd(ofd); + struct proc_dir_entry *entry; + int rc = 0; + + ENTRY; + /* lprocfs must be setup before the ofd so state can be safely added + * to /proc incrementally as the ofd is setup + */ + obd->obd_ktype.default_attrs = ofd_attrs; + obd->obd_vars = lprocfs_ofd_obd_vars; + rc = lprocfs_obd_setup(obd, false); + if (rc) { + CERROR("%s: lprocfs_obd_setup failed: %d.\n", + obd->obd_name, rc); + RETURN(rc); + } + + rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST); + if (rc) { + CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n", + obd->obd_name, rc); + GOTO(obd_cleanup, rc); + } + + entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL); + if (IS_ERR(entry)) { + rc = PTR_ERR(entry); + CERROR("%s: error %d setting up lprocfs for %s\n", + obd->obd_name, rc, "exports"); + GOTO(obd_free_stats, rc); + } + obd->obd_proc_exports_entry = entry; + + entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear", + obd, &lprocfs_nid_stats_clear_fops); + if (IS_ERR(entry)) { + rc = PTR_ERR(entry); + CERROR("%s: add proc entry 'clear' failed: %d.\n", + obd->obd_name, rc); + GOTO(obd_free_stats, rc); + } + + ofd_stats_counter_init(obd->obd_stats); + + rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST, + ofd_stats_counter_init); + if (rc) + GOTO(obd_free_stats, rc); + + RETURN(0); + +obd_free_stats: + lprocfs_free_obd_stats(obd); +obd_cleanup: + lprocfs_obd_cleanup(obd); + + return rc; } -#endif /* LPROCFS */ +#endif /* CONFIG_PROC_FS */