X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Fofd_obd.c;h=2e74bc4cb520ccdcdf031bf8d970d75ac523dc58;hp=a0d1787a93ae8ec1065888954d8857eadaeb6d42;hb=9b0ebf78f7919a144673edadc4a95bad84fae2d3;hpb=df00ee15534419070c559acdb913898fc9dc3a00 diff --git a/lustre/ofd/ofd_obd.c b/lustre/ofd/ofd_obd.c index a0d1787..2e74bc4 100644 --- a/lustre/ofd/ofd_obd.c +++ b/lustre/ofd/ofd_obd.c @@ -23,7 +23,7 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2016, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,10 +43,9 @@ #define DEBUG_SUBSYSTEM S_FILTER -#include #include "ofd_internal.h" #include -#include +#include #include #include #include @@ -71,9 +70,8 @@ static int ofd_export_stats_init(struct ofd_device *ofd, struct obd_device *obd = ofd_obd(ofd); struct nid_stat *stats; int rc; - ENTRY; - LASSERT(obd->obd_uses_nid_stats); + ENTRY; if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid)) /* Self-export gets no proc entry */ @@ -85,14 +83,11 @@ static int ofd_export_stats_init(struct ofd_device *ofd, RETURN(rc == -EALREADY ? 0 : rc); stats = exp->exp_nid_stats; - stats->nid_stats = lprocfs_alloc_stats(NUM_OBD_STATS + - LPROC_OFD_STATS_LAST, - LPROCFS_STATS_FLAG_NOPERCPU); - if (stats->nid_stats == NULL) + stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST, + LPROCFS_STATS_FLAG_NOPERCPU); + if (!stats->nid_stats) RETURN(-ENOMEM); - lprocfs_init_ops_stats(LPROC_OFD_STATS_LAST, stats->nid_stats); - ofd_stats_counter_init(stats->nid_stats); rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats); @@ -110,6 +105,93 @@ out: } /** + * Decide which checksums both client and OST support, possibly forcing + * the use of T10PI checksums if the hardware supports this. + * + * The clients that have no T10-PI RPC checksum support will use the same + * mechanism to select checksum type as before, and will not be affected by + * the following logic. + * + * For the clients that have T10-PI RPC checksum support: + * + * If the OST supports T10-PI feature and T10-PI checksum is enforced, clients + * will have no other choice for RPC checksum type other than using the T10PI + * checksum type. This is useful for enforcing end-to-end integrity in the + * whole system. + * + * If the OST doesn't support T10-PI feature and T10-PI checksum is enforced, + * together with other checksum with reasonably good speeds (e.g. crc32, + * crc32c, adler, etc.), all T10-PI checksum types understood by the client + * (t10ip512, t10ip4K, t10crc512, t10crc4K) will be added to the available + * checksum types, regardless of the speeds of T10-PI checksums. This is + * useful for testing T10-PI checksum of RPC. + * + * If the OST supports T10-PI feature and T10-PI checksum is NOT enforced, + * the corresponding T10-PI checksum type will be added to the checksum type + * list, regardless of the speed of the T10-PI checksum. This provides clients + * the flexibility to choose whether to enable end-to-end integrity or not. + * + * If the OST does NOT supports T10-PI feature and T10-PI checksum is NOT + * enforced, together with other checksums with reasonably good speeds, + * all the T10-PI checksum types with good speeds will be added into the + * checksum type list. Note that a T10-PI checksum type with a speed worse + * than half of Alder will NOT be added as a option. In this circumstance, + * T10-PI checksum types has the same behavior like other normal checksum + * types. + * + */ +static void +ofd_mask_cksum_types(struct ofd_device *ofd, enum cksum_types *cksum_types) +{ + bool enforce = ofd->ofd_checksum_t10pi_enforce; + enum cksum_types ofd_t10_cksum_type; + enum cksum_types client_t10_types = *cksum_types & OBD_CKSUM_T10_ALL; + enum cksum_types server_t10_types; + + /* + * The client set in ocd_cksum_types the checksum types it + * supports. We have to mask off the algorithms that we don't + * support. T10PI checksum types will be added later. + */ + *cksum_types &= (ofd->ofd_cksum_types_supported & ~OBD_CKSUM_T10_ALL); + server_t10_types = ofd->ofd_cksum_types_supported & OBD_CKSUM_T10_ALL; + ofd_t10_cksum_type = ofd->ofd_lut.lut_dt_conf.ddp_t10_cksum_type; + + /* Quick exit if no T10-PI support on client */ + if (!client_t10_types) + return; + + /* + * This OST has NO T10-PI feature. Add all supported T10-PI checksums + * as options if T10-PI checksum is enforced. If the T10-PI checksum is + * not enforced, only add them as options when speed is good. + */ + if (ofd_t10_cksum_type == 0) { + /* + * Server allows all T10PI checksums, and server_t10_types + * include quick ones. + */ + if (enforce) + *cksum_types |= client_t10_types; + else + *cksum_types |= client_t10_types & server_t10_types; + return; + } + + /* + * This OST has T10-PI feature. Disable all other checksum types if + * T10-PI checksum is enforced. If the T10-PI checksum is not enforced, + * add the checksum type as an option. + */ + if (client_t10_types & ofd_t10_cksum_type) { + if (enforce) + *cksum_types = ofd_t10_cksum_type; + else + *cksum_types |= ofd_t10_cksum_type; + } +} + +/** * Match client and OST server connection feature flags. * * Compute the compatibility flags for a connection request based on @@ -143,24 +225,20 @@ static int ofd_parse_connect_data(const struct lu_env *env, struct obd_connect_data *data, bool new_connection) { - struct ofd_device *ofd = ofd_exp(exp); - struct filter_export_data *fed = &exp->exp_filter_data; + struct ofd_device *ofd = ofd_exp(exp); + struct filter_export_data *fed = &exp->exp_filter_data; if (!data) RETURN(0); - CDEBUG(D_RPCTRACE, "%s: cli %s/%p ocd_connect_flags: %#llx" - " ocd_version: %x ocd_grant: %d ocd_index: %u" - " ocd_group %u\n", + CDEBUG(D_RPCTRACE, + "%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_group %u\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp, data->ocd_connect_flags, data->ocd_version, data->ocd_grant, data->ocd_index, data->ocd_group); if (fed->fed_group != 0 && fed->fed_group != data->ocd_group) { - CWARN("!!! This export (nid %s) used object group %d " - "earlier; now it's trying to use group %d! This could " - "be a bug in the MDS. Please report to " - "https://jira.hpdd.intel.com/\n", + CWARN("!!! This export (nid %s) used object group %d earlier; now it's trying to use group %d! This could be a bug in the MDS. Please report to https://jira.whamcloud.com/\n", obd_export_nid2str(exp), fed->fed_group, data->ocd_group); RETURN(-EPROTO); @@ -172,8 +250,6 @@ static int ofd_parse_connect_data(const struct lu_env *env, if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2) data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2; - data->ocd_version = LUSTRE_VERSION_CODE; - /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */ if (data->ocd_connect_flags & OBD_CONNECT_MDS) CDEBUG(D_HA, "%s: Received MDS connection for group %u\n", @@ -188,11 +264,7 @@ static int ofd_parse_connect_data(const struct lu_env *env, if (data->ocd_brw_size > ofd->ofd_brw_size) data->ocd_brw_size = ofd->ofd_brw_size; if (data->ocd_brw_size == 0) { - CERROR("%s: cli %s/%p ocd_connect_flags: %#llx" - " ocd_version: %x ocd_grant: %d ocd_index: %u " - "ocd_brw_size is unexpectedly zero, " - "network data corruption?" - "Refusing connection of this client\n", + CERROR("%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_brw_size is unexpectedly zero, network data corruption? Refusing connection of this client\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp, data->ocd_connect_flags, data->ocd_version, @@ -202,34 +274,37 @@ static int ofd_parse_connect_data(const struct lu_env *env, } if (OCD_HAS_FLAG(data, GRANT_PARAM)) { + struct dt_device_param *ddp = &ofd->ofd_lut.lut_dt_conf; + /* client is reporting its page size, for future use */ - exp->exp_filter_data.fed_pagebits = data->ocd_grant_blkbits; - data->ocd_grant_blkbits = ofd->ofd_blockbits; - /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs - * it's LDISKFS_DIR_REC_LEN(20) = 28. */ - data->ocd_grant_inobits = - fls(ofd->ofd_dt_conf.ddp_inodespace - 1); + exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits; + data->ocd_grant_blkbits = ofd->ofd_lut.lut_tgd.tgd_blockbits; + /* + * ddp_inodespace may not be power-of-two value, eg. for ldiskfs + * it's LDISKFS_DIR_REC_LEN(20) = 28. + */ + data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1); /* ocd_grant_tax_kb is in 1K byte blocks */ - data->ocd_grant_tax_kb = ofd->ofd_dt_conf.ddp_extent_tax >> 10; - data->ocd_grant_max_blks = ofd->ofd_dt_conf.ddp_max_extent_blks; + data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10; + data->ocd_grant_max_blks = ddp->ddp_max_extent_blks; } - if (OCD_HAS_FLAG(data, GRANT)) { - /* Save connect_data we have so far because ofd_grant_connect() - * uses it to calculate grant. */ - exp->exp_connect_data = *data; - ofd_grant_connect(env, exp, data, new_connection); - } + /* + * Save connect_data we have so far because tgt_grant_connect() + * uses it to calculate grant, and we want to save the client + * version before it is overwritten by LUSTRE_VERSION_CODE. + */ + exp->exp_connect_data = *data; + if (OCD_HAS_FLAG(data, GRANT)) + tgt_grant_connect(env, exp, data, new_connection); if (data->ocd_connect_flags & OBD_CONNECT_INDEX) { struct lr_server_data *lsd = &ofd->ofd_lut.lut_lsd; int index = lsd->lsd_osd_index; if (index != data->ocd_index) { - LCONSOLE_ERROR_MSG(0x136, "Connection from %s to index" - " %u doesn't match actual OST index" - " %u in last_rcvd file, bad " - "configuration?\n", + LCONSOLE_ERROR_MSG(0x136, + "Connection from %s to index %u doesn't match actual OST index %u in last_rcvd file, bad configuration?\n", obd_export_nid2str(exp), index, data->ocd_index); RETURN(-EBADF); @@ -237,8 +312,10 @@ static int ofd_parse_connect_data(const struct lu_env *env, if (!(lsd->lsd_feature_compat & OBD_COMPAT_OST)) { /* this will only happen on the first connect */ lsd->lsd_feature_compat |= OBD_COMPAT_OST; - /* sync is not needed here as tgt_client_new will - * set exp_need_sync flag */ + /* + * sync is not needed here as tgt_client_new will + * set exp_need_sync flag + */ tgt_server_data_update(env, &ofd->ofd_lut, 0); } } @@ -246,31 +323,32 @@ static int ofd_parse_connect_data(const struct lu_env *env, if (data->ocd_connect_flags & OBD_CONNECT_CKSUM) { __u32 cksum_types = data->ocd_cksum_types; - /* The client set in ocd_cksum_types the checksum types it - * supports. We have to mask off the algorithms that we don't - * support */ - data->ocd_cksum_types &= cksum_types_supported_server(); + ofd_mask_cksum_types(ofd, &data->ocd_cksum_types); if (unlikely(data->ocd_cksum_types == 0)) { - CERROR("%s: Connect with checksum support but no " - "ocd_cksum_types is set\n", + CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n", exp->exp_obd->obd_name); RETURN(-EPROTO); } - CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return " - "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp), + CDEBUG(D_RPCTRACE, + "%s: cli %s supports cksum type %x, return %x\n", + exp->exp_obd->obd_name, obd_export_nid2str(exp), cksum_types, data->ocd_cksum_types); } else { - /* This client does not support OBD_CONNECT_CKSUM. - * Report failure to negotiate checksum at connect */ - CDEBUG(D_RPCTRACE, "%s: cli %s does not support " - "OBD_CONNECT_CKSUM\n", + /* + * This client does not support OBD_CONNECT_CKSUM. + * Report failure to negotiate checksum at connect + */ + CDEBUG(D_RPCTRACE, + "%s: cli %s does not support OBD_CONNECT_CKSUM\n", exp->exp_obd->obd_name, obd_export_nid2str(exp)); } if (data->ocd_connect_flags & OBD_CONNECT_MAXBYTES) - data->ocd_maxbytes = ofd->ofd_dt_conf.ddp_maxbytes; + data->ocd_maxbytes = ofd->ofd_lut.lut_dt_conf.ddp_maxbytes; + + data->ocd_version = LUSTRE_VERSION_CODE; if (OCD_HAS_FLAG(data, PINGLESS)) { if (ptlrpc_pinger_suppress_pings()) { @@ -306,12 +384,12 @@ static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp, struct obd_connect_data *data, void *client_nid) { - struct ofd_device *ofd; - int rc; + struct ofd_device *ofd; + int rc; ENTRY; - if (exp == NULL || obd == NULL || cluuid == NULL) + if (!exp || !obd || !cluuid) RETURN(-EINVAL); rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp); @@ -350,13 +428,14 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { - struct obd_export *exp; - struct ofd_device *ofd; - struct lustre_handle conn = { 0 }; - int rc; + struct obd_export *exp; + struct ofd_device *ofd; + struct lustre_handle conn = { 0 }; + int rc; + ENTRY; - if (_exp == NULL || obd == NULL || cluuid == NULL) + if (!_exp || !obd || !cluuid) RETURN(-EINVAL); ofd = ofd_dev(obd->obd_lu_dev); @@ -368,13 +447,14 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp, exp = class_conn2export(&conn); LASSERT(exp != NULL); - if (localdata != NULL) { + if (localdata) { rc = nodemap_add_member(*(lnet_nid_t *)localdata, exp); if (rc != 0 && rc != -EEXIST) GOTO(out, rc); } else { - CDEBUG(D_HA, "%s: cannot find nodemap for client %s: " - "nid is null\n", obd->obd_name, cluuid->uuid); + CDEBUG(D_HA, + "%s: cannot find nodemap for client %s: nid is null\n", + obd->obd_name, cluuid->uuid); } rc = ofd_parse_connect_data(env, exp, data, true); @@ -419,9 +499,9 @@ out: */ int ofd_obd_disconnect(struct obd_export *exp) { - struct ofd_device *ofd = ofd_exp(exp); - struct lu_env env; - int rc; + struct ofd_device *ofd = ofd_exp(exp); + struct lu_env env; + int rc; ENTRY; @@ -429,11 +509,11 @@ int ofd_obd_disconnect(struct obd_export *exp) class_export_get(exp); if (!(exp->exp_flags & OBD_OPT_FORCE)) - ofd_grant_sanity_check(ofd_obd(ofd), __FUNCTION__); + tgt_grant_sanity_check(ofd_obd(ofd), __func__); rc = server_disconnect_export(exp); - ofd_grant_discard(exp); + tgt_grant_discard(exp); /* Do not erase record for recoverable client. */ if (exp->exp_obd->obd_replayable && @@ -466,8 +546,6 @@ static int ofd_init_export(struct obd_export *exp) { int rc; - spin_lock_init(&exp->exp_filter_data.fed_lock); - INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list); atomic_set(&exp->exp_filter_data.fed_soft_sync_count, 0); spin_lock(&exp->exp_lock); exp->exp_connecting = 1; @@ -502,10 +580,10 @@ static int ofd_destroy_export(struct obd_export *exp) { struct ofd_device *ofd = ofd_exp(exp); - if (exp->exp_filter_data.fed_pending) - CERROR("%s: cli %s/%p has %lu pending on destroyed export" - "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, - exp, exp->exp_filter_data.fed_pending); + if (exp->exp_target_data.ted_pending) + CERROR("%s: cli %s/%p has %lu pending on destroyed export\n", + exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, + exp, exp->exp_target_data.ted_pending); target_destroy_export(exp); @@ -516,22 +594,18 @@ static int ofd_destroy_export(struct obd_export *exp) ldlm_destroy_export(exp); tgt_client_free(exp); - ofd_fmd_cleanup(exp); - /* * discard grants once we're sure no more * interaction with the client is possible */ - ofd_grant_discard(exp); - ofd_fmd_cleanup(exp); + tgt_grant_discard(exp); if (exp_connect_flags(exp) & OBD_CONNECT_GRANT) - ofd->ofd_tot_granted_clients--; + ofd->ofd_lut.lut_tgd.tgd_tot_granted_clients--; if (!(exp->exp_flags & OBD_OPT_FORCE)) - ofd_grant_sanity_check(exp->exp_obd, __FUNCTION__); + tgt_grant_sanity_check(exp->exp_obd, __func__); - LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list)); return 0; } @@ -554,7 +628,7 @@ int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd) CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd)); - if (!ofd->ofd_skip_lfsck) { + if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) { struct lfsck_start_param lsp; lsp.lsp_start = NULL; @@ -581,9 +655,9 @@ int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd) */ static int ofd_obd_postrecov(struct obd_device *obd) { - struct lu_env env; - struct lu_device *ldev = obd->obd_lu_dev; - int rc; + struct lu_env env; + struct lu_device *ldev = obd->obd_lu_dev; + int rc; ENTRY; @@ -624,7 +698,7 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp, ENTRY; - if (exp->exp_obd == NULL) { + if (!exp->exp_obd) { CDEBUG(D_IOCTL, "invalid export %p\n", exp); RETURN(-EINVAL); } @@ -633,7 +707,7 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp, rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp)); } else { CERROR("%s: Unsupported key %s\n", - exp->exp_obd->obd_name, (char*)key); + exp->exp_obd->obd_name, (char *)key); rc = -EOPNOTSUPP; } RETURN(rc); @@ -660,15 +734,15 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp, static int ofd_get_info(const struct lu_env *env, struct obd_export *exp, __u32 keylen, void *key, __u32 *vallen, void *val) { - struct ofd_thread_info *info; - struct ofd_device *ofd; - struct ll_fiemap_info_key *fm_key = key; - struct fiemap *fiemap = val; - int rc = 0; + struct ofd_thread_info *info; + struct ofd_device *ofd; + struct ll_fiemap_info_key *fm_key = key; + struct fiemap *fiemap = val; + int rc = 0; ENTRY; - if (exp->exp_obd == NULL) { + if (!exp->exp_obd) { CDEBUG(D_IOCTL, "invalid client export %p\n", exp); RETURN(-EINVAL); } @@ -685,7 +759,8 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp, rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap); } else { - CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key); + CERROR("%s: not supported key %s\n", + ofd_name(ofd), (char *)key); rc = -EOPNOTSUPP; } @@ -693,105 +768,6 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp, } /** - * Get file system statistics of OST server. - * - * Helper function for ofd_statfs(), also used by grant code. - * Implements caching for statistics to avoid calling OSD device each time. - * - * \param[in] env execution environment - * \param[in] ofd OFD device - * \param[out] osfs statistic data to return - * \param[in] max_age maximum age for cached data - * \param[in] from_cache show that data was get from cache or not - * - * \retval 0 if successful - * \retval negative value on error - */ -int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd, - struct obd_statfs *osfs, __u64 max_age, int *from_cache) -{ - int rc = 0; - ENTRY; - - spin_lock(&ofd->ofd_osfs_lock); - if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) { - u64 unstable; - - /* statfs data are too old, get up-to-date one. - * we must be cautious here since multiple threads might be - * willing to update statfs data concurrently and we must - * grant that cached statfs data are always consistent */ - - if (ofd->ofd_statfs_inflight == 0) - /* clear inflight counter if no users, although it would - * take a while to overflow this 64-bit counter ... */ - ofd->ofd_osfs_inflight = 0; - /* notify ofd_grant_commit() that we want to track writes - * completed as of now */ - ofd->ofd_statfs_inflight++; - /* record value of inflight counter before running statfs to - * compute the diff once statfs is completed */ - unstable = ofd->ofd_osfs_inflight; - spin_unlock(&ofd->ofd_osfs_lock); - - /* statfs can sleep ... hopefully not for too long since we can - * call it fairly often as space fills up */ - rc = dt_statfs(env, ofd->ofd_osd, osfs); - if (unlikely(rc)) - GOTO(out, rc); - - spin_lock(&ofd->ofd_grant_lock); - spin_lock(&ofd->ofd_osfs_lock); - /* calculate how much space was written while we released the - * ofd_osfs_lock */ - unstable = ofd->ofd_osfs_inflight - unstable; - ofd->ofd_osfs_unstable = 0; - if (unstable) { - /* some writes committed while we were running statfs - * w/o the ofd_osfs_lock. Those ones got added to - * the cached statfs data that we are about to crunch. - * Take them into account in the new statfs data */ - osfs->os_bavail -= min_t(u64, osfs->os_bavail, - unstable >> ofd->ofd_blockbits); - /* However, we don't really know if those writes got - * accounted in the statfs call, so tell - * ofd_grant_space_left() there is some uncertainty - * on the accounting of those writes. - * The purpose is to prevent spurious error messages in - * ofd_grant_space_left() since those writes might be - * accounted twice. */ - ofd->ofd_osfs_unstable += unstable; - } - /* similarly, there is some uncertainty on write requests - * between prepare & commit */ - ofd->ofd_osfs_unstable += ofd->ofd_tot_pending; - spin_unlock(&ofd->ofd_grant_lock); - - /* finally udpate cached statfs data */ - ofd->ofd_osfs = *osfs; - ofd->ofd_osfs_age = cfs_time_current_64(); - - ofd->ofd_statfs_inflight--; /* stop tracking */ - if (ofd->ofd_statfs_inflight == 0) - ofd->ofd_osfs_inflight = 0; - spin_unlock(&ofd->ofd_osfs_lock); - - if (from_cache) - *from_cache = 0; - } else { - /* use cached statfs data */ - *osfs = ofd->ofd_osfs; - spin_unlock(&ofd->ofd_osfs_lock); - if (from_cache) - *from_cache = 1; - } - GOTO(out, rc); - -out: - return rc; -} - -/** * Implementation of obd_ops::o_statfs. * * This function returns information about a storage file system. @@ -813,43 +789,50 @@ out: * \retval negative value on error */ int ofd_statfs(const struct lu_env *env, struct obd_export *exp, - struct obd_statfs *osfs, __u64 max_age, __u32 flags) + struct obd_statfs *osfs, time64_t max_age, __u32 flags) { - struct obd_device *obd = class_exp2obd(exp); - struct ofd_device *ofd = ofd_exp(exp); - int rc; + struct obd_device *obd = class_exp2obd(exp); + struct ofd_device *ofd = ofd_exp(exp); + struct tg_grants_data *tgd = &ofd->ofd_lut.lut_tgd; + int rc; ENTRY; - rc = ofd_statfs_internal(env, ofd, osfs, max_age, NULL); + rc = tgt_statfs_internal(env, &ofd->ofd_lut, osfs, max_age, NULL); if (unlikely(rc)) GOTO(out, rc); - /* at least try to account for cached pages. its still racy and + /* + * at least try to account for cached pages. its still racy and * might be under-reporting if clients haven't announced their - * caches with brw recently */ + * caches with brw recently + */ - CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu" - " pending %llu free %llu avail %llu\n", - ofd->ofd_tot_dirty, ofd->ofd_tot_granted, ofd->ofd_tot_pending, - osfs->os_bfree << ofd->ofd_blockbits, - osfs->os_bavail << ofd->ofd_blockbits); + CDEBUG(D_SUPER | D_CACHE, + "blocks cached %llu granted %llu pending %llu free %llu avail %llu\n", + tgd->tgd_tot_dirty, tgd->tgd_tot_granted, + tgd->tgd_tot_pending, + osfs->os_bfree << tgd->tgd_blockbits, + osfs->os_bavail << tgd->tgd_blockbits); osfs->os_bavail -= min_t(u64, osfs->os_bavail, - ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending + - osfs->os_bsize - 1) >> ofd->ofd_blockbits)); + ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending + + osfs->os_bsize - 1) >> tgd->tgd_blockbits)); - /* The QoS code on the MDS does not care about space reserved for - * precreate, so take it out. */ + /* + * The QoS code on the MDS does not care about space reserved for + * precreate, so take it out. + */ if (exp_connect_flags(exp) & OBD_CONNECT_MDS) { - struct filter_export_data *fed; + struct tg_export_data *ted; - fed = &obd->obd_self_export->exp_filter_data; - osfs->os_bavail -= min_t(u64, osfs->os_bavail, - fed->fed_grant >> ofd->ofd_blockbits); + ted = &obd->obd_self_export->exp_target_data; + osfs->os_granted = min_t(u64, osfs->os_bavail, + ted->ted_grant >> tgd->tgd_blockbits); + osfs->os_bavail -= osfs->os_granted; } - ofd_grant_sanity_check(obd, __FUNCTION__); + tgt_grant_sanity_check(obd, __func__); CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; " "%llu objects: %llu free; state %x\n", osfs->os_blocks, osfs->os_bfree, osfs->os_bavail, @@ -866,16 +849,22 @@ int ofd_statfs(const struct lu_env *env, struct obd_export *exp, if (ofd->ofd_raid_degraded) osfs->os_state |= OS_STATE_DEGRADED; - if (obd->obd_self_export != exp && !ofd_grant_param_supp(exp) && - ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT) { - /* clients which don't support OBD_CONNECT_GRANT_PARAM + if (ofd->ofd_no_precreate) + osfs->os_state |= OS_STATE_NOPRECREATE; + + if (obd->obd_self_export != exp && !exp_grant_param_supp(exp) && + tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) { + /* + * clients which don't support OBD_CONNECT_GRANT_PARAM * should not see a block size > page size, otherwise * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12) * block size which is the biggest block size known to work - * with all client's page size. */ - osfs->os_blocks <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT; - osfs->os_bfree <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT; - osfs->os_bavail <<= ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT; + * with all client's page size. + */ + osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_bfree <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; + osfs->os_granted <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT; osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT; } @@ -908,14 +897,13 @@ out: static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp, struct obdo *oa) { - struct ofd_thread_info *info; - struct ofd_device *ofd = ofd_exp(exp); - struct ldlm_namespace *ns = ofd->ofd_namespace; - struct ldlm_resource *res; - struct ofd_object *fo; - struct lu_fid *fid = &oa->o_oi.oi_fid; - struct filter_fid *ff = NULL; - int rc = 0; + struct ofd_thread_info *info; + struct ofd_device *ofd = ofd_exp(exp); + struct ldlm_namespace *ns = ofd->ofd_namespace; + struct ldlm_resource *res; + struct ofd_object *fo; + struct lu_fid *fid = &oa->o_oi.oi_fid; + int rc = 0; ENTRY; @@ -923,8 +911,10 @@ static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp, ost_fid_build_resid(fid, &info->fti_resid); - /* This would be very bad - accidentally truncating a file when - * changing the time or similar - bug 12203. */ + /* + * This would be very bad - accidentally truncating a file when + * changing the time or similar - bug 12203. + */ if (oa->o_valid & OBD_MD_FLSIZE) { static char mdsinum[48]; @@ -951,13 +941,8 @@ static int ofd_echo_setattr(const struct lu_env *env, struct obd_export *exp, la_from_obdo(&info->fti_attr, oa, oa->o_valid); info->fti_attr.la_valid &= ~LA_TYPE; - if (oa->o_valid & OBD_MD_FLFID) { - ff = &info->fti_mds_fid; - ofd_prepare_fidea(ff, oa); - } - /* setting objects attributes (including owner/group) */ - rc = ofd_attr_set(env, fo, &info->fti_attr, ff); + rc = ofd_attr_set(env, fo, &info->fti_attr, oa); if (rc) GOTO(out_unlock, rc); @@ -967,12 +952,15 @@ out_unlock: ofd_object_put(env, fo); out: if (rc == 0) { - /* we do not call this before to avoid lu_object_find() in + /* + * we do not call this before to avoid lu_object_find() in * ->lvbo_update() holding another reference on the object. * otherwise concurrent destroy can make the object unavailable * for 2nd lu_object_find() waiting for the first reference - * to go... deadlock! */ - res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0); + * to go... deadlock! + */ + res = ldlm_resource_get(ns, NULL, &info->fti_resid, + LDLM_EXTENT, 0); if (!IS_ERR(res)) { ldlm_res_lvbo_update(res, NULL, 0); ldlm_resource_putref(res); @@ -987,7 +975,7 @@ out: * * Supplemental function to destroy object by FID, it is used by request * handler and by ofd_echo_destroy() below to find object by FID, lock it - * and call ofd_object_destroy() finally. + * and call ofd_destroy() finally. * * \param[in] env execution environment * \param[in] ofd OFD device @@ -1013,10 +1001,12 @@ int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd, if (IS_ERR(fo)) RETURN(PTR_ERR(fo)); - /* Tell the clients that the object is gone now and that they should - * throw away any cached pages. */ + /* + * Tell the clients that the object is gone now and that they should + * throw away any cached pages. + */ ost_fid_build_resid(fid, &info->fti_resid); - rc = ldlm_cli_enqueue_local(ofd->ofd_namespace, &info->fti_resid, + rc = ldlm_cli_enqueue_local(env, ofd->ofd_namespace, &info->fti_resid, LDLM_EXTENT, &policy, LCK_PW, &flags, ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL, 0, LVB_T_NONE, NULL, &lockh); @@ -1027,7 +1017,7 @@ int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd, LASSERT(fo != NULL); - rc = ofd_object_destroy(env, fo, orphan); + rc = ofd_destroy(env, fo, orphan); EXIT; ofd_object_put(env, fo); @@ -1054,9 +1044,9 @@ int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd, static int ofd_echo_destroy(const struct lu_env *env, struct obd_export *exp, struct obdo *oa) { - struct ofd_device *ofd = ofd_exp(exp); - struct lu_fid *fid = &oa->o_oi.oi_fid; - int rc = 0; + struct ofd_device *ofd = ofd_exp(exp); + struct lu_fid *fid = &oa->o_oi.oi_fid; + int rc = 0; ENTRY; @@ -1104,13 +1094,14 @@ out: static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp, struct obdo *oa) { - struct ofd_device *ofd = ofd_exp(exp); - u64 seq = ostid_seq(&oa->o_oi); - struct ofd_seq *oseq; - int rc = 0, diff = 1; - long granted; - u64 next_id; - int count; + struct ofd_device *ofd = ofd_exp(exp); + u64 seq = ostid_seq(&oa->o_oi); + struct ofd_seq *oseq; + long granted; + u64 next_id; + s64 diff = 1; + int rc = 0; + int count; ENTRY; @@ -1122,9 +1113,11 @@ static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp, CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi)); down_read(&ofd->ofd_lastid_rwsem); - /* Currently, for safe, we do not distinguish which LAST_ID is broken, + /* + * Currently, for safe, we do not distinguish which LAST_ID is broken, * we may do that in the future. - * Return -ENOSPC until the LAST_ID rebuilt. */ + * Return -ENOSPC until the LAST_ID rebuilt. + */ if (unlikely(ofd->ofd_lastid_rebuilding)) GOTO(out_sem, rc = -ENOSPC); @@ -1140,30 +1133,36 @@ static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp, } mutex_lock(&oseq->os_create_lock); - granted = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff); + granted = tgt_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff); if (granted < 0) { rc = granted; granted = 0; - CDEBUG(D_HA, "%s: failed to acquire grant space for " - "precreate (%d): rc = %d\n", ofd_name(ofd), diff, rc); + CDEBUG(D_HA, + "%s: failed to acquire grant space for precreate (%lld): rc = %d\n", + ofd_name(ofd), diff, rc); diff = 0; GOTO(out, rc); } next_id = ofd_seq_last_oid(oseq) + 1; - count = ofd_precreate_batch(ofd, diff); + count = ofd_precreate_batch(ofd, (int)diff); rc = ofd_precreate_objects(env, ofd, next_id, oseq, count, 0); if (rc < 0) { CERROR("%s: unable to precreate: rc = %d\n", ofd_name(ofd), rc); } else { - ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq)); + rc = ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq)); + if (rc) { + CERROR("%s: Bad %llu to set " DOSTID " : rc %d\n", + ofd_name(ofd), + (unsigned long long)ofd_seq_last_oid(oseq), + POSTID(&oa->o_oi), rc); + } oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP; - rc = 0; } - ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc); + tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc); out: mutex_unlock(&oseq->os_create_lock); ofd_seq_put(env, oseq); @@ -1191,11 +1190,11 @@ out_sem: static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp, struct obdo *oa) { - struct ofd_device *ofd = ofd_exp(exp); - struct ofd_thread_info *info; - struct lu_fid *fid = &oa->o_oi.oi_fid; - struct ofd_object *fo; - int rc = 0; + struct ofd_device *ofd = ofd_exp(exp); + struct ofd_thread_info *info; + struct lu_fid *fid = &oa->o_oi.oi_fid; + struct ofd_object *fo; + int rc = 0; ENTRY; @@ -1212,7 +1211,7 @@ static int ofd_echo_getattr(const struct lu_env *env, struct obd_export *exp, __u64 curr_version; obdo_from_la(oa, &info->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); + OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); /* Store object version in reply */ curr_version = dt_version_get(env, ofd_object_child(fo)); @@ -1244,26 +1243,28 @@ static int ofd_ioc_get_obj_version(const struct lu_env *env, struct ofd_device *ofd, void *karg) { struct obd_ioctl_data *data = karg; - struct lu_fid fid; - struct ofd_object *fo; - dt_obj_version_t version; - int rc = 0; + struct lu_fid fid; + struct ofd_object *fo; + dt_obj_version_t version; + int rc = 0; ENTRY; - if (data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version)) + if (!data->ioc_inlbuf2 || data->ioc_inllen2 != sizeof(version)) GOTO(out, rc = -EINVAL); - if (data->ioc_inlbuf1 != NULL && data->ioc_inllen1 == sizeof(fid)) { + if (data->ioc_inlbuf1 && data->ioc_inllen1 == sizeof(fid)) { fid = *(struct lu_fid *)data->ioc_inlbuf1; - } else if (data->ioc_inlbuf3 != NULL && + } else if (data->ioc_inlbuf3 && data->ioc_inllen3 == sizeof(__u64) && - data->ioc_inlbuf4 != NULL && + data->ioc_inlbuf4 && data->ioc_inllen4 == sizeof(__u64)) { - struct ost_id ostid; + struct ost_id ostid = { }; ostid_set_seq(&ostid, *(__u64 *)data->ioc_inlbuf4); - ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3); + rc = ostid_set_id(&ostid, *(__u64 *)data->ioc_inlbuf3); + if (rc) + GOTO(out, rc); rc = ostid_to_fid(&fid, &ostid, ofd->ofd_lut.lut_lsd.lsd_osd_index); if (rc != 0) @@ -1316,10 +1317,10 @@ out: static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void __user *uarg) { - struct lu_env env; - struct ofd_device *ofd = ofd_exp(exp); - struct obd_device *obd = ofd_obd(ofd); - int rc; + struct lu_env env; + struct ofd_device *ofd = ofd_exp(exp); + struct obd_device *obd = ofd_obd(ofd); + int rc; ENTRY; @@ -1347,7 +1348,7 @@ static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, struct obd_ioctl_data *data = karg; struct lfsck_start_param lsp; - if (unlikely(data == NULL)) { + if (unlikely(!data)) { rc = -EINVAL; break; } @@ -1363,7 +1364,7 @@ static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, stop.ls_status = LS_STOPPED; /* Old lfsck utils may pass NULL @stop. */ - if (data->ioc_inlbuf1 == NULL) + if (!data->ioc_inlbuf1) stop.ls_flags = 0; else stop.ls_flags = @@ -1403,26 +1404,6 @@ static int ofd_precleanup(struct obd_device *obd) } /** - * Implementation of obd_ops::o_ping. - * - * This is OFD-specific part of OBD_PING request handling. - * It controls Filter Modification Data (FMD) expiration each time PING is - * received. - * - * \see ofd_fmd_expire() and ofd_fmd.c for details - * - * \param[in] env execution environment - * \param[in] exp OBD export of client - * - * \retval 0 - */ -static int ofd_ping(const struct lu_env *env, struct obd_export *exp) -{ - ofd_fmd_expire(exp); - return 0; -} - -/** * Implementation of obd_ops::o_health_check. * * This function checks the OFD device health - ability to respond on @@ -1442,13 +1423,13 @@ static int ofd_ping(const struct lu_env *env, struct obd_export *exp) */ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) { - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - struct ofd_thread_info *info; - struct lu_env env; + struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); + struct ofd_thread_info *info; + struct lu_env env; #ifdef USE_HEALTH_CHECK_WRITE - struct thandle *th; + struct thandle *th; #endif - int rc = 0; + int rc = 0; /* obd_proc_read_health pass NULL env, we need real one */ rc = lu_env_init(&env, LCT_DT_THREAD); @@ -1465,7 +1446,7 @@ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) #ifdef USE_HEALTH_CHECK_WRITE OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE); - if (info->fti_buf.lb_buf == NULL) + if (!info->fti_buf.lb_buf) GOTO(out, rc = -ENOMEM); info->fti_buf.lb_len = PAGE_SIZE; @@ -1489,7 +1470,8 @@ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE); - CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc); + CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n", + rc); #endif out: lu_env_fini(&env); @@ -1513,7 +1495,6 @@ struct obd_ops ofd_obd_ops = { .o_getattr = ofd_echo_getattr, .o_iocontrol = ofd_iocontrol, .o_precleanup = ofd_precleanup, - .o_ping = ofd_ping, .o_health_check = ofd_health_check, .o_set_info_async = ofd_set_info_async, .o_get_info = ofd_get_info,