X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Fofd_obd.c;h=aa59c14ffb021c1f55b92f5309bf0ed4212036fb;hp=c21b7f8c1884d05c7865195b4e95e15ee15b74fe;hb=605a139a193e4ef212560673b4bb93eec763d088;hpb=7fd6681466dbc8007032f836808e077fc1fa244c diff --git a/lustre/ofd/ofd_obd.c b/lustre/ofd/ofd_obd.c index c21b7f8..aa59c14 100644 --- a/lustre/ofd/ofd_obd.c +++ b/lustre/ofd/ofd_obd.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 */ @@ -27,7 +23,7 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,89 +31,206 @@ * * lustre/ofd/ofd_obd.c * - * Author: Andreas Dilger - * Author: Alex Zhuravlev - * Author: Mike Pershin + * This file contains OBD API methods for OBD Filter Device (OFD) which are + * used for export handling, configuration purposes and recovery. + * Several methods are used by ECHO client only since it still uses OBD API. + * Such methods have _echo_ prefix in name. + * + * Author: Andreas Dilger + * Author: Alexey Zhuravlev + * Author: Mikhail Pershin */ #define DEBUG_SUBSYSTEM S_FILTER -#include #include "ofd_internal.h" #include +#include #include +#include +#include +/** + * Initialize OFD per-export statistics. + * + * This function sets up procfs entries for various OFD export counters. These + * counters are for per-client statistics tracked on the server. + * + * \param[in] ofd OFD device + * \param[in] exp OBD export + * \param[in] client_nid NID of client + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_export_stats_init(struct ofd_device *ofd, - struct obd_export *exp, void *client_nid) + struct obd_export *exp, + lnet_nid_t *client_nid) { struct obd_device *obd = ofd_obd(ofd); struct nid_stat *stats; - int num_stats; - int rc, newnid = 0; - + int rc; ENTRY; if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid)) /* Self-export gets no proc entry */ RETURN(0); - rc = lprocfs_exp_setup(exp, client_nid, &newnid); - if (rc) { - /* Mask error for already created - * /proc entries */ - if (rc == -EALREADY) - rc = 0; - RETURN(rc); - } - - if (newnid == 0) - RETURN(0); + rc = lprocfs_exp_setup(exp, client_nid); + if (rc != 0) + /* Mask error for already created /proc entries */ + RETURN(rc == -EALREADY ? 0 : rc); stats = exp->exp_nid_stats; - LASSERT(stats != NULL); - - num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) + - LPROC_OFD_LAST - 1; - - stats->nid_stats = lprocfs_alloc_stats(num_stats, + stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST, LPROCFS_STATS_FLAG_NOPERCPU); if (stats->nid_stats == NULL) - return -ENOMEM; + RETURN(-ENOMEM); - lprocfs_init_ops_stats(LPROC_OFD_LAST, stats->nid_stats); - lprocfs_counter_init(stats->nid_stats, LPROC_OFD_READ_BYTES, - LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes"); - lprocfs_counter_init(stats->nid_stats, LPROC_OFD_WRITE_BYTES, - LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes"); + ofd_stats_counter_init(stats->nid_stats); - rc = lprocfs_register_stats(stats->nid_proc, "stats", - stats->nid_stats); - if (rc) - GOTO(clean, rc); + rc = lprocfs_register_stats(stats->nid_proc, "stats", stats->nid_stats); + if (rc != 0) { + lprocfs_free_stats(&stats->nid_stats); + GOTO(out, rc); + } rc = lprocfs_nid_ldlm_stats_init(stats); - if (rc) { - lprocfs_free_stats(&stats->nid_stats); - GOTO(clean, rc); + if (rc != 0) + GOTO(out, rc); + +out: + RETURN(rc); +} + +/** + * 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; } - RETURN(0); -clean: - return rc; + /* + * 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 + * features mutually supported by client and server. + * + * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp + * must not be updated here, otherwise a partially initialized value may + * be exposed. After the connection request is successfully processed, + * the top-level tgt_connect() request handler atomically updates the export + * connect flags from the obd_connect_data::ocd_connect_flags field of the + * reply. \see tgt_connect(). + * + * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a + * full struct obd_connect_data. So care must be taken when accessing fields + * that are not present in struct obd_connect_data_v1. See LU-16. + * + * \param[in] env execution environment + * \param[in] exp the obd_export associated with this + * client/target pair + * \param[in] data stores data for this connect request + * \param[in] new_connection is this connection new or not + * + * \retval 0 if success + * \retval -EPROTO client and server feature requirements are + * incompatible + * \retval -EBADF OST index in connect request doesn't match + * real OST index + */ static int ofd_parse_connect_data(const struct lu_env *env, struct obd_export *exp, 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: "LPX64 + 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, @@ -128,7 +241,7 @@ static int ofd_parse_connect_data(const struct lu_env *env, 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 " - "http://bugs.whamcloud.com/\n", + "https://jira.whamcloud.com/\n", obd_export_nid2str(exp), fed->fed_group, data->ocd_group); RETURN(-EPROTO); @@ -136,8 +249,9 @@ static int ofd_parse_connect_data(const struct lu_env *env, fed->fed_group = data->ocd_group; data->ocd_connect_flags &= OST_CONNECT_SUPPORTED; - exp->exp_connect_data = *data; - data->ocd_version = LUSTRE_VERSION_CODE; + + if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2) + data->ocd_connect_flags2 &= OST_CONNECT_SUPPORTED2; /* Kindly make sure the SKIP_ORPHAN flag is from MDS. */ if (data->ocd_connect_flags & OBD_CONNECT_MDS) @@ -146,18 +260,46 @@ static int ofd_parse_connect_data(const struct lu_env *env, else if (data->ocd_connect_flags & OBD_CONNECT_SKIP_ORPHAN) RETURN(-EPROTO); - if (ofd_grant_param_supp(exp)) { - exp->exp_filter_data.fed_pagesize = data->ocd_blocksize; - /* ocd_{blocksize,inodespace} are log2 values */ - data->ocd_blocksize = ofd->ofd_blockbits; - data->ocd_inodespace = ofd->ofd_dt_conf.ddp_inodespace; - /* ocd_grant_extent is in 1K blocks */ - data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10; + /* Determine optimal brw size before calculating grant */ + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) { + data->ocd_brw_size = 65536; + } else if (OCD_HAS_FLAG(data, BRW_SIZE)) { + 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", + exp->exp_obd->obd_name, + exp->exp_client_uuid.uuid, + exp, data->ocd_connect_flags, data->ocd_version, + data->ocd_grant, data->ocd_index); + RETURN(-EPROTO); + } + } + + 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_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 = ddp->ddp_extent_tax >> 10; + data->ocd_grant_max_blks = ddp->ddp_max_extent_blks; } - if (exp_connect_flags(exp) & OBD_CONNECT_GRANT) - data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant, - 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; @@ -175,37 +317,16 @@ 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 lut_client_add will + /* sync is not needed here as tgt_client_new will * set exp_need_sync flag */ tgt_server_data_update(env, &ofd->ofd_lut, 0); } } - if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_SIZE)) { - data->ocd_brw_size = 65536; - } else if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) { - data->ocd_brw_size = min(data->ocd_brw_size, - (__u32)DT_MAX_BRW_SIZE); - if (data->ocd_brw_size == 0) { - CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64 - " 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, - data->ocd_grant, data->ocd_index); - RETURN(-EPROTO); - } - } 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 " @@ -218,15 +339,17 @@ static int ofd_parse_connect_data(const struct lu_env *env, "%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 - * fall back to CRC32 */ + /* 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, CRC32 will be used\n", + "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()) { @@ -241,9 +364,26 @@ static int ofd_parse_connect_data(const struct lu_env *env, RETURN(0); } +/** + * Re-initialize export upon client reconnection. + * + * This function parses connection data from reconnect and resets + * export statistics. + * + * \param[in] env execution environment + * \param[in] exp OBD export + * \param[in] obd OFD device + * \param[in] cluuid NID of client + * \param[in] data connection data from request + * \param[in] localdata client NID + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_obd_reconnect(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_connect_data *data, + void *client_nid) { struct ofd_device *ofd; int rc; @@ -253,22 +393,38 @@ static int ofd_obd_reconnect(const struct lu_env *env, struct obd_export *exp, if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); - ofd = ofd_dev(obd->obd_lu_dev); - - rc = lu_env_refill((struct lu_env *)env); - if (rc != 0) { - CERROR("Failure to refill session: '%d'\n", rc); + rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp); + if (rc != 0 && rc != -EEXIST) RETURN(rc); - } - ofd_info_init(env, exp); + ofd = ofd_dev(obd->obd_lu_dev); + rc = ofd_parse_connect_data(env, exp, data, false); if (rc == 0) - ofd_export_stats_init(ofd, exp, localdata); + ofd_export_stats_init(ofd, exp, client_nid); + else + nodemap_del_member(exp); RETURN(rc); } +/** + * Initialize new client connection. + * + * This function handles new connection to the OFD. The new export is + * created (in context of class_connect()) and persistent client data is + * initialized on storage. + * + * \param[in] env execution environment + * \param[out] _exp stores pointer to new export + * \param[in] obd OFD device + * \param[in] cluuid client UUID + * \param[in] data connection data from request + * \param[in] localdata client NID + * + * \retval 0 if successful + * \retval negative value on error + */ 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) @@ -291,14 +447,15 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp, exp = class_conn2export(&conn); LASSERT(exp != NULL); - rc = lu_env_refill((struct lu_env *)env); - if (rc != 0) { - CERROR("Failure to refill session: '%d'\n", rc); - GOTO(out, rc); + if (localdata != NULL) { + 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); } - ofd_info_init(env, exp); - rc = ofd_parse_connect_data(env, exp, data, true); if (rc) GOTO(out, rc); @@ -315,11 +472,12 @@ static int ofd_obd_connect(const struct lu_env *env, struct obd_export **_exp, } CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name, - data->ocd_group); + data ? data->ocd_group : -1); out: if (rc != 0) { class_disconnect(exp); + nodemap_del_member(exp); *_exp = NULL; } else { *_exp = exp; @@ -327,9 +485,20 @@ out: RETURN(rc); } -static int ofd_obd_disconnect(struct obd_export *exp) +/** + * Disconnect a connected client. + * + * This function terminates the client connection. The client export is + * disconnected (cleaned up) and client data on persistent storage is removed. + * + * \param[in] exp OBD export + * + * \retval 0 if successful + * \retval negative value on error + */ +int ofd_obd_disconnect(struct obd_export *exp) { - struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev); + struct ofd_device *ofd = ofd_exp(exp); struct lu_env env; int rc; @@ -339,32 +508,46 @@ static 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); - - rc = lu_env_init(&env, LCT_DT_THREAD); - if (rc) - RETURN(rc); + tgt_grant_discard(exp); /* Do not erase record for recoverable client. */ if (exp->exp_obd->obd_replayable && - (!exp->exp_obd->obd_fail || exp->exp_failed)) - tgt_client_del(&env, exp); - lu_env_fini(&env); + (!exp->exp_obd->obd_fail || exp->exp_failed)) { + rc = lu_env_init(&env, LCT_DT_THREAD); + if (rc) + GOTO(out, rc); + tgt_client_del(&env, exp); + lu_env_fini(&env); + } +out: + nodemap_del_member(exp); class_export_put(exp); RETURN(rc); } +/** + * Implementation of obd_ops::o_init_export. + * + * This function is called from class_new_export() and initializes + * the OFD-specific data for new export. + * + * \param[in] exp OBD export + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_init_export(struct obd_export *exp) { int rc; spin_lock_init(&exp->exp_filter_data.fed_lock); - CFS_INIT_LIST_HEAD(&exp->exp_filter_data.fed_mod_list); + 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; spin_unlock(&exp->exp_lock); @@ -383,14 +566,25 @@ static int ofd_init_export(struct obd_export *exp) return rc; } +/** + * Implementation of obd_ops::o_destroy_export. + * + * This function is called from class_export_destroy() to cleanup + * the OFD-specific data for export being destroyed. + * + * \param[in] exp OBD export + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_destroy_export(struct obd_export *exp) { - struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev); + struct ofd_device *ofd = ofd_exp(exp); - if (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_filter_data.fed_pending); + exp, exp->exp_target_data.ted_pending); target_destroy_export(exp); @@ -407,30 +601,64 @@ static int ofd_destroy_export(struct obd_export *exp) * discard grants once we're sure no more * interaction with the client is possible */ - ofd_grant_discard(exp); + tgt_grant_discard(exp); ofd_fmd_cleanup(exp); - if (exp_connect_flags(exp) & OBD_CONNECT_GRANT_SHRINK) { - if (ofd->ofd_tot_granted_clients > 0) - ofd->ofd_tot_granted_clients --; - } + if (exp_connect_flags(exp) & OBD_CONNECT_GRANT) + 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(cfs_list_empty(&exp->exp_filter_data.fed_mod_list)); + LASSERT(list_empty(&exp->exp_filter_data.fed_mod_list)); return 0; } +/** + * Notify all devices in server stack about recovery completion. + * + * This function calls ldo_recovery_complete() for all lower devices in the + * server stack so they will be prepared for normal operations. + * + * \param[in] env execution environment + * \param[in] ofd OFD device + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd) { struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev; + int rc; + + CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd)); + + if (!ofd->ofd_skip_lfsck && !ofd->ofd_osd->dd_rdonly) { + struct lfsck_start_param lsp; + + lsp.lsp_start = NULL; + lsp.lsp_index_valid = 0; + rc = lfsck_start(env, ofd->ofd_osd, &lsp); + if (rc != 0 && rc != -EALREADY) + CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", + ofd_name(ofd), rc); + } - CDEBUG(D_HA, "%s: recovery is over\n", ofd_obd(ofd)->obd_name); return ldev->ld_ops->ldo_recovery_complete(env, ldev); } -int ofd_obd_postrecov(struct obd_device *obd) +/** + * Implementation of obd_ops::o_postrecov. + * + * This function is called from target_finish_recovery() upon recovery + * completion. + * + * \param[in] obd OBD device of OFD + * + * \retval 0 if successful + * \retval negative value on error + */ +static int ofd_obd_postrecov(struct obd_device *obd) { struct lu_env env; struct lu_device *ldev = obd->obd_lu_dev; @@ -449,48 +677,29 @@ int ofd_obd_postrecov(struct obd_device *obd) RETURN(rc); } -static int ofd_adapt_sptlrpc_conf(const struct lu_env *env, - struct obd_device *obd, int initial) -{ - struct filter_obd *fo = &obd->u.filter; - struct sptlrpc_rule_set tmp_rset; - int rc; - - sptlrpc_rule_set_init(&tmp_rset); - rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial); - if (rc) { - CERROR("%s: failed get sptlrpc rules: rc = %d\n", - obd->obd_name, rc); - return rc; - } - - sptlrpc_target_update_exp_flavor(obd, &tmp_rset); - - write_lock(&fo->fo_sptlrpc_lock); - sptlrpc_rule_set_free(&fo->fo_sptlrpc_rset); - fo->fo_sptlrpc_rset = tmp_rset; - write_unlock(&fo->fo_sptlrpc_lock); - - return 0; -} - -static int ofd_set_mds_conn(struct obd_export *exp, void *val) -{ - int rc = 0; - - ENTRY; - - LCONSOLE_WARN("%s: received MDS connection from %s\n", - exp->exp_obd->obd_name, obd_export_nid2str(exp)); - RETURN(rc); -} - +/** + * Implementation of obd_ops::o_set_info_async. + * + * This function is not called from request handler, it is only used by + * class_notify_sptlrpc_conf() locally by direct obd_set_info_async() call. + * \see ofd_set_info_hdl() for request handler function. + * + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in] keylen length of \a key + * \param[in] key key name + * \param[in] vallen length of \a val + * \param[in] val the \a key value + * \param[in] set not used in OFD + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp, __u32 keylen, void *key, __u32 vallen, void *val, struct ptlrpc_request_set *set) { - struct ofd_device *ofd; - int rc = 0; + int rc = 0; ENTRY; @@ -499,38 +708,42 @@ static int ofd_set_info_async(const struct lu_env *env, struct obd_export *exp, RETURN(-EINVAL); } - ofd = ofd_exp(exp); - - if (KEY_IS(KEY_CAPA_KEY)) { - rc = ofd_update_capa_key(ofd, val); - if (rc) - CERROR("%s: update capability key failed: rc = %d\n", - exp->exp_obd->obd_name, rc); - } else if (KEY_IS(KEY_SPTLRPC_CONF)) { - ofd_adapt_sptlrpc_conf(env, exp->exp_obd, 0); - } else if (KEY_IS(KEY_MDS_CONN)) { - rc = ofd_set_mds_conn(exp, val); - } else if (KEY_IS(KEY_GRANT_SHRINK)) { - struct ost_body *body = val; - - ofd_info_init(env, exp); - /** handle grant shrink, similar to a read request */ - ofd_grant_prepare_read(env, exp, &body->oa); + if (KEY_IS(KEY_SPTLRPC_CONF)) { + rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp)); } else { CERROR("%s: Unsupported key %s\n", exp->exp_obd->obd_name, (char*)key); rc = -EOPNOTSUPP; } - RETURN(rc); } +/** + * Implementation of obd_ops::o_get_info. + * + * This function is not called from request handler, it is only used by + * direct call from nrs_orr_range_fill_physical() in ptlrpc, see LU-3239. + * + * \see ofd_get_info_hdl() for request handler function. + * + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in] keylen length of \a key + * \param[in] key key name + * \param[out] vallen length of key value + * \param[out] val the key value to return + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp, - __u32 keylen, void *key, __u32 *vallen, void *val, - struct lov_stripe_md *lsm) + __u32 keylen, void *key, __u32 *vallen, void *val) { - struct ofd_device *ofd; - 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; @@ -541,222 +754,55 @@ static int ofd_get_info(const struct lu_env *env, struct obd_export *exp, ofd = ofd_exp(exp); - if (KEY_IS(KEY_BLOCKSIZE)) { - __u32 *blocksize = val; - if (blocksize) { - if (*vallen < sizeof(*blocksize)) - RETURN(-EOVERFLOW); - *blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift; - } - *vallen = sizeof(*blocksize); - } else if (KEY_IS(KEY_BLOCKSIZE_BITS)) { - __u32 *blocksize_bits = val; - if (blocksize_bits) { - if (*vallen < sizeof(*blocksize_bits)) - RETURN(-EOVERFLOW); - *blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift; - } - *vallen = sizeof(*blocksize_bits); - } else if (KEY_IS(KEY_LAST_ID)) { - obd_id *last_id = val; - struct ofd_seq *oseq; - - if (val == NULL) { - *vallen = sizeof(obd_id); - RETURN(0); - } - ofd_info_init(env, exp); - oseq = ofd_seq_load(env, ofd, - (obd_seq)exp->exp_filter_data.fed_group); - LASSERT(!IS_ERR(oseq)); - if (last_id) { - if (*vallen < sizeof(*last_id)) { - ofd_seq_put(env, oseq); - RETURN(-EOVERFLOW); - } - *last_id = ofd_seq_last_oid(oseq); - } - ofd_seq_put(env, oseq); - *vallen = sizeof(*last_id); - } else if (KEY_IS(KEY_FIEMAP)) { - struct ofd_thread_info *info; - struct ofd_device *ofd = ofd_exp(exp); - struct ofd_object *fo; - struct ll_fiemap_info_key *fm_key = key; - - if (val == NULL) { - *vallen = fiemap_count_to_size( - fm_key->fiemap.fm_extent_count); - RETURN(0); - } - + if (KEY_IS(KEY_FIEMAP)) { info = ofd_info_init(env, exp); - rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0); - if (rc != 0) - RETURN(rc); - CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n", - PFID(&info->fti_fid)); - - fo = ofd_object_find(env, ofd, &info->fti_fid); - if (IS_ERR(fo)) { - CERROR("%s: error finding object "DFID"\n", - exp->exp_obd->obd_name, PFID(&info->fti_fid)); - rc = PTR_ERR(fo); - } else { - struct ll_user_fiemap *fiemap = val; - - ofd_read_lock(env, fo); - if (ofd_object_exists(fo)) { - *fiemap = fm_key->fiemap; - rc = dt_fiemap_get(env, - ofd_object_child(fo), - fiemap); - } else { - rc = -ENOENT; - } - ofd_read_unlock(env, fo); - ofd_object_put(env, fo); - } - } else if (KEY_IS(KEY_SYNC_LOCK_CANCEL)) { - *((__u32 *) val) = ofd->ofd_sync_lock_cancel; - *vallen = sizeof(__u32); - } else if (KEY_IS(KEY_LAST_FID)) { - struct lu_env env; - struct ofd_device *ofd = ofd_exp(exp); - struct ofd_seq *oseq; - struct lu_fid *fid = val; - int rc; - - if (fid == NULL) { - *vallen = sizeof(struct lu_fid); - RETURN(0); - } - - if (*vallen < sizeof(*fid)) - RETURN(-EOVERFLOW); - rc = lu_env_init(&env, LCT_DT_THREAD); + rc = ostid_to_fid(&info->fti_fid, &fm_key->lfik_oa.o_oi, + ofd->ofd_lut.lut_lsd.lsd_osd_index); if (rc != 0) RETURN(rc); - ofd_info_init(&env, exp); - - fid_le_to_cpu(fid, fid); - - oseq = ofd_seq_load(&env, ofd, ostid_seq((struct ost_id *)fid)); - if (IS_ERR(oseq)) - GOTO(out_fini, rc = PTR_ERR(oseq)); - rc = ostid_to_fid(fid, &oseq->os_oi, - ofd->ofd_lut.lut_lsd.lsd_osd_index); - if (rc != 0) - GOTO(out_put, rc); - - CDEBUG(D_HA, "%s: LAST FID is "DFID"\n", ofd_name(ofd), - PFID(fid)); - *vallen = sizeof(*fid); -out_put: - ofd_seq_put(&env, oseq); -out_fini: - lu_env_fini(&env); + rc = ofd_fiemap_get(env, ofd, &info->fti_fid, fiemap); } else { - CERROR("Not supported key %s\n", (char*)key); + CERROR("%s: not supported key %s\n", ofd_name(ofd), (char*)key); rc = -EOPNOTSUPP; } RETURN(rc); } -/** helper function for statfs, also used by grant code */ -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; - - spin_lock(&ofd->ofd_osfs_lock); - if (cfs_time_before_64(ofd->ofd_osfs_age, max_age) || max_age == 0) { - obd_size 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)) - return 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 completed 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(obd_size, 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; - } - return 0; -} - -static int ofd_statfs(const struct lu_env *env, struct obd_export *exp, - struct obd_statfs *osfs, __u64 max_age, __u32 flags) +/** + * Implementation of obd_ops::o_statfs. + * + * This function returns information about a storage file system. + * It is called from several places by using the OBD API as well as + * by direct call, e.g. from request handler. + * + * \see ofd_statfs_hdl() for request handler function. + * + * Report also the state of the OST to the caller in osfs->os_state + * (OS_STATE_READONLY, OS_STATE_DEGRADED). + * + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[out] osfs statistic data to return + * \param[in] max_age maximum age for cached data + * \param[in] flags not used in OFD + * + * \retval 0 if successful + * \retval negative value on error + */ +int ofd_statfs(const struct lu_env *env, struct obd_export *exp, + struct obd_statfs *osfs, time64_t max_age, __u32 flags) { struct obd_device *obd = class_exp2obd(exp); - struct ofd_device *ofd = ofd_dev(exp->exp_obd->obd_lu_dev); + 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); @@ -764,250 +810,184 @@ static int ofd_statfs(const struct lu_env *env, struct obd_export *exp, * might be under-reporting if clients haven't announced their * caches with brw recently */ - CDEBUG(D_SUPER | D_CACHE, "blocks cached "LPU64" granted "LPU64 - " pending "LPU64" free "LPU64" avail "LPU64"\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(obd_size, osfs->os_bavail, - ((ofd->ofd_tot_dirty + ofd->ofd_tot_pending + - osfs->os_bsize - 1) >> ofd->ofd_blockbits)); + osfs->os_bavail -= min_t(u64, osfs->os_bavail, + ((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. */ 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(obd_size, 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__); - CDEBUG(D_CACHE, LPU64" blocks: "LPU64" free, "LPU64" avail; " - LPU64" objects: "LPU64" free; state %x\n", + 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, osfs->os_files, osfs->os_ffree, osfs->os_state); if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO, - ofd->ofd_lut.lut_lsd.lsd_osd_index)) - osfs->os_ffree = 0; + ofd->ofd_lut.lut_lsd.lsd_osd_index)) { + /* Reduce free inode count to zero, but keep "used" intact */ + osfs->os_files -= osfs->os_ffree; + osfs->os_ffree -= osfs->os_ffree; + } /* OS_STATE_READONLY can be set by OSD already */ if (ofd->ofd_raid_degraded) osfs->os_state |= OS_STATE_DEGRADED; - if (obd->obd_self_export != exp && ofd_grant_compat(exp, ofd)) { + 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; + 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; } if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC, - ofd->ofd_lut.lut_lsd.lsd_osd_index)) - osfs->os_bfree = osfs->os_bavail = 2; + ofd->ofd_lut.lut_lsd.lsd_osd_index)) { + /* Reduce free blocks count near zero, but keep "used" intact */ + osfs->os_bavail -= osfs->os_bavail - 2; + osfs->os_blocks -= osfs->os_bfree - 2; + osfs->os_bfree -= osfs->os_bfree - 2; + } EXIT; out: return rc; } -int ofd_setattr(const struct lu_env *env, struct obd_export *exp, - struct obd_info *oinfo, struct obd_trans_info *oti) +/** + * Implementation of obd_ops::o_setattr. + * + * This function is only used by ECHO client when it is run on top of OFD, + * \see ofd_setattr_hdl() for request handler function. + + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in] oa setattr parameters + * + * \retval 0 if successful + * \retval negative value on error + */ +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 obdo *oa = oinfo->oi_oa; - struct filter_fid *ff = NULL; + struct lu_fid *fid = &oa->o_oi.oi_fid; int rc = 0; ENTRY; info = ofd_info_init(env, exp); - ofd_oti2info(info, oti); - - rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0); - if (rc != 0) - RETURN(rc); - ost_fid_build_resid(&info->fti_fid, &info->fti_resid); - rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oa->o_oi), - oinfo_capa(oinfo), CAPA_OPC_META_WRITE); - if (rc) - GOTO(out, rc); + 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. */ - if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE && - oinfo->oi_policy.l_extent.end != OBD_OBJECT_EOF) { + if (oa->o_valid & OBD_MD_FLSIZE) { static char mdsinum[48]; - if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) + if (oa->o_valid & OBD_MD_FLFID) snprintf(mdsinum, sizeof(mdsinum) - 1, - "of parent "DFID, oinfo->oi_oa->o_parent_seq, - oinfo->oi_oa->o_parent_oid, 0); + "of parent "DFID, oa->o_parent_seq, + oa->o_parent_oid, 0); else mdsinum[0] = '\0'; CERROR("%s: setattr from %s trying to truncate object "DFID - " %s\n", exp->exp_obd->obd_name, - obd_export_nid2str(exp), PFID(&info->fti_fid), mdsinum); + " %s\n", ofd_name(ofd), obd_export_nid2str(exp), + PFID(fid), mdsinum); GOTO(out, rc = -EPERM); } - fo = ofd_object_find(env, ofd, &info->fti_fid); + fo = ofd_object_find_exists(env, ofd, fid); if (IS_ERR(fo)) { CERROR("%s: can't find object "DFID"\n", - exp->exp_obd->obd_name, PFID(&info->fti_fid)); + ofd_name(ofd), PFID(fid)); GOTO(out, rc = PTR_ERR(fo)); } - la_from_obdo(&info->fti_attr, oinfo->oi_oa, oinfo->oi_oa->o_valid); + 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); - res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0); - if (res != NULL) { - ldlm_res_lvbo_update(res, NULL, 0); - ldlm_resource_putref(res); - } - - oinfo->oi_oa->o_valid = OBD_MD_FLID; - - /* Quota release needs uid/gid info */ - rc = ofd_attr_get(env, fo, &info->fti_attr); - obdo_from_la(oinfo->oi_oa, &info->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); - ofd_info2oti(info, oti); - - ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1); + ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, NULL, 1); EXIT; out_unlock: ofd_object_put(env, fo); out: - return rc; -} - -static int ofd_punch(const struct lu_env *env, struct obd_export *exp, - struct obd_info *oinfo, struct obd_trans_info *oti, - struct ptlrpc_request_set *rqset) -{ - 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 filter_fid *ff = NULL; - int rc = 0; - - ENTRY; - - info = ofd_info_init(env, exp); - ofd_oti2info(info, oti); - - rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0); - if (rc != 0) - RETURN(rc); - ost_fid_build_resid(&info->fti_fid, &info->fti_resid); - - CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64 - ", start = "LPD64", end = "LPD64"\n", PFID(&info->fti_fid), - oinfo->oi_oa->o_valid, oinfo->oi_policy.l_extent.start, - oinfo->oi_policy.l_extent.end); - - rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi), - oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC); - if (rc) - GOTO(out_env, rc); - - fo = ofd_object_find(env, ofd, &info->fti_fid); - if (IS_ERR(fo)) { - CERROR("%s: error finding object "DFID": rc = %ld\n", - exp->exp_obd->obd_name, PFID(&info->fti_fid), - PTR_ERR(fo)); - GOTO(out_env, rc = PTR_ERR(fo)); - } - - LASSERT(oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF); - if (oinfo->oi_policy.l_extent.end == OBD_OBJECT_EOF) { - /* Truncate case */ - oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start; - } else if (oinfo->oi_policy.l_extent.end >= oinfo->oi_oa->o_size) { - oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.end; - } - - la_from_obdo(&info->fti_attr, oinfo->oi_oa, - OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME); - info->fti_attr.la_valid &= ~LA_TYPE; - info->fti_attr.la_size = oinfo->oi_policy.l_extent.start; - info->fti_attr.la_valid |= LA_SIZE; - - if (oinfo->oi_oa->o_valid & OBD_MD_FLFID) { - ff = &info->fti_mds_fid; - ofd_prepare_fidea(ff, oinfo->oi_oa); - } - - rc = ofd_object_punch(env, fo, oinfo->oi_policy.l_extent.start, - oinfo->oi_policy.l_extent.end, &info->fti_attr, - ff); - if (rc) - GOTO(out, rc); - - res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0); - if (res != NULL) { - ldlm_res_lvbo_update(res, NULL, 0); - ldlm_resource_putref(res); + if (rc == 0) { + /* 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); + if (!IS_ERR(res)) { + ldlm_res_lvbo_update(res, NULL, 0); + ldlm_resource_putref(res); + } } - oinfo->oi_oa->o_valid = OBD_MD_FLID; - /* Quota release needs uid/gid info */ - rc = ofd_attr_get(env, fo, &info->fti_attr); - obdo_from_la(oinfo->oi_oa, &info->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); - ofd_info2oti(info, oti); - - ofd_counter_incr(exp, LPROC_OFD_STATS_PUNCH, oti->oti_jobid, 1); - EXIT; -out: - ofd_object_put(env, fo); -out_env: return rc; } -static int ofd_destroy_by_fid(const struct lu_env *env, - struct ofd_device *ofd, - const struct lu_fid *fid, int orphan) +/** + * Destroy OFD object by its FID. + * + * 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_destroy() finally. + * + * \param[in] env execution environment + * \param[in] ofd OFD device + * \param[in] fid FID of object + * \param[in] orphan set if object being destroyed is an orphan + * + * \retval 0 if successful + * \retval negative value on error + */ +int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd, + const struct lu_fid *fid, int orphan) { - struct ofd_thread_info *info = ofd_info(env); - struct lustre_handle lockh; - __u64 flags = LDLM_AST_DISCARD_DATA, rc = 0; - ldlm_policy_data_t policy = { - .l_extent = { 0, OBD_OBJECT_EOF } - }; - struct ofd_object *fo; + struct ofd_thread_info *info = ofd_info(env); + struct lustre_handle lockh; + union ldlm_policy_data policy = { .l_extent = { 0, OBD_OBJECT_EOF } }; + struct ofd_object *fo; + __u64 flags = LDLM_FL_AST_DISCARD_DATA; + __u64 rc = 0; ENTRY; - fo = ofd_object_find(env, ofd, fid); + fo = ofd_object_find_exists(env, ofd, fid); if (IS_ERR(fo)) RETURN(PTR_ERR(fo)); @@ -1025,434 +1005,225 @@ static int ofd_destroy_by_fid(const struct lu_env *env, LASSERT(fo != NULL); - rc = ofd_object_destroy(env, fo, orphan); + rc = ofd_destroy(env, fo, orphan); + EXIT; ofd_object_put(env, fo); RETURN(rc); } -int ofd_destroy(const struct lu_env *env, struct obd_export *exp, - struct obdo *oa, struct lov_stripe_md *md, - struct obd_trans_info *oti, struct obd_export *md_exp, - void *capa) +/** + * Implementation of obd_ops::o_destroy. + * + * This function is only used by ECHO client when it is run on top of OFD, + * \see ofd_destroy_hdl() for request handler function. + + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in] oa obdo structure with FID + * + * Note: this is OBD API method which is common API for server OBDs and + * client OBDs. Thus some parameters used in client OBDs may not be used + * on server OBDs and vice versa. + * + * \retval 0 if successful + * \retval negative value on error + */ +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 ofd_thread_info *info; - obd_count count; + struct lu_fid *fid = &oa->o_oi.oi_fid; int rc = 0; ENTRY; - info = ofd_info_init(env, exp); - ofd_oti2info(info, oti); - - if (!(oa->o_valid & OBD_MD_FLGROUP)) - ostid_set_seq_mdt0(&oa->o_oi); + ofd_info_init(env, exp); - /* check that o_misc makes sense */ - if (oa->o_valid & OBD_MD_FLOBJCOUNT) - count = oa->o_misc; - else - count = 1; /* default case - single destroy */ + rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi)); + if (rc != 0) + RETURN(rc); - /** - * There can be sequence of objects to destroy. Therefore this request - * may have multiple transaction involved in. It is OK, we need only - * the highest used transno to be reported back in reply but not for - * replays, they must report their transno - */ - if (info->fti_transno == 0) /* not replay */ - info->fti_mult_trans = 1; - - CDEBUG(D_HA, "%s: Destroy object "DOSTID" count %d\n", ofd_name(ofd), - POSTID(&oa->o_oi), count); - while (count > 0) { - int lrc; - - lrc = ostid_to_fid(&info->fti_fid, &oa->o_oi, 0); - if (lrc != 0) { - if (rc == 0) - rc = lrc; - GOTO(out, rc); - } - lrc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 0); - if (lrc == -ENOENT) { - CDEBUG(D_INODE, - "%s: destroying non-existent object "DFID"\n", - ofd_obd(ofd)->obd_name, PFID(&info->fti_fid)); - /* rewrite rc with -ENOENT only if it is 0 */ - if (rc == 0) - rc = lrc; - } else if (lrc != 0) { - CERROR("%s: error destroying object "DFID": %d\n", - ofd_obd(ofd)->obd_name, PFID(&info->fti_fid), - rc); - rc = lrc; - } - count--; - ostid_inc_id(&oa->o_oi); - } + CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid)); - /* if we have transaction then there were some deletions, we don't - * need to return ENOENT in that case because it will not wait - * for commit of these deletions. The ENOENT must be returned only - * if there were no transations. - */ + rc = ofd_destroy_by_fid(env, ofd, fid, 0); if (rc == -ENOENT) { - if (info->fti_transno != 0) - rc = 0; + CDEBUG(D_INODE, "%s: destroying non-existent object "DFID"\n", + ofd_name(ofd), PFID(fid)); + GOTO(out, rc); } else if (rc != 0) { - /* - * If we have at least one transaction then llog record - * on server will be removed upon commit, so for rc != 0 - * we return no transno and llog record will be reprocessed. - */ - info->fti_transno = 0; + CERROR("%s: error destroying object "DFID": %d\n", + ofd_name(ofd), PFID(fid), rc); + GOTO(out, rc); } - ofd_info2oti(info, oti); + EXIT; out: - RETURN(rc); -} - -static int ofd_orphans_destroy(const struct lu_env *env, - struct obd_export *exp, struct ofd_device *ofd, - struct obdo *oa) -{ - struct ofd_thread_info *info = ofd_info(env); - obd_id last; - int skip_orphan; - int rc = 0; - struct ost_id oi = oa->o_oi; - __u64 end_id = ostid_id(&oa->o_oi); - struct ofd_seq *oseq; - - ENTRY; - - oseq = ofd_seq_get(ofd, ostid_seq(&oa->o_oi)); - if (oseq == NULL) { - CERROR("%s: Can not find seq for "DOSTID"\n", - ofd_name(ofd), POSTID(&oa->o_oi)); - RETURN(-EINVAL); - } - - LASSERT(exp != NULL); - skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN); - - last = ofd_seq_last_oid(oseq); - LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64 - " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi), - end_id + 1, last); - - for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id; - ostid_dec_id(&oi)) { - rc = ostid_to_fid(&info->fti_fid, &oi, 0); - if (rc != 0) - GOTO(out_put, rc); - rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1); - if (rc && rc != -ENOENT) /* this is pretty fatal... */ - CEMERG("error destroying precreated id "DOSTID": %d\n", - POSTID(&oi), rc); - if (!skip_orphan) { - ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1); - /* update last_id on disk periodically so that if we - * restart * we don't need to re-scan all of the just - * deleted objects. */ - if ((ostid_id(&oi) & 511) == 0) - ofd_seq_last_oid_write(env, ofd, oseq); - } - } - CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n", - ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi)); - if (!skip_orphan) { - rc = ofd_seq_last_oid_write(env, ofd, oseq); - } else { - /* don't reuse orphan object, return last used objid */ - ostid_set_id(&oa->o_oi, last); - rc = 0; - } -out_put: - ofd_seq_put(env, oseq); - RETURN(rc); + return rc; } -int ofd_create(const struct lu_env *env, struct obd_export *exp, - struct obdo *oa, struct lov_stripe_md **ea, - struct obd_trans_info *oti) +/** + * Implementation of obd_ops::o_create. + * + * This function is only used by ECHO client when it is run on top of OFD + * and just creates an object. + * \see ofd_create_hdl() for request handler function. + * + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in] oa obdo structure with FID sequence to use + * + * Note: this is OBD API method which is common API for server OBDs and + * client OBDs. Thus some parameters used in client OBDs may not be used + * on server OBDs and vice versa. + * + * \retval 0 if successful + * \retval negative value on error + */ +static int ofd_echo_create(const struct lu_env *env, struct obd_export *exp, + struct obdo *oa) { struct ofd_device *ofd = ofd_exp(exp); - struct ofd_thread_info *info; - obd_seq seq = ostid_seq(&oa->o_oi); + u64 seq = ostid_seq(&oa->o_oi); struct ofd_seq *oseq; - int rc = 0, diff; - int sync_trans = 0; + long granted; + u64 next_id; + s64 diff = 1; + int rc = 0; + int count; ENTRY; - info = ofd_info_init(env, exp); - ofd_oti2info(info, oti); + ofd_info_init(env, exp); - LASSERT(ostid_seq(&oa->o_oi) >= FID_SEQ_OST_MDT0); + LASSERT(seq == FID_SEQ_ECHO); LASSERT(oa->o_valid & OBD_MD_FLGROUP); 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, + * we may do that in the future. + * Return -ENOSPC until the LAST_ID rebuilt. */ + if (unlikely(ofd->ofd_lastid_rebuilding)) + GOTO(out_sem, rc = -ENOSPC); + + rc = ofd_validate_seq(exp, seq); + if (rc != 0) + RETURN(rc); + oseq = ofd_seq_load(env, ofd, seq); if (IS_ERR(oseq)) { - CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n", + CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n", ofd_name(ofd), seq, PTR_ERR(oseq)); - RETURN(-EINVAL); + GOTO(out_sem, rc = -EINVAL); } - if ((oa->o_valid & OBD_MD_FLFLAGS) && - (oa->o_flags & OBD_FL_RECREATE_OBJS)) { - if (!ofd_obd(ofd)->obd_recovering || - ostid_id(&oa->o_oi) > ofd_seq_last_oid(oseq)) { - CERROR("recreate objid "DOSTID" > last id "LPU64"\n", - POSTID(&oa->o_oi), ofd_seq_last_oid(oseq)); - GOTO(out_nolock, rc = -EINVAL); - } - /* do nothing because we create objects during first write */ - GOTO(out_nolock, rc = 0); - } - /* former ofd_handle_precreate */ - if ((oa->o_valid & OBD_MD_FLFLAGS) && - (oa->o_flags & OBD_FL_DELORPHAN)) { - /* destroy orphans */ - if (oti->oti_conn_cnt < exp->exp_conn_cnt) { - CERROR("%s: dropping old orphan cleanup request\n", - ofd_name(ofd)); - GOTO(out_nolock, rc = 0); - } - /* This causes inflight precreates to abort and drop lock */ - oseq->os_destroys_in_progress = 1; - mutex_lock(&oseq->os_create_lock); - if (!oseq->os_destroys_in_progress) { - CERROR("%s:["LPU64"] destroys_in_progress already" - " cleared\n", exp->exp_obd->obd_name, - ostid_seq(&oa->o_oi)); - ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq)); - GOTO(out, rc = 0); - } - diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq); - CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n", - ofd_seq_last_oid(oseq), diff); - if (-diff > OST_MAX_PRECREATE) { - /* FIXME: should reset precreate_next_id on MDS */ - rc = 0; - } else if (diff < 0) { - rc = ofd_orphans_destroy(env, exp, ofd, oa); - oseq->os_destroys_in_progress = 0; - } else { - /* XXX: Used by MDS for the first time! */ - oseq->os_destroys_in_progress = 0; - } - } else { - mutex_lock(&oseq->os_create_lock); - if (oti->oti_conn_cnt < exp->exp_conn_cnt) { - CERROR("%s: dropping old precreate request\n", - ofd_obd(ofd)->obd_name); - GOTO(out, rc = 0); - } - /* only precreate if seq is 0, IDIF or normal and also o_id - * must be specfied */ - if ((!fid_seq_is_mdt(ostid_seq(&oa->o_oi)) && - !fid_seq_is_norm(ostid_seq(&oa->o_oi)) && - !fid_seq_is_idif(ostid_seq(&oa->o_oi))) || - ostid_id(&oa->o_oi) == 0) { - diff = 1; /* shouldn't we create this right now? */ - } else { - diff = ostid_id(&oa->o_oi) - ofd_seq_last_oid(oseq); - /* Do sync create if the seq is about to used up */ - if (fid_seq_is_idif(ostid_seq(&oa->o_oi)) || - fid_seq_is_mdt0(ostid_seq(&oa->o_oi))) { - if (unlikely(ostid_id(&oa->o_oi) >= IDIF_MAX_OID - 1)) - sync_trans = 1; - } else if (fid_seq_is_norm(ostid_seq(&oa->o_oi))) { - if (unlikely(ostid_id(&oa->o_oi) >= - LUSTRE_DATA_SEQ_MAX_WIDTH - 1)) - sync_trans = 1; - } else { - CERROR("%s : invalid o_seq "DOSTID": rc = %d\n", - ofd_name(ofd), POSTID(&oa->o_oi), -EINVAL); - GOTO(out, rc = -EINVAL); - } - } + mutex_lock(&oseq->os_create_lock); + 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 (%lld): rc = %d\n", ofd_name(ofd), diff, rc); + diff = 0; + GOTO(out, rc); } - if (diff > 0) { - cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT); - obd_id next_id; - int created = 0; - int count; - - if (!(oa->o_valid & OBD_MD_FLFLAGS) || - !(oa->o_flags & OBD_FL_DELORPHAN)) { - /* don't enforce grant during orphan recovery */ - rc = ofd_grant_create(env, - ofd_obd(ofd)->obd_self_export, - &diff); - if (rc) { - CDEBUG(D_HA, "%s: failed to acquire grant " - "space for precreate (%d): rc = %d\n", - ofd_name(ofd), diff, rc); - diff = 0; - } - } - while (diff > 0) { - next_id = ofd_seq_last_oid(oseq) + 1; - count = ofd_precreate_batch(ofd, diff); - - CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64 - " at "LPU64"\n", ofd_obd(ofd)->obd_name, - count, ostid_seq(&oa->o_oi), next_id); - - if (cfs_time_after(jiffies, enough_time)) { - LCONSOLE_WARN("%s: Slow creates, %d/%d objects" - " created at a rate of %d/s\n", - ofd_obd(ofd)->obd_name, - created, diff + created, - created / DISK_TIMEOUT); - break; - } - - rc = ofd_precreate_objects(env, ofd, next_id, - oseq, count, sync_trans); - if (rc > 0) { - created += rc; - diff -= rc; - } else if (rc < 0) { - break; - } - } - if (created > 0) - /* some objects got created, we can return - * them, even if last creation failed */ - rc = 0; - else - CERROR("%s: unable to precreate: rc = %d\n", - ofd_name(ofd), rc); + next_id = ofd_seq_last_oid(oseq) + 1; + count = ofd_precreate_batch(ofd, (int)diff); - ostid_set_id(&oa->o_oi, ofd_seq_last_oid(oseq)); + 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 { + 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; - - if (!(oa->o_valid & OBD_MD_FLFLAGS) || - !(oa->o_flags & OBD_FL_DELORPHAN)) - ofd_grant_commit(env, ofd_obd(ofd)->obd_self_export, - rc); } - ofd_info2oti(info, oti); + tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc); out: mutex_unlock(&oseq->os_create_lock); -out_nolock: - if (rc == 0 && ea != NULL) { - struct lov_stripe_md *lsm = *ea; - - lsm->lsm_oi = oa->o_oi; - } ofd_seq_put(env, oseq); + +out_sem: + up_read(&ofd->ofd_lastid_rwsem); RETURN(rc); } -int ofd_getattr(const struct lu_env *env, struct obd_export *exp, - struct obd_info *oinfo) +/** + * Implementation of obd_ops::o_getattr. + * + * This function is only used by ECHO client when it is run on top of OFD + * and returns attributes of object. + * \see ofd_getattr_hdl() for request handler function. + * + * \param[in] env execution environment + * \param[in] exp OBD export of OFD device + * \param[in,out] oa contains FID of object to get attributes from and + * is used to return attributes back + * + * \retval 0 if successful + * \retval negative value on error + */ +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; - __u64 curr_version; int rc = 0; ENTRY; info = ofd_info_init(env, exp); - rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0); - if (rc != 0) - GOTO(out, rc); - rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi), - oinfo_capa(oinfo), CAPA_OPC_META_READ); - if (rc) - GOTO(out, rc); - - fo = ofd_object_find(env, ofd, &info->fti_fid); + fo = ofd_object_find_exists(env, ofd, fid); if (IS_ERR(fo)) GOTO(out, rc = PTR_ERR(fo)); + LASSERT(fo != NULL); rc = ofd_attr_get(env, fo, &info->fti_attr); - oinfo->oi_oa->o_valid = OBD_MD_FLID; - if (rc == 0) - obdo_from_la(oinfo->oi_oa, &info->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); - - /* Store object version in reply */ - curr_version = dt_version_get(env, ofd_object_child(fo)); - if ((__s64)curr_version != -EOPNOTSUPP) { - oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION; - oinfo->oi_oa->o_data_version = curr_version; - } - ofd_object_put(env, fo); -out: - RETURN(rc); -} - -static int ofd_sync(const struct lu_env *env, struct obd_export *exp, - struct obd_info *oinfo, obd_size start, obd_size end, - struct ptlrpc_request_set *set) -{ - struct ofd_device *ofd = ofd_exp(exp); - struct ofd_thread_info *info; - struct ofd_object *fo; - int rc = 0; - - ENTRY; - - /* if no objid is specified, it means "sync whole filesystem" */ - if (oinfo->oi_oa == NULL || !(oinfo->oi_oa->o_valid & OBD_MD_FLID)) { - rc = dt_sync(env, ofd->ofd_osd); - GOTO(out, rc); - } - - info = ofd_info_init(env, exp); - rc = ostid_to_fid(&info->fti_fid, &oinfo->oi_oa->o_oi, 0); - if (rc != 0) - GOTO(out, rc); - - rc = ofd_auth_capa(exp, &info->fti_fid, ostid_seq(&oinfo->oi_oa->o_oi), - oinfo_capa(oinfo), CAPA_OPC_OSS_TRUNC); - if (rc) - GOTO(out, rc); - - fo = ofd_object_find(env, ofd, &info->fti_fid); - if (IS_ERR(fo)) { - CERROR("%s: error finding object "DFID": rc = %ld\n", - exp->exp_obd->obd_name, PFID(&info->fti_fid), - PTR_ERR(fo)); - GOTO(out, rc = PTR_ERR(fo)); - } + oa->o_valid = OBD_MD_FLID; + if (rc == 0) { + __u64 curr_version; - if (!ofd_object_exists(fo)) - GOTO(put, rc = -ENOENT); + obdo_from_la(oa, &info->fti_attr, + OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); - if (dt_version_get(env, ofd_object_child(fo)) > - ofd_obd(ofd)->obd_last_committed) { - rc = dt_object_sync(env, ofd_object_child(fo)); - if (rc) - GOTO(put, rc); + /* Store object version in reply */ + curr_version = dt_version_get(env, ofd_object_child(fo)); + if ((__s64)curr_version != -EOPNOTSUPP) { + oa->o_valid |= OBD_MD_FLDATAVERSION; + oa->o_data_version = curr_version; + } } - oinfo->oi_oa->o_valid = OBD_MD_FLID; - rc = ofd_attr_get(env, fo, &info->fti_attr); - obdo_from_la(oinfo->oi_oa, &info->fti_attr, OFD_VALID_FLAGS); - - ofd_counter_incr(exp, LPROC_OFD_STATS_SYNC, oinfo->oi_jobid, 1); - EXIT; -put: ofd_object_put(env, fo); out: - return rc; + RETURN(rc); } +/** + * Get object version for OBD_IOC_GET_OBJ_VERSION ioctl. + * + * This is supplemental function for ofd_iocontrol() to return object + * version for lctl tool. + * + * \param[in] env execution environment + * \param[in] ofd OFD device + * \param[out] karg ioctl data + * + * \retval 0 if successful + * \retval negative value on error + */ static int ofd_ioc_get_obj_version(const struct lu_env *env, struct ofd_device *ofd, void *karg) { @@ -1473,11 +1244,14 @@ static int ofd_ioc_get_obj_version(const struct lu_env *env, data->ioc_inllen3 == sizeof(__u64) && data->ioc_inlbuf4 != NULL && 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_to_fid(&fid, &ostid, 0); + 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) GOTO(out, rc); } else { @@ -1510,8 +1284,23 @@ out: return rc; } -int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, - void *karg, void *uarg) +/** + * Implementation of obd_ops::o_iocontrol. + * + * This is OFD ioctl handling function which is primary interface for + * Lustre tools like lfs, lctl and lfsck. + * + * \param[in] cmd ioctl command + * \param[in] exp OBD export of OFD + * \param[in] len not used + * \param[in,out] karg buffer with data + * \param[in] uarg not used + * + * \retval 0 if successful + * \retval negative value on error + */ +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); @@ -1528,6 +1317,7 @@ int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, switch (cmd) { case OBD_IOC_ABORT_RECOVERY: CERROR("%s: aborting recovery\n", obd->obd_name); + obd->obd_abort_recovery = 1; target_stop_recovery_thread(obd); break; case OBD_IOC_SYNC: @@ -1539,11 +1329,40 @@ int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (rc == 0) rc = dt_ro(&env, ofd->ofd_osd); break; + case OBD_IOC_START_LFSCK: { + struct obd_ioctl_data *data = karg; + struct lfsck_start_param lsp; + + if (unlikely(data == NULL)) { + rc = -EINVAL; + break; + } + + lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1); + lsp.lsp_index_valid = 0; + rc = lfsck_start(&env, ofd->ofd_osd, &lsp); + break; + } + case OBD_IOC_STOP_LFSCK: { + struct obd_ioctl_data *data = karg; + struct lfsck_stop stop; + + stop.ls_status = LS_STOPPED; + /* Old lfsck utils may pass NULL @stop. */ + if (data->ioc_inlbuf1 == NULL) + stop.ls_flags = 0; + else + stop.ls_flags = + ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags; + + rc = lfsck_stop(&env, ofd->ofd_osd, &stop); + break; + } case OBD_IOC_GET_OBJ_VERSION: rc = ofd_ioc_get_obj_version(&env, ofd, karg); break; default: - CERROR("%s: not supported cmd = %d\n", obd->obd_name, cmd); + CERROR("%s: not supported cmd = %#x\n", obd->obd_name, cmd); rc = -ENOTTY; } @@ -1551,28 +1370,62 @@ int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } -static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) +/** + * Implementation of obd_ops::o_precleanup. + * + * This function stops device activity before shutting it down. It is called + * from a cleanup function upon forceful device cleanup. For OFD there are no + * special actions, it just invokes target_recovery_cleanup(). + * + * \param[in] obd OBD device of OFD + * + * \retval 0 + */ +static int ofd_precleanup(struct obd_device *obd) { - int rc = 0; - ENTRY; - - switch(stage) { - case OBD_CLEANUP_EARLY: - break; - case OBD_CLEANUP_EXPORTS: - target_cleanup_recovery(obd); - break; - } - RETURN(rc); + target_cleanup_recovery(obd); + RETURN(0); } +/** + * 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 + * incoming requests. There are two health_check methods: + * - get statfs from the OSD. It checks just responsiveness of + * bottom device + * - do write attempt on bottom device to check it is fully operational and + * is not stuck. This is expensive method and requires special configuration + * option --enable-health-write while building Lustre, it is turned off + * by default. + * + * \param[in] nul not used + * \param[in] obd OBD device of OFD + * + * \retval 0 if successful + * \retval negative value in case of error + */ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) { struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); @@ -1593,15 +1446,15 @@ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) if (unlikely(rc)) GOTO(out, rc); - if (info->fti_u.osfs.os_state == OS_STATE_READONLY) + if (info->fti_u.osfs.os_state & OS_STATE_READONLY) GOTO(out, rc = -EROFS); #ifdef USE_HEALTH_CHECK_WRITE - OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE); + OBD_ALLOC(info->fti_buf.lb_buf, PAGE_SIZE); if (info->fti_buf.lb_buf == NULL) GOTO(out, rc = -ENOMEM); - info->fti_buf.lb_len = CFS_PAGE_SIZE; + info->fti_buf.lb_len = PAGE_SIZE; info->fti_off = 0; th = dt_trans_create(&env, ofd->ofd_osd); @@ -1609,7 +1462,7 @@ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) GOTO(out, rc = PTR_ERR(th)); rc = dt_declare_record_write(&env, ofd->ofd_health_check_file, - info->fti_buf.lb_len, info->fti_off, th); + &info->fti_buf, info->fti_off, th); if (rc == 0) { th->th_sync = 1; /* sync IO is needed */ rc = dt_trans_start_local(&env, ofd->ofd_osd, th); @@ -1620,7 +1473,7 @@ static int ofd_health_check(const struct lu_env *nul, struct obd_device *obd) } dt_trans_stop(&env, ofd->ofd_osd, th); - OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE); + OBD_FREE(info->fti_buf.lb_buf, PAGE_SIZE); CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc); #endif @@ -1629,58 +1482,25 @@ out: return !!rc; } -/* - * Handle quota control requests to consult current usage/limit. - * - * \param obd - is the obd device associated with the ofd - * \param exp - is the client's export - * \param oqctl - is the obd_quotactl request to be processed - */ -static int ofd_quotactl(struct obd_device *obd, struct obd_export *exp, - struct obd_quotactl *oqctl) -{ - struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev); - struct lu_env env; - int rc; - ENTRY; - - /* report success for quota on/off for interoperability with current MDT - * stack */ - if (oqctl->qc_cmd == Q_QUOTAON || oqctl->qc_cmd == Q_QUOTAOFF) - RETURN(0); - - rc = lu_env_init(&env, LCT_DT_THREAD); - if (rc) - RETURN(rc); - - rc = lquotactl_slv(&env, ofd->ofd_osd, oqctl); - lu_env_fini(&env); - - RETURN(rc); -} - struct obd_ops ofd_obd_ops = { .o_owner = THIS_MODULE, .o_connect = ofd_obd_connect, .o_reconnect = ofd_obd_reconnect, .o_disconnect = ofd_obd_disconnect, - .o_set_info_async = ofd_set_info_async, - .o_get_info = ofd_get_info, - .o_create = ofd_create, + .o_create = ofd_echo_create, .o_statfs = ofd_statfs, - .o_setattr = ofd_setattr, + .o_setattr = ofd_echo_setattr, .o_preprw = ofd_preprw, .o_commitrw = ofd_commitrw, - .o_destroy = ofd_destroy, + .o_destroy = ofd_echo_destroy, .o_init_export = ofd_init_export, .o_destroy_export = ofd_destroy_export, .o_postrecov = ofd_obd_postrecov, - .o_punch = ofd_punch, - .o_getattr = ofd_getattr, - .o_sync = ofd_sync, + .o_getattr = ofd_echo_getattr, .o_iocontrol = ofd_iocontrol, .o_precleanup = ofd_precleanup, .o_ping = ofd_ping, .o_health_check = ofd_health_check, - .o_quotactl = ofd_quotactl, + .o_set_info_async = ofd_set_info_async, + .o_get_info = ofd_get_info, };