* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2014 Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_FILTER
-#include <lustre/lustre_idl.h>
#include "ofd_internal.h"
#include <obd_cksum.h>
-#include <lustre_ioctl.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
#include <lustre_quota.h>
#include <lustre_lfsck.h>
#include <lustre_nodemap.h>
int rc;
ENTRY;
- LASSERT(obd->obd_uses_nid_stats);
-
if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
/* Self-export gets no proc entry */
RETURN(0);
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);
+ stats->nid_stats = lprocfs_alloc_stats(LPROC_OFD_STATS_LAST,
+ LPROCFS_STATS_FLAG_NOPERCPU);
if (stats->nid_stats == NULL)
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);
}
/**
+ * 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
* 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
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,
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",
+ "https://jira.whamcloud.com/\n",
obd_export_nid2str(exp), fed->fed_group,
data->ocd_group);
RETURN(-EPROTO);
fed->fed_group = data->ocd_group;
data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
- 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)
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 (data->ocd_connect_flags & OBD_CONNECT_GRANT)
- data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
- new_connection);
+ 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;
+ }
+
+ /* 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;
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 "
"%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()) {
if (exp == NULL || obd == NULL || cluuid == NULL)
RETURN(-EINVAL);
+ rc = nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
+ if (rc != 0 && rc != -EEXIST)
+ RETURN(rc);
+
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, client_nid);
-
- nodemap_add_member(*(lnet_nid_t *)client_nid, exp);
+ else
+ nodemap_del_member(exp);
RETURN(rc);
}
struct ofd_device *ofd;
struct lustre_handle conn = { 0 };
int rc;
- lnet_nid_t *client_nid;
ENTRY;
if (_exp == NULL || obd == NULL || cluuid == NULL)
exp = class_conn2export(&conn);
LASSERT(exp != NULL);
+ 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);
+ }
+
rc = ofd_parse_connect_data(env, exp, data, true);
if (rc)
GOTO(out, rc);
- if (localdata != NULL) {
- client_nid = localdata;
- nodemap_add_member(*client_nid, exp);
- }
-
if (obd->obd_replayable) {
struct tg_export_data *ted = &exp->exp_target_data;
out:
if (rc != 0) {
- nodemap_del_member(exp);
class_disconnect(exp);
+ nodemap_del_member(exp);
*_exp = NULL;
} else {
*_exp = 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__);
- nodemap_del_member(exp);
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 &&
lu_env_fini(&env);
}
out:
+ nodemap_del_member(exp);
class_export_put(exp);
RETURN(rc);
}
{
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);
* 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(list_empty(&exp->exp_filter_data.fed_mod_list));
return 0;
int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd)
{
struct lu_device *ldev = &ofd->ofd_dt_dev.dd_lu_dev;
- struct lfsck_start_param lsp;
int rc;
CDEBUG(D_HA, "%s: recovery is over\n", ofd_name(ofd));
- 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);
+ 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);
+ }
return ldev->ld_ops->ldo_recovery_complete(env, ldev);
}
}
if (KEY_IS(KEY_SPTLRPC_CONF)) {
- rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
+ rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
} else {
CERROR("%s: Unsupported key %s\n",
exp->exp_obd->obd_name, (char*)key);
}
/**
- * 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;
-
- 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, rc = -ENOSPC);
-
- 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 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(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:
- up_read(&ofd->ofd_lastid_rwsem);
-
- return rc;
-}
-
-/**
* Implementation of obd_ops::o_statfs.
*
* This function returns information about a storage file system.
* \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);
+ 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);
* 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(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. */
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__);
- 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:
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;
ENTRY;
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);
*
* 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
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_FL_AST_DISCARD_DATA;
- __u64 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;
LASSERT(fo != NULL);
- rc = ofd_object_destroy(env, fo, orphan);
+ rc = ofd_destroy(env, fo, orphan);
EXIT;
ofd_object_put(env, fo);
ofd_info_init(env, exp);
+ rc = ofd_validate_seq(exp, ostid_seq(&oa->o_oi));
+ if (rc != 0)
+ RETURN(rc);
+
CDEBUG(D_HA, "%s: Destroy object "DFID"\n", ofd_name(ofd), PFID(fid));
rc = ofd_destroy_by_fid(env, ofd, fid, 0);
struct obdo *oa)
{
struct ofd_device *ofd = ofd_exp(exp);
- struct ofd_thread_info *info;
u64 seq = ostid_seq(&oa->o_oi);
struct ofd_seq *oseq;
- int rc = 0, diff = 1;
+ long granted;
u64 next_id;
+ s64 diff = 1;
+ int rc = 0;
int count;
ENTRY;
- info = ofd_info_init(env, exp);
+ ofd_info_init(env, exp);
LASSERT(seq == FID_SEQ_ECHO);
LASSERT(oa->o_valid & OBD_MD_FLGROUP);
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));
GOTO(out_sem, rc = -EINVAL);
}
mutex_lock(&oseq->os_create_lock);
- rc = ofd_grant_create(env, ofd_obd(ofd)->obd_self_export, &diff);
- if (rc < 0) {
+ 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);
+ "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(env, ofd_obd(ofd)->obd_self_export, rc);
+ tgt_grant_commit(ofd_obd(ofd)->obd_self_export, granted, rc);
out:
mutex_unlock(&oseq->os_create_lock);
ofd_seq_put(env, oseq);
__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));
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_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)
* \retval negative value on error
*/
static int ofd_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
- void *karg, void *uarg)
+ void *karg, void __user *uarg)
{
struct lu_env env;
struct ofd_device *ofd = ofd_exp(exp);
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:
* special actions, it just invokes target_recovery_cleanup().
*
* \param[in] obd OBD device of OFD
- * \param[in] stage cleanup stage
*
* \retval 0
*/
-static int ofd_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
+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);
}
/**
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, PAGE_CACHE_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 = PAGE_CACHE_SIZE;
+ info->fti_buf.lb_len = PAGE_SIZE;
info->fti_off = 0;
th = dt_trans_create(&env, ofd->ofd_osd);
}
dt_trans_stop(&env, ofd->ofd_osd, th);
- OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_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