*
* 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
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*
* lustre/ofd/ofd_io.c
*
- * Author: Alex Tomas <bzzz@whamcloud.com>
+ * This file provides functions to handle IO requests from clients and
+ * also LFSCK routines to check parent file identifier (PFID) consistency.
+ *
+ * Author: Alexey Zhuravlev <alexey.zhuravlev@intel.com>
+ * Author: Fan Yong <fan.yong@intel.com>
*/
#define DEBUG_SUBSYSTEM S_FILTER
+#include <linux/kthread.h>
#include "ofd_internal.h"
+#include <lustre_nodemap.h>
struct ofd_inconsistency_item {
struct list_head oii_list;
struct ofd_object *oii_obj;
- struct lu_fid oii_pfid;
+ struct filter_fid oii_ff;
};
+/**
+ * Verify single object for parent FID consistency.
+ *
+ * Part of LFSCK processing which checks single object PFID stored in extended
+ * attribute (XATTR) against real FID of MDT parent object received by LFSCK.
+ * This verifies that the OST object is being referenced by only a single MDT
+ * object.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] oii object-related local data
+ * \param[in] lrl LFSCK request data
+ */
static void ofd_inconsistency_verify_one(const struct lu_env *env,
struct ofd_device *ofd,
struct ofd_inconsistency_item *oii,
- struct lfsck_request *lr)
+ struct lfsck_req_local *lrl)
{
- struct ofd_object *fo = oii->oii_obj;
- struct lu_fid *pfid = &fo->ofo_pfid;
- int rc;
+ struct ofd_object *fo = oii->oii_obj;
+ struct filter_fid *client_ff = &oii->oii_ff;
+ struct filter_fid *local_ff = &fo->ofo_ff;
+ int rc;
LASSERT(fo->ofo_pfid_checking);
LASSERT(!fo->ofo_pfid_verified);
- lr->lr_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
- lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */
- lr->lr_fid3 = *pfid; /* OST local stored PFID. */
+ lrl->lrl_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
+ lrl->lrl_ff_client = *client_ff; /* client given PFID. */
+ lrl->lrl_ff_local = *local_ff; /* OST local stored PFID. */
- rc = lfsck_in_notify(env, ofd->ofd_osd, lr);
+ rc = lfsck_in_notify_local(env, ofd->ofd_osd, lrl, NULL);
ofd_write_lock(env, fo);
- switch (lr->lr_status) {
+ switch (lrl->lrl_status) {
case LPVS_INIT:
LASSERT(rc <= 0);
"PFID xattr for "DFID", the client given PFID "
DFID", OST local stored PFID "DFID": rc = %d\n",
ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- PFID(&oii->oii_pfid), PFID(pfid), rc);
+ PFID(&client_ff->ff_parent),
+ PFID(&local_ff->ff_parent), rc);
else
fo->ofo_pfid_verified = 1;
break;
"PFID for "DFID", the client given PFID "DFID
", local stored PFID "DFID": rc = %d\n",
ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- PFID(&oii->oii_pfid), PFID(pfid), rc);
+ PFID(&client_ff->ff_parent),
+ PFID(&local_ff->ff_parent), rc);
else
CDEBUG(D_LFSCK, "%s: both the client given PFID and "
"the OST local stored PFID are stale for the "
"OST-object "DFID", client given PFID is "DFID
", local stored PFID is "DFID"\n",
ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- PFID(&oii->oii_pfid), PFID(pfid));
+ PFID(&client_ff->ff_parent),
+ PFID(&local_ff->ff_parent));
break;
case LPVS_INCONSISTENT_TOFIX:
ofd->ofd_inconsistency_self_detected++;
"for "DFID", with the client given PFID "DFID
", the old stored PFID "DFID"\n",
ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- PFID(&oii->oii_pfid), PFID(pfid));
- } else {
+ PFID(&client_ff->ff_parent),
+ PFID(&local_ff->ff_parent));
+ } else if (rc < 0) {
CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
"for "DFID", client given PFID "DFID", local "
"stored PFID "DFID": rc = %d\n",
ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- PFID(&oii->oii_pfid), PFID(pfid), rc);
+ PFID(&client_ff->ff_parent),
+ PFID(&local_ff->ff_parent), rc);
}
- *pfid = oii->oii_pfid;
+ local_ff->ff_parent = client_ff->ff_parent;
fo->ofo_pfid_verified = 1;
break;
default:
fo->ofo_pfid_checking = 0;
ofd_write_unlock(env, fo);
- lu_object_put(env, &fo->ofo_obj.do_lu);
+ ofd_object_put(env, fo);
OBD_FREE_PTR(oii);
}
+/**
+ * Verification thread to check parent FID consistency.
+ *
+ * Kernel thread to check consistency of parent FID for any
+ * new item added for checking by ofd_add_inconsistency_item().
+ *
+ * \param[in] args OFD device
+ *
+ * \retval 0 on successful thread termination
+ * \retval negative value if thread can't start
+ */
static int ofd_inconsistency_verification_main(void *args)
{
- struct lu_env env;
- struct ofd_device *ofd = args;
- struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
+ struct lu_env env;
+ struct ofd_device *ofd = args;
+ struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
struct ofd_inconsistency_item *oii;
- struct lfsck_request *lr = NULL;
- struct l_wait_info lwi = { 0 };
- int rc;
+ struct lfsck_req_local *lrl = NULL;
+ struct l_wait_info lwi = { 0 };
+ int rc;
ENTRY;
rc = lu_env_init(&env, LCT_DT_THREAD);
spin_lock(&ofd->ofd_inconsistency_lock);
- thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
+ thread_set_flags(thread, rc ? SVC_STOPPED : SVC_RUNNING);
wake_up_all(&thread->t_ctl_waitq);
spin_unlock(&ofd->ofd_inconsistency_lock);
- if (rc != 0)
+ if (rc)
RETURN(rc);
- OBD_ALLOC_PTR(lr);
- if (unlikely(lr == NULL))
- GOTO(out, rc = -ENOMEM);
+ OBD_ALLOC_PTR(lrl);
+ if (unlikely(!lrl))
+ GOTO(out_unlocked, rc = -ENOMEM);
- lr->lr_event = LE_PAIRS_VERIFY;
- lr->lr_active = LT_LAYOUT;
+ lrl->lrl_event = LEL_PAIRS_VERIFY_LOCAL;
+ lrl->lrl_active = LFSCK_TYPE_LAYOUT;
spin_lock(&ofd->ofd_inconsistency_lock);
while (1) {
oii_list);
list_del_init(&oii->oii_list);
spin_unlock(&ofd->ofd_inconsistency_lock);
- ofd_inconsistency_verify_one(&env, ofd, oii, lr);
+ ofd_inconsistency_verify_one(&env, ofd, oii, lrl);
spin_lock(&ofd->ofd_inconsistency_lock);
}
fo->ofo_pfid_checking = 0;
ofd_write_unlock(&env, fo);
- lu_object_put(&env, &fo->ofo_obj.do_lu);
+ ofd_object_put(&env, fo);
OBD_FREE_PTR(oii);
spin_lock(&ofd->ofd_inconsistency_lock);
}
- OBD_FREE_PTR(lr);
+ OBD_FREE_PTR(lrl);
GOTO(out, rc = 0);
+out_unlocked:
+ spin_lock(&ofd->ofd_inconsistency_lock);
out:
thread_set_flags(thread, SVC_STOPPED);
wake_up_all(&thread->t_ctl_waitq);
return rc;
}
+/**
+ * Start parent FID verification thread.
+ *
+ * See ofd_inconsistency_verification_main().
+ *
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 on successful start of thread
+ * \retval negative value on error
+ */
int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd)
{
struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
return rc;
}
+/**
+ * Stop parent FID verification thread.
+ *
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 on successful start of thread
+ * \retval -EALREADY if thread is already stopped
+ */
int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
{
struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
return 0;
}
+/**
+ * Add new item for parent FID verification.
+ *
+ * Prepare new verification item and pass it to the dedicated
+ * verification thread for further processing.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] oa OBDO structure with PFID
+ */
static void ofd_add_inconsistency_item(const struct lu_env *env,
struct ofd_object *fo, struct obdo *oa)
{
- struct ofd_device *ofd = ofd_obj2dev(fo);
- struct ofd_inconsistency_item *oii;
- bool wakeup = false;
+ struct ofd_device *ofd = ofd_obj2dev(fo);
+ struct ofd_inconsistency_item *oii;
+ struct filter_fid *ff;
+ bool wakeup = false;
OBD_ALLOC_PTR(oii);
- if (oii == NULL) {
- CERROR("%s: cannot alloc memory for verify OST-object "
- "consistency for "DFID", client given PFID "DFID
- ", local stored PFID "DFID"\n",
- ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
- oa->o_parent_seq, oa->o_parent_oid, oa->o_stripe_idx,
- PFID(&fo->ofo_pfid));
-
+ if (oii == NULL)
return;
- }
INIT_LIST_HEAD(&oii->oii_list);
lu_object_get(&fo->ofo_obj.do_lu);
oii->oii_obj = fo;
- oii->oii_pfid.f_seq = oa->o_parent_seq;
- oii->oii_pfid.f_oid = oa->o_parent_oid;
- oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
+ ff = &oii->oii_ff;
+ ff->ff_parent.f_seq = oa->o_parent_seq;
+ ff->ff_parent.f_oid = oa->o_parent_oid;
+ ff->ff_parent.f_stripe_idx = oa->o_stripe_idx;
+ ff->ff_layout = oa->o_layout;
spin_lock(&ofd->ofd_inconsistency_lock);
if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
* inconsistency within the whole system. */
}
+/**
+ * Verify parent FID of an object.
+ *
+ * Check the parent FID is sane and start extended
+ * verification procedure otherwise.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] oa OBDO structure with PFID
+ *
+ * \retval 0 on successful verification
+ * \retval -EINPROGRESS if PFID is being repaired
+ * \retval -EPERM if PFID was verified but still insane
+ */
int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
struct obdo *oa)
{
- struct lu_fid *pfid = &fo->ofo_pfid;
- int rc = 0;
+ struct lu_fid *pfid = &fo->ofo_ff.ff_parent;
+ int rc = 0;
ENTRY;
if (fid_is_sane(pfid)) {
RETURN(-EINPROGRESS);
}
+/**
+ * FLR: verify the layout version of object.
+ *
+ * \param[in] env execution environment
+ * \param[in] fo OFD object
+ * \param[in] oa OBDO structure with layout version
+ *
+ * \retval 0 on successful verification
+ * \retval -EINPROGRESS layout version is in transfer
+ * \retval -ESTALE the layout version on client is stale
+ */
+int ofd_verify_layout_version(const struct lu_env *env,
+ struct ofd_object *fo, const struct obdo *oa)
+{
+ __u32 layout_version;
+ int rc;
+ ENTRY;
+
+ if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_SKIP_LV_CHECK)))
+ GOTO(out, rc = 0);
+
+ rc = ofd_object_ff_load(env, fo);
+ if (rc < 0) {
+ if (rc == -ENODATA)
+ rc = -EINPROGRESS;
+ GOTO(out, rc);
+ }
+
+ layout_version = fo->ofo_ff.ff_layout_version;
+ if (oa->o_layout_version >= layout_version &&
+ oa->o_layout_version <= layout_version + fo->ofo_ff.ff_range)
+ GOTO(out, rc = 0);
+
+ /* normal traffic, decide if to return ESTALE or EINPROGRESS */
+ layout_version &= ~LU_LAYOUT_RESYNC;
+
+ /* this update is not legitimate */
+ if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) <= layout_version)
+ GOTO(out, rc = -ESTALE);
+
+ /* layout version may not be transmitted yet */
+ if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) > layout_version)
+ GOTO(out, rc = -EINPROGRESS);
+
+ EXIT;
+
+out:
+ CDEBUG(D_INODE, DFID " verify layout version: %u vs. %u/%u, rc: %d\n",
+ PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
+ oa->o_layout_version, fo->ofo_ff.ff_layout_version,
+ fo->ofo_ff.ff_range, rc);
+ return rc;
+
+}
+
+/**
+ * Prepare buffers for read request processing.
+ *
+ * This function converts remote buffers from client to local buffers
+ * and prepares the latter.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of client
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of object
+ * \param[in] la object attributes
+ * \param[in] oa OBDO structure from client
+ * \param[in] niocount number of remote buffers
+ * \param[in] rnb remote buffers
+ * \param[in] nr_local number of local buffers
+ * \param[in] lnb local buffers
+ * \param[in] jobid job ID name
+ *
+ * \retval 0 on successful prepare
+ * \retval negative value on error
+ */
static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, const struct lu_fid *fid,
struct lu_attr *la, struct obdo *oa, int niocount,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
- struct ofd_object *fo;
- int i, j, rc, tot_bytes = 0;
+ struct ofd_object *fo;
+ int i, j, rc, tot_bytes = 0;
+ enum dt_bufs_type dbt = DT_BUFS_TYPE_READ;
ENTRY;
LASSERT(env != NULL);
GOTO(unlock, rc);
}
- /* parse remote buffers to local buffers and prepare the latter */
- *nr_local = 0;
- for (i = 0, j = 0; i < niocount; i++) {
+ if (ptlrpc_connection_is_local(exp->exp_connection))
+ dbt |= DT_BUFS_TYPE_LOCAL;
+
+ for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
- lnb + j, 0, ofd_object_capa(env, fo));
+ lnb + j, dbt);
if (unlikely(rc < 0))
GOTO(buf_put, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
}
LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
- rc = dt_attr_get(env, ofd_object_child(fo), la,
- ofd_object_capa(env, fo));
+ rc = dt_attr_get(env, ofd_object_child(fo), la);
if (unlikely(rc))
GOTO(buf_put, rc);
return rc;
}
+/**
+ * Prepare buffers for write request processing.
+ *
+ * This function converts remote buffers from client to local buffers
+ * and prepares the latter. If there is recovery in progress and required
+ * object is missing then it can be re-created before write.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of client
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of object
+ * \param[in] la object attributes
+ * \param[in] oa OBDO structure from client
+ * \param[in] objcount always 1
+ * \param[in] obj object data
+ * \param[in] rnb remote buffers
+ * \param[in] nr_local number of local buffers
+ * \param[in] lnb local buffers
+ * \param[in] jobid job ID name
+ *
+ * \retval 0 on successful prepare
+ * \retval negative value on error
+ */
static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, const struct lu_fid *fid,
struct lu_attr *la, struct obdo *oa,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
- struct ofd_object *fo;
- int i, j, k, rc = 0, tot_bytes = 0;
+ struct ofd_object *fo;
+ int i, j, k, rc = 0, tot_bytes = 0;
+ enum dt_bufs_type dbt = DT_BUFS_TYPE_WRITE;
ENTRY;
LASSERT(env != NULL);
LASSERT(objcount == 1);
if (unlikely(exp->exp_obd->obd_recovering)) {
- obd_seq seq = fid_seq(fid);
- obd_id oid = fid_oid(fid);
+ u64 seq = fid_seq(fid);
+ u64 oid = fid_oid(fid);
struct ofd_seq *oseq;
oseq = ofd_seq_load(env, ofd, seq);
if (IS_ERR(oseq)) {
- CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
+ CERROR("%s: Can't find FID Sequence %#llx: rc = %d\n",
ofd_name(ofd), seq, (int)PTR_ERR(oseq));
GOTO(out, rc = -EINVAL);
}
}
while (diff > 0) {
- obd_id next_id = ofd_seq_last_oid(oseq) + 1;
+ u64 next_id = ofd_seq_last_oid(oseq) + 1;
int count = ofd_precreate_batch(ofd, diff);
rc = ofd_precreate_objects(env, ofd, next_id,
ofd_seq_put(env, oseq);
}
+ /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
+ * space back if possible, we have to do this outside of the lock as
+ * grant preparation may need to sync whole fs thus wait for all the
+ * transactions to complete. */
+ tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
+
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo))
GOTO(out, rc = PTR_ERR(fo));
}
}
- /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
- * space back if possible */
- ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
+ /* need to verify layout version */
+ if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
+ rc = ofd_verify_layout_version(env, fo, oa);
+ if (rc) {
+ ofd_read_unlock(env, fo);
+ ofd_object_put(env, fo);
+ GOTO(out, rc);
+ }
+
+ oa->o_valid &= ~OBD_MD_LAYOUT_VERSION;
+ }
+
+ if (ptlrpc_connection_is_local(exp->exp_connection))
+ dbt |= DT_BUFS_TYPE_LOCAL;
/* parse remote buffers to local buffers and prepare the latter */
- *nr_local = 0;
- for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
+ for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo),
- rnb + i, lnb + j, 1,
- ofd_object_capa(env, fo));
+ rnb + i, lnb + j, dbt);
if (unlikely(rc < 0))
GOTO(err, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
/* correct index for local buffers to continue with */
for (k = 0; k < rc; k++) {
lnb[j+k].lnb_flags = rnb[i].rnb_flags;
+ lnb[j+k].lnb_flags &= ~OBD_BRW_LOCALS;
if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
lnb[j+k].lnb_rc = -ENOSPC;
-
- /* remote client can't break through quota */
- if (exp_connect_rmtclient(exp))
- lnb[j+k].lnb_flags &= ~OBD_BRW_NOQUOTA;
}
j += rc;
*nr_local += rc;
if (unlikely(rc != 0))
GOTO(err, rc);
+ ofd_read_unlock(env, fo);
ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
RETURN(0);
err:
dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
ofd_read_unlock(env, fo);
- /* ofd_grant_prepare_write() was called, so we must commit */
- ofd_grant_commit(env, exp, rc);
+ ofd_object_put(env, fo);
+ /* tgt_grant_prepare_write() was called, so we must commit */
+ tgt_grant_commit(exp, oa->o_grant_used, rc);
out:
/* let's still process incoming grant information packed in the oa,
* but without enforcing grant since we won't proceed with the write.
* Just like a read request actually. */
- ofd_grant_prepare_read(env, exp, oa);
+ tgt_grant_prepare_read(env, exp, oa);
return rc;
}
+/**
+ * Prepare bulk IO requests for processing.
+ *
+ * This function does initial checks of IO and calls corresponding
+ * functions for read/write processing.
+ *
+ * \param[in] env execution environment
+ * \param[in] cmd IO type (read/write)
+ * \param[in] exp OBD export of client
+ * \param[in] oa OBDO structure from request
+ * \param[in] objcount always 1
+ * \param[in] obj object data
+ * \param[in] rnb remote buffers
+ * \param[in] nr_local number of local buffers
+ * \param[in] lnb local buffers
+ *
+ * \retval 0 on successful prepare
+ * \retval negative value on error
+ */
int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
- struct niobuf_local *lnb, struct obd_trans_info *oti,
- struct lustre_capa *capa)
+ struct niobuf_local *lnb)
{
struct tgt_session_info *tsi = tgt_ses_info(env);
struct ofd_device *ofd = ofd_exp(exp);
}
if (tgt_ses_req(tsi) == NULL) { /* echo client case */
- LASSERT(oti != NULL);
info = ofd_info_init(env, exp);
- ofd_oti2info(info, oti);
jobid = NULL;
} else {
info = tsi2ofd_info(tsi);
LASSERT(oa != NULL);
- if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) {
+ if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) {
struct ofd_seq *oseq;
oseq = ofd_seq_load(env, ofd, ostid_seq(&oa->o_oi));
LASSERT(obj->ioo_bufcnt > 0);
if (cmd == OBD_BRW_WRITE) {
- rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
- capa, CAPA_OPC_OSS_WRITE);
- if (rc == 0) {
- la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
- rc = ofd_preprw_write(env, exp, ofd, fid,
- &info->fti_attr, oa, objcount,
- obj, rnb, nr_local, lnb, jobid);
- }
+ la_from_obdo(&info->fti_attr, oa, OBD_MD_FLGETATTR);
+ rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
+ objcount, obj, rnb, nr_local, lnb, jobid);
} else if (cmd == OBD_BRW_READ) {
- rc = ofd_auth_capa(exp, fid, ostid_seq(&oa->o_oi),
- capa, CAPA_OPC_OSS_READ);
- if (rc == 0) {
- ofd_grant_prepare_read(env, exp, oa);
- rc = ofd_preprw_read(env, exp, ofd, fid,
- &info->fti_attr, oa,
- obj->ioo_bufcnt, rnb, nr_local,
- lnb, jobid);
- obdo_from_la(oa, &info->fti_attr, LA_ATIME);
- }
+ tgt_grant_prepare_read(env, exp, oa);
+ rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
+ obj->ioo_bufcnt, rnb, nr_local, lnb,
+ jobid);
+ obdo_from_la(oa, &info->fti_attr, LA_ATIME);
} else {
CERROR("%s: wrong cmd %d received!\n",
exp->exp_obd->obd_name, cmd);
RETURN(rc);
}
+/**
+ * Drop reference on local buffers for read bulk IO.
+ *
+ * This will free all local buffers use by this read request.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] fid object FID
+ * \param[in] objcount always 1
+ * \param[in] niocount number of local buffers
+ * \param[in] lnb local buffers
+ *
+ * \retval 0 on successful execution
+ * \retval negative value on error
+ */
static int
ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd,
const struct lu_fid *fid, int objcount, int niocount,
RETURN(0);
}
+/**
+ * Set attributes of object during write bulk IO processing.
+ *
+ * Change object attributes and write parent FID into extended
+ * attributes when needed.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] ofd_obj OFD object
+ * \param[in] la object attributes
+ * \param[in] oa obdo
+ *
+ * \retval 0 on successful attributes update
+ * \retval negative value on error
+ */
static int
ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *ofd_obj, struct lu_attr *la,
- struct filter_fid *ff)
+ struct obdo *oa)
{
struct ofd_thread_info *info = ofd_info(env);
+ struct filter_fid *ff = &info->fti_mds_fid;
__u64 valid = la->la_valid;
- int rc;
struct thandle *th;
struct dt_object *dt_obj;
- int ff_needed = 0;
+ int fl = 0;
+ int rc;
ENTRY;
dt_obj = ofd_object_child(ofd_obj);
LASSERT(dt_obj != NULL);
- la->la_valid &= LA_UID | LA_GID;
+ la->la_valid &= LA_UID | LA_GID | LA_PROJID;
- rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
+ rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(out, rc);
- if (ff != NULL) {
- rc = ofd_object_ff_load(env, ofd_obj);
- if (rc == -ENODATA)
- ff_needed = 1;
- else if (rc < 0)
- GOTO(out, rc);
- }
-
- if (!la->la_valid && !ff_needed)
+ if (!la->la_valid && !(oa->o_valid &
+ (OBD_MD_FLFID | OBD_MD_FLOSTLAYOUT | OBD_MD_LAYOUT_VERSION)))
/* no attributes to set */
GOTO(out, rc = 0);
GOTO(out_tx, rc);
}
- if (ff_needed) {
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
- ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
- ff->ff_parent.f_oid =
- cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
-
- info->fti_buf.lb_buf = ff;
- info->fti_buf.lb_len = sizeof(*ff);
- rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
- XATTR_NAME_FID, 0, th);
- if (rc)
- GOTO(out_tx, rc);
- }
+ rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
+ XATTR_NAME_FID, 0, th);
+ if (rc)
+ GOTO(out_tx, rc);
/* We don't need a transno for this operation which will be re-executed
* anyway when the OST_WRITE (with a transno assigned) is replayed */
if (rc)
GOTO(out_tx, rc);
- /* set uid/gid */
+ ofd_read_lock(env, ofd_obj);
+
+ /* set uid/gid/projid */
if (la->la_valid) {
- rc = dt_attr_set(env, dt_obj, la, th,
- ofd_object_capa(env, ofd_obj));
+ rc = dt_attr_set(env, dt_obj, la, th);
if (rc)
- GOTO(out_tx, rc);
- }
-
- /* set filter fid EA */
- if (ff_needed) {
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
- GOTO(out_tx, rc);
-
- rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
- 0, th, BYPASS_CAPA);
- if (rc == 0) {
- ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
- ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
- /* Currently, the filter_fid::ff_parent::f_ver is not
- * the real parent MDT-object's FID::f_ver, instead it
- * is the OST-object index in its parent MDT-object's
- * layout EA. */
- ofd_obj->ofo_pfid.f_stripe_idx =
- le32_to_cpu(ff->ff_parent.f_stripe_idx);
- }
+ GOTO(out_unlock, rc);
}
- GOTO(out_tx, rc);
-
+ fl = ofd_object_ff_update(env, ofd_obj, oa, ff);
+ if (fl <= 0)
+ GOTO(out_unlock, rc = fl);
+
+ /* set filter fid EA.
+ * FIXME: it holds read lock of ofd object to modify the XATTR_NAME_FID
+ * while the write lock should be held. However, it should work because
+ * write RPCs only modify ff_{parent,layout} and those information will
+ * be the same from all the write RPCs. The reason that fl is not used
+ * in dt_xattr_set() is to allow this race. */
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
+ GOTO(out_unlock, rc);
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
+ ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
+ else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
+ le32_add_cpu(&ff->ff_parent.f_oid, -1);
+
+ info->fti_buf.lb_buf = ff;
+ info->fti_buf.lb_len = sizeof(*ff);
+ rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID, 0, th);
+ if (rc == 0)
+ filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
+
+ GOTO(out_unlock, rc);
+
+out_unlock:
+ ofd_read_unlock(env, ofd_obj);
out_tx:
dt_trans_stop(env, ofd->ofd_osd, th);
out:
struct obd_export *ossc_exp;
};
+/**
+ * Callback function for "soft sync" update.
+ *
+ * Reset fed_soft_sync_count upon committing the "soft_sync" update.
+ * See ofd_soft_sync_cb_add() below for more details on soft sync.
+ *
+ * \param[in] env execution environment
+ * \param[in] th transaction handle
+ * \param[in] cb callback data
+ * \param[in] err error code
+ */
static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th,
struct dt_txn_commit_cb *cb, int err)
{
OBD_FREE_PTR(ossc);
}
+/**
+ * Add callback for "soft sync" processing.
+ *
+ * The "soft sync" mechanism does asynchronous commit when OBD_BRW_SOFT_SYNC
+ * flag is set in client buffers. The intention is for this operation to
+ * commit pages belonging to a client which has "too many" outstanding
+ * unstable pages in its cache. See LU-2139 for details.
+ *
+ * This function adds callback to be called when commit is done.
+ *
+ * \param[in] th transaction handle
+ * \param[in] exp OBD export of client
+ *
+ * \retval 0 on successful callback adding
+ * \retval negative value on error
+ */
static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp)
{
struct ofd_soft_sync_callback *ossc;
dcb = &ossc->ossc_cb;
dcb->dcb_func = ofd_cb_soft_sync;
- CFS_INIT_LIST_HEAD(&dcb->dcb_linkage);
- strncpy(dcb->dcb_name, "ofd_cb_soft_sync", MAX_COMMIT_CB_STR_LEN);
- dcb->dcb_name[MAX_COMMIT_CB_STR_LEN - 1] = '\0';
+ INIT_LIST_HEAD(&dcb->dcb_linkage);
+ strlcpy(dcb->dcb_name, "ofd_cb_soft_sync", sizeof(dcb->dcb_name));
rc = dt_trans_cb_add(th, dcb);
if (rc) {
return rc;
}
+/**
+ * Commit bulk IO buffers to the storage.
+ *
+ * This function finalizes write IO processing by writing data to the disk.
+ * That write can be synchronous or asynchronous depending on buffers flags.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export of client
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of object
+ * \param[in] la object attributes
+ * \param[in] ff parent FID of object
+ * \param[in] objcount always 1
+ * \param[in] niocount number of local buffers
+ * \param[in] lnb local buffers
+ * \param[in] granted grant space consumed for the bulk I/O
+ * \param[in] old_rc result of processing at this point
+ *
+ * \retval 0 on successful commit
+ * \retval negative value on error
+ */
static int
ofd_commitrw_write(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, const struct lu_fid *fid,
- struct lu_attr *la, struct filter_fid *ff, int objcount,
- int niocount, struct niobuf_local *lnb, int old_rc)
+ struct lu_attr *la, struct obdo *oa, int objcount,
+ int niocount, struct niobuf_local *lnb,
+ unsigned long granted, int old_rc)
{
- struct ofd_thread_info *info = ofd_info(env);
- struct ofd_object *fo;
- struct dt_object *o;
- struct thandle *th;
- int rc = 0;
- int retries = 0;
- int i;
struct filter_export_data *fed = &exp->exp_filter_data;
- bool soft_sync = false;
- bool cb_registered = false;
+ struct ofd_object *fo;
+ struct dt_object *o;
+ struct thandle *th;
+ int rc = 0;
+ int rc2 = 0;
+ int retries = 0;
+ int i;
+ bool soft_sync = false;
+ bool cb_registered = false;
+ bool fake_write = false;
ENTRY;
fo = ofd_object_find(env, ofd, fid);
LASSERT(fo != NULL);
- LASSERT(ofd_object_exists(fo));
o = ofd_object_child(fo);
LASSERT(o != NULL);
if (old_rc)
GOTO(out, rc = old_rc);
+ if (!ofd_object_exists(fo))
+ GOTO(out, rc = -ENOENT);
/*
* The first write to each object must set some attributes. It is
* dt_declare_write_commit() since quota enforcement is now handled in
* declare phases.
*/
- rc = ofd_write_attr_set(env, ofd, fo, la, ff);
+ rc = ofd_write_attr_set(env, ofd, fo, la, oa);
if (rc)
GOTO(out, rc);
la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
+ /* do fake write, to simulate the write case for performance testing */
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) {
+ struct niobuf_local *last = &lnb[niocount - 1];
+ __u64 file_size = last->lnb_file_offset + last->lnb_len;
+ __u64 valid = la->la_valid;
+
+ la->la_valid = LA_SIZE;
+ la->la_size = 0;
+ rc = dt_attr_get(env, o, la);
+ if (rc < 0 && rc != -ENOENT)
+ GOTO(out, rc);
+
+ if (file_size < la->la_size)
+ file_size = la->la_size;
+
+ /* dirty inode by setting file size */
+ la->la_valid = valid | LA_SIZE;
+ la->la_size = file_size;
+
+ fake_write = true;
+ }
+
retry:
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
- th->th_sync |= ofd->ofd_syncjournal;
+ th->th_sync |= ofd->ofd_sync_journal;
if (th->th_sync == 0) {
for (i = 0; i < niocount; i++) {
if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
GOTO(out_stop, rc = -EINPROGRESS);
- rc = dt_declare_write_commit(env, o, lnb, niocount, th);
- if (rc)
- GOTO(out_stop, rc);
+ if (likely(!fake_write)) {
+ rc = dt_declare_write_commit(env, o, lnb, niocount, th);
+ if (rc)
+ GOTO(out_stop, rc);
+ }
if (la->la_valid) {
/* update [mac]time if needed */
if (rc)
GOTO(out_stop, rc);
- rc = dt_write_commit(env, o, lnb, niocount, th);
- if (rc)
- GOTO(out_stop, rc);
+ ofd_read_lock(env, fo);
+ if (!ofd_object_exists(fo))
+ GOTO(out_unlock, rc = -ENOENT);
+
+ if (likely(!fake_write)) {
+ rc = dt_write_commit(env, o, lnb, niocount, th);
+ if (rc)
+ GOTO(out_unlock, rc);
+ }
if (la->la_valid) {
- rc = dt_attr_set(env, o, la, th, ofd_object_capa(env, fo));
+ rc = dt_attr_set(env, o, la, th);
if (rc)
- GOTO(out_stop, rc);
+ GOTO(out_unlock, rc);
}
/* get attr to return */
- rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo));
+ rc = dt_attr_get(env, o, la);
+out_unlock:
+ ofd_read_unlock(env, fo);
out_stop:
/* Force commit to make the just-deleted blocks
* reusable. LU-456 */
cb_registered = true;
}
- ofd_trans_stop(env, ofd, th, rc);
+ if (rc == 0 && granted > 0) {
+ if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
+ granted = 0;
+ }
+
+ rc2 = ofd_trans_stop(env, ofd, th, rc);
+ if (!rc)
+ rc = rc2;
if (rc == -ENOSPC && retries++ < 3) {
CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
retries);
out:
dt_bufs_put(env, o, lnb, niocount);
- ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
/* second put is pair to object_get in ofd_preprw_write */
ofd_object_put(env, fo);
- ofd_grant_commit(env, info->fti_exp, old_rc);
+ if (granted > 0)
+ tgt_grant_commit(exp, granted, old_rc);
RETURN(rc);
}
+/**
+ * Commit bulk IO to the storage.
+ *
+ * This is companion function to the ofd_preprw(). It finishes bulk IO
+ * request processing by committing buffers to the storage (WRITE) and/or
+ * freeing those buffers (read/write). See ofd_commitrw_read() and
+ * ofd_commitrw_write() for details about each type of IO.
+ *
+ * \param[in] env execution environment
+ * \param[in] cmd IO type (READ/WRITE)
+ * \param[in] exp OBD export of client
+ * \param[in] oa OBDO structure from client
+ * \param[in] objcount always 1
+ * \param[in] obj object data
+ * \param[in] rnb remote buffers
+ * \param[in] npages number of local buffers
+ * \param[in] lnb local buffers
+ * \param[in] old_rc result of processing at this point
+ *
+ * \retval 0 on successful commit
+ * \retval negative value on error
+ */
int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int npages,
- struct niobuf_local *lnb, struct obd_trans_info *oti,
- int old_rc)
+ struct niobuf_local *lnb, int old_rc)
{
- struct ofd_thread_info *info = ofd_info(env);
- struct ofd_mod_data *fmd;
- __u64 valid;
- struct ofd_device *ofd = ofd_exp(exp);
- struct filter_fid *ff = NULL;
- const struct lu_fid *fid = &oa->o_oi.oi_fid;
- int rc = 0;
+ struct ofd_thread_info *info = ofd_info(env);
+ struct ofd_device *ofd = ofd_exp(exp);
+ const struct lu_fid *fid = &oa->o_oi.oi_fid;
+ __u64 valid;
+ int rc = 0;
LASSERT(npages > 0);
if (cmd == OBD_BRW_WRITE) {
+ struct lu_nodemap *nodemap;
+
/* Don't update timestamps if this write is older than a
* setattr which modifies the timestamps. b=10150 */
-
- /* XXX when we start having persistent reservations this needs
- * to be changed to ofd_fmd_get() to create the fmd if it
- * doesn't already exist so we can store the reservation handle
- * there. */
- valid = OBD_MD_FLUID | OBD_MD_FLGID;
- fmd = ofd_fmd_find(exp, fid);
- if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
+ valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
+ if (tgt_fmd_check(exp, fid, info->fti_xid))
valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
OBD_MD_FLCTIME;
- ofd_fmd_put(exp, fmd);
- la_from_obdo(&info->fti_attr, oa, valid);
- if (oa->o_valid & OBD_MD_FLFID) {
- ff = &info->fti_mds_fid;
- ofd_prepare_fidea(ff, oa);
- }
+ la_from_obdo(&info->fti_attr, oa, valid);
rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr,
- ff, objcount, npages, lnb, old_rc);
+ oa, objcount, npages, lnb,
+ oa->o_grant_used, old_rc);
if (rc == 0)
obdo_from_la(oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_GID | LA_UID);
+ OFD_VALID_FLAGS | LA_GID | LA_UID |
+ LA_PROJID);
else
- obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
+ obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
+ LA_PROJID);
/* don't report overquota flag if we failed before reaching
* commit */
else
oa->o_flags = OBD_FL_NO_GRPQUOTA;
}
+ if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
+ if (oa->o_valid & OBD_MD_FLFLAGS)
+ oa->o_flags |= OBD_FL_NO_PRJQUOTA;
+ else
+ oa->o_flags = OBD_FL_NO_PRJQUOTA;
+ }
oa->o_valid |= OBD_MD_FLFLAGS;
- oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
+ oa->o_valid |= OBD_MD_FLALLQUOTA;
+ }
+
+ /* Convert back to client IDs. LU-9671.
+ * nodemap_get_from_exp() may fail due to nodemap deactivated,
+ * server ID will be returned back to client in that case. */
+ nodemap = nodemap_get_from_exp(exp);
+ if (nodemap != NULL && !IS_ERR(nodemap)) {
+ oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
+ NODEMAP_FS_TO_CLIENT,
+ oa->o_uid);
+ oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
+ NODEMAP_FS_TO_CLIENT,
+ oa->o_gid);
+ nodemap_putref(nodemap);
}
} else if (cmd == OBD_BRW_READ) {
struct ldlm_namespace *ns = ofd->ofd_namespace;
ost_fid_build_resid(fid, &info->fti_resid);
rs = ldlm_resource_get(ns, NULL, &info->fti_resid,
LDLM_EXTENT, 0);
- if (rs != NULL) {
- ns->ns_lvbo->lvbo_update(rs, NULL, 1);
+ if (!IS_ERR(rs)) {
+ ldlm_res_lvbo_update(rs, NULL, 1);
ldlm_resource_putref(rs);
}
}
rc = -EPROTO;
}
- if (oti != NULL)
- ofd_info2oti(info, oti);
RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff