X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_io.c;h=0608cd99e4445f5e33e1c0a506930501a9ada229;hb=b4d4a448b2ed93a8bc73db81bbf9c8ba08250e3a;hp=599e59f2449d35ca60251e910f4ca520b252c617;hpb=4c4f74b0a581e139f3bf3db1a41d2b11594c039f;p=fs%2Flustre-release.git diff --git a/lustre/ofd/ofd_io.c b/lustre/ofd/ofd_io.c index 599e59f..0608cd9 100644 --- a/lustre/ofd/ofd_io.c +++ b/lustre/ofd/ofd_io.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, 2013, Intel Corporation. + * Copyright (c) 2012, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,11 +31,16 @@ * * lustre/ofd/ofd_io.c * - * Author: Alex Tomas + * This file provides functions to handle IO requests from clients and + * also LFSCK routines to check parent file identifier (PFID) consistency. + * + * Author: Alexey Zhuravlev + * Author: Fan Yong */ #define DEBUG_SUBSYSTEM S_FILTER +#include #include "ofd_internal.h" struct ofd_inconsistency_item { @@ -48,6 +49,19 @@ struct ofd_inconsistency_item { struct lu_fid oii_pfid; }; +/** + * 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] lr LFSCK request data + */ static void ofd_inconsistency_verify_one(const struct lu_env *env, struct ofd_device *ofd, struct ofd_inconsistency_item *oii, @@ -64,7 +78,7 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env, lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */ lr->lr_fid3 = *pfid; /* OST local stored PFID. */ - rc = lfsck_in_notify(env, ofd->ofd_osd, lr); + rc = lfsck_in_notify(env, ofd->ofd_osd, lr, NULL); ofd_write_lock(env, fo); switch (lr->lr_status) { case LPVS_INIT: @@ -106,7 +120,7 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env, ", the old stored PFID "DFID"\n", ofd_name(ofd), PFID(&fo->ofo_header.loh_fid), PFID(&oii->oii_pfid), PFID(pfid)); - } else { + } 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", @@ -126,6 +140,17 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env, 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; @@ -147,7 +172,7 @@ static int ofd_inconsistency_verification_main(void *args) OBD_ALLOC_PTR(lr); if (unlikely(lr == NULL)) - GOTO(out, rc = -ENOMEM); + GOTO(out_unlocked, rc = -ENOMEM); lr->lr_event = LE_PAIRS_VERIFY; lr->lr_active = LFSCK_TYPE_LAYOUT; @@ -198,6 +223,8 @@ static int ofd_inconsistency_verification_main(void *args) 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); @@ -207,6 +234,16 @@ out: 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; @@ -240,6 +277,14 @@ int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd) 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; @@ -262,6 +307,16 @@ int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd) 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) { @@ -270,16 +325,8 @@ static void ofd_add_inconsistency_item(const struct lu_env *env, 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); @@ -314,6 +361,20 @@ static void ofd_add_inconsistency_item(const struct lu_env *env, * 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) { @@ -355,6 +416,27 @@ int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo, RETURN(-EINPROGRESS); } +/** + * 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, @@ -382,11 +464,10 @@ static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp, GOTO(unlock, rc); } - /* parse remote buffers to local buffers and prepare the latter */ *nr_local = 0; for (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, 0); if (unlikely(rc < 0)) GOTO(buf_put, rc); LASSERT(rc <= PTLRPC_MAX_BRW_PAGES); @@ -398,8 +479,7 @@ static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp, } 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); @@ -418,6 +498,29 @@ unlock: 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, @@ -433,8 +536,8 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, 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); @@ -468,7 +571,7 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, } 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, @@ -519,8 +622,7 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, *nr_local = 0; for (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, 1); if (unlikely(rc < 0)) GOTO(err, rc); LASSERT(rc <= PTLRPC_MAX_BRW_PAGES); @@ -550,8 +652,9 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp, err: dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local); ofd_read_unlock(env, fo); + ofd_object_put(env, fo); /* ofd_grant_prepare_write() was called, so we must commit */ - ofd_grant_commit(env, exp, rc); + ofd_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. @@ -560,11 +663,29 @@ out: 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); @@ -581,9 +702,7 @@ int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *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); @@ -592,7 +711,7 @@ int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp, 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)); @@ -617,25 +736,15 @@ int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp, 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); - } + 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); } else { CERROR("%s: wrong cmd %d received!\n", exp->exp_obd->obd_name, cmd); @@ -644,6 +753,21 @@ int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp, 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, @@ -670,6 +794,21 @@ ofd_commitrw_read(const struct lu_env *env, struct ofd_device *ofd, 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] ff parent FID + * + * \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, @@ -740,8 +879,7 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd, /* set uid/gid */ 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); } @@ -752,7 +890,7 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd, GOTO(out_tx, rc); rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID, - 0, th, BYPASS_CAPA); + 0, th); 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); @@ -779,6 +917,17 @@ struct ofd_soft_sync_callback { 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) { @@ -793,6 +942,22 @@ static void ofd_cb_soft_sync(struct lu_env *env, struct thandle *th, 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; @@ -807,7 +972,7 @@ static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp) dcb = &ossc->ossc_cb; dcb->dcb_func = ofd_cb_soft_sync; - CFS_INIT_LIST_HEAD(&dcb->dcb_linkage); + 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); @@ -819,13 +984,34 @@ static int ofd_soft_sync_cb_add(struct thandle *th, struct obd_export *exp) 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) + 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; @@ -902,13 +1088,13 @@ retry: GOTO(out_stop, 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); } /* get attr to return */ - rc = dt_attr_get(env, o, la, ofd_object_capa(env, fo)); + rc = dt_attr_get(env, o, la); out_stop: /* Force commit to make the just-deleted blocks @@ -922,6 +1108,11 @@ out_stop: cb_registered = true; } + if (rc == 0 && granted > 0) { + if (ofd_grant_commit_cb_add(th, exp, granted) == 0) + granted = 0; + } + ofd_trans_stop(env, ofd, th, rc); if (rc == -ENOSPC && retries++ < 3) { CDEBUG(D_INODE, "retry after force commit, retries:%d\n", @@ -942,15 +1133,37 @@ out: 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) + ofd_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; @@ -984,7 +1197,8 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, } rc = ofd_commitrw_write(env, exp, ofd, fid, &info->fti_attr, - ff, objcount, npages, lnb, old_rc); + ff, 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); @@ -1025,7 +1239,7 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, rs = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0); if (!IS_ERR(rs)) { - ns->ns_lvbo->lvbo_update(rs, NULL, 1); + ldlm_res_lvbo_update(rs, NULL, 1); ldlm_resource_putref(rs); } } @@ -1038,7 +1252,5 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp, rc = -EPROTO; } - if (oti != NULL) - ofd_info2oti(info, oti); RETURN(rc); }