X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fllog_cat.c;h=c5b15f1df527bc399e5ca12f8947adda805a7331;hp=cc065c14437de6829f989fce0d9d2c1aa028fe21;hb=7f4a1dd92a900a55cbb980bac3d20c32968334ec;hpb=7e253000b0feb3fd7580d7f8991ee377973b862f diff --git a/lustre/obdclass/llog_cat.c b/lustre/obdclass/llog_cat.c index cc065c1..c5b15f1 100644 --- a/lustre/obdclass/llog_cat.c +++ b/lustre/obdclass/llog_cat.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) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel, Inc. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -48,14 +44,20 @@ #define DEBUG_SUBSYSTEM S_LOG -#ifndef __KERNEL__ -#include -#endif #include #include "llog_internal.h" + +/** + * lockdep markers for nested struct llog_handle::lgh_lock locking. + */ +enum { + LLOGH_CAT, + LLOGH_LOG, +}; + /* Create a new log handle and add it to the open list. * This log handle will be closed when all of the records in it are removed. * @@ -66,92 +68,305 @@ static int llog_cat_new_log(const struct lu_env *env, struct llog_handle *loghandle, struct thandle *th) { + struct llog_thread_info *lgi = llog_info(env); + struct llog_logid_rec *rec = &lgi->lgi_logid; + struct thandle *handle = NULL; + struct dt_device *dt = NULL; + struct llog_log_hdr *llh = cathandle->lgh_hdr; + int rc, index; - struct llog_log_hdr *llh; - struct llog_logid_rec rec = { { 0 }, }; - int rc, index, bitmap_size; - ENTRY; - - llh = cathandle->lgh_hdr; - bitmap_size = LLOG_BITMAP_SIZE(llh); - - index = (cathandle->lgh_last_idx + 1) % bitmap_size; + ENTRY; - /* maximum number of available slots in catlog is bitmap_size - 2 */ - if (llh->llh_cat_idx == index) { - CERROR("no free catalog slots for log...\n"); + index = (cathandle->lgh_last_idx + 1) % + (OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS) ? (cfs_fail_val + 1) : + LLOG_HDR_BITMAP_SIZE(llh)); + + /* check that new llog index will not overlap with the first one. + * - llh_cat_idx is the index just before the first/oldest still in-use + * index in catalog + * - lgh_last_idx is the last/newest used index in catalog + * + * When catalog is not wrapped yet then lgh_last_idx is always larger + * than llh_cat_idx. After the wrap around lgh_last_idx re-starts + * from 0 and llh_cat_idx becomes the upper limit for it + * + * Check if catalog has already wrapped around or not by comparing + * last_idx and cat_idx */ + if ((index == llh->llh_cat_idx + 1 && llh->llh_count > 1) || + (index == 0 && llh->llh_cat_idx == 0)) { + if (cathandle->lgh_name == NULL) { + CWARN("%s: there are no more free slots in catalog " + DFID":%x\n", + loghandle->lgh_ctxt->loc_obd->obd_name, + PFID(&cathandle->lgh_id.lgl_oi.oi_fid), + cathandle->lgh_id.lgl_ogen); + } else { + CWARN("%s: there are no more free slots in " + "catalog %s\n", + loghandle->lgh_ctxt->loc_obd->obd_name, + cathandle->lgh_name); + } RETURN(-ENOSPC); } if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED)) RETURN(-ENOSPC); + if (loghandle->lgh_hdr != NULL) { + /* If llog object is remote and creation is failed, lgh_hdr + * might be left over here, free it first */ + LASSERT(!llog_exist(loghandle)); + OBD_FREE_LARGE(loghandle->lgh_hdr, loghandle->lgh_hdr_size); + loghandle->lgh_hdr = NULL; + } + + if (th == NULL) { + dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev); + + handle = dt_trans_create(env, dt); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); + + /* Create update llog object synchronously, which + * happens during inialization process see + * lod_sub_prep_llog(), to make sure the update + * llog object is created before corss-MDT writing + * updates into the llog object */ + if (cathandle->lgh_ctxt->loc_flags & LLOG_CTXT_FLAG_NORMAL_FID) + handle->th_sync = 1; + + handle->th_wait_submit = 1; + + rc = llog_declare_create(env, loghandle, handle); + if (rc != 0) + GOTO(out, rc); + + rec->lid_hdr.lrh_len = sizeof(*rec); + rec->lid_hdr.lrh_type = LLOG_LOGID_MAGIC; + rec->lid_id = loghandle->lgh_id; + rc = llog_declare_write_rec(env, cathandle, &rec->lid_hdr, -1, + handle); + if (rc != 0) + GOTO(out, rc); + + rc = dt_trans_start_local(env, dt, handle); + if (rc != 0) + GOTO(out, rc); + + th = handle; + } + rc = llog_create(env, loghandle, th); /* if llog is already created, no need to initialize it */ if (rc == -EEXIST) { - RETURN(0); + GOTO(out, rc = 0); } else if (rc != 0) { CERROR("%s: can't create new plain llog in catalog: rc = %d\n", loghandle->lgh_ctxt->loc_obd->obd_name, rc); - RETURN(rc); + GOTO(out, rc); } rc = llog_init_handle(env, loghandle, - LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY, - &cathandle->lgh_hdr->llh_tgtuuid); - if (rc) - GOTO(out_destroy, rc); - - if (index == 0) - index = 1; - - cfs_spin_lock(&loghandle->lgh_hdr_lock); - llh->llh_count++; - if (ext2_set_bit(index, llh->llh_bitmap)) { - CERROR("argh, index %u already set in log bitmap?\n", - index); - cfs_spin_unlock(&loghandle->lgh_hdr_lock); - LBUG(); /* should never happen */ - } - cfs_spin_unlock(&loghandle->lgh_hdr_lock); - - cathandle->lgh_last_idx = index; - llh->llh_tail.lrt_index = index; - - CDEBUG(D_RPCTRACE,"new recovery log "LPX64":%x for index %u of catalog " - LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen, - index, cathandle->lgh_id.lgl_oid); - /* build the record for this log in the catalog */ - rec.lid_hdr.lrh_len = sizeof(rec); - rec.lid_hdr.lrh_index = index; - rec.lid_hdr.lrh_type = LLOG_LOGID_MAGIC; - rec.lid_id = loghandle->lgh_id; - rec.lid_tail.lrt_len = sizeof(rec); - rec.lid_tail.lrt_index = index; - - /* update the catalog: header and record */ - rc = llog_write_rec(env, cathandle, &rec.lid_hdr, - &loghandle->u.phd.phd_cookie, 1, NULL, index, th); + LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY, + &cathandle->lgh_hdr->llh_tgtuuid); + if (rc < 0) + GOTO(out, rc); + + /* build the record for this log in the catalog */ + rec->lid_hdr.lrh_len = sizeof(*rec); + rec->lid_hdr.lrh_type = LLOG_LOGID_MAGIC; + rec->lid_id = loghandle->lgh_id; + + /* append the new record into catalog. The new index will be + * assigned to the record and updated in rec header */ + rc = llog_write_rec(env, cathandle, &rec->lid_hdr, + &loghandle->u.phd.phd_cookie, LLOG_NEXT_IDX, th); if (rc < 0) GOTO(out_destroy, rc); - loghandle->lgh_hdr->llh_cat_idx = index; - RETURN(0); + CDEBUG(D_OTHER, "new plain log "DFID".%u of catalog "DFID"\n", + PFID(&loghandle->lgh_id.lgl_oi.oi_fid), rec->lid_hdr.lrh_index, + PFID(&cathandle->lgh_id.lgl_oi.oi_fid)); + + loghandle->lgh_hdr->llh_cat_idx = rec->lid_hdr.lrh_index; + + /* limit max size of plain llog so that space can be + * released sooner, especially on small filesystems */ + /* 2MB for the cases when free space hasn't been learned yet */ + loghandle->lgh_max_size = 2 << 20; + dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev); + rc = dt_statfs(env, dt, &lgi->lgi_statfs); + if (rc == 0 && lgi->lgi_statfs.os_bfree > 0) { + __u64 freespace = (lgi->lgi_statfs.os_bfree * + lgi->lgi_statfs.os_bsize) >> 6; + if (freespace < loghandle->lgh_max_size) + loghandle->lgh_max_size = freespace; + /* shouldn't be > 128MB in any case? + * it's 256K records of 512 bytes each */ + if (freespace > (128 << 20)) + loghandle->lgh_max_size = 128 << 20; + } + rc = 0; + +out: + if (handle != NULL) { + handle->th_result = rc >= 0 ? 0 : rc; + dt_trans_stop(env, dt, handle); + } + RETURN(rc); + out_destroy: - llog_destroy(env, loghandle); + /* to signal llog_cat_close() it shouldn't try to destroy the llog, + * we want to destroy it in this transaction, otherwise the object + * becomes an orphan */ + loghandle->lgh_hdr->llh_flags &= ~LLOG_F_ZAP_WHEN_EMPTY; + /* this is to mimic full log, so another llog_cat_current_log() + * can skip it and ask for another onet */ + loghandle->lgh_last_idx = LLOG_HDR_BITMAP_SIZE(loghandle->lgh_hdr) + 1; + llog_trans_destroy(env, loghandle, th); + if (handle != NULL) + dt_trans_stop(env, dt, handle); RETURN(rc); } +static int llog_cat_refresh(const struct lu_env *env, + struct llog_handle *cathandle) +{ + struct llog_handle *loghandle; + int rc; + + down_write(&cathandle->lgh_lock); + list_for_each_entry(loghandle, &cathandle->u.chd.chd_head, + u.phd.phd_entry) { + if (!llog_exist(loghandle)) + continue; + + rc = llog_read_header(env, loghandle, NULL); + if (rc) + goto unlock; + } + + rc = llog_read_header(env, cathandle, NULL); +unlock: + up_write(&loghandle->lgh_lock); + + return rc; +} + +/* + * prepare current/next log for catalog. + * + * if \a *ploghandle is NULL, open it, and declare create, NB, if \a + * *ploghandle is remote, create it synchronously here, see comments + * below. + * + * \a cathandle->lgh_lock is down_read-ed, it gets down_write-ed if \a + * *ploghandle has to be opened. + */ +static int llog_cat_prep_log(const struct lu_env *env, + struct llog_handle *cathandle, + struct llog_handle **ploghandle, + struct thandle *th) +{ + int rc; + int sem_upgraded; + +start: + rc = 0; + sem_upgraded = 0; + if (IS_ERR_OR_NULL(*ploghandle)) { + up_read(&cathandle->lgh_lock); + down_write(&cathandle->lgh_lock); + sem_upgraded = 1; + if (IS_ERR_OR_NULL(*ploghandle)) { + struct llog_handle *loghandle; + + rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, + NULL, NULL, LLOG_OPEN_NEW); + if (!rc) { + *ploghandle = loghandle; + list_add_tail(&loghandle->u.phd.phd_entry, + &cathandle->u.chd.chd_head); + } + } + if (rc) + GOTO(out, rc); + } + + rc = llog_exist(*ploghandle); + if (rc < 0) + GOTO(out, rc); + if (rc) + GOTO(out, rc = 0); + + if (dt_object_remote(cathandle->lgh_obj)) { + down_write_nested(&(*ploghandle)->lgh_lock, LLOGH_LOG); + if (!llog_exist(*ploghandle)) { + /* For remote operation, if we put the llog object + * creation in the current transaction, then the + * llog object will not be created on the remote + * target until the transaction stop, if other + * operations start before the transaction stop, + * and use the same llog object, will be dependent + * on the success of this transaction. So let's + * create the llog object synchronously here to + * remove the dependency. */ + rc = llog_cat_new_log(env, cathandle, *ploghandle, + NULL); + if (rc == -ESTALE) { + up_write(&(*ploghandle)->lgh_lock); + if (sem_upgraded) + up_write(&cathandle->lgh_lock); + else + up_read(&cathandle->lgh_lock); + + rc = llog_cat_refresh(env, cathandle); + down_read_nested(&cathandle->lgh_lock, + LLOGH_CAT); + if (rc) + return rc; + /* *ploghandle might become NULL, restart */ + goto start; + } + } + up_write(&(*ploghandle)->lgh_lock); + } else { + struct llog_thread_info *lgi = llog_info(env); + struct llog_logid_rec *lirec = &lgi->lgi_logid; + + rc = llog_declare_create(env, *ploghandle, th); + if (rc) + GOTO(out, rc); + + lirec->lid_hdr.lrh_len = sizeof(*lirec); + rc = llog_declare_write_rec(env, cathandle, &lirec->lid_hdr, -1, + th); + } + +out: + if (sem_upgraded) { + up_write(&cathandle->lgh_lock); + down_read_nested(&cathandle->lgh_lock, LLOGH_CAT); + if (rc == 0) + goto start; + } + return rc; +} + /* Open an existent log handle and add it to the open list. * This log handle will be closed when all of the records in it are removed. * * Assumes caller has already pushed us into the kernel context and is locking. * We return a lock on the handle to ensure nobody yanks it from us. + * + * This takes extra reference on llog_handle via llog_handle_get() and require + * this reference to be put by caller using llog_handle_put() */ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle, struct llog_handle **res, struct llog_logid *logid) { struct llog_handle *loghandle; + enum llog_flag fmt; int rc = 0; ENTRY; @@ -159,44 +374,47 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle, if (cathandle == NULL) RETURN(-EBADF); - cfs_down_write(&cathandle->lgh_lock); - cfs_list_for_each_entry(loghandle, &cathandle->u.chd.chd_head, - u.phd.phd_entry) { + fmt = cathandle->lgh_hdr->llh_flags & LLOG_F_EXT_MASK; + down_write(&cathandle->lgh_lock); + list_for_each_entry(loghandle, &cathandle->u.chd.chd_head, + u.phd.phd_entry) { struct llog_logid *cgl = &loghandle->lgh_id; - if (cgl->lgl_oid == logid->lgl_oid) { + if (ostid_id(&cgl->lgl_oi) == ostid_id(&logid->lgl_oi) && + ostid_seq(&cgl->lgl_oi) == ostid_seq(&logid->lgl_oi)) { if (cgl->lgl_ogen != logid->lgl_ogen) { - CERROR("%s: log "LPX64" generation %x != %x\n", - loghandle->lgh_ctxt->loc_obd->obd_name, - logid->lgl_oid, cgl->lgl_ogen, - logid->lgl_ogen); + CWARN("%s: log "DFID" generation %x != %x\n", + loghandle->lgh_ctxt->loc_obd->obd_name, + PFID(&logid->lgl_oi.oi_fid), + cgl->lgl_ogen, logid->lgl_ogen); continue; } loghandle->u.phd.phd_cat_handle = cathandle; - cfs_up_write(&cathandle->lgh_lock); + up_write(&cathandle->lgh_lock); GOTO(out, rc = 0); } } - cfs_up_write(&cathandle->lgh_lock); + up_write(&cathandle->lgh_lock); rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, logid, NULL, LLOG_OPEN_EXISTS); if (rc < 0) { - CERROR("%s: error opening log id "LPX64":%x: rc = %d\n", + CERROR("%s: error opening log id "DFID":%x: rc = %d\n", cathandle->lgh_ctxt->loc_obd->obd_name, - logid->lgl_oid, logid->lgl_ogen, rc); - GOTO(out, rc); + PFID(&logid->lgl_oi.oi_fid), logid->lgl_ogen, rc); + RETURN(rc); } - rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN | fmt, NULL); if (rc < 0) { llog_close(env, loghandle); - GOTO(out, rc); + loghandle = NULL; + RETURN(rc); } - cfs_down_write(&cathandle->lgh_lock); - cfs_list_add(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head); - cfs_up_write(&cathandle->lgh_lock); + down_write(&cathandle->lgh_lock); + list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head); + up_write(&cathandle->lgh_lock); loghandle->u.phd.phd_cat_handle = cathandle; loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id; @@ -204,8 +422,9 @@ int llog_cat_id2handle(const struct lu_env *env, struct llog_handle *cathandle, loghandle->lgh_hdr->llh_cat_idx; EXIT; out: + llog_handle_get(loghandle); *res = loghandle; - return rc; + return 0; } int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle) @@ -215,13 +434,13 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle) ENTRY; - cfs_list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head, - u.phd.phd_entry) { + list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head, + u.phd.phd_entry) { struct llog_log_hdr *llh = loghandle->lgh_hdr; int index; /* unlink open-not-created llogs */ - cfs_list_del_init(&loghandle->u.phd.phd_entry); + list_del_init(&loghandle->u.phd.phd_entry); llh = loghandle->lgh_hdr; if (loghandle->lgh_obj != NULL && llh != NULL && (llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) && @@ -234,15 +453,7 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle) rc); index = loghandle->u.phd.phd_cookie.lgc_index; - - LASSERT(index); - llog_cat_set_first_idx(cathandle, index); - rc = llog_cancel_rec(env, cathandle, index); - if (rc == 0) - CDEBUG(D_RPCTRACE, - "cancel plain log at index %u of " - "catalog "LPX64"\n", - index, cathandle->lgh_id.lgl_oid); + llog_cat_cleanup(env, cathandle, NULL, index); } llog_close(env, loghandle); } @@ -254,14 +465,6 @@ int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle) } EXPORT_SYMBOL(llog_cat_close); -/** - * lockdep markers for nested struct llog_handle::lgh_lock locking. - */ -enum { - LLOGH_CAT, - LLOGH_LOG -}; - /** Return the currently active log handle. If the current log handle doesn't * have enough space left for the current record, start a new one. * @@ -278,49 +481,73 @@ static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle, struct llog_handle *loghandle = NULL; ENTRY; - cfs_down_read_nested(&cathandle->lgh_lock, LLOGH_CAT); + + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED2)) { + down_write_nested(&cathandle->lgh_lock, LLOGH_CAT); + GOTO(next, loghandle); + } + + down_read_nested(&cathandle->lgh_lock, LLOGH_CAT); loghandle = cathandle->u.chd.chd_current_log; if (loghandle) { struct llog_log_hdr *llh; - cfs_down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); + down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); llh = loghandle->lgh_hdr; - if (llh == NULL || - loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) { - cfs_up_read(&cathandle->lgh_lock); + if (llh == NULL || !llog_is_full(loghandle)) { + up_read(&cathandle->lgh_lock); RETURN(loghandle); } else { - cfs_up_write(&loghandle->lgh_lock); + up_write(&loghandle->lgh_lock); } } - cfs_up_read(&cathandle->lgh_lock); + up_read(&cathandle->lgh_lock); /* time to use next log */ /* first, we have to make sure the state hasn't changed */ - cfs_down_write_nested(&cathandle->lgh_lock, LLOGH_CAT); + down_write_nested(&cathandle->lgh_lock, LLOGH_CAT); loghandle = cathandle->u.chd.chd_current_log; if (loghandle) { struct llog_log_hdr *llh; - cfs_down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); + down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); llh = loghandle->lgh_hdr; - LASSERT(llh); - if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) { - cfs_up_write(&cathandle->lgh_lock); - RETURN(loghandle); - } else { - cfs_up_write(&loghandle->lgh_lock); - } - } + if (llh == NULL || !llog_is_full(loghandle)) + GOTO(out_unlock, loghandle); + else + up_write(&loghandle->lgh_lock); + } + +next: + /* Sigh, the chd_next_log and chd_current_log is initialized + * in declare phase, and we do not serialize the catlog + * accessing, so it might be possible the llog creation + * thread (see llog_cat_declare_add_rec()) did not create + * llog successfully, then the following thread might + * meet this situation. */ + if (IS_ERR_OR_NULL(cathandle->u.chd.chd_next_log)) { + CERROR("%s: next log does not exist!\n", + cathandle->lgh_ctxt->loc_obd->obd_name); + loghandle = ERR_PTR(-EIO); + if (cathandle->u.chd.chd_next_log == NULL) { + /* Store the error in chd_next_log, so + * the following process can get correct + * failure value */ + cathandle->u.chd.chd_next_log = loghandle; + } + GOTO(out_unlock, loghandle); + } CDEBUG(D_INODE, "use next log\n"); loghandle = cathandle->u.chd.chd_next_log; cathandle->u.chd.chd_current_log = loghandle; cathandle->u.chd.chd_next_log = NULL; - cfs_down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); - cfs_up_write(&cathandle->lgh_lock); + down_write_nested(&loghandle->lgh_lock, LLOGH_LOG); + +out_unlock: + up_write(&cathandle->lgh_lock); LASSERT(loghandle); RETURN(loghandle); } @@ -332,47 +559,53 @@ static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle, */ int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle, struct llog_rec_hdr *rec, struct llog_cookie *reccookie, - void *buf, struct thandle *th) + struct thandle *th) { struct llog_handle *loghandle; - int rc; - ENTRY; + int rc, retried = 0; + ENTRY; + + LASSERT(rec->lrh_len <= cathandle->lgh_ctxt->loc_chunk_size); - LASSERT(rec->lrh_len <= LLOG_CHUNK_SIZE); +retry: loghandle = llog_cat_current_log(cathandle, th); - LASSERT(!IS_ERR(loghandle)); + if (IS_ERR(loghandle)) + RETURN(PTR_ERR(loghandle)); /* loghandle is already locked by llog_cat_current_log() for us */ if (!llog_exist(loghandle)) { rc = llog_cat_new_log(env, cathandle, loghandle, th); if (rc < 0) { - cfs_up_write(&loghandle->lgh_lock); + up_write(&loghandle->lgh_lock); + /* nobody should be trying to use this llog */ + down_write(&cathandle->lgh_lock); + if (cathandle->u.chd.chd_current_log == loghandle) + cathandle->u.chd.chd_current_log = NULL; + up_write(&cathandle->lgh_lock); RETURN(rc); } } /* now let's try to add the record */ - rc = llog_write_rec(env, loghandle, rec, reccookie, 1, buf, -1, th); - if (rc < 0) - CERROR("llog_write_rec %d: lh=%p\n", rc, loghandle); - cfs_up_write(&loghandle->lgh_lock); - if (rc == -ENOSPC) { - /* try to use next log */ - loghandle = llog_cat_current_log(cathandle, th); - LASSERT(!IS_ERR(loghandle)); - /* new llog can be created concurrently */ - if (!llog_exist(loghandle)) { - rc = llog_cat_new_log(env, cathandle, loghandle, th); - if (rc < 0) { - cfs_up_write(&loghandle->lgh_lock); - RETURN(rc); - } - } - /* now let's try to add the record */ - rc = llog_write_rec(env, loghandle, rec, reccookie, 1, buf, - -1, th); - if (rc < 0) - CERROR("llog_write_rec %d: lh=%p\n", rc, loghandle); - cfs_up_write(&loghandle->lgh_lock); + rc = llog_write_rec(env, loghandle, rec, reccookie, LLOG_NEXT_IDX, th); + if (rc < 0) { + CDEBUG_LIMIT(rc == -ENOSPC ? D_HA : D_ERROR, + "llog_write_rec %d: lh=%p\n", rc, loghandle); + /* -ENOSPC is returned if no empty records left + * and when it's lack of space on the stogage. + * there is no point to try again if it's the second + * case. many callers (like llog test) expect ENOSPC, + * so we preserve this error code, but look for the + * actual cause here */ + if (rc == -ENOSPC && llog_is_full(loghandle)) + rc = -ENOBUFS; + } + up_write(&loghandle->lgh_lock); + + if (rc == -ENOBUFS) { + if (retried++ == 0) + GOTO(retry, rc); + CERROR("%s: error on 2nd llog: rc = %d\n", + cathandle->lgh_ctxt->loc_obd->obd_name, rc); } RETURN(rc); @@ -383,70 +616,49 @@ int llog_cat_declare_add_rec(const struct lu_env *env, struct llog_handle *cathandle, struct llog_rec_hdr *rec, struct thandle *th) { - struct llog_handle *loghandle, *next; - int rc = 0; + int rc; ENTRY; - if (cathandle->u.chd.chd_current_log == NULL) { - /* declare new plain llog */ - cfs_down_write(&cathandle->lgh_lock); - if (cathandle->u.chd.chd_current_log == NULL) { - rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, - NULL, NULL, LLOG_OPEN_NEW); - if (rc == 0) { - cathandle->u.chd.chd_current_log = loghandle; - cfs_list_add_tail(&loghandle->u.phd.phd_entry, - &cathandle->u.chd.chd_head); - } - } - cfs_up_write(&cathandle->lgh_lock); - } else if (cathandle->u.chd.chd_next_log == NULL) { - /* declare next plain llog */ - cfs_down_write(&cathandle->lgh_lock); - if (cathandle->u.chd.chd_next_log == NULL) { - rc = llog_open(env, cathandle->lgh_ctxt, &loghandle, - NULL, NULL, LLOG_OPEN_NEW); - if (rc == 0) { - cathandle->u.chd.chd_next_log = loghandle; - cfs_list_add_tail(&loghandle->u.phd.phd_entry, - &cathandle->u.chd.chd_head); - } - } - cfs_up_write(&cathandle->lgh_lock); - } +start: + down_read_nested(&cathandle->lgh_lock, LLOGH_CAT); + rc = llog_cat_prep_log(env, cathandle, + &cathandle->u.chd.chd_current_log, th); if (rc) - GOTO(out, rc); + GOTO(unlock, rc); - if (!llog_exist(cathandle->u.chd.chd_current_log)) { - rc = llog_declare_create(env, cathandle->u.chd.chd_current_log, - th); - if (rc) - GOTO(out, rc); - llog_declare_write_rec(env, cathandle, NULL, -1, th); - } - /* declare records in the llogs */ - rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log, - rec, -1, th); + rc = llog_cat_prep_log(env, cathandle, &cathandle->u.chd.chd_next_log, + th); if (rc) - GOTO(out, rc); + GOTO(unlock, rc); - next = cathandle->u.chd.chd_next_log; - if (next) { - if (!llog_exist(next)) { - rc = llog_declare_create(env, next, th); - llog_declare_write_rec(env, cathandle, NULL, -1, th); - } - llog_declare_write_rec(env, next, rec, -1, th); + rc = llog_declare_write_rec(env, cathandle->u.chd.chd_current_log, + rec, -1, th); + if (rc == -ESTALE && dt_object_remote(cathandle->lgh_obj)) { + up_read(&cathandle->lgh_lock); + rc = llog_cat_refresh(env, cathandle); + if (rc) + RETURN(rc); + goto start; } -out: + +#if 0 + /* + * XXX: we hope for declarations made for existing llog this might be + * not correct with some backends where declarations are expected + * against specific object like ZFS with full debugging enabled. + */ + rc = llog_declare_write_rec(env, cathandle->u.chd.chd_next_log, rec, -1, + th); +#endif +unlock: + up_read(&cathandle->lgh_lock); RETURN(rc); } EXPORT_SYMBOL(llog_cat_declare_add_rec); int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle, - struct llog_rec_hdr *rec, struct llog_cookie *reccookie, - void *buf) + struct llog_rec_hdr *rec, struct llog_cookie *reccookie) { struct llog_ctxt *ctxt; struct dt_device *dt; @@ -457,35 +669,84 @@ int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle, LASSERT(ctxt); LASSERT(ctxt->loc_exp); - if (cathandle->lgh_obj != NULL) { - dt = ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt; - LASSERT(dt); + LASSERT(cathandle->lgh_obj != NULL); + dt = lu2dt_dev(cathandle->lgh_obj->do_lu.lo_dev); - th = dt_trans_create(env, dt); - if (IS_ERR(th)) - RETURN(PTR_ERR(th)); + th = dt_trans_create(env, dt); + if (IS_ERR(th)) + RETURN(PTR_ERR(th)); - rc = llog_cat_declare_add_rec(env, cathandle, rec, th); - if (rc) - GOTO(out_trans, rc); + rc = llog_cat_declare_add_rec(env, cathandle, rec, th); + if (rc) + GOTO(out_trans, rc); - rc = dt_trans_start_local(env, dt, th); - if (rc) - GOTO(out_trans, rc); - rc = llog_cat_add_rec(env, cathandle, rec, reccookie, buf, th); + rc = dt_trans_start_local(env, dt, th); + if (rc) + GOTO(out_trans, rc); + rc = llog_cat_add_rec(env, cathandle, rec, reccookie, th); out_trans: - dt_trans_stop(env, dt, th); - } else { /* lvfs compat code */ - LASSERT(cathandle->lgh_file != NULL); - rc = llog_cat_declare_add_rec(env, cathandle, rec, th); - if (rc == 0) - rc = llog_cat_add_rec(env, cathandle, rec, reccookie, - buf, th); - } + dt_trans_stop(env, dt, th); RETURN(rc); } EXPORT_SYMBOL(llog_cat_add); +int llog_cat_cancel_arr_rec(const struct lu_env *env, + struct llog_handle *cathandle, + struct llog_logid *lgl, int count, int *index) +{ + struct llog_handle *loghandle; + int rc; + + ENTRY; + rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl); + if (rc) { + CDEBUG(D_HA, "%s: cannot find llog for handle "DFID":%x" + ": rc = %d\n", + cathandle->lgh_ctxt->loc_obd->obd_name, + PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc); + RETURN(rc); + } + + if ((cathandle->lgh_ctxt->loc_flags & + LLOG_CTXT_FLAG_NORMAL_FID) && !llog_exist(loghandle)) { + /* For update log, some of loghandles of cathandle + * might not exist because remote llog creation might + * be failed, so let's skip the record cancellation + * for these non-exist llogs. + */ + rc = -ENOENT; + CDEBUG(D_HA, "%s: llog "DFID":%x does not exist" + ": rc = %d\n", + cathandle->lgh_ctxt->loc_obd->obd_name, + PFID(&lgl->lgl_oi.oi_fid), lgl->lgl_ogen, rc); + + llog_handle_put(loghandle); + RETURN(rc); + } + + rc = llog_cancel_arr_rec(env, loghandle, count, index); + if (rc == LLOG_DEL_PLAIN) { /* log has been destroyed */ + int cat_index; + + cat_index = loghandle->u.phd.phd_cookie.lgc_index; + rc = llog_cat_cleanup(env, cathandle, loghandle, cat_index); + if (rc) + CERROR("%s: fail to cancel catalog record: rc = %d\n", + cathandle->lgh_ctxt->loc_obd->obd_name, rc); + rc = 0; + + } + llog_handle_put(loghandle); + + if (rc) + CERROR("%s: fail to cancel %d llog-records: rc = %d\n", + cathandle->lgh_ctxt->loc_obd->obd_name, count, + rc); + + RETURN(rc); +} +EXPORT_SYMBOL(llog_cat_cancel_arr_rec); + /* For each cookie in the cookie array, we clear the log in-use bit and either: * - the log is empty, so mark it free in the catalog header and delete it * - the log is not empty, just write out the log header @@ -499,241 +760,300 @@ int llog_cat_cancel_records(const struct lu_env *env, struct llog_handle *cathandle, int count, struct llog_cookie *cookies) { - int i, index, rc = 0, failed = 0; + int i, rc = 0, failed = 0; ENTRY; for (i = 0; i < count; i++, cookies++) { - struct llog_handle *loghandle; - struct llog_logid *lgl = &cookies->lgc_lgl; - int lrc; - - rc = llog_cat_id2handle(env, cathandle, &loghandle, lgl); - if (rc) { - CERROR("Cannot find log "LPX64"\n", lgl->lgl_oid); - break; - } + int lrc; - lrc = llog_cancel_rec(env, loghandle, cookies->lgc_index); - if (lrc == 1) { /* log has been destroyed */ - index = loghandle->u.phd.phd_cookie.lgc_index; - cfs_down_write(&cathandle->lgh_lock); - if (cathandle->u.chd.chd_current_log == loghandle) - cathandle->u.chd.chd_current_log = NULL; - cfs_up_write(&cathandle->lgh_lock); - llog_close(env, loghandle); - - LASSERT(index); - llog_cat_set_first_idx(cathandle, index); - lrc = llog_cancel_rec(env, cathandle, index); - if (lrc == 0) - CDEBUG(D_RPCTRACE, "cancel plain log at index" - " %u of catalog "LPX64"\n", - index, cathandle->lgh_id.lgl_oid); - } else if (lrc == -ENOENT) { - if (rc == 0) /* ENOENT shouldn't rewrite any error */ - rc = lrc; - } else if (lrc < 0) { + lrc = llog_cat_cancel_arr_rec(env, cathandle, &cookies->lgc_lgl, + 1, &cookies->lgc_index); + if (lrc) { failed++; - rc = lrc; + if (!rc) + rc = lrc; } } - if (rc) + if (failed) CERROR("%s: fail to cancel %d of %d llog-records: rc = %d\n", cathandle->lgh_ctxt->loc_obd->obd_name, failed, count, rc); - RETURN(rc); } EXPORT_SYMBOL(llog_cat_cancel_records); -int llog_cat_process_cb(const struct lu_env *env, struct llog_handle *cat_llh, - struct llog_rec_hdr *rec, void *data) +static int llog_cat_process_common(const struct lu_env *env, + struct llog_handle *cat_llh, + struct llog_rec_hdr *rec, + struct llog_handle **llhp) { - struct llog_process_data *d = data; - struct llog_logid_rec *lir = (struct llog_logid_rec *)rec; - struct llog_handle *llh; - int rc; + struct llog_logid_rec *lir = container_of(rec, typeof(*lir), lid_hdr); + struct llog_log_hdr *hdr; + int rc; - ENTRY; - if (rec->lrh_type != LLOG_LOGID_MAGIC) { - CERROR("invalid record in catalog\n"); - RETURN(-EINVAL); - } - CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog " - LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen, - rec->lrh_index, cat_llh->lgh_id.lgl_oid); - - rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id); - if (rc) { - CERROR("Cannot find handle for log "LPX64"\n", - lir->lid_id.lgl_oid); - RETURN(rc); - } + ENTRY; + if (rec->lrh_type != le32_to_cpu(LLOG_LOGID_MAGIC)) { + rc = -EINVAL; + CWARN("%s: invalid record in catalog "DFID":%x: rc = %d\n", + cat_llh->lgh_ctxt->loc_obd->obd_name, + PFID(&cat_llh->lgh_id.lgl_oi.oi_fid), + cat_llh->lgh_id.lgl_ogen, rc); + RETURN(rc); + } + CDEBUG(D_HA, "processing log "DFID":%x at index %u of catalog "DFID"\n", + PFID(&lir->lid_id.lgl_oi.oi_fid), lir->lid_id.lgl_ogen, + le32_to_cpu(rec->lrh_index), + PFID(&cat_llh->lgh_id.lgl_oi.oi_fid)); + + rc = llog_cat_id2handle(env, cat_llh, llhp, &lir->lid_id); + if (rc) { + /* After a server crash, a stub of index record in catlog could + * be kept, because plain log destroy + catlog index record + * deletion are not atomic. So we end up with an index but no + * actual record. Destroy the index and move on. */ + if (rc == -ENOENT || rc == -ESTALE) + rc = LLOG_DEL_RECORD; + else if (rc) + CWARN("%s: can't find llog handle "DFID":%x: rc = %d\n", + cat_llh->lgh_ctxt->loc_obd->obd_name, + PFID(&lir->lid_id.lgl_oi.oi_fid), + lir->lid_id.lgl_ogen, rc); + + RETURN(rc); + } - if (rec->lrh_index < d->lpd_startcat) - /* Skip processing of the logs until startcat */ - RETURN(0); + /* clean old empty llogs, do not consider current llog in use */ + /* ignore remote (lgh_obj == NULL) llogs */ + hdr = (*llhp)->lgh_hdr; + if ((hdr->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) && + hdr->llh_count == 1 && cat_llh->lgh_obj != NULL && + *llhp != cat_llh->u.chd.chd_current_log) { + rc = llog_destroy(env, *llhp); + if (rc) + CWARN("%s: can't destroy empty log "DFID": rc = %d\n", + (*llhp)->lgh_ctxt->loc_obd->obd_name, + PFID(&lir->lid_id.lgl_oi.oi_fid), rc); + rc = LLOG_DEL_PLAIN; + } - if (d->lpd_startidx > 0) { + RETURN(rc); +} + +static int llog_cat_process_cb(const struct lu_env *env, + struct llog_handle *cat_llh, + struct llog_rec_hdr *rec, void *data) +{ + struct llog_process_data *d = data; + struct llog_handle *llh = NULL; + int rc; + + ENTRY; + rc = llog_cat_process_common(env, cat_llh, rec, &llh); + if (rc) + GOTO(out, rc); + + if (rec->lrh_index < d->lpd_startcat) { + /* Skip processing of the logs until startcat */ + rc = 0; + } else if (d->lpd_startidx > 0) { struct llog_process_cat_data cd; cd.lpcd_first_idx = d->lpd_startidx; cd.lpcd_last_idx = 0; rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data, &cd, false); - /* Continue processing the next log from idx 0 */ - d->lpd_startidx = 0; - } else { + /* Continue processing the next log from idx 0 */ + d->lpd_startidx = 0; + } else { rc = llog_process_or_fork(env, llh, d->lpd_cb, d->lpd_data, NULL, false); - } + } - RETURN(rc); +out: + /* The empty plain log was destroyed while processing */ + if (rc == LLOG_DEL_PLAIN) { + rc = llog_cat_cleanup(env, cat_llh, llh, + llh->u.phd.phd_cookie.lgc_index); + } else if (rc == LLOG_DEL_RECORD) { + /* clear wrong catalog entry */ + rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index); + } + + if (llh) + llog_handle_put(llh); + + RETURN(rc); } int llog_cat_process_or_fork(const struct lu_env *env, - struct llog_handle *cat_llh, + struct llog_handle *cat_llh, llog_cb_t cat_cb, llog_cb_t cb, void *data, int startcat, int startidx, bool fork) { - struct llog_process_data d; - struct llog_log_hdr *llh = cat_llh->lgh_hdr; - int rc; - ENTRY; - - LASSERT(llh->llh_flags & LLOG_F_IS_CAT); - d.lpd_data = data; - d.lpd_cb = cb; - d.lpd_startcat = startcat; - d.lpd_startidx = startidx; - - if (llh->llh_cat_idx > cat_llh->lgh_last_idx) { - struct llog_process_cat_data cd; - - CWARN("catlog "LPX64" crosses index zero\n", - cat_llh->lgh_id.lgl_oid); + struct llog_process_data d; + struct llog_log_hdr *llh = cat_llh->lgh_hdr; + int rc; - cd.lpcd_first_idx = llh->llh_cat_idx; - cd.lpcd_last_idx = 0; - rc = llog_process_or_fork(env, cat_llh, llog_cat_process_cb, - &d, &cd, fork); - if (rc != 0) - RETURN(rc); + ENTRY; - cd.lpcd_first_idx = 0; - cd.lpcd_last_idx = cat_llh->lgh_last_idx; - rc = llog_process_or_fork(env, cat_llh, llog_cat_process_cb, + LASSERT(llh->llh_flags & LLOG_F_IS_CAT); + d.lpd_data = data; + d.lpd_cb = cb; + d.lpd_startcat = (startcat == LLOG_CAT_FIRST ? 0 : startcat); + d.lpd_startidx = startidx; + + if (llh->llh_cat_idx >= cat_llh->lgh_last_idx && + llh->llh_count > 1) { + struct llog_process_cat_data cd; + + CWARN("%s: catlog "DFID" crosses index zero\n", + cat_llh->lgh_ctxt->loc_obd->obd_name, + PFID(&cat_llh->lgh_id.lgl_oi.oi_fid)); + /*startcat = 0 is default value for general processing */ + if ((startcat != LLOG_CAT_FIRST && + startcat >= llh->llh_cat_idx) || !startcat) { + /* processing the catalog part at the end */ + cd.lpcd_first_idx = (startcat ? startcat : + llh->llh_cat_idx); + if (OBD_FAIL_PRECHECK(OBD_FAIL_CAT_RECORDS)) + cd.lpcd_last_idx = cfs_fail_val; + else + cd.lpcd_last_idx = 0; + rc = llog_process_or_fork(env, cat_llh, cat_cb, + &d, &cd, fork); + /* Reset the startcat becasue it has already reached + * catalog bottom. + */ + startcat = 0; + if (rc != 0) + RETURN(rc); + } + /* processing the catalog part at the begining */ + cd.lpcd_first_idx = (startcat == LLOG_CAT_FIRST) ? 0 : startcat; + /* Note, the processing will stop at the lgh_last_idx value, + * and it could be increased during processing. So records + * between current lgh_last_idx and lgh_last_idx in future + * would left unprocessed. + */ + cd.lpcd_last_idx = cat_llh->lgh_last_idx; + rc = llog_process_or_fork(env, cat_llh, cat_cb, &d, &cd, fork); - } else { - rc = llog_process_or_fork(env, cat_llh, llog_cat_process_cb, + } else { + rc = llog_process_or_fork(env, cat_llh, cat_cb, &d, NULL, fork); - } + } - RETURN(rc); + RETURN(rc); } EXPORT_SYMBOL(llog_cat_process_or_fork); int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh, llog_cb_t cb, void *data, int startcat, int startidx) { - return llog_cat_process_or_fork(env, cat_llh, cb, data, startcat, - startidx, false); + return llog_cat_process_or_fork(env, cat_llh, llog_cat_process_cb, + cb, data, startcat, startidx, false); } EXPORT_SYMBOL(llog_cat_process); -#ifdef __KERNEL__ -int llog_cat_process_thread(void *data) +static int llog_cat_size_cb(const struct lu_env *env, + struct llog_handle *cat_llh, + struct llog_rec_hdr *rec, void *data) { - struct llog_process_cat_args *args = data; - struct llog_ctxt *ctxt = args->lpca_ctxt; - struct llog_handle *llh = NULL; - llog_cb_t cb = args->lpca_cb; - struct llog_thread_info *lgi; - struct lu_env env; - int rc; - ENTRY; - - cfs_daemonize_ctxt("ll_log_process"); + struct llog_process_data *d = data; + struct llog_handle *llh = NULL; + __u64 *cum_size = d->lpd_data; + __u64 size; + int rc; - rc = lu_env_init(&env, LCT_LOCAL); - if (rc) - GOTO(out, rc); - lgi = llog_info(&env); - LASSERT(lgi); + ENTRY; + rc = llog_cat_process_common(env, cat_llh, rec, &llh); + + if (rc == LLOG_DEL_PLAIN) { + /* empty log was deleted, don't count it */ + rc = llog_cat_cleanup(env, cat_llh, llh, + llh->u.phd.phd_cookie.lgc_index); + } else if (rc == LLOG_DEL_RECORD) { + /* clear wrong catalog entry */ + rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index); + } else { + size = llog_size(env, llh); + *cum_size += size; - lgi->lgi_logid = *(struct llog_logid *)(args->lpca_arg); - rc = llog_open(&env, ctxt, &llh, &lgi->lgi_logid, NULL, - LLOG_OPEN_EXISTS); - if (rc) { - CERROR("%s: cannot open llog "LPX64":%x: rc = %d\n", - ctxt->loc_obd->obd_name, lgi->lgi_logid.lgl_oid, - lgi->lgi_logid.lgl_ogen, rc); - GOTO(out_env, rc); - } - rc = llog_init_handle(&env, llh, LLOG_F_IS_CAT, NULL); - if (rc) { - CERROR("%s: llog_init_handle failed: rc = %d\n", - llh->lgh_ctxt->loc_obd->obd_name, rc); - GOTO(release_llh, rc); + CDEBUG(D_INFO, "Add llog entry "DFID" size=%llu, tot=%llu\n", + PFID(&llh->lgh_id.lgl_oi.oi_fid), size, *cum_size); } - if (cb) { - rc = llog_cat_process(&env, llh, cb, NULL, 0, 0); - if (rc != LLOG_PROC_BREAK && rc != 0) - CERROR("%s: llog_cat_process() failed: rc = %d\n", - llh->lgh_ctxt->loc_obd->obd_name, rc); - cb(&env, llh, NULL, NULL); - } else { - CWARN("No callback function for recovery\n"); - } + if (llh != NULL) + llog_handle_put(llh); - /* - * Make sure that all cached data is sent. - */ - llog_sync(ctxt, NULL, 0); - GOTO(release_llh, rc); -release_llh: - rc = llog_cat_close(&env, llh); - if (rc) - CERROR("%s: llog_cat_close() failed: rc = %d\n", - llh->lgh_ctxt->loc_obd->obd_name, rc); -out_env: - lu_env_fini(&env); -out: - llog_ctxt_put(ctxt); - OBD_FREE_PTR(args); - return rc; + RETURN(0); } -EXPORT_SYMBOL(llog_cat_process_thread); -#endif + +__u64 llog_cat_size(const struct lu_env *env, struct llog_handle *cat_llh) +{ + __u64 size = llog_size(env, cat_llh); + + llog_cat_process_or_fork(env, cat_llh, llog_cat_size_cb, + NULL, &size, 0, 0, false); + + return size; +} +EXPORT_SYMBOL(llog_cat_size); + +/* currently returns the number of "free" entries in catalog, + * ie the available entries for a new plain LLOG file creation, + * even if catalog has wrapped + */ +__u32 llog_cat_free_space(struct llog_handle *cat_llh) +{ + /* simulate almost full Catalog */ + if (OBD_FAIL_CHECK(OBD_FAIL_CAT_FREE_RECORDS)) + return cfs_fail_val; + + if (cat_llh->lgh_hdr->llh_count == 1) + return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1; + + if (cat_llh->lgh_last_idx > cat_llh->lgh_hdr->llh_cat_idx) + return LLOG_HDR_BITMAP_SIZE(cat_llh->lgh_hdr) - 1 + + cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx; + + /* catalog is presently wrapped */ + return cat_llh->lgh_hdr->llh_cat_idx - cat_llh->lgh_last_idx; +} +EXPORT_SYMBOL(llog_cat_free_space); static int llog_cat_reverse_process_cb(const struct lu_env *env, struct llog_handle *cat_llh, struct llog_rec_hdr *rec, void *data) { - struct llog_process_data *d = data; - struct llog_logid_rec *lir = (struct llog_logid_rec *)rec; - struct llog_handle *llh; - int rc; + struct llog_process_data *d = data; + struct llog_handle *llh; + int rc; - if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) { - CERROR("invalid record in catalog\n"); - RETURN(-EINVAL); - } - CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog " - LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen, - le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid); - - rc = llog_cat_id2handle(env, cat_llh, &llh, &lir->lid_id); - if (rc) { - CERROR("Cannot find handle for log "LPX64"\n", - lir->lid_id.lgl_oid); - RETURN(rc); - } + ENTRY; + rc = llog_cat_process_common(env, cat_llh, rec, &llh); + + /* The empty plain log was destroyed while processing */ + if (rc == LLOG_DEL_PLAIN) { + rc = llog_cat_cleanup(env, cat_llh, llh, + llh->u.phd.phd_cookie.lgc_index); + } else if (rc == LLOG_DEL_RECORD) { + /* clear wrong catalog entry */ + rc = llog_cat_cleanup(env, cat_llh, NULL, rec->lrh_index); + } + if (rc) + RETURN(rc); rc = llog_reverse_process(env, llh, d->lpd_cb, d->lpd_data, NULL); - RETURN(rc); + + /* The empty plain was destroyed while processing */ + if (rc == LLOG_DEL_PLAIN) + rc = llog_cat_cleanup(env, cat_llh, llh, + llh->u.phd.phd_cookie.lgc_index); + + llog_handle_put(llh); + RETURN(rc); } int llog_cat_reverse_process(const struct lu_env *env, @@ -750,20 +1070,22 @@ int llog_cat_reverse_process(const struct lu_env *env, d.lpd_data = data; d.lpd_cb = cb; - if (llh->llh_cat_idx > cat_llh->lgh_last_idx) { - CWARN("catalog "LPX64" crosses index zero\n", - cat_llh->lgh_id.lgl_oid); + if (llh->llh_cat_idx >= cat_llh->lgh_last_idx && + llh->llh_count > 1) { + CWARN("%s: catalog "DFID" crosses index zero\n", + cat_llh->lgh_ctxt->loc_obd->obd_name, + PFID(&cat_llh->lgh_id.lgl_oi.oi_fid)); - cd.lpcd_first_idx = 0; - cd.lpcd_last_idx = cat_llh->lgh_last_idx; + cd.lpcd_first_idx = 0; + cd.lpcd_last_idx = cat_llh->lgh_last_idx; rc = llog_reverse_process(env, cat_llh, llog_cat_reverse_process_cb, &d, &cd); - if (rc != 0) - RETURN(rc); + if (rc != 0) + RETURN(rc); - cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx); - cd.lpcd_last_idx = 0; + cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx); + cd.lpcd_last_idx = 0; rc = llog_reverse_process(env, cat_llh, llog_cat_reverse_process_cb, &d, &cd); @@ -777,111 +1099,75 @@ int llog_cat_reverse_process(const struct lu_env *env, } EXPORT_SYMBOL(llog_cat_reverse_process); -int llog_cat_set_first_idx(struct llog_handle *cathandle, int index) +static int llog_cat_set_first_idx(struct llog_handle *cathandle, int idx) { - struct llog_log_hdr *llh = cathandle->lgh_hdr; - int i, bitmap_size, idx; - ENTRY; - - bitmap_size = LLOG_BITMAP_SIZE(llh); - if (llh->llh_cat_idx == (index - 1)) { - idx = llh->llh_cat_idx + 1; - llh->llh_cat_idx = idx; - if (idx == cathandle->lgh_last_idx) - goto out; - for (i = (index + 1) % bitmap_size; - i != cathandle->lgh_last_idx; - i = (i + 1) % bitmap_size) { - if (!ext2_test_bit(i, llh->llh_bitmap)) { - idx = llh->llh_cat_idx + 1; - llh->llh_cat_idx = idx; - } else if (i == 0) { - llh->llh_cat_idx = 0; - } else { - break; - } - } -out: - CDEBUG(D_RPCTRACE, "set catlog "LPX64" first idx %u\n", - cathandle->lgh_id.lgl_oid, llh->llh_cat_idx); - } - - RETURN(0); -} - -int cat_cancel_cb(const struct lu_env *env, struct llog_handle *cathandle, - struct llog_rec_hdr *rec, void *data) -{ - struct llog_logid_rec *lir = (struct llog_logid_rec *)rec; - struct llog_handle *loghandle; - struct llog_log_hdr *llh; - int rc, index; + struct llog_log_hdr *llh = cathandle->lgh_hdr; + int bitmap_size; ENTRY; - if (rec->lrh_type != LLOG_LOGID_MAGIC) { - CERROR("%s: invalid record in catalog\n", - loghandle->lgh_ctxt->loc_obd->obd_name); - RETURN(-EINVAL); - } - CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog " - LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen, - rec->lrh_index, cathandle->lgh_id.lgl_oid); - - rc = llog_cat_id2handle(env, cathandle, &loghandle, &lir->lid_id); - if (rc) { - CERROR("%s: cannot find handle for llog "LPX64"\n: %d", - loghandle->lgh_ctxt->loc_obd->obd_name, - lir->lid_id.lgl_oid, rc); - if (rc == -ENOENT || rc == -ESTALE) { - index = rec->lrh_index; - goto cat_cleanup; + bitmap_size = LLOG_HDR_BITMAP_SIZE(llh); + /* + * The llh_cat_idx equals to the first used index minus 1 + * so if we canceled the first index then llh_cat_idx + * must be renewed. + */ + if (llh->llh_cat_idx == (idx - 1)) { + llh->llh_cat_idx = idx; + + while (idx != cathandle->lgh_last_idx) { + idx = (idx + 1) % bitmap_size; + if (!ext2_test_bit(idx, LLOG_HDR_BITMAP(llh))) { + /* update llh_cat_idx for each unset bit, + * expecting the next one is set */ + llh->llh_cat_idx = idx; + } else if (idx == 0) { + /* skip header bit */ + llh->llh_cat_idx = 0; + continue; + } else { + /* the first index is found */ + break; + } } - RETURN(rc); - } - llh = loghandle->lgh_hdr; - if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) && - (llh->llh_count == 1)) { - rc = llog_destroy(env, loghandle); - if (rc) - CERROR("%s: fail to destroy empty log: rc = %d\n", - loghandle->lgh_ctxt->loc_obd->obd_name, rc); - - index = loghandle->u.phd.phd_cookie.lgc_index; - llog_close(env, loghandle); - -cat_cleanup: - LASSERT(index); - llog_cat_set_first_idx(cathandle, index); - rc = llog_cancel_rec(env, cathandle, index); - if (rc == 0) - CDEBUG(D_HA, - "cancel log "LPX64":%x at index %u of catalog " - LPX64"\n", lir->lid_id.lgl_oid, - lir->lid_id.lgl_ogen, rec->lrh_index, - cathandle->lgh_id.lgl_oid); + CDEBUG(D_RPCTRACE, "catlog "DFID" first idx %u, last_idx %u\n", + PFID(&cathandle->lgh_id.lgl_oi.oi_fid), + llh->llh_cat_idx, cathandle->lgh_last_idx); } - RETURN(rc); + RETURN(0); } -EXPORT_SYMBOL(cat_cancel_cb); -/* helper to initialize catalog llog and process it to cancel */ -int llog_cat_init_and_process(const struct lu_env *env, - struct llog_handle *llh) +/* Cleanup deleted plain llog traces from catalog */ +int llog_cat_cleanup(const struct lu_env *env, struct llog_handle *cathandle, + struct llog_handle *loghandle, int index) { int rc; - rc = llog_init_handle(env, llh, LLOG_F_IS_CAT, NULL); - if (rc) - RETURN(rc); + LASSERT(index); + if (loghandle != NULL) { + /* remove destroyed llog from catalog list and + * chd_current_log variable */ + down_write(&cathandle->lgh_lock); + if (cathandle->u.chd.chd_current_log == loghandle) + cathandle->u.chd.chd_current_log = NULL; + list_del_init(&loghandle->u.phd.phd_entry); + up_write(&cathandle->lgh_lock); + LASSERT(index == loghandle->u.phd.phd_cookie.lgc_index); + /* llog was opened and keep in a list, close it now */ + llog_close(env, loghandle); + } - rc = llog_process(env, llh, cat_cancel_cb, NULL, NULL); - if (rc) - CERROR("%s: llog_process() with cat_cancel_cb failed: rc = " - "%d\n", llh->lgh_ctxt->loc_obd->obd_name, rc); - RETURN(0); -} -EXPORT_SYMBOL(llog_cat_init_and_process); + /* do not attempt to cleanup on-disk llog if on client side */ + if (cathandle->lgh_obj == NULL) + return 0; + /* remove plain llog entry from catalog by index */ + llog_cat_set_first_idx(cathandle, index); + rc = llog_cancel_rec(env, cathandle, index); + if (rc == 0) + CDEBUG(D_HA, "cancel plain log at index %u of catalog "DFID"\n", + index, PFID(&cathandle->lgh_id.lgl_oi.oi_fid)); + return rc; +}