X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fquota%2Fquota_interface.c;h=87c14cffb0e3ef55c08daca3283efeb6e5b1ad6a;hb=d8506f4b3a03b5605fc927409ce16f55ad5bffd5;hp=5e1c51b7ab609173dab8fd35a0254b4e6dcd58c1;hpb=11127cbfc4b9a184376a054a60a268d3e4b214b9;p=fs%2Flustre-release.git diff --git a/lustre/quota/quota_interface.c b/lustre/quota/quota_interface.c index 5e1c51b..87c14cf 100644 --- a/lustre/quota/quota_interface.c +++ b/lustre/quota/quota_interface.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -45,14 +45,10 @@ # include # include # include -# if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)) -# include -# include -# include -# include -# else -# include -# endif +# include +# include +# include +# include #else /* __KERNEL__ */ # include #endif @@ -72,7 +68,7 @@ #ifdef HAVE_QUOTA_SUPPORT static cfs_time_t last_print = 0; -static spinlock_t last_print_lock = SPIN_LOCK_UNLOCKED; +static cfs_spinlock_t last_print_lock = CFS_SPIN_LOCK_UNLOCKED; static int filter_quota_setup(struct obd_device *obd) { @@ -80,9 +76,9 @@ static int filter_quota_setup(struct obd_device *obd) struct obd_device_target *obt = &obd->u.obt; ENTRY; - init_rwsem(&obt->obt_rwsem); + cfs_init_rwsem(&obt->obt_rwsem); obt->obt_qfmt = LUSTRE_QUOTA_V2; - atomic_set(&obt->obt_quotachecking, 1); + cfs_sema_init(&obt->obt_quotachecking, 1); rc = qctxt_init(obd, NULL); if (rc) CERROR("initialize quota context failed! (rc:%d)\n", rc); @@ -101,48 +97,65 @@ static int filter_quota_setinfo(struct obd_device *obd, void *data) { struct obd_export *exp = data; struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; - struct obd_import *imp; + struct obd_import *imp = exp->exp_imp_reverse; ENTRY; + LASSERT(imp != NULL); + /* setup the quota context import */ - spin_lock(&qctxt->lqc_lock); - qctxt->lqc_import = exp->exp_imp_reverse; - spin_unlock(&qctxt->lqc_lock); - CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated now, \n", - obd->obd_name,exp->exp_imp_reverse, obd); - - /* make imp's connect flags equal relative exp's connect flags - * adding it to avoid the scan export list - */ - imp = qctxt->lqc_import; - if (likely(imp)) + cfs_spin_lock(&qctxt->lqc_lock); + if (qctxt->lqc_import != NULL) { + cfs_spin_unlock(&qctxt->lqc_lock); + if (qctxt->lqc_import == imp) + CDEBUG(D_WARNING, "%s: lqc_import(%p) of obd(%p) was " + "activated already.\n", obd->obd_name, imp, obd); + else + CERROR("%s: lqc_import(%p:%p) of obd(%p) was " + "activated by others.\n", obd->obd_name, + qctxt->lqc_import, imp, obd); + } else { + qctxt->lqc_import = imp; + /* make imp's connect flags equal relative exp's connect flags + * adding it to avoid the scan export list */ imp->imp_connect_data.ocd_connect_flags |= - (exp->exp_connect_flags & - (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS)); - - cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster); - /* start quota slave recovery thread. (release high limits) */ - qslave_start_recovery(obd, qctxt); + (exp->exp_connect_flags & + (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS)); + cfs_spin_unlock(&qctxt->lqc_lock); + CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated " + "now.\n", obd->obd_name, imp, obd); + + cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster); + /* start quota slave recovery thread. (release high limits) */ + qslave_start_recovery(obd, qctxt); + } RETURN(0); } static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd) { struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; + struct obd_import *imp = exp->exp_imp_reverse; ENTRY; /* lquota may be not set up before destroying export, b=14896 */ if (!obd->obd_set_up) RETURN(0); + if (unlikely(imp == NULL)) + RETURN(0); + /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import * should be invalid b=12374 */ - if (qctxt->lqc_import && qctxt->lqc_import == exp->exp_imp_reverse) { - spin_lock(&qctxt->lqc_lock); + cfs_spin_lock(&qctxt->lqc_lock); + if (qctxt->lqc_import == imp) { qctxt->lqc_import = NULL; - spin_unlock(&qctxt->lqc_lock); - CDEBUG(D_QUOTA, "%s: lqc_import of obd(%p) is invalid now.\n", - obd->obd_name, obd); + cfs_spin_unlock(&qctxt->lqc_lock); + CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is invalid now.\n", + obd->obd_name, imp, obd); + ptlrpc_cleanup_imp(imp); + dqacq_interrupt(qctxt); + } else { + cfs_spin_unlock(&qctxt->lqc_lock); } RETURN(0); } @@ -151,7 +164,7 @@ static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore) { ENTRY; - if (!sb_any_quota_enabled(obd->u.obt.obt_sb)) + if (!ll_sb_any_quota_active(obd->u.obt.obt_sb)) RETURN(0); if (ignore) { @@ -164,6 +177,7 @@ static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore) RETURN(0); } +#define GET_OA_ID(flag, oa) (flag == USRQUOTA ? oa->o_uid : oa->o_gid) static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa) { struct obd_device_target *obt = &obd->u.obt; @@ -172,33 +186,37 @@ static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa) struct obd_quotactl *oqctl; ENTRY; - if (!sb_any_quota_enabled(obt->obt_sb)) + if (!ll_sb_any_quota_active(obt->obt_sb)) RETURN(0); OBD_ALLOC_PTR(oqctl); - if (!oqctl) { - CERROR("Not enough memory!"); + if (!oqctl) RETURN(-ENOMEM); - } /* set over quota flags for a uid/gid */ oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA; oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA); for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - struct quota_adjust_qunit oqaq_tmp; struct lustre_qunit_size *lqs = NULL; - oqaq_tmp.qaq_flags = cnt; - oqaq_tmp.qaq_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid; - - quota_search_lqs(NULL, &oqaq_tmp, qctxt, &lqs); - if (lqs) { - spin_lock(&lqs->lqs_lock); + lqs = quota_search_lqs(LQS_KEY(cnt, GET_OA_ID(cnt, oa)), + qctxt, 0); + if (lqs == NULL || IS_ERR(lqs)) { + rc = PTR_ERR(lqs); + if (rc) + CDEBUG(D_QUOTA, "search lqs for %s %d failed, " + "(rc = %d)\n", + cnt == USRQUOTA ? "user" : "group", + cnt == USRQUOTA ? oa->o_uid : oa->o_gid, + rc); + break; + } else { + cfs_spin_lock(&lqs->lqs_lock); if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) { oa->o_flags |= (cnt == USRQUOTA) ? OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA; - spin_unlock(&lqs->lqs_lock); + cfs_spin_unlock(&lqs->lqs_lock); CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), " "sync_blk(%d)\n", lqs->lqs_bunit_sz, qctxt->lqc_sync_blk); @@ -206,7 +224,7 @@ static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa) lqs_putref(lqs); continue; } - spin_unlock(&lqs->lqs_lock); + cfs_spin_unlock(&lqs->lqs_lock); /* this is for quota_search_lqs */ lqs_putref(lqs); } @@ -222,14 +240,22 @@ static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa) rc = err; oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA); + CDEBUG(D_QUOTA, "fsfilt getquota for %s %d failed, " + "(rc = %d)\n", + cnt == USRQUOTA ? "user" : "group", + cnt == USRQUOTA ? oa->o_uid : oa->o_gid, err); continue; } if (oqctl->qc_dqblk.dqb_bhardlimit && (toqb(oqctl->qc_dqblk.dqb_curspace) >= - oqctl->qc_dqblk.dqb_bhardlimit)) + oqctl->qc_dqblk.dqb_bhardlimit)) { oa->o_flags |= (cnt == USRQUOTA) ? OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA; + CDEBUG(D_QUOTA, "out of quota for %s %d\n", + cnt == USRQUOTA ? "user" : "group", + cnt == USRQUOTA ? oa->o_uid : oa->o_gid); + } } OBD_FREE_PTR(oqctl); RETURN(rc); @@ -239,26 +265,23 @@ static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa) * check whether the left quota of certain uid and gid can satisfy a block_write * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA */ -static int quota_check_common(struct obd_device *obd, unsigned int uid, - unsigned int gid, int count, int cycle, int isblk) +static int quota_check_common(struct obd_device *obd, const unsigned int id[], + int pending[], int count, int cycle, int isblk, + struct inode *inode, int frags) { struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; int i; - __u32 id[MAXQUOTAS] = { uid, gid }; struct qunit_data qdata[MAXQUOTAS]; + int mb = 0; int rc = 0, rc2[2] = { 0, 0 }; ENTRY; - CLASSERT(MAXQUOTAS < 4); - if (!sb_any_quota_enabled(qctxt->lqc_sb)) - RETURN(rc); - - spin_lock(&qctxt->lqc_lock); + cfs_spin_lock(&qctxt->lqc_lock); if (!qctxt->lqc_valid){ - spin_unlock(&qctxt->lqc_lock); + cfs_spin_unlock(&qctxt->lqc_lock); RETURN(rc); } - spin_unlock(&qctxt->lqc_lock); + cfs_spin_unlock(&qctxt->lqc_lock); for (i = 0; i < MAXQUOTAS; i++) { struct lustre_qunit_size *lqs = NULL; @@ -273,32 +296,81 @@ static int quota_check_common(struct obd_device *obd, unsigned int uid, if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i])) continue; - quota_search_lqs(&qdata[i], NULL, qctxt, &lqs); - if (!lqs) + lqs = quota_search_lqs(LQS_KEY(i, id[i]), qctxt, 0); + if (lqs == NULL || IS_ERR(lqs)) continue; + if (IS_ERR(lqs)) { + CERROR("can not find lqs for check_common: " + "[id %u] [%c] [isblk %d] [count %d] [rc %ld]\n", + id[i], i % 2 ? 'g': 'u', isblk, count, + PTR_ERR(lqs)); + RETURN(PTR_ERR(lqs)); + } + rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk); - spin_lock(&lqs->lqs_lock); + cfs_spin_lock(&lqs->lqs_lock); if (!cycle) { - rc = QUOTA_RET_INC_PENDING; - if (isblk) - lqs->lqs_bwrite_pending += count; + if (isblk) { + pending[i] = count * CFS_PAGE_SIZE; + /* in order to complete this write, we need extra + * meta blocks. This function can get it through + * data needed to be written b=16542 */ + if (inode) { + mb = pending[i]; + rc = fsfilt_get_mblk(obd, qctxt->lqc_sb, + &mb, inode, + frags); + if (rc) + CERROR("%s: can't get extra " + "meta blocks\n", + obd->obd_name); + else + pending[i] += mb; + } + lqs->lqs_bwrite_pending += pending[i]; + } else { + pending[i] = count; + lqs->lqs_iwrite_pending += pending[i]; + } + } + + /* if xx_rec < 0, that means quota are releasing, + * and it may return before we use quota. So if + * we find this situation, we assuming it has + * returned b=18491 */ + if (isblk && lqs->lqs_blk_rec < 0) { + if (qdata[i].qd_count < -lqs->lqs_blk_rec) + qdata[i].qd_count = 0; + else + qdata[i].qd_count += lqs->lqs_blk_rec; + } + if (!isblk && lqs->lqs_ino_rec < 0) { + if (qdata[i].qd_count < -lqs->lqs_ino_rec) + qdata[i].qd_count = 0; else - lqs->lqs_iwrite_pending += count; + qdata[i].qd_count += lqs->lqs_ino_rec; } + + CDEBUG(D_QUOTA, "[id %u] [%c] [isblk %d] [count %d]" + " [lqs pending: %lu] [qd_count: "LPU64"] [metablocks: %d]" + " [pending: %d]\n", id[i], i % 2 ? 'g': 'u', isblk, count, + isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending, + qdata[i].qd_count, mb, pending[i]); if (rc2[i] == QUOTA_RET_OK) { - if (isblk && qdata[i].qd_count < - lqs->lqs_bwrite_pending * CFS_PAGE_SIZE) + if (isblk && qdata[i].qd_count < lqs->lqs_bwrite_pending) rc2[i] = QUOTA_RET_ACQUOTA; if (!isblk && qdata[i].qd_count < lqs->lqs_iwrite_pending) rc2[i] = QUOTA_RET_ACQUOTA; } - spin_unlock(&lqs->lqs_lock); - CDEBUG(D_QUOTA, "count: %d, write pending: %lu, qd_count: "LPU64 - ".\n", count, - isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending, - qdata[i].qd_count); + + cfs_spin_unlock(&lqs->lqs_lock); + + if (lqs->lqs_blk_rec < 0 && + qdata[i].qd_count < + lqs->lqs_bwrite_pending - lqs->lqs_blk_rec - mb) + OBD_FAIL_TIMEOUT(OBD_FAIL_QUOTA_DELAY_REL, 5); /* When cycle is zero, lqs_*_pending will be changed. We will * get reference of the lqs here and put reference of lqs in @@ -311,15 +383,37 @@ static int quota_check_common(struct obd_device *obd, unsigned int uid, } if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA) - RETURN(rc | QUOTA_RET_ACQUOTA); + RETURN(QUOTA_RET_ACQUOTA); else RETURN(rc); } -static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid, - unsigned int gid, int count, int *pending, - quota_acquire acquire, - struct obd_trans_info *oti, int isblk) +int quota_is_set(struct obd_device *obd, const unsigned int id[], int flag) +{ + struct lustre_qunit_size *lqs; + int i, q_set = 0; + + if (!ll_sb_any_quota_active(obd->u.obt.obt_qctxt.lqc_sb)) + RETURN(0); + + for (i = 0; i < MAXQUOTAS; i++) { + lqs = quota_search_lqs(LQS_KEY(i, id[i]), + &obd->u.obt.obt_qctxt, 0); + if (lqs && !IS_ERR(lqs)) { + if (lqs->lqs_flags & flag) + q_set = 1; + lqs_putref(lqs); + } + } + + return q_set; +} + +static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp, + const unsigned int id[], int pending[], + int count, quota_acquire acquire, + struct obd_trans_info *oti, int isblk, + struct inode *inode, int frags) { struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; struct timeval work_start; @@ -329,19 +423,29 @@ static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid, int rc = 0, cycle = 0, count_err = 1; ENTRY; + if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET)) + RETURN(0); + + if (isblk && (exp->exp_failed || exp->exp_abort_active_req)) + /* If the client has been evicted or if it + * timed out and tried to reconnect already, + * abort the request immediately */ + RETURN(-ENOTCONN); + CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name); - *pending = 0; + pending[USRQUOTA] = pending[GRPQUOTA] = 0; /* Unfortunately, if quota master is too busy to handle the * pre-dqacq in time and quota hash on ost is used up, we * have to wait for the completion of in flight dqacq/dqrel, * in order to get enough quota for write b=12588 */ - do_gettimeofday(&work_start); - while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk)) & + cfs_gettimeofday(&work_start); + while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk, + inode, frags)) & QUOTA_RET_ACQUOTA) { - spin_lock(&qctxt->lqc_lock); + cfs_spin_lock(&qctxt->lqc_lock); if (!qctxt->lqc_import && oti) { - spin_unlock(&qctxt->lqc_lock); + cfs_spin_unlock(&qctxt->lqc_lock); LASSERT(oti && oti->oti_thread && oti->oti_thread->t_watchdog); @@ -351,24 +455,22 @@ static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid, l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt), &lwi); CDEBUG(D_QUOTA, "wake up when quota master is back\n"); - lc_watchdog_touch(oti->oti_thread->t_watchdog); + lc_watchdog_touch(oti->oti_thread->t_watchdog, + CFS_GET_TIMEOUT(oti->oti_thread->t_svc)); } else { - spin_unlock(&qctxt->lqc_lock); + cfs_spin_unlock(&qctxt->lqc_lock); } - if (rc & QUOTA_RET_INC_PENDING) - *pending = 1; - cycle++; if (isblk) OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90); /* after acquire(), we should run quota_check_common again * so that we confirm there are enough quota to finish write */ - rc = acquire(obd, uid, gid, oti, isblk); + rc = acquire(obd, id, oti, isblk); /* please reference to dqacq_completion for the below */ /* a new request is finished, try again */ - if (rc == -EAGAIN) { + if (rc == QUOTA_REQ_RETURNED) { CDEBUG(D_QUOTA, "finish a quota req, try again\n"); continue; } @@ -379,52 +481,63 @@ static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid, break; } + /* Related quota has been disabled by master, but enabled by + * slave, do not try again. */ + if (unlikely(rc == -ESRCH)) { + CERROR("mismatched quota configuration, stop try.\n"); + break; + } + + if (isblk && (exp->exp_failed || exp->exp_abort_active_req)) + /* The client has been evicted or tried to + * to reconnect already, abort the request */ + RETURN(-ENOTCONN); + /* -EBUSY and others, wait a second and try again */ if (rc < 0) { cfs_waitq_t waitq; struct l_wait_info lwi; if (oti && oti->oti_thread && oti->oti_thread->t_watchdog) - lc_watchdog_touch(oti->oti_thread->t_watchdog); + lc_watchdog_touch(oti->oti_thread->t_watchdog, + CFS_GET_TIMEOUT(oti->oti_thread->t_svc)); CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc, count_err++); - init_waitqueue_head(&waitq); + cfs_waitq_init(&waitq); lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL, NULL); l_wait_event(waitq, 0, &lwi); } - if (rc < 0 || cycle % 10 == 2) { - spin_lock(&last_print_lock); + if (rc < 0 || cycle % 10 == 0) { + cfs_spin_lock(&last_print_lock); if (last_print == 0 || cfs_time_before((last_print + cfs_time_seconds(30)), cfs_time_current())) { last_print = cfs_time_current(); - spin_unlock(&last_print_lock); + cfs_spin_unlock(&last_print_lock); CWARN("still haven't managed to acquire quota " "space from the quota master after %d " "retries (err=%d, rc=%d)\n", cycle, count_err - 1, rc); } else { - spin_unlock(&last_print_lock); + cfs_spin_unlock(&last_print_lock); } } CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc, cycle); } - - if (!cycle && rc & QUOTA_RET_INC_PENDING) - *pending = 1; - - do_gettimeofday(&work_end); + cfs_gettimeofday(&work_end); timediff = cfs_timeval_sub(&work_end, &work_start, NULL); lprocfs_counter_add(qctxt->lqc_stats, isblk ? LQUOTA_WAIT_FOR_CHK_BLK : LQUOTA_WAIT_FOR_CHK_INO, timediff); + if (rc > 0) + rc = 0; RETURN(rc); } @@ -432,27 +545,30 @@ static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid, * when a block_write or inode_create rpc is finished, adjust the record for * pending blocks and inodes */ -static int quota_pending_commit(struct obd_device *obd, unsigned int uid, - unsigned int gid, int count, int isblk) +static int quota_pending_commit(struct obd_device *obd, const unsigned int id[], + int pending[], int isblk) { struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; struct timeval work_start; struct timeval work_end; long timediff; int i; - __u32 id[MAXQUOTAS] = { uid, gid }; struct qunit_data qdata[MAXQUOTAS]; ENTRY; CDEBUG(D_QUOTA, "commit pending quota for %s\n", obd->obd_name); CLASSERT(MAXQUOTAS < 4); - if (!sb_any_quota_enabled(qctxt->lqc_sb)) + if (!ll_sb_any_quota_active(qctxt->lqc_sb)) RETURN(0); - do_gettimeofday(&work_start); + cfs_gettimeofday(&work_start); for (i = 0; i < MAXQUOTAS; i++) { struct lustre_qunit_size *lqs = NULL; + LASSERT(pending[i] >= 0); + if (pending[i] == 0) + continue; + qdata[i].qd_id = id[i]; qdata[i].qd_flags = i; if (isblk) @@ -462,43 +578,46 @@ static int quota_pending_commit(struct obd_device *obd, unsigned int uid, if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i])) continue; - quota_search_lqs(&qdata[i], NULL, qctxt, &lqs); - if (lqs) { - int flag = 0; - CDEBUG(D_QUOTA, "pending: %lu, count: %d.\n", - isblk ? lqs->lqs_bwrite_pending : - lqs->lqs_iwrite_pending, count); - spin_lock(&lqs->lqs_lock); - if (isblk) { - if (lqs->lqs_bwrite_pending >= count) { - lqs->lqs_bwrite_pending -= count; - spin_unlock(&lqs->lqs_lock); - flag = 1; - } else { - spin_unlock(&lqs->lqs_lock); - CDEBUG(D_ERROR, - "there are too many blocks!\n"); - } - } else { - if (lqs->lqs_iwrite_pending >= count) { - lqs->lqs_iwrite_pending -= count; - spin_unlock(&lqs->lqs_lock); - flag = 1; - } else { - spin_unlock(&lqs->lqs_lock); - CDEBUG(D_ERROR, - "there are too many files!\n"); - } - } + lqs = quota_search_lqs(LQS_KEY(i, qdata[i].qd_id), qctxt, 0); + if (lqs == NULL || IS_ERR(lqs)) { + CERROR("can not find lqs for pending_commit: " + "[id %u] [%c] [pending %u] [isblk %d] (rc %ld), " + "maybe cause unexpected lqs refcount error!\n", + id[i], i ? 'g': 'u', pending[i], isblk, + lqs ? PTR_ERR(lqs) : -1); + continue; + } - lqs_putref(lqs); - /* When lqs_*_pening is changed back, we'll putref lqs - * here b=14784 */ - if (flag) - lqs_putref(lqs); + cfs_spin_lock(&lqs->lqs_lock); + if (isblk) { + LASSERTF(lqs->lqs_bwrite_pending >= pending[i], + "there are too many blocks! [id %u] [%c] " + "[bwrite_pending %lu] [pending %u]\n", + id[i], i % 2 ? 'g' : 'u', + lqs->lqs_bwrite_pending, pending[i]); + + lqs->lqs_bwrite_pending -= pending[i]; + } else { + LASSERTF(lqs->lqs_iwrite_pending >= pending[i], + "there are too many files! [id %u] [%c] " + "[iwrite_pending %lu] [pending %u]\n", + id[i], i % 2 ? 'g' : 'u', + lqs->lqs_iwrite_pending, pending[i]); + + lqs->lqs_iwrite_pending -= pending[i]; } + CDEBUG(D_QUOTA, "%s: lqs_pending=%lu pending[%d]=%d isblk=%d\n", + obd->obd_name, + isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending, + i, pending[i], isblk); + cfs_spin_unlock(&lqs->lqs_lock); + + /* for quota_search_lqs in pending_commit */ + lqs_putref(lqs); + /* for quota_search_lqs in quota_check */ + lqs_putref(lqs); } - do_gettimeofday(&work_end); + cfs_gettimeofday(&work_end); timediff = cfs_timeval_sub(&work_end, &work_start, NULL); lprocfs_counter_add(qctxt->lqc_stats, isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK : @@ -531,15 +650,16 @@ static int mds_quota_setup(struct obd_device *obd) RETURN(0); } - init_rwsem(&obt->obt_rwsem); + cfs_init_rwsem(&obt->obt_rwsem); obt->obt_qfmt = LUSTRE_QUOTA_V2; mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2; - atomic_set(&obt->obt_quotachecking, 1); + cfs_sema_init(&obt->obt_quotachecking, 1); /* initialize quota master and quota context */ - sema_init(&mds->mds_qonoff_sem, 1); + cfs_init_rwsem(&mds->mds_qonoff_sem); rc = qctxt_init(obd, dqacq_handler); if (rc) { - CERROR("initialize quota context failed! (rc:%d)\n", rc); + CERROR("%s: initialize quota context failed! (rc:%d)\n", + obd->obd_name, rc); RETURN(rc); } mds->mds_quota = 1; @@ -584,21 +704,20 @@ static int mds_quota_fs_cleanup(struct obd_device *obd) memset(&oqctl, 0, sizeof(oqctl)); oqctl.qc_type = UGQUOTA; - down(&mds->mds_qonoff_sem); + cfs_down_write(&mds->mds_qonoff_sem); mds_admin_quota_off(obd, &oqctl); - up(&mds->mds_qonoff_sem); + cfs_up_write(&mds->mds_qonoff_sem); RETURN(0); } -static int quota_acquire_common(struct obd_device *obd, unsigned int uid, - unsigned int gid, struct obd_trans_info *oti, - int isblk) +static int quota_acquire_common(struct obd_device *obd, const unsigned int id[], + struct obd_trans_info *oti, int isblk) { struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt; int rc; ENTRY; - rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, isblk, 1, oti); + rc = qctxt_adjust_qunit(obd, qctxt, id, isblk, 1, oti); RETURN(rc); } @@ -606,15 +725,15 @@ static int quota_acquire_common(struct obd_device *obd, unsigned int uid, #endif /* __KERNEL__ */ struct osc_quota_info { - struct list_head oqi_hash; /* hash list */ + cfs_list_t oqi_hash; /* hash list */ struct client_obd *oqi_cli; /* osc obd */ unsigned int oqi_id; /* uid/gid of a file */ short oqi_type; /* quota type */ }; -spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED; +cfs_spinlock_t qinfo_list_lock = CFS_SPIN_LOCK_UNLOCKED; -static struct list_head qinfo_hash[NR_DQHASH]; +static cfs_list_t qinfo_hash[NR_DQHASH]; /* SLAB cache for client quota context */ cfs_mem_cache_t *qinfo_cachep = NULL; @@ -631,18 +750,18 @@ static inline int hashfn(struct client_obd *cli, unsigned long id, int type) /* caller must hold qinfo_list_lock */ static inline void insert_qinfo_hash(struct osc_quota_info *oqi) { - struct list_head *head = qinfo_hash + + cfs_list_t *head = qinfo_hash + hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type); LASSERT_SPIN_LOCKED(&qinfo_list_lock); - list_add(&oqi->oqi_hash, head); + cfs_list_add(&oqi->oqi_hash, head); } /* caller must hold qinfo_list_lock */ static inline void remove_qinfo_hash(struct osc_quota_info *oqi) { LASSERT_SPIN_LOCKED(&qinfo_list_lock); - list_del_init(&oqi->oqi_hash); + cfs_list_del_init(&oqi->oqi_hash); } /* caller must hold qinfo_list_lock */ @@ -654,7 +773,7 @@ static inline struct osc_quota_info *find_qinfo(struct client_obd *cli, ENTRY; LASSERT_SPIN_LOCKED(&qinfo_list_lock); - list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) { + cfs_list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) { if (oqi->oqi_cli == cli && oqi->oqi_id == id && oqi->oqi_type == type) return oqi; @@ -668,7 +787,7 @@ static struct osc_quota_info *alloc_qinfo(struct client_obd *cli, struct osc_quota_info *oqi; ENTRY; - OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi)); + OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_IO, sizeof(*oqi)); if(!oqi) RETURN(NULL); @@ -685,29 +804,32 @@ static void free_qinfo(struct osc_quota_info *oqi) OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi)); } -int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid) +int osc_quota_chkdq(struct client_obd *cli, const unsigned int qid[]) { unsigned int id; int cnt, rc = QUOTA_OK; ENTRY; - spin_lock(&qinfo_list_lock); + cfs_spin_lock(&qinfo_list_lock); for (cnt = 0; cnt < MAXQUOTAS; cnt++) { struct osc_quota_info *oqi = NULL; - id = (cnt == USRQUOTA) ? uid : gid; + id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA]; oqi = find_qinfo(cli, id, cnt); if (oqi) { rc = NO_QUOTA; break; } } - spin_unlock(&qinfo_list_lock); + cfs_spin_unlock(&qinfo_list_lock); + if (rc == NO_QUOTA) + CDEBUG(D_QUOTA, "chkdq found noquota for %s %d\n", + cnt == USRQUOTA ? "user" : "group", id); RETURN(rc); } -int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid, +int osc_quota_setdq(struct client_obd *cli, const unsigned int qid[], obd_flag valid, obd_flag flags) { unsigned int id; @@ -717,36 +839,48 @@ int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid, for (cnt = 0; cnt < MAXQUOTAS; cnt++) { - struct osc_quota_info *oqi, *old; + struct osc_quota_info *oqi = NULL, *old; if (!(valid & ((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA))) continue; - id = (cnt == USRQUOTA) ? uid : gid; + id = (cnt == USRQUOTA) ? qid[USRQUOTA] : qid[GRPQUOTA]; noquota = (cnt == USRQUOTA) ? (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA); - oqi = alloc_qinfo(cli, id, cnt); - if (oqi) { - spin_lock(&qinfo_list_lock); - - old = find_qinfo(cli, id, cnt); - if (old && !noquota) - remove_qinfo_hash(old); - else if (!old && noquota) - insert_qinfo_hash(oqi); - - spin_unlock(&qinfo_list_lock); + if (noquota) { + oqi = alloc_qinfo(cli, id, cnt); + if (!oqi) { + rc = -ENOMEM; + CDEBUG(D_QUOTA, "setdq for %s %d failed, " + "(rc = %d)\n", + cnt == USRQUOTA ? "user" : "group", + id, rc); + break; + } + } - if (old || !noquota) + cfs_spin_lock(&qinfo_list_lock); + old = find_qinfo(cli, id, cnt); + if (old && !noquota) + remove_qinfo_hash(old); + else if (!old && noquota) + insert_qinfo_hash(oqi); + cfs_spin_unlock(&qinfo_list_lock); + + if (old && !noquota) + CDEBUG(D_QUOTA, "setdq to remove for %s %d\n", + cnt == USRQUOTA ? "user" : "group", id); + else if (!old && noquota) + CDEBUG(D_QUOTA, "setdq to insert for %s %d\n", + cnt == USRQUOTA ? "user" : "group", id); + + if (old) { + if (noquota) free_qinfo(oqi); - if (old && !noquota) + else free_qinfo(old); - } else { - CERROR("not enough mem!\n"); - rc = -ENOMEM; - break; } } @@ -760,16 +894,16 @@ int osc_quota_cleanup(struct obd_device *obd) int i; ENTRY; - spin_lock(&qinfo_list_lock); + cfs_spin_lock(&qinfo_list_lock); for (i = 0; i < NR_DQHASH; i++) { - list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) { + cfs_list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) { if (oqi->oqi_cli != cli) continue; remove_qinfo_hash(oqi); free_qinfo(oqi); } } - spin_unlock(&qinfo_list_lock); + cfs_spin_unlock(&qinfo_list_lock); RETURN(0); } @@ -798,14 +932,14 @@ int osc_quota_exit(void) int i, rc; ENTRY; - spin_lock(&qinfo_list_lock); + cfs_spin_lock(&qinfo_list_lock); for (i = 0; i < NR_DQHASH; i++) { - list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) { + cfs_list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) { remove_qinfo_hash(oqi); free_qinfo(oqi); } } - spin_unlock(&qinfo_list_lock); + cfs_spin_unlock(&qinfo_list_lock); rc = cfs_mem_cache_destroy(qinfo_cachep); LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");