lustre/quota/quota_adjust_qunit.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #ifndef EXPORT_SYMTAB
  37 # define EXPORT_SYMTAB
  38 #endif
  39 #define DEBUG_SUBSYSTEM S_LQUOTA
  40
  41 #ifdef __KERNEL__
  42 # include <linux/version.h>
  43 # include <linux/module.h>
  44 # include <linux/init.h>
  45 # include <linux/fs.h>
  46 # include <linux/jbd.h>
  47 # include <linux/quota.h>
  48 # include <linux/smp_lock.h>
  49 # include <linux/buffer_head.h>
  50 # include <linux/workqueue.h>
  51 # include <linux/mount.h>
  52 #else /* __KERNEL__ */
  53 # include <liblustre.h>
  54 #endif
  55
  56 #include <obd_class.h>
  57 #include <lustre_mds.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_cfg.h>
  60 #include <obd_ost.h>
  61 #include <lustre_fsfilt.h>
  62 #include <linux/lustre_quota.h>
  63 #include "quota_internal.h"
  64
  65 #ifdef HAVE_QUOTA_SUPPORT
  66
  67 #ifdef __KERNEL__
  68 /**
  69  * This function is charge of recording lqs_ino_rec and
  70  * lqs_blk_rec. when a lquota slave checks a quota
  71  * request(check_cur_qunit) and finishes a quota
  72  * request(dqacq_completion), it will be called.
  73  * is_chk: whether it is checking quota; otherwise, it is finishing
  74  * is_acq: whether it is acquiring; otherwise, it is releasing
  75  */
  76 void quota_compute_lqs(struct qunit_data *qdata, struct lustre_qunit_size *lqs,
  77                        int is_chk, int is_acq)
  78 {
  79         long long *rec;
  80
  81         LASSERT(qdata && lqs);
  82         LASSERT_SPIN_LOCKED(&lqs->lqs_lock);
  83
  84         rec = QDATA_IS_BLK(qdata) ? &lqs->lqs_blk_rec : &lqs->lqs_ino_rec;
  85
  86         if (!!is_chk + !!is_acq == 1)
  87                 *rec -= qdata->qd_count;
  88         else
  89                 *rec += qdata->qd_count;
  90
  91 }
  92
  93 struct lustre_qunit_size *quota_search_lqs(unsigned long long lqs_key,
  94                                            struct lustre_quota_ctxt *qctxt,
  95                                            int create)
  96 {
  97         struct lustre_qunit_size *lqs;
  98         struct lustre_qunit_size *lqs2;
  99         cfs_hash_t *hs = NULL;
 100         int rc = 0;
 101
 102         cfs_spin_lock(&qctxt->lqc_lock);
 103         if (qctxt->lqc_valid) {
 104                 LASSERT(qctxt->lqc_lqs_hash != NULL);
 105                 hs = cfs_hash_getref(qctxt->lqc_lqs_hash);
 106         }
 107         cfs_spin_unlock(&qctxt->lqc_lock);
 108
 109         if (hs == NULL) {
 110                 rc = -EBUSY;
 111                 goto out;
 112         }
 113
 114         /* cfs_hash_lookup will +1 refcount for caller */
 115         lqs = cfs_hash_lookup(qctxt->lqc_lqs_hash, &lqs_key);
 116         if (lqs != NULL) /* found */
 117                 goto out_put;
 118
 119         if (!create)
 120                 goto out_put;
 121
 122         OBD_ALLOC_PTR(lqs);
 123         if (!lqs) {
 124                 rc = -ENOMEM;
 125                 goto out_put;
 126         }
 127
 128         lqs->lqs_key = lqs_key;
 129
 130         cfs_spin_lock_init(&lqs->lqs_lock);
 131
 132         lqs->lqs_bwrite_pending = 0;
 133         lqs->lqs_iwrite_pending = 0;
 134         lqs->lqs_ino_rec = 0;
 135         lqs->lqs_blk_rec = 0;
 136         lqs->lqs_id = LQS_KEY_ID(lqs->lqs_key);
 137         lqs->lqs_flags = LQS_KEY_GRP(lqs->lqs_key) ? LQUOTA_FLAGS_GRP : 0;
 138         lqs->lqs_bunit_sz = qctxt->lqc_bunit_sz;
 139         lqs->lqs_iunit_sz = qctxt->lqc_iunit_sz;
 140         lqs->lqs_btune_sz = qctxt->lqc_btune_sz;
 141         lqs->lqs_itune_sz = qctxt->lqc_itune_sz;
 142         if (qctxt->lqc_handler) {
 143                 lqs->lqs_last_bshrink  = 0;
 144                 lqs->lqs_last_ishrink  = 0;
 145         }
 146
 147         lqs->lqs_ctxt = qctxt; /* must be called before lqs_initref */
 148         cfs_atomic_set(&lqs->lqs_refcount, 1); /* 1 for caller */
 149         cfs_atomic_inc(&lqs->lqs_ctxt->lqc_lqs);
 150
 151         /* lqc_lqs_hash will take +1 refcount on lqs on adding */
 152         lqs2 = cfs_hash_findadd_unique(qctxt->lqc_lqs_hash,
 153                                        &lqs->lqs_key, &lqs->lqs_hash);
 154         if (lqs2 == lqs) /* added to hash */
 155                 goto out_put;
 156
 157         create = 0;
 158         lqs_putref(lqs);
 159         lqs = lqs2;
 160
 161  out_put:
 162         cfs_hash_putref(hs);
 163  out:
 164         if (rc != 0) { /* error */
 165                 CERROR("get lqs error(rc: %d)\n", rc);
 166                 return ERR_PTR(rc);
 167         }
 168
 169         if (lqs != NULL) {
 170                 LQS_DEBUG(lqs, "%s\n",
 171                           (create == 1 ? "create lqs" : "search lqs"));
 172         }
 173         return lqs;
 174 }
 175
 176 int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
 177                            struct lustre_quota_ctxt *qctxt)
 178 {
 179         struct lustre_qunit_size *lqs = NULL;
 180         unsigned long *unit, *tune;
 181         signed long tmp = 0;
 182         cfs_time_t time_limit = 0, *shrink;
 183         int i, rc = 0;
 184         ENTRY;
 185
 186         LASSERT(qctxt);
 187         lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
 188                                qctxt, QAQ_IS_CREATE_LQS(oqaq) ? 1 : 0);
 189         if (lqs == NULL || IS_ERR(lqs)){
 190                 CERROR("fail to find a lqs for %sid %u!\n",
 191                        QAQ_IS_GRP(oqaq) ? "g" : "u", oqaq->qaq_id);
 192                 RETURN(PTR_ERR(lqs));
 193         }
 194
 195         CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n",
 196                lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
 197         cfs_spin_lock(&lqs->lqs_lock);
 198         for (i = 0; i < 2; i++) {
 199                 if (i == 0 && !QAQ_IS_ADJBLK(oqaq))
 200                         continue;
 201
 202                 if (i == 1 && !QAQ_IS_ADJINO(oqaq))
 203                         continue;
 204
 205                 tmp = i ? (lqs->lqs_iunit_sz - oqaq->qaq_iunit_sz) :
 206                           (lqs->lqs_bunit_sz - oqaq->qaq_bunit_sz);
 207                 shrink = i ? &lqs->lqs_last_ishrink :
 208                              &lqs->lqs_last_bshrink;
 209                 time_limit = cfs_time_add(i ? lqs->lqs_last_ishrink :
 210                                               lqs->lqs_last_bshrink,
 211                                    cfs_time_seconds(qctxt->lqc_switch_seconds));
 212                 unit = i ? &lqs->lqs_iunit_sz : &lqs->lqs_bunit_sz;
 213                 tune = i ? &lqs->lqs_itune_sz : &lqs->lqs_btune_sz;
 214
 215                 /* quota master shrinks */
 216                 if (qctxt->lqc_handler && tmp > 0)
 217                         *shrink = cfs_time_current();
 218
 219                 /* quota master enlarges */
 220                 if (qctxt->lqc_handler && tmp < 0) {
 221                         /* in case of ping-pong effect, don't enlarge lqs
 222                          * in a short time */
 223                         if (*shrink &&
 224                             cfs_time_before(cfs_time_current(), time_limit))
 225                                 tmp = 0;
 226                 }
 227
 228                 /* when setquota, don't enlarge lqs b=18616 */
 229                 if (QAQ_IS_CREATE_LQS(oqaq) && tmp < 0)
 230                         tmp = 0;
 231
 232                 if (tmp != 0) {
 233                         *unit = i ? oqaq->qaq_iunit_sz : oqaq->qaq_bunit_sz;
 234                         *tune = (*unit) / 2;
 235                 }
 236
 237
 238                 if (tmp > 0)
 239                         rc |= i ? LQS_INO_DECREASE : LQS_BLK_DECREASE;
 240                 if (tmp < 0)
 241                         rc |= i ? LQS_INO_INCREASE : LQS_BLK_INCREASE;
 242         }
 243         cfs_spin_unlock(&lqs->lqs_lock);
 244         CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n",
 245                lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
 246
 247         lqs_putref(lqs);
 248
 249         RETURN(rc);
 250 }
 251
 252 int filter_quota_adjust_qunit(struct obd_export *exp,
 253                               struct quota_adjust_qunit *oqaq,
 254                               struct lustre_quota_ctxt *qctxt,
 255                               struct ptlrpc_request_set *rqset)
 256 {
 257         struct obd_device *obd = exp->exp_obd;
 258         unsigned int id[MAXQUOTAS] = { 0, 0 };
 259         int rc = 0;
 260         ENTRY;
 261
 262         LASSERT(oqaq);
 263         LASSERT(QAQ_IS_ADJBLK(oqaq));
 264         rc = quota_adjust_slave_lqs(oqaq, qctxt);
 265         if (rc < 0) {
 266                 CERROR("adjust mds slave's qunit size failed!(rc:%d)\n", rc);
 267                 RETURN(rc);
 268         }
 269         if (QAQ_IS_GRP(oqaq))
 270                 id[GRPQUOTA] = oqaq->qaq_id;
 271         else
 272                 id[USRQUOTA] = oqaq->qaq_id;
 273
 274         if (rc > 0) {
 275                 rc = qctxt_adjust_qunit(obd, qctxt, id, 1, 0, NULL);
 276                 if (rc == -EDQUOT || rc == -EBUSY ||
 277                     rc == QUOTA_REQ_RETURNED || rc == -EAGAIN) {
 278                         CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 279                         rc = 0;
 280                 }
 281                 if (rc)
 282                         CERROR("slave adjust block quota failed!(rc:%d)\n", rc);
 283         }
 284         RETURN(rc);
 285 }
 286 #endif /* __KERNEL__ */
 287 #endif
 288
 289 int client_quota_adjust_qunit(struct obd_export *exp,
 290                               struct quota_adjust_qunit *oqaq,
 291                               struct lustre_quota_ctxt *qctxt,
 292                               struct ptlrpc_request_set *rqset)
 293 {
 294         struct ptlrpc_request *req;
 295         struct quota_adjust_qunit *oqa;
 296         int rc = 0;
 297         ENTRY;
 298
 299         /* client don't support this kind of operation, abort it */
 300         if (!(exp->exp_connect_flags & OBD_CONNECT_CHANGE_QS)) {
 301                 CDEBUG(D_QUOTA, "osc: %s don't support change qunit size\n",
 302                        exp->exp_obd->obd_name);
 303                 RETURN(rc);
 304         }
 305         if (strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_OSC_NAME))
 306                 RETURN(-EINVAL);
 307
 308         LASSERT(rqset);
 309
 310         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 311                                         &RQF_OST_QUOTA_ADJUST_QUNIT,
 312                                         LUSTRE_OST_VERSION,
 313                                         OST_QUOTA_ADJUST_QUNIT);
 314         if (req == NULL)
 315                 RETURN(-ENOMEM);
 316
 317         oqa = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
 318         *oqa = *oqaq;
 319
 320         ptlrpc_request_set_replen(req);
 321
 322         ptlrpc_set_add_req(rqset, req);
 323         RETURN (rc);
 324 }
 325
 326 int lov_quota_adjust_qunit(struct obd_export *exp,
 327                            struct quota_adjust_qunit *oqaq,
 328                            struct lustre_quota_ctxt *qctxt,
 329                            struct ptlrpc_request_set *rqset)
 330 {
 331         struct obd_device *obd = class_exp2obd(exp);
 332         struct lov_obd *lov = &obd->u.lov;
 333         int i, err, rc = 0;
 334         unsigned no_set = 0;
 335         ENTRY;
 336
 337         if (!QAQ_IS_ADJBLK(oqaq)) {
 338                 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
 339                 RETURN(-EFAULT);
 340         }
 341
 342
 343         if (rqset == NULL) {
 344                 rqset = ptlrpc_prep_set();
 345                 if (!rqset)
 346                         RETURN(-ENOMEM);
 347                 no_set = 1;
 348         }
 349
 350         obd_getref(obd);
 351         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 352
 353                 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active) {
 354                         CDEBUG(D_HA, "ost %d is inactive\n", i);
 355                         continue;
 356                 }
 357
 358                 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
 359                                              NULL, rqset);
 360                 if (err) {
 361                         if (lov->lov_tgts[i]->ltd_active && !rc)
 362                                 rc = err;
 363                         continue;
 364                 }
 365         }
 366
 367         err = ptlrpc_set_wait(rqset);
 368         if (!rc)
 369                 rc = err;
 370
 371         /* Destroy the set if none was provided by the caller */
 372         if (no_set)
 373                 ptlrpc_set_destroy(rqset);
 374
 375         obd_putref(obd);
 376         RETURN(rc);
 377 }