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  2008 Sun Microsystems, Inc. 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 <class_hash.h>
  64 #include "quota_internal.h"
  65
  66 #ifdef HAVE_QUOTA_SUPPORT
  67
  68 #ifdef __KERNEL__
  69 /**
  70  * This function is charge of recording lqs_ino_rec and
  71  * lqs_blk_rec. when a lquota slave checks a quota
  72  * request(check_cur_qunit) and finishes a quota
  73  * request(dqacq_completion), it will be called.
  74  * is_chk: whether it is checking quota; otherwise, it is finishing
  75  * is_acq: whether it is acquiring; otherwise, it is releasing
  76  */
  77 void quota_compute_lqs(struct qunit_data *qdata, struct lustre_qunit_size *lqs,
  78                        int is_chk, int is_acq)
  79 {
  80         long long *rec;
  81
  82         LASSERT(qdata && lqs);
  83         LASSERT_SPIN_LOCKED(&lqs->lqs_lock);
  84
  85         rec = QDATA_IS_BLK(qdata) ? &lqs->lqs_blk_rec : &lqs->lqs_ino_rec;
  86
  87         if (!!is_chk + !!is_acq == 1)
  88                 *rec -= qdata->qd_count;
  89         else
  90                 *rec += qdata->qd_count;
  91
  92 }
  93
  94 static struct lustre_qunit_size *
  95 quota_create_lqs(unsigned long long lqs_key, struct lustre_quota_ctxt *qctxt)
  96 {
  97         struct lustre_qunit_size *lqs = NULL;
  98         int rc = 0;
  99
 100         OBD_ALLOC_PTR(lqs);
 101         if (!lqs)
 102                 GOTO(out, rc = -ENOMEM);
 103
 104         lqs->lqs_key = lqs_key;
 105
 106         spin_lock_init(&lqs->lqs_lock);
 107         lqs->lqs_bwrite_pending = 0;
 108         lqs->lqs_iwrite_pending = 0;
 109         lqs->lqs_ino_rec = 0;
 110         lqs->lqs_blk_rec = 0;
 111         lqs->lqs_id = LQS_KEY_ID(lqs->lqs_key);
 112         lqs->lqs_flags = LQS_KEY_GRP(lqs->lqs_key) ? LQUOTA_FLAGS_GRP : 0;
 113         lqs->lqs_bunit_sz = qctxt->lqc_bunit_sz;
 114         lqs->lqs_iunit_sz = qctxt->lqc_iunit_sz;
 115         lqs->lqs_btune_sz = qctxt->lqc_btune_sz;
 116         lqs->lqs_itune_sz = qctxt->lqc_itune_sz;
 117         lqs->lqs_ctxt = qctxt;
 118         if (qctxt->lqc_handler) {
 119                 lqs->lqs_last_bshrink  = 0;
 120                 lqs->lqs_last_ishrink  = 0;
 121         }
 122         lqs_initref(lqs);
 123         rc = lustre_hash_add_unique(qctxt->lqc_lqs_hash, &lqs->lqs_key,
 124                                     &lqs->lqs_hash);
 125         if (!rc)
 126                 lqs_getref(lqs);
 127
 128  out:
 129         if (rc && lqs)
 130                 OBD_FREE_PTR(lqs);
 131
 132         if (rc)
 133                 return ERR_PTR(rc);
 134         else
 135                 return lqs;
 136 }
 137
 138 struct lustre_qunit_size *quota_search_lqs(unsigned long long lqs_key,
 139                                            struct lustre_quota_ctxt *qctxt,
 140                                            int create)
 141 {
 142         int rc = 0;
 143         struct lustre_qunit_size *lqs;
 144
 145  search_lqs:
 146         lqs = lustre_hash_lookup(qctxt->lqc_lqs_hash, &lqs_key);
 147         if (lqs == NULL && create) {
 148                 lqs = quota_create_lqs(lqs_key, qctxt);
 149                 if (IS_ERR(lqs))
 150                         rc = PTR_ERR(lqs);
 151                 if (rc == -EALREADY) {
 152                         rc = 0;
 153                         goto search_lqs;
 154                 }
 155         }
 156
 157         if (lqs)
 158                 LQS_DEBUG(lqs, "%s\n",
 159                           (create == 1 ? "create lqs" : "search lqs"));
 160
 161         if (rc == 0) {
 162                 return lqs;
 163         } else {
 164                 CDEBUG(D_ERROR, "get lqs error(rc: %d)\n", rc);
 165                 return ERR_PTR(rc);
 166         }
 167 }
 168
 169 int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
 170                            struct lustre_quota_ctxt *qctxt)
 171 {
 172         struct lustre_qunit_size *lqs = NULL;
 173         unsigned long *lbunit, *liunit, *lbtune, *litune;
 174         signed long b_tmp = 0, i_tmp = 0;
 175         cfs_time_t time_limit = 0;
 176         int rc = 0;
 177         ENTRY;
 178
 179         LASSERT(qctxt);
 180         lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
 181                                qctxt, 1);
 182         if (lqs == NULL || IS_ERR(lqs))
 183                 RETURN(PTR_ERR(lqs));
 184
 185         /* deleting the lqs, because a user sets lfs quota 0 0 0 0  */
 186         if (!oqaq->qaq_bunit_sz && !oqaq->qaq_iunit_sz && QAQ_IS_ADJBLK(oqaq) &&
 187             QAQ_IS_ADJINO(oqaq)) {
 188                 LQS_DEBUG(lqs, "release lqs\n");
 189                 /* this is for quota_search_lqs */
 190                 lqs_putref(lqs);
 191                 /* kill lqs */
 192                 lqs_putref(lqs);
 193                 RETURN(rc);
 194         }
 195
 196         lbunit = &lqs->lqs_bunit_sz;
 197         liunit = &lqs->lqs_iunit_sz;
 198         lbtune = &lqs->lqs_btune_sz;
 199         litune = &lqs->lqs_itune_sz;
 200
 201         CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n", *lbunit, *liunit);
 202         spin_lock(&lqs->lqs_lock);
 203         /* adjust the slave's block qunit size */
 204         if (QAQ_IS_ADJBLK(oqaq)) {
 205                 cfs_duration_t sec = cfs_time_seconds(qctxt->lqc_switch_seconds);
 206
 207                 b_tmp = *lbunit - oqaq->qaq_bunit_sz;
 208
 209                 if (qctxt->lqc_handler && b_tmp > 0)
 210                         lqs->lqs_last_bshrink = cfs_time_current();
 211
 212                 if (qctxt->lqc_handler && b_tmp < 0) {
 213                         time_limit = cfs_time_add(lqs->lqs_last_bshrink, sec);
 214                         if (!lqs->lqs_last_bshrink ||
 215                             cfs_time_after(cfs_time_current(), time_limit)) {
 216                                 *lbunit = oqaq->qaq_bunit_sz;
 217                                 *lbtune = (*lbunit) / 2;
 218                         } else {
 219                                 b_tmp = 0;
 220                         }
 221                 } else {
 222                         *lbunit = oqaq->qaq_bunit_sz;
 223                         *lbtune = (*lbunit) / 2;
 224                 }
 225         }
 226
 227         /* adjust the slave's file qunit size */
 228         if (QAQ_IS_ADJINO(oqaq)) {
 229                 i_tmp = *liunit - oqaq->qaq_iunit_sz;
 230
 231                 if (qctxt->lqc_handler && i_tmp > 0)
 232                         lqs->lqs_last_ishrink  = cfs_time_current();
 233
 234                 if (qctxt->lqc_handler && i_tmp < 0) {
 235                         time_limit = cfs_time_add(lqs->lqs_last_ishrink,
 236                                                   cfs_time_seconds(qctxt->
 237                                                   lqc_switch_seconds));
 238                         if (!lqs->lqs_last_ishrink ||
 239                             cfs_time_after(cfs_time_current(), time_limit)) {
 240                                 *liunit = oqaq->qaq_iunit_sz;
 241                                 *litune = (*liunit) / 2;
 242                         } else {
 243                                 i_tmp = 0;
 244                         }
 245                 } else {
 246                         *liunit = oqaq->qaq_iunit_sz;
 247                         *litune = (*liunit) / 2;
 248                 }
 249         }
 250         spin_unlock(&lqs->lqs_lock);
 251         CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n", *lbunit, *liunit);
 252
 253         lqs_putref(lqs);
 254
 255         if (b_tmp > 0)
 256                 rc |= LQS_BLK_DECREASE;
 257         else if (b_tmp < 0)
 258                 rc |= LQS_BLK_INCREASE;
 259
 260         if (i_tmp > 0)
 261                 rc |= LQS_INO_DECREASE;
 262         else if (i_tmp < 0)
 263                 rc |= LQS_INO_INCREASE;
 264
 265         RETURN(rc);
 266 }
 267
 268 int filter_quota_adjust_qunit(struct obd_export *exp,
 269                               struct quota_adjust_qunit *oqaq,
 270                               struct lustre_quota_ctxt *qctxt)
 271 {
 272         struct obd_device *obd = exp->exp_obd;
 273         unsigned int id[MAXQUOTAS] = { 0, 0 };
 274         int rc = 0;
 275         ENTRY;
 276
 277         LASSERT(oqaq);
 278         LASSERT(QAQ_IS_ADJBLK(oqaq));
 279         rc = quota_adjust_slave_lqs(oqaq, qctxt);
 280         if (rc < 0) {
 281                 CERROR("adjust mds slave's qunit size failed!(rc:%d)\n", rc);
 282                 RETURN(rc);
 283         }
 284         if (QAQ_IS_GRP(oqaq))
 285                 id[GRPQUOTA] = oqaq->qaq_id;
 286         else
 287                 id[USRQUOTA] = oqaq->qaq_id;
 288
 289         if (rc > 0) {
 290                 rc = qctxt_adjust_qunit(obd, qctxt, id, 1, 0, NULL);
 291                 if (rc == -EDQUOT || rc == -EBUSY ||
 292                     rc == QUOTA_REQ_RETURNED || rc == -EAGAIN) {
 293                         CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 294                         rc = 0;
 295                 }
 296                 if (rc)
 297                         CERROR("slave adjust block quota failed!(rc:%d)\n", rc);
 298         }
 299         RETURN(rc);
 300 }
 301 #endif /* __KERNEL__ */
 302 #endif
 303
 304 int client_quota_adjust_qunit(struct obd_export *exp,
 305                               struct quota_adjust_qunit *oqaq,
 306                               struct lustre_quota_ctxt *qctxt)
 307 {
 308         struct ptlrpc_request *req;
 309         struct quota_adjust_qunit *oqa;
 310         int rc = 0;
 311         ENTRY;
 312
 313         /* client don't support this kind of operation, abort it */
 314         if (!(exp->exp_connect_flags & OBD_CONNECT_CHANGE_QS)) {
 315                 CDEBUG(D_QUOTA, "osc: %s don't support change qunit size\n",
 316                        exp->exp_obd->obd_name);
 317                 RETURN(rc);
 318         }
 319         if (strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_OSC_NAME))
 320                 RETURN(-EINVAL);
 321
 322         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 323                                         &RQF_OST_QUOTA_ADJUST_QUNIT,
 324                                         LUSTRE_OST_VERSION,
 325                                         OST_QUOTA_ADJUST_QUNIT);
 326         if (req == NULL)
 327                 RETURN(-ENOMEM);
 328
 329         oqa = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
 330         *oqa = *oqaq;
 331
 332         ptlrpc_request_set_replen(req);
 333
 334         rc = ptlrpc_queue_wait(req);
 335         if (rc)
 336                 CERROR("%s: %s failed: rc = %d\n", exp->exp_obd->obd_name,
 337                        __FUNCTION__, rc);
 338         ptlrpc_req_finished(req);
 339         RETURN (rc);
 340 }
 341
 342 int lov_quota_adjust_qunit(struct obd_export *exp,
 343                            struct quota_adjust_qunit *oqaq,
 344                            struct lustre_quota_ctxt *qctxt)
 345 {
 346         struct obd_device *obd = class_exp2obd(exp);
 347         struct lov_obd *lov = &obd->u.lov;
 348         int i, rc = 0;
 349         ENTRY;
 350
 351         if (!QAQ_IS_ADJBLK(oqaq)) {
 352                 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
 353                 RETURN(-EFAULT);
 354         }
 355
 356         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 357                 int err;
 358
 359                 if (!lov->lov_tgts[i]->ltd_active) {
 360                         CDEBUG(D_HA, "ost %d is inactive\n", i);
 361                         continue;
 362                 }
 363
 364                 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
 365                                              NULL);
 366                 if (err) {
 367                         if (lov->lov_tgts[i]->ltd_active && !rc)
 368                                 rc = err;
 369                         continue;
 370                 }
 371         }
 372         RETURN(rc);
 373 }