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 /* 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 static struct lustre_qunit_size *
  94 quota_create_lqs(unsigned long long lqs_key, struct lustre_quota_ctxt *qctxt)
  95 {
  96         struct lustre_qunit_size *lqs = NULL;
  97         int rc = 0;
  98
  99         OBD_ALLOC_PTR(lqs);
 100         if (!lqs)
 101                 GOTO(out, rc = -ENOMEM);
 102
 103         lqs->lqs_key = lqs_key;
 104
 105         spin_lock_init(&lqs->lqs_lock);
 106         lqs->lqs_bwrite_pending = 0;
 107         lqs->lqs_iwrite_pending = 0;
 108         lqs->lqs_ino_rec = 0;
 109         lqs->lqs_blk_rec = 0;
 110         lqs->lqs_id = LQS_KEY_ID(lqs->lqs_key);
 111         lqs->lqs_flags = LQS_KEY_GRP(lqs->lqs_key) ? LQUOTA_FLAGS_GRP : 0;
 112         lqs->lqs_bunit_sz = qctxt->lqc_bunit_sz;
 113         lqs->lqs_iunit_sz = qctxt->lqc_iunit_sz;
 114         lqs->lqs_btune_sz = qctxt->lqc_btune_sz;
 115         lqs->lqs_itune_sz = qctxt->lqc_itune_sz;
 116         lqs->lqs_ctxt = qctxt;
 117         if (qctxt->lqc_handler) {
 118                 lqs->lqs_last_bshrink  = 0;
 119                 lqs->lqs_last_ishrink  = 0;
 120         }
 121         lqs_initref(lqs);
 122
 123         spin_lock(&qctxt->lqc_lock);
 124         if (!qctxt->lqc_valid)
 125                 rc = -EBUSY;
 126         else
 127                 rc = lustre_hash_add_unique(qctxt->lqc_lqs_hash,
 128                                     &lqs->lqs_key, &lqs->lqs_hash);
 129         spin_unlock(&qctxt->lqc_lock);
 130
 131         if (!rc)
 132                 lqs_getref(lqs);
 133
 134  out:
 135         if (rc && lqs)
 136                 OBD_FREE_PTR(lqs);
 137
 138         if (rc)
 139                 return ERR_PTR(rc);
 140         else
 141                 return lqs;
 142 }
 143
 144 struct lustre_qunit_size *quota_search_lqs(unsigned long long lqs_key,
 145                                            struct lustre_quota_ctxt *qctxt,
 146                                            int create)
 147 {
 148         struct lustre_qunit_size *lqs;
 149         int rc = 0;
 150
 151  search_lqs:
 152         lqs = lustre_hash_lookup(qctxt->lqc_lqs_hash, &lqs_key);
 153         if (IS_ERR(lqs))
 154                 GOTO(out, rc = PTR_ERR(lqs));
 155
 156         if (create && lqs == NULL) {
 157                 /* if quota_create_lqs is successful, it will get a
 158                  * ref to the lqs. The ref will be released when
 159                  * qctxt_cleanup() or quota is nullified */
 160                 lqs = quota_create_lqs(lqs_key, qctxt);
 161                 if (IS_ERR(lqs))
 162                         rc = PTR_ERR(lqs);
 163                 if (rc == -EALREADY)
 164                         GOTO(search_lqs, rc = 0);
 165                 /* get a reference for the caller when creating lqs
 166                  * successfully */
 167                 if (rc == 0)
 168                         lqs_getref(lqs);
 169         }
 170
 171         if (lqs && rc == 0)
 172                 LQS_DEBUG(lqs, "%s\n",
 173                           (create == 1 ? "create lqs" : "search lqs"));
 174
 175  out:
 176         if (rc == 0) {
 177                 return lqs;
 178         } else {
 179                 CDEBUG(D_ERROR, "get lqs error(rc: %d)\n", rc);
 180                 return ERR_PTR(rc);
 181         }
 182 }
 183
 184 int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
 185                            struct lustre_quota_ctxt *qctxt)
 186 {
 187         struct lustre_qunit_size *lqs = NULL;
 188         unsigned long *lbunit, *liunit, *lbtune, *litune;
 189         signed long b_tmp = 0, i_tmp = 0;
 190         cfs_time_t time_limit = 0;
 191         int rc = 0;
 192         ENTRY;
 193
 194         LASSERT(qctxt);
 195         lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
 196                                qctxt, 0);
 197         if (lqs == NULL || IS_ERR(lqs)){
 198                 CDEBUG(D_ERROR, "fail to find a lqs(%s id: %u)!\n",
 199                        QAQ_IS_GRP(oqaq) ? "group" : "user", oqaq->qaq_id);
 200                 RETURN(PTR_ERR(lqs));
 201         }
 202
 203         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_WITHOUT_CHANGE_QS)) {
 204                 lqs->lqs_bunit_sz = qctxt->lqc_bunit_sz;
 205                 lqs->lqs_btune_sz = qctxt->lqc_btune_sz;
 206                 lqs->lqs_iunit_sz = qctxt->lqc_iunit_sz;
 207                 lqs->lqs_itune_sz = qctxt->lqc_itune_sz;
 208                 lqs_putref(lqs);
 209                 RETURN(0);
 210         }
 211
 212         lbunit = &lqs->lqs_bunit_sz;
 213         liunit = &lqs->lqs_iunit_sz;
 214         lbtune = &lqs->lqs_btune_sz;
 215         litune = &lqs->lqs_itune_sz;
 216
 217         spin_lock(&lqs->lqs_lock);
 218         CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n", *lbunit, *liunit);
 219         /* adjust the slave's block qunit size */
 220         if (QAQ_IS_ADJBLK(oqaq)) {
 221                 cfs_duration_t sec = cfs_time_seconds(qctxt->lqc_switch_seconds);
 222
 223                 b_tmp = *lbunit - oqaq->qaq_bunit_sz;
 224
 225                 if (qctxt->lqc_handler && b_tmp > 0)
 226                         lqs->lqs_last_bshrink = cfs_time_current();
 227
 228                 if (qctxt->lqc_handler && b_tmp < 0) {
 229                         time_limit = cfs_time_add(lqs->lqs_last_bshrink, sec);
 230                         if (!lqs->lqs_last_bshrink ||
 231                             cfs_time_after(cfs_time_current(), time_limit)) {
 232                                 *lbunit = oqaq->qaq_bunit_sz;
 233                                 *lbtune = (*lbunit) / 2;
 234                         } else {
 235                                 b_tmp = 0;
 236                         }
 237                 } else {
 238                         *lbunit = oqaq->qaq_bunit_sz;
 239                         *lbtune = (*lbunit) / 2;
 240                 }
 241         }
 242
 243         /* adjust the slave's file qunit size */
 244         if (QAQ_IS_ADJINO(oqaq)) {
 245                 i_tmp = *liunit - oqaq->qaq_iunit_sz;
 246
 247                 if (qctxt->lqc_handler && i_tmp > 0)
 248                         lqs->lqs_last_ishrink  = cfs_time_current();
 249
 250                 if (qctxt->lqc_handler && i_tmp < 0) {
 251                         time_limit = cfs_time_add(lqs->lqs_last_ishrink,
 252                                                   cfs_time_seconds(qctxt->
 253                                                   lqc_switch_seconds));
 254                         if (!lqs->lqs_last_ishrink ||
 255                             cfs_time_after(cfs_time_current(), time_limit)) {
 256                                 *liunit = oqaq->qaq_iunit_sz;
 257                                 *litune = (*liunit) / 2;
 258                         } else {
 259                                 i_tmp = 0;
 260                         }
 261                 } else {
 262                         *liunit = oqaq->qaq_iunit_sz;
 263                         *litune = (*liunit) / 2;
 264                 }
 265         }
 266         CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n", *lbunit, *liunit);
 267         spin_unlock(&lqs->lqs_lock);
 268
 269         lqs_putref(lqs);
 270         if (b_tmp > 0)
 271                 rc |= LQS_BLK_DECREASE;
 272         else if (b_tmp < 0)
 273                 rc |= LQS_BLK_INCREASE;
 274
 275         if (i_tmp > 0)
 276                 rc |= LQS_INO_DECREASE;
 277         else if (i_tmp < 0)
 278                 rc |= LQS_INO_INCREASE;
 279
 280         RETURN(rc);
 281 }
 282
 283 int filter_quota_adjust_qunit(struct obd_export *exp,
 284                               struct quota_adjust_qunit *oqaq,
 285                               struct lustre_quota_ctxt *qctxt)
 286 {
 287         struct obd_device *obd = exp->exp_obd;
 288         unsigned int uid = 0, gid = 0;
 289         int rc = 0;
 290         ENTRY;
 291
 292         LASSERT(oqaq);
 293         LASSERT(QAQ_IS_ADJBLK(oqaq));
 294         rc = quota_adjust_slave_lqs(oqaq, qctxt);
 295         if (rc < 0) {
 296                 CERROR("adjust mds slave's qunit size failed!(rc:%d)\n", rc);
 297                 RETURN(rc);
 298         }
 299         if (QAQ_IS_GRP(oqaq))
 300                 gid = oqaq->qaq_id;
 301         else
 302                 uid = oqaq->qaq_id;
 303
 304         if (rc > 0) {
 305                 rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, 1, 0, NULL);
 306                 if (rc == -EDQUOT || rc == -EBUSY ||
 307                     rc == QUOTA_REQ_RETURNED || rc == -EAGAIN) {
 308                         CDEBUG(D_QUOTA, "rc: %d.\n", rc);
 309                         rc = 0;
 310                 }
 311                 if (rc)
 312                         CERROR("slave adjust block quota failed!(rc:%d)\n", rc);
 313         }
 314         RETURN(rc);
 315 }
 316 #endif /* __KERNEL__ */
 317 #endif
 318
 319 int client_quota_adjust_qunit(struct obd_export *exp,
 320                               struct quota_adjust_qunit *oqaq,
 321                               struct lustre_quota_ctxt *qctxt)
 322 {
 323         struct ptlrpc_request *req;
 324         struct quota_adjust_qunit *oqa;
 325         __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*oqaq) };
 326         int rc = 0;
 327         ENTRY;
 328
 329         /* client don't support this kind of operation, abort it */
 330         if (!(exp->exp_connect_flags & OBD_CONNECT_CHANGE_QS)||
 331             OBD_FAIL_CHECK(OBD_FAIL_QUOTA_WITHOUT_CHANGE_QS)) {
 332                 CDEBUG(D_QUOTA, "osc: %s don't support change qunit size\n",
 333                        exp->exp_obd->obd_name);
 334                 RETURN(rc);
 335         }
 336         if (strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_OSC_NAME))
 337                 RETURN(-EINVAL);
 338
 339         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OST_VERSION,
 340                               OST_QUOTA_ADJUST_QUNIT, 2, size, NULL);
 341         if (!req)
 342                 GOTO(out, rc = -ENOMEM);
 343
 344         oqa = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*oqaq));
 345         *oqa = *oqaq;
 346
 347         ptlrpc_req_set_repsize(req, 2, size);
 348
 349         rc = ptlrpc_queue_wait(req);
 350         if (rc) {
 351                 CERROR("%s: %s failed: rc = %d\n", exp->exp_obd->obd_name,
 352                        __FUNCTION__, rc);
 353                 GOTO(out, rc);
 354         }
 355         ptlrpc_req_finished(req);
 356 out:
 357         RETURN (rc);
 358 }
 359
 360 int lov_quota_adjust_qunit(struct obd_export *exp,
 361                            struct quota_adjust_qunit *oqaq,
 362                            struct lustre_quota_ctxt *qctxt)
 363 {
 364         struct obd_device *obd = class_exp2obd(exp);
 365         struct lov_obd *lov = &obd->u.lov;
 366         int i, rc = 0;
 367         ENTRY;
 368
 369         if (!QAQ_IS_ADJBLK(oqaq)) {
 370                 CERROR("bad qaq_flags %x for lov obd.\n", oqaq->qaq_flags);
 371                 RETURN(-EFAULT);
 372         }
 373
 374         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
 375                 int err;
 376
 377                 if (!lov->lov_tgts[i]->ltd_active) {
 378                         CDEBUG(D_HA, "ost %d is inactive\n", i);
 379                         continue;
 380                 }
 381
 382                 err = obd_quota_adjust_qunit(lov->lov_tgts[i]->ltd_exp, oqaq,
 383                                              NULL);
 384                 if (err) {
 385                         if (lov->lov_tgts[i]->ltd_active && !rc)
 386                                 rc = err;
 387                         continue;
 388                 }
 389         }
 390         RETURN(rc);
 391 }