lustre/quota/qsd_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, 2017, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 /*
  32  * Quota Slave Driver (QSD) management.
  33  *
  34  * The quota slave feature is implemented under the form of a library called
  35  * QSD. Each OSD device should create a QSD instance via qsd_init() which will
  36  * be used to manage quota enforcement for this device. This implies:
  37  * - completing the reintegration procedure with the quota master (aka QMT, see
  38  *   qmt_dev.c) to retrieve the latest quota settings and space distribution.
  39  * - managing quota locks in order to be notified of configuration changes.
  40  * - acquiring space from the QMT when quota space for a given user/group is
  41  *   close to exhaustion.
  42  * - allocating quota space to service threads for local request processing.
  43  *
  44  * Once the QSD instance created, the OSD device should invoke qsd_start()
  45  * when recovery is completed. This notifies the QSD that we are about to
  46  * process new requests on which quota should be strictly enforced.
  47  * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
  48  * for/after each operation until shutdown where the QSD instance should be
  49  * freed via qsd_fini().
  50  */
  51
  52 #define DEBUG_SUBSYSTEM S_LQUOTA
  53
  54 #include <obd_class.h>
  55 #include "qsd_internal.h"
  56
  57 struct kmem_cache *upd_kmem;
  58
  59 struct lu_kmem_descr qsd_caches[] = {
  60         {
  61                 .ckd_cache = &upd_kmem,
  62                 .ckd_name  = "upd_kmem",
  63                 .ckd_size  = sizeof(struct qsd_upd_rec)
  64         },
  65         {
  66                 .ckd_cache = NULL
  67         }
  68 };
  69
  70 /* define qsd thread key */
  71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  73 LU_KEY_INIT_GENERIC(qsd);
  74
  75 /* some procfs helpers */
  76 static int qsd_state_seq_show(struct seq_file *m, void *data)
  77 {
  78         struct qsd_instance     *qsd = m->private;
  79         char                     enabled[5];
  80
  81         LASSERT(qsd != NULL);
  82
  83         memset(enabled, 0, sizeof(enabled));
  84         if (qsd_type_enabled(qsd, USRQUOTA))
  85                 strcat(enabled, "u");
  86         if (qsd_type_enabled(qsd, GRPQUOTA))
  87                 strcat(enabled, "g");
  88         if (qsd_type_enabled(qsd, PRJQUOTA))
  89                 strncat(enabled, "p", 1);
  90         if (strlen(enabled) == 0)
  91                 strcat(enabled, "none");
  92
  93         seq_printf(m, "target name:    %s\n"
  94                    "pool ID:        %d\n"
  95                    "type:           %s\n"
  96                    "quota enabled:  %s\n"
  97                    "conn to master: %s\n",
  98                    qsd->qsd_svname, qsd->qsd_pool_id,
  99                    qsd->qsd_is_md ? "md" : "dt", enabled,
 100                    qsd->qsd_exp_valid ? "setup" : "not setup yet");
 101
 102         if (qsd->qsd_prepared) {
 103                 memset(enabled, 0, sizeof(enabled));
 104                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 105                         strcat(enabled, "u");
 106                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 107                         strcat(enabled, "g");
 108                 if (qsd->qsd_type_array[PRJQUOTA]->qqi_acct_obj != NULL)
 109                         strncat(enabled, "p", 1);
 110                 if (strlen(enabled) == 0)
 111                         strcat(enabled, "none");
 112                 seq_printf(m, "space acct:     %s\n"
 113                            "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 114                            "group uptodate: glb[%d],slv[%d],reint[%d]\n"
 115                            "project uptodate: glb[%d],slv[%d],reint[%d]\n",
 116                            enabled,
 117                            qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 118                            qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 119                            qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 120                            qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 121                            qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 122                            qsd->qsd_type_array[GRPQUOTA]->qqi_reint,
 123                            qsd->qsd_type_array[PRJQUOTA]->qqi_glb_uptodate,
 124                            qsd->qsd_type_array[PRJQUOTA]->qqi_slv_uptodate,
 125                            qsd->qsd_type_array[PRJQUOTA]->qqi_reint);
 126         }
 127         return 0;
 128 }
 129 LPROC_SEQ_FOPS_RO(qsd_state);
 130
 131 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 132 {
 133         struct qsd_instance     *qsd = m->private;
 134         char                     enabled[5];
 135
 136         LASSERT(qsd != NULL);
 137
 138         memset(enabled, 0, sizeof(enabled));
 139         if (qsd_type_enabled(qsd, USRQUOTA))
 140                 strcat(enabled, "u");
 141         if (qsd_type_enabled(qsd, GRPQUOTA))
 142                 strcat(enabled, "g");
 143         if (qsd_type_enabled(qsd, PRJQUOTA))
 144                 strncat(enabled, "p", 1);
 145         if (strlen(enabled) == 0)
 146                 strcat(enabled, "none");
 147
 148         seq_printf(m, "%s\n", enabled);
 149         return 0;
 150 }
 151
 152 static ssize_t qsd_enabled_seq_write(struct file *file,
 153                                      const char __user *buffer,
 154                                      size_t count, loff_t *off)
 155 {
 156         struct seq_file *m = file->private_data;
 157         struct qsd_instance *qsd = m->private;
 158         char fsname[LUSTRE_MAXFSNAME];
 159         int enabled = 0;
 160         char valstr[5];
 161         int pool, rc;
 162
 163         if (count > 4)
 164                 return -E2BIG;
 165
 166         if (copy_from_user(valstr, buffer, count))
 167                 GOTO(out, count = -EFAULT);
 168
 169         if (strchr(valstr, 'u'))
 170                 enabled |= BIT(USRQUOTA);
 171         if (strchr(valstr, 'g'))
 172                 enabled |= BIT(GRPQUOTA);
 173         if (strchr(valstr, 'p'))
 174                 enabled |= BIT(PRJQUOTA);
 175
 176         if (enabled == 0 && strcmp(valstr, "none"))
 177                 GOTO(out, count = -EINVAL);
 178
 179         if (qsd->qsd_is_md)
 180                 pool = LQUOTA_RES_MD;
 181         else
 182                 pool = LQUOTA_RES_DT;
 183
 184         if (server_name2fsname(qsd->qsd_svname, fsname, NULL))
 185                 GOTO(out, count = -EINVAL);
 186
 187         rc = qsd_config(valstr, fsname, pool);
 188         if (rc)
 189                 count = rc;
 190 out:
 191         return count;
 192 }
 193 LPROC_SEQ_FOPS(qsd_enabled);
 194
 195 /* force reintegration procedure to be executed.
 196  * Used for test/debugging purpose */
 197 static ssize_t
 198 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 199                                 size_t count, loff_t *off)
 200 {
 201         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 202         int                  rc = 0, qtype;
 203
 204         LASSERT(qsd != NULL);
 205
 206         write_lock(&qsd->qsd_lock);
 207         if (qsd->qsd_stopping) {
 208                 /* don't mess up with shutdown procedure, it is already
 209                  * complicated enough */
 210                 rc = -ESHUTDOWN;
 211         } else if (!qsd->qsd_prepared) {
 212                 rc = -EAGAIN;
 213         } else {
 214                 /* mark all indexes as stale */
 215                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 216                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 217                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 218                 }
 219         }
 220         write_unlock(&qsd->qsd_lock);
 221
 222         if (rc)
 223                 return rc;
 224
 225         /* kick off reintegration */
 226         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 227                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 228                 if (rc)
 229                         break;
 230         }
 231         return rc == 0 ? count : rc;
 232 }
 233 LPROC_SEQ_FOPS_WR_ONLY(qsd, force_reint);
 234
 235 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 236 {
 237         struct qsd_instance *qsd = m->private;
 238         LASSERT(qsd != NULL);
 239
 240         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 241         return 0;
 242 }
 243
 244 static ssize_t
 245 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 246                         size_t count, loff_t *off)
 247 {
 248         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 249         time64_t timeout;
 250         int rc;
 251
 252         LASSERT(qsd != NULL);
 253         rc = kstrtoll_from_user(buffer, count, 0, &timeout);
 254         if (rc)
 255                 return rc;
 256
 257         if (timeout < 0)
 258                 return -EINVAL;
 259
 260         qsd->qsd_timeout = timeout;
 261         return count;
 262 }
 263 LPROC_SEQ_FOPS(qsd_timeout);
 264
 265 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 266         { .name =       "info",
 267           .fops =       &qsd_state_fops         },
 268         { .name =       "enabled",
 269           .fops =       &qsd_enabled_fops       },
 270         { .name =       "force_reint",
 271           .fops =       &qsd_force_reint_fops   },
 272         { .name =       "timeout",
 273           .fops =       &qsd_timeout_fops       },
 274         { NULL }
 275 };
 276
 277 /*
 278  * Callback function invoked by the OSP layer when the connection to the master
 279  * has been set up.
 280  *
 281  * \param data - is a pointer to the qsd_instance
 282  *
 283  * \retval - 0 on success, appropriate error on failure
 284  */
 285 static int qsd_conn_callback(void *data)
 286 {
 287         struct qsd_instance *qsd = (struct qsd_instance *)data;
 288         int                  type;
 289         ENTRY;
 290
 291         /* qsd_exp should now be valid */
 292         LASSERT(qsd->qsd_exp);
 293
 294         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 295
 296         write_lock(&qsd->qsd_lock);
 297         /* notify that qsd_exp is now valid */
 298         qsd->qsd_exp_valid = true;
 299         write_unlock(&qsd->qsd_lock);
 300
 301         /* Now that the connection to master is setup, we can initiate the
 302          * reintegration procedure for quota types which are enabled.
 303          * It is worth noting that, if the qsd_instance hasn't been started
 304          * already, then we can only complete the first two steps of the
 305          * reintegration procedure (i.e. global lock enqueue and slave
 306          * index transfer) since the space usage reconciliation (i.e.
 307          * step 3) will have to wait for qsd_start() to be called */
 308         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 309                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 310                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 311         }
 312
 313         RETURN(0);
 314 }
 315
 316 /*
 317  * Release qsd_qtype_info structure which contains data associated with a
 318  * given quota type. This releases the accounting objects.
 319  * It's called on OSD cleanup when the qsd instance is released.
 320  *
 321  * \param env - is the environment passed by the caller
 322  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 323  *              to be released
 324  * \param qtype - is the quota type to be shutdown
 325  */
 326 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 327                            int qtype)
 328 {
 329         struct qsd_qtype_info   *qqi;
 330         int repeat = 0;
 331         ENTRY;
 332
 333         if (qsd->qsd_type_array[qtype] == NULL)
 334                 RETURN_EXIT;
 335         qqi = qsd->qsd_type_array[qtype];
 336         qsd->qsd_type_array[qtype] = NULL;
 337
 338         /* all deferred work lists should be empty */
 339         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 340         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 341
 342         /* shutdown lquota site */
 343         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 344                 lquota_site_free(env, qqi->qqi_site);
 345                 qqi->qqi_site = NULL;
 346         }
 347
 348         /* The qqi may still be holding by global locks which are being
 349          * canceled asynchronously (LU-4365), see the following steps:
 350          *
 351          * - On server umount, we try to clear all quota locks first by
 352          *   disconnecting LWP (which will invalidate import and cleanup
 353          *   all locks on it), however, if quota reint process is holding
 354          *   the global lock for reintegration at that time, global lock
 355          *   will fail to be cleared on LWP disconnection.
 356          *
 357          * - Umount process goes on and stops reint process, the global
 358          *   lock will be dropped on reint process exit, however, the lock
 359          *   cancel in done in asynchronous way, so the
 360          *   qsd_glb_blocking_ast() might haven't been called yet when we
 361          *   get here.
 362          */
 363         while (atomic_read(&qqi->qqi_ref) > 1) {
 364                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 365                        atomic_read(&qqi->qqi_ref), repeat);
 366                 repeat++;
 367                 set_current_state(TASK_INTERRUPTIBLE);
 368                 schedule_timeout(cfs_time_seconds(1));
 369         }
 370
 371         /* by now, all qqi users should have gone away */
 372         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 373         lu_ref_fini(&qqi->qqi_reference);
 374
 375         /* release accounting object */
 376         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 377                 dt_object_put(env, qqi->qqi_acct_obj);
 378                 qqi->qqi_acct_obj = NULL;
 379         }
 380
 381         /* release slv index */
 382         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 383                 dt_object_put(env, qqi->qqi_slv_obj);
 384                 qqi->qqi_slv_obj = NULL;
 385                 qqi->qqi_slv_ver = 0;
 386         }
 387
 388         /* release global index */
 389         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 390                 dt_object_put(env, qqi->qqi_glb_obj);
 391                 qqi->qqi_glb_obj = NULL;
 392                 qqi->qqi_glb_ver = 0;
 393         }
 394
 395         OBD_FREE_PTR(qqi);
 396         EXIT;
 397 }
 398
 399 static const char *qtype2acct_name(int qtype)
 400 {
 401         static char unknown[24];
 402
 403         switch (qtype) {
 404         case USRQUOTA:
 405                 return "acct_user";
 406         case GRPQUOTA:
 407                 return "acct_group";
 408         case PRJQUOTA:
 409                 return "acct_project";
 410         }
 411
 412         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 413         return unknown;
 414 }
 415
 416 static const char *qtype2glb_name(int qtype)
 417 {
 418         static char unknown[24];
 419
 420         switch (qtype) {
 421         case USRQUOTA:
 422                 return "limit_user";
 423         case GRPQUOTA:
 424                 return "limit_group";
 425         case PRJQUOTA:
 426                 return "limit_project";
 427         }
 428
 429         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 430         return unknown;
 431 }
 432
 433 /*
 434  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 435  * This opens the accounting object and initializes the proc file.
 436  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 437  * instance.
 438  *
 439  * \param env  - the environment passed by the caller
 440  * \param qsd  - is the qsd instance which will be in charge of the new
 441  *               qsd_qtype_info instance.
 442  * \param qtype - is quota type to set up
 443  *
 444  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 445  *           appropriate error on failure
 446  */
 447 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 448                           int qtype)
 449 {
 450         struct qsd_qtype_info   *qqi;
 451         int                      rc;
 452         struct obd_uuid          uuid;
 453         ENTRY;
 454
 455         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 456
 457         /* allocate structure for this quota type */
 458         OBD_ALLOC_PTR(qqi);
 459         if (qqi == NULL)
 460                 RETURN(-ENOMEM);
 461         qsd->qsd_type_array[qtype] = qqi;
 462         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 463
 464         /* set backpointer and other parameters */
 465         qqi->qqi_qsd   = qsd;
 466         qqi->qqi_qtype = qtype;
 467         lu_ref_init(&qqi->qqi_reference);
 468         qqi->qqi_glb_uptodate = false;
 469         qqi->qqi_slv_uptodate = false;
 470         qqi->qqi_reint        = false;
 471         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 472         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 473         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 474         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 475         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id,
 476                             QSD_RES_TYPE(qsd), qtype);
 477
 478         /* open accounting object */
 479         LASSERT(qqi->qqi_acct_obj == NULL);
 480         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 481         if (IS_ERR(qqi->qqi_acct_obj)) {
 482                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 483                        qsd->qsd_svname, qtype_name(qtype),
 484                        PTR_ERR(qqi->qqi_acct_obj));
 485                 qqi->qqi_acct_obj = NULL;
 486                 qqi->qqi_acct_failed = true;
 487         }
 488
 489         /* open global index copy */
 490         LASSERT(qqi->qqi_glb_obj == NULL);
 491         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 492                                                        qsd->qsd_root,
 493                                                        &qqi->qqi_fid, true);
 494         if (IS_ERR(qqi->qqi_glb_obj)) {
 495                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 496                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 497                        PTR_ERR(qqi->qqi_glb_obj));
 498                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 499         }
 500         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 501
 502         /* open slave index copy */
 503         LASSERT(qqi->qqi_slv_obj == NULL);
 504         obd_str2uuid(&uuid, qsd->qsd_svname);
 505         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 506                                                        qsd->qsd_root,
 507                                                        &qqi->qqi_fid, &uuid,
 508                                                        true);
 509         if (IS_ERR(qqi->qqi_slv_obj)) {
 510                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 511                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 512                        PTR_ERR(qqi->qqi_slv_obj));
 513                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 514         }
 515         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 516
 517         /* allocate site */
 518         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 519         if (IS_ERR(qqi->qqi_site)) {
 520                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 521                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 522                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 523         }
 524
 525         /* register proc entry for accounting & global index copy objects */
 526         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 527                                 0444, &lprocfs_quota_seq_fops,
 528                                 qqi->qqi_acct_obj);
 529         if (rc) {
 530                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 531                        qsd->qsd_svname, rc);
 532                 GOTO(out, rc);
 533         }
 534
 535         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 536                                 0444, &lprocfs_quota_seq_fops,
 537                                 qqi->qqi_glb_obj);
 538         if (rc) {
 539                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 540                        qsd->qsd_svname, rc);
 541                 GOTO(out, rc);
 542         }
 543         EXIT;
 544 out:
 545         if (rc)
 546                 qsd_qtype_fini(env, qsd, qtype);
 547         return rc;
 548 }
 549
 550 /*
 551  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 552  * structures associated with the quota slave (on-disk objects, lquota entry
 553  * tables, ...).
 554  * This function should be called when the OSD is shutting down.
 555  *
 556  * \param env - is the environment passed by the caller
 557  * \param qsd - is the qsd instance to shutdown
 558  */
 559 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 560 {
 561         int     qtype;
 562         ENTRY;
 563
 564         if (unlikely(qsd == NULL))
 565                 RETURN_EXIT;
 566
 567         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 568         write_lock(&qsd->qsd_lock);
 569         qsd->qsd_stopping = true;
 570         write_unlock(&qsd->qsd_lock);
 571
 572         /* remove qsd proc entry */
 573         if (qsd->qsd_proc != NULL) {
 574                 lprocfs_remove(&qsd->qsd_proc);
 575                 qsd->qsd_proc = NULL;
 576         }
 577
 578         /* stop the writeback thread */
 579         qsd_stop_upd_thread(qsd);
 580
 581         /* shutdown the reintegration threads */
 582         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 583                 if (qsd->qsd_type_array[qtype] == NULL)
 584                         continue;
 585                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 586         }
 587
 588         if (qsd->qsd_ns != NULL) {
 589                 qsd->qsd_ns = NULL;
 590         }
 591
 592         /* free per-quota type data */
 593         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 594                 qsd_qtype_fini(env, qsd, qtype);
 595
 596         /* deregister connection to the quota master */
 597         qsd->qsd_exp_valid = false;
 598         lustre_deregister_lwp_item(&qsd->qsd_exp);
 599
 600         /* release per-filesystem information */
 601         if (qsd->qsd_fsinfo != NULL) {
 602                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 603                 /* remove from the list of fsinfo */
 604                 list_del_init(&qsd->qsd_link);
 605                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 606                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 607                 qsd->qsd_fsinfo = NULL;
 608         }
 609
 610         /* release quota root directory */
 611         if (qsd->qsd_root != NULL) {
 612                 dt_object_put(env, qsd->qsd_root);
 613                 qsd->qsd_root = NULL;
 614         }
 615
 616         /* release reference on dt_device */
 617         if (qsd->qsd_dev != NULL) {
 618                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 619                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 620                 qsd->qsd_dev = NULL;
 621         }
 622
 623         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 624         OBD_FREE_PTR(qsd);
 625         EXIT;
 626 }
 627 EXPORT_SYMBOL(qsd_fini);
 628
 629 /*
 630  * Create a new qsd_instance to be associated with backend osd device
 631  * identified by \dev.
 632  *
 633  * \param env    - the environment passed by the caller
 634  * \param svname - is the service name of the OSD device creating this instance
 635  * \param dev    - is the dt_device where to store quota index files
 636  * \param osd_proc - is the procfs parent directory where to create procfs file
 637  *                   related to this new qsd instance
 638  *
 639  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 640  *           appropriate error on failure
 641  */
 642 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 643                               struct dt_device *dev,
 644                               struct proc_dir_entry *osd_proc)
 645 {
 646         struct qsd_thread_info  *qti = qsd_info(env);
 647         struct qsd_instance     *qsd;
 648         int                      rc, type, idx;
 649         ENTRY;
 650
 651         /* only configure qsd for MDT & OST */
 652         type = server_name2index(svname, &idx, NULL);
 653         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 654                 RETURN(NULL);
 655
 656         /* allocate qsd instance */
 657         OBD_ALLOC_PTR(qsd);
 658         if (qsd == NULL)
 659                 RETURN(ERR_PTR(-ENOMEM));
 660
 661         /* generic initializations */
 662         rwlock_init(&qsd->qsd_lock);
 663         INIT_LIST_HEAD(&qsd->qsd_link);
 664         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 665         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 666         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 667         spin_lock_init(&qsd->qsd_adjust_lock);
 668         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 669         qsd->qsd_prepared = false;
 670         qsd->qsd_started = false;
 671
 672         /* copy service name */
 673         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 674             >= sizeof(qsd->qsd_svname))
 675                 GOTO(out, rc = -E2BIG);
 676
 677         /* grab reference on osd device */
 678         lu_device_get(&dev->dd_lu_dev);
 679         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 680         qsd->qsd_dev = dev;
 681
 682         /* we only support pool ID 0 (default data or metadata pool) for the
 683          * time being. A different pool ID could be assigned to this target via
 684          * the configuration log in the future */
 685         qsd->qsd_pool_id  = 0;
 686
 687         /* get fsname from svname */
 688         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 689         if (rc) {
 690                 CERROR("%s: fail to extract filesystem name\n", svname);
 691                 GOTO(out, rc);
 692         }
 693
 694         /* look up quota setting for the filesystem the target belongs to */
 695         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 696         if (qsd->qsd_fsinfo == NULL) {
 697                 CERROR("%s: failed to locate filesystem information\n", svname);
 698                 GOTO(out, rc = -EINVAL);
 699         }
 700
 701         /* add in the list of lquota_fsinfo */
 702         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 703         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 704         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 705
 706         /* register procfs directory */
 707         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 708                                          lprocfs_quota_qsd_vars, qsd);
 709         if (IS_ERR(qsd->qsd_proc)) {
 710                 rc = PTR_ERR(qsd->qsd_proc);
 711                 qsd->qsd_proc = NULL;
 712                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 713                        svname, rc);
 714                 GOTO(out, rc);
 715         }
 716         EXIT;
 717 out:
 718         if (rc) {
 719                 qsd_fini(env, qsd);
 720                 return ERR_PTR(rc);
 721         }
 722         RETURN(qsd);
 723 }
 724 EXPORT_SYMBOL(qsd_init);
 725
 726 /*
 727  * Initialize on-disk structures in order to manage quota enforcement for
 728  * the target associated with the qsd instance \qsd and starts the reintegration
 729  * procedure for each quota type as soon as possible.
 730  * The last step of the reintegration will be completed once qsd_start() is
 731  * called, at which points the space reconciliation with the master will be
 732  * executed.
 733  * This function must be called when the server stack is fully configured,
 734  * typically when ->ldo_prepare is called across the stack.
 735  *
 736  * \param env - the environment passed by the caller
 737  * \param qsd - is qsd_instance to prepare
 738  *
 739  * \retval - 0 on success, appropriate error on failure
 740  */
 741 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 742 {
 743         struct qsd_thread_info  *qti = qsd_info(env);
 744         int                      qtype, rc = 0;
 745         ENTRY;
 746
 747         if (unlikely(qsd == NULL))
 748                 RETURN(0);
 749
 750         read_lock(&qsd->qsd_lock);
 751         if (qsd->qsd_prepared) {
 752                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 753                 rc = -EALREADY;
 754         }
 755         read_unlock(&qsd->qsd_lock);
 756         if (rc)
 757                 RETURN(rc);
 758
 759         /* Record whether this qsd instance is managing quota enforcement for a
 760          * MDT (i.e. inode quota) or OST (block quota) */
 761         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 762                 qsd->qsd_is_md = true;
 763                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 764         } else {
 765                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 766         }
 767
 768         /* look-up on-disk directory for the quota slave */
 769         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 770                                                     QSD_DIR);
 771         if (IS_ERR(qsd->qsd_root)) {
 772                 rc = PTR_ERR(qsd->qsd_root);
 773                 qsd->qsd_root = NULL;
 774                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 775                        qsd->qsd_svname, rc);
 776                 RETURN(rc);
 777         }
 778
 779         /* initialize per-quota type data */
 780         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 781                 rc = qsd_qtype_init(env, qsd, qtype);
 782                 if (rc)
 783                         RETURN(rc);
 784         }
 785
 786         /* pools successfully setup, mark the qsd as prepared */
 787         write_lock(&qsd->qsd_lock);
 788         qsd->qsd_prepared = true;
 789         write_unlock(&qsd->qsd_lock);
 790
 791         if (qsd->qsd_dev->dd_rdonly)
 792                 RETURN(0);
 793
 794         /* start reintegration thread for each type, if required */
 795         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 796                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 797
 798                 if (qsd_type_enabled(qsd, qtype) &&
 799                     qqi->qqi_acct_failed) {
 800                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 801                                        "since space accounting isn't functional"
 802                                        ". Please run tunefs.lustre --quota on "
 803                                        "an unmounted filesystem if not done "
 804                                        "already\n", qsd->qsd_svname);
 805                         break;
 806                 }
 807
 808                 rc = qsd_start_reint_thread(qqi);
 809                 if (rc) {
 810                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 811                                 qsd->qsd_svname, qtype_name(qtype), rc);
 812                         RETURN(rc);
 813                 }
 814         }
 815
 816         /* start writeback thread */
 817         rc = qsd_start_upd_thread(qsd);
 818         if (rc) {
 819                 CERROR("%s: failed to start writeback thread (%d)\n",
 820                        qsd->qsd_svname, rc);
 821                 RETURN(rc);
 822         }
 823
 824         /* generate osp name */
 825         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 826                                MTI_NAME_MAXLEN, 0);
 827         if (rc) {
 828                 CERROR("%s: failed to generate ospname (%d)\n",
 829                        qsd->qsd_svname, rc);
 830                 RETURN(rc);
 831         }
 832
 833         /* the connection callback will start the reintegration
 834          * procedure if quota is enabled */
 835         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 836                                       qsd_conn_callback, (void *)qsd);
 837         if (rc) {
 838                 CERROR("%s: fail to get connection to master (%d)\n",
 839                        qsd->qsd_svname, rc);
 840                 RETURN(rc);
 841         }
 842
 843         RETURN(0);
 844 }
 845 EXPORT_SYMBOL(qsd_prepare);
 846
 847 /*
 848  * Start a qsd instance. This will complete the last step of the reintegration
 849  * procedure as soon as possible (provided that the master is reachable).
 850  * This should be called when recovery has been completed and quota should now
 851  * be enforced on every operations.
 852  *
 853  * \param env - the environment passed by the caller
 854  * \param qsd - is the qsd instance associated with the osd device to start
 855  */
 856 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 857 {
 858         int     type, rc = 0;
 859         ENTRY;
 860
 861         if (unlikely(qsd == NULL))
 862                 RETURN(0);
 863
 864         write_lock(&qsd->qsd_lock);
 865         if (!qsd->qsd_prepared) {
 866                 CERROR("%s: can't start qsd instance since it wasn't properly "
 867                        "initialized\n", qsd->qsd_svname);
 868                 rc = -EFAULT;
 869         } else if (qsd->qsd_started) {
 870                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 871                 rc = -EALREADY;
 872         } else {
 873                 /* notify that the qsd_instance is now started */
 874                 qsd->qsd_started = true;
 875         }
 876         write_unlock(&qsd->qsd_lock);
 877
 878         if (rc)
 879                 RETURN(rc);
 880
 881         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 882          * up to usage; If usage < granted, release down to usage.  */
 883         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 884                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 885                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 886         }
 887
 888         RETURN(rc);
 889 }
 890 EXPORT_SYMBOL(qsd_start);
 891
 892 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 893
 894 /*
 895  * Global initialization performed at module load time
 896  */
 897 int qsd_glb_init(void)
 898 {
 899         int     rc;
 900
 901         rc = lu_kmem_init(qsd_caches);
 902         if (rc)
 903                 return rc;
 904
 905         qsd_key_init_generic(&qsd_thread_key, NULL);
 906         lu_context_key_register(&qsd_thread_key);
 907         lustre_register_quota_process_config(qsd_process_config);
 908
 909         return 0;
 910 }
 911
 912 /*
 913  * Companion of qsd_glb_init() called at module unload time
 914  */
 915 void qsd_glb_fini(void)
 916 {
 917         lustre_register_quota_process_config(NULL);
 918         lu_kmem_fini(qsd_caches);
 919         lu_context_key_degister(&qsd_thread_key);
 920 }