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         /* TODO: further pool ID should be removed or
  94          * replaced with pool Name */
  95         seq_printf(m, "target name:    %s\n"
  96                    "pool ID:        %d\n"
  97                    "type:           %s\n"
  98                    "quota enabled:  %s\n"
  99                    "conn to master: %s\n",
 100                    qsd->qsd_svname, 0,
 101                    qsd->qsd_is_md ? "md" : "dt", enabled,
 102                    qsd->qsd_exp_valid ? "setup" : "not setup yet");
 103
 104         if (qsd->qsd_prepared) {
 105                 memset(enabled, 0, sizeof(enabled));
 106                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 107                         strcat(enabled, "u");
 108                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 109                         strcat(enabled, "g");
 110                 if (qsd->qsd_type_array[PRJQUOTA]->qqi_acct_obj != NULL)
 111                         strncat(enabled, "p", 1);
 112                 if (strlen(enabled) == 0)
 113                         strcat(enabled, "none");
 114                 seq_printf(m, "space acct:     %s\n"
 115                            "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 116                            "group uptodate: glb[%d],slv[%d],reint[%d]\n"
 117                            "project uptodate: glb[%d],slv[%d],reint[%d]\n",
 118                            enabled,
 119                            qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 120                            qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 121                            qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 122                            qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 123                            qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 124                            qsd->qsd_type_array[GRPQUOTA]->qqi_reint,
 125                            qsd->qsd_type_array[PRJQUOTA]->qqi_glb_uptodate,
 126                            qsd->qsd_type_array[PRJQUOTA]->qqi_slv_uptodate,
 127                            qsd->qsd_type_array[PRJQUOTA]->qqi_reint);
 128         }
 129         return 0;
 130 }
 131 LPROC_SEQ_FOPS_RO(qsd_state);
 132
 133 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 134 {
 135         struct qsd_instance     *qsd = m->private;
 136         char                     enabled[5];
 137
 138         LASSERT(qsd != NULL);
 139
 140         memset(enabled, 0, sizeof(enabled));
 141         if (qsd_type_enabled(qsd, USRQUOTA))
 142                 strncat(enabled, "u", sizeof(enabled) - strlen(enabled));
 143         if (qsd_type_enabled(qsd, GRPQUOTA))
 144                 strncat(enabled, "g", sizeof(enabled) - strlen(enabled));
 145         if (qsd_type_enabled(qsd, PRJQUOTA))
 146                 strncat(enabled, "p", sizeof(enabled) - strlen(enabled));
 147         if (strlen(enabled) == 0)
 148                 strncat(enabled, "none", sizeof(enabled) - strlen(enabled));
 149
 150         seq_printf(m, "%s\n", enabled);
 151         return 0;
 152 }
 153
 154 static ssize_t qsd_enabled_seq_write(struct file *file,
 155                                      const char __user *buffer,
 156                                      size_t count, loff_t *off)
 157 {
 158         struct seq_file *m = file->private_data;
 159         struct qsd_instance *qsd = m->private;
 160         char fsname[LUSTRE_MAXFSNAME + 1];
 161         int enabled = 0;
 162         char valstr[5];
 163         int pool, rc;
 164
 165         if (count > 4)
 166                 return -E2BIG;
 167
 168         if (copy_from_user(valstr, buffer, count))
 169                 GOTO(out, count = -EFAULT);
 170
 171         if (strchr(valstr, 'u'))
 172                 enabled |= BIT(USRQUOTA);
 173         if (strchr(valstr, 'g'))
 174                 enabled |= BIT(GRPQUOTA);
 175         if (strchr(valstr, 'p'))
 176                 enabled |= BIT(PRJQUOTA);
 177
 178         if (enabled == 0 && strcmp(valstr, "none"))
 179                 GOTO(out, count = -EINVAL);
 180
 181         if (qsd->qsd_is_md)
 182                 pool = LQUOTA_RES_MD;
 183         else
 184                 pool = LQUOTA_RES_DT;
 185
 186         if (server_name2fsname(qsd->qsd_svname, fsname, NULL))
 187                 GOTO(out, count = -EINVAL);
 188
 189         rc = qsd_config(valstr, fsname, pool);
 190         if (rc)
 191                 count = rc;
 192 out:
 193         return count;
 194 }
 195 LPROC_SEQ_FOPS(qsd_enabled);
 196
 197 /* force reintegration procedure to be executed.
 198  * Used for test/debugging purpose */
 199 static ssize_t
 200 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 201                                 size_t count, loff_t *off)
 202 {
 203         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 204         int                  rc = 0, qtype;
 205
 206         LASSERT(qsd != NULL);
 207
 208         write_lock(&qsd->qsd_lock);
 209         if (qsd->qsd_stopping) {
 210                 /* don't mess up with shutdown procedure, it is already
 211                  * complicated enough */
 212                 rc = -ESHUTDOWN;
 213         } else if (!qsd->qsd_prepared) {
 214                 rc = -EAGAIN;
 215         } else {
 216                 /* mark all indexes as stale */
 217                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 218                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 219                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 220                 }
 221         }
 222         write_unlock(&qsd->qsd_lock);
 223
 224         if (rc)
 225                 return rc;
 226
 227         /* kick off reintegration */
 228         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 229                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 230                 if (rc)
 231                         break;
 232         }
 233         return rc == 0 ? count : rc;
 234 }
 235 LPROC_SEQ_FOPS_WR_ONLY(qsd, force_reint);
 236
 237 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 238 {
 239         struct qsd_instance *qsd = m->private;
 240         LASSERT(qsd != NULL);
 241
 242         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 243         return 0;
 244 }
 245
 246 static ssize_t
 247 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 248                         size_t count, loff_t *off)
 249 {
 250         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 251         time64_t timeout;
 252         int rc;
 253
 254         LASSERT(qsd != NULL);
 255         rc = kstrtoll_from_user(buffer, count, 0, &timeout);
 256         if (rc)
 257                 return rc;
 258
 259         if (timeout < 0)
 260                 return -EINVAL;
 261
 262         qsd->qsd_timeout = timeout;
 263         return count;
 264 }
 265 LPROC_SEQ_FOPS(qsd_timeout);
 266
 267 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 268         { .name =       "info",
 269           .fops =       &qsd_state_fops         },
 270         { .name =       "enabled",
 271           .fops =       &qsd_enabled_fops       },
 272         { .name =       "force_reint",
 273           .fops =       &qsd_force_reint_fops   },
 274         { .name =       "timeout",
 275           .fops =       &qsd_timeout_fops       },
 276         { NULL }
 277 };
 278
 279 /*
 280  * Callback function invoked by the OSP layer when the connection to the master
 281  * has been set up.
 282  *
 283  * \param data - is a pointer to the qsd_instance
 284  *
 285  * \retval - 0 on success, appropriate error on failure
 286  */
 287 static int qsd_conn_callback(void *data)
 288 {
 289         struct qsd_instance *qsd = (struct qsd_instance *)data;
 290         int                  type;
 291         ENTRY;
 292
 293         /* qsd_exp should now be valid */
 294         LASSERT(qsd->qsd_exp);
 295
 296         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 297
 298         write_lock(&qsd->qsd_lock);
 299         /* notify that qsd_exp is now valid */
 300         qsd->qsd_exp_valid = true;
 301         write_unlock(&qsd->qsd_lock);
 302
 303         /* Now that the connection to master is setup, we can initiate the
 304          * reintegration procedure for quota types which are enabled.
 305          * It is worth noting that, if the qsd_instance hasn't been started
 306          * already, then we can only complete the first two steps of the
 307          * reintegration procedure (i.e. global lock enqueue and slave
 308          * index transfer) since the space usage reconciliation (i.e.
 309          * step 3) will have to wait for qsd_start() to be called */
 310         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 311                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 312                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 313         }
 314
 315         RETURN(0);
 316 }
 317
 318 /*
 319  * Release qsd_qtype_info structure which contains data associated with a
 320  * given quota type. This releases the accounting objects.
 321  * It's called on OSD cleanup when the qsd instance is released.
 322  *
 323  * \param env - is the environment passed by the caller
 324  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 325  *              to be released
 326  * \param qtype - is the quota type to be shutdown
 327  */
 328 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 329                            int qtype)
 330 {
 331         struct qsd_qtype_info   *qqi;
 332         int repeat = 0;
 333         ENTRY;
 334
 335         if (qsd->qsd_type_array[qtype] == NULL)
 336                 RETURN_EXIT;
 337         qqi = qsd->qsd_type_array[qtype];
 338         qsd->qsd_type_array[qtype] = NULL;
 339
 340         /* all deferred work lists should be empty */
 341         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 342         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 343
 344         /* shutdown lquota site */
 345         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 346                 lquota_site_free(env, qqi->qqi_site);
 347                 qqi->qqi_site = NULL;
 348         }
 349
 350         /* The qqi may still be holding by global locks which are being
 351          * canceled asynchronously (LU-4365), see the following steps:
 352          *
 353          * - On server umount, we try to clear all quota locks first by
 354          *   disconnecting LWP (which will invalidate import and cleanup
 355          *   all locks on it), however, if quota reint process is holding
 356          *   the global lock for reintegration at that time, global lock
 357          *   will fail to be cleared on LWP disconnection.
 358          *
 359          * - Umount process goes on and stops reint process, the global
 360          *   lock will be dropped on reint process exit, however, the lock
 361          *   cancel in done in asynchronous way, so the
 362          *   qsd_glb_blocking_ast() might haven't been called yet when we
 363          *   get here.
 364          */
 365         while (atomic_read(&qqi->qqi_ref) > 1) {
 366                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 367                        atomic_read(&qqi->qqi_ref), repeat);
 368                 repeat++;
 369                 set_current_state(TASK_INTERRUPTIBLE);
 370                 schedule_timeout(cfs_time_seconds(1));
 371         }
 372
 373         /* by now, all qqi users should have gone away */
 374         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 375         lu_ref_fini(&qqi->qqi_reference);
 376
 377         /* release accounting object */
 378         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 379                 dt_object_put(env, qqi->qqi_acct_obj);
 380                 qqi->qqi_acct_obj = NULL;
 381         }
 382
 383         /* release slv index */
 384         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 385                 dt_object_put(env, qqi->qqi_slv_obj);
 386                 qqi->qqi_slv_obj = NULL;
 387                 qqi->qqi_slv_ver = 0;
 388         }
 389
 390         /* release global index */
 391         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 392                 dt_object_put(env, qqi->qqi_glb_obj);
 393                 qqi->qqi_glb_obj = NULL;
 394                 qqi->qqi_glb_ver = 0;
 395         }
 396
 397         OBD_FREE_PTR(qqi);
 398         EXIT;
 399 }
 400
 401 static const char *qtype2acct_name(int qtype)
 402 {
 403         static char unknown[24];
 404
 405         switch (qtype) {
 406         case USRQUOTA:
 407                 return "acct_user";
 408         case GRPQUOTA:
 409                 return "acct_group";
 410         case PRJQUOTA:
 411                 return "acct_project";
 412         }
 413
 414         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 415         return unknown;
 416 }
 417
 418 static const char *qtype2glb_name(int qtype)
 419 {
 420         static char unknown[24];
 421
 422         switch (qtype) {
 423         case USRQUOTA:
 424                 return "limit_user";
 425         case GRPQUOTA:
 426                 return "limit_group";
 427         case PRJQUOTA:
 428                 return "limit_project";
 429         }
 430
 431         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 432         return unknown;
 433 }
 434
 435 /*
 436  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 437  * This opens the accounting object and initializes the proc file.
 438  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 439  * instance.
 440  *
 441  * \param env  - the environment passed by the caller
 442  * \param qsd  - is the qsd instance which will be in charge of the new
 443  *               qsd_qtype_info instance.
 444  * \param qtype - is quota type to set up
 445  *
 446  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 447  *           appropriate error on failure
 448  */
 449 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 450                           int qtype)
 451 {
 452         struct qsd_qtype_info   *qqi;
 453         int                      rc;
 454         struct obd_uuid          uuid;
 455         ENTRY;
 456
 457         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 458
 459         /* allocate structure for this quota type */
 460         OBD_ALLOC_PTR(qqi);
 461         if (qqi == NULL)
 462                 RETURN(-ENOMEM);
 463         qsd->qsd_type_array[qtype] = qqi;
 464         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 465
 466         /* set backpointer and other parameters */
 467         qqi->qqi_qsd   = qsd;
 468         qqi->qqi_qtype = qtype;
 469         lu_ref_init(&qqi->qqi_reference);
 470         qqi->qqi_glb_uptodate = false;
 471         qqi->qqi_slv_uptodate = false;
 472         qqi->qqi_reint        = false;
 473         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 474         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 475         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 476         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 477         lquota_generate_fid(&qqi->qqi_fid, QSD_RES_TYPE(qsd), qtype);
 478
 479         /* open accounting object */
 480         LASSERT(qqi->qqi_acct_obj == NULL);
 481         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 482         if (IS_ERR(qqi->qqi_acct_obj)) {
 483                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 484                        qsd->qsd_svname, qtype_name(qtype),
 485                        PTR_ERR(qqi->qqi_acct_obj));
 486                 qqi->qqi_acct_obj = NULL;
 487                 qqi->qqi_acct_failed = true;
 488         }
 489
 490         /* open global index copy */
 491         LASSERT(qqi->qqi_glb_obj == NULL);
 492         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 493                                                        qsd->qsd_root,
 494                                                        &qqi->qqi_fid, true);
 495         if (IS_ERR(qqi->qqi_glb_obj)) {
 496                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 497                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 498                        PTR_ERR(qqi->qqi_glb_obj));
 499                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 500         }
 501         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 502
 503         /* open slave index copy */
 504         LASSERT(qqi->qqi_slv_obj == NULL);
 505         obd_str2uuid(&uuid, qsd->qsd_svname);
 506         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 507                                                        qsd->qsd_root,
 508                                                        &qqi->qqi_fid, &uuid,
 509                                                        true);
 510         if (IS_ERR(qqi->qqi_slv_obj)) {
 511                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 512                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 513                        PTR_ERR(qqi->qqi_slv_obj));
 514                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 515         }
 516         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 517
 518         /* allocate site */
 519         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 520         if (IS_ERR(qqi->qqi_site)) {
 521                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 522                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 523                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 524         }
 525
 526         /* register proc entry for accounting & global index copy objects */
 527         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 528                                 0444, &lprocfs_quota_seq_fops,
 529                                 qqi->qqi_acct_obj);
 530         if (rc) {
 531                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 532                        qsd->qsd_svname, rc);
 533                 GOTO(out, rc);
 534         }
 535
 536         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 537                                 0444, &lprocfs_quota_seq_fops,
 538                                 qqi->qqi_glb_obj);
 539         if (rc) {
 540                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 541                        qsd->qsd_svname, rc);
 542                 GOTO(out, rc);
 543         }
 544         EXIT;
 545 out:
 546         if (rc)
 547                 qsd_qtype_fini(env, qsd, qtype);
 548         return rc;
 549 }
 550
 551 /*
 552  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 553  * structures associated with the quota slave (on-disk objects, lquota entry
 554  * tables, ...).
 555  * This function should be called when the OSD is shutting down.
 556  *
 557  * \param env - is the environment passed by the caller
 558  * \param qsd - is the qsd instance to shutdown
 559  */
 560 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 561 {
 562         int     qtype;
 563         ENTRY;
 564
 565         if (unlikely(qsd == NULL))
 566                 RETURN_EXIT;
 567
 568         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 569         write_lock(&qsd->qsd_lock);
 570         qsd->qsd_stopping = true;
 571         write_unlock(&qsd->qsd_lock);
 572
 573         /* remove qsd proc entry */
 574         if (qsd->qsd_proc != NULL) {
 575                 lprocfs_remove(&qsd->qsd_proc);
 576                 qsd->qsd_proc = NULL;
 577         }
 578
 579         /* stop the writeback thread */
 580         qsd_stop_upd_thread(qsd);
 581
 582         /* shutdown the reintegration threads */
 583         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 584                 if (qsd->qsd_type_array[qtype] == NULL)
 585                         continue;
 586                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 587         }
 588
 589         if (qsd->qsd_ns != NULL) {
 590                 qsd->qsd_ns = NULL;
 591         }
 592
 593         /* release per-filesystem information */
 594         if (qsd->qsd_fsinfo != NULL) {
 595                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 596                 /* remove from the list of fsinfo */
 597                 list_del_init(&qsd->qsd_link);
 598                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 599                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 600                 qsd->qsd_fsinfo = NULL;
 601         }
 602
 603         /* free per-quota type data */
 604         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 605                 qsd_qtype_fini(env, qsd, qtype);
 606
 607         /* deregister connection to the quota master */
 608         qsd->qsd_exp_valid = false;
 609         lustre_deregister_lwp_item(&qsd->qsd_exp);
 610
 611         /* release quota root directory */
 612         if (qsd->qsd_root != NULL) {
 613                 dt_object_put(env, qsd->qsd_root);
 614                 qsd->qsd_root = NULL;
 615         }
 616
 617         /* release reference on dt_device */
 618         if (qsd->qsd_dev != NULL) {
 619                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 620                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 621                 qsd->qsd_dev = NULL;
 622         }
 623
 624         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 625         OBD_FREE_PTR(qsd);
 626         EXIT;
 627 }
 628 EXPORT_SYMBOL(qsd_fini);
 629
 630 /*
 631  * Create a new qsd_instance to be associated with backend osd device
 632  * identified by \dev.
 633  *
 634  * \param env    - the environment passed by the caller
 635  * \param svname - is the service name of the OSD device creating this instance
 636  * \param dev    - is the dt_device where to store quota index files
 637  * \param osd_proc - is the procfs parent directory where to create procfs file
 638  *                   related to this new qsd instance
 639  *
 640  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 641  *           appropriate error on failure
 642  */
 643 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 644                               struct dt_device *dev,
 645                               struct proc_dir_entry *osd_proc, bool is_md)
 646 {
 647         struct qsd_thread_info  *qti = qsd_info(env);
 648         struct qsd_instance     *qsd;
 649         int                      rc, type, idx;
 650         ENTRY;
 651
 652         /* only configure qsd for MDT & OST */
 653         type = server_name2index(svname, &idx, NULL);
 654         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 655                 RETURN(NULL);
 656
 657         /* allocate qsd instance */
 658         OBD_ALLOC_PTR(qsd);
 659         if (qsd == NULL)
 660                 RETURN(ERR_PTR(-ENOMEM));
 661
 662         /* generic initializations */
 663         rwlock_init(&qsd->qsd_lock);
 664         INIT_LIST_HEAD(&qsd->qsd_link);
 665         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 666         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 667         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 668         spin_lock_init(&qsd->qsd_adjust_lock);
 669         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 670         qsd->qsd_prepared = false;
 671         qsd->qsd_started = false;
 672         qsd->qsd_is_md = is_md;
 673
 674         /* copy service name */
 675         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 676             >= sizeof(qsd->qsd_svname))
 677                 GOTO(out, rc = -E2BIG);
 678
 679         /* grab reference on osd device */
 680         lu_device_get(&dev->dd_lu_dev);
 681         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 682         qsd->qsd_dev = dev;
 683
 684         /* get fsname from svname */
 685         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 686         if (rc) {
 687                 CERROR("%s: fail to extract filesystem name\n", svname);
 688                 GOTO(out, rc);
 689         }
 690
 691         /* look up quota setting for the filesystem the target belongs to */
 692         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 693         if (qsd->qsd_fsinfo == NULL) {
 694                 CERROR("%s: failed to locate filesystem information\n", svname);
 695                 GOTO(out, rc = -EINVAL);
 696         }
 697
 698         /* add in the list of lquota_fsinfo */
 699         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 700         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 701         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 702
 703         /* register procfs directory */
 704         if (qsd->qsd_is_md)
 705                 qsd->qsd_proc = lprocfs_register(QSD_DIR_MD, osd_proc,
 706                                                  lprocfs_quota_qsd_vars, qsd);
 707         else
 708                 qsd->qsd_proc = lprocfs_register(QSD_DIR_DT, osd_proc,
 709                                                  lprocfs_quota_qsd_vars, qsd);
 710
 711         if (type == LDD_F_SV_TYPE_MDT && qsd->qsd_is_md)
 712                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_MD);
 713         else if (type == LDD_F_SV_TYPE_OST && !qsd->qsd_is_md)
 714                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_DT);
 715
 716         if (IS_ERR(qsd->qsd_proc)) {
 717                 rc = PTR_ERR(qsd->qsd_proc);
 718                 qsd->qsd_proc = NULL;
 719                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 720                        svname, rc);
 721                 GOTO(out, rc);
 722         }
 723         EXIT;
 724 out:
 725         if (rc) {
 726                 qsd_fini(env, qsd);
 727                 return ERR_PTR(rc);
 728         }
 729         RETURN(qsd);
 730 }
 731 EXPORT_SYMBOL(qsd_init);
 732
 733 /*
 734  * Initialize on-disk structures in order to manage quota enforcement for
 735  * the target associated with the qsd instance \qsd and starts the reintegration
 736  * procedure for each quota type as soon as possible.
 737  * The last step of the reintegration will be completed once qsd_start() is
 738  * called, at which points the space reconciliation with the master will be
 739  * executed.
 740  * This function must be called when the server stack is fully configured,
 741  * typically when ->ldo_prepare is called across the stack.
 742  *
 743  * \param env - the environment passed by the caller
 744  * \param qsd - is qsd_instance to prepare
 745  *
 746  * \retval - 0 on success, appropriate error on failure
 747  */
 748 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 749 {
 750         struct qsd_thread_info  *qti = qsd_info(env);
 751         int                      qtype, rc = 0;
 752         ENTRY;
 753
 754         if (unlikely(qsd == NULL))
 755                 RETURN(0);
 756
 757         read_lock(&qsd->qsd_lock);
 758         if (qsd->qsd_prepared) {
 759                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 760                 rc = -EALREADY;
 761         }
 762         read_unlock(&qsd->qsd_lock);
 763         if (rc)
 764                 RETURN(rc);
 765
 766         /* Record whether this qsd instance is managing quota enforcement for a
 767          * MDT (i.e. inode quota) or OST (block quota) */
 768         if (qsd->qsd_is_md)
 769                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 770         else
 771                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 772
 773         /* look-up on-disk directory for the quota slave */
 774         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 775                                                     QSD_DIR);
 776         if (IS_ERR(qsd->qsd_root)) {
 777                 rc = PTR_ERR(qsd->qsd_root);
 778                 qsd->qsd_root = NULL;
 779                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 780                        qsd->qsd_svname, rc);
 781                 RETURN(rc);
 782         }
 783
 784         /* initialize per-quota type data */
 785         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 786                 rc = qsd_qtype_init(env, qsd, qtype);
 787                 if (rc)
 788                         RETURN(rc);
 789         }
 790
 791         /* pools successfully setup, mark the qsd as prepared */
 792         write_lock(&qsd->qsd_lock);
 793         qsd->qsd_prepared = true;
 794         write_unlock(&qsd->qsd_lock);
 795
 796         if (qsd->qsd_dev->dd_rdonly)
 797                 RETURN(0);
 798
 799         /* start reintegration thread for each type, if required */
 800         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 801                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 802
 803                 if (qsd_type_enabled(qsd, qtype) &&
 804                     qqi->qqi_acct_failed) {
 805                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 806                                        "since space accounting isn't functional"
 807                                        ". Please run tunefs.lustre --quota on "
 808                                        "an unmounted filesystem if not done "
 809                                        "already\n", qsd->qsd_svname);
 810                         continue;
 811                 }
 812
 813                 rc = qsd_start_reint_thread(qqi);
 814                 if (rc) {
 815                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 816                                 qsd->qsd_svname, qtype_name(qtype), rc);
 817                         RETURN(rc);
 818                 }
 819         }
 820
 821         /* start writeback thread */
 822         rc = qsd_start_upd_thread(qsd);
 823         if (rc) {
 824                 CERROR("%s: failed to start writeback thread (%d)\n",
 825                        qsd->qsd_svname, rc);
 826                 RETURN(rc);
 827         }
 828
 829         /* generate osp name */
 830         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 831                                MTI_NAME_MAXLEN, 0);
 832         if (rc) {
 833                 CERROR("%s: failed to generate ospname (%d)\n",
 834                        qsd->qsd_svname, rc);
 835                 RETURN(rc);
 836         }
 837
 838         /* the connection callback will start the reintegration
 839          * procedure if quota is enabled */
 840         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 841                                       qsd_conn_callback, (void *)qsd);
 842         if (rc) {
 843                 CERROR("%s: fail to get connection to master (%d)\n",
 844                        qsd->qsd_svname, rc);
 845                 RETURN(rc);
 846         }
 847
 848         RETURN(0);
 849 }
 850 EXPORT_SYMBOL(qsd_prepare);
 851
 852 /*
 853  * Start a qsd instance. This will complete the last step of the reintegration
 854  * procedure as soon as possible (provided that the master is reachable).
 855  * This should be called when recovery has been completed and quota should now
 856  * be enforced on every operations.
 857  *
 858  * \param env - the environment passed by the caller
 859  * \param qsd - is the qsd instance associated with the osd device to start
 860  */
 861 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 862 {
 863         int     type, rc = 0;
 864         ENTRY;
 865
 866         if (unlikely(qsd == NULL))
 867                 RETURN(0);
 868
 869         write_lock(&qsd->qsd_lock);
 870         if (!qsd->qsd_prepared) {
 871                 CERROR("%s: can't start qsd instance since it wasn't properly "
 872                        "initialized\n", qsd->qsd_svname);
 873                 rc = -EFAULT;
 874         } else if (qsd->qsd_started) {
 875                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 876                 rc = -EALREADY;
 877         } else {
 878                 /* notify that the qsd_instance is now started */
 879                 qsd->qsd_started = true;
 880         }
 881         write_unlock(&qsd->qsd_lock);
 882
 883         if (rc)
 884                 RETURN(rc);
 885
 886         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 887          * up to usage; If usage < granted, release down to usage.  */
 888         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 889                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 890                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 891         }
 892
 893         RETURN(rc);
 894 }
 895 EXPORT_SYMBOL(qsd_start);
 896
 897 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 898
 899 /*
 900  * Global initialization performed at module load time
 901  */
 902 int qsd_glb_init(void)
 903 {
 904         int     rc;
 905
 906         rc = lu_kmem_init(qsd_caches);
 907         if (rc)
 908                 return rc;
 909
 910         qsd_key_init_generic(&qsd_thread_key, NULL);
 911         lu_context_key_register(&qsd_thread_key);
 912         lustre_register_quota_process_config(qsd_process_config);
 913
 914         return 0;
 915 }
 916
 917 /*
 918  * Companion of qsd_glb_init() called at module unload time
 919  */
 920 void qsd_glb_fini(void)
 921 {
 922         lustre_register_quota_process_config(NULL);
 923         lu_kmem_fini(qsd_caches);
 924         lu_context_key_degister(&qsd_thread_key);
 925 }