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         int rc;
 250         __s64 timeout;
 251         LASSERT(qsd != NULL);
 252
 253         rc = lprocfs_str_to_s64(buffer, count, &timeout);
 254         if (rc)
 255                 return rc;
 256         if (timeout < 0 || timeout > INT_MAX)
 257                 return -EINVAL;
 258
 259         qsd->qsd_timeout = timeout;
 260         return count;
 261 }
 262 LPROC_SEQ_FOPS(qsd_timeout);
 263
 264 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 265         { .name =       "info",
 266           .fops =       &qsd_state_fops         },
 267         { .name =       "enabled",
 268           .fops =       &qsd_enabled_fops       },
 269         { .name =       "force_reint",
 270           .fops =       &qsd_force_reint_fops   },
 271         { .name =       "timeout",
 272           .fops =       &qsd_timeout_fops       },
 273         { NULL }
 274 };
 275
 276 /*
 277  * Callback function invoked by the OSP layer when the connection to the master
 278  * has been set up.
 279  *
 280  * \param data - is a pointer to the qsd_instance
 281  *
 282  * \retval - 0 on success, appropriate error on failure
 283  */
 284 static int qsd_conn_callback(void *data)
 285 {
 286         struct qsd_instance *qsd = (struct qsd_instance *)data;
 287         int                  type;
 288         ENTRY;
 289
 290         /* qsd_exp should now be valid */
 291         LASSERT(qsd->qsd_exp);
 292
 293         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 294
 295         write_lock(&qsd->qsd_lock);
 296         /* notify that qsd_exp is now valid */
 297         qsd->qsd_exp_valid = true;
 298         write_unlock(&qsd->qsd_lock);
 299
 300         /* Now that the connection to master is setup, we can initiate the
 301          * reintegration procedure for quota types which are enabled.
 302          * It is worth noting that, if the qsd_instance hasn't been started
 303          * already, then we can only complete the first two steps of the
 304          * reintegration procedure (i.e. global lock enqueue and slave
 305          * index transfer) since the space usage reconciliation (i.e.
 306          * step 3) will have to wait for qsd_start() to be called */
 307         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 308                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 309                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 310         }
 311
 312         RETURN(0);
 313 }
 314
 315 /*
 316  * Release qsd_qtype_info structure which contains data associated with a
 317  * given quota type. This releases the accounting objects.
 318  * It's called on OSD cleanup when the qsd instance is released.
 319  *
 320  * \param env - is the environment passed by the caller
 321  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 322  *              to be released
 323  * \param qtype - is the quota type to be shutdown
 324  */
 325 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 326                            int qtype)
 327 {
 328         struct qsd_qtype_info   *qqi;
 329         int repeat = 0;
 330         ENTRY;
 331
 332         if (qsd->qsd_type_array[qtype] == NULL)
 333                 RETURN_EXIT;
 334         qqi = qsd->qsd_type_array[qtype];
 335         qsd->qsd_type_array[qtype] = NULL;
 336
 337         /* all deferred work lists should be empty */
 338         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 339         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 340
 341         /* shutdown lquota site */
 342         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 343                 lquota_site_free(env, qqi->qqi_site);
 344                 qqi->qqi_site = NULL;
 345         }
 346
 347         /* The qqi may still be holding by global locks which are being
 348          * canceled asynchronously (LU-4365), see the following steps:
 349          *
 350          * - On server umount, we try to clear all quota locks first by
 351          *   disconnecting LWP (which will invalidate import and cleanup
 352          *   all locks on it), however, if quota reint process is holding
 353          *   the global lock for reintegration at that time, global lock
 354          *   will fail to be cleared on LWP disconnection.
 355          *
 356          * - Umount process goes on and stops reint process, the global
 357          *   lock will be dropped on reint process exit, however, the lock
 358          *   cancel in done in asynchronous way, so the
 359          *   qsd_glb_blocking_ast() might haven't been called yet when we
 360          *   get here.
 361          */
 362         while (atomic_read(&qqi->qqi_ref) > 1) {
 363                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 364                        atomic_read(&qqi->qqi_ref), repeat);
 365                 repeat++;
 366                 set_current_state(TASK_INTERRUPTIBLE);
 367                 schedule_timeout(cfs_time_seconds(1));
 368         }
 369
 370         /* by now, all qqi users should have gone away */
 371         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 372         lu_ref_fini(&qqi->qqi_reference);
 373
 374         /* release accounting object */
 375         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 376                 dt_object_put(env, qqi->qqi_acct_obj);
 377                 qqi->qqi_acct_obj = NULL;
 378         }
 379
 380         /* release slv index */
 381         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 382                 dt_object_put(env, qqi->qqi_slv_obj);
 383                 qqi->qqi_slv_obj = NULL;
 384                 qqi->qqi_slv_ver = 0;
 385         }
 386
 387         /* release global index */
 388         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 389                 dt_object_put(env, qqi->qqi_glb_obj);
 390                 qqi->qqi_glb_obj = NULL;
 391                 qqi->qqi_glb_ver = 0;
 392         }
 393
 394         OBD_FREE_PTR(qqi);
 395         EXIT;
 396 }
 397
 398 static const char *qtype2acct_name(int qtype)
 399 {
 400         static char unknown[24];
 401
 402         switch (qtype) {
 403         case USRQUOTA:
 404                 return "acct_user";
 405         case GRPQUOTA:
 406                 return "acct_group";
 407         case PRJQUOTA:
 408                 return "acct_project";
 409         }
 410
 411         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 412         return unknown;
 413 }
 414
 415 static const char *qtype2glb_name(int qtype)
 416 {
 417         static char unknown[24];
 418
 419         switch (qtype) {
 420         case USRQUOTA:
 421                 return "limit_user";
 422         case GRPQUOTA:
 423                 return "limit_group";
 424         case PRJQUOTA:
 425                 return "limit_project";
 426         }
 427
 428         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 429         return unknown;
 430 }
 431
 432 /*
 433  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 434  * This opens the accounting object and initializes the proc file.
 435  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 436  * instance.
 437  *
 438  * \param env  - the environment passed by the caller
 439  * \param qsd  - is the qsd instance which will be in charge of the new
 440  *               qsd_qtype_info instance.
 441  * \param qtype - is quota type to set up
 442  *
 443  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 444  *           appropriate error on failure
 445  */
 446 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 447                           int qtype)
 448 {
 449         struct qsd_qtype_info   *qqi;
 450         int                      rc;
 451         struct obd_uuid          uuid;
 452         ENTRY;
 453
 454         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 455
 456         /* allocate structure for this quota type */
 457         OBD_ALLOC_PTR(qqi);
 458         if (qqi == NULL)
 459                 RETURN(-ENOMEM);
 460         qsd->qsd_type_array[qtype] = qqi;
 461         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 462
 463         /* set backpointer and other parameters */
 464         qqi->qqi_qsd   = qsd;
 465         qqi->qqi_qtype = qtype;
 466         lu_ref_init(&qqi->qqi_reference);
 467         qqi->qqi_glb_uptodate = false;
 468         qqi->qqi_slv_uptodate = false;
 469         qqi->qqi_reint        = false;
 470         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 471         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 472         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 473         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 474         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id,
 475                             QSD_RES_TYPE(qsd), qtype);
 476
 477         /* open accounting object */
 478         LASSERT(qqi->qqi_acct_obj == NULL);
 479         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 480         if (IS_ERR(qqi->qqi_acct_obj)) {
 481                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 482                        qsd->qsd_svname, qtype_name(qtype),
 483                        PTR_ERR(qqi->qqi_acct_obj));
 484                 qqi->qqi_acct_obj = NULL;
 485                 qqi->qqi_acct_failed = true;
 486         }
 487
 488         /* open global index copy */
 489         LASSERT(qqi->qqi_glb_obj == NULL);
 490         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 491                                                        qsd->qsd_root,
 492                                                        &qqi->qqi_fid, true);
 493         if (IS_ERR(qqi->qqi_glb_obj)) {
 494                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 495                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 496                        PTR_ERR(qqi->qqi_glb_obj));
 497                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 498         }
 499         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 500
 501         /* open slave index copy */
 502         LASSERT(qqi->qqi_slv_obj == NULL);
 503         obd_str2uuid(&uuid, qsd->qsd_svname);
 504         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 505                                                        qsd->qsd_root,
 506                                                        &qqi->qqi_fid, &uuid,
 507                                                        true);
 508         if (IS_ERR(qqi->qqi_slv_obj)) {
 509                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 510                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 511                        PTR_ERR(qqi->qqi_slv_obj));
 512                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 513         }
 514         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 515
 516         /* allocate site */
 517         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 518         if (IS_ERR(qqi->qqi_site)) {
 519                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 520                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 521                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 522         }
 523
 524         /* register proc entry for accounting & global index copy objects */
 525         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 526                                 0444, &lprocfs_quota_seq_fops,
 527                                 qqi->qqi_acct_obj);
 528         if (rc) {
 529                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 530                        qsd->qsd_svname, rc);
 531                 GOTO(out, rc);
 532         }
 533
 534         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 535                                 0444, &lprocfs_quota_seq_fops,
 536                                 qqi->qqi_glb_obj);
 537         if (rc) {
 538                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 539                        qsd->qsd_svname, rc);
 540                 GOTO(out, rc);
 541         }
 542         EXIT;
 543 out:
 544         if (rc)
 545                 qsd_qtype_fini(env, qsd, qtype);
 546         return rc;
 547 }
 548
 549 /*
 550  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 551  * structures associated with the quota slave (on-disk objects, lquota entry
 552  * tables, ...).
 553  * This function should be called when the OSD is shutting down.
 554  *
 555  * \param env - is the environment passed by the caller
 556  * \param qsd - is the qsd instance to shutdown
 557  */
 558 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 559 {
 560         int     qtype;
 561         ENTRY;
 562
 563         if (unlikely(qsd == NULL))
 564                 RETURN_EXIT;
 565
 566         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 567         write_lock(&qsd->qsd_lock);
 568         qsd->qsd_stopping = true;
 569         write_unlock(&qsd->qsd_lock);
 570
 571         /* remove qsd proc entry */
 572         if (qsd->qsd_proc != NULL) {
 573                 lprocfs_remove(&qsd->qsd_proc);
 574                 qsd->qsd_proc = NULL;
 575         }
 576
 577         /* stop the writeback thread */
 578         qsd_stop_upd_thread(qsd);
 579
 580         /* shutdown the reintegration threads */
 581         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 582                 if (qsd->qsd_type_array[qtype] == NULL)
 583                         continue;
 584                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 585         }
 586
 587         if (qsd->qsd_ns != NULL) {
 588                 qsd->qsd_ns = NULL;
 589         }
 590
 591         /* free per-quota type data */
 592         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 593                 qsd_qtype_fini(env, qsd, qtype);
 594
 595         /* deregister connection to the quota master */
 596         qsd->qsd_exp_valid = false;
 597         lustre_deregister_lwp_item(&qsd->qsd_exp);
 598
 599         /* release per-filesystem information */
 600         if (qsd->qsd_fsinfo != NULL) {
 601                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 602                 /* remove from the list of fsinfo */
 603                 list_del_init(&qsd->qsd_link);
 604                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 605                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 606                 qsd->qsd_fsinfo = NULL;
 607         }
 608
 609         /* release quota root directory */
 610         if (qsd->qsd_root != NULL) {
 611                 dt_object_put(env, qsd->qsd_root);
 612                 qsd->qsd_root = NULL;
 613         }
 614
 615         /* release reference on dt_device */
 616         if (qsd->qsd_dev != NULL) {
 617                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 618                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 619                 qsd->qsd_dev = NULL;
 620         }
 621
 622         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 623         OBD_FREE_PTR(qsd);
 624         EXIT;
 625 }
 626 EXPORT_SYMBOL(qsd_fini);
 627
 628 /*
 629  * Create a new qsd_instance to be associated with backend osd device
 630  * identified by \dev.
 631  *
 632  * \param env    - the environment passed by the caller
 633  * \param svname - is the service name of the OSD device creating this instance
 634  * \param dev    - is the dt_device where to store quota index files
 635  * \param osd_proc - is the procfs parent directory where to create procfs file
 636  *                   related to this new qsd instance
 637  *
 638  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 639  *           appropriate error on failure
 640  */
 641 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 642                               struct dt_device *dev,
 643                               struct proc_dir_entry *osd_proc)
 644 {
 645         struct qsd_thread_info  *qti = qsd_info(env);
 646         struct qsd_instance     *qsd;
 647         int                      rc, type, idx;
 648         ENTRY;
 649
 650         /* only configure qsd for MDT & OST */
 651         type = server_name2index(svname, &idx, NULL);
 652         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 653                 RETURN(NULL);
 654
 655         /* allocate qsd instance */
 656         OBD_ALLOC_PTR(qsd);
 657         if (qsd == NULL)
 658                 RETURN(ERR_PTR(-ENOMEM));
 659
 660         /* generic initializations */
 661         rwlock_init(&qsd->qsd_lock);
 662         INIT_LIST_HEAD(&qsd->qsd_link);
 663         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 664         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 665         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 666         spin_lock_init(&qsd->qsd_adjust_lock);
 667         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 668         qsd->qsd_prepared = false;
 669         qsd->qsd_started = false;
 670
 671         /* copy service name */
 672         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 673             >= sizeof(qsd->qsd_svname))
 674                 GOTO(out, rc = -E2BIG);
 675
 676         /* grab reference on osd device */
 677         lu_device_get(&dev->dd_lu_dev);
 678         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 679         qsd->qsd_dev = dev;
 680
 681         /* we only support pool ID 0 (default data or metadata pool) for the
 682          * time being. A different pool ID could be assigned to this target via
 683          * the configuration log in the future */
 684         qsd->qsd_pool_id  = 0;
 685
 686         /* get fsname from svname */
 687         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 688         if (rc) {
 689                 CERROR("%s: fail to extract filesystem name\n", svname);
 690                 GOTO(out, rc);
 691         }
 692
 693         /* look up quota setting for the filesystem the target belongs to */
 694         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 695         if (qsd->qsd_fsinfo == NULL) {
 696                 CERROR("%s: failed to locate filesystem information\n", svname);
 697                 GOTO(out, rc = -EINVAL);
 698         }
 699
 700         /* add in the list of lquota_fsinfo */
 701         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 702         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 703         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 704
 705         /* register procfs directory */
 706         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 707                                          lprocfs_quota_qsd_vars, qsd);
 708         if (IS_ERR(qsd->qsd_proc)) {
 709                 rc = PTR_ERR(qsd->qsd_proc);
 710                 qsd->qsd_proc = NULL;
 711                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 712                        svname, rc);
 713                 GOTO(out, rc);
 714         }
 715         EXIT;
 716 out:
 717         if (rc) {
 718                 qsd_fini(env, qsd);
 719                 return ERR_PTR(rc);
 720         }
 721         RETURN(qsd);
 722 }
 723 EXPORT_SYMBOL(qsd_init);
 724
 725 /*
 726  * Initialize on-disk structures in order to manage quota enforcement for
 727  * the target associated with the qsd instance \qsd and starts the reintegration
 728  * procedure for each quota type as soon as possible.
 729  * The last step of the reintegration will be completed once qsd_start() is
 730  * called, at which points the space reconciliation with the master will be
 731  * executed.
 732  * This function must be called when the server stack is fully configured,
 733  * typically when ->ldo_prepare is called across the stack.
 734  *
 735  * \param env - the environment passed by the caller
 736  * \param qsd - is qsd_instance to prepare
 737  *
 738  * \retval - 0 on success, appropriate error on failure
 739  */
 740 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 741 {
 742         struct qsd_thread_info  *qti = qsd_info(env);
 743         int                      qtype, rc = 0;
 744         ENTRY;
 745
 746         if (unlikely(qsd == NULL))
 747                 RETURN(0);
 748
 749         read_lock(&qsd->qsd_lock);
 750         if (qsd->qsd_prepared) {
 751                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 752                 rc = -EALREADY;
 753         }
 754         read_unlock(&qsd->qsd_lock);
 755         if (rc)
 756                 RETURN(rc);
 757
 758         /* Record whether this qsd instance is managing quota enforcement for a
 759          * MDT (i.e. inode quota) or OST (block quota) */
 760         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 761                 qsd->qsd_is_md = true;
 762                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 763         } else {
 764                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 765         }
 766
 767         /* look-up on-disk directory for the quota slave */
 768         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 769                                                     QSD_DIR);
 770         if (IS_ERR(qsd->qsd_root)) {
 771                 rc = PTR_ERR(qsd->qsd_root);
 772                 qsd->qsd_root = NULL;
 773                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 774                        qsd->qsd_svname, rc);
 775                 RETURN(rc);
 776         }
 777
 778         /* initialize per-quota type data */
 779         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 780                 rc = qsd_qtype_init(env, qsd, qtype);
 781                 if (rc)
 782                         RETURN(rc);
 783         }
 784
 785         /* pools successfully setup, mark the qsd as prepared */
 786         write_lock(&qsd->qsd_lock);
 787         qsd->qsd_prepared = true;
 788         write_unlock(&qsd->qsd_lock);
 789
 790         if (qsd->qsd_dev->dd_rdonly)
 791                 RETURN(0);
 792
 793         /* start reintegration thread for each type, if required */
 794         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 795                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 796
 797                 if (qsd_type_enabled(qsd, qtype) &&
 798                     qqi->qqi_acct_failed) {
 799                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 800                                        "since space accounting isn't functional"
 801                                        ". Please run tunefs.lustre --quota on "
 802                                        "an unmounted filesystem if not done "
 803                                        "already\n", qsd->qsd_svname);
 804                         break;
 805                 }
 806
 807                 rc = qsd_start_reint_thread(qqi);
 808                 if (rc) {
 809                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 810                                 qsd->qsd_svname, qtype_name(qtype), rc);
 811                         RETURN(rc);
 812                 }
 813         }
 814
 815         /* start writeback thread */
 816         rc = qsd_start_upd_thread(qsd);
 817         if (rc) {
 818                 CERROR("%s: failed to start writeback thread (%d)\n",
 819                        qsd->qsd_svname, rc);
 820                 RETURN(rc);
 821         }
 822
 823         /* generate osp name */
 824         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 825                                MTI_NAME_MAXLEN, 0);
 826         if (rc) {
 827                 CERROR("%s: failed to generate ospname (%d)\n",
 828                        qsd->qsd_svname, rc);
 829                 RETURN(rc);
 830         }
 831
 832         /* the connection callback will start the reintegration
 833          * procedure if quota is enabled */
 834         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 835                                       qsd_conn_callback, (void *)qsd);
 836         if (rc) {
 837                 CERROR("%s: fail to get connection to master (%d)\n",
 838                        qsd->qsd_svname, rc);
 839                 RETURN(rc);
 840         }
 841
 842         RETURN(0);
 843 }
 844 EXPORT_SYMBOL(qsd_prepare);
 845
 846 /*
 847  * Start a qsd instance. This will complete the last step of the reintegration
 848  * procedure as soon as possible (provided that the master is reachable).
 849  * This should be called when recovery has been completed and quota should now
 850  * be enforced on every operations.
 851  *
 852  * \param env - the environment passed by the caller
 853  * \param qsd - is the qsd instance associated with the osd device to start
 854  */
 855 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 856 {
 857         int     type, rc = 0;
 858         ENTRY;
 859
 860         if (unlikely(qsd == NULL))
 861                 RETURN(0);
 862
 863         write_lock(&qsd->qsd_lock);
 864         if (!qsd->qsd_prepared) {
 865                 CERROR("%s: can't start qsd instance since it wasn't properly "
 866                        "initialized\n", qsd->qsd_svname);
 867                 rc = -EFAULT;
 868         } else if (qsd->qsd_started) {
 869                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 870                 rc = -EALREADY;
 871         } else {
 872                 /* notify that the qsd_instance is now started */
 873                 qsd->qsd_started = true;
 874         }
 875         write_unlock(&qsd->qsd_lock);
 876
 877         if (rc)
 878                 RETURN(rc);
 879
 880         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 881          * up to usage; If usage < granted, release down to usage.  */
 882         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 883                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 884                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 885         }
 886
 887         RETURN(rc);
 888 }
 889 EXPORT_SYMBOL(qsd_start);
 890
 891 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 892
 893 /*
 894  * Global initialization performed at module load time
 895  */
 896 int qsd_glb_init(void)
 897 {
 898         int     rc;
 899
 900         rc = lu_kmem_init(qsd_caches);
 901         if (rc)
 902                 return rc;
 903
 904         qsd_key_init_generic(&qsd_thread_key, NULL);
 905         lu_context_key_register(&qsd_thread_key);
 906         lustre_register_quota_process_config(qsd_process_config);
 907
 908         return 0;
 909 }
 910
 911 /*
 912  * Companion of qsd_glb_init() called at module unload time
 913  */
 914 void qsd_glb_fini(void)
 915 {
 916         lustre_register_quota_process_config(NULL);
 917         lu_kmem_fini(qsd_caches);
 918         lu_context_key_degister(&qsd_thread_key);
 919 }