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, 2016, 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 (strlen(enabled) == 0)
  89                 strcat(enabled, "none");
  90
  91         seq_printf(m, "target name:    %s\n"
  92                    "pool ID:        %d\n"
  93                    "type:           %s\n"
  94                    "quota enabled:  %s\n"
  95                    "conn to master: %s\n",
  96                    qsd->qsd_svname, qsd->qsd_pool_id,
  97                    qsd->qsd_is_md ? "md" : "dt", enabled,
  98                    qsd->qsd_exp_valid ? "setup" : "not setup yet");
  99
 100         if (qsd->qsd_prepared) {
 101                 memset(enabled, 0, sizeof(enabled));
 102                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 103                         strcat(enabled, "u");
 104                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 105                         strcat(enabled, "g");
 106                 if (strlen(enabled) == 0)
 107                         strcat(enabled, "none");
 108                 seq_printf(m, "space acct:     %s\n"
 109                            "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 110                            "group uptodate: glb[%d],slv[%d],reint[%d]\n",
 111                            enabled,
 112                            qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 113                            qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 114                            qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 115                            qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 116                            qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 117                            qsd->qsd_type_array[GRPQUOTA]->qqi_reint);
 118         }
 119         return 0;
 120 }
 121 LPROC_SEQ_FOPS_RO(qsd_state);
 122
 123 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 124 {
 125         struct qsd_instance     *qsd = m->private;
 126         char                     enabled[5];
 127
 128         LASSERT(qsd != NULL);
 129
 130         memset(enabled, 0, sizeof(enabled));
 131         if (qsd_type_enabled(qsd, USRQUOTA))
 132                 strcat(enabled, "u");
 133         if (qsd_type_enabled(qsd, GRPQUOTA))
 134                 strcat(enabled, "g");
 135         if (strlen(enabled) == 0)
 136                 strcat(enabled, "none");
 137
 138         seq_printf(m, "%s\n", enabled);
 139         return 0;
 140 }
 141 LPROC_SEQ_FOPS_RO(qsd_enabled);
 142
 143 /* force reintegration procedure to be executed.
 144  * Used for test/debugging purpose */
 145 static ssize_t
 146 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 147                                 size_t count, loff_t *off)
 148 {
 149         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 150         int                  rc = 0, qtype;
 151
 152         LASSERT(qsd != NULL);
 153
 154         write_lock(&qsd->qsd_lock);
 155         if (qsd->qsd_stopping) {
 156                 /* don't mess up with shutdown procedure, it is already
 157                  * complicated enough */
 158                 rc = -ESHUTDOWN;
 159         } else if (!qsd->qsd_prepared) {
 160                 rc = -EAGAIN;
 161         } else {
 162                 /* mark all indexes as stale */
 163                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 164                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 165                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 166                 }
 167         }
 168         write_unlock(&qsd->qsd_lock);
 169
 170         if (rc)
 171                 return rc;
 172
 173         /* kick off reintegration */
 174         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 175                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 176                 if (rc)
 177                         break;
 178         }
 179         return rc == 0 ? count : rc;
 180 }
 181 LPROC_SEQ_FOPS_WO_TYPE(qsd, force_reint);
 182
 183 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 184 {
 185         struct qsd_instance *qsd = m->private;
 186         LASSERT(qsd != NULL);
 187
 188         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 189         return 0;
 190 }
 191
 192 static ssize_t
 193 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 194                         size_t count, loff_t *off)
 195 {
 196         struct qsd_instance *qsd = ((struct seq_file *)file->private_data)->private;
 197         int rc;
 198         __s64 timeout;
 199         LASSERT(qsd != NULL);
 200
 201         rc = lprocfs_str_to_s64(buffer, count, &timeout);
 202         if (rc)
 203                 return rc;
 204         if (timeout < 0 || timeout > INT_MAX)
 205                 return -EINVAL;
 206
 207         qsd->qsd_timeout = timeout;
 208         return count;
 209 }
 210 LPROC_SEQ_FOPS(qsd_timeout);
 211
 212 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 213         { .name =       "info",
 214           .fops =       &qsd_state_fops         },
 215         { .name =       "enabled",
 216           .fops =       &qsd_enabled_fops       },
 217         { .name =       "force_reint",
 218           .fops =       &qsd_force_reint_fops   },
 219         { .name =       "timeout",
 220           .fops =       &qsd_timeout_fops       },
 221         { NULL }
 222 };
 223
 224 /*
 225  * Callback function invoked by the OSP layer when the connection to the master
 226  * has been set up.
 227  *
 228  * \param data - is a pointer to the qsd_instance
 229  *
 230  * \retval - 0 on success, appropriate error on failure
 231  */
 232 static int qsd_conn_callback(void *data)
 233 {
 234         struct qsd_instance *qsd = (struct qsd_instance *)data;
 235         int                  type;
 236         ENTRY;
 237
 238         /* qsd_exp should now be valid */
 239         LASSERT(qsd->qsd_exp);
 240
 241         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 242
 243         write_lock(&qsd->qsd_lock);
 244         /* notify that qsd_exp is now valid */
 245         qsd->qsd_exp_valid = true;
 246         write_unlock(&qsd->qsd_lock);
 247
 248         /* Now that the connection to master is setup, we can initiate the
 249          * reintegration procedure for quota types which are enabled.
 250          * It is worth noting that, if the qsd_instance hasn't been started
 251          * already, then we can only complete the first two steps of the
 252          * reintegration procedure (i.e. global lock enqueue and slave
 253          * index transfer) since the space usage reconciliation (i.e.
 254          * step 3) will have to wait for qsd_start() to be called */
 255         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 256                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 257                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 258         }
 259
 260         RETURN(0);
 261 }
 262
 263 /*
 264  * Release qsd_qtype_info structure which contains data associated with a
 265  * given quota type. This releases the accounting objects.
 266  * It's called on OSD cleanup when the qsd instance is released.
 267  *
 268  * \param env - is the environment passed by the caller
 269  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 270  *              to be released
 271  * \param qtype - is the quota type to be shutdown
 272  */
 273 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 274                            int qtype)
 275 {
 276         struct qsd_qtype_info   *qqi;
 277         int repeat = 0;
 278         ENTRY;
 279
 280         if (qsd->qsd_type_array[qtype] == NULL)
 281                 RETURN_EXIT;
 282         qqi = qsd->qsd_type_array[qtype];
 283         qsd->qsd_type_array[qtype] = NULL;
 284
 285         /* all deferred work lists should be empty */
 286         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 287         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 288
 289         /* shutdown lquota site */
 290         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 291                 lquota_site_free(env, qqi->qqi_site);
 292                 qqi->qqi_site = NULL;
 293         }
 294
 295         /* The qqi may still be holding by global locks which are being
 296          * canceled asynchronously (LU-4365), see the following steps:
 297          *
 298          * - On server umount, we try to clear all quota locks first by
 299          *   disconnecting LWP (which will invalidate import and cleanup
 300          *   all locks on it), however, if quota reint process is holding
 301          *   the global lock for reintegration at that time, global lock
 302          *   will fail to be cleared on LWP disconnection.
 303          *
 304          * - Umount process goes on and stops reint process, the global
 305          *   lock will be dropped on reint process exit, however, the lock
 306          *   cancel in done in asynchronous way, so the
 307          *   qsd_glb_blocking_ast() might haven't been called yet when we
 308          *   get here.
 309          */
 310         while (atomic_read(&qqi->qqi_ref) > 1) {
 311                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 312                        atomic_read(&qqi->qqi_ref), repeat);
 313                 repeat++;
 314                 set_current_state(TASK_INTERRUPTIBLE);
 315                 schedule_timeout(cfs_time_seconds(1));
 316         }
 317
 318         /* by now, all qqi users should have gone away */
 319         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 320         lu_ref_fini(&qqi->qqi_reference);
 321
 322         /* release accounting object */
 323         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 324                 dt_object_put(env, qqi->qqi_acct_obj);
 325                 qqi->qqi_acct_obj = NULL;
 326         }
 327
 328         /* release slv index */
 329         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 330                 dt_object_put(env, qqi->qqi_slv_obj);
 331                 qqi->qqi_slv_obj = NULL;
 332                 qqi->qqi_slv_ver = 0;
 333         }
 334
 335         /* release global index */
 336         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 337                 dt_object_put(env, qqi->qqi_glb_obj);
 338                 qqi->qqi_glb_obj = NULL;
 339                 qqi->qqi_glb_ver = 0;
 340         }
 341
 342         OBD_FREE_PTR(qqi);
 343         EXIT;
 344 }
 345
 346 static const char *qtype2acct_name(int qtype)
 347 {
 348         switch (qtype) {
 349         case USRQUOTA:
 350                 return "acct_user";
 351         case GRPQUOTA:
 352                 return "acct_group";
 353         }
 354
 355         LASSERTF(0, "invalid quota type: %d", qtype);
 356         return NULL;
 357 }
 358
 359 static const char *qtype2glb_name(int qtype)
 360 {
 361         switch (qtype) {
 362         case USRQUOTA:
 363                 return "limit_user";
 364         case GRPQUOTA:
 365                 return "limit_group";
 366         }
 367
 368         LASSERTF(0, "invalid quota type: %d", qtype);
 369         return NULL;
 370 }
 371
 372 /*
 373  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 374  * This opens the accounting object and initializes the proc file.
 375  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 376  * instance.
 377  *
 378  * \param env  - the environment passed by the caller
 379  * \param qsd  - is the qsd instance which will be in charge of the new
 380  *               qsd_qtype_info instance.
 381  * \param qtype - is quota type to set up
 382  *
 383  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 384  *           appropriate error on failure
 385  */
 386 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 387                           int qtype)
 388 {
 389         struct qsd_qtype_info   *qqi;
 390         int                      rc;
 391         struct obd_uuid          uuid;
 392         ENTRY;
 393
 394         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 395
 396         /* allocate structure for this quota type */
 397         OBD_ALLOC_PTR(qqi);
 398         if (qqi == NULL)
 399                 RETURN(-ENOMEM);
 400         qsd->qsd_type_array[qtype] = qqi;
 401         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 402
 403         /* set backpointer and other parameters */
 404         qqi->qqi_qsd   = qsd;
 405         qqi->qqi_qtype = qtype;
 406         lu_ref_init(&qqi->qqi_reference);
 407         qqi->qqi_glb_uptodate = false;
 408         qqi->qqi_slv_uptodate = false;
 409         qqi->qqi_reint        = false;
 410         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 411         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 412         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 413         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 414         lquota_generate_fid(&qqi->qqi_fid, qsd->qsd_pool_id,
 415                             QSD_RES_TYPE(qsd), qtype);
 416
 417         /* open accounting object */
 418         LASSERT(qqi->qqi_acct_obj == NULL);
 419         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 420         if (IS_ERR(qqi->qqi_acct_obj)) {
 421                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 422                        qsd->qsd_svname, qtype_name(qtype),
 423                        PTR_ERR(qqi->qqi_acct_obj));
 424                 qqi->qqi_acct_obj = NULL;
 425                 qsd->qsd_acct_failed = true;
 426         }
 427
 428         /* open global index copy */
 429         LASSERT(qqi->qqi_glb_obj == NULL);
 430         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 431                                                        qsd->qsd_root,
 432                                                        &qqi->qqi_fid, true);
 433         if (IS_ERR(qqi->qqi_glb_obj)) {
 434                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 435                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 436                        PTR_ERR(qqi->qqi_glb_obj));
 437                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 438         }
 439         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 440
 441         /* open slave index copy */
 442         LASSERT(qqi->qqi_slv_obj == NULL);
 443         obd_str2uuid(&uuid, qsd->qsd_svname);
 444         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 445                                                        qsd->qsd_root,
 446                                                        &qqi->qqi_fid, &uuid,
 447                                                        true);
 448         if (IS_ERR(qqi->qqi_slv_obj)) {
 449                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 450                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 451                        PTR_ERR(qqi->qqi_slv_obj));
 452                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 453         }
 454         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 455
 456         /* allocate site */
 457         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 458         if (IS_ERR(qqi->qqi_site)) {
 459                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 460                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 461                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 462         }
 463
 464         /* register proc entry for accounting & global index copy objects */
 465         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 466                                 0444, &lprocfs_quota_seq_fops,
 467                                 qqi->qqi_acct_obj);
 468         if (rc) {
 469                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 470                        qsd->qsd_svname, rc);
 471                 GOTO(out, rc);
 472         }
 473
 474         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 475                                 0444, &lprocfs_quota_seq_fops,
 476                                 qqi->qqi_glb_obj);
 477         if (rc) {
 478                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 479                        qsd->qsd_svname, rc);
 480                 GOTO(out, rc);
 481         }
 482         EXIT;
 483 out:
 484         if (rc)
 485                 qsd_qtype_fini(env, qsd, qtype);
 486         return rc;
 487 }
 488
 489 /*
 490  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 491  * structures associated with the quota slave (on-disk objects, lquota entry
 492  * tables, ...).
 493  * This function should be called when the OSD is shutting down.
 494  *
 495  * \param env - is the environment passed by the caller
 496  * \param qsd - is the qsd instance to shutdown
 497  */
 498 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 499 {
 500         int     qtype;
 501         ENTRY;
 502
 503         if (unlikely(qsd == NULL))
 504                 RETURN_EXIT;
 505
 506         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 507         write_lock(&qsd->qsd_lock);
 508         qsd->qsd_stopping = true;
 509         write_unlock(&qsd->qsd_lock);
 510
 511         /* remove qsd proc entry */
 512         if (qsd->qsd_proc != NULL) {
 513                 lprocfs_remove(&qsd->qsd_proc);
 514                 qsd->qsd_proc = NULL;
 515         }
 516
 517         /* stop the writeback thread */
 518         qsd_stop_upd_thread(qsd);
 519
 520         /* shutdown the reintegration threads */
 521         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 522                 if (qsd->qsd_type_array[qtype] == NULL)
 523                         continue;
 524                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 525         }
 526
 527         if (qsd->qsd_ns != NULL) {
 528                 qsd->qsd_ns = NULL;
 529         }
 530
 531         /* free per-quota type data */
 532         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 533                 qsd_qtype_fini(env, qsd, qtype);
 534
 535         if (qsd->qsd_exp) {
 536                 /* deregister connection to the quota master */
 537                 qsd->qsd_exp_valid = false;
 538                 lustre_deregister_lwp_item(&qsd->qsd_exp);
 539         }
 540
 541         /* release per-filesystem information */
 542         if (qsd->qsd_fsinfo != NULL) {
 543                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 544                 /* remove from the list of fsinfo */
 545                 list_del_init(&qsd->qsd_link);
 546                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 547                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 548                 qsd->qsd_fsinfo = NULL;
 549         }
 550
 551         /* release quota root directory */
 552         if (qsd->qsd_root != NULL) {
 553                 dt_object_put(env, qsd->qsd_root);
 554                 qsd->qsd_root = NULL;
 555         }
 556
 557         /* release reference on dt_device */
 558         if (qsd->qsd_dev != NULL) {
 559                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 560                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 561                 qsd->qsd_dev = NULL;
 562         }
 563
 564         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 565         OBD_FREE_PTR(qsd);
 566         EXIT;
 567 }
 568 EXPORT_SYMBOL(qsd_fini);
 569
 570 /*
 571  * Create a new qsd_instance to be associated with backend osd device
 572  * identified by \dev.
 573  *
 574  * \param env    - the environment passed by the caller
 575  * \param svname - is the service name of the OSD device creating this instance
 576  * \param dev    - is the dt_device where to store quota index files
 577  * \param osd_proc - is the procfs parent directory where to create procfs file
 578  *                   related to this new qsd instance
 579  *
 580  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 581  *           appropriate error on failure
 582  */
 583 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 584                               struct dt_device *dev,
 585                               struct proc_dir_entry *osd_proc)
 586 {
 587         struct qsd_thread_info  *qti = qsd_info(env);
 588         struct qsd_instance     *qsd;
 589         int                      rc, type, idx;
 590         ENTRY;
 591
 592         /* only configure qsd for MDT & OST */
 593         type = server_name2index(svname, &idx, NULL);
 594         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 595                 RETURN(NULL);
 596
 597         /* allocate qsd instance */
 598         OBD_ALLOC_PTR(qsd);
 599         if (qsd == NULL)
 600                 RETURN(ERR_PTR(-ENOMEM));
 601
 602         /* generic initializations */
 603         rwlock_init(&qsd->qsd_lock);
 604         INIT_LIST_HEAD(&qsd->qsd_link);
 605         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 606         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 607         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 608         spin_lock_init(&qsd->qsd_adjust_lock);
 609         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 610         qsd->qsd_prepared = false;
 611         qsd->qsd_started = false;
 612
 613         /* copy service name */
 614         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 615             >= sizeof(qsd->qsd_svname))
 616                 GOTO(out, rc = -E2BIG);
 617
 618         /* grab reference on osd device */
 619         lu_device_get(&dev->dd_lu_dev);
 620         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 621         qsd->qsd_dev = dev;
 622
 623         /* we only support pool ID 0 (default data or metadata pool) for the
 624          * time being. A different pool ID could be assigned to this target via
 625          * the configuration log in the future */
 626         qsd->qsd_pool_id  = 0;
 627
 628         /* get fsname from svname */
 629         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 630         if (rc) {
 631                 CERROR("%s: fail to extract filesystem name\n", svname);
 632                 GOTO(out, rc);
 633         }
 634
 635         /* look up quota setting for the filesystem the target belongs to */
 636         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 637         if (qsd->qsd_fsinfo == NULL) {
 638                 CERROR("%s: failed to locate filesystem information\n", svname);
 639                 GOTO(out, rc = -EINVAL);
 640         }
 641
 642         /* add in the list of lquota_fsinfo */
 643         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 644         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 645         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 646
 647         /* register procfs directory */
 648         qsd->qsd_proc = lprocfs_register(QSD_DIR, osd_proc,
 649                                          lprocfs_quota_qsd_vars, qsd);
 650         if (IS_ERR(qsd->qsd_proc)) {
 651                 rc = PTR_ERR(qsd->qsd_proc);
 652                 qsd->qsd_proc = NULL;
 653                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 654                        svname, rc);
 655                 GOTO(out, rc);
 656         }
 657         EXIT;
 658 out:
 659         if (rc) {
 660                 qsd_fini(env, qsd);
 661                 return ERR_PTR(rc);
 662         }
 663         RETURN(qsd);
 664 }
 665 EXPORT_SYMBOL(qsd_init);
 666
 667 /*
 668  * Initialize on-disk structures in order to manage quota enforcement for
 669  * the target associated with the qsd instance \qsd and starts the reintegration
 670  * procedure for each quota type as soon as possible.
 671  * The last step of the reintegration will be completed once qsd_start() is
 672  * called, at which points the space reconciliation with the master will be
 673  * executed.
 674  * This function must be called when the server stack is fully configured,
 675  * typically when ->ldo_prepare is called across the stack.
 676  *
 677  * \param env - the environment passed by the caller
 678  * \param qsd - is qsd_instance to prepare
 679  *
 680  * \retval - 0 on success, appropriate error on failure
 681  */
 682 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 683 {
 684         struct qsd_thread_info  *qti = qsd_info(env);
 685         int                      qtype, rc = 0;
 686         ENTRY;
 687
 688         if (unlikely(qsd == NULL))
 689                 RETURN(0);
 690
 691         read_lock(&qsd->qsd_lock);
 692         if (qsd->qsd_prepared) {
 693                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 694                 rc = -EALREADY;
 695         }
 696         read_unlock(&qsd->qsd_lock);
 697         if (rc)
 698                 RETURN(rc);
 699
 700         /* Record whether this qsd instance is managing quota enforcement for a
 701          * MDT (i.e. inode quota) or OST (block quota) */
 702         if (lu_device_is_md(qsd->qsd_dev->dd_lu_dev.ld_site->ls_top_dev)) {
 703                 qsd->qsd_is_md = true;
 704                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 705         } else {
 706                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 707         }
 708
 709         /* look-up on-disk directory for the quota slave */
 710         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 711                                                     QSD_DIR);
 712         if (IS_ERR(qsd->qsd_root)) {
 713                 rc = PTR_ERR(qsd->qsd_root);
 714                 qsd->qsd_root = NULL;
 715                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 716                        qsd->qsd_svname, rc);
 717                 RETURN(rc);
 718         }
 719
 720         /* initialize per-quota type data */
 721         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 722                 rc = qsd_qtype_init(env, qsd, qtype);
 723                 if (rc)
 724                         RETURN(rc);
 725         }
 726
 727         /* pools successfully setup, mark the qsd as prepared */
 728         write_lock(&qsd->qsd_lock);
 729         qsd->qsd_prepared = true;
 730         write_unlock(&qsd->qsd_lock);
 731
 732         if (qsd->qsd_dev->dd_rdonly)
 733                 RETURN(0);
 734
 735         /* start reintegration thread for each type, if required */
 736         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 737                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 738
 739                 if (qsd_type_enabled(qsd, qtype) && qsd->qsd_acct_failed) {
 740                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 741                                        "since space accounting isn't functional"
 742                                        ". Please run tunefs.lustre --quota on "
 743                                        "an unmounted filesystem if not done "
 744                                        "already\n", qsd->qsd_svname);
 745                         break;
 746                 }
 747
 748                 rc = qsd_start_reint_thread(qqi);
 749                 if (rc) {
 750                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 751                                 qsd->qsd_svname, qtype_name(qtype), rc);
 752                         RETURN(rc);
 753                 }
 754         }
 755
 756         /* start writeback thread */
 757         rc = qsd_start_upd_thread(qsd);
 758         if (rc) {
 759                 CERROR("%s: failed to start writeback thread (%d)\n",
 760                        qsd->qsd_svname, rc);
 761                 RETURN(rc);
 762         }
 763
 764         /* generate osp name */
 765         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 766                                MTI_NAME_MAXLEN, 0);
 767         if (rc) {
 768                 CERROR("%s: failed to generate ospname (%d)\n",
 769                        qsd->qsd_svname, rc);
 770                 RETURN(rc);
 771         }
 772
 773         /* the connection callback will start the reintegration
 774          * procedure if quota is enabled */
 775         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 776                                       qsd_conn_callback, (void *)qsd);
 777         if (rc) {
 778                 CERROR("%s: fail to get connection to master (%d)\n",
 779                        qsd->qsd_svname, rc);
 780                 RETURN(rc);
 781         }
 782
 783         RETURN(0);
 784 }
 785 EXPORT_SYMBOL(qsd_prepare);
 786
 787 /*
 788  * Start a qsd instance. This will complete the last step of the reintegration
 789  * procedure as soon as possible (provided that the master is reachable).
 790  * This should be called when recovery has been completed and quota should now
 791  * be enforced on every operations.
 792  *
 793  * \param env - the environment passed by the caller
 794  * \param qsd - is the qsd instance associated with the osd device to start
 795  */
 796 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 797 {
 798         int     type, rc = 0;
 799         ENTRY;
 800
 801         if (unlikely(qsd == NULL))
 802                 RETURN(0);
 803
 804         write_lock(&qsd->qsd_lock);
 805         if (!qsd->qsd_prepared) {
 806                 CERROR("%s: can't start qsd instance since it wasn't properly "
 807                        "initialized\n", qsd->qsd_svname);
 808                 rc = -EFAULT;
 809         } else if (qsd->qsd_started) {
 810                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 811                 rc = -EALREADY;
 812         } else {
 813                 /* notify that the qsd_instance is now started */
 814                 qsd->qsd_started = true;
 815         }
 816         write_unlock(&qsd->qsd_lock);
 817
 818         if (rc)
 819                 RETURN(rc);
 820
 821         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 822          * up to usage; If usage < granted, release down to usage.  */
 823         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 824                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 825                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 826         }
 827
 828         RETURN(rc);
 829 }
 830 EXPORT_SYMBOL(qsd_start);
 831
 832 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 833
 834 /*
 835  * Global initialization performed at module load time
 836  */
 837 int qsd_glb_init(void)
 838 {
 839         int     rc;
 840
 841         rc = lu_kmem_init(qsd_caches);
 842         if (rc)
 843                 return rc;
 844
 845         qsd_key_init_generic(&qsd_thread_key, NULL);
 846         lu_context_key_register(&qsd_thread_key);
 847         lustre_register_quota_process_config(qsd_process_config);
 848
 849         return 0;
 850 }
 851
 852 /*
 853  * Companion of qsd_glb_init() called at module unload time
 854  */
 855 void qsd_glb_fini(void)
 856 {
 857         lustre_register_quota_process_config(NULL);
 858         lu_kmem_fini(qsd_caches);
 859         lu_context_key_degister(&qsd_thread_key);
 860 }