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_MG_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         int types = 0;
 138
 139         LASSERT(qsd != NULL);
 140
 141         memset(enabled, 0, sizeof(enabled));
 142         if (qsd_type_enabled(qsd, USRQUOTA))
 143                 enabled[types++] = 'u';
 144         if (qsd_type_enabled(qsd, GRPQUOTA))
 145                 enabled[types++] = 'g';
 146         if (qsd_type_enabled(qsd, PRJQUOTA))
 147                 enabled[types++] = 'p';
 148         if (!types)
 149                 strncpy(enabled, "none", 4);
 150
 151         seq_printf(m, "%s\n", enabled);
 152         return 0;
 153 }
 154
 155 static ssize_t qsd_enabled_seq_write(struct file *file,
 156                                      const char __user *buffer,
 157                                      size_t count, loff_t *off)
 158 {
 159         struct seq_file *m = file->private_data;
 160         struct qsd_instance *qsd = m->private;
 161         char fsname[LUSTRE_MAXFSNAME + 1];
 162         int enabled = 0;
 163         char valstr[5];
 164         int pool, rc;
 165
 166         if (count > 4)
 167                 return -E2BIG;
 168
 169         if (copy_from_user(valstr, buffer, count))
 170                 GOTO(out, count = -EFAULT);
 171
 172         valstr[sizeof(valstr) - 1] = 0;
 173         if (strchr(valstr, 'u'))
 174                 enabled |= BIT(USRQUOTA);
 175         if (strchr(valstr, 'g'))
 176                 enabled |= BIT(GRPQUOTA);
 177         if (strchr(valstr, 'p'))
 178                 enabled |= BIT(PRJQUOTA);
 179
 180         if (enabled == 0 && strcmp(valstr, "none"))
 181                 GOTO(out, count = -EINVAL);
 182
 183         if (qsd->qsd_is_md)
 184                 pool = LQUOTA_RES_MD;
 185         else
 186                 pool = LQUOTA_RES_DT;
 187
 188         if (server_name2fsname(qsd->qsd_svname, fsname, NULL))
 189                 GOTO(out, count = -EINVAL);
 190
 191         rc = qsd_config(valstr, fsname, pool);
 192         if (rc)
 193                 count = rc;
 194 out:
 195         return count;
 196 }
 197 LPROC_SEQ_FOPS(qsd_enabled);
 198
 199 /* force reintegration procedure to be executed.
 200  * Used for test/debugging purpose */
 201 static ssize_t
 202 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 203                                 size_t count, loff_t *off)
 204 {
 205         struct seq_file     *m = file->private_data;
 206         struct qsd_instance *qsd = m->private;
 207         int                  rc = 0, qtype;
 208
 209         LASSERT(qsd != NULL);
 210
 211         write_lock(&qsd->qsd_lock);
 212         if (qsd->qsd_stopping) {
 213                 /* don't mess up with shutdown procedure, it is already
 214                  * complicated enough */
 215                 rc = -ESHUTDOWN;
 216         } else if (!qsd->qsd_prepared) {
 217                 rc = -EAGAIN;
 218         } else {
 219                 /* mark all indexes as stale */
 220                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 221                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 222                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 223                 }
 224         }
 225         write_unlock(&qsd->qsd_lock);
 226
 227         if (rc)
 228                 return rc;
 229
 230         /* kick off reintegration */
 231         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 232                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 233                 if (rc)
 234                         break;
 235         }
 236         return rc == 0 ? count : rc;
 237 }
 238 LPROC_SEQ_FOPS_WR_ONLY(qsd, force_reint);
 239
 240 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 241 {
 242         struct qsd_instance *qsd = m->private;
 243         LASSERT(qsd != NULL);
 244
 245         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 246         return 0;
 247 }
 248
 249 static ssize_t
 250 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 251                         size_t count, loff_t *off)
 252 {
 253         struct seq_file *m = file->private_data;
 254         struct qsd_instance *qsd = m->private;
 255         time64_t timeout;
 256         int rc;
 257
 258         LASSERT(qsd != NULL);
 259         rc = kstrtoll_from_user(buffer, count, 0, &timeout);
 260         if (rc)
 261                 return rc;
 262
 263         if (timeout < 0)
 264                 return -EINVAL;
 265
 266         qsd->qsd_timeout = timeout;
 267         return count;
 268 }
 269 LPROC_SEQ_FOPS(qsd_timeout);
 270
 271 static int qsd_root_prj_enable_seq_show(struct seq_file *m, void *data)
 272 {
 273         struct qsd_instance *qsd = m->private;
 274
 275         LASSERT(qsd != NULL);
 276         seq_printf(m, "%d\n", qsd->qsd_root_prj_enable);
 277         return 0;
 278 }
 279
 280 static ssize_t
 281 qsd_root_prj_enable_seq_write(struct file *file, const char __user *buffer,
 282                         size_t count, loff_t *off)
 283 {
 284         struct seq_file *m = file->private_data;
 285         struct qsd_instance *qsd = m->private;
 286         bool enable;
 287         int rc;
 288
 289         LASSERT(qsd != NULL);
 290         rc = kstrtobool_from_user(buffer, count, &enable);
 291         if (rc)
 292                 return rc;
 293
 294         qsd->qsd_root_prj_enable = enable;
 295         return count;
 296 }
 297 LPROC_SEQ_FOPS(qsd_root_prj_enable);
 298
 299 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 300         { .name =       "info",
 301           .fops =       &qsd_state_fops         },
 302         { .name =       "enabled",
 303           .fops =       &qsd_enabled_fops       },
 304         { .name =       "force_reint",
 305           .fops =       &qsd_force_reint_fops   },
 306         { .name =       "timeout",
 307           .fops =       &qsd_timeout_fops       },
 308         { .name =       "root_prj_enable",
 309           .fops =       &qsd_root_prj_enable_fops       },
 310         { NULL }
 311 };
 312
 313 /*
 314  * Callback function invoked by the OSP layer when the connection to the master
 315  * has been set up.
 316  *
 317  * \param data - is a pointer to the qsd_instance
 318  *
 319  * \retval - 0 on success, appropriate error on failure
 320  */
 321 static int qsd_conn_callback(void *data)
 322 {
 323         struct qsd_instance *qsd = (struct qsd_instance *)data;
 324         int                  type;
 325         ENTRY;
 326
 327         /* qsd_exp should now be valid */
 328         LASSERT(qsd->qsd_exp);
 329
 330         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 331
 332         write_lock(&qsd->qsd_lock);
 333         /* notify that qsd_exp is now valid */
 334         qsd->qsd_exp_valid = true;
 335         write_unlock(&qsd->qsd_lock);
 336
 337         /* Now that the connection to master is setup, we can initiate the
 338          * reintegration procedure for quota types which are enabled.
 339          * It is worth noting that, if the qsd_instance hasn't been started
 340          * already, then we can only complete the first two steps of the
 341          * reintegration procedure (i.e. global lock enqueue and slave
 342          * index transfer) since the space usage reconciliation (i.e.
 343          * step 3) will have to wait for qsd_start() to be called */
 344         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 345                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 346                 struct task_struct *t;
 347
 348                 /* qqi_reint_task can be set to NULL at any time,
 349                  * so we need to be careful.
 350                  */
 351                 rcu_read_lock();
 352                 t = rcu_dereference(qqi->qqi_reint_task);
 353                 if (t)
 354                         wake_up_process(t);
 355                 rcu_read_unlock();
 356         }
 357
 358         RETURN(0);
 359 }
 360
 361 /*
 362  * Release qsd_qtype_info structure which contains data associated with a
 363  * given quota type. This releases the accounting objects.
 364  * It's called on OSD cleanup when the qsd instance is released.
 365  *
 366  * \param env - is the environment passed by the caller
 367  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 368  *              to be released
 369  * \param qtype - is the quota type to be shutdown
 370  */
 371 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 372                            int qtype)
 373 {
 374         struct qsd_qtype_info   *qqi;
 375         int repeat = 0;
 376         ENTRY;
 377
 378         if (qsd->qsd_type_array[qtype] == NULL)
 379                 RETURN_EXIT;
 380         qqi = qsd->qsd_type_array[qtype];
 381         qsd->qsd_type_array[qtype] = NULL;
 382
 383         /* all deferred work lists should be empty */
 384         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 385         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 386
 387         /* shutdown lquota site */
 388         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 389                 lquota_site_free(env, qqi->qqi_site);
 390                 qqi->qqi_site = NULL;
 391         }
 392
 393         /* The qqi may still be holding by global locks which are being
 394          * canceled asynchronously (LU-4365), see the following steps:
 395          *
 396          * - On server umount, we try to clear all quota locks first by
 397          *   disconnecting LWP (which will invalidate import and cleanup
 398          *   all locks on it), however, if quota reint process is holding
 399          *   the global lock for reintegration at that time, global lock
 400          *   will fail to be cleared on LWP disconnection.
 401          *
 402          * - Umount process goes on and stops reint process, the global
 403          *   lock will be dropped on reint process exit, however, the lock
 404          *   cancel in done in asynchronous way, so the
 405          *   qsd_glb_blocking_ast() might haven't been called yet when we
 406          *   get here.
 407          */
 408         while (atomic_read(&qqi->qqi_ref) > 1) {
 409                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 410                        atomic_read(&qqi->qqi_ref), repeat);
 411                 repeat++;
 412                 schedule_timeout_interruptible(cfs_time_seconds(1));
 413         }
 414
 415         /* by now, all qqi users should have gone away */
 416         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 417         lu_ref_fini(&qqi->qqi_reference);
 418
 419         /* release accounting object */
 420         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 421                 dt_object_put(env, qqi->qqi_acct_obj);
 422                 qqi->qqi_acct_obj = NULL;
 423         }
 424
 425         /* release slv index */
 426         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 427                 dt_object_put(env, qqi->qqi_slv_obj);
 428                 qqi->qqi_slv_obj = NULL;
 429                 qqi->qqi_slv_ver = 0;
 430         }
 431
 432         /* release global index */
 433         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 434                 dt_object_put(env, qqi->qqi_glb_obj);
 435                 qqi->qqi_glb_obj = NULL;
 436                 qqi->qqi_glb_ver = 0;
 437         }
 438
 439         OBD_FREE_PTR(qqi);
 440         EXIT;
 441 }
 442
 443 static const char *qtype2acct_name(int qtype)
 444 {
 445         static char unknown[24];
 446
 447         switch (qtype) {
 448         case USRQUOTA:
 449                 return "acct_user";
 450         case GRPQUOTA:
 451                 return "acct_group";
 452         case PRJQUOTA:
 453                 return "acct_project";
 454         }
 455
 456         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 457         return unknown;
 458 }
 459
 460 static const char *qtype2glb_name(int qtype)
 461 {
 462         static char unknown[24];
 463
 464         switch (qtype) {
 465         case USRQUOTA:
 466                 return "limit_user";
 467         case GRPQUOTA:
 468                 return "limit_group";
 469         case PRJQUOTA:
 470                 return "limit_project";
 471         }
 472
 473         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 474         return unknown;
 475 }
 476
 477 /*
 478  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 479  * This opens the accounting object and initializes the proc file.
 480  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 481  * instance.
 482  *
 483  * \param env  - the environment passed by the caller
 484  * \param qsd  - is the qsd instance which will be in charge of the new
 485  *               qsd_qtype_info instance.
 486  * \param qtype - is quota type to set up
 487  *
 488  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 489  *           appropriate error on failure
 490  */
 491 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 492                           int qtype)
 493 {
 494         struct qsd_qtype_info   *qqi;
 495         int                      rc;
 496         struct obd_uuid          uuid;
 497         ENTRY;
 498
 499         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 500
 501         /* allocate structure for this quota type */
 502         OBD_ALLOC_PTR(qqi);
 503         if (qqi == NULL)
 504                 RETURN(-ENOMEM);
 505         qsd->qsd_type_array[qtype] = qqi;
 506         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 507
 508         /* set backpointer and other parameters */
 509         qqi->qqi_qsd   = qsd;
 510         qqi->qqi_qtype = qtype;
 511         lu_ref_init(&qqi->qqi_reference);
 512         qqi->qqi_glb_uptodate = false;
 513         qqi->qqi_slv_uptodate = false;
 514         qqi->qqi_reint        = false;
 515         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 516         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 517         lquota_generate_fid(&qqi->qqi_fid, QSD_RES_TYPE(qsd), qtype);
 518
 519         /* open accounting object */
 520         LASSERT(qqi->qqi_acct_obj == NULL);
 521         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 522         if (IS_ERR(qqi->qqi_acct_obj)) {
 523                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 524                        qsd->qsd_svname, qtype_name(qtype),
 525                        PTR_ERR(qqi->qqi_acct_obj));
 526                 qqi->qqi_acct_obj = NULL;
 527                 qqi->qqi_acct_failed = true;
 528         }
 529
 530         /* open global index copy */
 531         LASSERT(qqi->qqi_glb_obj == NULL);
 532         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 533                                                        qsd->qsd_root,
 534                                                        &qqi->qqi_fid, true);
 535         if (IS_ERR(qqi->qqi_glb_obj)) {
 536                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 537                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 538                        PTR_ERR(qqi->qqi_glb_obj));
 539                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 540         }
 541         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 542
 543         /* open slave index copy */
 544         LASSERT(qqi->qqi_slv_obj == NULL);
 545         obd_str2uuid(&uuid, qsd->qsd_svname);
 546         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 547                                                        qsd->qsd_root,
 548                                                        &qqi->qqi_fid, &uuid,
 549                                                        true);
 550         if (IS_ERR(qqi->qqi_slv_obj)) {
 551                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 552                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 553                        PTR_ERR(qqi->qqi_slv_obj));
 554                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 555         }
 556         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 557
 558         /* allocate site */
 559         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 560         if (IS_ERR(qqi->qqi_site)) {
 561                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 562                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 563                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 564         }
 565
 566         /* register proc entry for accounting & global index copy objects */
 567         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 568                                 0444, &lprocfs_quota_seq_fops,
 569                                 qqi->qqi_acct_obj);
 570         if (rc) {
 571                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 572                        qsd->qsd_svname, rc);
 573                 GOTO(out, rc);
 574         }
 575
 576         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 577                                 0444, &lprocfs_quota_seq_fops,
 578                                 qqi->qqi_glb_obj);
 579         if (rc) {
 580                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 581                        qsd->qsd_svname, rc);
 582                 GOTO(out, rc);
 583         }
 584         EXIT;
 585 out:
 586         if (rc)
 587                 qsd_qtype_fini(env, qsd, qtype);
 588         return rc;
 589 }
 590
 591 /*
 592  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 593  * structures associated with the quota slave (on-disk objects, lquota entry
 594  * tables, ...).
 595  * This function should be called when the OSD is shutting down.
 596  *
 597  * \param env - is the environment passed by the caller
 598  * \param qsd - is the qsd instance to shutdown
 599  */
 600 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 601 {
 602         int     qtype;
 603         ENTRY;
 604
 605         if (unlikely(qsd == NULL))
 606                 RETURN_EXIT;
 607
 608         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 609         write_lock(&qsd->qsd_lock);
 610         qsd->qsd_stopping = true;
 611         write_unlock(&qsd->qsd_lock);
 612
 613         /* remove qsd proc entry */
 614         if (qsd->qsd_proc != NULL) {
 615                 lprocfs_remove(&qsd->qsd_proc);
 616                 qsd->qsd_proc = NULL;
 617         }
 618
 619         /* stop the writeback thread */
 620         qsd_stop_upd_thread(qsd);
 621
 622         /* shutdown the reintegration threads */
 623         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 624                 if (qsd->qsd_type_array[qtype] == NULL)
 625                         continue;
 626                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 627         }
 628
 629         if (qsd->qsd_ns != NULL) {
 630                 qsd->qsd_ns = NULL;
 631         }
 632
 633         /* release per-filesystem information */
 634         if (qsd->qsd_fsinfo != NULL) {
 635                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 636                 /* remove from the list of fsinfo */
 637                 list_del_init(&qsd->qsd_link);
 638                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 639                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 640                 qsd->qsd_fsinfo = NULL;
 641         }
 642
 643         /* free per-quota type data */
 644         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 645                 qsd_qtype_fini(env, qsd, qtype);
 646
 647         /* deregister connection to the quota master */
 648         qsd->qsd_exp_valid = false;
 649         lustre_deregister_lwp_item(&qsd->qsd_exp);
 650
 651         /* release quota root directory */
 652         if (qsd->qsd_root != NULL) {
 653                 dt_object_put(env, qsd->qsd_root);
 654                 qsd->qsd_root = NULL;
 655         }
 656
 657         /* release reference on dt_device */
 658         if (qsd->qsd_dev != NULL) {
 659                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 660                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 661                 qsd->qsd_dev = NULL;
 662         }
 663
 664         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 665         OBD_FREE_PTR(qsd);
 666         EXIT;
 667 }
 668 EXPORT_SYMBOL(qsd_fini);
 669
 670 /*
 671  * Create a new qsd_instance to be associated with backend osd device
 672  * identified by \dev.
 673  *
 674  * \param env    - the environment passed by the caller
 675  * \param svname - is the service name of the OSD device creating this instance
 676  * \param dev    - is the dt_device where to store quota index files
 677  * \param osd_proc - is the procfs parent directory where to create procfs file
 678  *                   related to this new qsd instance
 679  *
 680  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 681  *           appropriate error on failure
 682  */
 683 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 684                               struct dt_device *dev,
 685                               struct proc_dir_entry *osd_proc,
 686                               bool is_md, bool excl)
 687 {
 688         struct qsd_thread_info  *qti = qsd_info(env);
 689         struct qsd_instance     *qsd;
 690         int                      rc, type, idx;
 691         ENTRY;
 692
 693         /* only configure qsd for MDT & OST */
 694         type = server_name2index(svname, &idx, NULL);
 695         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 696                 RETURN(NULL);
 697
 698         /* allocate qsd instance */
 699         OBD_ALLOC_PTR(qsd);
 700         if (qsd == NULL)
 701                 RETURN(ERR_PTR(-ENOMEM));
 702
 703         /* generic initializations */
 704         rwlock_init(&qsd->qsd_lock);
 705         INIT_LIST_HEAD(&qsd->qsd_link);
 706         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 707         spin_lock_init(&qsd->qsd_adjust_lock);
 708         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 709         qsd->qsd_prepared = false;
 710         qsd->qsd_started = false;
 711         qsd->qsd_is_md = is_md;
 712         qsd->qsd_updating = false;
 713         qsd->qsd_exclusive = excl;
 714
 715         /* copy service name */
 716         rc = strscpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname));
 717         if (rc < 0)
 718                 GOTO(out, rc);
 719
 720         /* grab reference on osd device */
 721         lu_device_get(&dev->dd_lu_dev);
 722         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 723         qsd->qsd_dev = dev;
 724
 725         /* get fsname from svname */
 726         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 727         if (rc) {
 728                 CERROR("%s: fail to extract filesystem name\n", svname);
 729                 GOTO(out, rc);
 730         }
 731
 732         /* look up quota setting for the filesystem the target belongs to */
 733         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 734         if (qsd->qsd_fsinfo == NULL) {
 735                 CERROR("%s: failed to locate filesystem information\n", svname);
 736                 GOTO(out, rc = -EINVAL);
 737         }
 738
 739         /* add in the list of lquota_fsinfo */
 740         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 741         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 742         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 743
 744         /* register procfs directory */
 745         if (qsd->qsd_is_md)
 746                 qsd->qsd_proc = lprocfs_register(QSD_DIR_MD, osd_proc,
 747                                                  lprocfs_quota_qsd_vars, qsd);
 748         else
 749                 qsd->qsd_proc = lprocfs_register(QSD_DIR_DT, osd_proc,
 750                                                  lprocfs_quota_qsd_vars, qsd);
 751
 752         if (type == LDD_F_SV_TYPE_MDT && qsd->qsd_is_md)
 753                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_MD);
 754         else if (type == LDD_F_SV_TYPE_OST && !qsd->qsd_is_md)
 755                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_DT);
 756
 757         if (IS_ERR(qsd->qsd_proc)) {
 758                 rc = PTR_ERR(qsd->qsd_proc);
 759                 qsd->qsd_proc = NULL;
 760                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 761                        svname, rc);
 762                 GOTO(out, rc);
 763         }
 764         EXIT;
 765 out:
 766         if (rc) {
 767                 qsd_fini(env, qsd);
 768                 return ERR_PTR(rc);
 769         }
 770         RETURN(qsd);
 771 }
 772 EXPORT_SYMBOL(qsd_init);
 773
 774 /*
 775  * Initialize on-disk structures in order to manage quota enforcement for
 776  * the target associated with the qsd instance \qsd and starts the reintegration
 777  * procedure for each quota type as soon as possible.
 778  * The last step of the reintegration will be completed once qsd_start() is
 779  * called, at which points the space reconciliation with the master will be
 780  * executed.
 781  * This function must be called when the server stack is fully configured,
 782  * typically when ->ldo_prepare is called across the stack.
 783  *
 784  * \param env - the environment passed by the caller
 785  * \param qsd - is qsd_instance to prepare
 786  *
 787  * \retval - 0 on success, appropriate error on failure
 788  */
 789 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 790 {
 791         struct qsd_thread_info  *qti = qsd_info(env);
 792         int                      qtype, rc = 0;
 793         ENTRY;
 794
 795         if (unlikely(qsd == NULL))
 796                 RETURN(0);
 797
 798         read_lock(&qsd->qsd_lock);
 799         if (qsd->qsd_prepared) {
 800                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 801                 rc = -EALREADY;
 802         }
 803         read_unlock(&qsd->qsd_lock);
 804         if (rc)
 805                 RETURN(rc);
 806
 807         /* Record whether this qsd instance is managing quota enforcement for a
 808          * MDT (i.e. inode quota) or OST (block quota) */
 809         if (qsd->qsd_is_md)
 810                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 811         else
 812                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 813
 814         /* look-up on-disk directory for the quota slave */
 815         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 816                                                     QSD_DIR);
 817         if (IS_ERR(qsd->qsd_root)) {
 818                 rc = PTR_ERR(qsd->qsd_root);
 819                 qsd->qsd_root = NULL;
 820                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 821                        qsd->qsd_svname, rc);
 822                 RETURN(rc);
 823         }
 824
 825         /* initialize per-quota type data */
 826         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 827                 rc = qsd_qtype_init(env, qsd, qtype);
 828                 if (rc)
 829                         RETURN(rc);
 830         }
 831
 832         /* pools successfully setup, mark the qsd as prepared */
 833         write_lock(&qsd->qsd_lock);
 834         qsd->qsd_prepared = true;
 835         write_unlock(&qsd->qsd_lock);
 836
 837         if (qsd->qsd_dev->dd_rdonly)
 838                 RETURN(0);
 839
 840         /* start reintegration thread for each type, if required */
 841         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 842                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 843
 844                 if (qsd_type_enabled(qsd, qtype) &&
 845                     qqi->qqi_acct_failed) {
 846                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 847                                        "since space accounting isn't functional"
 848                                        ". Please run tunefs.lustre --quota on "
 849                                        "an unmounted filesystem if not done "
 850                                        "already\n", qsd->qsd_svname);
 851                         continue;
 852                 }
 853
 854                 rc = qsd_start_reint_thread(qqi);
 855                 if (rc) {
 856                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 857                                 qsd->qsd_svname, qtype_name(qtype), rc);
 858                         RETURN(rc);
 859                 }
 860         }
 861
 862         /* start writeback thread */
 863         rc = qsd_start_upd_thread(qsd);
 864         if (rc) {
 865                 CERROR("%s: failed to start writeback thread (%d)\n",
 866                        qsd->qsd_svname, rc);
 867                 RETURN(rc);
 868         }
 869
 870         /* generate osp name */
 871         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 872                                MTI_NAME_MAXLEN, 0);
 873         if (rc) {
 874                 CERROR("%s: failed to generate ospname (%d)\n",
 875                        qsd->qsd_svname, rc);
 876                 RETURN(rc);
 877         }
 878
 879         /* the connection callback will start the reintegration
 880          * procedure if quota is enabled */
 881         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 882                                       qsd_conn_callback, (void *)qsd);
 883         if (rc) {
 884                 CERROR("%s: fail to get connection to master (%d)\n",
 885                        qsd->qsd_svname, rc);
 886                 RETURN(rc);
 887         }
 888
 889         RETURN(0);
 890 }
 891 EXPORT_SYMBOL(qsd_prepare);
 892
 893 /*
 894  * Start a qsd instance. This will complete the last step of the reintegration
 895  * procedure as soon as possible (provided that the master is reachable).
 896  * This should be called when recovery has been completed and quota should now
 897  * be enforced on every operations.
 898  *
 899  * \param env - the environment passed by the caller
 900  * \param qsd - is the qsd instance associated with the osd device to start
 901  */
 902 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 903 {
 904         int     type, rc = 0;
 905         ENTRY;
 906
 907         if (unlikely(qsd == NULL))
 908                 RETURN(0);
 909
 910         write_lock(&qsd->qsd_lock);
 911         if (!qsd->qsd_prepared) {
 912                 CERROR("%s: can't start qsd instance since it wasn't properly "
 913                        "initialized\n", qsd->qsd_svname);
 914                 rc = -EFAULT;
 915         } else if (qsd->qsd_started) {
 916                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 917                 rc = -EALREADY;
 918         } else {
 919                 /* notify that the qsd_instance is now started */
 920                 qsd->qsd_started = true;
 921         }
 922         write_unlock(&qsd->qsd_lock);
 923
 924         if (rc)
 925                 RETURN(rc);
 926
 927         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 928          * up to usage; If usage < granted, release down to usage.  */
 929         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 930                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 931                 struct task_struct *t;
 932
 933                 /* qqi_reint_task can be set to NULL at any time,
 934                  * so we need to be careful.
 935                  */
 936                 rcu_read_lock();
 937                 t = rcu_dereference(qqi->qqi_reint_task);
 938                 if (t)
 939                         wake_up_process(t);
 940                 rcu_read_unlock();
 941         }
 942
 943         RETURN(rc);
 944 }
 945 EXPORT_SYMBOL(qsd_start);
 946
 947 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 948
 949 /*
 950  * Global initialization performed at module load time
 951  */
 952 int qsd_glb_init(void)
 953 {
 954         int     rc;
 955
 956         rc = lu_kmem_init(qsd_caches);
 957         if (rc)
 958                 return rc;
 959
 960         qsd_key_init_generic(&qsd_thread_key, NULL);
 961         lu_context_key_register(&qsd_thread_key);
 962         lustre_register_quota_process_config(qsd_process_config);
 963
 964         return 0;
 965 }
 966
 967 /*
 968  * Companion of qsd_glb_init() called at module unload time
 969  */
 970 void qsd_glb_fini(void)
 971 {
 972         lustre_register_quota_process_config(NULL);
 973         lu_kmem_fini(qsd_caches);
 974         lu_context_key_degister(&qsd_thread_key);
 975 }