lustre/quota/qsd_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2012, 2017, Intel Corporation.
  25  * Use is subject to license terms.
  26  *
  27  * Author: Johann Lombardi <johann.lombardi@intel.com>
  28  * Author: Niu    Yawei    <yawei.niu@intel.com>
  29  */
  30
  31 /*
  32  * Quota Slave Driver (QSD) management.
  33  *
  34  * The quota slave feature is implemented under the form of a library called
  35  * QSD. Each OSD device should create a QSD instance via qsd_init() which will
  36  * be used to manage quota enforcement for this device. This implies:
  37  * - completing the reintegration procedure with the quota master (aka QMT, see
  38  *   qmt_dev.c) to retrieve the latest quota settings and space distribution.
  39  * - managing quota locks in order to be notified of configuration changes.
  40  * - acquiring space from the QMT when quota space for a given user/group is
  41  *   close to exhaustion.
  42  * - allocating quota space to service threads for local request processing.
  43  *
  44  * Once the QSD instance created, the OSD device should invoke qsd_start()
  45  * when recovery is completed. This notifies the QSD that we are about to
  46  * process new requests on which quota should be strictly enforced.
  47  * Then, qsd_op_begin/end can be used to reserve/release/pre-acquire quota space
  48  * for/after each operation until shutdown where the QSD instance should be
  49  * freed via qsd_fini().
  50  */
  51
  52 #define DEBUG_SUBSYSTEM S_LQUOTA
  53
  54 #include <obd_class.h>
  55 #include "qsd_internal.h"
  56
  57 struct kmem_cache *upd_kmem;
  58
  59 struct lu_kmem_descr qsd_caches[] = {
  60         {
  61                 .ckd_cache = &upd_kmem,
  62                 .ckd_name  = "upd_kmem",
  63                 .ckd_size  = sizeof(struct qsd_upd_rec)
  64         },
  65         {
  66                 .ckd_cache = NULL
  67         }
  68 };
  69
  70 /* define qsd thread key */
  71 LU_KEY_INIT_FINI(qsd, struct qsd_thread_info);
  72 LU_CONTEXT_KEY_DEFINE(qsd, LCT_MD_THREAD | LCT_DT_THREAD | LCT_LOCAL);
  73 LU_KEY_INIT_GENERIC(qsd);
  74
  75 /* some procfs helpers */
  76 static int qsd_state_seq_show(struct seq_file *m, void *data)
  77 {
  78         struct qsd_instance     *qsd = m->private;
  79         char                     enabled[5];
  80
  81         LASSERT(qsd != NULL);
  82
  83         memset(enabled, 0, sizeof(enabled));
  84         if (qsd_type_enabled(qsd, USRQUOTA))
  85                 strcat(enabled, "u");
  86         if (qsd_type_enabled(qsd, GRPQUOTA))
  87                 strcat(enabled, "g");
  88         if (qsd_type_enabled(qsd, PRJQUOTA))
  89                 strncat(enabled, "p", 1);
  90         if (strlen(enabled) == 0)
  91                 strcat(enabled, "none");
  92
  93         /* TODO: further pool ID should be removed or
  94          * replaced with pool Name */
  95         seq_printf(m, "target name:    %s\n"
  96                    "pool ID:        %d\n"
  97                    "type:           %s\n"
  98                    "quota enabled:  %s\n"
  99                    "conn to master: %s\n",
 100                    qsd->qsd_svname, 0,
 101                    qsd->qsd_is_md ? "md" : "dt", enabled,
 102                    qsd->qsd_exp_valid ? "setup" : "not setup yet");
 103
 104         if (qsd->qsd_prepared) {
 105                 memset(enabled, 0, sizeof(enabled));
 106                 if (qsd->qsd_type_array[USRQUOTA]->qqi_acct_obj != NULL)
 107                         strcat(enabled, "u");
 108                 if (qsd->qsd_type_array[GRPQUOTA]->qqi_acct_obj != NULL)
 109                         strcat(enabled, "g");
 110                 if (qsd->qsd_type_array[PRJQUOTA]->qqi_acct_obj != NULL)
 111                         strncat(enabled, "p", 1);
 112                 if (strlen(enabled) == 0)
 113                         strcat(enabled, "none");
 114                 seq_printf(m, "space acct:     %s\n"
 115                            "user uptodate:  glb[%d],slv[%d],reint[%d]\n"
 116                            "group uptodate: glb[%d],slv[%d],reint[%d]\n"
 117                            "project uptodate: glb[%d],slv[%d],reint[%d]\n",
 118                            enabled,
 119                            qsd->qsd_type_array[USRQUOTA]->qqi_glb_uptodate,
 120                            qsd->qsd_type_array[USRQUOTA]->qqi_slv_uptodate,
 121                            qsd->qsd_type_array[USRQUOTA]->qqi_reint,
 122                            qsd->qsd_type_array[GRPQUOTA]->qqi_glb_uptodate,
 123                            qsd->qsd_type_array[GRPQUOTA]->qqi_slv_uptodate,
 124                            qsd->qsd_type_array[GRPQUOTA]->qqi_reint,
 125                            qsd->qsd_type_array[PRJQUOTA]->qqi_glb_uptodate,
 126                            qsd->qsd_type_array[PRJQUOTA]->qqi_slv_uptodate,
 127                            qsd->qsd_type_array[PRJQUOTA]->qqi_reint);
 128         }
 129         return 0;
 130 }
 131 LPROC_SEQ_FOPS_RO(qsd_state);
 132
 133 static int qsd_enabled_seq_show(struct seq_file *m, void *data)
 134 {
 135         struct qsd_instance     *qsd = m->private;
 136         char                     enabled[5];
 137
 138         LASSERT(qsd != NULL);
 139
 140         memset(enabled, 0, sizeof(enabled));
 141         if (qsd_type_enabled(qsd, USRQUOTA))
 142                 strncat(enabled, "u", sizeof(enabled) - strlen(enabled));
 143         if (qsd_type_enabled(qsd, GRPQUOTA))
 144                 strncat(enabled, "g", sizeof(enabled) - strlen(enabled));
 145         if (qsd_type_enabled(qsd, PRJQUOTA))
 146                 strncat(enabled, "p", sizeof(enabled) - strlen(enabled));
 147         if (strlen(enabled) == 0)
 148                 strncat(enabled, "none", sizeof(enabled) - strlen(enabled));
 149
 150         seq_printf(m, "%s\n", enabled);
 151         return 0;
 152 }
 153
 154 static ssize_t qsd_enabled_seq_write(struct file *file,
 155                                      const char __user *buffer,
 156                                      size_t count, loff_t *off)
 157 {
 158         struct seq_file *m = file->private_data;
 159         struct qsd_instance *qsd = m->private;
 160         char fsname[LUSTRE_MAXFSNAME + 1];
 161         int enabled = 0;
 162         char valstr[5];
 163         int pool, rc;
 164
 165         if (count > 4)
 166                 return -E2BIG;
 167
 168         if (copy_from_user(valstr, buffer, count))
 169                 GOTO(out, count = -EFAULT);
 170
 171         valstr[sizeof(valstr) - 1] = 0;
 172         if (strchr(valstr, 'u'))
 173                 enabled |= BIT(USRQUOTA);
 174         if (strchr(valstr, 'g'))
 175                 enabled |= BIT(GRPQUOTA);
 176         if (strchr(valstr, 'p'))
 177                 enabled |= BIT(PRJQUOTA);
 178
 179         if (enabled == 0 && strcmp(valstr, "none"))
 180                 GOTO(out, count = -EINVAL);
 181
 182         if (qsd->qsd_is_md)
 183                 pool = LQUOTA_RES_MD;
 184         else
 185                 pool = LQUOTA_RES_DT;
 186
 187         if (server_name2fsname(qsd->qsd_svname, fsname, NULL))
 188                 GOTO(out, count = -EINVAL);
 189
 190         rc = qsd_config(valstr, fsname, pool);
 191         if (rc)
 192                 count = rc;
 193 out:
 194         return count;
 195 }
 196 LPROC_SEQ_FOPS(qsd_enabled);
 197
 198 /* force reintegration procedure to be executed.
 199  * Used for test/debugging purpose */
 200 static ssize_t
 201 lprocfs_force_reint_seq_write(struct file *file, const char __user *buffer,
 202                                 size_t count, loff_t *off)
 203 {
 204         struct seq_file     *m = file->private_data;
 205         struct qsd_instance *qsd = m->private;
 206         int                  rc = 0, qtype;
 207
 208         LASSERT(qsd != NULL);
 209
 210         write_lock(&qsd->qsd_lock);
 211         if (qsd->qsd_stopping) {
 212                 /* don't mess up with shutdown procedure, it is already
 213                  * complicated enough */
 214                 rc = -ESHUTDOWN;
 215         } else if (!qsd->qsd_prepared) {
 216                 rc = -EAGAIN;
 217         } else {
 218                 /* mark all indexes as stale */
 219                 for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 220                         qsd->qsd_type_array[qtype]->qqi_glb_uptodate = false;
 221                         qsd->qsd_type_array[qtype]->qqi_slv_uptodate = false;
 222                 }
 223         }
 224         write_unlock(&qsd->qsd_lock);
 225
 226         if (rc)
 227                 return rc;
 228
 229         /* kick off reintegration */
 230         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 231                 rc = qsd_start_reint_thread(qsd->qsd_type_array[qtype]);
 232                 if (rc)
 233                         break;
 234         }
 235         return rc == 0 ? count : rc;
 236 }
 237 LPROC_SEQ_FOPS_WR_ONLY(qsd, force_reint);
 238
 239 static int qsd_timeout_seq_show(struct seq_file *m, void *data)
 240 {
 241         struct qsd_instance *qsd = m->private;
 242         LASSERT(qsd != NULL);
 243
 244         seq_printf(m, "%d\n", qsd_wait_timeout(qsd));
 245         return 0;
 246 }
 247
 248 static ssize_t
 249 qsd_timeout_seq_write(struct file *file, const char __user *buffer,
 250                         size_t count, loff_t *off)
 251 {
 252         struct seq_file *m = file->private_data;
 253         struct qsd_instance *qsd = m->private;
 254         time64_t timeout;
 255         int rc;
 256
 257         LASSERT(qsd != NULL);
 258         rc = kstrtoll_from_user(buffer, count, 0, &timeout);
 259         if (rc)
 260                 return rc;
 261
 262         if (timeout < 0)
 263                 return -EINVAL;
 264
 265         qsd->qsd_timeout = timeout;
 266         return count;
 267 }
 268 LPROC_SEQ_FOPS(qsd_timeout);
 269
 270 static struct lprocfs_vars lprocfs_quota_qsd_vars[] = {
 271         { .name =       "info",
 272           .fops =       &qsd_state_fops         },
 273         { .name =       "enabled",
 274           .fops =       &qsd_enabled_fops       },
 275         { .name =       "force_reint",
 276           .fops =       &qsd_force_reint_fops   },
 277         { .name =       "timeout",
 278           .fops =       &qsd_timeout_fops       },
 279         { NULL }
 280 };
 281
 282 /*
 283  * Callback function invoked by the OSP layer when the connection to the master
 284  * has been set up.
 285  *
 286  * \param data - is a pointer to the qsd_instance
 287  *
 288  * \retval - 0 on success, appropriate error on failure
 289  */
 290 static int qsd_conn_callback(void *data)
 291 {
 292         struct qsd_instance *qsd = (struct qsd_instance *)data;
 293         int                  type;
 294         ENTRY;
 295
 296         /* qsd_exp should now be valid */
 297         LASSERT(qsd->qsd_exp);
 298
 299         qsd->qsd_ns = class_exp2obd(qsd->qsd_exp)->obd_namespace;
 300
 301         write_lock(&qsd->qsd_lock);
 302         /* notify that qsd_exp is now valid */
 303         qsd->qsd_exp_valid = true;
 304         write_unlock(&qsd->qsd_lock);
 305
 306         /* Now that the connection to master is setup, we can initiate the
 307          * reintegration procedure for quota types which are enabled.
 308          * It is worth noting that, if the qsd_instance hasn't been started
 309          * already, then we can only complete the first two steps of the
 310          * reintegration procedure (i.e. global lock enqueue and slave
 311          * index transfer) since the space usage reconciliation (i.e.
 312          * step 3) will have to wait for qsd_start() to be called */
 313         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 314                 struct qsd_qtype_info *qqi = qsd->qsd_type_array[type];
 315                 struct task_struct *t;
 316
 317                 /* qqi_reint_task can be set to NULL at any time,
 318                  * so we need to be careful.
 319                  */
 320                 rcu_read_lock();
 321                 t = rcu_dereference(qqi->qqi_reint_task);
 322                 if (t)
 323                         wake_up_process(t);
 324                 rcu_read_unlock();
 325         }
 326
 327         RETURN(0);
 328 }
 329
 330 /*
 331  * Release qsd_qtype_info structure which contains data associated with a
 332  * given quota type. This releases the accounting objects.
 333  * It's called on OSD cleanup when the qsd instance is released.
 334  *
 335  * \param env - is the environment passed by the caller
 336  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 337  *              to be released
 338  * \param qtype - is the quota type to be shutdown
 339  */
 340 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 341                            int qtype)
 342 {
 343         struct qsd_qtype_info   *qqi;
 344         int repeat = 0;
 345         ENTRY;
 346
 347         if (qsd->qsd_type_array[qtype] == NULL)
 348                 RETURN_EXIT;
 349         qqi = qsd->qsd_type_array[qtype];
 350         qsd->qsd_type_array[qtype] = NULL;
 351
 352         /* all deferred work lists should be empty */
 353         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 354         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 355
 356         /* shutdown lquota site */
 357         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 358                 lquota_site_free(env, qqi->qqi_site);
 359                 qqi->qqi_site = NULL;
 360         }
 361
 362         /* The qqi may still be holding by global locks which are being
 363          * canceled asynchronously (LU-4365), see the following steps:
 364          *
 365          * - On server umount, we try to clear all quota locks first by
 366          *   disconnecting LWP (which will invalidate import and cleanup
 367          *   all locks on it), however, if quota reint process is holding
 368          *   the global lock for reintegration at that time, global lock
 369          *   will fail to be cleared on LWP disconnection.
 370          *
 371          * - Umount process goes on and stops reint process, the global
 372          *   lock will be dropped on reint process exit, however, the lock
 373          *   cancel in done in asynchronous way, so the
 374          *   qsd_glb_blocking_ast() might haven't been called yet when we
 375          *   get here.
 376          */
 377         while (atomic_read(&qqi->qqi_ref) > 1) {
 378                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 379                        atomic_read(&qqi->qqi_ref), repeat);
 380                 repeat++;
 381                 schedule_timeout_interruptible(cfs_time_seconds(1));
 382         }
 383
 384         /* by now, all qqi users should have gone away */
 385         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 386         lu_ref_fini(&qqi->qqi_reference);
 387
 388         /* release accounting object */
 389         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 390                 dt_object_put(env, qqi->qqi_acct_obj);
 391                 qqi->qqi_acct_obj = NULL;
 392         }
 393
 394         /* release slv index */
 395         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 396                 dt_object_put(env, qqi->qqi_slv_obj);
 397                 qqi->qqi_slv_obj = NULL;
 398                 qqi->qqi_slv_ver = 0;
 399         }
 400
 401         /* release global index */
 402         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 403                 dt_object_put(env, qqi->qqi_glb_obj);
 404                 qqi->qqi_glb_obj = NULL;
 405                 qqi->qqi_glb_ver = 0;
 406         }
 407
 408         OBD_FREE_PTR(qqi);
 409         EXIT;
 410 }
 411
 412 static const char *qtype2acct_name(int qtype)
 413 {
 414         static char unknown[24];
 415
 416         switch (qtype) {
 417         case USRQUOTA:
 418                 return "acct_user";
 419         case GRPQUOTA:
 420                 return "acct_group";
 421         case PRJQUOTA:
 422                 return "acct_project";
 423         }
 424
 425         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 426         return unknown;
 427 }
 428
 429 static const char *qtype2glb_name(int qtype)
 430 {
 431         static char unknown[24];
 432
 433         switch (qtype) {
 434         case USRQUOTA:
 435                 return "limit_user";
 436         case GRPQUOTA:
 437                 return "limit_group";
 438         case PRJQUOTA:
 439                 return "limit_project";
 440         }
 441
 442         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 443         return unknown;
 444 }
 445
 446 /*
 447  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 448  * This opens the accounting object and initializes the proc file.
 449  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 450  * instance.
 451  *
 452  * \param env  - the environment passed by the caller
 453  * \param qsd  - is the qsd instance which will be in charge of the new
 454  *               qsd_qtype_info instance.
 455  * \param qtype - is quota type to set up
 456  *
 457  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 458  *           appropriate error on failure
 459  */
 460 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 461                           int qtype)
 462 {
 463         struct qsd_qtype_info   *qqi;
 464         int                      rc;
 465         struct obd_uuid          uuid;
 466         ENTRY;
 467
 468         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 469
 470         /* allocate structure for this quota type */
 471         OBD_ALLOC_PTR(qqi);
 472         if (qqi == NULL)
 473                 RETURN(-ENOMEM);
 474         qsd->qsd_type_array[qtype] = qqi;
 475         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 476
 477         /* set backpointer and other parameters */
 478         qqi->qqi_qsd   = qsd;
 479         qqi->qqi_qtype = qtype;
 480         lu_ref_init(&qqi->qqi_reference);
 481         qqi->qqi_glb_uptodate = false;
 482         qqi->qqi_slv_uptodate = false;
 483         qqi->qqi_reint        = false;
 484         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 485         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 486         lquota_generate_fid(&qqi->qqi_fid, QSD_RES_TYPE(qsd), qtype);
 487
 488         /* open accounting object */
 489         LASSERT(qqi->qqi_acct_obj == NULL);
 490         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 491         if (IS_ERR(qqi->qqi_acct_obj)) {
 492                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 493                        qsd->qsd_svname, qtype_name(qtype),
 494                        PTR_ERR(qqi->qqi_acct_obj));
 495                 qqi->qqi_acct_obj = NULL;
 496                 qqi->qqi_acct_failed = true;
 497         }
 498
 499         /* open global index copy */
 500         LASSERT(qqi->qqi_glb_obj == NULL);
 501         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 502                                                        qsd->qsd_root,
 503                                                        &qqi->qqi_fid, true);
 504         if (IS_ERR(qqi->qqi_glb_obj)) {
 505                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 506                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 507                        PTR_ERR(qqi->qqi_glb_obj));
 508                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 509         }
 510         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 511
 512         /* open slave index copy */
 513         LASSERT(qqi->qqi_slv_obj == NULL);
 514         obd_str2uuid(&uuid, qsd->qsd_svname);
 515         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 516                                                        qsd->qsd_root,
 517                                                        &qqi->qqi_fid, &uuid,
 518                                                        true);
 519         if (IS_ERR(qqi->qqi_slv_obj)) {
 520                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 521                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 522                        PTR_ERR(qqi->qqi_slv_obj));
 523                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 524         }
 525         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 526
 527         /* allocate site */
 528         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 529         if (IS_ERR(qqi->qqi_site)) {
 530                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 531                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 532                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 533         }
 534
 535         /* register proc entry for accounting & global index copy objects */
 536         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 537                                 0444, &lprocfs_quota_seq_fops,
 538                                 qqi->qqi_acct_obj);
 539         if (rc) {
 540                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 541                        qsd->qsd_svname, rc);
 542                 GOTO(out, rc);
 543         }
 544
 545         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 546                                 0444, &lprocfs_quota_seq_fops,
 547                                 qqi->qqi_glb_obj);
 548         if (rc) {
 549                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 550                        qsd->qsd_svname, rc);
 551                 GOTO(out, rc);
 552         }
 553         EXIT;
 554 out:
 555         if (rc)
 556                 qsd_qtype_fini(env, qsd, qtype);
 557         return rc;
 558 }
 559
 560 /*
 561  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 562  * structures associated with the quota slave (on-disk objects, lquota entry
 563  * tables, ...).
 564  * This function should be called when the OSD is shutting down.
 565  *
 566  * \param env - is the environment passed by the caller
 567  * \param qsd - is the qsd instance to shutdown
 568  */
 569 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 570 {
 571         int     qtype;
 572         ENTRY;
 573
 574         if (unlikely(qsd == NULL))
 575                 RETURN_EXIT;
 576
 577         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 578         write_lock(&qsd->qsd_lock);
 579         qsd->qsd_stopping = true;
 580         write_unlock(&qsd->qsd_lock);
 581
 582         /* remove qsd proc entry */
 583         if (qsd->qsd_proc != NULL) {
 584                 lprocfs_remove(&qsd->qsd_proc);
 585                 qsd->qsd_proc = NULL;
 586         }
 587
 588         /* stop the writeback thread */
 589         qsd_stop_upd_thread(qsd);
 590
 591         /* shutdown the reintegration threads */
 592         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 593                 if (qsd->qsd_type_array[qtype] == NULL)
 594                         continue;
 595                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 596         }
 597
 598         if (qsd->qsd_ns != NULL) {
 599                 qsd->qsd_ns = NULL;
 600         }
 601
 602         /* release per-filesystem information */
 603         if (qsd->qsd_fsinfo != NULL) {
 604                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 605                 /* remove from the list of fsinfo */
 606                 list_del_init(&qsd->qsd_link);
 607                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 608                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 609                 qsd->qsd_fsinfo = NULL;
 610         }
 611
 612         /* free per-quota type data */
 613         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 614                 qsd_qtype_fini(env, qsd, qtype);
 615
 616         /* deregister connection to the quota master */
 617         qsd->qsd_exp_valid = false;
 618         lustre_deregister_lwp_item(&qsd->qsd_exp);
 619
 620         /* release quota root directory */
 621         if (qsd->qsd_root != NULL) {
 622                 dt_object_put(env, qsd->qsd_root);
 623                 qsd->qsd_root = NULL;
 624         }
 625
 626         /* release reference on dt_device */
 627         if (qsd->qsd_dev != NULL) {
 628                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 629                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 630                 qsd->qsd_dev = NULL;
 631         }
 632
 633         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 634         OBD_FREE_PTR(qsd);
 635         EXIT;
 636 }
 637 EXPORT_SYMBOL(qsd_fini);
 638
 639 /*
 640  * Create a new qsd_instance to be associated with backend osd device
 641  * identified by \dev.
 642  *
 643  * \param env    - the environment passed by the caller
 644  * \param svname - is the service name of the OSD device creating this instance
 645  * \param dev    - is the dt_device where to store quota index files
 646  * \param osd_proc - is the procfs parent directory where to create procfs file
 647  *                   related to this new qsd instance
 648  *
 649  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 650  *           appropriate error on failure
 651  */
 652 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 653                               struct dt_device *dev,
 654                               struct proc_dir_entry *osd_proc, bool is_md)
 655 {
 656         struct qsd_thread_info  *qti = qsd_info(env);
 657         struct qsd_instance     *qsd;
 658         int                      rc, type, idx;
 659         ENTRY;
 660
 661         /* only configure qsd for MDT & OST */
 662         type = server_name2index(svname, &idx, NULL);
 663         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 664                 RETURN(NULL);
 665
 666         /* allocate qsd instance */
 667         OBD_ALLOC_PTR(qsd);
 668         if (qsd == NULL)
 669                 RETURN(ERR_PTR(-ENOMEM));
 670
 671         /* generic initializations */
 672         rwlock_init(&qsd->qsd_lock);
 673         INIT_LIST_HEAD(&qsd->qsd_link);
 674         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 675         spin_lock_init(&qsd->qsd_adjust_lock);
 676         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 677         qsd->qsd_prepared = false;
 678         qsd->qsd_started = false;
 679         qsd->qsd_is_md = is_md;
 680
 681         /* copy service name */
 682         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 683             >= sizeof(qsd->qsd_svname))
 684                 GOTO(out, rc = -E2BIG);
 685
 686         /* grab reference on osd device */
 687         lu_device_get(&dev->dd_lu_dev);
 688         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 689         qsd->qsd_dev = dev;
 690
 691         /* get fsname from svname */
 692         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 693         if (rc) {
 694                 CERROR("%s: fail to extract filesystem name\n", svname);
 695                 GOTO(out, rc);
 696         }
 697
 698         /* look up quota setting for the filesystem the target belongs to */
 699         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 700         if (qsd->qsd_fsinfo == NULL) {
 701                 CERROR("%s: failed to locate filesystem information\n", svname);
 702                 GOTO(out, rc = -EINVAL);
 703         }
 704
 705         /* add in the list of lquota_fsinfo */
 706         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 707         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 708         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 709
 710         /* register procfs directory */
 711         if (qsd->qsd_is_md)
 712                 qsd->qsd_proc = lprocfs_register(QSD_DIR_MD, osd_proc,
 713                                                  lprocfs_quota_qsd_vars, qsd);
 714         else
 715                 qsd->qsd_proc = lprocfs_register(QSD_DIR_DT, osd_proc,
 716                                                  lprocfs_quota_qsd_vars, qsd);
 717
 718         if (type == LDD_F_SV_TYPE_MDT && qsd->qsd_is_md)
 719                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_MD);
 720         else if (type == LDD_F_SV_TYPE_OST && !qsd->qsd_is_md)
 721                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_DT);
 722
 723         if (IS_ERR(qsd->qsd_proc)) {
 724                 rc = PTR_ERR(qsd->qsd_proc);
 725                 qsd->qsd_proc = NULL;
 726                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 727                        svname, rc);
 728                 GOTO(out, rc);
 729         }
 730         EXIT;
 731 out:
 732         if (rc) {
 733                 qsd_fini(env, qsd);
 734                 return ERR_PTR(rc);
 735         }
 736         RETURN(qsd);
 737 }
 738 EXPORT_SYMBOL(qsd_init);
 739
 740 /*
 741  * Initialize on-disk structures in order to manage quota enforcement for
 742  * the target associated with the qsd instance \qsd and starts the reintegration
 743  * procedure for each quota type as soon as possible.
 744  * The last step of the reintegration will be completed once qsd_start() is
 745  * called, at which points the space reconciliation with the master will be
 746  * executed.
 747  * This function must be called when the server stack is fully configured,
 748  * typically when ->ldo_prepare is called across the stack.
 749  *
 750  * \param env - the environment passed by the caller
 751  * \param qsd - is qsd_instance to prepare
 752  *
 753  * \retval - 0 on success, appropriate error on failure
 754  */
 755 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 756 {
 757         struct qsd_thread_info  *qti = qsd_info(env);
 758         int                      qtype, rc = 0;
 759         ENTRY;
 760
 761         if (unlikely(qsd == NULL))
 762                 RETURN(0);
 763
 764         read_lock(&qsd->qsd_lock);
 765         if (qsd->qsd_prepared) {
 766                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 767                 rc = -EALREADY;
 768         }
 769         read_unlock(&qsd->qsd_lock);
 770         if (rc)
 771                 RETURN(rc);
 772
 773         /* Record whether this qsd instance is managing quota enforcement for a
 774          * MDT (i.e. inode quota) or OST (block quota) */
 775         if (qsd->qsd_is_md)
 776                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 777         else
 778                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 779
 780         /* look-up on-disk directory for the quota slave */
 781         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 782                                                     QSD_DIR);
 783         if (IS_ERR(qsd->qsd_root)) {
 784                 rc = PTR_ERR(qsd->qsd_root);
 785                 qsd->qsd_root = NULL;
 786                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 787                        qsd->qsd_svname, rc);
 788                 RETURN(rc);
 789         }
 790
 791         /* initialize per-quota type data */
 792         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 793                 rc = qsd_qtype_init(env, qsd, qtype);
 794                 if (rc)
 795                         RETURN(rc);
 796         }
 797
 798         /* pools successfully setup, mark the qsd as prepared */
 799         write_lock(&qsd->qsd_lock);
 800         qsd->qsd_prepared = true;
 801         write_unlock(&qsd->qsd_lock);
 802
 803         if (qsd->qsd_dev->dd_rdonly)
 804                 RETURN(0);
 805
 806         /* start reintegration thread for each type, if required */
 807         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 808                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 809
 810                 if (qsd_type_enabled(qsd, qtype) &&
 811                     qqi->qqi_acct_failed) {
 812                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 813                                        "since space accounting isn't functional"
 814                                        ". Please run tunefs.lustre --quota on "
 815                                        "an unmounted filesystem if not done "
 816                                        "already\n", qsd->qsd_svname);
 817                         continue;
 818                 }
 819
 820                 rc = qsd_start_reint_thread(qqi);
 821                 if (rc) {
 822                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 823                                 qsd->qsd_svname, qtype_name(qtype), rc);
 824                         RETURN(rc);
 825                 }
 826         }
 827
 828         /* start writeback thread */
 829         rc = qsd_start_upd_thread(qsd);
 830         if (rc) {
 831                 CERROR("%s: failed to start writeback thread (%d)\n",
 832                        qsd->qsd_svname, rc);
 833                 RETURN(rc);
 834         }
 835
 836         /* generate osp name */
 837         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 838                                MTI_NAME_MAXLEN, 0);
 839         if (rc) {
 840                 CERROR("%s: failed to generate ospname (%d)\n",
 841                        qsd->qsd_svname, rc);
 842                 RETURN(rc);
 843         }
 844
 845         /* the connection callback will start the reintegration
 846          * procedure if quota is enabled */
 847         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 848                                       qsd_conn_callback, (void *)qsd);
 849         if (rc) {
 850                 CERROR("%s: fail to get connection to master (%d)\n",
 851                        qsd->qsd_svname, rc);
 852                 RETURN(rc);
 853         }
 854
 855         RETURN(0);
 856 }
 857 EXPORT_SYMBOL(qsd_prepare);
 858
 859 /*
 860  * Start a qsd instance. This will complete the last step of the reintegration
 861  * procedure as soon as possible (provided that the master is reachable).
 862  * This should be called when recovery has been completed and quota should now
 863  * be enforced on every operations.
 864  *
 865  * \param env - the environment passed by the caller
 866  * \param qsd - is the qsd instance associated with the osd device to start
 867  */
 868 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 869 {
 870         int     type, rc = 0;
 871         ENTRY;
 872
 873         if (unlikely(qsd == NULL))
 874                 RETURN(0);
 875
 876         write_lock(&qsd->qsd_lock);
 877         if (!qsd->qsd_prepared) {
 878                 CERROR("%s: can't start qsd instance since it wasn't properly "
 879                        "initialized\n", qsd->qsd_svname);
 880                 rc = -EFAULT;
 881         } else if (qsd->qsd_started) {
 882                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 883                 rc = -EALREADY;
 884         } else {
 885                 /* notify that the qsd_instance is now started */
 886                 qsd->qsd_started = true;
 887         }
 888         write_unlock(&qsd->qsd_lock);
 889
 890         if (rc)
 891                 RETURN(rc);
 892
 893         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 894          * up to usage; If usage < granted, release down to usage.  */
 895         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 896                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 897                 struct task_struct *t;
 898
 899                 /* qqi_reint_task can be set to NULL at any time,
 900                  * so we need to be careful.
 901                  */
 902                 rcu_read_lock();
 903                 t = rcu_dereference(qqi->qqi_reint_task);
 904                 if (t)
 905                         wake_up_process(t);
 906                 rcu_read_unlock();
 907         }
 908
 909         RETURN(rc);
 910 }
 911 EXPORT_SYMBOL(qsd_start);
 912
 913 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 914
 915 /*
 916  * Global initialization performed at module load time
 917  */
 918 int qsd_glb_init(void)
 919 {
 920         int     rc;
 921
 922         rc = lu_kmem_init(qsd_caches);
 923         if (rc)
 924                 return rc;
 925
 926         qsd_key_init_generic(&qsd_thread_key, NULL);
 927         lu_context_key_register(&qsd_thread_key);
 928         lustre_register_quota_process_config(qsd_process_config);
 929
 930         return 0;
 931 }
 932
 933 /*
 934  * Companion of qsd_glb_init() called at module unload time
 935  */
 936 void qsd_glb_fini(void)
 937 {
 938         lustre_register_quota_process_config(NULL);
 939         lu_kmem_fini(qsd_caches);
 940         lu_context_key_degister(&qsd_thread_key);
 941 }