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