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                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 316         }
 317
 318         RETURN(0);
 319 }
 320
 321 /*
 322  * Release qsd_qtype_info structure which contains data associated with a
 323  * given quota type. This releases the accounting objects.
 324  * It's called on OSD cleanup when the qsd instance is released.
 325  *
 326  * \param env - is the environment passed by the caller
 327  * \param qsd - is the qsd instance managing the qsd_qtype_info structure
 328  *              to be released
 329  * \param qtype - is the quota type to be shutdown
 330  */
 331 static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
 332                            int qtype)
 333 {
 334         struct qsd_qtype_info   *qqi;
 335         int repeat = 0;
 336         ENTRY;
 337
 338         if (qsd->qsd_type_array[qtype] == NULL)
 339                 RETURN_EXIT;
 340         qqi = qsd->qsd_type_array[qtype];
 341         qsd->qsd_type_array[qtype] = NULL;
 342
 343         /* all deferred work lists should be empty */
 344         LASSERT(list_empty(&qqi->qqi_deferred_glb));
 345         LASSERT(list_empty(&qqi->qqi_deferred_slv));
 346
 347         /* shutdown lquota site */
 348         if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
 349                 lquota_site_free(env, qqi->qqi_site);
 350                 qqi->qqi_site = NULL;
 351         }
 352
 353         /* The qqi may still be holding by global locks which are being
 354          * canceled asynchronously (LU-4365), see the following steps:
 355          *
 356          * - On server umount, we try to clear all quota locks first by
 357          *   disconnecting LWP (which will invalidate import and cleanup
 358          *   all locks on it), however, if quota reint process is holding
 359          *   the global lock for reintegration at that time, global lock
 360          *   will fail to be cleared on LWP disconnection.
 361          *
 362          * - Umount process goes on and stops reint process, the global
 363          *   lock will be dropped on reint process exit, however, the lock
 364          *   cancel in done in asynchronous way, so the
 365          *   qsd_glb_blocking_ast() might haven't been called yet when we
 366          *   get here.
 367          */
 368         while (atomic_read(&qqi->qqi_ref) > 1) {
 369                 CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
 370                        atomic_read(&qqi->qqi_ref), repeat);
 371                 repeat++;
 372                 set_current_state(TASK_INTERRUPTIBLE);
 373                 schedule_timeout(cfs_time_seconds(1));
 374         }
 375
 376         /* by now, all qqi users should have gone away */
 377         LASSERT(atomic_read(&qqi->qqi_ref) == 1);
 378         lu_ref_fini(&qqi->qqi_reference);
 379
 380         /* release accounting object */
 381         if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
 382                 dt_object_put(env, qqi->qqi_acct_obj);
 383                 qqi->qqi_acct_obj = NULL;
 384         }
 385
 386         /* release slv index */
 387         if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
 388                 dt_object_put(env, qqi->qqi_slv_obj);
 389                 qqi->qqi_slv_obj = NULL;
 390                 qqi->qqi_slv_ver = 0;
 391         }
 392
 393         /* release global index */
 394         if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
 395                 dt_object_put(env, qqi->qqi_glb_obj);
 396                 qqi->qqi_glb_obj = NULL;
 397                 qqi->qqi_glb_ver = 0;
 398         }
 399
 400         OBD_FREE_PTR(qqi);
 401         EXIT;
 402 }
 403
 404 static const char *qtype2acct_name(int qtype)
 405 {
 406         static char unknown[24];
 407
 408         switch (qtype) {
 409         case USRQUOTA:
 410                 return "acct_user";
 411         case GRPQUOTA:
 412                 return "acct_group";
 413         case PRJQUOTA:
 414                 return "acct_project";
 415         }
 416
 417         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 418         return unknown;
 419 }
 420
 421 static const char *qtype2glb_name(int qtype)
 422 {
 423         static char unknown[24];
 424
 425         switch (qtype) {
 426         case USRQUOTA:
 427                 return "limit_user";
 428         case GRPQUOTA:
 429                 return "limit_group";
 430         case PRJQUOTA:
 431                 return "limit_project";
 432         }
 433
 434         snprintf(unknown, sizeof(unknown), "acct_unknown_%u", qtype);
 435         return unknown;
 436 }
 437
 438 /*
 439  * Allocate and initialize a qsd_qtype_info structure for quota type \qtype.
 440  * This opens the accounting object and initializes the proc file.
 441  * It's called on OSD start when the qsd_prepare() is invoked on the qsd
 442  * instance.
 443  *
 444  * \param env  - the environment passed by the caller
 445  * \param qsd  - is the qsd instance which will be in charge of the new
 446  *               qsd_qtype_info instance.
 447  * \param qtype - is quota type to set up
 448  *
 449  * \retval - 0 on success and qsd->qsd_type_array[qtype] is allocated,
 450  *           appropriate error on failure
 451  */
 452 static int qsd_qtype_init(const struct lu_env *env, struct qsd_instance *qsd,
 453                           int qtype)
 454 {
 455         struct qsd_qtype_info   *qqi;
 456         int                      rc;
 457         struct obd_uuid          uuid;
 458         ENTRY;
 459
 460         LASSERT(qsd->qsd_type_array[qtype] == NULL);
 461
 462         /* allocate structure for this quota type */
 463         OBD_ALLOC_PTR(qqi);
 464         if (qqi == NULL)
 465                 RETURN(-ENOMEM);
 466         qsd->qsd_type_array[qtype] = qqi;
 467         atomic_set(&qqi->qqi_ref, 1); /* referenced from qsd */
 468
 469         /* set backpointer and other parameters */
 470         qqi->qqi_qsd   = qsd;
 471         qqi->qqi_qtype = qtype;
 472         lu_ref_init(&qqi->qqi_reference);
 473         qqi->qqi_glb_uptodate = false;
 474         qqi->qqi_slv_uptodate = false;
 475         qqi->qqi_reint        = false;
 476         init_waitqueue_head(&qqi->qqi_reint_thread.t_ctl_waitq);
 477         thread_set_flags(&qqi->qqi_reint_thread, SVC_STOPPED);
 478         INIT_LIST_HEAD(&qqi->qqi_deferred_glb);
 479         INIT_LIST_HEAD(&qqi->qqi_deferred_slv);
 480         lquota_generate_fid(&qqi->qqi_fid, QSD_RES_TYPE(qsd), qtype);
 481
 482         /* open accounting object */
 483         LASSERT(qqi->qqi_acct_obj == NULL);
 484         qqi->qqi_acct_obj = acct_obj_lookup(env, qsd->qsd_dev, qtype);
 485         if (IS_ERR(qqi->qqi_acct_obj)) {
 486                 CDEBUG(D_QUOTA, "%s: no %s space accounting support: rc = %ld\n",
 487                        qsd->qsd_svname, qtype_name(qtype),
 488                        PTR_ERR(qqi->qqi_acct_obj));
 489                 qqi->qqi_acct_obj = NULL;
 490                 qqi->qqi_acct_failed = true;
 491         }
 492
 493         /* open global index copy */
 494         LASSERT(qqi->qqi_glb_obj == NULL);
 495         qqi->qqi_glb_obj = lquota_disk_glb_find_create(env, qsd->qsd_dev,
 496                                                        qsd->qsd_root,
 497                                                        &qqi->qqi_fid, true);
 498         if (IS_ERR(qqi->qqi_glb_obj)) {
 499                 CERROR("%s: can't open global index copy "DFID" %ld\n",
 500                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 501                        PTR_ERR(qqi->qqi_glb_obj));
 502                 GOTO(out, rc = PTR_ERR(qqi->qqi_glb_obj));
 503         }
 504         qqi->qqi_glb_ver = dt_version_get(env, qqi->qqi_glb_obj);
 505
 506         /* open slave index copy */
 507         LASSERT(qqi->qqi_slv_obj == NULL);
 508         obd_str2uuid(&uuid, qsd->qsd_svname);
 509         qqi->qqi_slv_obj = lquota_disk_slv_find_create(env, qsd->qsd_dev,
 510                                                        qsd->qsd_root,
 511                                                        &qqi->qqi_fid, &uuid,
 512                                                        true);
 513         if (IS_ERR(qqi->qqi_slv_obj)) {
 514                 CERROR("%s: can't open slave index copy "DFID" %ld\n",
 515                        qsd->qsd_svname, PFID(&qqi->qqi_fid),
 516                        PTR_ERR(qqi->qqi_slv_obj));
 517                 GOTO(out, rc = PTR_ERR(qqi->qqi_slv_obj));
 518         }
 519         qqi->qqi_slv_ver = dt_version_get(env, qqi->qqi_slv_obj);
 520
 521         /* allocate site */
 522         qqi->qqi_site = lquota_site_alloc(env, qqi, false, qtype, &qsd_lqe_ops);
 523         if (IS_ERR(qqi->qqi_site)) {
 524                 CERROR("%s: can't allocate site "DFID" %ld\n", qsd->qsd_svname,
 525                        PFID(&qqi->qqi_fid), PTR_ERR(qqi->qqi_site));
 526                 GOTO(out, rc = PTR_ERR(qqi->qqi_site));
 527         }
 528
 529         /* register proc entry for accounting & global index copy objects */
 530         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2acct_name(qtype),
 531                                 0444, &lprocfs_quota_seq_fops,
 532                                 qqi->qqi_acct_obj);
 533         if (rc) {
 534                 CERROR("%s: can't add procfs entry for accounting file %d\n",
 535                        qsd->qsd_svname, rc);
 536                 GOTO(out, rc);
 537         }
 538
 539         rc = lprocfs_seq_create(qsd->qsd_proc, qtype2glb_name(qtype),
 540                                 0444, &lprocfs_quota_seq_fops,
 541                                 qqi->qqi_glb_obj);
 542         if (rc) {
 543                 CERROR("%s: can't add procfs entry for global index copy %d\n",
 544                        qsd->qsd_svname, rc);
 545                 GOTO(out, rc);
 546         }
 547         EXIT;
 548 out:
 549         if (rc)
 550                 qsd_qtype_fini(env, qsd, qtype);
 551         return rc;
 552 }
 553
 554 /*
 555  * Release a qsd_instance. Companion of qsd_init(). This releases all data
 556  * structures associated with the quota slave (on-disk objects, lquota entry
 557  * tables, ...).
 558  * This function should be called when the OSD is shutting down.
 559  *
 560  * \param env - is the environment passed by the caller
 561  * \param qsd - is the qsd instance to shutdown
 562  */
 563 void qsd_fini(const struct lu_env *env, struct qsd_instance *qsd)
 564 {
 565         int     qtype;
 566         ENTRY;
 567
 568         if (unlikely(qsd == NULL))
 569                 RETURN_EXIT;
 570
 571         CDEBUG(D_QUOTA, "%s: initiating QSD shutdown\n", qsd->qsd_svname);
 572         write_lock(&qsd->qsd_lock);
 573         qsd->qsd_stopping = true;
 574         write_unlock(&qsd->qsd_lock);
 575
 576         /* remove qsd proc entry */
 577         if (qsd->qsd_proc != NULL) {
 578                 lprocfs_remove(&qsd->qsd_proc);
 579                 qsd->qsd_proc = NULL;
 580         }
 581
 582         /* stop the writeback thread */
 583         qsd_stop_upd_thread(qsd);
 584
 585         /* shutdown the reintegration threads */
 586         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 587                 if (qsd->qsd_type_array[qtype] == NULL)
 588                         continue;
 589                 qsd_stop_reint_thread(qsd->qsd_type_array[qtype]);
 590         }
 591
 592         if (qsd->qsd_ns != NULL) {
 593                 qsd->qsd_ns = NULL;
 594         }
 595
 596         /* release per-filesystem information */
 597         if (qsd->qsd_fsinfo != NULL) {
 598                 mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 599                 /* remove from the list of fsinfo */
 600                 list_del_init(&qsd->qsd_link);
 601                 mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 602                 qsd_put_fsinfo(qsd->qsd_fsinfo);
 603                 qsd->qsd_fsinfo = NULL;
 604         }
 605
 606         /* free per-quota type data */
 607         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++)
 608                 qsd_qtype_fini(env, qsd, qtype);
 609
 610         /* deregister connection to the quota master */
 611         qsd->qsd_exp_valid = false;
 612         lustre_deregister_lwp_item(&qsd->qsd_exp);
 613
 614         /* release quota root directory */
 615         if (qsd->qsd_root != NULL) {
 616                 dt_object_put(env, qsd->qsd_root);
 617                 qsd->qsd_root = NULL;
 618         }
 619
 620         /* release reference on dt_device */
 621         if (qsd->qsd_dev != NULL) {
 622                 lu_ref_del(&qsd->qsd_dev->dd_lu_dev.ld_reference, "qsd", qsd);
 623                 lu_device_put(&qsd->qsd_dev->dd_lu_dev);
 624                 qsd->qsd_dev = NULL;
 625         }
 626
 627         CDEBUG(D_QUOTA, "%s: QSD shutdown completed\n", qsd->qsd_svname);
 628         OBD_FREE_PTR(qsd);
 629         EXIT;
 630 }
 631 EXPORT_SYMBOL(qsd_fini);
 632
 633 /*
 634  * Create a new qsd_instance to be associated with backend osd device
 635  * identified by \dev.
 636  *
 637  * \param env    - the environment passed by the caller
 638  * \param svname - is the service name of the OSD device creating this instance
 639  * \param dev    - is the dt_device where to store quota index files
 640  * \param osd_proc - is the procfs parent directory where to create procfs file
 641  *                   related to this new qsd instance
 642  *
 643  * \retval - pointer to new qsd_instance associated with dev \dev on success,
 644  *           appropriate error on failure
 645  */
 646 struct qsd_instance *qsd_init(const struct lu_env *env, char *svname,
 647                               struct dt_device *dev,
 648                               struct proc_dir_entry *osd_proc, bool is_md)
 649 {
 650         struct qsd_thread_info  *qti = qsd_info(env);
 651         struct qsd_instance     *qsd;
 652         int                      rc, type, idx;
 653         ENTRY;
 654
 655         /* only configure qsd for MDT & OST */
 656         type = server_name2index(svname, &idx, NULL);
 657         if (type != LDD_F_SV_TYPE_MDT && type != LDD_F_SV_TYPE_OST)
 658                 RETURN(NULL);
 659
 660         /* allocate qsd instance */
 661         OBD_ALLOC_PTR(qsd);
 662         if (qsd == NULL)
 663                 RETURN(ERR_PTR(-ENOMEM));
 664
 665         /* generic initializations */
 666         rwlock_init(&qsd->qsd_lock);
 667         INIT_LIST_HEAD(&qsd->qsd_link);
 668         thread_set_flags(&qsd->qsd_upd_thread, SVC_STOPPED);
 669         init_waitqueue_head(&qsd->qsd_upd_thread.t_ctl_waitq);
 670         INIT_LIST_HEAD(&qsd->qsd_upd_list);
 671         spin_lock_init(&qsd->qsd_adjust_lock);
 672         INIT_LIST_HEAD(&qsd->qsd_adjust_list);
 673         qsd->qsd_prepared = false;
 674         qsd->qsd_started = false;
 675         qsd->qsd_is_md = is_md;
 676
 677         /* copy service name */
 678         if (strlcpy(qsd->qsd_svname, svname, sizeof(qsd->qsd_svname))
 679             >= sizeof(qsd->qsd_svname))
 680                 GOTO(out, rc = -E2BIG);
 681
 682         /* grab reference on osd device */
 683         lu_device_get(&dev->dd_lu_dev);
 684         lu_ref_add(&dev->dd_lu_dev.ld_reference, "qsd", qsd);
 685         qsd->qsd_dev = dev;
 686
 687         /* get fsname from svname */
 688         rc = server_name2fsname(svname, qti->qti_buf, NULL);
 689         if (rc) {
 690                 CERROR("%s: fail to extract filesystem name\n", svname);
 691                 GOTO(out, rc);
 692         }
 693
 694         /* look up quota setting for the filesystem the target belongs to */
 695         qsd->qsd_fsinfo = qsd_get_fsinfo(qti->qti_buf, 1);
 696         if (qsd->qsd_fsinfo == NULL) {
 697                 CERROR("%s: failed to locate filesystem information\n", svname);
 698                 GOTO(out, rc = -EINVAL);
 699         }
 700
 701         /* add in the list of lquota_fsinfo */
 702         mutex_lock(&qsd->qsd_fsinfo->qfs_mutex);
 703         list_add_tail(&qsd->qsd_link, &qsd->qsd_fsinfo->qfs_qsd_list);
 704         mutex_unlock(&qsd->qsd_fsinfo->qfs_mutex);
 705
 706         /* register procfs directory */
 707         if (qsd->qsd_is_md)
 708                 qsd->qsd_proc = lprocfs_register(QSD_DIR_MD, osd_proc,
 709                                                  lprocfs_quota_qsd_vars, qsd);
 710         else
 711                 qsd->qsd_proc = lprocfs_register(QSD_DIR_DT, osd_proc,
 712                                                  lprocfs_quota_qsd_vars, qsd);
 713
 714         if (type == LDD_F_SV_TYPE_MDT && qsd->qsd_is_md)
 715                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_MD);
 716         else if (type == LDD_F_SV_TYPE_OST && !qsd->qsd_is_md)
 717                 lprocfs_add_symlink(QSD_DIR, osd_proc, "./%s", QSD_DIR_DT);
 718
 719         if (IS_ERR(qsd->qsd_proc)) {
 720                 rc = PTR_ERR(qsd->qsd_proc);
 721                 qsd->qsd_proc = NULL;
 722                 CERROR("%s: fail to create quota slave proc entry (%d)\n",
 723                        svname, rc);
 724                 GOTO(out, rc);
 725         }
 726         EXIT;
 727 out:
 728         if (rc) {
 729                 qsd_fini(env, qsd);
 730                 return ERR_PTR(rc);
 731         }
 732         RETURN(qsd);
 733 }
 734 EXPORT_SYMBOL(qsd_init);
 735
 736 /*
 737  * Initialize on-disk structures in order to manage quota enforcement for
 738  * the target associated with the qsd instance \qsd and starts the reintegration
 739  * procedure for each quota type as soon as possible.
 740  * The last step of the reintegration will be completed once qsd_start() is
 741  * called, at which points the space reconciliation with the master will be
 742  * executed.
 743  * This function must be called when the server stack is fully configured,
 744  * typically when ->ldo_prepare is called across the stack.
 745  *
 746  * \param env - the environment passed by the caller
 747  * \param qsd - is qsd_instance to prepare
 748  *
 749  * \retval - 0 on success, appropriate error on failure
 750  */
 751 int qsd_prepare(const struct lu_env *env, struct qsd_instance *qsd)
 752 {
 753         struct qsd_thread_info  *qti = qsd_info(env);
 754         int                      qtype, rc = 0;
 755         ENTRY;
 756
 757         if (unlikely(qsd == NULL))
 758                 RETURN(0);
 759
 760         read_lock(&qsd->qsd_lock);
 761         if (qsd->qsd_prepared) {
 762                 CERROR("%s: qsd instance already prepared\n", qsd->qsd_svname);
 763                 rc = -EALREADY;
 764         }
 765         read_unlock(&qsd->qsd_lock);
 766         if (rc)
 767                 RETURN(rc);
 768
 769         /* Record whether this qsd instance is managing quota enforcement for a
 770          * MDT (i.e. inode quota) or OST (block quota) */
 771         if (qsd->qsd_is_md)
 772                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_MD);
 773         else
 774                 qsd->qsd_sync_threshold = LQUOTA_LEAST_QUNIT(LQUOTA_RES_DT);
 775
 776         /* look-up on-disk directory for the quota slave */
 777         qsd->qsd_root = lquota_disk_dir_find_create(env, qsd->qsd_dev, NULL,
 778                                                     QSD_DIR);
 779         if (IS_ERR(qsd->qsd_root)) {
 780                 rc = PTR_ERR(qsd->qsd_root);
 781                 qsd->qsd_root = NULL;
 782                 CERROR("%s: failed to create quota slave root dir (%d)\n",
 783                        qsd->qsd_svname, rc);
 784                 RETURN(rc);
 785         }
 786
 787         /* initialize per-quota type data */
 788         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 789                 rc = qsd_qtype_init(env, qsd, qtype);
 790                 if (rc)
 791                         RETURN(rc);
 792         }
 793
 794         /* pools successfully setup, mark the qsd as prepared */
 795         write_lock(&qsd->qsd_lock);
 796         qsd->qsd_prepared = true;
 797         write_unlock(&qsd->qsd_lock);
 798
 799         if (qsd->qsd_dev->dd_rdonly)
 800                 RETURN(0);
 801
 802         /* start reintegration thread for each type, if required */
 803         for (qtype = USRQUOTA; qtype < LL_MAXQUOTAS; qtype++) {
 804                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[qtype];
 805
 806                 if (qsd_type_enabled(qsd, qtype) &&
 807                     qqi->qqi_acct_failed) {
 808                         LCONSOLE_ERROR("%s: can't enable quota enforcement "
 809                                        "since space accounting isn't functional"
 810                                        ". Please run tunefs.lustre --quota on "
 811                                        "an unmounted filesystem if not done "
 812                                        "already\n", qsd->qsd_svname);
 813                         continue;
 814                 }
 815
 816                 rc = qsd_start_reint_thread(qqi);
 817                 if (rc) {
 818                         CERROR("%s: failed to start reint thread for type %s: rc = %d\n",
 819                                 qsd->qsd_svname, qtype_name(qtype), rc);
 820                         RETURN(rc);
 821                 }
 822         }
 823
 824         /* start writeback thread */
 825         rc = qsd_start_upd_thread(qsd);
 826         if (rc) {
 827                 CERROR("%s: failed to start writeback thread (%d)\n",
 828                        qsd->qsd_svname, rc);
 829                 RETURN(rc);
 830         }
 831
 832         /* generate osp name */
 833         rc = tgt_name2lwp_name(qsd->qsd_svname, qti->qti_buf,
 834                                MTI_NAME_MAXLEN, 0);
 835         if (rc) {
 836                 CERROR("%s: failed to generate ospname (%d)\n",
 837                        qsd->qsd_svname, rc);
 838                 RETURN(rc);
 839         }
 840
 841         /* the connection callback will start the reintegration
 842          * procedure if quota is enabled */
 843         rc = lustre_register_lwp_item(qti->qti_buf, &qsd->qsd_exp,
 844                                       qsd_conn_callback, (void *)qsd);
 845         if (rc) {
 846                 CERROR("%s: fail to get connection to master (%d)\n",
 847                        qsd->qsd_svname, rc);
 848                 RETURN(rc);
 849         }
 850
 851         RETURN(0);
 852 }
 853 EXPORT_SYMBOL(qsd_prepare);
 854
 855 /*
 856  * Start a qsd instance. This will complete the last step of the reintegration
 857  * procedure as soon as possible (provided that the master is reachable).
 858  * This should be called when recovery has been completed and quota should now
 859  * be enforced on every operations.
 860  *
 861  * \param env - the environment passed by the caller
 862  * \param qsd - is the qsd instance associated with the osd device to start
 863  */
 864 int qsd_start(const struct lu_env *env, struct qsd_instance *qsd)
 865 {
 866         int     type, rc = 0;
 867         ENTRY;
 868
 869         if (unlikely(qsd == NULL))
 870                 RETURN(0);
 871
 872         write_lock(&qsd->qsd_lock);
 873         if (!qsd->qsd_prepared) {
 874                 CERROR("%s: can't start qsd instance since it wasn't properly "
 875                        "initialized\n", qsd->qsd_svname);
 876                 rc = -EFAULT;
 877         } else if (qsd->qsd_started) {
 878                 CERROR("%s: qsd instance already started\n", qsd->qsd_svname);
 879                 rc = -EALREADY;
 880         } else {
 881                 /* notify that the qsd_instance is now started */
 882                 qsd->qsd_started = true;
 883         }
 884         write_unlock(&qsd->qsd_lock);
 885
 886         if (rc)
 887                 RETURN(rc);
 888
 889         /* Trigger the 3rd step of reintegration: If usage > granted, acquire
 890          * up to usage; If usage < granted, release down to usage.  */
 891         for (type = USRQUOTA; type < LL_MAXQUOTAS; type++) {
 892                 struct qsd_qtype_info   *qqi = qsd->qsd_type_array[type];
 893                 wake_up(&qqi->qqi_reint_thread.t_ctl_waitq);
 894         }
 895
 896         RETURN(rc);
 897 }
 898 EXPORT_SYMBOL(qsd_start);
 899
 900 void lustre_register_quota_process_config(int (*qpc)(struct lustre_cfg *lcfg));
 901
 902 /*
 903  * Global initialization performed at module load time
 904  */
 905 int qsd_glb_init(void)
 906 {
 907         int     rc;
 908
 909         rc = lu_kmem_init(qsd_caches);
 910         if (rc)
 911                 return rc;
 912
 913         qsd_key_init_generic(&qsd_thread_key, NULL);
 914         lu_context_key_register(&qsd_thread_key);
 915         lustre_register_quota_process_config(qsd_process_config);
 916
 917         return 0;
 918 }
 919
 920 /*
 921  * Companion of qsd_glb_init() called at module unload time
 922  */
 923 void qsd_glb_fini(void)
 924 {
 925         lustre_register_quota_process_config(NULL);
 926         lu_kmem_fini(qsd_caches);
 927         lu_context_key_degister(&qsd_thread_key);
 928 }