lustre/lov/lovsub_lock.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, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * Implementation of cl_lock for LOVSUB layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LOV
  42
  43 #include "lov_cl_internal.h"
  44
  45 /** \addtogroup lov
  46  *  @{
  47  */
  48
  49 /*****************************************************************************
  50  *
  51  * Lovsub lock operations.
  52  *
  53  */
  54
  55 static void lovsub_lock_fini(const struct lu_env *env,
  56                              struct cl_lock_slice *slice)
  57 {
  58         struct lovsub_lock   *lsl;
  59
  60         ENTRY;
  61         lsl = cl2lovsub_lock(slice);
  62         LASSERT(list_empty(&lsl->lss_parents));
  63         OBD_SLAB_FREE_PTR(lsl, lovsub_lock_kmem);
  64         EXIT;
  65 }
  66
  67 static void lovsub_parent_lock(const struct lu_env *env, struct lov_lock *lov)
  68 {
  69         struct cl_lock *parent;
  70
  71         ENTRY;
  72         parent = lov->lls_cl.cls_lock;
  73         cl_lock_get(parent);
  74         lu_ref_add(&parent->cll_reference, "lovsub-parent", current);
  75         cl_lock_mutex_get(env, parent);
  76         EXIT;
  77 }
  78
  79 static void lovsub_parent_unlock(const struct lu_env *env, struct lov_lock *lov)
  80 {
  81         struct cl_lock *parent;
  82
  83         ENTRY;
  84         parent = lov->lls_cl.cls_lock;
  85         cl_lock_mutex_put(env, lov->lls_cl.cls_lock);
  86         lu_ref_del(&parent->cll_reference, "lovsub-parent", current);
  87         cl_lock_put(env, parent);
  88         EXIT;
  89 }
  90
  91 /**
  92  * Implements cl_lock_operations::clo_state() method for lovsub layer, which
  93  * method is called whenever sub-lock state changes. Propagates state change
  94  * to the top-locks.
  95  */
  96 static void lovsub_lock_state(const struct lu_env *env,
  97                               const struct cl_lock_slice *slice,
  98                               enum cl_lock_state state)
  99 {
 100         struct lovsub_lock   *sub = cl2lovsub_lock(slice);
 101         struct lov_lock_link *scan;
 102
 103         LASSERT(cl_lock_is_mutexed(slice->cls_lock));
 104         ENTRY;
 105
 106         list_for_each_entry(scan, &sub->lss_parents, lll_list) {
 107                 struct lov_lock *lov    = scan->lll_super;
 108                 struct cl_lock  *parent = lov->lls_cl.cls_lock;
 109
 110                 if (sub->lss_active != parent) {
 111                         lovsub_parent_lock(env, lov);
 112                         cl_lock_signal(env, parent);
 113                         lovsub_parent_unlock(env, lov);
 114                 }
 115         }
 116         EXIT;
 117 }
 118
 119 /**
 120  * Implementation of cl_lock_operation::clo_weigh() estimating lock weight by
 121  * asking parent lock.
 122  */
 123 static unsigned long lovsub_lock_weigh(const struct lu_env *env,
 124                                        const struct cl_lock_slice *slice)
 125 {
 126         struct lovsub_lock *lock = cl2lovsub_lock(slice);
 127         struct lov_lock    *lov;
 128         unsigned long       dumbbell;
 129
 130         ENTRY;
 131
 132         LASSERT(cl_lock_is_mutexed(slice->cls_lock));
 133
 134         if (!list_empty(&lock->lss_parents)) {
 135                 /*
 136                  * It is not clear whether all parents have to be asked and
 137                  * their estimations summed, or it is enough to ask one. For
 138                  * the current usages, one is always enough.
 139                  */
 140                 lov = container_of(lock->lss_parents.next,
 141                                    struct lov_lock_link, lll_list)->lll_super;
 142
 143                 lovsub_parent_lock(env, lov);
 144                 dumbbell = cl_lock_weigh(env, lov->lls_cl.cls_lock);
 145                 lovsub_parent_unlock(env, lov);
 146         } else
 147                 dumbbell = 0;
 148
 149         RETURN(dumbbell);
 150 }
 151
 152 /**
 153  * Maps start/end offsets within a stripe, to offsets within a file.
 154  */
 155 static void lovsub_lock_descr_map(const struct cl_lock_descr *in,
 156                                   struct lov_object *lov,
 157                                   int stripe, struct cl_lock_descr *out)
 158 {
 159         pgoff_t size; /* stripe size in pages */
 160         pgoff_t skip; /* how many pages in every stripe are occupied by
 161                        * "other" stripes */
 162         pgoff_t start;
 163         pgoff_t end;
 164
 165         ENTRY;
 166         start = in->cld_start;
 167         end   = in->cld_end;
 168
 169         if (lov->lo_lsm->lsm_stripe_count > 1) {
 170                 size = cl_index(lov2cl(lov), lov->lo_lsm->lsm_stripe_size);
 171                 skip = (lov->lo_lsm->lsm_stripe_count - 1) * size;
 172
 173                 /* XXX overflow check here? */
 174                 start += start/size * skip + stripe * size;
 175
 176                 if (end != CL_PAGE_EOF) {
 177                         end += end/size * skip + stripe * size;
 178                         /*
 179                          * And check for overflow...
 180                          */
 181                         if (end < in->cld_end)
 182                                 end = CL_PAGE_EOF;
 183                 }
 184         }
 185         out->cld_start = start;
 186         out->cld_end   = end;
 187         EXIT;
 188 }
 189
 190 /**
 191  * Adjusts parent lock extent when a sub-lock is attached to a parent. This is
 192  * called in two ways:
 193  *
 194  *     - as part of receive call-back, when server returns granted extent to
 195  *       the client, and
 196  *
 197  *     - when top-lock finds existing sub-lock in the cache.
 198  *
 199  * Note, that lock mode is not propagated to the parent: i.e., if CLM_READ
 200  * top-lock matches CLM_WRITE sub-lock, top-lock is still CLM_READ.
 201  */
 202 int lov_sublock_modify(const struct lu_env *env, struct lov_lock *lov,
 203                        struct lovsub_lock *sublock,
 204                        const struct cl_lock_descr *d, int idx)
 205 {
 206         struct cl_lock       *parent;
 207         struct lovsub_object *subobj;
 208         struct cl_lock_descr *pd;
 209         struct cl_lock_descr *parent_descr;
 210         int                   result;
 211
 212         parent       = lov->lls_cl.cls_lock;
 213         parent_descr = &parent->cll_descr;
 214         LASSERT(cl_lock_mode_match(d->cld_mode, parent_descr->cld_mode));
 215
 216         subobj = cl2lovsub(sublock->lss_cl.cls_obj);
 217         pd     = &lov_env_info(env)->lti_ldescr;
 218
 219         pd->cld_obj  = parent_descr->cld_obj;
 220         pd->cld_mode = parent_descr->cld_mode;
 221         pd->cld_gid  = parent_descr->cld_gid;
 222         lovsub_lock_descr_map(d, subobj->lso_super, subobj->lso_index, pd);
 223
 224         /* LU-3027: only update extent of lock, plus the change in
 225          * lovsub_lock_delete() that lock extent is modified after a sublock
 226          * is canceled, we can make sure that the lock extent won't be updated
 227          * any more. Therefore, lov_lock_fits_into() will always find feasible
 228          * locks */
 229         lov->lls_sub[idx].sub_got.cld_start = d->cld_start;
 230         lov->lls_sub[idx].sub_got.cld_end = d->cld_end;
 231         /*
 232          * Notify top-lock about modification, if lock description changes
 233          * materially.
 234          */
 235         if (!cl_lock_ext_match(parent_descr, pd))
 236                 result = cl_lock_modify(env, parent, pd);
 237         else
 238                 result = 0;
 239         return result;
 240 }
 241
 242 static int lovsub_lock_modify(const struct lu_env *env,
 243                               const struct cl_lock_slice *s,
 244                               const struct cl_lock_descr *d)
 245 {
 246         struct lovsub_lock   *lock   = cl2lovsub_lock(s);
 247         struct lov_lock_link *scan;
 248         struct lov_lock      *lov;
 249         int result                   = 0;
 250
 251         ENTRY;
 252
 253         LASSERT(cl_lock_mode_match(d->cld_mode,
 254                                    s->cls_lock->cll_descr.cld_mode));
 255         list_for_each_entry(scan, &lock->lss_parents, lll_list) {
 256                 int rc;
 257
 258                 lov = scan->lll_super;
 259                 lovsub_parent_lock(env, lov);
 260                 rc = lov_sublock_modify(env, lov, lock, d, scan->lll_idx);
 261                 lovsub_parent_unlock(env, lov);
 262                 result = result ?: rc;
 263         }
 264         RETURN(result);
 265 }
 266
 267 static int lovsub_lock_closure(const struct lu_env *env,
 268                                const struct cl_lock_slice *slice,
 269                                struct cl_lock_closure *closure)
 270 {
 271         struct lovsub_lock   *sub;
 272         struct cl_lock       *parent;
 273         struct lov_lock_link *scan;
 274         int                   result;
 275
 276         LASSERT(cl_lock_is_mutexed(slice->cls_lock));
 277         ENTRY;
 278
 279         sub    = cl2lovsub_lock(slice);
 280         result = 0;
 281
 282         list_for_each_entry(scan, &sub->lss_parents, lll_list) {
 283                 parent = scan->lll_super->lls_cl.cls_lock;
 284                 result = cl_lock_closure_build(env, parent, closure);
 285                 if (result != 0)
 286                         break;
 287         }
 288         RETURN(result);
 289 }
 290
 291 /**
 292  * A helper function for lovsub_lock_delete() that deals with a given parent
 293  * top-lock.
 294  */
 295 static int lovsub_lock_delete_one(const struct lu_env *env,
 296                                   struct cl_lock *child, struct lov_lock *lov)
 297 {
 298         struct cl_lock *parent;
 299         int             result;
 300         ENTRY;
 301
 302         parent = lov->lls_cl.cls_lock;
 303         if (parent->cll_error)
 304                 RETURN(0);
 305
 306         result = 0;
 307         lov->lls_ever_canceled = 1;
 308         switch (parent->cll_state) {
 309         case CLS_ENQUEUED:
 310                 /* See LU-1355 for the case that a glimpse lock is
 311                  * interrupted by signal */
 312                 LASSERT(parent->cll_flags & CLF_CANCELLED);
 313                 break;
 314         case CLS_QUEUING:
 315         case CLS_FREEING:
 316                 cl_lock_signal(env, parent);
 317                 break;
 318         case CLS_INTRANSIT:
 319                 /*
 320                  * Here lies a problem: a sub-lock is canceled while top-lock
 321                  * is being unlocked. Top-lock cannot be moved into CLS_NEW
 322                  * state, because unlocking has to succeed eventually by
 323                  * placing lock into CLS_CACHED (or failing it), see
 324                  * cl_unuse_try(). Nor can top-lock be left in CLS_CACHED
 325                  * state, because lov maintains an invariant that all
 326                  * sub-locks exist in CLS_CACHED (this allows cached top-lock
 327                  * to be reused immediately). Nor can we wait for top-lock
 328                  * state to change, because this can be synchronous to the
 329                  * current thread.
 330                  *
 331                  * We know for sure that lov_lock_unuse() will be called at
 332                  * least one more time to finish un-using, so leave a mark on
 333                  * the top-lock, that will be seen by the next call to
 334                  * lov_lock_unuse().
 335                  */
 336                 if (cl_lock_is_intransit(parent))
 337                         lov->lls_cancel_race = 1;
 338                 break;
 339         case CLS_CACHED:
 340                 /*
 341                  * if a sub-lock is canceled move its top-lock into CLS_NEW
 342                  * state to preserve an invariant that a top-lock in
 343                  * CLS_CACHED is immediately ready for re-use (i.e., has all
 344                  * sub-locks), and so that next attempt to re-use the top-lock
 345                  * enqueues missing sub-lock.
 346                  */
 347                 cl_lock_state_set(env, parent, CLS_NEW);
 348                 /* fall through */
 349         case CLS_NEW:
 350                 /*
 351                  * if last sub-lock is canceled, destroy the top-lock (which
 352                  * is now `empty') proactively.
 353                  */
 354                 if (lov->lls_nr_filled == 0) {
 355                         /* ... but unfortunately, this cannot be done easily,
 356                          * as cancellation of a top-lock might acquire mutices
 357                          * of its other sub-locks, violating lock ordering,
 358                          * see cl_lock_{cancel,delete}() preconditions.
 359                          *
 360                          * To work around this, the mutex of this sub-lock is
 361                          * released, top-lock is destroyed, and sub-lock mutex
 362                          * acquired again. The list of parents has to be
 363                          * re-scanned from the beginning after this.
 364                          *
 365                          * Only do this if no mutices other than on @child and
 366                          * @parent are held by the current thread.
 367                          *
 368                          * TODO: The lock modal here is too complex, because
 369                          * the lock may be canceled and deleted by voluntarily:
 370                          *    cl_lock_request
 371                          *      -> osc_lock_enqueue_wait
 372                          *        -> osc_lock_cancel_wait
 373                          *          -> cl_lock_delete
 374                          *            -> lovsub_lock_delete
 375                          *              -> cl_lock_cancel/delete
 376                          *                -> ...
 377                          *
 378                          * The better choice is to spawn a kernel thread for
 379                          * this purpose. -jay
 380                          */
 381                         if (cl_lock_nr_mutexed(env) == 2) {
 382                                 cl_lock_mutex_put(env, child);
 383                                 cl_lock_cancel(env, parent);
 384                                 cl_lock_delete(env, parent);
 385                                 result = 1;
 386                         }
 387                 }
 388                 break;
 389         case CLS_HELD:
 390                 CL_LOCK_DEBUG(D_ERROR, env, parent, "Delete CLS_HELD lock\n");
 391                 /* falling through */
 392         default:
 393                 CERROR("Impossible state: %d\n", parent->cll_state);
 394                 LBUG();
 395                 break;
 396         }
 397
 398         RETURN(result);
 399 }
 400
 401 /**
 402  * An implementation of cl_lock_operations::clo_delete() method. This is
 403  * invoked in "bottom-to-top" delete, when lock destruction starts from the
 404  * sub-lock (e.g, as a result of ldlm lock LRU policy).
 405  */
 406 static void lovsub_lock_delete(const struct lu_env *env,
 407                                const struct cl_lock_slice *slice)
 408 {
 409         struct cl_lock     *child = slice->cls_lock;
 410         struct lovsub_lock *sub   = cl2lovsub_lock(slice);
 411         int restart;
 412
 413         LASSERT(cl_lock_is_mutexed(child));
 414
 415         ENTRY;
 416         /*
 417          * Destruction of a sub-lock might take multiple iterations, because
 418          * when the last sub-lock of a given top-lock is deleted, top-lock is
 419          * canceled proactively, and this requires to release sub-lock
 420          * mutex. Once sub-lock mutex has been released, list of its parents
 421          * has to be re-scanned from the beginning.
 422          */
 423         do {
 424                 struct lov_lock      *lov;
 425                 struct lov_lock_link *scan;
 426                 struct lov_lock_link *temp;
 427
 428                 restart = 0;
 429                 list_for_each_entry_safe(scan, temp,
 430                                              &sub->lss_parents, lll_list) {
 431                         lov     = scan->lll_super;
 432                         lovsub_parent_lock(env, lov);
 433                         lov_lock_unlink(env, scan, sub);
 434                         restart = lovsub_lock_delete_one(env, child, lov);
 435                         lovsub_parent_unlock(env, lov);
 436
 437                         if (restart) {
 438                                 cl_lock_mutex_get(env, child);
 439                                 break;
 440                         }
 441                }
 442         } while (restart);
 443         EXIT;
 444 }
 445
 446 static int lovsub_lock_print(const struct lu_env *env, void *cookie,
 447                              lu_printer_t p, const struct cl_lock_slice *slice)
 448 {
 449         struct lovsub_lock   *sub = cl2lovsub_lock(slice);
 450         struct lov_lock      *lov;
 451         struct lov_lock_link *scan;
 452
 453         list_for_each_entry(scan, &sub->lss_parents, lll_list) {
 454                 lov = scan->lll_super;
 455                 (*p)(env, cookie, "[%d %p ", scan->lll_idx, lov);
 456                 if (lov != NULL)
 457                         cl_lock_descr_print(env, cookie, p,
 458                                             &lov->lls_cl.cls_lock->cll_descr);
 459                 (*p)(env, cookie, "] ");
 460         }
 461         return 0;
 462 }
 463
 464 static const struct cl_lock_operations lovsub_lock_ops = {
 465         .clo_fini    = lovsub_lock_fini,
 466         .clo_state   = lovsub_lock_state,
 467         .clo_delete  = lovsub_lock_delete,
 468         .clo_modify  = lovsub_lock_modify,
 469         .clo_closure = lovsub_lock_closure,
 470         .clo_weigh   = lovsub_lock_weigh,
 471         .clo_print   = lovsub_lock_print
 472 };
 473
 474 int lovsub_lock_init(const struct lu_env *env, struct cl_object *obj,
 475                      struct cl_lock *lock, const struct cl_io *io)
 476 {
 477         struct lovsub_lock *lsk;
 478         int result;
 479
 480         ENTRY;
 481         OBD_SLAB_ALLOC_PTR_GFP(lsk, lovsub_lock_kmem, GFP_NOFS);
 482         if (lsk != NULL) {
 483                 INIT_LIST_HEAD(&lsk->lss_parents);
 484                 cl_lock_slice_add(lock, &lsk->lss_cl, obj, &lovsub_lock_ops);
 485                 result = 0;
 486         } else
 487                 result = -ENOMEM;
 488         RETURN(result);
 489 }
 490
 491 /** @} lov */