These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / lov / lov_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 LOV 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 static struct cl_lock_closure *lov_closure_get(const struct lu_env *env,
50                                                struct cl_lock *parent);
51
52 static int lov_lock_unuse(const struct lu_env *env,
53                           const struct cl_lock_slice *slice);
54 /*****************************************************************************
55  *
56  * Lov lock operations.
57  *
58  */
59
60 static struct lov_sublock_env *lov_sublock_env_get(const struct lu_env *env,
61                                                    struct cl_lock *parent,
62                                                    struct lov_lock_sub *lls)
63 {
64         struct lov_sublock_env *subenv;
65         struct lov_io     *lio    = lov_env_io(env);
66         struct cl_io       *io     = lio->lis_cl.cis_io;
67         struct lov_io_sub      *sub;
68
69         subenv = &lov_env_session(env)->ls_subenv;
70
71         /*
72          * FIXME: We tend to use the subio's env & io to call the sublock
73          * lock operations because osc lock sometimes stores some control
74          * variables in thread's IO information(Now only lockless information).
75          * However, if the lock's host(object) is different from the object
76          * for current IO, we have no way to get the subenv and subio because
77          * they are not initialized at all. As a temp fix, in this case,
78          * we still borrow the parent's env to call sublock operations.
79          */
80         if (!io || !cl_object_same(io->ci_obj, parent->cll_descr.cld_obj)) {
81                 subenv->lse_env = env;
82                 subenv->lse_io  = io;
83                 subenv->lse_sub = NULL;
84         } else {
85                 sub = lov_sub_get(env, lio, lls->sub_stripe);
86                 if (!IS_ERR(sub)) {
87                         subenv->lse_env = sub->sub_env;
88                         subenv->lse_io  = sub->sub_io;
89                         subenv->lse_sub = sub;
90                 } else {
91                         subenv = (void *)sub;
92                 }
93         }
94         return subenv;
95 }
96
97 static void lov_sublock_env_put(struct lov_sublock_env *subenv)
98 {
99         if (subenv && subenv->lse_sub)
100                 lov_sub_put(subenv->lse_sub);
101 }
102
103 static void lov_sublock_adopt(const struct lu_env *env, struct lov_lock *lck,
104                               struct cl_lock *sublock, int idx,
105                               struct lov_lock_link *link)
106 {
107         struct lovsub_lock *lsl;
108         struct cl_lock     *parent = lck->lls_cl.cls_lock;
109         int              rc;
110
111         LASSERT(cl_lock_is_mutexed(parent));
112         LASSERT(cl_lock_is_mutexed(sublock));
113
114         lsl = cl2sub_lock(sublock);
115         /*
116          * check that sub-lock doesn't have lock link to this top-lock.
117          */
118         LASSERT(lov_lock_link_find(env, lck, lsl) == NULL);
119         LASSERT(idx < lck->lls_nr);
120
121         lck->lls_sub[idx].sub_lock = lsl;
122         lck->lls_nr_filled++;
123         LASSERT(lck->lls_nr_filled <= lck->lls_nr);
124         list_add_tail(&link->lll_list, &lsl->lss_parents);
125         link->lll_idx = idx;
126         link->lll_super = lck;
127         cl_lock_get(parent);
128         lu_ref_add(&parent->cll_reference, "lov-child", sublock);
129         lck->lls_sub[idx].sub_flags |= LSF_HELD;
130         cl_lock_user_add(env, sublock);
131
132         rc = lov_sublock_modify(env, lck, lsl, &sublock->cll_descr, idx);
133         LASSERT(rc == 0); /* there is no way this can fail, currently */
134 }
135
136 static struct cl_lock *lov_sublock_alloc(const struct lu_env *env,
137                                          const struct cl_io *io,
138                                          struct lov_lock *lck,
139                                          int idx, struct lov_lock_link **out)
140 {
141         struct cl_lock       *sublock;
142         struct cl_lock       *parent;
143         struct lov_lock_link *link;
144
145         LASSERT(idx < lck->lls_nr);
146
147         link = kmem_cache_alloc(lov_lock_link_kmem, GFP_NOFS | __GFP_ZERO);
148         if (link != NULL) {
149                 struct lov_sublock_env *subenv;
150                 struct lov_lock_sub  *lls;
151                 struct cl_lock_descr *descr;
152
153                 parent = lck->lls_cl.cls_lock;
154                 lls    = &lck->lls_sub[idx];
155                 descr  = &lls->sub_got;
156
157                 subenv = lov_sublock_env_get(env, parent, lls);
158                 if (!IS_ERR(subenv)) {
159                         /* CAVEAT: Don't try to add a field in lov_lock_sub
160                          * to remember the subio. This is because lock is able
161                          * to be cached, but this is not true for IO. This
162                          * further means a sublock might be referenced in
163                          * different io context. -jay */
164
165                         sublock = cl_lock_hold(subenv->lse_env, subenv->lse_io,
166                                                descr, "lov-parent", parent);
167                         lov_sublock_env_put(subenv);
168                 } else {
169                         /* error occurs. */
170                         sublock = (void *)subenv;
171                 }
172
173                 if (!IS_ERR(sublock))
174                         *out = link;
175                 else
176                         kmem_cache_free(lov_lock_link_kmem, link);
177         } else
178                 sublock = ERR_PTR(-ENOMEM);
179         return sublock;
180 }
181
182 static void lov_sublock_unlock(const struct lu_env *env,
183                                struct lovsub_lock *lsl,
184                                struct cl_lock_closure *closure,
185                                struct lov_sublock_env *subenv)
186 {
187         lov_sublock_env_put(subenv);
188         lsl->lss_active = NULL;
189         cl_lock_disclosure(env, closure);
190 }
191
192 static int lov_sublock_lock(const struct lu_env *env,
193                             struct lov_lock *lck,
194                             struct lov_lock_sub *lls,
195                             struct cl_lock_closure *closure,
196                             struct lov_sublock_env **lsep)
197 {
198         struct lovsub_lock *sublock;
199         struct cl_lock     *child;
200         int              result = 0;
201
202         LASSERT(list_empty(&closure->clc_list));
203
204         sublock = lls->sub_lock;
205         child = sublock->lss_cl.cls_lock;
206         result = cl_lock_closure_build(env, child, closure);
207         if (result == 0) {
208                 struct cl_lock *parent = closure->clc_origin;
209
210                 LASSERT(cl_lock_is_mutexed(child));
211                 sublock->lss_active = parent;
212
213                 if (unlikely((child->cll_state == CLS_FREEING) ||
214                              (child->cll_flags & CLF_CANCELLED))) {
215                         struct lov_lock_link *link;
216                         /*
217                          * we could race with lock deletion which temporarily
218                          * put the lock in freeing state, bug 19080.
219                          */
220                         LASSERT(!(lls->sub_flags & LSF_HELD));
221
222                         link = lov_lock_link_find(env, lck, sublock);
223                         LASSERT(link != NULL);
224                         lov_lock_unlink(env, link, sublock);
225                         lov_sublock_unlock(env, sublock, closure, NULL);
226                         lck->lls_cancel_race = 1;
227                         result = CLO_REPEAT;
228                 } else if (lsep) {
229                         struct lov_sublock_env *subenv;
230
231                         subenv = lov_sublock_env_get(env, parent, lls);
232                         if (IS_ERR(subenv)) {
233                                 lov_sublock_unlock(env, sublock,
234                                                    closure, NULL);
235                                 result = PTR_ERR(subenv);
236                         } else {
237                                 *lsep = subenv;
238                         }
239                 }
240         }
241         return result;
242 }
243
244 /**
245  * Updates the result of a top-lock operation from a result of sub-lock
246  * sub-operations. Top-operations like lov_lock_{enqueue,use,unuse}() iterate
247  * over sub-locks and lov_subresult() is used to calculate return value of a
248  * top-operation. To this end, possible return values of sub-operations are
249  * ordered as
250  *
251  *     - 0                success
252  *     - CLO_WAIT          wait for event
253  *     - CLO_REPEAT      repeat top-operation
254  *     - -ne            fundamental error
255  *
256  * Top-level return code can only go down through this list. CLO_REPEAT
257  * overwrites CLO_WAIT, because lock mutex was released and sleeping condition
258  * has to be rechecked by the upper layer.
259  */
260 static int lov_subresult(int result, int rc)
261 {
262         int result_rank;
263         int rc_rank;
264
265         LASSERTF(result <= 0 || result == CLO_REPEAT || result == CLO_WAIT,
266                  "result = %d", result);
267         LASSERTF(rc <= 0 || rc == CLO_REPEAT || rc == CLO_WAIT,
268                  "rc = %d\n", rc);
269         CLASSERT(CLO_WAIT < CLO_REPEAT);
270
271         /* calculate ranks in the ordering above */
272         result_rank = result < 0 ? 1 + CLO_REPEAT : result;
273         rc_rank = rc < 0 ? 1 + CLO_REPEAT : rc;
274
275         if (result_rank < rc_rank)
276                 result = rc;
277         return result;
278 }
279
280 /**
281  * Creates sub-locks for a given lov_lock for the first time.
282  *
283  * Goes through all sub-objects of top-object, and creates sub-locks on every
284  * sub-object intersecting with top-lock extent. This is complicated by the
285  * fact that top-lock (that is being created) can be accessed concurrently
286  * through already created sub-locks (possibly shared with other top-locks).
287  */
288 static int lov_lock_sub_init(const struct lu_env *env,
289                              struct lov_lock *lck, const struct cl_io *io)
290 {
291         int result = 0;
292         int i;
293         int nr;
294         u64 start;
295         u64 end;
296         u64 file_start;
297         u64 file_end;
298
299         struct lov_object       *loo    = cl2lov(lck->lls_cl.cls_obj);
300         struct lov_layout_raid0 *r0     = lov_r0(loo);
301         struct cl_lock    *parent = lck->lls_cl.cls_lock;
302
303         lck->lls_orig = parent->cll_descr;
304         file_start = cl_offset(lov2cl(loo), parent->cll_descr.cld_start);
305         file_end   = cl_offset(lov2cl(loo), parent->cll_descr.cld_end + 1) - 1;
306
307         for (i = 0, nr = 0; i < r0->lo_nr; i++) {
308                 /*
309                  * XXX for wide striping smarter algorithm is desirable,
310                  * breaking out of the loop, early.
311                  */
312                 if (likely(r0->lo_sub[i] != NULL) &&
313                     lov_stripe_intersects(loo->lo_lsm, i,
314                                           file_start, file_end, &start, &end))
315                         nr++;
316         }
317         LASSERT(nr > 0);
318         lck->lls_sub = libcfs_kvzalloc(nr * sizeof(lck->lls_sub[0]), GFP_NOFS);
319         if (lck->lls_sub == NULL)
320                 return -ENOMEM;
321
322         lck->lls_nr = nr;
323         /*
324          * First, fill in sub-lock descriptions in
325          * lck->lls_sub[].sub_descr. They are used by lov_sublock_alloc()
326          * (called below in this function, and by lov_lock_enqueue()) to
327          * create sub-locks. At this moment, no other thread can access
328          * top-lock.
329          */
330         for (i = 0, nr = 0; i < r0->lo_nr; ++i) {
331                 if (likely(r0->lo_sub[i] != NULL) &&
332                     lov_stripe_intersects(loo->lo_lsm, i,
333                                           file_start, file_end, &start, &end)) {
334                         struct cl_lock_descr *descr;
335
336                         descr = &lck->lls_sub[nr].sub_descr;
337
338                         LASSERT(descr->cld_obj == NULL);
339                         descr->cld_obj   = lovsub2cl(r0->lo_sub[i]);
340                         descr->cld_start = cl_index(descr->cld_obj, start);
341                         descr->cld_end   = cl_index(descr->cld_obj, end);
342                         descr->cld_mode  = parent->cll_descr.cld_mode;
343                         descr->cld_gid   = parent->cll_descr.cld_gid;
344                         descr->cld_enq_flags   = parent->cll_descr.cld_enq_flags;
345                         /* XXX has no effect */
346                         lck->lls_sub[nr].sub_got = *descr;
347                         lck->lls_sub[nr].sub_stripe = i;
348                         nr++;
349                 }
350         }
351         LASSERT(nr == lck->lls_nr);
352
353         /*
354          * Some sub-locks can be missing at this point. This is not a problem,
355          * because enqueue will create them anyway. Main duty of this function
356          * is to fill in sub-lock descriptions in a race free manner.
357          */
358         return result;
359 }
360
361 static int lov_sublock_release(const struct lu_env *env, struct lov_lock *lck,
362                                int i, int deluser, int rc)
363 {
364         struct cl_lock *parent = lck->lls_cl.cls_lock;
365
366         LASSERT(cl_lock_is_mutexed(parent));
367
368         if (lck->lls_sub[i].sub_flags & LSF_HELD) {
369                 struct cl_lock    *sublock;
370                 int dying;
371
372                 LASSERT(lck->lls_sub[i].sub_lock != NULL);
373                 sublock = lck->lls_sub[i].sub_lock->lss_cl.cls_lock;
374                 LASSERT(cl_lock_is_mutexed(sublock));
375
376                 lck->lls_sub[i].sub_flags &= ~LSF_HELD;
377                 if (deluser)
378                         cl_lock_user_del(env, sublock);
379                 /*
380                  * If the last hold is released, and cancellation is pending
381                  * for a sub-lock, release parent mutex, to avoid keeping it
382                  * while sub-lock is being paged out.
383                  */
384                 dying = (sublock->cll_descr.cld_mode == CLM_PHANTOM ||
385                          sublock->cll_descr.cld_mode == CLM_GROUP ||
386                          (sublock->cll_flags & (CLF_CANCELPEND|CLF_DOOMED))) &&
387                         sublock->cll_holds == 1;
388                 if (dying)
389                         cl_lock_mutex_put(env, parent);
390                 cl_lock_unhold(env, sublock, "lov-parent", parent);
391                 if (dying) {
392                         cl_lock_mutex_get(env, parent);
393                         rc = lov_subresult(rc, CLO_REPEAT);
394                 }
395                 /*
396                  * From now on lck->lls_sub[i].sub_lock is a "weak" pointer,
397                  * not backed by a reference on a
398                  * sub-lock. lovsub_lock_delete() will clear
399                  * lck->lls_sub[i].sub_lock under semaphores, just before
400                  * sub-lock is destroyed.
401                  */
402         }
403         return rc;
404 }
405
406 static void lov_sublock_hold(const struct lu_env *env, struct lov_lock *lck,
407                              int i)
408 {
409         struct cl_lock *parent = lck->lls_cl.cls_lock;
410
411         LASSERT(cl_lock_is_mutexed(parent));
412
413         if (!(lck->lls_sub[i].sub_flags & LSF_HELD)) {
414                 struct cl_lock *sublock;
415
416                 LASSERT(lck->lls_sub[i].sub_lock != NULL);
417                 sublock = lck->lls_sub[i].sub_lock->lss_cl.cls_lock;
418                 LASSERT(cl_lock_is_mutexed(sublock));
419                 LASSERT(sublock->cll_state != CLS_FREEING);
420
421                 lck->lls_sub[i].sub_flags |= LSF_HELD;
422
423                 cl_lock_get_trust(sublock);
424                 cl_lock_hold_add(env, sublock, "lov-parent", parent);
425                 cl_lock_user_add(env, sublock);
426                 cl_lock_put(env, sublock);
427         }
428 }
429
430 static void lov_lock_fini(const struct lu_env *env,
431                           struct cl_lock_slice *slice)
432 {
433         struct lov_lock *lck;
434         int i;
435
436         lck = cl2lov_lock(slice);
437         LASSERT(lck->lls_nr_filled == 0);
438         if (lck->lls_sub != NULL) {
439                 for (i = 0; i < lck->lls_nr; ++i)
440                         /*
441                          * No sub-locks exists at this point, as sub-lock has
442                          * a reference on its parent.
443                          */
444                         LASSERT(lck->lls_sub[i].sub_lock == NULL);
445                 kvfree(lck->lls_sub);
446         }
447         kmem_cache_free(lov_lock_kmem, lck);
448 }
449
450 static int lov_lock_enqueue_wait(const struct lu_env *env,
451                                  struct lov_lock *lck,
452                                  struct cl_lock *sublock)
453 {
454         struct cl_lock *lock = lck->lls_cl.cls_lock;
455         int          result;
456
457         LASSERT(cl_lock_is_mutexed(lock));
458
459         cl_lock_mutex_put(env, lock);
460         result = cl_lock_enqueue_wait(env, sublock, 0);
461         cl_lock_mutex_get(env, lock);
462         return result ?: CLO_REPEAT;
463 }
464
465 /**
466  * Tries to advance a state machine of a given sub-lock toward enqueuing of
467  * the top-lock.
468  *
469  * \retval 0 if state-transition can proceed
470  * \retval -ve otherwise.
471  */
472 static int lov_lock_enqueue_one(const struct lu_env *env, struct lov_lock *lck,
473                                 struct cl_lock *sublock,
474                                 struct cl_io *io, __u32 enqflags, int last)
475 {
476         int result;
477
478         /* first, try to enqueue a sub-lock ... */
479         result = cl_enqueue_try(env, sublock, io, enqflags);
480         if ((sublock->cll_state == CLS_ENQUEUED) && !(enqflags & CEF_AGL)) {
481                 /* if it is enqueued, try to `wait' on it---maybe it's already
482                  * granted */
483                 result = cl_wait_try(env, sublock);
484                 if (result == CLO_REENQUEUED)
485                         result = CLO_WAIT;
486         }
487         /*
488          * If CEF_ASYNC flag is set, then all sub-locks can be enqueued in
489          * parallel, otherwise---enqueue has to wait until sub-lock is granted
490          * before proceeding to the next one.
491          */
492         if ((result == CLO_WAIT) && (sublock->cll_state <= CLS_HELD) &&
493             (enqflags & CEF_ASYNC) && (!last || (enqflags & CEF_AGL)))
494                 result = 0;
495         return result;
496 }
497
498 /**
499  * Helper function for lov_lock_enqueue() that creates missing sub-lock.
500  */
501 static int lov_sublock_fill(const struct lu_env *env, struct cl_lock *parent,
502                             struct cl_io *io, struct lov_lock *lck, int idx)
503 {
504         struct lov_lock_link *link = NULL;
505         struct cl_lock       *sublock;
506         int                result;
507
508         LASSERT(parent->cll_depth == 1);
509         cl_lock_mutex_put(env, parent);
510         sublock = lov_sublock_alloc(env, io, lck, idx, &link);
511         if (!IS_ERR(sublock))
512                 cl_lock_mutex_get(env, sublock);
513         cl_lock_mutex_get(env, parent);
514
515         if (!IS_ERR(sublock)) {
516                 cl_lock_get_trust(sublock);
517                 if (parent->cll_state == CLS_QUEUING &&
518                     lck->lls_sub[idx].sub_lock == NULL) {
519                         lov_sublock_adopt(env, lck, sublock, idx, link);
520                 } else {
521                         kmem_cache_free(lov_lock_link_kmem, link);
522                         /* other thread allocated sub-lock, or enqueue is no
523                          * longer going on */
524                         cl_lock_mutex_put(env, parent);
525                         cl_lock_unhold(env, sublock, "lov-parent", parent);
526                         cl_lock_mutex_get(env, parent);
527                 }
528                 cl_lock_mutex_put(env, sublock);
529                 cl_lock_put(env, sublock);
530                 result = CLO_REPEAT;
531         } else
532                 result = PTR_ERR(sublock);
533         return result;
534 }
535
536 /**
537  * Implementation of cl_lock_operations::clo_enqueue() for lov layer. This
538  * function is rather subtle, as it enqueues top-lock (i.e., advances top-lock
539  * state machine from CLS_QUEUING to CLS_ENQUEUED states) by juggling sub-lock
540  * state machines in the face of sub-locks sharing (by multiple top-locks),
541  * and concurrent sub-lock cancellations.
542  */
543 static int lov_lock_enqueue(const struct lu_env *env,
544                             const struct cl_lock_slice *slice,
545                             struct cl_io *io, __u32 enqflags)
546 {
547         struct cl_lock   *lock    = slice->cls_lock;
548         struct lov_lock *lck     = cl2lov_lock(slice);
549         struct cl_lock_closure *closure = lov_closure_get(env, lock);
550         int i;
551         int result;
552         enum cl_lock_state minstate;
553
554         for (result = 0, minstate = CLS_FREEING, i = 0; i < lck->lls_nr; ++i) {
555                 int rc;
556                 struct lovsub_lock     *sub;
557                 struct lov_lock_sub    *lls;
558                 struct cl_lock   *sublock;
559                 struct lov_sublock_env *subenv;
560
561                 if (lock->cll_state != CLS_QUEUING) {
562                         /*
563                          * Lock might have left QUEUING state if previous
564                          * iteration released its mutex. Stop enqueing in this
565                          * case and let the upper layer to decide what to do.
566                          */
567                         LASSERT(i > 0 && result != 0);
568                         break;
569                 }
570
571                 lls = &lck->lls_sub[i];
572                 sub = lls->sub_lock;
573                 /*
574                  * Sub-lock might have been canceled, while top-lock was
575                  * cached.
576                  */
577                 if (sub == NULL) {
578                         result = lov_sublock_fill(env, lock, io, lck, i);
579                         /* lov_sublock_fill() released @lock mutex,
580                          * restart. */
581                         break;
582                 }
583                 sublock = sub->lss_cl.cls_lock;
584                 rc = lov_sublock_lock(env, lck, lls, closure, &subenv);
585                 if (rc == 0) {
586                         lov_sublock_hold(env, lck, i);
587                         rc = lov_lock_enqueue_one(subenv->lse_env, lck, sublock,
588                                                   subenv->lse_io, enqflags,
589                                                   i == lck->lls_nr - 1);
590                         minstate = min(minstate, sublock->cll_state);
591                         if (rc == CLO_WAIT) {
592                                 switch (sublock->cll_state) {
593                                 case CLS_QUEUING:
594                                         /* take recursive mutex, the lock is
595                                          * released in lov_lock_enqueue_wait.
596                                          */
597                                         cl_lock_mutex_get(env, sublock);
598                                         lov_sublock_unlock(env, sub, closure,
599                                                            subenv);
600                                         rc = lov_lock_enqueue_wait(env, lck,
601                                                                    sublock);
602                                         break;
603                                 case CLS_CACHED:
604                                         cl_lock_get(sublock);
605                                         /* take recursive mutex of sublock */
606                                         cl_lock_mutex_get(env, sublock);
607                                         /* need to release all locks in closure
608                                          * otherwise it may deadlock. LU-2683.*/
609                                         lov_sublock_unlock(env, sub, closure,
610                                                            subenv);
611                                         /* sublock and parent are held. */
612                                         rc = lov_sublock_release(env, lck, i,
613                                                                  1, rc);
614                                         cl_lock_mutex_put(env, sublock);
615                                         cl_lock_put(env, sublock);
616                                         break;
617                                 default:
618                                         lov_sublock_unlock(env, sub, closure,
619                                                            subenv);
620                                         break;
621                                 }
622                         } else {
623                                 LASSERT(sublock->cll_conflict == NULL);
624                                 lov_sublock_unlock(env, sub, closure, subenv);
625                         }
626                 }
627                 result = lov_subresult(result, rc);
628                 if (result != 0)
629                         break;
630         }
631         cl_lock_closure_fini(closure);
632         return result ?: minstate >= CLS_ENQUEUED ? 0 : CLO_WAIT;
633 }
634
635 static int lov_lock_unuse(const struct lu_env *env,
636                           const struct cl_lock_slice *slice)
637 {
638         struct lov_lock *lck     = cl2lov_lock(slice);
639         struct cl_lock_closure *closure = lov_closure_get(env, slice->cls_lock);
640         int i;
641         int result;
642
643         for (result = 0, i = 0; i < lck->lls_nr; ++i) {
644                 int rc;
645                 struct lovsub_lock     *sub;
646                 struct cl_lock   *sublock;
647                 struct lov_lock_sub    *lls;
648                 struct lov_sublock_env *subenv;
649
650                 /* top-lock state cannot change concurrently, because single
651                  * thread (one that released the last hold) carries unlocking
652                  * to the completion. */
653                 LASSERT(slice->cls_lock->cll_state == CLS_INTRANSIT);
654                 lls = &lck->lls_sub[i];
655                 sub = lls->sub_lock;
656                 if (sub == NULL)
657                         continue;
658
659                 sublock = sub->lss_cl.cls_lock;
660                 rc = lov_sublock_lock(env, lck, lls, closure, &subenv);
661                 if (rc == 0) {
662                         if (lls->sub_flags & LSF_HELD) {
663                                 LASSERT(sublock->cll_state == CLS_HELD ||
664                                         sublock->cll_state == CLS_ENQUEUED);
665                                 rc = cl_unuse_try(subenv->lse_env, sublock);
666                                 rc = lov_sublock_release(env, lck, i, 0, rc);
667                         }
668                         lov_sublock_unlock(env, sub, closure, subenv);
669                 }
670                 result = lov_subresult(result, rc);
671         }
672
673         if (result == 0 && lck->lls_cancel_race) {
674                 lck->lls_cancel_race = 0;
675                 result = -ESTALE;
676         }
677         cl_lock_closure_fini(closure);
678         return result;
679 }
680
681 static void lov_lock_cancel(const struct lu_env *env,
682                            const struct cl_lock_slice *slice)
683 {
684         struct lov_lock *lck     = cl2lov_lock(slice);
685         struct cl_lock_closure *closure = lov_closure_get(env, slice->cls_lock);
686         int i;
687         int result;
688
689         for (result = 0, i = 0; i < lck->lls_nr; ++i) {
690                 int rc;
691                 struct lovsub_lock     *sub;
692                 struct cl_lock   *sublock;
693                 struct lov_lock_sub    *lls;
694                 struct lov_sublock_env *subenv;
695
696                 /* top-lock state cannot change concurrently, because single
697                  * thread (one that released the last hold) carries unlocking
698                  * to the completion. */
699                 lls = &lck->lls_sub[i];
700                 sub = lls->sub_lock;
701                 if (sub == NULL)
702                         continue;
703
704                 sublock = sub->lss_cl.cls_lock;
705                 rc = lov_sublock_lock(env, lck, lls, closure, &subenv);
706                 if (rc == 0) {
707                         if (!(lls->sub_flags & LSF_HELD)) {
708                                 lov_sublock_unlock(env, sub, closure, subenv);
709                                 continue;
710                         }
711
712                         switch (sublock->cll_state) {
713                         case CLS_HELD:
714                                 rc = cl_unuse_try(subenv->lse_env, sublock);
715                                 lov_sublock_release(env, lck, i, 0, 0);
716                                 break;
717                         default:
718                                 lov_sublock_release(env, lck, i, 1, 0);
719                                 break;
720                         }
721                         lov_sublock_unlock(env, sub, closure, subenv);
722                 }
723
724                 if (rc == CLO_REPEAT) {
725                         --i;
726                         continue;
727                 }
728
729                 result = lov_subresult(result, rc);
730         }
731
732         if (result)
733                 CL_LOCK_DEBUG(D_ERROR, env, slice->cls_lock,
734                               "lov_lock_cancel fails with %d.\n", result);
735
736         cl_lock_closure_fini(closure);
737 }
738
739 static int lov_lock_wait(const struct lu_env *env,
740                          const struct cl_lock_slice *slice)
741 {
742         struct lov_lock *lck     = cl2lov_lock(slice);
743         struct cl_lock_closure *closure = lov_closure_get(env, slice->cls_lock);
744         enum cl_lock_state      minstate;
745         int                  reenqueued;
746         int                  result;
747         int                  i;
748
749 again:
750         for (result = 0, minstate = CLS_FREEING, i = 0, reenqueued = 0;
751              i < lck->lls_nr; ++i) {
752                 int rc;
753                 struct lovsub_lock     *sub;
754                 struct cl_lock   *sublock;
755                 struct lov_lock_sub    *lls;
756                 struct lov_sublock_env *subenv;
757
758                 lls = &lck->lls_sub[i];
759                 sub = lls->sub_lock;
760                 LASSERT(sub != NULL);
761                 sublock = sub->lss_cl.cls_lock;
762                 rc = lov_sublock_lock(env, lck, lls, closure, &subenv);
763                 if (rc == 0) {
764                         LASSERT(sublock->cll_state >= CLS_ENQUEUED);
765                         if (sublock->cll_state < CLS_HELD)
766                                 rc = cl_wait_try(env, sublock);
767
768                         minstate = min(minstate, sublock->cll_state);
769                         lov_sublock_unlock(env, sub, closure, subenv);
770                 }
771                 if (rc == CLO_REENQUEUED) {
772                         reenqueued++;
773                         rc = 0;
774                 }
775                 result = lov_subresult(result, rc);
776                 if (result != 0)
777                         break;
778         }
779         /* Each sublock only can be reenqueued once, so will not loop for
780          * ever. */
781         if (result == 0 && reenqueued != 0)
782                 goto again;
783         cl_lock_closure_fini(closure);
784         return result ?: minstate >= CLS_HELD ? 0 : CLO_WAIT;
785 }
786
787 static int lov_lock_use(const struct lu_env *env,
788                         const struct cl_lock_slice *slice)
789 {
790         struct lov_lock *lck     = cl2lov_lock(slice);
791         struct cl_lock_closure *closure = lov_closure_get(env, slice->cls_lock);
792         int                  result;
793         int                  i;
794
795         LASSERT(slice->cls_lock->cll_state == CLS_INTRANSIT);
796
797         for (result = 0, i = 0; i < lck->lls_nr; ++i) {
798                 int rc;
799                 struct lovsub_lock     *sub;
800                 struct cl_lock   *sublock;
801                 struct lov_lock_sub    *lls;
802                 struct lov_sublock_env *subenv;
803
804                 LASSERT(slice->cls_lock->cll_state == CLS_INTRANSIT);
805
806                 lls = &lck->lls_sub[i];
807                 sub = lls->sub_lock;
808                 if (sub == NULL) {
809                         /*
810                          * Sub-lock might have been canceled, while top-lock was
811                          * cached.
812                          */
813                         result = -ESTALE;
814                         break;
815                 }
816
817                 sublock = sub->lss_cl.cls_lock;
818                 rc = lov_sublock_lock(env, lck, lls, closure, &subenv);
819                 if (rc == 0) {
820                         LASSERT(sublock->cll_state != CLS_FREEING);
821                         lov_sublock_hold(env, lck, i);
822                         if (sublock->cll_state == CLS_CACHED) {
823                                 rc = cl_use_try(subenv->lse_env, sublock, 0);
824                                 if (rc != 0)
825                                         rc = lov_sublock_release(env, lck,
826                                                                  i, 1, rc);
827                         } else if (sublock->cll_state == CLS_NEW) {
828                                 /* Sub-lock might have been canceled, while
829                                  * top-lock was cached. */
830                                 result = -ESTALE;
831                                 lov_sublock_release(env, lck, i, 1, result);
832                         }
833                         lov_sublock_unlock(env, sub, closure, subenv);
834                 }
835                 result = lov_subresult(result, rc);
836                 if (result != 0)
837                         break;
838         }
839
840         if (lck->lls_cancel_race) {
841                 /*
842                  * If there is unlocking happened at the same time, then
843                  * sublock_lock state should be FREEING, and lov_sublock_lock
844                  * should return CLO_REPEAT. In this case, it should return
845                  * ESTALE, and up layer should reset the lock state to be NEW.
846                  */
847                 lck->lls_cancel_race = 0;
848                 LASSERT(result != 0);
849                 result = -ESTALE;
850         }
851         cl_lock_closure_fini(closure);
852         return result;
853 }
854
855 #if 0
856 static int lock_lock_multi_match()
857 {
858         struct cl_lock    *lock    = slice->cls_lock;
859         struct cl_lock_descr    *subneed = &lov_env_info(env)->lti_ldescr;
860         struct lov_object       *loo     = cl2lov(lov->lls_cl.cls_obj);
861         struct lov_layout_raid0 *r0      = lov_r0(loo);
862         struct lov_lock_sub     *sub;
863         struct cl_object        *subobj;
864         u64  fstart;
865         u64  fend;
866         u64  start;
867         u64  end;
868         int i;
869
870         fstart = cl_offset(need->cld_obj, need->cld_start);
871         fend   = cl_offset(need->cld_obj, need->cld_end + 1) - 1;
872         subneed->cld_mode = need->cld_mode;
873         cl_lock_mutex_get(env, lock);
874         for (i = 0; i < lov->lls_nr; ++i) {
875                 sub = &lov->lls_sub[i];
876                 if (sub->sub_lock == NULL)
877                         continue;
878                 subobj = sub->sub_descr.cld_obj;
879                 if (!lov_stripe_intersects(loo->lo_lsm, sub->sub_stripe,
880                                            fstart, fend, &start, &end))
881                         continue;
882                 subneed->cld_start = cl_index(subobj, start);
883                 subneed->cld_end   = cl_index(subobj, end);
884                 subneed->cld_obj   = subobj;
885                 if (!cl_lock_ext_match(&sub->sub_got, subneed)) {
886                         result = 0;
887                         break;
888                 }
889         }
890         cl_lock_mutex_put(env, lock);
891 }
892 #endif
893
894 /**
895  * Check if the extent region \a descr is covered by \a child against the
896  * specific \a stripe.
897  */
898 static int lov_lock_stripe_is_matching(const struct lu_env *env,
899                                        struct lov_object *lov, int stripe,
900                                        const struct cl_lock_descr *child,
901                                        const struct cl_lock_descr *descr)
902 {
903         struct lov_stripe_md *lsm = lov->lo_lsm;
904         u64 start;
905         u64 end;
906         int result;
907
908         if (lov_r0(lov)->lo_nr == 1)
909                 return cl_lock_ext_match(child, descr);
910
911         /*
912          * For a multi-stripes object:
913          * - make sure the descr only covers child's stripe, and
914          * - check if extent is matching.
915          */
916         start = cl_offset(&lov->lo_cl, descr->cld_start);
917         end   = cl_offset(&lov->lo_cl, descr->cld_end + 1) - 1;
918         result = 0;
919         /* glimpse should work on the object with LOV EA hole. */
920         if (end - start <= lsm->lsm_stripe_size) {
921                 int idx;
922
923                 idx = lov_stripe_number(lsm, start);
924                 if (idx == stripe ||
925                     unlikely(lov_r0(lov)->lo_sub[idx] == NULL)) {
926                         idx = lov_stripe_number(lsm, end);
927                         if (idx == stripe ||
928                             unlikely(lov_r0(lov)->lo_sub[idx] == NULL))
929                                 result = 1;
930                 }
931         }
932
933         if (result != 0) {
934                 struct cl_lock_descr *subd = &lov_env_info(env)->lti_ldescr;
935                 u64 sub_start;
936                 u64 sub_end;
937
938                 subd->cld_obj  = NULL;   /* don't need sub object at all */
939                 subd->cld_mode = descr->cld_mode;
940                 subd->cld_gid  = descr->cld_gid;
941                 result = lov_stripe_intersects(lsm, stripe, start, end,
942                                                &sub_start, &sub_end);
943                 LASSERT(result);
944                 subd->cld_start = cl_index(child->cld_obj, sub_start);
945                 subd->cld_end   = cl_index(child->cld_obj, sub_end);
946                 result = cl_lock_ext_match(child, subd);
947         }
948         return result;
949 }
950
951 /**
952  * An implementation of cl_lock_operations::clo_fits_into() method.
953  *
954  * Checks whether a lock (given by \a slice) is suitable for \a
955  * io. Multi-stripe locks can be used only for "quick" io, like truncate, or
956  * O_APPEND write.
957  *
958  * \see ccc_lock_fits_into().
959  */
960 static int lov_lock_fits_into(const struct lu_env *env,
961                               const struct cl_lock_slice *slice,
962                               const struct cl_lock_descr *need,
963                               const struct cl_io *io)
964 {
965         struct lov_lock   *lov = cl2lov_lock(slice);
966         struct lov_object *obj = cl2lov(slice->cls_obj);
967         int result;
968
969         LASSERT(cl_object_same(need->cld_obj, slice->cls_obj));
970         LASSERT(lov->lls_nr > 0);
971
972         /* for top lock, it's necessary to match enq flags otherwise it will
973          * run into problem if a sublock is missing and reenqueue. */
974         if (need->cld_enq_flags != lov->lls_orig.cld_enq_flags)
975                 return 0;
976
977         if (need->cld_mode == CLM_GROUP)
978                 /*
979                  * always allow to match group lock.
980                  */
981                 result = cl_lock_ext_match(&lov->lls_orig, need);
982         else if (lov->lls_nr == 1) {
983                 struct cl_lock_descr *got = &lov->lls_sub[0].sub_got;
984
985                 result = lov_lock_stripe_is_matching(env,
986                                                      cl2lov(slice->cls_obj),
987                                                      lov->lls_sub[0].sub_stripe,
988                                                      got, need);
989         } else if (io->ci_type != CIT_SETATTR && io->ci_type != CIT_MISC &&
990                    !cl_io_is_append(io) && need->cld_mode != CLM_PHANTOM)
991                 /*
992                  * Multi-stripe locks are only suitable for `quick' IO and for
993                  * glimpse.
994                  */
995                 result = 0;
996         else
997                 /*
998                  * Most general case: multi-stripe existing lock, and
999                  * (potentially) multi-stripe @need lock. Check that @need is
1000                  * covered by @lov's sub-locks.
1001                  *
1002                  * For now, ignore lock expansions made by the server, and
1003                  * match against original lock extent.
1004                  */
1005                 result = cl_lock_ext_match(&lov->lls_orig, need);
1006         CDEBUG(D_DLMTRACE, DDESCR"/"DDESCR" %d %d/%d: %d\n",
1007                PDESCR(&lov->lls_orig), PDESCR(&lov->lls_sub[0].sub_got),
1008                lov->lls_sub[0].sub_stripe, lov->lls_nr, lov_r0(obj)->lo_nr,
1009                result);
1010         return result;
1011 }
1012
1013 void lov_lock_unlink(const struct lu_env *env,
1014                      struct lov_lock_link *link, struct lovsub_lock *sub)
1015 {
1016         struct lov_lock *lck    = link->lll_super;
1017         struct cl_lock  *parent = lck->lls_cl.cls_lock;
1018
1019         LASSERT(cl_lock_is_mutexed(parent));
1020         LASSERT(cl_lock_is_mutexed(sub->lss_cl.cls_lock));
1021
1022         list_del_init(&link->lll_list);
1023         LASSERT(lck->lls_sub[link->lll_idx].sub_lock == sub);
1024         /* yank this sub-lock from parent's array */
1025         lck->lls_sub[link->lll_idx].sub_lock = NULL;
1026         LASSERT(lck->lls_nr_filled > 0);
1027         lck->lls_nr_filled--;
1028         lu_ref_del(&parent->cll_reference, "lov-child", sub->lss_cl.cls_lock);
1029         cl_lock_put(env, parent);
1030         kmem_cache_free(lov_lock_link_kmem, link);
1031 }
1032
1033 struct lov_lock_link *lov_lock_link_find(const struct lu_env *env,
1034                                          struct lov_lock *lck,
1035                                          struct lovsub_lock *sub)
1036 {
1037         struct lov_lock_link *scan;
1038
1039         LASSERT(cl_lock_is_mutexed(sub->lss_cl.cls_lock));
1040
1041         list_for_each_entry(scan, &sub->lss_parents, lll_list) {
1042                 if (scan->lll_super == lck)
1043                         return scan;
1044         }
1045         return NULL;
1046 }
1047
1048 /**
1049  * An implementation of cl_lock_operations::clo_delete() method. This is
1050  * invoked for "top-to-bottom" delete, when lock destruction starts from the
1051  * top-lock, e.g., as a result of inode destruction.
1052  *
1053  * Unlinks top-lock from all its sub-locks. Sub-locks are not deleted there:
1054  * this is done separately elsewhere:
1055  *
1056  *     - for inode destruction, lov_object_delete() calls cl_object_kill() for
1057  *       each sub-object, purging its locks;
1058  *
1059  *     - in other cases (e.g., a fatal error with a top-lock) sub-locks are
1060  *       left in the cache.
1061  */
1062 static void lov_lock_delete(const struct lu_env *env,
1063                             const struct cl_lock_slice *slice)
1064 {
1065         struct lov_lock *lck     = cl2lov_lock(slice);
1066         struct cl_lock_closure *closure = lov_closure_get(env, slice->cls_lock);
1067         struct lov_lock_link   *link;
1068         int                  rc;
1069         int                  i;
1070
1071         LASSERT(slice->cls_lock->cll_state == CLS_FREEING);
1072
1073         for (i = 0; i < lck->lls_nr; ++i) {
1074                 struct lov_lock_sub *lls = &lck->lls_sub[i];
1075                 struct lovsub_lock  *lsl = lls->sub_lock;
1076
1077                 if (lsl == NULL) /* already removed */
1078                         continue;
1079
1080                 rc = lov_sublock_lock(env, lck, lls, closure, NULL);
1081                 if (rc == CLO_REPEAT) {
1082                         --i;
1083                         continue;
1084                 }
1085
1086                 LASSERT(rc == 0);
1087                 LASSERT(lsl->lss_cl.cls_lock->cll_state < CLS_FREEING);
1088
1089                 if (lls->sub_flags & LSF_HELD)
1090                         lov_sublock_release(env, lck, i, 1, 0);
1091
1092                 link = lov_lock_link_find(env, lck, lsl);
1093                 LASSERT(link != NULL);
1094                 lov_lock_unlink(env, link, lsl);
1095                 LASSERT(lck->lls_sub[i].sub_lock == NULL);
1096
1097                 lov_sublock_unlock(env, lsl, closure, NULL);
1098         }
1099
1100         cl_lock_closure_fini(closure);
1101 }
1102
1103 static int lov_lock_print(const struct lu_env *env, void *cookie,
1104                           lu_printer_t p, const struct cl_lock_slice *slice)
1105 {
1106         struct lov_lock *lck = cl2lov_lock(slice);
1107         int           i;
1108
1109         (*p)(env, cookie, "%d\n", lck->lls_nr);
1110         for (i = 0; i < lck->lls_nr; ++i) {
1111                 struct lov_lock_sub *sub;
1112
1113                 sub = &lck->lls_sub[i];
1114                 (*p)(env, cookie, "    %d %x: ", i, sub->sub_flags);
1115                 if (sub->sub_lock != NULL)
1116                         cl_lock_print(env, cookie, p,
1117                                       sub->sub_lock->lss_cl.cls_lock);
1118                 else
1119                         (*p)(env, cookie, "---\n");
1120         }
1121         return 0;
1122 }
1123
1124 static const struct cl_lock_operations lov_lock_ops = {
1125         .clo_fini      = lov_lock_fini,
1126         .clo_enqueue   = lov_lock_enqueue,
1127         .clo_wait      = lov_lock_wait,
1128         .clo_use       = lov_lock_use,
1129         .clo_unuse     = lov_lock_unuse,
1130         .clo_cancel    = lov_lock_cancel,
1131         .clo_fits_into = lov_lock_fits_into,
1132         .clo_delete    = lov_lock_delete,
1133         .clo_print     = lov_lock_print
1134 };
1135
1136 int lov_lock_init_raid0(const struct lu_env *env, struct cl_object *obj,
1137                         struct cl_lock *lock, const struct cl_io *io)
1138 {
1139         struct lov_lock *lck;
1140         int result;
1141
1142         lck = kmem_cache_alloc(lov_lock_kmem, GFP_NOFS | __GFP_ZERO);
1143         if (lck != NULL) {
1144                 cl_lock_slice_add(lock, &lck->lls_cl, obj, &lov_lock_ops);
1145                 result = lov_lock_sub_init(env, lck, io);
1146         } else
1147                 result = -ENOMEM;
1148         return result;
1149 }
1150
1151 static void lov_empty_lock_fini(const struct lu_env *env,
1152                                 struct cl_lock_slice *slice)
1153 {
1154         struct lov_lock *lck = cl2lov_lock(slice);
1155
1156         kmem_cache_free(lov_lock_kmem, lck);
1157 }
1158
1159 static int lov_empty_lock_print(const struct lu_env *env, void *cookie,
1160                         lu_printer_t p, const struct cl_lock_slice *slice)
1161 {
1162         (*p)(env, cookie, "empty\n");
1163         return 0;
1164 }
1165
1166 /* XXX: more methods will be added later. */
1167 static const struct cl_lock_operations lov_empty_lock_ops = {
1168         .clo_fini  = lov_empty_lock_fini,
1169         .clo_print = lov_empty_lock_print
1170 };
1171
1172 int lov_lock_init_empty(const struct lu_env *env, struct cl_object *obj,
1173                 struct cl_lock *lock, const struct cl_io *io)
1174 {
1175         struct lov_lock *lck;
1176         int result = -ENOMEM;
1177
1178         lck = kmem_cache_alloc(lov_lock_kmem, GFP_NOFS | __GFP_ZERO);
1179         if (lck != NULL) {
1180                 cl_lock_slice_add(lock, &lck->lls_cl, obj, &lov_empty_lock_ops);
1181                 lck->lls_orig = lock->cll_descr;
1182                 result = 0;
1183         }
1184         return result;
1185 }
1186
1187 static struct cl_lock_closure *lov_closure_get(const struct lu_env *env,
1188                                                struct cl_lock *parent)
1189 {
1190         struct cl_lock_closure *closure;
1191
1192         closure = &lov_env_info(env)->lti_closure;
1193         LASSERT(list_empty(&closure->clc_list));
1194         cl_lock_closure_init(env, closure, parent, 1);
1195         return closure;
1196 }
1197
1198 /** @} lov */