Whamcloud - gitweb
4b1d3afddbc4c6f3cfe9d7fae6069cd0d5a4cbb9
[fs/lustre-release.git] / lustre / lov / lov_object.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_object for LOV layer.
37  *
38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
39  *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
40  */
41
42 #define DEBUG_SUBSYSTEM S_LOV
43
44 #include "lov_cl_internal.h"
45 #include <lustre_debug.h>
46
47 /** \addtogroup lov
48  *  @{
49  */
50
51 /*****************************************************************************
52  *
53  * Layout operations.
54  *
55  */
56
57 struct lov_layout_operations {
58         int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
59                         struct lov_object *lov,
60                         const struct cl_object_conf *conf,
61                         union lov_layout_state *state);
62         int (*llo_delete)(const struct lu_env *env, struct lov_object *lov,
63                            union lov_layout_state *state);
64         void (*llo_fini)(const struct lu_env *env, struct lov_object *lov,
65                          union lov_layout_state *state);
66         void (*llo_install)(const struct lu_env *env, struct lov_object *lov,
67                             union lov_layout_state *state);
68         int  (*llo_print)(const struct lu_env *env, void *cookie,
69                           lu_printer_t p, const struct lu_object *o);
70         int  (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
71                                 struct cl_page *page, cfs_page_t *vmpage);
72         int  (*llo_lock_init)(const struct lu_env *env,
73                               struct cl_object *obj, struct cl_lock *lock,
74                               const struct cl_io *io);
75         int  (*llo_io_init)(const struct lu_env *env,
76                             struct cl_object *obj, struct cl_io *io);
77         int  (*llo_getattr)(const struct lu_env *env, struct cl_object *obj,
78                             struct cl_attr *attr);
79 };
80
81 /*****************************************************************************
82  *
83  * Lov object layout operations.
84  *
85  */
86
87 static void lov_install_empty(const struct lu_env *env,
88                               struct lov_object *lov,
89                               union  lov_layout_state *state)
90 {
91         /*
92          * File without objects.
93          */
94 }
95
96 static int lov_init_empty(const struct lu_env *env,
97                           struct lov_device *dev, struct lov_object *lov,
98                           const struct cl_object_conf *conf,
99                           union  lov_layout_state *state)
100 {
101         return 0;
102 }
103
104 static void lov_install_raid0(const struct lu_env *env,
105                               struct lov_object *lov,
106                               union  lov_layout_state *state)
107 {
108 }
109
110 static struct cl_object *lov_sub_find(const struct lu_env *env,
111                                       struct cl_device *dev,
112                                       const struct lu_fid *fid,
113                                       const struct cl_object_conf *conf)
114 {
115         struct lu_object *o;
116
117         ENTRY;
118         o = lu_object_find_at(env, cl2lu_dev(dev), fid, &conf->coc_lu);
119         LASSERT(ergo(!IS_ERR(o), o->lo_dev->ld_type == &lovsub_device_type));
120         RETURN(lu2cl(o));
121 }
122
123 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
124                         struct cl_object *stripe,
125                         struct lov_layout_raid0 *r0, int idx)
126 {
127         struct cl_object_header *hdr;
128         struct cl_object_header *subhdr;
129         struct cl_object_header *parent;
130         struct lov_oinfo        *oinfo;
131         int result;
132
133         if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
134                 /* For sanity:test_206.
135                  * Do not leave the object in cache to avoid accessing
136                  * freed memory. This is because osc_object is referring to
137                  * lov_oinfo of lsm_stripe_data which will be freed due to
138                  * this failure. */
139                 cl_object_kill(env, stripe);
140                 cl_object_put(env, stripe);
141                 return -EIO;
142         }
143
144         hdr    = cl_object_header(lov2cl(lov));
145         subhdr = cl_object_header(stripe);
146         parent = subhdr->coh_parent;
147
148         oinfo = lov->lo_lsm->lsm_oinfo[idx];
149         CDEBUG(D_INODE, DFID"@%p[%d] -> "DFID"@%p: id: "LPU64" seq: "LPU64
150                " idx: %d gen: %d\n",
151                PFID(&subhdr->coh_lu.loh_fid), subhdr, idx,
152                PFID(&hdr->coh_lu.loh_fid), hdr,
153                oinfo->loi_id, oinfo->loi_seq,
154                oinfo->loi_ost_idx, oinfo->loi_ost_gen);
155
156         if (parent == NULL) {
157                 subhdr->coh_parent = hdr;
158                 subhdr->coh_nesting = hdr->coh_nesting + 1;
159                 lu_object_ref_add(&stripe->co_lu, "lov-parent", lov);
160                 r0->lo_sub[idx] = cl2lovsub(stripe);
161                 r0->lo_sub[idx]->lso_super = lov;
162                 r0->lo_sub[idx]->lso_index = idx;
163                 result = 0;
164         } else {
165                 CERROR("Stripe is already owned by other file (%d).\n", idx);
166                 LU_OBJECT_DEBUG(D_ERROR, env, &stripe->co_lu, "\n");
167                 LU_OBJECT_DEBUG(D_ERROR, env, lu_object_top(&parent->coh_lu),
168                                 "old\n");
169                 LU_OBJECT_HEADER(D_ERROR, env, lov2lu(lov), "new\n");
170                 cl_object_put(env, stripe);
171                 result = -EIO;
172         }
173         return result;
174 }
175
176 static int lov_init_raid0(const struct lu_env *env,
177                           struct lov_device *dev, struct lov_object *lov,
178                           const struct cl_object_conf *conf,
179                           union  lov_layout_state *state)
180 {
181         int result;
182         int i;
183
184         struct cl_object        *stripe;
185         struct lov_thread_info  *lti     = lov_env_info(env);
186         struct cl_object_conf   *subconf = &lti->lti_stripe_conf;
187         struct lov_stripe_md    *lsm     = conf->u.coc_md->lsm;
188         struct lu_fid           *ofid    = &lti->lti_fid;
189         struct lov_layout_raid0 *r0      = &state->raid0;
190
191         ENTRY;
192
193         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3) {
194                 dump_lsm(D_ERROR, lsm);
195                 LASSERTF(0, "magic mismatch, expected %d/%d, actual %d.\n",
196                          LOV_MAGIC_V1, LOV_MAGIC_V3, lsm->lsm_magic);
197         }
198
199         LASSERT(lov->lo_lsm == NULL);
200         lov->lo_lsm = lsm_addref(lsm);
201         r0->lo_nr  = lsm->lsm_stripe_count;
202         LASSERT(r0->lo_nr <= lov_targets_nr(dev));
203
204         OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
205         if (r0->lo_sub != NULL) {
206                 result = 0;
207                 subconf->coc_inode = conf->coc_inode;
208                 spin_lock_init(&r0->lo_sub_lock);
209                 /*
210                  * Create stripe cl_objects.
211                  */
212                 for (i = 0; i < r0->lo_nr && result == 0; ++i) {
213                         struct cl_device *subdev;
214                         struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
215                         int ost_idx = oinfo->loi_ost_idx;
216
217                         fid_ostid_unpack(ofid, &oinfo->loi_oi,
218                                          oinfo->loi_ost_idx);
219                         subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
220                         subconf->u.coc_oinfo = oinfo;
221                         LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
222                         /* In the function below, .hs_keycmp resolves to
223                          * lu_obj_hop_keycmp() */
224                         /* coverity[overrun-buffer-val] */
225                         stripe = lov_sub_find(env, subdev, ofid, subconf);
226                         if (!IS_ERR(stripe))
227                                 result = lov_init_sub(env, lov, stripe, r0, i);
228                         else
229                                 result = PTR_ERR(stripe);
230                 }
231         } else
232                 result = -ENOMEM;
233         RETURN(result);
234 }
235
236 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
237                             union lov_layout_state *state)
238 {
239         LASSERT(lov->lo_type == LLT_EMPTY);
240         cl_object_prune(env, &lov->lo_cl);
241         return 0;
242 }
243
244 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
245                                struct lovsub_object *los, int idx)
246 {
247         struct cl_object        *sub;
248         struct lov_layout_raid0 *r0;
249         struct lu_site          *site;
250         struct lu_site_bkt_data *bkt;
251         cfs_waitlink_t          *waiter;
252
253         r0  = &lov->u.raid0;
254         LASSERT(r0->lo_sub[idx] == los);
255
256         sub  = lovsub2cl(los);
257         site = sub->co_lu.lo_dev->ld_site;
258         bkt  = lu_site_bkt_from_fid(site, &sub->co_lu.lo_header->loh_fid);
259
260         cl_object_kill(env, sub);
261         /* release a reference to the sub-object and ... */
262         lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
263         cl_object_put(env, sub);
264
265         /* ... wait until it is actually destroyed---sub-object clears its
266          * ->lo_sub[] slot in lovsub_object_fini() */
267         if (r0->lo_sub[idx] == los) {
268                 waiter = &lov_env_info(env)->lti_waiter;
269                 cfs_waitlink_init(waiter);
270                 cfs_waitq_add(&bkt->lsb_marche_funebre, waiter);
271                 cfs_set_current_state(CFS_TASK_UNINT);
272                 while (1) {
273                         /* this wait-queue is signaled at the end of
274                          * lu_object_free(). */
275                         cfs_set_current_state(CFS_TASK_UNINT);
276                         spin_lock(&r0->lo_sub_lock);
277                         if (r0->lo_sub[idx] == los) {
278                                 spin_unlock(&r0->lo_sub_lock);
279                                 cfs_waitq_wait(waiter, CFS_TASK_UNINT);
280                         } else {
281                                 spin_unlock(&r0->lo_sub_lock);
282                                 cfs_set_current_state(CFS_TASK_RUNNING);
283                                 break;
284                         }
285                 }
286                 cfs_waitq_del(&bkt->lsb_marche_funebre, waiter);
287         }
288         LASSERT(r0->lo_sub[idx] == NULL);
289 }
290
291 static int lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
292                             union lov_layout_state *state)
293 {
294         struct lov_layout_raid0 *r0 = &state->raid0;
295         struct lov_stripe_md    *lsm = lov->lo_lsm;
296         int i;
297
298         ENTRY;
299
300         dump_lsm(D_INODE, lsm);
301         if (lov->lo_lsm_invalid && cfs_atomic_read(&lsm->lsm_refc) > 1)
302                 RETURN(-EBUSY);
303
304         if (r0->lo_sub != NULL) {
305                 for (i = 0; i < r0->lo_nr; ++i) {
306                         struct lovsub_object *los = r0->lo_sub[i];
307
308                         if (los != NULL) {
309                                 cl_locks_prune(env, &los->lso_cl, 1);
310                                 /*
311                                  * If top-level object is to be evicted from
312                                  * the cache, so are its sub-objects.
313                                  */
314                                 lov_subobject_kill(env, lov, los, i);
315                         }
316                 }
317         }
318         RETURN(0);
319 }
320
321 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
322                            union lov_layout_state *state)
323 {
324         LASSERT(lov->lo_type == LLT_EMPTY);
325 }
326
327 static void lov_fini_raid0(const struct lu_env *env, struct lov_object *lov,
328                            union lov_layout_state *state)
329 {
330         struct lov_layout_raid0 *r0 = &state->raid0;
331         ENTRY;
332
333         if (r0->lo_sub != NULL) {
334                 OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
335                 r0->lo_sub = NULL;
336         }
337
338         dump_lsm(D_INODE, lov->lo_lsm);
339         lov_free_memmd(&lov->lo_lsm);
340
341         EXIT;
342 }
343
344 static int lov_print_empty(const struct lu_env *env, void *cookie,
345                            lu_printer_t p, const struct lu_object *o)
346 {
347         (*p)(env, cookie, "empty\n");
348         return 0;
349 }
350
351 static int lov_print_raid0(const struct lu_env *env, void *cookie,
352                            lu_printer_t p, const struct lu_object *o)
353 {
354         struct lov_object       *lov = lu2lov(o);
355         struct lov_layout_raid0 *r0  = lov_r0(lov);
356         int i;
357
358         (*p)(env, cookie, "stripes: %d:\n", r0->lo_nr);
359         for (i = 0; i < r0->lo_nr; ++i) {
360                 struct lu_object *sub;
361
362                 if (r0->lo_sub[i] != NULL) {
363                         sub = lovsub2lu(r0->lo_sub[i]);
364                         lu_object_print(env, cookie, p, sub);
365                 } else
366                         (*p)(env, cookie, "sub %d absent\n", i);
367         }
368         return 0;
369 }
370
371 /**
372  * Implements cl_object_operations::coo_attr_get() method for an object
373  * without stripes (LLT_EMPTY layout type).
374  *
375  * The only attributes this layer is authoritative in this case is
376  * cl_attr::cat_blocks---it's 0.
377  */
378 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
379                               struct cl_attr *attr)
380 {
381         attr->cat_blocks = 0;
382         return 0;
383 }
384
385 static int lov_attr_get_raid0(const struct lu_env *env, struct cl_object *obj,
386                               struct cl_attr *attr)
387 {
388         struct lov_object       *lov = cl2lov(obj);
389         struct lov_layout_raid0 *r0 = lov_r0(lov);
390         struct lov_stripe_md    *lsm = lov->lo_lsm;
391         struct ost_lvb          *lvb = &lov_env_info(env)->lti_lvb;
392         __u64                    kms;
393         int                      result = 0;
394
395         ENTRY;
396
397         /* this is called w/o holding type guard mutex, so it must be inside
398          * an on going IO otherwise lsm may be replaced.
399          * LU-2117: it turns out there exists one exception. For mmaped files,
400          * the lock of those files may be requested in the other file's IO
401          * context, and this function is called in ccc_lock_state(), it will
402          * hit this assertion.
403          * Anyway, it's still okay to call attr_get w/o type guard as layout
404          * can't go if locks exist. */
405         /* LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 1); */
406
407         if (!r0->lo_attr_valid) {
408                 /*
409                  * Fill LVB with attributes already initialized by the upper
410                  * layer.
411                  */
412                 cl_attr2lvb(lvb, attr);
413                 kms = attr->cat_kms;
414
415                 /*
416                  * XXX that should be replaced with a loop over sub-objects,
417                  * doing cl_object_attr_get() on them. But for now, let's
418                  * reuse old lov code.
419                  */
420
421                 /*
422                  * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
423                  * happy. It's not needed, because new code uses
424                  * ->coh_attr_guard spin-lock to protect consistency of
425                  * sub-object attributes.
426                  */
427                 lov_stripe_lock(lsm);
428                 result = lov_merge_lvb_kms(lsm, lvb, &kms);
429                 lov_stripe_unlock(lsm);
430                 if (result == 0) {
431                         cl_lvb2attr(attr, lvb);
432                         attr->cat_kms = kms;
433                         r0->lo_attr_valid = 1;
434                         r0->lo_attr = *attr;
435                 }
436         } else
437                 *attr = r0->lo_attr;
438         RETURN(result);
439 }
440
441 const static struct lov_layout_operations lov_dispatch[] = {
442         [LLT_EMPTY] = {
443                 .llo_init      = lov_init_empty,
444                 .llo_delete    = lov_delete_empty,
445                 .llo_fini      = lov_fini_empty,
446                 .llo_install   = lov_install_empty,
447                 .llo_print     = lov_print_empty,
448                 .llo_page_init = lov_page_init_empty,
449                 .llo_lock_init = lov_lock_init_empty,
450                 .llo_io_init   = lov_io_init_empty,
451                 .llo_getattr   = lov_attr_get_empty
452         },
453         [LLT_RAID0] = {
454                 .llo_init      = lov_init_raid0,
455                 .llo_delete    = lov_delete_raid0,
456                 .llo_fini      = lov_fini_raid0,
457                 .llo_install   = lov_install_raid0,
458                 .llo_print     = lov_print_raid0,
459                 .llo_page_init = lov_page_init_raid0,
460                 .llo_lock_init = lov_lock_init_raid0,
461                 .llo_io_init   = lov_io_init_raid0,
462                 .llo_getattr   = lov_attr_get_raid0
463         }
464 };
465
466
467 /**
468  * Performs a double-dispatch based on the layout type of an object.
469  */
470 #define LOV_2DISPATCH_NOLOCK(obj, op, ...)                              \
471 ({                                                                      \
472         struct lov_object                      *__obj = (obj);          \
473         enum lov_layout_type                    __llt;                  \
474                                                                         \
475         __llt = __obj->lo_type;                                         \
476         LASSERT(0 <= __llt && __llt < ARRAY_SIZE(lov_dispatch));        \
477         lov_dispatch[__llt].op(__VA_ARGS__);                            \
478 })
479
480 static inline void lov_conf_freeze(struct lov_object *lov)
481 {
482         if (lov->lo_owner != cfs_current())
483                 down_read(&lov->lo_type_guard);
484 }
485
486 static inline void lov_conf_thaw(struct lov_object *lov)
487 {
488         if (lov->lo_owner != cfs_current())
489                 up_read(&lov->lo_type_guard);
490 }
491
492 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...)                       \
493 ({                                                                      \
494         struct lov_object                      *__obj = (obj);          \
495         int                                     __lock = !!(lock);      \
496         typeof(lov_dispatch[0].op(__VA_ARGS__)) __result;               \
497                                                                         \
498         if (__lock)                                                     \
499                 lov_conf_freeze(__obj);                                 \
500         __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__);          \
501         if (__lock)                                                     \
502                 lov_conf_thaw(__obj);                                   \
503         __result;                                                       \
504 })
505
506 /**
507  * Performs a locked double-dispatch based on the layout type of an object.
508  */
509 #define LOV_2DISPATCH(obj, op, ...)                     \
510         LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
511
512 #define LOV_2DISPATCH_VOID(obj, op, ...)                                \
513 do {                                                                    \
514         struct lov_object                      *__obj = (obj);          \
515         enum lov_layout_type                    __llt;                  \
516                                                                         \
517         lov_conf_freeze(__obj);                                         \
518         __llt = __obj->lo_type;                                         \
519         LASSERT(0 <= __llt && __llt < ARRAY_SIZE(lov_dispatch));        \
520         lov_dispatch[__llt].op(__VA_ARGS__);                            \
521         lov_conf_thaw(__obj);                                           \
522 } while (0)
523
524 static void lov_conf_lock(struct lov_object *lov)
525 {
526         LASSERT(lov->lo_owner != cfs_current());
527         down_write(&lov->lo_type_guard);
528         LASSERT(lov->lo_owner == NULL);
529         lov->lo_owner = cfs_current();
530 }
531
532 static void lov_conf_unlock(struct lov_object *lov)
533 {
534         lov->lo_owner = NULL;
535         up_write(&lov->lo_type_guard);
536 }
537
538 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
539 {
540         struct l_wait_info lwi = { 0 };
541         struct lov_stripe_md *lsm = lov->lo_lsm;
542         ENTRY;
543
544         if (!lov->lo_lsm_invalid || lsm == NULL)
545                 RETURN(0);
546
547         LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 0);
548         while (cfs_atomic_read(&lsm->lsm_refc) > 1 && lov->lo_lsm_invalid) {
549                 lov_conf_unlock(lov);
550
551                 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
552                         PFID(lu_object_fid(lov2lu(lov))),
553                         cfs_atomic_read(&lsm->lsm_refc));
554
555                 l_wait_event(lov->lo_waitq,
556                              cfs_atomic_read(&lsm->lsm_refc) == 1, &lwi);
557                 lov_conf_lock(lov);
558         }
559         RETURN(0);
560 }
561
562 static int lov_layout_change(const struct lu_env *unused,
563                              struct lov_object *lov, enum lov_layout_type llt,
564                              const struct cl_object_conf *conf)
565 {
566         int result;
567         union lov_layout_state *state = &lov->u;
568         const struct lov_layout_operations *old_ops;
569         const struct lov_layout_operations *new_ops;
570
571         struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
572         void *cookie;
573         struct lu_env *env;
574         int refcheck;
575
576         LASSERT(0 <= lov->lo_type && lov->lo_type < ARRAY_SIZE(lov_dispatch));
577         LASSERT(0 <= llt && llt < ARRAY_SIZE(lov_dispatch));
578         ENTRY;
579
580         cookie = cl_env_reenter();
581         env = cl_env_get(&refcheck);
582         if (IS_ERR(env)) {
583                 cl_env_reexit(cookie);
584                 RETURN(PTR_ERR(env));
585         }
586
587         old_ops = &lov_dispatch[lov->lo_type];
588         new_ops = &lov_dispatch[llt];
589
590         result = old_ops->llo_delete(env, lov, &lov->u);
591         if (result == 0) {
592                 old_ops->llo_fini(env, lov, &lov->u);
593                 LASSERT(cfs_list_empty(&hdr->coh_locks));
594                 LASSERT(hdr->coh_tree.rnode == NULL);
595                 LASSERT(hdr->coh_pages == 0);
596
597                 lov->lo_type = LLT_EMPTY;
598                 result = new_ops->llo_init(env,
599                                         lu2lov_dev(lov->lo_cl.co_lu.lo_dev),
600                                         lov, conf, state);
601                 if (result == 0) {
602                         new_ops->llo_install(env, lov, state);
603                         lov->lo_type = llt;
604                 } else {
605                         new_ops->llo_delete(env, lov, state);
606                         new_ops->llo_fini(env, lov, state);
607                         /* this file becomes an EMPTY file. */
608                 }
609         }
610
611         cl_env_put(env, &refcheck);
612         cl_env_reexit(cookie);
613         RETURN(result);
614 }
615
616 /*****************************************************************************
617  *
618  * Lov object operations.
619  *
620  */
621
622 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
623                     const struct lu_object_conf *conf)
624 {
625         struct lov_device            *dev   = lu2lov_dev(obj->lo_dev);
626         struct lov_object            *lov   = lu2lov(obj);
627         const struct cl_object_conf  *cconf = lu2cl_conf(conf);
628         union  lov_layout_state      *set   = &lov->u;
629         const struct lov_layout_operations *ops;
630         int result;
631
632         ENTRY;
633         init_rwsem(&lov->lo_type_guard);
634         cfs_waitq_init(&lov->lo_waitq);
635
636         cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
637
638         /* no locking is necessary, as object is being created */
639         lov->lo_type = cconf->u.coc_md->lsm != NULL ? LLT_RAID0 : LLT_EMPTY;
640         ops = &lov_dispatch[lov->lo_type];
641         result = ops->llo_init(env, dev, lov, cconf, set);
642         if (result == 0)
643                 ops->llo_install(env, lov, set);
644         RETURN(result);
645 }
646
647 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
648                         const struct cl_object_conf *conf)
649 {
650         struct lov_stripe_md *lsm = NULL;
651         struct lov_object *lov = cl2lov(obj);
652         int result = 0;
653         ENTRY;
654
655         lov_conf_lock(lov);
656         if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
657                 lov->lo_lsm_invalid = 1;
658                 GOTO(out, result = 0);
659         }
660
661         if (conf->coc_opc == OBJECT_CONF_WAIT) {
662                 result = lov_layout_wait(env, lov);
663                 GOTO(out, result);
664         }
665
666         LASSERT(conf->coc_opc == OBJECT_CONF_SET);
667
668         if (conf->u.coc_md != NULL)
669                 lsm = conf->u.coc_md->lsm;
670         if ((lsm == NULL && lov->lo_lsm == NULL) ||
671             (lsm != NULL && lov->lo_lsm != NULL &&
672              lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen)) {
673                 /* same version of layout */
674                 lov->lo_lsm_invalid = 0;
675                 GOTO(out, result = 0);
676         }
677
678         /* will change layout - check if there still exists active IO. */
679         if (lov->lo_lsm != NULL &&
680             cfs_atomic_read(&lov->lo_lsm->lsm_refc) > 1) {
681                 lov->lo_lsm_invalid = 1;
682                 GOTO(out, result = -EBUSY);
683         }
684
685         /*
686          * Only LLT_EMPTY <-> LLT_RAID0 transitions are supported.
687          */
688         switch (lov->lo_type) {
689         case LLT_EMPTY:
690                 if (lsm != NULL)
691                         result = lov_layout_change(env, lov, LLT_RAID0, conf);
692                 break;
693         case LLT_RAID0:
694                 if (lsm == NULL)
695                         result = lov_layout_change(env, lov, LLT_EMPTY, conf);
696                 else if (lov_stripe_md_cmp(lov->lo_lsm, lsm))
697                         result = -EOPNOTSUPP;
698                 break;
699         default:
700                 LBUG();
701         }
702         lov->lo_lsm_invalid = result != 0;
703         EXIT;
704
705 out:
706         lov_conf_unlock(lov);
707         RETURN(result);
708 }
709
710 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
711 {
712         struct lov_object *lov = lu2lov(obj);
713
714         ENTRY;
715         LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
716         EXIT;
717 }
718
719 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
720 {
721         struct lov_object *lov = lu2lov(obj);
722
723         ENTRY;
724         LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
725         lu_object_fini(obj);
726         OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
727         EXIT;
728 }
729
730 static int lov_object_print(const struct lu_env *env, void *cookie,
731                             lu_printer_t p, const struct lu_object *o)
732 {
733         return LOV_2DISPATCH(lu2lov(o), llo_print, env, cookie, p, o);
734 }
735
736 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
737                 struct cl_page *page, cfs_page_t *vmpage)
738 {
739         return LOV_2DISPATCH_NOLOCK(cl2lov(obj),
740                                     llo_page_init, env, obj, page, vmpage);
741 }
742
743 /**
744  * Implements cl_object_operations::clo_io_init() method for lov
745  * layer. Dispatches to the appropriate layout io initialization method.
746  */
747 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
748                 struct cl_io *io)
749 {
750         CL_IO_SLICE_CLEAN(lov_env_io(env), lis_cl);
751         return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
752                                      !io->ci_ignore_layout, env, obj, io);
753 }
754
755 /**
756  * An implementation of cl_object_operations::clo_attr_get() method for lov
757  * layer. For raid0 layout this collects and merges attributes of all
758  * sub-objects.
759  */
760 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
761                         struct cl_attr *attr)
762 {
763         /* do not take lock, as this function is called under a
764          * spin-lock. Layout is protected from changing by ongoing IO. */
765         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
766 }
767
768 static int lov_attr_set(const struct lu_env *env, struct cl_object *obj,
769                         const struct cl_attr *attr, unsigned valid)
770 {
771         /*
772          * No dispatch is required here, as no layout implements this.
773          */
774         return 0;
775 }
776
777 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
778                   struct cl_lock *lock, const struct cl_io *io)
779 {
780         /* No need to lock because we've taken one refcount of layout.  */
781         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
782                                     io);
783 }
784
785 static const struct cl_object_operations lov_ops = {
786         .coo_page_init = lov_page_init,
787         .coo_lock_init = lov_lock_init,
788         .coo_io_init   = lov_io_init,
789         .coo_attr_get  = lov_attr_get,
790         .coo_attr_set  = lov_attr_set,
791         .coo_conf_set  = lov_conf_set
792 };
793
794 static const struct lu_object_operations lov_lu_obj_ops = {
795         .loo_object_init      = lov_object_init,
796         .loo_object_delete    = lov_object_delete,
797         .loo_object_release   = NULL,
798         .loo_object_free      = lov_object_free,
799         .loo_object_print     = lov_object_print,
800         .loo_object_invariant = NULL
801 };
802
803 struct lu_object *lov_object_alloc(const struct lu_env *env,
804                                    const struct lu_object_header *unused,
805                                    struct lu_device *dev)
806 {
807         struct lov_object *lov;
808         struct lu_object  *obj;
809
810         ENTRY;
811         OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, CFS_ALLOC_IO);
812         if (lov != NULL) {
813                 obj = lov2lu(lov);
814                 lu_object_init(obj, NULL, dev);
815                 lov->lo_cl.co_ops = &lov_ops;
816                 lov->lo_type = -1; /* invalid, to catch uninitialized type */
817                 /*
818                  * object io operation vector (cl_object::co_iop) is installed
819                  * later in lov_object_init(), as different vectors are used
820                  * for object with different layouts.
821                  */
822                 obj->lo_ops = &lov_lu_obj_ops;
823         } else
824                 obj = NULL;
825         RETURN(obj);
826 }
827
828 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
829 {
830         struct lov_stripe_md *lsm = NULL;
831
832         lov_conf_freeze(lov);
833         if (lov->lo_lsm != NULL) {
834                 lsm = lsm_addref(lov->lo_lsm);
835                 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
836                         lsm, cfs_atomic_read(&lsm->lsm_refc),
837                         lov->lo_lsm_invalid, cfs_current());
838         }
839         lov_conf_thaw(lov);
840         return lsm;
841 }
842
843 void lov_lsm_decref(struct lov_object *lov, struct lov_stripe_md *lsm)
844 {
845         if (lsm == NULL)
846                 return;
847
848         CDEBUG(D_INODE, "lsm %p decref %d by %p.\n",
849                 lsm, cfs_atomic_read(&lsm->lsm_refc), cfs_current());
850
851         if (lov_free_memmd(&lsm) <= 1 && lov->lo_lsm_invalid)
852                 cfs_waitq_signal(&lov->lo_waitq);
853 }
854
855 struct lov_stripe_md *lov_lsm_get(struct cl_object *clobj)
856 {
857         struct lu_object *luobj;
858         struct lov_stripe_md *lsm = NULL;
859
860         if (clobj == NULL)
861                 return NULL;
862
863         luobj = lu_object_locate(&cl_object_header(clobj)->coh_lu,
864                                  &lov_device_type);
865         if (luobj != NULL)
866                 lsm = lov_lsm_addref(lu2lov(luobj));
867         return lsm;
868 }
869 EXPORT_SYMBOL(lov_lsm_get);
870
871 void lov_lsm_put(struct cl_object *clobj, struct lov_stripe_md *lsm)
872 {
873         struct lu_object *luobj;
874
875         if (clobj == NULL || lsm == NULL)
876                 return;
877
878         luobj = lu_object_locate(&cl_object_header(clobj)->coh_lu,
879                                  &lov_device_type);
880         LASSERT(luobj != NULL);
881
882         lov_lsm_decref(lu2lov(luobj), lsm);
883 }
884 EXPORT_SYMBOL(lov_lsm_put);
885
886 int lov_read_and_clear_async_rc(struct cl_object *clob)
887 {
888         struct lu_object *luobj;
889         int rc = 0;
890         ENTRY;
891
892         luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
893                                  &lov_device_type);
894         if (luobj != NULL) {
895                 struct lov_object *lov = lu2lov(luobj);
896
897                 lov_conf_freeze(lov);
898                 switch (lov->lo_type) {
899                 case LLT_RAID0: {
900                         struct lov_stripe_md *lsm;
901                         int i;
902
903                         lsm = lov->lo_lsm;
904                         LASSERT(lsm != NULL);
905                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
906                                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
907                                 if (loi->loi_ar.ar_rc && !rc)
908                                         rc = loi->loi_ar.ar_rc;
909                                 loi->loi_ar.ar_rc = 0;
910                         }
911                 }
912                 case LLT_EMPTY:
913                         break;
914                 default:
915                         LBUG();
916                 }
917                 lov_conf_thaw(lov);
918         }
919         RETURN(rc);
920 }
921 EXPORT_SYMBOL(lov_read_and_clear_async_rc);
922
923 /** @} lov */