Whamcloud - gitweb
LU-12516 mdd: support for volatile creation in .lustre
[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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * Implementation of cl_object for LOV layer.
33  *
34  *   Author: Nikita Danilov <nikita.danilov@sun.com>
35  *   Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
36  */
37
38 #define DEBUG_SUBSYSTEM S_LOV
39
40 #include <linux/random.h>
41
42 #include "lov_cl_internal.h"
43
44 static inline struct lov_device *lov_object_dev(struct lov_object *obj)
45 {
46         return lu2lov_dev(obj->lo_cl.co_lu.lo_dev);
47 }
48
49 /** \addtogroup lov
50  *  @{
51  */
52
53 /*****************************************************************************
54  *
55  * Layout operations.
56  *
57  */
58
59 struct lov_layout_operations {
60         int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
61                         struct lov_object *lov, struct lov_stripe_md *lsm,
62                         const struct cl_object_conf *conf,
63                         union lov_layout_state *state);
64         int (*llo_delete)(const struct lu_env *env, struct lov_object *lov,
65                            union lov_layout_state *state);
66         void (*llo_fini)(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, pgoff_t index);
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 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov);
82
83 static void lov_lsm_put(struct lov_stripe_md *lsm)
84 {
85         if (lsm != NULL)
86                 lov_free_memmd(&lsm);
87 }
88
89 /*****************************************************************************
90  *
91  * Lov object layout operations.
92  *
93  */
94
95 static struct cl_object *lov_sub_find(const struct lu_env *env,
96                                       struct cl_device *dev,
97                                       const struct lu_fid *fid,
98                                       const struct cl_object_conf *conf)
99 {
100         struct lu_object *o;
101
102         ENTRY;
103
104         o = lu_object_find_at(env, cl2lu_dev(dev), fid, &conf->coc_lu);
105         LASSERT(ergo(!IS_ERR(o), o->lo_dev->ld_type == &lovsub_device_type));
106         RETURN(lu2cl(o));
107 }
108
109 static int lov_page_slice_fixup(struct lov_object *lov,
110                                 struct cl_object *stripe)
111 {
112         struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
113         struct cl_object *o;
114
115         if (stripe == NULL)
116                 return hdr->coh_page_bufsize - lov->lo_cl.co_slice_off -
117                        cfs_size_round(sizeof(struct lov_page));
118
119         cl_object_for_each(o, stripe)
120                 o->co_slice_off += hdr->coh_page_bufsize;
121
122         return cl_object_header(stripe)->coh_page_bufsize;
123 }
124
125 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
126                         struct cl_object *subobj, struct lov_oinfo *oinfo,
127                         int idx)
128 {
129         struct cl_object_header *hdr;
130         struct cl_object_header *subhdr;
131         struct cl_object_header *parent;
132         int entry = lov_comp_entry(idx);
133         int stripe = lov_comp_stripe(idx);
134         int result;
135
136         if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
137                 /* For sanity:test_206.
138                  * Do not leave the object in cache to avoid accessing
139                  * freed memory. This is because osc_object is referring to
140                  * lov_oinfo of lsm_stripe_data which will be freed due to
141                  * this failure. */
142                 cl_object_kill(env, subobj);
143                 cl_object_put(env, subobj);
144                 return -EIO;
145         }
146
147         hdr = cl_object_header(lov2cl(lov));
148         subhdr = cl_object_header(subobj);
149
150         CDEBUG(D_INODE, DFID"@%p[%d:%d] -> "DFID"@%p: ostid: "DOSTID
151                " ost idx: %d gen: %d\n",
152                PFID(lu_object_fid(&subobj->co_lu)), subhdr, entry, stripe,
153                PFID(lu_object_fid(lov2lu(lov))), hdr, POSTID(&oinfo->loi_oi),
154                oinfo->loi_ost_idx, oinfo->loi_ost_gen);
155
156         /* reuse ->coh_attr_guard to protect coh_parent change */
157         spin_lock(&subhdr->coh_attr_guard);
158         parent = subhdr->coh_parent;
159         if (parent == NULL) {
160                 struct lovsub_object *lso = cl2lovsub(subobj);
161
162                 subhdr->coh_parent = hdr;
163                 spin_unlock(&subhdr->coh_attr_guard);
164                 subhdr->coh_nesting = hdr->coh_nesting + 1;
165                 lu_object_ref_add(&subobj->co_lu, "lov-parent", lov);
166                 lso->lso_super = lov;
167                 lso->lso_index = idx;
168                 result = 0;
169         } else {
170                 struct lu_object  *old_obj;
171                 struct lov_object *old_lov;
172                 unsigned int mask = D_INODE;
173
174                 spin_unlock(&subhdr->coh_attr_guard);
175                 old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
176                 LASSERT(old_obj != NULL);
177                 old_lov = cl2lov(lu2cl(old_obj));
178                 if (old_lov->lo_layout_invalid) {
179                         /* the object's layout has already changed but isn't
180                          * refreshed */
181                         lu_object_unhash(env, &subobj->co_lu);
182                         result = -EAGAIN;
183                 } else {
184                         mask = D_ERROR;
185                         result = -EIO;
186                 }
187
188                 LU_OBJECT_DEBUG(mask, env, &subobj->co_lu,
189                                 "stripe %d is already owned.", idx);
190                 LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
191                 LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
192                 cl_object_put(env, subobj);
193         }
194         return result;
195 }
196
197 static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
198                           struct lov_object *lov, unsigned int index,
199                           const struct cl_object_conf *conf,
200                           struct lov_layout_entry *lle)
201 {
202         struct lov_layout_raid0 *r0 = &lle->lle_raid0;
203         struct lov_thread_info *lti = lov_env_info(env);
204         struct cl_object_conf *subconf = &lti->lti_stripe_conf;
205         struct lu_fid *ofid = &lti->lti_fid;
206         struct cl_object *stripe;
207         struct lov_stripe_md_entry *lse  = lov_lse(lov, index);
208         int result;
209         int psz, sz;
210         int i;
211
212         ENTRY;
213
214         spin_lock_init(&r0->lo_sub_lock);
215         r0->lo_nr = lse->lsme_stripe_count;
216
217         OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof(r0->lo_sub[0]));
218         if (r0->lo_sub == NULL)
219                 GOTO(out, result = -ENOMEM);
220
221         psz = 0;
222         result = 0;
223         memset(subconf, 0, sizeof(*subconf));
224
225         /*
226          * Create stripe cl_objects.
227          */
228         for (i = 0; i < r0->lo_nr; ++i) {
229                 struct cl_device *subdev;
230                 struct lov_oinfo *oinfo = lse->lsme_oinfo[i];
231                 int ost_idx = oinfo->loi_ost_idx;
232
233                 if (lov_oinfo_is_dummy(oinfo))
234                         continue;
235
236                 result = ostid_to_fid(ofid, &oinfo->loi_oi, oinfo->loi_ost_idx);
237                 if (result != 0)
238                         GOTO(out, result);
239
240                 if (dev->ld_target[ost_idx] == NULL) {
241                         CERROR("%s: OST %04x is not initialized\n",
242                                lov2obd(dev->ld_lov)->obd_name, ost_idx);
243                         GOTO(out, result = -EIO);
244                 }
245
246                 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
247                 subconf->u.coc_oinfo = oinfo;
248                 LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
249                 /* In the function below, .hs_keycmp resolves to
250                  * lu_obj_hop_keycmp() */
251                 /* coverity[overrun-buffer-val] */
252                 stripe = lov_sub_find(env, subdev, ofid, subconf);
253                 if (IS_ERR(stripe))
254                         GOTO(out, result = PTR_ERR(stripe));
255
256                 result = lov_init_sub(env, lov, stripe, oinfo,
257                                       lov_comp_index(index, i));
258                 if (result == -EAGAIN) { /* try again */
259                         --i;
260                         result = 0;
261                         continue;
262                 }
263
264                 if (result == 0) {
265                         r0->lo_sub[i] = cl2lovsub(stripe);
266
267                         sz = lov_page_slice_fixup(lov, stripe);
268                         LASSERT(ergo(psz > 0, psz == sz));
269                         psz = sz;
270                 }
271         }
272         if (result == 0)
273                 result = psz;
274 out:
275         RETURN(result);
276 }
277
278 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
279                                struct lov_layout_raid0 *r0,
280                                struct lovsub_object *los, int idx)
281 {
282         struct cl_object        *sub;
283         struct lu_site          *site;
284         wait_queue_head_t *wq;
285         wait_queue_entry_t *waiter;
286
287         LASSERT(r0->lo_sub[idx] == los);
288
289         sub = lovsub2cl(los);
290         site = sub->co_lu.lo_dev->ld_site;
291         wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
292
293         cl_object_kill(env, sub);
294         /* release a reference to the sub-object and ... */
295         lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
296         cl_object_put(env, sub);
297
298         /* ... wait until it is actually destroyed---sub-object clears its
299          * ->lo_sub[] slot in lovsub_object_free() */
300         if (r0->lo_sub[idx] == los) {
301                 waiter = &lov_env_info(env)->lti_waiter;
302                 init_waitqueue_entry(waiter, current);
303                 add_wait_queue(wq, waiter);
304                 set_current_state(TASK_UNINTERRUPTIBLE);
305                 while (1) {
306                         /* this wait-queue is signaled at the end of
307                          * lu_object_free(). */
308                         set_current_state(TASK_UNINTERRUPTIBLE);
309                         spin_lock(&r0->lo_sub_lock);
310                         if (r0->lo_sub[idx] == los) {
311                                 spin_unlock(&r0->lo_sub_lock);
312                                 schedule();
313                         } else {
314                                 spin_unlock(&r0->lo_sub_lock);
315                                 set_current_state(TASK_RUNNING);
316                                 break;
317                         }
318                 }
319                 remove_wait_queue(wq, waiter);
320         }
321         LASSERT(r0->lo_sub[idx] == NULL);
322 }
323
324 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
325                              struct lov_layout_entry *lle)
326 {
327         struct lov_layout_raid0 *r0 = &lle->lle_raid0;
328
329         ENTRY;
330
331         if (r0->lo_sub != NULL) {
332                 int i;
333
334                 for (i = 0; i < r0->lo_nr; ++i) {
335                         struct lovsub_object *los = r0->lo_sub[i];
336
337                         if (los != NULL) {
338                                 cl_object_prune(env, &los->lso_cl);
339                                 /*
340                                  * If top-level object is to be evicted from
341                                  * the cache, so are its sub-objects.
342                                  */
343                                 lov_subobject_kill(env, lov, r0, los, i);
344                         }
345                 }
346         }
347
348         EXIT;
349 }
350
351 static void lov_fini_raid0(const struct lu_env *env,
352                            struct lov_layout_entry *lle)
353 {
354         struct lov_layout_raid0 *r0 = &lle->lle_raid0;
355
356         if (r0->lo_sub != NULL) {
357                 OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
358                 r0->lo_sub = NULL;
359         }
360 }
361
362 static int lov_print_raid0(const struct lu_env *env, void *cookie,
363                            lu_printer_t p, const struct lov_layout_entry *lle)
364 {
365         const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
366         int i;
367
368         for (i = 0; i < r0->lo_nr; ++i) {
369                 struct lu_object *sub;
370
371                 if (r0->lo_sub[i] != NULL) {
372                         sub = lovsub2lu(r0->lo_sub[i]);
373                         lu_object_print(env, cookie, p, sub);
374                 } else {
375                         (*p)(env, cookie, "sub %d absent\n", i);
376                 }
377         }
378         return 0;
379 }
380
381 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
382                               unsigned int index, struct lov_layout_entry *lle,
383                               struct cl_attr **lov_attr)
384 {
385         struct lov_layout_raid0 *r0 = &lle->lle_raid0;
386         struct lov_stripe_md *lsm = lov->lo_lsm;
387         struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
388         struct cl_attr *attr = &r0->lo_attr;
389         __u64 kms = 0;
390         int result = 0;
391
392         if (r0->lo_attr_valid) {
393                 *lov_attr = attr;
394                 return 0;
395         }
396
397         memset(lvb, 0, sizeof(*lvb));
398
399         /* XXX: timestamps can be negative by sanity:test_39m,
400          * how can it be? */
401         lvb->lvb_atime = LLONG_MIN;
402         lvb->lvb_ctime = LLONG_MIN;
403         lvb->lvb_mtime = LLONG_MIN;
404
405         /*
406          * XXX that should be replaced with a loop over sub-objects,
407          * doing cl_object_attr_get() on them. But for now, let's
408          * reuse old lov code.
409          */
410
411         /*
412          * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
413          * happy. It's not needed, because new code uses
414          * ->coh_attr_guard spin-lock to protect consistency of
415          * sub-object attributes.
416          */
417         lov_stripe_lock(lsm);
418         result = lov_merge_lvb_kms(lsm, index, lvb, &kms);
419         lov_stripe_unlock(lsm);
420         if (result == 0) {
421                 cl_lvb2attr(attr, lvb);
422                 attr->cat_kms = kms;
423                 r0->lo_attr_valid = 1;
424                 *lov_attr = attr;
425         }
426
427         return result;
428 }
429
430 static struct lov_comp_layout_entry_ops raid0_ops = {
431         .lco_init      = lov_init_raid0,
432         .lco_fini      = lov_fini_raid0,
433         .lco_getattr   = lov_attr_get_raid0,
434 };
435
436 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
437                             unsigned int index, struct lov_layout_entry *lle,
438                             struct cl_attr **lov_attr)
439 {
440         struct lov_layout_dom *dom = &lle->lle_dom;
441         struct lov_oinfo *loi = dom->lo_loi;
442         struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
443
444         if (dom->lo_dom_r0.lo_attr_valid) {
445                 *lov_attr = attr;
446                 return 0;
447         }
448
449         if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
450                 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
451
452         cl_lvb2attr(attr, &loi->loi_lvb);
453
454         /* DoM component size can be bigger than stripe size after
455          * client's setattr RPC, so do not count anything beyond
456          * component end. Alternatively, check that limit on server
457          * and do not allow size overflow there. */
458         if (attr->cat_size > lle->lle_extent->e_end)
459                 attr->cat_size = lle->lle_extent->e_end;
460
461         attr->cat_kms = attr->cat_size;
462
463         dom->lo_dom_r0.lo_attr_valid = 1;
464         *lov_attr = attr;
465
466         return 0;
467 }
468
469 /**
470  * Lookup FLD to get MDS index of the given DOM object FID.
471  *
472  * \param[in]  ld       LOV device
473  * \param[in]  fid      FID to lookup
474  * \param[out] nr       index in MDC array to return back
475  *
476  * \retval              0 and \a mds filled with MDS index if successful
477  * \retval              negative value on error
478  */
479 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
480                           __u32 *nr)
481 {
482         __u32 mds_idx;
483         int i, rc;
484
485         ENTRY;
486
487         rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
488                                &mds_idx, LU_SEQ_RANGE_MDT, NULL);
489         if (rc) {
490                 CERROR("%s: error while looking for mds number. Seq %#llx"
491                        ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
492                        fid_seq(fid), rc);
493                 RETURN(rc);
494         }
495
496         CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
497                mds_idx, PFID(fid));
498
499         /* find proper MDC device in the array */
500         for (i = 0; i < ld->ld_md_tgts_nr; i++) {
501                 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
502                     ld->ld_md_tgts[i].ldm_idx == mds_idx)
503                         break;
504         }
505
506         if (i == ld->ld_md_tgts_nr) {
507                 CERROR("%s: cannot find corresponding MDC device for mds #%x "
508                        "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
509                        mds_idx, PFID(fid));
510                 rc = -EINVAL;
511         } else {
512                 *nr = i;
513         }
514         RETURN(rc);
515 }
516
517 /**
518  * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
519  *
520  * Init the DOM object for the first time. It prepares also RAID0 entry
521  * for it to use in common methods with ordinary RAID0 layout entries.
522  *
523  * \param[in] env       execution environment
524  * \param[in] dev       LOV device
525  * \param[in] lov       LOV object
526  * \param[in] index     Composite layout entry index in LSM
527  * \param[in] lle       Composite LOV layout entry
528  */
529 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
530                         struct lov_object *lov, unsigned int index,
531                         const struct cl_object_conf *conf,
532                         struct lov_layout_entry *lle)
533 {
534         struct lov_thread_info *lti = lov_env_info(env);
535         struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
536         struct cl_object *clo;
537         struct lu_object *o = lov2lu(lov);
538         const struct lu_fid *fid = lu_object_fid(o);
539         struct cl_device *mdcdev;
540         struct lov_oinfo *loi = NULL;
541         struct cl_object_conf *sconf = &lti->lti_stripe_conf;
542
543         int rc;
544         __u32 idx = 0;
545
546         ENTRY;
547
548         LASSERT(index == 0);
549
550         /* find proper MDS device */
551         rc = lov_fld_lookup(dev, fid, &idx);
552         if (rc)
553                 RETURN(rc);
554
555         LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
556                  "LOV md target[%u] is NULL\n", idx);
557
558         /* check lsm is DOM, more checks are needed */
559         LASSERT(lsme->lsme_stripe_count == 0);
560
561         /*
562          * Create lower cl_objects.
563          */
564         mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
565
566         LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
567
568         /* DoM object has no oinfo in LSM entry, create it exclusively */
569         OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
570         if (loi == NULL)
571                 RETURN(-ENOMEM);
572
573         fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
574
575         sconf->u.coc_oinfo = loi;
576 again:
577         clo = lov_sub_find(env, mdcdev, fid, sconf);
578         if (IS_ERR(clo))
579                 GOTO(out, rc = PTR_ERR(clo));
580
581         rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
582         if (rc == -EAGAIN) /* try again */
583                 goto again;
584         else if (rc != 0)
585                 GOTO(out, rc);
586
587         lle->lle_dom.lo_dom = cl2lovsub(clo);
588         spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
589         lle->lle_dom.lo_dom_r0.lo_nr = 1;
590         lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
591         lle->lle_dom.lo_loi = loi;
592
593         rc = lov_page_slice_fixup(lov, clo);
594         RETURN(rc);
595
596 out:
597         if (loi != NULL)
598                 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
599         return rc;
600 }
601
602 /**
603  * Implementation of lov_layout_operations::llo_fini for DOM object.
604  *
605  * Finish the DOM object and free related memory.
606  *
607  * \param[in] env       execution environment
608  * \param[in] lov       LOV object
609  * \param[in] state     LOV layout state
610  */
611 static void lov_fini_dom(const struct lu_env *env,
612                          struct lov_layout_entry *lle)
613 {
614         if (lle->lle_dom.lo_dom != NULL)
615                 lle->lle_dom.lo_dom = NULL;
616         if (lle->lle_dom.lo_loi != NULL)
617                 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
618 }
619
620 static struct lov_comp_layout_entry_ops dom_ops = {
621         .lco_init = lov_init_dom,
622         .lco_fini = lov_fini_dom,
623         .lco_getattr = lov_attr_get_dom,
624 };
625
626 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
627                               struct lov_object *lov, struct lov_stripe_md *lsm,
628                               const struct cl_object_conf *conf,
629                               union lov_layout_state *state)
630 {
631         struct lov_layout_composite *comp = &state->composite;
632         struct lov_layout_entry *lle;
633         struct lov_mirror_entry *lre;
634         unsigned int entry_count;
635         unsigned int psz = 0;
636         unsigned int mirror_count;
637         int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
638         int result = 0;
639         unsigned int seq;
640         int i, j;
641
642         ENTRY;
643
644         LASSERT(lsm->lsm_entry_count > 0);
645         LASSERT(lov->lo_lsm == NULL);
646         lov->lo_lsm = lsm_addref(lsm);
647         lov->lo_layout_invalid = true;
648
649         dump_lsm(D_INODE, lsm);
650
651         entry_count = lsm->lsm_entry_count;
652
653         spin_lock_init(&comp->lo_write_lock);
654         comp->lo_flags = lsm->lsm_flags;
655         comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
656         comp->lo_entry_count = lsm->lsm_entry_count;
657         comp->lo_preferred_mirror = -1;
658
659         if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
660                 RETURN(-EINVAL);
661
662         OBD_ALLOC(comp->lo_mirrors,
663                   comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
664         if (comp->lo_mirrors == NULL)
665                 RETURN(-ENOMEM);
666
667         OBD_ALLOC(comp->lo_entries, entry_count * sizeof(*comp->lo_entries));
668         if (comp->lo_entries == NULL)
669                 RETURN(-ENOMEM);
670
671         /* Initiate all entry types and extents data at first */
672         for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
673                 int mirror_id = 0;
674
675                 lle = &comp->lo_entries[i];
676
677                 lle->lle_lsme = lsm->lsm_entries[i];
678                 lle->lle_type = lov_entry_type(lle->lle_lsme);
679                 switch (lle->lle_type) {
680                 case LOV_PATTERN_RAID0:
681                         lle->lle_comp_ops = &raid0_ops;
682                         break;
683                 case LOV_PATTERN_MDT:
684                         lle->lle_comp_ops = &dom_ops;
685                         break;
686                 default:
687                         CERROR("%s: unknown composite layout entry type %i\n",
688                                lov2obd(dev->ld_lov)->obd_name,
689                                lsm->lsm_entries[i]->lsme_pattern);
690                         dump_lsm(D_ERROR, lsm);
691                         RETURN(-EIO);
692                 }
693
694                 lle->lle_extent = &lle->lle_lsme->lsme_extent;
695                 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
696
697                 if (flr_state != LCM_FL_NONE)
698                         mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
699
700                 lre = &comp->lo_mirrors[j];
701                 if (i > 0) {
702                         if (mirror_id == lre->lre_mirror_id) {
703                                 lre->lre_valid |= lle->lle_valid;
704                                 lre->lre_stale |= !lle->lle_valid;
705                                 lre->lre_end = i;
706                                 continue;
707                         }
708
709                         /* new mirror detected, assume that the mirrors
710                          * are shorted in layout */
711                         ++mirror_count;
712                         ++j;
713                         if (j >= comp->lo_mirror_count)
714                                 break;
715
716                         lre = &comp->lo_mirrors[j];
717                 }
718
719                 /* entries must be sorted by mirrors */
720                 lre->lre_mirror_id = mirror_id;
721                 lre->lre_start = lre->lre_end = i;
722                 lre->lre_preferred = !!(lle->lle_lsme->lsme_flags &
723                                         LCME_FL_PREF_RD);
724                 lre->lre_valid = lle->lle_valid;
725                 lre->lre_stale = !lle->lle_valid;
726         }
727
728         /* sanity check for FLR */
729         if (mirror_count != comp->lo_mirror_count) {
730                 CDEBUG(D_INODE, DFID
731                        " doesn't have the # of mirrors it claims, %u/%u\n",
732                        PFID(lu_object_fid(lov2lu(lov))), mirror_count,
733                        comp->lo_mirror_count + 1);
734
735                 GOTO(out, result = -EINVAL);
736         }
737
738         lov_foreach_layout_entry(lov, lle) {
739                 int index = lov_layout_entry_index(lov, lle);
740
741                 /**
742                  * If the component has not been init-ed on MDS side, for
743                  * PFL layout, we'd know that the components beyond this one
744                  * will be dynamically init-ed later on file write/trunc ops.
745                  */
746                 if (!lsme_inited(lle->lle_lsme))
747                         continue;
748
749                 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
750                                                      conf, lle);
751                 if (result < 0)
752                         break;
753
754                 LASSERT(ergo(psz > 0, psz == result));
755                 psz = result;
756         }
757
758         if (psz > 0)
759                 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
760
761         /* decide the preferred mirror. It uses the hash value of lov_object
762          * so that different clients would use different mirrors for read. */
763         mirror_count = 0;
764         seq = hash_long((unsigned long)lov, 8);
765         for (i = 0; i < comp->lo_mirror_count; i++) {
766                 unsigned int idx = (i + seq) % comp->lo_mirror_count;
767
768                 lre = lov_mirror_entry(lov, idx);
769                 if (lre->lre_stale)
770                         continue;
771
772                 mirror_count++; /* valid mirror */
773
774                 if (lre->lre_preferred || comp->lo_preferred_mirror < 0)
775                         comp->lo_preferred_mirror = idx;
776         }
777         if (!mirror_count) {
778                 CDEBUG(D_INODE, DFID
779                        " doesn't have any valid mirrors\n",
780                        PFID(lu_object_fid(lov2lu(lov))));
781
782                 comp->lo_preferred_mirror = 0;
783         }
784
785         LASSERT(comp->lo_preferred_mirror >= 0);
786
787         EXIT;
788 out:
789         return result > 0 ? 0 : result;
790 }
791
792 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
793                           struct lov_object *lov, struct lov_stripe_md *lsm,
794                           const struct cl_object_conf *conf,
795                           union lov_layout_state *state)
796 {
797         return 0;
798 }
799
800 static int lov_init_released(const struct lu_env *env,
801                              struct lov_device *dev, struct lov_object *lov,
802                              struct lov_stripe_md *lsm,
803                              const struct cl_object_conf *conf,
804                              union lov_layout_state *state)
805 {
806         LASSERT(lsm != NULL);
807         LASSERT(lsm->lsm_is_released);
808         LASSERT(lov->lo_lsm == NULL);
809
810         lov->lo_lsm = lsm_addref(lsm);
811         return 0;
812 }
813
814 static int lov_init_foreign(const struct lu_env *env,
815                             struct lov_device *dev, struct lov_object *lov,
816                             struct lov_stripe_md *lsm,
817                             const struct cl_object_conf *conf,
818                             union lov_layout_state *state)
819 {
820         LASSERT(lsm != NULL);
821         LASSERT(lov->lo_type == LLT_FOREIGN);
822         LASSERT(lov->lo_lsm == NULL);
823
824         lov->lo_lsm = lsm_addref(lsm);
825         return 0;
826 }
827
828 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
829                             union lov_layout_state *state)
830 {
831         LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED ||
832                 lov->lo_type == LLT_FOREIGN);
833
834         lov_layout_wait(env, lov);
835         return 0;
836 }
837
838 static int lov_delete_composite(const struct lu_env *env,
839                                 struct lov_object *lov,
840                                 union lov_layout_state *state)
841 {
842         struct lov_layout_entry *entry;
843         struct lov_layout_composite *comp = &state->composite;
844
845         ENTRY;
846
847         dump_lsm(D_INODE, lov->lo_lsm);
848
849         lov_layout_wait(env, lov);
850         if (comp->lo_entries)
851                 lov_foreach_layout_entry(lov, entry)
852                         lov_delete_raid0(env, lov, entry);
853
854         RETURN(0);
855 }
856
857 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
858                            union lov_layout_state *state)
859 {
860         LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
861 }
862
863 static void lov_fini_composite(const struct lu_env *env,
864                                struct lov_object *lov,
865                                union lov_layout_state *state)
866 {
867         struct lov_layout_composite *comp = &state->composite;
868         ENTRY;
869
870         if (comp->lo_entries != NULL) {
871                 struct lov_layout_entry *entry;
872
873                 lov_foreach_layout_entry(lov, entry)
874                         entry->lle_comp_ops->lco_fini(env, entry);
875
876                 OBD_FREE(comp->lo_entries,
877                          comp->lo_entry_count * sizeof(*comp->lo_entries));
878                 comp->lo_entries = NULL;
879         }
880
881         if (comp->lo_mirrors != NULL) {
882                 OBD_FREE(comp->lo_mirrors,
883                          comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
884                 comp->lo_mirrors = NULL;
885         }
886
887         memset(comp, 0, sizeof(*comp));
888
889         dump_lsm(D_INODE, lov->lo_lsm);
890         lov_free_memmd(&lov->lo_lsm);
891
892         EXIT;
893 }
894
895 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
896                                 union lov_layout_state *state)
897 {
898         ENTRY;
899         dump_lsm(D_INODE, lov->lo_lsm);
900         lov_free_memmd(&lov->lo_lsm);
901         EXIT;
902 }
903
904 static int lov_print_empty(const struct lu_env *env, void *cookie,
905                            lu_printer_t p, const struct lu_object *o)
906 {
907         (*p)(env, cookie, "empty %d\n", lu2lov(o)->lo_layout_invalid);
908         return 0;
909 }
910
911 static int lov_print_composite(const struct lu_env *env, void *cookie,
912                                lu_printer_t p, const struct lu_object *o)
913 {
914         struct lov_object *lov = lu2lov(o);
915         struct lov_stripe_md *lsm = lov->lo_lsm;
916         int i;
917
918         (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
919              lsm->lsm_entry_count,
920              lov->lo_layout_invalid ? "invalid" : "valid", lsm,
921              lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
922              lsm->lsm_layout_gen);
923
924         for (i = 0; i < lsm->lsm_entry_count; i++) {
925                 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
926                 struct lov_layout_entry *lle = lov_entry(lov, i);
927
928                 (*p)(env, cookie,
929                      DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
930                      PEXT(&lse->lsme_extent), lse->lsme_magic,
931                      lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
932                      lse->lsme_flags, lse->lsme_stripe_count,
933                      lse->lsme_stripe_size);
934                 lov_print_raid0(env, cookie, p, lle);
935         }
936
937         return 0;
938 }
939
940 static int lov_print_released(const struct lu_env *env, void *cookie,
941                                 lu_printer_t p, const struct lu_object *o)
942 {
943         struct lov_object       *lov = lu2lov(o);
944         struct lov_stripe_md    *lsm = lov->lo_lsm;
945
946         (*p)(env, cookie,
947                 "released: %s, lsm{%p 0x%08X %d %u}:\n",
948                 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
949                 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
950                 lsm->lsm_layout_gen);
951         return 0;
952 }
953
954 static int lov_print_foreign(const struct lu_env *env, void *cookie,
955                                 lu_printer_t p, const struct lu_object *o)
956 {
957         struct lov_object       *lov = lu2lov(o);
958         struct lov_stripe_md    *lsm = lov->lo_lsm;
959
960         (*p)(env, cookie,
961                 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
962                 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
963                 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
964                 lsm->lsm_layout_gen);
965         (*p)(env, cookie,
966                 "raw_ea_content '%.*s'\n",
967                 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
968         return 0;
969 }
970
971 /**
972  * Implements cl_object_operations::coo_attr_get() method for an object
973  * without stripes (LLT_EMPTY layout type).
974  *
975  * The only attributes this layer is authoritative in this case is
976  * cl_attr::cat_blocks---it's 0.
977  */
978 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
979                               struct cl_attr *attr)
980 {
981         attr->cat_blocks = 0;
982         return 0;
983 }
984
985 static int lov_attr_get_composite(const struct lu_env *env,
986                                   struct cl_object *obj,
987                                   struct cl_attr *attr)
988 {
989         struct lov_object       *lov = cl2lov(obj);
990         struct lov_layout_entry *entry;
991         int                      result = 0;
992
993         ENTRY;
994
995         attr->cat_size = 0;
996         attr->cat_blocks = 0;
997         lov_foreach_layout_entry(lov, entry) {
998                 struct cl_attr *lov_attr = NULL;
999                 int index = lov_layout_entry_index(lov, entry);
1000
1001                 if (!entry->lle_valid)
1002                         continue;
1003
1004                 /* PFL: This component has not been init-ed. */
1005                 if (!lsm_entry_inited(lov->lo_lsm, index))
1006                         continue;
1007
1008                 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1009                                                           entry, &lov_attr);
1010                 if (result < 0)
1011                         RETURN(result);
1012
1013                 if (lov_attr == NULL)
1014                         continue;
1015
1016                 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1017                        "b=%llu\n", index - 1, lov_attr->cat_size,
1018                        lov_attr->cat_mtime, lov_attr->cat_atime,
1019                        lov_attr->cat_ctime, lov_attr->cat_blocks);
1020
1021                 /* merge results */
1022                 attr->cat_blocks += lov_attr->cat_blocks;
1023                 if (attr->cat_size < lov_attr->cat_size)
1024                         attr->cat_size = lov_attr->cat_size;
1025                 if (attr->cat_kms < lov_attr->cat_kms)
1026                         attr->cat_kms = lov_attr->cat_kms;
1027                 if (attr->cat_atime < lov_attr->cat_atime)
1028                         attr->cat_atime = lov_attr->cat_atime;
1029                 if (attr->cat_ctime < lov_attr->cat_ctime)
1030                         attr->cat_ctime = lov_attr->cat_ctime;
1031                 if (attr->cat_mtime < lov_attr->cat_mtime)
1032                         attr->cat_mtime = lov_attr->cat_mtime;
1033         }
1034
1035         RETURN(0);
1036 }
1037
1038 const static struct lov_layout_operations lov_dispatch[] = {
1039         [LLT_EMPTY] = {
1040                 .llo_init      = lov_init_empty,
1041                 .llo_delete    = lov_delete_empty,
1042                 .llo_fini      = lov_fini_empty,
1043                 .llo_print     = lov_print_empty,
1044                 .llo_page_init = lov_page_init_empty,
1045                 .llo_lock_init = lov_lock_init_empty,
1046                 .llo_io_init   = lov_io_init_empty,
1047                 .llo_getattr   = lov_attr_get_empty,
1048         },
1049         [LLT_RELEASED] = {
1050                 .llo_init      = lov_init_released,
1051                 .llo_delete    = lov_delete_empty,
1052                 .llo_fini      = lov_fini_released,
1053                 .llo_print     = lov_print_released,
1054                 .llo_page_init = lov_page_init_empty,
1055                 .llo_lock_init = lov_lock_init_empty,
1056                 .llo_io_init   = lov_io_init_released,
1057                 .llo_getattr   = lov_attr_get_empty,
1058         },
1059         [LLT_COMP] = {
1060                 .llo_init      = lov_init_composite,
1061                 .llo_delete    = lov_delete_composite,
1062                 .llo_fini      = lov_fini_composite,
1063                 .llo_print     = lov_print_composite,
1064                 .llo_page_init = lov_page_init_composite,
1065                 .llo_lock_init = lov_lock_init_composite,
1066                 .llo_io_init   = lov_io_init_composite,
1067                 .llo_getattr   = lov_attr_get_composite,
1068         },
1069         [LLT_FOREIGN] = {
1070                 .llo_init      = lov_init_foreign,
1071                 .llo_delete    = lov_delete_empty,
1072                 .llo_fini      = lov_fini_released,
1073                 .llo_print     = lov_print_foreign,
1074                 .llo_page_init = lov_page_init_foreign,
1075                 .llo_lock_init = lov_lock_init_empty,
1076                 .llo_io_init   = lov_io_init_empty,
1077                 .llo_getattr   = lov_attr_get_empty,
1078         },
1079 };
1080
1081 /**
1082  * Performs a double-dispatch based on the layout type of an object.
1083  */
1084 #define LOV_2DISPATCH_NOLOCK(obj, op, ...)              \
1085 ({                                                      \
1086         struct lov_object *__obj = (obj);               \
1087         enum lov_layout_type __llt;                     \
1088                                                         \
1089         __llt = __obj->lo_type;                         \
1090         LASSERT(__llt < ARRAY_SIZE(lov_dispatch));      \
1091         lov_dispatch[__llt].op(__VA_ARGS__);            \
1092 })
1093
1094 /**
1095  * Return lov_layout_type associated with a given lsm
1096  */
1097 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1098 {
1099         if (lsm == NULL)
1100                 return LLT_EMPTY;
1101
1102         if (lsm->lsm_is_released)
1103                 return LLT_RELEASED;
1104
1105         if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1106             lsm->lsm_magic == LOV_MAGIC_V3 ||
1107             lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1108                 return LLT_COMP;
1109
1110         if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1111                 return LLT_FOREIGN;
1112
1113         return LLT_EMPTY;
1114 }
1115
1116 static inline void lov_conf_freeze(struct lov_object *lov)
1117 {
1118         CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1119                 lov, lov->lo_owner, current);
1120         if (lov->lo_owner != current)
1121                 down_read(&lov->lo_type_guard);
1122 }
1123
1124 static inline void lov_conf_thaw(struct lov_object *lov)
1125 {
1126         CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1127                 lov, lov->lo_owner, current);
1128         if (lov->lo_owner != current)
1129                 up_read(&lov->lo_type_guard);
1130 }
1131
1132 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...)                       \
1133 ({                                                                      \
1134         struct lov_object                      *__obj = (obj);          \
1135         int                                     __lock = !!(lock);      \
1136         typeof(lov_dispatch[0].op(__VA_ARGS__)) __result;               \
1137                                                                         \
1138         if (__lock)                                                     \
1139                 lov_conf_freeze(__obj);                                 \
1140         __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__);          \
1141         if (__lock)                                                     \
1142                 lov_conf_thaw(__obj);                                   \
1143         __result;                                                       \
1144 })
1145
1146 /**
1147  * Performs a locked double-dispatch based on the layout type of an object.
1148  */
1149 #define LOV_2DISPATCH(obj, op, ...)                     \
1150         LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1151
1152 #define LOV_2DISPATCH_VOID(obj, op, ...)                                \
1153 do {                                                                    \
1154         struct lov_object                      *__obj = (obj);          \
1155         enum lov_layout_type                    __llt;                  \
1156                                                                         \
1157         lov_conf_freeze(__obj);                                         \
1158         __llt = __obj->lo_type;                                         \
1159         LASSERT(__llt < ARRAY_SIZE(lov_dispatch));                      \
1160         lov_dispatch[__llt].op(__VA_ARGS__);                            \
1161         lov_conf_thaw(__obj);                                           \
1162 } while (0)
1163
1164 static void lov_conf_lock(struct lov_object *lov)
1165 {
1166         LASSERT(lov->lo_owner != current);
1167         down_write(&lov->lo_type_guard);
1168         LASSERT(lov->lo_owner == NULL);
1169         lov->lo_owner = current;
1170         CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1171                 lov, lov->lo_owner);
1172 }
1173
1174 static void lov_conf_unlock(struct lov_object *lov)
1175 {
1176         CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1177                 lov, lov->lo_owner);
1178         lov->lo_owner = NULL;
1179         up_write(&lov->lo_type_guard);
1180 }
1181
1182 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1183 {
1184         struct l_wait_info lwi = { 0 };
1185         ENTRY;
1186
1187         while (atomic_read(&lov->lo_active_ios) > 0) {
1188                 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1189                         PFID(lu_object_fid(lov2lu(lov))),
1190                         atomic_read(&lov->lo_active_ios));
1191
1192                 l_wait_event(lov->lo_waitq,
1193                              atomic_read(&lov->lo_active_ios) == 0, &lwi);
1194         }
1195         RETURN(0);
1196 }
1197
1198 static int lov_layout_change(const struct lu_env *unused,
1199                              struct lov_object *lov, struct lov_stripe_md *lsm,
1200                              const struct cl_object_conf *conf)
1201 {
1202         enum lov_layout_type llt = lov_type(lsm);
1203         union lov_layout_state *state = &lov->u;
1204         const struct lov_layout_operations *old_ops;
1205         const struct lov_layout_operations *new_ops;
1206         struct lov_device *lov_dev = lov_object_dev(lov);
1207         struct lu_env *env;
1208         __u16 refcheck;
1209         int rc;
1210         ENTRY;
1211
1212         LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1213
1214         env = cl_env_get(&refcheck);
1215         if (IS_ERR(env))
1216                 RETURN(PTR_ERR(env));
1217
1218         LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1219
1220         CDEBUG(D_INODE, DFID" from %s to %s\n",
1221                PFID(lu_object_fid(lov2lu(lov))),
1222                llt2str(lov->lo_type), llt2str(llt));
1223
1224         old_ops = &lov_dispatch[lov->lo_type];
1225         new_ops = &lov_dispatch[llt];
1226
1227         rc = cl_object_prune(env, &lov->lo_cl);
1228         if (rc != 0)
1229                 GOTO(out, rc);
1230
1231         rc = old_ops->llo_delete(env, lov, &lov->u);
1232         if (rc != 0)
1233                 GOTO(out, rc);
1234
1235         old_ops->llo_fini(env, lov, &lov->u);
1236
1237         LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1238
1239         CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1240                PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1241
1242         /* page bufsize fixup */
1243         cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1244                 lov_page_slice_fixup(lov, NULL);
1245
1246         lov->lo_type = llt;
1247         rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1248         if (rc != 0) {
1249                 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1250
1251                 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1252                        obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1253                 new_ops->llo_delete(env, lov, state);
1254                 new_ops->llo_fini(env, lov, state);
1255                 /* this file becomes an EMPTY file. */
1256                 lov->lo_type = LLT_EMPTY;
1257                 GOTO(out, rc);
1258         }
1259
1260 out:
1261         cl_env_put(env, &refcheck);
1262         RETURN(rc);
1263 }
1264
1265 /*****************************************************************************
1266  *
1267  * Lov object operations.
1268  *
1269  */
1270 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1271                     const struct lu_object_conf *conf)
1272 {
1273         struct lov_object            *lov   = lu2lov(obj);
1274         struct lov_device            *dev   = lov_object_dev(lov);
1275         const struct cl_object_conf  *cconf = lu2cl_conf(conf);
1276         union lov_layout_state       *set   = &lov->u;
1277         const struct lov_layout_operations *ops;
1278         struct lov_stripe_md *lsm = NULL;
1279         int rc;
1280         ENTRY;
1281
1282         init_rwsem(&lov->lo_type_guard);
1283         atomic_set(&lov->lo_active_ios, 0);
1284         init_waitqueue_head(&lov->lo_waitq);
1285         cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
1286
1287         lov->lo_type = LLT_EMPTY;
1288         if (cconf->u.coc_layout.lb_buf != NULL) {
1289                 lsm = lov_unpackmd(dev->ld_lov,
1290                                    cconf->u.coc_layout.lb_buf,
1291                                    cconf->u.coc_layout.lb_len);
1292                 if (IS_ERR(lsm))
1293                         RETURN(PTR_ERR(lsm));
1294
1295                 dump_lsm(D_INODE, lsm);
1296         }
1297
1298         /* no locking is necessary, as object is being created */
1299         lov->lo_type = lov_type(lsm);
1300         ops = &lov_dispatch[lov->lo_type];
1301         rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1302         if (rc != 0)
1303                 GOTO(out_lsm, rc);
1304
1305 out_lsm:
1306         lov_lsm_put(lsm);
1307
1308         RETURN(rc);
1309 }
1310
1311 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1312                         const struct cl_object_conf *conf)
1313 {
1314         struct lov_stripe_md    *lsm = NULL;
1315         struct lov_object       *lov = cl2lov(obj);
1316         int                      result = 0;
1317         ENTRY;
1318
1319         if (conf->coc_opc == OBJECT_CONF_SET &&
1320             conf->u.coc_layout.lb_buf != NULL) {
1321                 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1322                                    conf->u.coc_layout.lb_buf,
1323                                    conf->u.coc_layout.lb_len);
1324                 if (IS_ERR(lsm))
1325                         RETURN(PTR_ERR(lsm));
1326                 dump_lsm(D_INODE, lsm);
1327         }
1328
1329         lov_conf_lock(lov);
1330         if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1331                 lov->lo_layout_invalid = true;
1332                 GOTO(out, result = 0);
1333         }
1334
1335         if (conf->coc_opc == OBJECT_CONF_WAIT) {
1336                 if (lov->lo_layout_invalid &&
1337                     atomic_read(&lov->lo_active_ios) > 0) {
1338                         lov_conf_unlock(lov);
1339                         result = lov_layout_wait(env, lov);
1340                         lov_conf_lock(lov);
1341                 }
1342                 GOTO(out, result);
1343         }
1344
1345         LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1346
1347         if ((lsm == NULL && lov->lo_lsm == NULL) ||
1348             ((lsm != NULL && lov->lo_lsm != NULL) &&
1349              (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1350              (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1351               lsm->lsm_entries[0]->lsme_pattern))) {
1352                 /* same version of layout */
1353                 lov->lo_layout_invalid = false;
1354                 GOTO(out, result = 0);
1355         }
1356
1357         /* will change layout - check if there still exists active IO. */
1358         if (atomic_read(&lov->lo_active_ios) > 0) {
1359                 lov->lo_layout_invalid = true;
1360                 GOTO(out, result = -EBUSY);
1361         }
1362
1363         result = lov_layout_change(env, lov, lsm, conf);
1364         lov->lo_layout_invalid = result != 0;
1365         EXIT;
1366
1367 out:
1368         lov_conf_unlock(lov);
1369         lov_lsm_put(lsm);
1370         CDEBUG(D_INODE, DFID" lo_layout_invalid=%d\n",
1371                PFID(lu_object_fid(lov2lu(lov))), lov->lo_layout_invalid);
1372         RETURN(result);
1373 }
1374
1375 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1376 {
1377         struct lov_object *lov = lu2lov(obj);
1378
1379         ENTRY;
1380         LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1381         EXIT;
1382 }
1383
1384 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1385 {
1386         struct lov_object *lov = lu2lov(obj);
1387
1388         ENTRY;
1389         LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1390         lu_object_fini(obj);
1391         OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1392         EXIT;
1393 }
1394
1395 static int lov_object_print(const struct lu_env *env, void *cookie,
1396                             lu_printer_t p, const struct lu_object *o)
1397 {
1398         return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1399 }
1400
1401 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1402                   struct cl_page *page, pgoff_t index)
1403 {
1404         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1405                                     index);
1406 }
1407
1408 /**
1409  * Implements cl_object_operations::clo_io_init() method for lov
1410  * layer. Dispatches to the appropriate layout io initialization method.
1411  */
1412 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1413                 struct cl_io *io)
1414 {
1415         CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1416
1417         CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1418                PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1419                io->ci_ignore_layout, io->ci_verify_layout);
1420
1421         /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1422          * the OSC layer. It shouldn't take lov layout conf lock in that case,
1423          * because as long as the OSC object exists, the layout can't be
1424          * reconfigured. */
1425         return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1426                         !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1427                         env, obj, io);
1428 }
1429
1430 /**
1431  * An implementation of cl_object_operations::clo_attr_get() method for lov
1432  * layer. For raid0 layout this collects and merges attributes of all
1433  * sub-objects.
1434  */
1435 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1436                         struct cl_attr *attr)
1437 {
1438         /* do not take lock, as this function is called under a
1439          * spin-lock. Layout is protected from changing by ongoing IO. */
1440         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1441 }
1442
1443 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1444                            const struct cl_attr *attr, unsigned valid)
1445 {
1446         /*
1447          * No dispatch is required here, as no layout implements this.
1448          */
1449         return 0;
1450 }
1451
1452 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1453                   struct cl_lock *lock, const struct cl_io *io)
1454 {
1455         /* No need to lock because we've taken one refcount of layout.  */
1456         return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1457                                     io);
1458 }
1459
1460 /**
1461  * We calculate on which OST the mapping will end. If the length of mapping
1462  * is greater than (stripe_size * stripe_count) then the last_stripe will
1463  * will be one just before start_stripe. Else we check if the mapping
1464  * intersects each OST and find last_stripe.
1465  * This function returns the last_stripe and also sets the stripe_count
1466  * over which the mapping is spread
1467  *
1468  * \param lsm [in]              striping information for the file
1469  * \param index [in]            stripe component index
1470  * \param ext [in]              logical extent of mapping
1471  * \param start_stripe [in]     starting stripe of the mapping
1472  * \param stripe_count [out]    the number of stripes across which to map is
1473  *                              returned
1474  *
1475  * \retval last_stripe          return the last stripe of the mapping
1476  */
1477 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1478                                    struct lu_extent *ext,
1479                                    int start_stripe, int *stripe_count)
1480 {
1481         struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1482         int last_stripe;
1483         u64 obd_start;
1484         u64 obd_end;
1485         int i, j;
1486
1487         if (ext->e_end - ext->e_start >
1488             lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1489                 last_stripe = (start_stripe < 1 ? lsme->lsme_stripe_count - 1 :
1490                                                   start_stripe - 1);
1491                 *stripe_count = lsme->lsme_stripe_count;
1492         } else {
1493                 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1494                      i = (i + 1) % lsme->lsme_stripe_count, j++) {
1495                         if ((lov_stripe_intersects(lsm, index,  i, ext,
1496                                                    &obd_start, &obd_end)) == 0)
1497                                 break;
1498                 }
1499                 *stripe_count = j;
1500                 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1501         }
1502
1503         return last_stripe;
1504 }
1505
1506 /**
1507  * Set fe_device and copy extents from local buffer into main return buffer.
1508  *
1509  * \param fiemap [out]          fiemap to hold all extents
1510  * \param lcl_fm_ext [in]       array of fiemap extents get from OSC layer
1511  * \param ost_index [in]        OST index to be written into the fm_device
1512  *                              field for each extent
1513  * \param ext_count [in]        number of extents to be copied
1514  * \param current_extent [in]   where to start copying in the extent array
1515  */
1516 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1517                                          struct fiemap_extent *lcl_fm_ext,
1518                                          int ost_index, unsigned int ext_count,
1519                                          int current_extent)
1520 {
1521         char            *to;
1522         unsigned int    ext;
1523
1524         for (ext = 0; ext < ext_count; ext++) {
1525                 lcl_fm_ext[ext].fe_device = ost_index;
1526                 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1527         }
1528
1529         /* Copy fm_extent's from fm_local to return buffer */
1530         to = (char *)fiemap + fiemap_count_to_size(current_extent);
1531         memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1532 }
1533
1534 #define FIEMAP_BUFFER_SIZE 4096
1535
1536 /**
1537  * Non-zero fe_logical indicates that this is a continuation FIEMAP
1538  * call. The local end offset and the device are sent in the first
1539  * fm_extent. This function calculates the stripe number from the index.
1540  * This function returns a stripe_no on which mapping is to be restarted.
1541  *
1542  * This function returns fm_end_offset which is the in-OST offset at which
1543  * mapping should be restarted. If fm_end_offset=0 is returned then caller
1544  * will re-calculate proper offset in next stripe.
1545  * Note that the first extent is passed to lov_get_info via the value field.
1546  *
1547  * \param fiemap [in]           fiemap request header
1548  * \param lsm [in]              striping information for the file
1549  * \param index [in]            stripe component index
1550  * \param ext [in]              logical extent of mapping
1551  * \param start_stripe [out]    starting stripe will be returned in this
1552  */
1553 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1554                                      struct lov_stripe_md *lsm,
1555                                      int index, struct lu_extent *ext,
1556                                      int *start_stripe)
1557 {
1558         struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1559         u64 local_end = fiemap->fm_extents[0].fe_logical;
1560         u64 lun_start;
1561         u64 lun_end;
1562         u64 fm_end_offset;
1563         int stripe_no = -1;
1564         int i;
1565
1566         if (fiemap->fm_extent_count == 0 ||
1567             fiemap->fm_extents[0].fe_logical == 0)
1568                 return 0;
1569
1570         /* Find out stripe_no from ost_index saved in the fe_device */
1571         for (i = 0; i < lsme->lsme_stripe_count; i++) {
1572                 struct lov_oinfo *oinfo = lsme->lsme_oinfo[i];
1573
1574                 if (lov_oinfo_is_dummy(oinfo))
1575                         continue;
1576
1577                 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1578                         stripe_no = i;
1579                         break;
1580                 }
1581         }
1582
1583         if (stripe_no == -1)
1584                 return -EINVAL;
1585
1586         /* If we have finished mapping on previous device, shift logical
1587          * offset to start of next device */
1588         if (lov_stripe_intersects(lsm, index, stripe_no, ext,
1589                                    &lun_start, &lun_end) != 0 &&
1590             local_end < lun_end) {
1591                 fm_end_offset = local_end;
1592                 *start_stripe = stripe_no;
1593         } else {
1594                 /* This is a special value to indicate that caller should
1595                  * calculate offset in next stripe. */
1596                 fm_end_offset = 0;
1597                 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1598         }
1599
1600         return fm_end_offset;
1601 }
1602
1603 struct fiemap_state {
1604         struct fiemap           *fs_fm;
1605         struct lu_extent        fs_ext;
1606         u64                     fs_length;
1607         u64                     fs_end_offset;
1608         int                     fs_cur_extent;
1609         int                     fs_cnt_need;
1610         int                     fs_start_stripe;
1611         int                     fs_last_stripe;
1612         bool                    fs_device_done;
1613         bool                    fs_finish_stripe;
1614         bool                    fs_enough;
1615 };
1616
1617 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1618                                          struct lov_object *lov,
1619                                          struct lov_stripe_md *lsm,
1620                                          int index)
1621 {
1622         struct lov_device       *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1623         struct lov_thread_info  *lti = lov_env_info(env);
1624         struct lu_fid           *ofid = &lti->lti_fid;
1625         struct lov_oinfo        *oinfo;
1626         struct cl_device        *subdev;
1627         int                     entry = lov_comp_entry(index);
1628         int                     stripe = lov_comp_stripe(index);
1629         int                     ost_idx;
1630         int                     rc;
1631         struct cl_object        *result;
1632
1633         if (lov->lo_type != LLT_COMP)
1634                 GOTO(out, result = NULL);
1635
1636         if (entry >= lsm->lsm_entry_count ||
1637             stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1638                 GOTO(out, result = NULL);
1639
1640         oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1641         ost_idx = oinfo->loi_ost_idx;
1642         rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1643         if (rc != 0)
1644                 GOTO(out, result = NULL);
1645
1646         subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1647         result = lov_sub_find(env, subdev, ofid, NULL);
1648 out:
1649         if (result == NULL)
1650                 result = ERR_PTR(-EINVAL);
1651         return result;
1652 }
1653
1654 int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1655                       struct lov_stripe_md *lsm, struct fiemap *fiemap,
1656                       size_t *buflen, struct ll_fiemap_info_key *fmkey,
1657                       int index, int stripeno, struct fiemap_state *fs)
1658 {
1659         struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1660         struct cl_object *subobj;
1661         struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1662         struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1663         u64 req_fm_len; /* Stores length of required mapping */
1664         u64 len_mapped_single_call;
1665         u64 lun_start;
1666         u64 lun_end;
1667         u64 obd_object_end;
1668         unsigned int ext_count;
1669         /* EOF for object */
1670         bool ost_eof = false;
1671         /* done with required mapping for this OST? */
1672         bool ost_done = false;
1673         int ost_index;
1674         int rc = 0;
1675
1676         fs->fs_device_done = false;
1677         /* Find out range of mapping on this stripe */
1678         if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1679                                    &lun_start, &obd_object_end)) == 0)
1680                 return 0;
1681
1682         if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1683                 return -EIO;
1684
1685         /* If this is a continuation FIEMAP call and we are on
1686          * starting stripe then lun_start needs to be set to
1687          * end_offset */
1688         if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1689                 lun_start = fs->fs_end_offset;
1690         lun_end = lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno);
1691         if (lun_start == lun_end)
1692                 return 0;
1693
1694         req_fm_len = obd_object_end - lun_start + 1;
1695         fs->fs_fm->fm_length = 0;
1696         len_mapped_single_call = 0;
1697
1698         /* find lobsub object */
1699         subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1700                                  lov_comp_index(index, stripeno));
1701         if (IS_ERR(subobj))
1702                 return PTR_ERR(subobj);
1703         /* If the output buffer is very large and the objects have many
1704          * extents we may need to loop on a single OST repeatedly */
1705         do {
1706                 if (fiemap->fm_extent_count > 0) {
1707                         /* Don't get too many extents. */
1708                         if (fs->fs_cur_extent + fs->fs_cnt_need >
1709                             fiemap->fm_extent_count)
1710                                 fs->fs_cnt_need = fiemap->fm_extent_count -
1711                                                   fs->fs_cur_extent;
1712                 }
1713
1714                 lun_start += len_mapped_single_call;
1715                 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1716                 req_fm_len = fs->fs_fm->fm_length;
1717                 /**
1718                  * If we've collected enough extent map, we'd request 1 more,
1719                  * to see whether we coincidentally finished all available
1720                  * extent map, so that FIEMAP_EXTENT_LAST would be set.
1721                  */
1722                 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1723                                              1 : fs->fs_cnt_need;
1724                 fs->fs_fm->fm_mapped_extents = 0;
1725                 fs->fs_fm->fm_flags = fiemap->fm_flags;
1726
1727                 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1728
1729                 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1730                         GOTO(obj_put, rc = -EINVAL);
1731                 /* If OST is inactive, return extent with UNKNOWN flag. */
1732                 if (!lov->lov_tgts[ost_index]->ltd_active) {
1733                         fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1734                         fs->fs_fm->fm_mapped_extents = 1;
1735
1736                         fm_ext[0].fe_logical = lun_start;
1737                         fm_ext[0].fe_length = obd_object_end - lun_start + 1;
1738                         fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1739
1740                         goto inactive_tgt;
1741                 }
1742
1743                 fs->fs_fm->fm_start = lun_start;
1744                 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1745                 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1746                 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1747
1748                 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1749                 if (rc != 0)
1750                         GOTO(obj_put, rc);
1751 inactive_tgt:
1752                 ext_count = fs->fs_fm->fm_mapped_extents;
1753                 if (ext_count == 0) {
1754                         ost_done = true;
1755                         fs->fs_device_done = true;
1756                         /* If last stripe has hold at the end,
1757                          * we need to return */
1758                         if (stripeno == fs->fs_last_stripe) {
1759                                 fiemap->fm_mapped_extents = 0;
1760                                 fs->fs_finish_stripe = true;
1761                                 GOTO(obj_put, rc);
1762                         }
1763                         break;
1764                 } else if (fs->fs_enough) {
1765                         /*
1766                          * We've collected enough extents and there are
1767                          * more extents after it.
1768                          */
1769                         GOTO(obj_put, rc);
1770                 }
1771
1772                 /* If we just need num of extents, got to next device */
1773                 if (fiemap->fm_extent_count == 0) {
1774                         fs->fs_cur_extent += ext_count;
1775                         break;
1776                 }
1777
1778                 /* prepare to copy retrived map extents */
1779                 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1780                                          fm_ext[ext_count - 1].fe_length -
1781                                          lun_start;
1782
1783                 /* Have we finished mapping on this device? */
1784                 if (req_fm_len <= len_mapped_single_call) {
1785                         ost_done = true;
1786                         fs->fs_device_done = true;
1787                 }
1788
1789                 /* Clear the EXTENT_LAST flag which can be present on
1790                  * the last extent */
1791                 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1792                         fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1793                 if (lov_stripe_size(lsm, index,
1794                                     fm_ext[ext_count - 1].fe_logical +
1795                                     fm_ext[ext_count - 1].fe_length,
1796                                     stripeno) >= fmkey->lfik_oa.o_size) {
1797                         ost_eof = true;
1798                         fs->fs_device_done = true;
1799                 }
1800
1801                 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1802                                              ext_count, fs->fs_cur_extent);
1803                 fs->fs_cur_extent += ext_count;
1804
1805                 /* Ran out of available extents? */
1806                 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1807                         fs->fs_enough = true;
1808         } while (!ost_done && !ost_eof);
1809
1810         if (stripeno == fs->fs_last_stripe)
1811                 fs->fs_finish_stripe = true;
1812 obj_put:
1813         cl_object_put(env, subobj);
1814
1815         return rc;
1816 }
1817
1818 /**
1819  * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1820  * This also handles the restarting of FIEMAP calls in case mapping overflows
1821  * the available number of extents in single call.
1822  *
1823  * \param env [in]              lustre environment
1824  * \param obj [in]              file object
1825  * \param fmkey [in]            fiemap request header and other info
1826  * \param fiemap [out]          fiemap buffer holding retrived map extents
1827  * \param buflen [in/out]       max buffer length of @fiemap, when iterate
1828  *                              each OST, it is used to limit max map needed
1829  * \retval 0    success
1830  * \retval < 0  error
1831  */
1832 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1833                              struct ll_fiemap_info_key *fmkey,
1834                              struct fiemap *fiemap, size_t *buflen)
1835 {
1836         struct lov_stripe_md_entry *lsme;
1837         struct lov_stripe_md *lsm;
1838         struct fiemap *fm_local = NULL;
1839         loff_t whole_start;
1840         loff_t whole_end;
1841         int entry;
1842         int start_entry;
1843         int end_entry;
1844         int cur_stripe = 0;
1845         int stripe_count;
1846         unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1847         int rc = 0;
1848         struct fiemap_state fs = { 0 };
1849         ENTRY;
1850
1851         lsm = lov_lsm_addref(cl2lov(obj));
1852         if (lsm == NULL) {
1853                 /* no extent: there is no object for mapping */
1854                 fiemap->fm_mapped_extents = 0;
1855                 return 0;
1856         }
1857
1858         if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1859                 /**
1860                  * If the entry count > 1 or stripe_count > 1 and the
1861                  * application does not understand DEVICE_ORDER flag,
1862                  * it cannot interpret the extents correctly.
1863                  */
1864                 if (lsm->lsm_entry_count > 1 ||
1865                     (lsm->lsm_entry_count == 1 &&
1866                      lsm->lsm_entries[0]->lsme_stripe_count > 1))
1867                         GOTO(out_lsm, rc = -ENOTSUPP);
1868         }
1869
1870         /* No support for DOM layout yet. */
1871         if (lsme_is_dom(lsm->lsm_entries[0]))
1872                 GOTO(out_lsm, rc = -ENOTSUPP);
1873
1874         if (lsm->lsm_is_released) {
1875                 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1876                         /**
1877                          * released file, return a minimal FIEMAP if
1878                          * request fits in file-size.
1879                          */
1880                         fiemap->fm_mapped_extents = 1;
1881                         fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1882                         if (fiemap->fm_start + fiemap->fm_length <
1883                             fmkey->lfik_oa.o_size)
1884                                 fiemap->fm_extents[0].fe_length =
1885                                         fiemap->fm_length;
1886                         else
1887                                 fiemap->fm_extents[0].fe_length =
1888                                         fmkey->lfik_oa.o_size -
1889                                         fiemap->fm_start;
1890                         fiemap->fm_extents[0].fe_flags |=
1891                                 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1892                 }
1893                 GOTO(out_lsm, rc = 0);
1894         }
1895
1896         /* buffer_size is small to hold fm_extent_count of extents. */
1897         if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1898                 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1899
1900         OBD_ALLOC_LARGE(fm_local, buffer_size);
1901         if (fm_local == NULL)
1902                 GOTO(out_lsm, rc = -ENOMEM);
1903
1904         /**
1905          * Requested extent count exceeds the fiemap buffer size, shrink our
1906          * ambition.
1907          */
1908         if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1909                 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1910         if (fiemap->fm_extent_count == 0)
1911                 fs.fs_cnt_need = 0;
1912
1913         fs.fs_enough = false;
1914         fs.fs_cur_extent = 0;
1915         fs.fs_fm = fm_local;
1916         fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1917
1918         whole_start = fiemap->fm_start;
1919         /* whole_start is beyond the end of the file */
1920         if (whole_start > fmkey->lfik_oa.o_size)
1921                 GOTO(out_fm_local, rc = -EINVAL);
1922         whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1923                                         fmkey->lfik_oa.o_size :
1924                                         whole_start + fiemap->fm_length - 1;
1925         /**
1926          * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
1927          * size
1928          */
1929         if (whole_end > fmkey->lfik_oa.o_size)
1930                 whole_end = fmkey->lfik_oa.o_size;
1931
1932         start_entry = lov_lsm_entry(lsm, whole_start);
1933         end_entry = lov_lsm_entry(lsm, whole_end);
1934         if (end_entry == -1)
1935                 end_entry = lsm->lsm_entry_count - 1;
1936
1937         if (start_entry == -1 || end_entry == -1)
1938                 GOTO(out_fm_local, rc = -EINVAL);
1939
1940         /* TODO: rewrite it with lov_foreach_io_layout() */
1941         for (entry = start_entry; entry <= end_entry; entry++) {
1942                 lsme = lsm->lsm_entries[entry];
1943
1944                 if (!lsme_inited(lsme))
1945                         break;
1946
1947                 if (entry == start_entry)
1948                         fs.fs_ext.e_start = whole_start;
1949                 else
1950                         fs.fs_ext.e_start = lsme->lsme_extent.e_start;
1951                 if (entry == end_entry)
1952                         fs.fs_ext.e_end = whole_end;
1953                 else
1954                         fs.fs_ext.e_end = lsme->lsme_extent.e_end - 1;
1955                 fs.fs_length = fs.fs_ext.e_end - fs.fs_ext.e_start + 1;
1956
1957                 /* Calculate start stripe, last stripe and length of mapping */
1958                 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
1959                                                        fs.fs_ext.e_start);
1960                 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
1961                                         &fs.fs_ext, fs.fs_start_stripe,
1962                                         &stripe_count);
1963                 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, entry,
1964                                         &fs.fs_ext, &fs.fs_start_stripe);
1965                 /* Check each stripe */
1966                 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
1967                      --stripe_count,
1968                      cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
1969                         rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
1970                                                fmkey, entry, cur_stripe, &fs);
1971                         if (rc < 0)
1972                                 GOTO(out_fm_local, rc);
1973                         if (fs.fs_enough)
1974                                 GOTO(finish, rc);
1975                         if (fs.fs_finish_stripe)
1976                                 break;
1977                 } /* for each stripe */
1978         } /* for covering layout component */
1979         /*
1980          * We've traversed all components, set @entry to the last component
1981          * entry, it's for the last stripe check.
1982          */
1983         entry--;
1984 finish:
1985         /* Indicate that we are returning device offsets unless file just has
1986          * single stripe */
1987         if (lsm->lsm_entry_count > 1 ||
1988             (lsm->lsm_entry_count == 1 &&
1989              lsm->lsm_entries[0]->lsme_stripe_count > 1))
1990                 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1991
1992         if (fiemap->fm_extent_count == 0)
1993                 goto skip_last_device_calc;
1994
1995         /* Check if we have reached the last stripe and whether mapping for that
1996          * stripe is done. */
1997         if ((cur_stripe == fs.fs_last_stripe) && fs.fs_device_done)
1998                 fiemap->fm_extents[fs.fs_cur_extent - 1].fe_flags |=
1999                                                              FIEMAP_EXTENT_LAST;
2000 skip_last_device_calc:
2001         fiemap->fm_mapped_extents = fs.fs_cur_extent;
2002 out_fm_local:
2003         OBD_FREE_LARGE(fm_local, buffer_size);
2004
2005 out_lsm:
2006         lov_lsm_put(lsm);
2007         return rc;
2008 }
2009
2010 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2011                                 struct lov_user_md __user *lum, size_t size)
2012 {
2013         struct lov_object       *lov = cl2lov(obj);
2014         struct lov_stripe_md    *lsm;
2015         int                     rc = 0;
2016         ENTRY;
2017
2018         lsm = lov_lsm_addref(lov);
2019         if (lsm == NULL)
2020                 RETURN(-ENODATA);
2021
2022         rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2023         lov_lsm_put(lsm);
2024         RETURN(rc);
2025 }
2026
2027 static int lov_object_layout_get(const struct lu_env *env,
2028                                  struct cl_object *obj,
2029                                  struct cl_layout *cl)
2030 {
2031         struct lov_object *lov = cl2lov(obj);
2032         struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2033         struct lu_buf *buf = &cl->cl_buf;
2034         ssize_t rc;
2035         ENTRY;
2036
2037         if (lsm == NULL) {
2038                 cl->cl_size = 0;
2039                 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2040
2041                 RETURN(0);
2042         }
2043
2044         cl->cl_size = lov_comp_md_size(lsm);
2045         cl->cl_layout_gen = lsm->lsm_layout_gen;
2046         cl->cl_dom_comp_size = 0;
2047         cl->cl_is_released = lsm->lsm_is_released;
2048         if (lsm_is_composite(lsm->lsm_magic)) {
2049                 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[0];
2050
2051                 cl->cl_is_composite = true;
2052
2053                 if (lsme_is_dom(lsme))
2054                         cl->cl_dom_comp_size = lsme->lsme_extent.e_end;
2055         } else {
2056                 cl->cl_is_composite = false;
2057         }
2058
2059         rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2060         lov_lsm_put(lsm);
2061
2062         RETURN(rc < 0 ? rc : 0);
2063 }
2064
2065 static loff_t lov_object_maxbytes(struct cl_object *obj)
2066 {
2067         struct lov_object *lov = cl2lov(obj);
2068         struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2069         loff_t maxbytes;
2070
2071         if (lsm == NULL)
2072                 return LLONG_MAX;
2073
2074         maxbytes = lsm->lsm_maxbytes;
2075
2076         lov_lsm_put(lsm);
2077
2078         return maxbytes;
2079 }
2080
2081 static const struct cl_object_operations lov_ops = {
2082         .coo_page_init    = lov_page_init,
2083         .coo_lock_init    = lov_lock_init,
2084         .coo_io_init      = lov_io_init,
2085         .coo_attr_get     = lov_attr_get,
2086         .coo_attr_update  = lov_attr_update,
2087         .coo_conf_set     = lov_conf_set,
2088         .coo_getstripe    = lov_object_getstripe,
2089         .coo_layout_get   = lov_object_layout_get,
2090         .coo_maxbytes     = lov_object_maxbytes,
2091         .coo_fiemap       = lov_object_fiemap,
2092 };
2093
2094 static const struct lu_object_operations lov_lu_obj_ops = {
2095         .loo_object_init      = lov_object_init,
2096         .loo_object_delete    = lov_object_delete,
2097         .loo_object_release   = NULL,
2098         .loo_object_free      = lov_object_free,
2099         .loo_object_print     = lov_object_print,
2100         .loo_object_invariant = NULL
2101 };
2102
2103 struct lu_object *lov_object_alloc(const struct lu_env *env,
2104                                    const struct lu_object_header *unused,
2105                                    struct lu_device *dev)
2106 {
2107         struct lov_object *lov;
2108         struct lu_object  *obj;
2109
2110         ENTRY;
2111         OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2112         if (lov != NULL) {
2113                 obj = lov2lu(lov);
2114                 lu_object_init(obj, NULL, dev);
2115                 lov->lo_cl.co_ops = &lov_ops;
2116                 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2117                 /*
2118                  * object io operation vector (cl_object::co_iop) is installed
2119                  * later in lov_object_init(), as different vectors are used
2120                  * for object with different layouts.
2121                  */
2122                 obj->lo_ops = &lov_lu_obj_ops;
2123         } else
2124                 obj = NULL;
2125         RETURN(obj);
2126 }
2127
2128 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2129 {
2130         struct lov_stripe_md *lsm = NULL;
2131
2132         lov_conf_freeze(lov);
2133         if (lov->lo_lsm != NULL) {
2134                 lsm = lsm_addref(lov->lo_lsm);
2135                 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2136                         lsm, atomic_read(&lsm->lsm_refc),
2137                         lov->lo_layout_invalid, current);
2138         }
2139         lov_conf_thaw(lov);
2140         return lsm;
2141 }
2142
2143 int lov_read_and_clear_async_rc(struct cl_object *clob)
2144 {
2145         struct lu_object *luobj;
2146         int rc = 0;
2147         ENTRY;
2148
2149         luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2150                                  &lov_device_type);
2151         if (luobj != NULL) {
2152                 struct lov_object *lov = lu2lov(luobj);
2153
2154                 lov_conf_freeze(lov);
2155                 switch (lov->lo_type) {
2156                 case LLT_COMP: {
2157                         struct lov_stripe_md *lsm;
2158                         int i;
2159
2160                         lsm = lov->lo_lsm;
2161                         LASSERT(lsm != NULL);
2162                         for (i = 0; i < lsm->lsm_entry_count; i++) {
2163                                 struct lov_stripe_md_entry *lse =
2164                                                 lsm->lsm_entries[i];
2165                                 int j;
2166
2167                                 if (!lsme_inited(lse))
2168                                         break;
2169
2170                                 for (j = 0; j < lse->lsme_stripe_count; j++) {
2171                                         struct lov_oinfo *loi =
2172                                                         lse->lsme_oinfo[j];
2173
2174                                         if (lov_oinfo_is_dummy(loi))
2175                                                 continue;
2176
2177                                         if (loi->loi_ar.ar_rc && !rc)
2178                                                 rc = loi->loi_ar.ar_rc;
2179                                         loi->loi_ar.ar_rc = 0;
2180                                 }
2181                         }
2182                 }
2183                 case LLT_RELEASED:
2184                 case LLT_EMPTY:
2185                         /* fall through */
2186                 case LLT_FOREIGN:
2187                         break;
2188                 default:
2189                         LBUG();
2190                 }
2191                 lov_conf_thaw(lov);
2192         }
2193         RETURN(rc);
2194 }
2195 EXPORT_SYMBOL(lov_read_and_clear_async_rc);
2196
2197 /** @} lov */