4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_object for LOV layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_LOV
40 #include "lov_cl_internal.h"
42 static inline struct lov_device *lov_object_dev(struct lov_object *obj)
44 return lu2lov_dev(obj->lo_cl.co_lu.lo_dev);
51 /*****************************************************************************
57 struct lov_layout_operations {
58 int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
59 struct lov_object *lov, struct lov_stripe_md *lsm,
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 int (*llo_print)(const struct lu_env *env, void *cookie,
67 lu_printer_t p, const struct lu_object *o);
68 int (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
69 struct cl_page *page, pgoff_t index);
70 int (*llo_lock_init)(const struct lu_env *env,
71 struct cl_object *obj, struct cl_lock *lock,
72 const struct cl_io *io);
73 int (*llo_io_init)(const struct lu_env *env,
74 struct cl_object *obj, struct cl_io *io);
75 int (*llo_getattr)(const struct lu_env *env, struct cl_object *obj,
76 struct cl_attr *attr);
79 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov);
81 static void lov_lsm_put(struct lov_stripe_md *lsm)
87 /*****************************************************************************
89 * Lov object layout operations.
93 static struct cl_object *lov_sub_find(const struct lu_env *env,
94 struct cl_device *dev,
95 const struct lu_fid *fid,
96 const struct cl_object_conf *conf)
102 o = lu_object_find_at(env, cl2lu_dev(dev), fid, &conf->coc_lu);
103 LASSERT(ergo(!IS_ERR(o), o->lo_dev->ld_type == &lovsub_device_type));
107 static int lov_page_slice_fixup(struct lov_object *lov,
108 struct cl_object *stripe)
110 struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
114 return hdr->coh_page_bufsize - lov->lo_cl.co_slice_off -
115 cfs_size_round(sizeof(struct lov_page));
117 cl_object_for_each(o, stripe)
118 o->co_slice_off += hdr->coh_page_bufsize;
120 return cl_object_header(stripe)->coh_page_bufsize;
123 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
124 struct cl_object *subobj, struct lov_oinfo *oinfo,
127 struct cl_object_header *hdr;
128 struct cl_object_header *subhdr;
129 struct cl_object_header *parent;
130 int entry = lov_comp_entry(idx);
131 int stripe = lov_comp_stripe(idx);
134 if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
135 /* For sanity:test_206.
136 * Do not leave the object in cache to avoid accessing
137 * freed memory. This is because osc_object is referring to
138 * lov_oinfo of lsm_stripe_data which will be freed due to
140 cl_object_kill(env, subobj);
141 cl_object_put(env, subobj);
145 hdr = cl_object_header(lov2cl(lov));
146 subhdr = cl_object_header(subobj);
148 CDEBUG(D_INODE, DFID"@%p[%d:%d] -> "DFID"@%p: ostid: "DOSTID
149 " ost idx: %d gen: %d\n",
150 PFID(lu_object_fid(&subobj->co_lu)), subhdr, entry, stripe,
151 PFID(lu_object_fid(lov2lu(lov))), hdr, POSTID(&oinfo->loi_oi),
152 oinfo->loi_ost_idx, oinfo->loi_ost_gen);
154 /* reuse ->coh_attr_guard to protect coh_parent change */
155 spin_lock(&subhdr->coh_attr_guard);
156 parent = subhdr->coh_parent;
157 if (parent == NULL) {
158 struct lovsub_object *lso = cl2lovsub(subobj);
160 subhdr->coh_parent = hdr;
161 spin_unlock(&subhdr->coh_attr_guard);
162 subhdr->coh_nesting = hdr->coh_nesting + 1;
163 lu_object_ref_add(&subobj->co_lu, "lov-parent", lov);
164 lso->lso_super = lov;
165 lso->lso_index = idx;
168 struct lu_object *old_obj;
169 struct lov_object *old_lov;
170 unsigned int mask = D_INODE;
172 spin_unlock(&subhdr->coh_attr_guard);
173 old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
174 LASSERT(old_obj != NULL);
175 old_lov = cl2lov(lu2cl(old_obj));
176 if (old_lov->lo_layout_invalid) {
177 /* the object's layout has already changed but isn't
179 lu_object_unhash(env, &subobj->co_lu);
186 LU_OBJECT_DEBUG(mask, env, &subobj->co_lu,
187 "stripe %d is already owned.", idx);
188 LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
189 LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
190 cl_object_put(env, subobj);
195 static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
196 struct lov_object *lov, unsigned int index,
197 const struct cl_object_conf *conf,
198 struct lov_layout_entry *lle)
200 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
201 struct lov_thread_info *lti = lov_env_info(env);
202 struct cl_object_conf *subconf = <i->lti_stripe_conf;
203 struct lu_fid *ofid = <i->lti_fid;
204 struct cl_object *stripe;
205 struct lov_stripe_md_entry *lse = lov_lse(lov, index);
212 spin_lock_init(&r0->lo_sub_lock);
213 r0->lo_nr = lse->lsme_stripe_count;
214 LASSERT(r0->lo_nr <= lov_targets_nr(dev));
216 OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
217 if (r0->lo_sub == NULL)
218 GOTO(out, result = -ENOMEM);
222 memset(subconf, 0, sizeof(*subconf));
225 * Create stripe cl_objects.
227 for (i = 0; i < r0->lo_nr; ++i) {
228 struct cl_device *subdev;
229 struct lov_oinfo *oinfo = lse->lsme_oinfo[i];
230 int ost_idx = oinfo->loi_ost_idx;
232 if (lov_oinfo_is_dummy(oinfo))
235 result = ostid_to_fid(ofid, &oinfo->loi_oi, oinfo->loi_ost_idx);
239 if (dev->ld_target[ost_idx] == NULL) {
240 CERROR("%s: OST %04x is not initialized\n",
241 lov2obd(dev->ld_lov)->obd_name, ost_idx);
242 GOTO(out, result = -EIO);
245 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
246 subconf->u.coc_oinfo = oinfo;
247 LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
248 /* In the function below, .hs_keycmp resolves to
249 * lu_obj_hop_keycmp() */
250 /* coverity[overrun-buffer-val] */
251 stripe = lov_sub_find(env, subdev, ofid, subconf);
253 GOTO(out, result = PTR_ERR(stripe));
255 result = lov_init_sub(env, lov, stripe, oinfo,
256 lov_comp_index(index, i));
257 if (result == -EAGAIN) { /* try again */
264 r0->lo_sub[i] = cl2lovsub(stripe);
266 sz = lov_page_slice_fixup(lov, stripe);
267 LASSERT(ergo(psz > 0, psz == sz));
277 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
278 struct lov_layout_raid0 *r0,
279 struct lovsub_object *los, int idx)
281 struct cl_object *sub;
282 struct lu_site *site;
283 struct lu_site_bkt_data *bkt;
284 wait_queue_t *waiter;
286 LASSERT(r0->lo_sub[idx] == los);
288 sub = lovsub2cl(los);
289 site = sub->co_lu.lo_dev->ld_site;
290 bkt = lu_site_bkt_from_fid(site, &sub->co_lu.lo_header->loh_fid);
292 cl_object_kill(env, sub);
293 /* release a reference to the sub-object and ... */
294 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
295 cl_object_put(env, sub);
297 /* ... wait until it is actually destroyed---sub-object clears its
298 * ->lo_sub[] slot in lovsub_object_free() */
299 if (r0->lo_sub[idx] == los) {
300 waiter = &lov_env_info(env)->lti_waiter;
301 init_waitqueue_entry(waiter, current);
302 add_wait_queue(&bkt->lsb_marche_funebre, waiter);
303 set_current_state(TASK_UNINTERRUPTIBLE);
305 /* this wait-queue is signaled at the end of
306 * lu_object_free(). */
307 set_current_state(TASK_UNINTERRUPTIBLE);
308 spin_lock(&r0->lo_sub_lock);
309 if (r0->lo_sub[idx] == los) {
310 spin_unlock(&r0->lo_sub_lock);
313 spin_unlock(&r0->lo_sub_lock);
314 set_current_state(TASK_RUNNING);
318 remove_wait_queue(&bkt->lsb_marche_funebre, waiter);
320 LASSERT(r0->lo_sub[idx] == NULL);
323 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
324 struct lov_layout_entry *lle)
326 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
330 if (r0->lo_sub != NULL) {
333 for (i = 0; i < r0->lo_nr; ++i) {
334 struct lovsub_object *los = r0->lo_sub[i];
337 cl_object_prune(env, &los->lso_cl);
339 * If top-level object is to be evicted from
340 * the cache, so are its sub-objects.
342 lov_subobject_kill(env, lov, r0, los, i);
350 static void lov_fini_raid0(const struct lu_env *env,
351 struct lov_layout_entry *lle)
353 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
355 if (r0->lo_sub != NULL) {
356 OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
361 static int lov_print_raid0(const struct lu_env *env, void *cookie,
362 lu_printer_t p, const struct lov_layout_entry *lle)
364 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
367 for (i = 0; i < r0->lo_nr; ++i) {
368 struct lu_object *sub;
370 if (r0->lo_sub[i] != NULL) {
371 sub = lovsub2lu(r0->lo_sub[i]);
372 lu_object_print(env, cookie, p, sub);
374 (*p)(env, cookie, "sub %d absent\n", i);
380 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
381 unsigned int index, struct lov_layout_entry *lle,
382 struct cl_attr **lov_attr)
384 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
385 struct lov_stripe_md *lsm = lov->lo_lsm;
386 struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
387 struct cl_attr *attr = &r0->lo_attr;
391 if (r0->lo_attr_valid) {
396 memset(lvb, 0, sizeof(*lvb));
398 /* XXX: timestamps can be negative by sanity:test_39m,
400 lvb->lvb_atime = LLONG_MIN;
401 lvb->lvb_ctime = LLONG_MIN;
402 lvb->lvb_mtime = LLONG_MIN;
405 * XXX that should be replaced with a loop over sub-objects,
406 * doing cl_object_attr_get() on them. But for now, let's
407 * reuse old lov code.
411 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
412 * happy. It's not needed, because new code uses
413 * ->coh_attr_guard spin-lock to protect consistency of
414 * sub-object attributes.
416 lov_stripe_lock(lsm);
417 result = lov_merge_lvb_kms(lsm, index, lvb, &kms);
418 lov_stripe_unlock(lsm);
420 cl_lvb2attr(attr, lvb);
422 r0->lo_attr_valid = 1;
429 static struct lov_comp_layout_entry_ops raid0_ops = {
430 .lco_init = lov_init_raid0,
431 .lco_fini = lov_fini_raid0,
432 .lco_getattr = lov_attr_get_raid0,
435 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
436 unsigned int index, struct lov_layout_entry *lle,
437 struct cl_attr **lov_attr)
439 struct lov_layout_dom *dom = &lle->lle_dom;
440 struct lov_oinfo *loi = dom->lo_loi;
441 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
443 if (dom->lo_dom_r0.lo_attr_valid) {
448 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
449 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
451 cl_lvb2attr(attr, &loi->loi_lvb);
453 /* DoM component size can be bigger than stripe size after
454 * client's setattr RPC, so do not count anything beyond
455 * component end. Alternatively, check that limit on server
456 * and do not allow size overflow there. */
457 if (attr->cat_size > lle->lle_extent->e_end)
458 attr->cat_size = lle->lle_extent->e_end;
460 attr->cat_kms = attr->cat_size;
462 dom->lo_dom_r0.lo_attr_valid = 1;
469 * Lookup FLD to get MDS index of the given DOM object FID.
471 * \param[in] ld LOV device
472 * \param[in] fid FID to lookup
473 * \param[out] nr index in MDC array to return back
475 * \retval 0 and \a mds filled with MDS index if successful
476 * \retval negative value on error
478 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
486 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
487 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
489 CERROR("%s: error while looking for mds number. Seq %#llx"
490 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
495 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
498 /* find proper MDC device in the array */
499 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
500 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
501 ld->ld_md_tgts[i].ldm_idx == mds_idx)
505 if (i == ld->ld_md_tgts_nr) {
506 CERROR("%s: cannot find corresponding MDC device for mds #%x "
507 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
517 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
519 * Init the DOM object for the first time. It prepares also RAID0 entry
520 * for it to use in common methods with ordinary RAID0 layout entries.
522 * \param[in] env execution environment
523 * \param[in] dev LOV device
524 * \param[in] lov LOV object
525 * \param[in] index Composite layout entry index in LSM
526 * \param[in] lle Composite LOV layout entry
528 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
529 struct lov_object *lov, unsigned int index,
530 const struct cl_object_conf *conf,
531 struct lov_layout_entry *lle)
533 struct lov_thread_info *lti = lov_env_info(env);
534 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
535 struct cl_object *clo;
536 struct lu_object *o = lov2lu(lov);
537 const struct lu_fid *fid = lu_object_fid(o);
538 struct cl_device *mdcdev;
539 struct lov_oinfo *loi = NULL;
540 struct cl_object_conf *sconf = <i->lti_stripe_conf;
549 /* find proper MDS device */
550 rc = lov_fld_lookup(dev, fid, &idx);
554 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
555 "LOV md target[%u] is NULL\n", idx);
557 /* check lsm is DOM, more checks are needed */
558 LASSERT(lsme->lsme_stripe_count == 0);
561 * Create lower cl_objects.
563 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
565 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
567 /* DoM object has no oinfo in LSM entry, create it exclusively */
568 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
572 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
574 sconf->u.coc_oinfo = loi;
576 clo = lov_sub_find(env, mdcdev, fid, sconf);
578 GOTO(out, rc = PTR_ERR(clo));
580 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
581 if (rc == -EAGAIN) /* try again */
586 lle->lle_dom.lo_dom = cl2lovsub(clo);
587 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
588 lle->lle_dom.lo_dom_r0.lo_nr = 1;
589 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
590 lle->lle_dom.lo_loi = loi;
592 rc = lov_page_slice_fixup(lov, clo);
597 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
602 * Implementation of lov_layout_operations::llo_fini for DOM object.
604 * Finish the DOM object and free related memory.
606 * \param[in] env execution environment
607 * \param[in] lov LOV object
608 * \param[in] state LOV layout state
610 static void lov_fini_dom(const struct lu_env *env,
611 struct lov_layout_entry *lle)
613 if (lle->lle_dom.lo_dom != NULL)
614 lle->lle_dom.lo_dom = NULL;
615 if (lle->lle_dom.lo_loi != NULL)
616 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
619 static struct lov_comp_layout_entry_ops dom_ops = {
620 .lco_init = lov_init_dom,
621 .lco_fini = lov_fini_dom,
622 .lco_getattr = lov_attr_get_dom,
625 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
626 struct lov_object *lov, struct lov_stripe_md *lsm,
627 const struct cl_object_conf *conf,
628 union lov_layout_state *state)
630 struct lov_layout_composite *comp = &state->composite;
631 struct lov_layout_entry *lle;
632 struct lov_mirror_entry *lre;
633 unsigned int entry_count;
634 unsigned int psz = 0;
635 unsigned int mirror_count;
636 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
642 LASSERT(lsm->lsm_entry_count > 0);
643 LASSERT(lov->lo_lsm == NULL);
644 lov->lo_lsm = lsm_addref(lsm);
645 lov->lo_layout_invalid = true;
647 dump_lsm(D_INODE, lsm);
649 entry_count = lsm->lsm_entry_count;
651 spin_lock_init(&comp->lo_write_lock);
652 comp->lo_flags = lsm->lsm_flags;
653 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
654 comp->lo_entry_count = lsm->lsm_entry_count;
655 comp->lo_preferred_mirror = -1;
657 if (equi(flr_state == LCM_FL_NOT_FLR, comp->lo_mirror_count > 1))
660 OBD_ALLOC(comp->lo_mirrors,
661 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
662 if (comp->lo_mirrors == NULL)
665 OBD_ALLOC(comp->lo_entries, entry_count * sizeof(*comp->lo_entries));
666 if (comp->lo_entries == NULL)
669 /* Initiate all entry types and extents data at first */
670 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
673 lle = &comp->lo_entries[i];
675 lle->lle_lsme = lsm->lsm_entries[i];
676 lle->lle_type = lov_entry_type(lle->lle_lsme);
677 switch (lle->lle_type) {
678 case LOV_PATTERN_RAID0:
679 lle->lle_comp_ops = &raid0_ops;
681 case LOV_PATTERN_MDT:
682 lle->lle_comp_ops = &dom_ops;
685 CERROR("%s: unknown composite layout entry type %i\n",
686 lov2obd(dev->ld_lov)->obd_name,
687 lsm->lsm_entries[i]->lsme_pattern);
688 dump_lsm(D_ERROR, lsm);
692 lle->lle_extent = &lle->lle_lsme->lsme_extent;
693 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
695 if (flr_state != LCM_FL_NOT_FLR)
696 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
698 lre = &comp->lo_mirrors[j];
700 if (mirror_id == lre->lre_mirror_id) {
701 lre->lre_valid |= lle->lle_valid;
702 lre->lre_stale |= !lle->lle_valid;
707 /* new mirror detected, assume that the mirrors
708 * are shorted in layout */
711 if (j >= comp->lo_mirror_count)
714 lre = &comp->lo_mirrors[j];
717 /* entries must be sorted by mirrors */
718 lre->lre_mirror_id = mirror_id;
719 lre->lre_start = lre->lre_end = i;
720 lre->lre_preferred = (lle->lle_lsme->lsme_flags &
722 lre->lre_valid = lle->lle_valid;
723 lre->lre_stale = !lle->lle_valid;
726 /* sanity check for FLR */
727 if (mirror_count != comp->lo_mirror_count) {
729 " doesn't have the # of mirrors it claims, %u/%u\n",
730 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
731 comp->lo_mirror_count + 1);
733 GOTO(out, result = -EINVAL);
736 lov_foreach_layout_entry(lov, lle) {
737 int index = lov_layout_entry_index(lov, lle);
740 * If the component has not been init-ed on MDS side, for
741 * PFL layout, we'd know that the components beyond this one
742 * will be dynamically init-ed later on file write/trunc ops.
744 if (!lsme_inited(lle->lle_lsme))
747 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
752 LASSERT(ergo(psz > 0, psz == result));
757 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
759 /* decide the preferred mirror */
760 mirror_count = 0, i = 0;
761 lov_foreach_mirror_entry(lov, lre) {
766 mirror_count++; /* valid mirror */
768 if (lre->lre_preferred || comp->lo_preferred_mirror < 0)
769 comp->lo_preferred_mirror = i - 1;
771 if (mirror_count == 0) {
773 " doesn't have any valid mirrors\n",
774 PFID(lu_object_fid(lov2lu(lov))));
776 GOTO(out, result = -EINVAL);
779 LASSERT(comp->lo_preferred_mirror >= 0);
783 return result > 0 ? 0 : result;
786 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
787 struct lov_object *lov, struct lov_stripe_md *lsm,
788 const struct cl_object_conf *conf,
789 union lov_layout_state *state)
794 static int lov_init_released(const struct lu_env *env,
795 struct lov_device *dev, struct lov_object *lov,
796 struct lov_stripe_md *lsm,
797 const struct cl_object_conf *conf,
798 union lov_layout_state *state)
800 LASSERT(lsm != NULL);
801 LASSERT(lsm->lsm_is_released);
802 LASSERT(lov->lo_lsm == NULL);
804 lov->lo_lsm = lsm_addref(lsm);
808 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
809 union lov_layout_state *state)
811 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
813 lov_layout_wait(env, lov);
817 static int lov_delete_composite(const struct lu_env *env,
818 struct lov_object *lov,
819 union lov_layout_state *state)
821 struct lov_layout_entry *entry;
822 struct lov_layout_composite *comp = &state->composite;
826 dump_lsm(D_INODE, lov->lo_lsm);
828 lov_layout_wait(env, lov);
829 if (comp->lo_entries)
830 lov_foreach_layout_entry(lov, entry)
831 lov_delete_raid0(env, lov, entry);
836 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
837 union lov_layout_state *state)
839 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
842 static void lov_fini_composite(const struct lu_env *env,
843 struct lov_object *lov,
844 union lov_layout_state *state)
846 struct lov_layout_composite *comp = &state->composite;
849 if (comp->lo_entries != NULL) {
850 struct lov_layout_entry *entry;
852 lov_foreach_layout_entry(lov, entry)
853 entry->lle_comp_ops->lco_fini(env, entry);
855 OBD_FREE(comp->lo_entries,
856 comp->lo_entry_count * sizeof(*comp->lo_entries));
857 comp->lo_entries = NULL;
860 if (comp->lo_mirrors != NULL) {
861 OBD_FREE(comp->lo_mirrors,
862 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
863 comp->lo_mirrors = NULL;
866 memset(comp, 0, sizeof(*comp));
868 dump_lsm(D_INODE, lov->lo_lsm);
869 lov_free_memmd(&lov->lo_lsm);
874 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
875 union lov_layout_state *state)
878 dump_lsm(D_INODE, lov->lo_lsm);
879 lov_free_memmd(&lov->lo_lsm);
883 static int lov_print_empty(const struct lu_env *env, void *cookie,
884 lu_printer_t p, const struct lu_object *o)
886 (*p)(env, cookie, "empty %d\n", lu2lov(o)->lo_layout_invalid);
890 static int lov_print_composite(const struct lu_env *env, void *cookie,
891 lu_printer_t p, const struct lu_object *o)
893 struct lov_object *lov = lu2lov(o);
894 struct lov_stripe_md *lsm = lov->lo_lsm;
897 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
898 lsm->lsm_entry_count,
899 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
900 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
901 lsm->lsm_layout_gen);
903 for (i = 0; i < lsm->lsm_entry_count; i++) {
904 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
905 struct lov_layout_entry *lle = lov_entry(lov, i);
908 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
909 PEXT(&lse->lsme_extent), lse->lsme_magic,
910 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
911 lse->lsme_flags, lse->lsme_stripe_count,
912 lse->lsme_stripe_size);
913 lov_print_raid0(env, cookie, p, lle);
919 static int lov_print_released(const struct lu_env *env, void *cookie,
920 lu_printer_t p, const struct lu_object *o)
922 struct lov_object *lov = lu2lov(o);
923 struct lov_stripe_md *lsm = lov->lo_lsm;
926 "released: %s, lsm{%p 0x%08X %d %u}:\n",
927 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
928 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
929 lsm->lsm_layout_gen);
934 * Implements cl_object_operations::coo_attr_get() method for an object
935 * without stripes (LLT_EMPTY layout type).
937 * The only attributes this layer is authoritative in this case is
938 * cl_attr::cat_blocks---it's 0.
940 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
941 struct cl_attr *attr)
943 attr->cat_blocks = 0;
947 static int lov_attr_get_composite(const struct lu_env *env,
948 struct cl_object *obj,
949 struct cl_attr *attr)
951 struct lov_object *lov = cl2lov(obj);
952 struct lov_layout_entry *entry;
958 attr->cat_blocks = 0;
959 lov_foreach_layout_entry(lov, entry) {
960 struct cl_attr *lov_attr = NULL;
961 int index = lov_layout_entry_index(lov, entry);
963 if (!entry->lle_valid)
966 /* PFL: This component has not been init-ed. */
967 if (!lsm_entry_inited(lov->lo_lsm, index))
970 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
975 if (lov_attr == NULL)
978 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
979 "b=%llu\n", index - 1, lov_attr->cat_size,
980 lov_attr->cat_mtime, lov_attr->cat_atime,
981 lov_attr->cat_ctime, lov_attr->cat_blocks);
984 attr->cat_blocks += lov_attr->cat_blocks;
985 if (attr->cat_size < lov_attr->cat_size)
986 attr->cat_size = lov_attr->cat_size;
987 if (attr->cat_kms < lov_attr->cat_kms)
988 attr->cat_kms = lov_attr->cat_kms;
989 if (attr->cat_atime < lov_attr->cat_atime)
990 attr->cat_atime = lov_attr->cat_atime;
991 if (attr->cat_ctime < lov_attr->cat_ctime)
992 attr->cat_ctime = lov_attr->cat_ctime;
993 if (attr->cat_mtime < lov_attr->cat_mtime)
994 attr->cat_mtime = lov_attr->cat_mtime;
1000 const static struct lov_layout_operations lov_dispatch[] = {
1002 .llo_init = lov_init_empty,
1003 .llo_delete = lov_delete_empty,
1004 .llo_fini = lov_fini_empty,
1005 .llo_print = lov_print_empty,
1006 .llo_page_init = lov_page_init_empty,
1007 .llo_lock_init = lov_lock_init_empty,
1008 .llo_io_init = lov_io_init_empty,
1009 .llo_getattr = lov_attr_get_empty,
1012 .llo_init = lov_init_released,
1013 .llo_delete = lov_delete_empty,
1014 .llo_fini = lov_fini_released,
1015 .llo_print = lov_print_released,
1016 .llo_page_init = lov_page_init_empty,
1017 .llo_lock_init = lov_lock_init_empty,
1018 .llo_io_init = lov_io_init_released,
1019 .llo_getattr = lov_attr_get_empty,
1022 .llo_init = lov_init_composite,
1023 .llo_delete = lov_delete_composite,
1024 .llo_fini = lov_fini_composite,
1025 .llo_print = lov_print_composite,
1026 .llo_page_init = lov_page_init_composite,
1027 .llo_lock_init = lov_lock_init_composite,
1028 .llo_io_init = lov_io_init_composite,
1029 .llo_getattr = lov_attr_get_composite,
1034 * Performs a double-dispatch based on the layout type of an object.
1036 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1038 struct lov_object *__obj = (obj); \
1039 enum lov_layout_type __llt; \
1041 __llt = __obj->lo_type; \
1042 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1043 lov_dispatch[__llt].op(__VA_ARGS__); \
1047 * Return lov_layout_type associated with a given lsm
1049 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1054 if (lsm->lsm_is_released)
1055 return LLT_RELEASED;
1057 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1058 lsm->lsm_magic == LOV_MAGIC_V3 ||
1059 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1065 static inline void lov_conf_freeze(struct lov_object *lov)
1067 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1068 lov, lov->lo_owner, current);
1069 if (lov->lo_owner != current)
1070 down_read(&lov->lo_type_guard);
1073 static inline void lov_conf_thaw(struct lov_object *lov)
1075 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1076 lov, lov->lo_owner, current);
1077 if (lov->lo_owner != current)
1078 up_read(&lov->lo_type_guard);
1081 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1083 struct lov_object *__obj = (obj); \
1084 int __lock = !!(lock); \
1085 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1088 lov_conf_freeze(__obj); \
1089 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1091 lov_conf_thaw(__obj); \
1096 * Performs a locked double-dispatch based on the layout type of an object.
1098 #define LOV_2DISPATCH(obj, op, ...) \
1099 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1101 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1103 struct lov_object *__obj = (obj); \
1104 enum lov_layout_type __llt; \
1106 lov_conf_freeze(__obj); \
1107 __llt = __obj->lo_type; \
1108 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1109 lov_dispatch[__llt].op(__VA_ARGS__); \
1110 lov_conf_thaw(__obj); \
1113 static void lov_conf_lock(struct lov_object *lov)
1115 LASSERT(lov->lo_owner != current);
1116 down_write(&lov->lo_type_guard);
1117 LASSERT(lov->lo_owner == NULL);
1118 lov->lo_owner = current;
1119 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1120 lov, lov->lo_owner);
1123 static void lov_conf_unlock(struct lov_object *lov)
1125 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1126 lov, lov->lo_owner);
1127 lov->lo_owner = NULL;
1128 up_write(&lov->lo_type_guard);
1131 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1133 struct l_wait_info lwi = { 0 };
1136 while (atomic_read(&lov->lo_active_ios) > 0) {
1137 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1138 PFID(lu_object_fid(lov2lu(lov))),
1139 atomic_read(&lov->lo_active_ios));
1141 l_wait_event(lov->lo_waitq,
1142 atomic_read(&lov->lo_active_ios) == 0, &lwi);
1147 static int lov_layout_change(const struct lu_env *unused,
1148 struct lov_object *lov, struct lov_stripe_md *lsm,
1149 const struct cl_object_conf *conf)
1151 enum lov_layout_type llt = lov_type(lsm);
1152 union lov_layout_state *state = &lov->u;
1153 const struct lov_layout_operations *old_ops;
1154 const struct lov_layout_operations *new_ops;
1155 struct lov_device *lov_dev = lov_object_dev(lov);
1161 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1163 env = cl_env_get(&refcheck);
1165 RETURN(PTR_ERR(env));
1167 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1169 CDEBUG(D_INODE, DFID" from %s to %s\n",
1170 PFID(lu_object_fid(lov2lu(lov))),
1171 llt2str(lov->lo_type), llt2str(llt));
1173 old_ops = &lov_dispatch[lov->lo_type];
1174 new_ops = &lov_dispatch[llt];
1176 rc = cl_object_prune(env, &lov->lo_cl);
1180 rc = old_ops->llo_delete(env, lov, &lov->u);
1184 old_ops->llo_fini(env, lov, &lov->u);
1186 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1188 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1189 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1191 /* page bufsize fixup */
1192 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1193 lov_page_slice_fixup(lov, NULL);
1196 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1198 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1200 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1201 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1202 new_ops->llo_delete(env, lov, state);
1203 new_ops->llo_fini(env, lov, state);
1204 /* this file becomes an EMPTY file. */
1205 lov->lo_type = LLT_EMPTY;
1210 cl_env_put(env, &refcheck);
1214 /*****************************************************************************
1216 * Lov object operations.
1219 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1220 const struct lu_object_conf *conf)
1222 struct lov_object *lov = lu2lov(obj);
1223 struct lov_device *dev = lov_object_dev(lov);
1224 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1225 union lov_layout_state *set = &lov->u;
1226 const struct lov_layout_operations *ops;
1227 struct lov_stripe_md *lsm = NULL;
1231 init_rwsem(&lov->lo_type_guard);
1232 atomic_set(&lov->lo_active_ios, 0);
1233 init_waitqueue_head(&lov->lo_waitq);
1234 cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
1236 lov->lo_type = LLT_EMPTY;
1237 if (cconf->u.coc_layout.lb_buf != NULL) {
1238 lsm = lov_unpackmd(dev->ld_lov,
1239 cconf->u.coc_layout.lb_buf,
1240 cconf->u.coc_layout.lb_len);
1242 RETURN(PTR_ERR(lsm));
1244 dump_lsm(D_INODE, lsm);
1247 /* no locking is necessary, as object is being created */
1248 lov->lo_type = lov_type(lsm);
1249 ops = &lov_dispatch[lov->lo_type];
1250 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1260 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1261 const struct cl_object_conf *conf)
1263 struct lov_stripe_md *lsm = NULL;
1264 struct lov_object *lov = cl2lov(obj);
1268 if (conf->coc_opc == OBJECT_CONF_SET &&
1269 conf->u.coc_layout.lb_buf != NULL) {
1270 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1271 conf->u.coc_layout.lb_buf,
1272 conf->u.coc_layout.lb_len);
1274 RETURN(PTR_ERR(lsm));
1275 dump_lsm(D_INODE, lsm);
1279 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1280 lov->lo_layout_invalid = true;
1281 GOTO(out, result = 0);
1284 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1285 if (lov->lo_layout_invalid &&
1286 atomic_read(&lov->lo_active_ios) > 0) {
1287 lov_conf_unlock(lov);
1288 result = lov_layout_wait(env, lov);
1294 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1296 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1297 ((lsm != NULL && lov->lo_lsm != NULL) &&
1298 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1299 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1300 lsm->lsm_entries[0]->lsme_pattern))) {
1301 /* same version of layout */
1302 lov->lo_layout_invalid = false;
1303 GOTO(out, result = 0);
1306 /* will change layout - check if there still exists active IO. */
1307 if (atomic_read(&lov->lo_active_ios) > 0) {
1308 lov->lo_layout_invalid = true;
1309 GOTO(out, result = -EBUSY);
1312 result = lov_layout_change(env, lov, lsm, conf);
1313 lov->lo_layout_invalid = result != 0;
1317 lov_conf_unlock(lov);
1319 CDEBUG(D_INODE, DFID" lo_layout_invalid=%d\n",
1320 PFID(lu_object_fid(lov2lu(lov))), lov->lo_layout_invalid);
1324 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1326 struct lov_object *lov = lu2lov(obj);
1329 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1333 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1335 struct lov_object *lov = lu2lov(obj);
1338 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1339 lu_object_fini(obj);
1340 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1344 static int lov_object_print(const struct lu_env *env, void *cookie,
1345 lu_printer_t p, const struct lu_object *o)
1347 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1350 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1351 struct cl_page *page, pgoff_t index)
1353 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1358 * Implements cl_object_operations::clo_io_init() method for lov
1359 * layer. Dispatches to the appropriate layout io initialization method.
1361 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1364 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1366 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1367 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1368 io->ci_ignore_layout, io->ci_verify_layout);
1370 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1371 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1372 * because as long as the OSC object exists, the layout can't be
1374 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1375 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1380 * An implementation of cl_object_operations::clo_attr_get() method for lov
1381 * layer. For raid0 layout this collects and merges attributes of all
1384 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1385 struct cl_attr *attr)
1387 /* do not take lock, as this function is called under a
1388 * spin-lock. Layout is protected from changing by ongoing IO. */
1389 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1392 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1393 const struct cl_attr *attr, unsigned valid)
1396 * No dispatch is required here, as no layout implements this.
1401 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1402 struct cl_lock *lock, const struct cl_io *io)
1404 /* No need to lock because we've taken one refcount of layout. */
1405 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1410 * We calculate on which OST the mapping will end. If the length of mapping
1411 * is greater than (stripe_size * stripe_count) then the last_stripe will
1412 * will be one just before start_stripe. Else we check if the mapping
1413 * intersects each OST and find last_stripe.
1414 * This function returns the last_stripe and also sets the stripe_count
1415 * over which the mapping is spread
1417 * \param lsm [in] striping information for the file
1418 * \param index [in] stripe component index
1419 * \param ext [in] logical extent of mapping
1420 * \param start_stripe [in] starting stripe of the mapping
1421 * \param stripe_count [out] the number of stripes across which to map is
1424 * \retval last_stripe return the last stripe of the mapping
1426 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1427 struct lu_extent *ext,
1428 int start_stripe, int *stripe_count)
1430 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1436 if (ext->e_end - ext->e_start >
1437 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1438 last_stripe = (start_stripe < 1 ? lsme->lsme_stripe_count - 1 :
1440 *stripe_count = lsme->lsme_stripe_count;
1442 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1443 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1444 if ((lov_stripe_intersects(lsm, index, i, ext,
1445 &obd_start, &obd_end)) == 0)
1449 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1456 * Set fe_device and copy extents from local buffer into main return buffer.
1458 * \param fiemap [out] fiemap to hold all extents
1459 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1460 * \param ost_index [in] OST index to be written into the fm_device
1461 * field for each extent
1462 * \param ext_count [in] number of extents to be copied
1463 * \param current_extent [in] where to start copying in the extent array
1465 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1466 struct fiemap_extent *lcl_fm_ext,
1467 int ost_index, unsigned int ext_count,
1473 for (ext = 0; ext < ext_count; ext++) {
1474 lcl_fm_ext[ext].fe_device = ost_index;
1475 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1478 /* Copy fm_extent's from fm_local to return buffer */
1479 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1480 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1483 #define FIEMAP_BUFFER_SIZE 4096
1486 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1487 * call. The local end offset and the device are sent in the first
1488 * fm_extent. This function calculates the stripe number from the index.
1489 * This function returns a stripe_no on which mapping is to be restarted.
1491 * This function returns fm_end_offset which is the in-OST offset at which
1492 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1493 * will re-calculate proper offset in next stripe.
1494 * Note that the first extent is passed to lov_get_info via the value field.
1496 * \param fiemap [in] fiemap request header
1497 * \param lsm [in] striping information for the file
1498 * \param index [in] stripe component index
1499 * \param ext [in] logical extent of mapping
1500 * \param start_stripe [out] starting stripe will be returned in this
1502 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1503 struct lov_stripe_md *lsm,
1504 int index, struct lu_extent *ext,
1507 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1508 u64 local_end = fiemap->fm_extents[0].fe_logical;
1515 if (fiemap->fm_extent_count == 0 ||
1516 fiemap->fm_extents[0].fe_logical == 0)
1519 /* Find out stripe_no from ost_index saved in the fe_device */
1520 for (i = 0; i < lsme->lsme_stripe_count; i++) {
1521 struct lov_oinfo *oinfo = lsme->lsme_oinfo[i];
1523 if (lov_oinfo_is_dummy(oinfo))
1526 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1532 if (stripe_no == -1)
1535 /* If we have finished mapping on previous device, shift logical
1536 * offset to start of next device */
1537 if (lov_stripe_intersects(lsm, index, stripe_no, ext,
1538 &lun_start, &lun_end) != 0 &&
1539 local_end < lun_end) {
1540 fm_end_offset = local_end;
1541 *start_stripe = stripe_no;
1543 /* This is a special value to indicate that caller should
1544 * calculate offset in next stripe. */
1546 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1549 return fm_end_offset;
1552 struct fiemap_state {
1553 struct fiemap *fs_fm;
1554 struct lu_extent fs_ext;
1559 int fs_start_stripe;
1561 bool fs_device_done;
1562 bool fs_finish_stripe;
1566 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1567 struct lov_object *lov,
1568 struct lov_stripe_md *lsm,
1571 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1572 struct lov_thread_info *lti = lov_env_info(env);
1573 struct lu_fid *ofid = <i->lti_fid;
1574 struct lov_oinfo *oinfo;
1575 struct cl_device *subdev;
1576 int entry = lov_comp_entry(index);
1577 int stripe = lov_comp_stripe(index);
1580 struct cl_object *result;
1582 if (lov->lo_type != LLT_COMP)
1583 GOTO(out, result = NULL);
1585 if (entry >= lsm->lsm_entry_count ||
1586 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1587 GOTO(out, result = NULL);
1589 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1590 ost_idx = oinfo->loi_ost_idx;
1591 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1593 GOTO(out, result = NULL);
1595 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1596 result = lov_sub_find(env, subdev, ofid, NULL);
1599 result = ERR_PTR(-EINVAL);
1603 int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1604 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1605 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1606 int index, int stripeno, struct fiemap_state *fs)
1608 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1609 struct cl_object *subobj;
1610 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1611 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1612 u64 req_fm_len; /* Stores length of required mapping */
1613 u64 len_mapped_single_call;
1617 unsigned int ext_count;
1618 /* EOF for object */
1619 bool ost_eof = false;
1620 /* done with required mapping for this OST? */
1621 bool ost_done = false;
1625 fs->fs_device_done = false;
1626 /* Find out range of mapping on this stripe */
1627 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1628 &lun_start, &obd_object_end)) == 0)
1631 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1634 /* If this is a continuation FIEMAP call and we are on
1635 * starting stripe then lun_start needs to be set to
1637 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1638 lun_start = fs->fs_end_offset;
1639 lun_end = lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno);
1640 if (lun_start == lun_end)
1643 req_fm_len = obd_object_end - lun_start;
1644 fs->fs_fm->fm_length = 0;
1645 len_mapped_single_call = 0;
1647 /* find lobsub object */
1648 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1649 lov_comp_index(index, stripeno));
1651 return PTR_ERR(subobj);
1652 /* If the output buffer is very large and the objects have many
1653 * extents we may need to loop on a single OST repeatedly */
1655 if (fiemap->fm_extent_count > 0) {
1656 /* Don't get too many extents. */
1657 if (fs->fs_cur_extent + fs->fs_cnt_need >
1658 fiemap->fm_extent_count)
1659 fs->fs_cnt_need = fiemap->fm_extent_count -
1663 lun_start += len_mapped_single_call;
1664 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1665 req_fm_len = fs->fs_fm->fm_length;
1667 * If we've collected enough extent map, we'd request 1 more,
1668 * to see whether we coincidentally finished all available
1669 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1671 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1672 1 : fs->fs_cnt_need;
1673 fs->fs_fm->fm_mapped_extents = 0;
1674 fs->fs_fm->fm_flags = fiemap->fm_flags;
1676 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1678 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1679 GOTO(obj_put, rc = -EINVAL);
1680 /* If OST is inactive, return extent with UNKNOWN flag. */
1681 if (!lov->lov_tgts[ost_index]->ltd_active) {
1682 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1683 fs->fs_fm->fm_mapped_extents = 1;
1685 fm_ext[0].fe_logical = lun_start;
1686 fm_ext[0].fe_length = obd_object_end - lun_start;
1687 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1692 fs->fs_fm->fm_start = lun_start;
1693 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1694 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1695 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1697 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1701 ext_count = fs->fs_fm->fm_mapped_extents;
1702 if (ext_count == 0) {
1704 fs->fs_device_done = true;
1705 /* If last stripe has hold at the end,
1706 * we need to return */
1707 if (stripeno == fs->fs_last_stripe) {
1708 fiemap->fm_mapped_extents = 0;
1709 fs->fs_finish_stripe = true;
1713 } else if (fs->fs_enough) {
1715 * We've collected enough extents and there are
1716 * more extents after it.
1721 /* If we just need num of extents, got to next device */
1722 if (fiemap->fm_extent_count == 0) {
1723 fs->fs_cur_extent += ext_count;
1727 /* prepare to copy retrived map extents */
1728 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1729 fm_ext[ext_count - 1].fe_length -
1732 /* Have we finished mapping on this device? */
1733 if (req_fm_len <= len_mapped_single_call) {
1735 fs->fs_device_done = true;
1738 /* Clear the EXTENT_LAST flag which can be present on
1739 * the last extent */
1740 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1741 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1742 if (lov_stripe_size(lsm, index,
1743 fm_ext[ext_count - 1].fe_logical +
1744 fm_ext[ext_count - 1].fe_length,
1745 stripeno) >= fmkey->lfik_oa.o_size) {
1747 fs->fs_device_done = true;
1750 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1751 ext_count, fs->fs_cur_extent);
1752 fs->fs_cur_extent += ext_count;
1754 /* Ran out of available extents? */
1755 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1756 fs->fs_enough = true;
1757 } while (!ost_done && !ost_eof);
1759 if (stripeno == fs->fs_last_stripe)
1760 fs->fs_finish_stripe = true;
1762 cl_object_put(env, subobj);
1768 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1769 * This also handles the restarting of FIEMAP calls in case mapping overflows
1770 * the available number of extents in single call.
1772 * \param env [in] lustre environment
1773 * \param obj [in] file object
1774 * \param fmkey [in] fiemap request header and other info
1775 * \param fiemap [out] fiemap buffer holding retrived map extents
1776 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1777 * each OST, it is used to limit max map needed
1781 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1782 struct ll_fiemap_info_key *fmkey,
1783 struct fiemap *fiemap, size_t *buflen)
1785 struct lov_stripe_md_entry *lsme;
1786 struct lov_stripe_md *lsm;
1787 struct fiemap *fm_local = NULL;
1795 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1797 struct fiemap_state fs = { 0 };
1800 lsm = lov_lsm_addref(cl2lov(obj));
1804 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1806 * If the entry count > 1 or stripe_count > 1 and the
1807 * application does not understand DEVICE_ORDER flag,
1808 * it cannot interpret the extents correctly.
1810 if (lsm->lsm_entry_count > 1 ||
1811 (lsm->lsm_entry_count == 1 &&
1812 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1813 GOTO(out_lsm, rc = -ENOTSUPP);
1816 /* No support for DOM layout yet. */
1817 if (lsme_is_dom(lsm->lsm_entries[0]))
1818 GOTO(out_lsm, rc = -ENOTSUPP);
1820 if (lsm->lsm_is_released) {
1821 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1823 * released file, return a minimal FIEMAP if
1824 * request fits in file-size.
1826 fiemap->fm_mapped_extents = 1;
1827 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1828 if (fiemap->fm_start + fiemap->fm_length <
1829 fmkey->lfik_oa.o_size)
1830 fiemap->fm_extents[0].fe_length =
1833 fiemap->fm_extents[0].fe_length =
1834 fmkey->lfik_oa.o_size -
1836 fiemap->fm_extents[0].fe_flags |=
1837 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1839 GOTO(out_lsm, rc = 0);
1842 /* buffer_size is small to hold fm_extent_count of extents. */
1843 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1844 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1846 OBD_ALLOC_LARGE(fm_local, buffer_size);
1847 if (fm_local == NULL)
1848 GOTO(out_lsm, rc = -ENOMEM);
1851 * Requested extent count exceeds the fiemap buffer size, shrink our
1854 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1855 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1856 if (fiemap->fm_extent_count == 0)
1859 fs.fs_enough = false;
1860 fs.fs_cur_extent = 0;
1861 fs.fs_fm = fm_local;
1862 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1864 whole_start = fiemap->fm_start;
1865 /* whole_start is beyond the end of the file */
1866 if (whole_start > fmkey->lfik_oa.o_size)
1867 GOTO(out_fm_local, rc = -EINVAL);
1868 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1869 fmkey->lfik_oa.o_size :
1870 whole_start + fiemap->fm_length - 1;
1872 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
1875 if (whole_end > fmkey->lfik_oa.o_size)
1876 whole_end = fmkey->lfik_oa.o_size;
1878 start_entry = lov_lsm_entry(lsm, whole_start);
1879 end_entry = lov_lsm_entry(lsm, whole_end);
1880 if (end_entry == -1)
1881 end_entry = lsm->lsm_entry_count - 1;
1883 if (start_entry == -1 || end_entry == -1)
1884 GOTO(out_fm_local, rc = -EINVAL);
1886 /* TODO: rewrite it with lov_foreach_io_layout() */
1887 for (entry = start_entry; entry <= end_entry; entry++) {
1888 lsme = lsm->lsm_entries[entry];
1890 if (!lsme_inited(lsme))
1893 if (entry == start_entry)
1894 fs.fs_ext.e_start = whole_start;
1896 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
1897 if (entry == end_entry)
1898 fs.fs_ext.e_end = whole_end;
1900 fs.fs_ext.e_end = lsme->lsme_extent.e_end - 1;
1901 fs.fs_length = fs.fs_ext.e_end - fs.fs_ext.e_start + 1;
1903 /* Calculate start stripe, last stripe and length of mapping */
1904 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
1906 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
1907 &fs.fs_ext, fs.fs_start_stripe,
1909 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, entry,
1910 &fs.fs_ext, &fs.fs_start_stripe);
1911 /* Check each stripe */
1912 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
1914 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
1915 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
1916 fmkey, entry, cur_stripe, &fs);
1918 GOTO(out_fm_local, rc);
1921 if (fs.fs_finish_stripe)
1923 } /* for each stripe */
1924 } /* for covering layout component */
1926 * We've traversed all components, set @entry to the last component
1927 * entry, it's for the last stripe check.
1931 /* Indicate that we are returning device offsets unless file just has
1933 if (lsm->lsm_entry_count > 1 ||
1934 (lsm->lsm_entry_count == 1 &&
1935 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1936 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1938 if (fiemap->fm_extent_count == 0)
1939 goto skip_last_device_calc;
1941 /* Check if we have reached the last stripe and whether mapping for that
1942 * stripe is done. */
1943 if ((cur_stripe == fs.fs_last_stripe) && fs.fs_device_done)
1944 fiemap->fm_extents[fs.fs_cur_extent - 1].fe_flags |=
1946 skip_last_device_calc:
1947 fiemap->fm_mapped_extents = fs.fs_cur_extent;
1949 OBD_FREE_LARGE(fm_local, buffer_size);
1956 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
1957 struct lov_user_md __user *lum, size_t size)
1959 struct lov_object *lov = cl2lov(obj);
1960 struct lov_stripe_md *lsm;
1964 lsm = lov_lsm_addref(lov);
1968 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
1973 static int lov_object_layout_get(const struct lu_env *env,
1974 struct cl_object *obj,
1975 struct cl_layout *cl)
1977 struct lov_object *lov = cl2lov(obj);
1978 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
1979 struct lu_buf *buf = &cl->cl_buf;
1985 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
1990 cl->cl_size = lov_comp_md_size(lsm);
1991 cl->cl_layout_gen = lsm->lsm_layout_gen;
1992 cl->cl_dom_comp_size = 0;
1993 if (lsm_is_composite(lsm->lsm_magic)) {
1994 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[0];
1996 cl->cl_is_composite = true;
1998 if (lsme_is_dom(lsme))
1999 cl->cl_dom_comp_size = lsme->lsme_extent.e_end;
2001 cl->cl_is_composite = false;
2004 rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2007 RETURN(rc < 0 ? rc : 0);
2010 static loff_t lov_object_maxbytes(struct cl_object *obj)
2012 struct lov_object *lov = cl2lov(obj);
2013 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2019 maxbytes = lsm->lsm_maxbytes;
2026 static const struct cl_object_operations lov_ops = {
2027 .coo_page_init = lov_page_init,
2028 .coo_lock_init = lov_lock_init,
2029 .coo_io_init = lov_io_init,
2030 .coo_attr_get = lov_attr_get,
2031 .coo_attr_update = lov_attr_update,
2032 .coo_conf_set = lov_conf_set,
2033 .coo_getstripe = lov_object_getstripe,
2034 .coo_layout_get = lov_object_layout_get,
2035 .coo_maxbytes = lov_object_maxbytes,
2036 .coo_fiemap = lov_object_fiemap,
2039 static const struct lu_object_operations lov_lu_obj_ops = {
2040 .loo_object_init = lov_object_init,
2041 .loo_object_delete = lov_object_delete,
2042 .loo_object_release = NULL,
2043 .loo_object_free = lov_object_free,
2044 .loo_object_print = lov_object_print,
2045 .loo_object_invariant = NULL
2048 struct lu_object *lov_object_alloc(const struct lu_env *env,
2049 const struct lu_object_header *unused,
2050 struct lu_device *dev)
2052 struct lov_object *lov;
2053 struct lu_object *obj;
2056 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2059 lu_object_init(obj, NULL, dev);
2060 lov->lo_cl.co_ops = &lov_ops;
2061 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2063 * object io operation vector (cl_object::co_iop) is installed
2064 * later in lov_object_init(), as different vectors are used
2065 * for object with different layouts.
2067 obj->lo_ops = &lov_lu_obj_ops;
2073 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2075 struct lov_stripe_md *lsm = NULL;
2077 lov_conf_freeze(lov);
2078 if (lov->lo_lsm != NULL) {
2079 lsm = lsm_addref(lov->lo_lsm);
2080 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2081 lsm, atomic_read(&lsm->lsm_refc),
2082 lov->lo_layout_invalid, current);
2088 int lov_read_and_clear_async_rc(struct cl_object *clob)
2090 struct lu_object *luobj;
2094 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2096 if (luobj != NULL) {
2097 struct lov_object *lov = lu2lov(luobj);
2099 lov_conf_freeze(lov);
2100 switch (lov->lo_type) {
2102 struct lov_stripe_md *lsm;
2106 LASSERT(lsm != NULL);
2107 for (i = 0; i < lsm->lsm_entry_count; i++) {
2108 struct lov_stripe_md_entry *lse =
2109 lsm->lsm_entries[i];
2112 if (!lsme_inited(lse))
2115 for (j = 0; j < lse->lsme_stripe_count; j++) {
2116 struct lov_oinfo *loi =
2119 if (lov_oinfo_is_dummy(loi))
2122 if (loi->loi_ar.ar_rc && !rc)
2123 rc = loi->loi_ar.ar_rc;
2124 loi->loi_ar.ar_rc = 0;
2138 EXPORT_SYMBOL(lov_read_and_clear_async_rc);