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, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * Implementation of cl_object for LOV layer.
33 * Author: Nikita Danilov <nikita.danilov@sun.com>
34 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
37 #define DEBUG_SUBSYSTEM S_LOV
39 #include <linux/random.h>
41 #include "lov_cl_internal.h"
43 static inline struct lov_device *lov_object_dev(struct lov_object *obj)
45 return lu2lov_dev(obj->lo_cl.co_lu.lo_dev);
52 /*****************************************************************************
58 struct lov_layout_operations {
59 int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
60 struct lov_object *lov, struct lov_stripe_md *lsm,
61 const struct cl_object_conf *conf,
62 union lov_layout_state *state);
63 int (*llo_delete)(const struct lu_env *env, struct lov_object *lov,
64 union lov_layout_state *state);
65 void (*llo_fini)(const struct lu_env *env, struct lov_object *lov,
66 union lov_layout_state *state);
67 int (*llo_print)(const struct lu_env *env, void *cookie,
68 lu_printer_t p, const struct lu_object *o);
69 int (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
70 struct cl_page *page, pgoff_t index);
71 int (*llo_lock_init)(const struct lu_env *env,
72 struct cl_object *obj, struct cl_lock *lock,
73 const struct cl_io *io);
74 int (*llo_io_init)(const struct lu_env *env,
75 struct cl_object *obj, struct cl_io *io);
76 int (*llo_getattr)(const struct lu_env *env, struct cl_object *obj,
77 struct cl_attr *attr);
78 int (*llo_flush)(const struct lu_env *env, struct cl_object *obj,
79 struct ldlm_lock *lock);
82 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov);
83 static struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov);
85 static void lov_lsm_put(struct lov_stripe_md *lsm)
91 /*****************************************************************************
93 * Lov object layout operations.
97 static struct cl_object *lov_sub_find(const struct lu_env *env,
98 struct cl_device *dev,
99 const struct lu_fid *fid,
100 const struct cl_object_conf *conf)
106 o = lu_object_find_at(env, cl2lu_dev(dev), fid, &conf->coc_lu);
107 LASSERT(ergo(!IS_ERR(o), o->lo_dev->ld_type == &lovsub_device_type));
111 static int lov_page_slice_fixup(struct lov_object *lov,
112 struct cl_object *stripe)
114 struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
118 return hdr->coh_page_bufsize - lov->lo_cl.co_slice_off -
119 cfs_size_round(sizeof(struct lov_page));
121 cl_object_for_each(o, stripe)
122 o->co_slice_off += hdr->coh_page_bufsize;
124 return cl_object_header(stripe)->coh_page_bufsize;
127 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
128 struct cl_object *subobj, struct lov_oinfo *oinfo,
131 struct cl_object_header *hdr;
132 struct cl_object_header *subhdr;
133 struct cl_object_header *parent;
134 int entry = lov_comp_entry(idx);
135 int stripe = lov_comp_stripe(idx);
138 if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
139 /* For sanity:test_206.
140 * Do not leave the object in cache to avoid accessing
141 * freed memory. This is because osc_object is referring to
142 * lov_oinfo of lsm_stripe_data which will be freed due to
144 cl_object_kill(env, subobj);
145 cl_object_put(env, subobj);
149 hdr = cl_object_header(lov2cl(lov));
150 subhdr = cl_object_header(subobj);
152 CDEBUG(D_INODE, DFID"@%p[%d:%d] -> "DFID"@%p: ostid: "DOSTID
153 " ost idx: %d gen: %d\n",
154 PFID(lu_object_fid(&subobj->co_lu)), subhdr, entry, stripe,
155 PFID(lu_object_fid(lov2lu(lov))), hdr, POSTID(&oinfo->loi_oi),
156 oinfo->loi_ost_idx, oinfo->loi_ost_gen);
158 /* reuse ->coh_attr_guard to protect coh_parent change */
159 spin_lock(&subhdr->coh_attr_guard);
160 parent = subhdr->coh_parent;
161 if (parent == NULL) {
162 struct lovsub_object *lso = cl2lovsub(subobj);
164 subhdr->coh_parent = hdr;
165 spin_unlock(&subhdr->coh_attr_guard);
166 subhdr->coh_nesting = hdr->coh_nesting + 1;
167 lu_object_ref_add(&subobj->co_lu, "lov-parent", lov);
168 lso->lso_super = lov;
169 lso->lso_index = idx;
172 struct lu_object *old_obj;
173 struct lov_object *old_lov;
174 unsigned int mask = D_INODE;
176 spin_unlock(&subhdr->coh_attr_guard);
177 old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
178 LASSERT(old_obj != NULL);
179 old_lov = cl2lov(lu2cl(old_obj));
180 if (test_bit(LO_LAYOUT_INVALID, &old_lov->lo_obj_flags)) {
181 /* the object's layout has already changed but isn't
183 lu_object_unhash(env, &subobj->co_lu);
190 LU_OBJECT_DEBUG(mask, env, &subobj->co_lu,
191 "stripe %d is already owned.", idx);
192 LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
193 LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
194 cl_object_put(env, subobj);
199 static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
200 struct lov_object *lov, unsigned int index,
201 const struct cl_object_conf *conf,
202 struct lov_layout_entry *lle)
204 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
205 struct lov_thread_info *lti = lov_env_info(env);
206 struct cl_object_conf *subconf = <i->lti_stripe_conf;
207 struct lu_fid *ofid = <i->lti_fid;
208 struct cl_object *stripe;
209 struct lov_stripe_md_entry *lse = lov_lse(lov, index);
216 spin_lock_init(&r0->lo_sub_lock);
217 r0->lo_nr = lse->lsme_stripe_count;
219 OBD_ALLOC_PTR_ARRAY_LARGE(r0->lo_sub, r0->lo_nr);
220 if (r0->lo_sub == NULL)
221 GOTO(out, result = -ENOMEM);
225 memset(subconf, 0, sizeof(*subconf));
228 * Create stripe cl_objects.
230 for (i = 0; i < r0->lo_nr; ++i) {
231 struct cl_device *subdev;
232 struct lov_oinfo *oinfo = lse->lsme_oinfo[i];
233 int ost_idx = oinfo->loi_ost_idx;
234 struct obd_export *exp;
236 if (lov_oinfo_is_dummy(oinfo))
239 result = ostid_to_fid(ofid, &oinfo->loi_oi, oinfo->loi_ost_idx);
243 if (dev->ld_target[ost_idx] == NULL) {
244 CERROR("%s: OST %04x is not initialized\n",
245 lov2obd(dev->ld_lov)->obd_name, ost_idx);
246 GOTO(out, result = -EIO);
249 exp = dev->ld_lov->lov_tgts[ost_idx]->ltd_exp;
251 /* the more fast OSTs the better */
252 if (exp->exp_obd->obd_osfs.os_state & OS_STATFS_NONROT)
253 lle->lle_preference++;
256 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
257 subconf->u.coc_oinfo = oinfo;
258 LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
259 /* In the function below, .hs_keycmp resolves to
260 * lu_obj_hop_keycmp() */
261 /* coverity[overrun-buffer-val] */
262 stripe = lov_sub_find(env, subdev, ofid, subconf);
264 GOTO(out, result = PTR_ERR(stripe));
266 result = lov_init_sub(env, lov, stripe, oinfo,
267 lov_comp_index(index, i));
268 if (result == -EAGAIN) { /* try again */
275 r0->lo_sub[i] = cl2lovsub(stripe);
277 sz = lov_page_slice_fixup(lov, stripe);
278 LASSERT(ergo(psz > 0, psz == sz));
288 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
289 struct lov_layout_raid0 *r0,
290 struct lovsub_object *los, int idx)
292 struct cl_object *sub;
293 struct lu_site *site;
294 wait_queue_head_t *wq;
296 LASSERT(r0->lo_sub[idx] == los);
298 sub = lovsub2cl(los);
299 site = sub->co_lu.lo_dev->ld_site;
300 wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
302 cl_object_kill(env, sub);
303 /* release a reference to the sub-object and ... */
304 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
305 cl_object_put(env, sub);
307 /* ... wait until it is actually destroyed---sub-object clears its
308 * ->lo_sub[] slot in lovsub_object_free() */
309 wait_event(*wq, r0->lo_sub[idx] != los);
310 LASSERT(r0->lo_sub[idx] == NULL);
313 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
314 struct lov_layout_entry *lle)
316 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
320 if (r0->lo_sub != NULL) {
323 for (i = 0; i < r0->lo_nr; ++i) {
324 struct lovsub_object *los = r0->lo_sub[i];
327 cl_object_prune(env, &los->lso_cl);
329 * If top-level object is to be evicted from
330 * the cache, so are its sub-objects.
332 lov_subobject_kill(env, lov, r0, los, i);
340 static void lov_fini_raid0(const struct lu_env *env,
341 struct lov_layout_entry *lle)
343 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
345 if (r0->lo_sub != NULL) {
346 OBD_FREE_PTR_ARRAY_LARGE(r0->lo_sub, r0->lo_nr);
351 static int lov_print_raid0(const struct lu_env *env, void *cookie,
352 lu_printer_t p, const struct lov_layout_entry *lle)
354 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
357 for (i = 0; i < r0->lo_nr; ++i) {
358 struct lu_object *sub;
360 if (r0->lo_sub[i] != NULL) {
361 sub = lovsub2lu(r0->lo_sub[i]);
362 lu_object_print(env, cookie, p, sub);
364 (*p)(env, cookie, "sub %d absent\n", i);
370 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
371 unsigned int index, struct lov_layout_entry *lle,
372 struct cl_attr **lov_attr)
374 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
375 struct lov_stripe_md *lsm = lov->lo_lsm;
376 struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
377 struct cl_attr *attr = &r0->lo_attr;
381 if (r0->lo_attr_valid) {
386 memset(lvb, 0, sizeof(*lvb));
388 /* XXX: timestamps can be negative by sanity:test_39m,
390 lvb->lvb_atime = LLONG_MIN;
391 lvb->lvb_ctime = LLONG_MIN;
392 lvb->lvb_mtime = LLONG_MIN;
395 * XXX that should be replaced with a loop over sub-objects,
396 * doing cl_object_attr_get() on them. But for now, let's
397 * reuse old lov code.
401 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
402 * happy. It's not needed, because new code uses
403 * ->coh_attr_guard spin-lock to protect consistency of
404 * sub-object attributes.
406 lov_stripe_lock(lsm);
407 result = lov_merge_lvb_kms(lsm, index, lvb, &kms);
408 lov_stripe_unlock(lsm);
410 cl_lvb2attr(attr, lvb);
412 r0->lo_attr_valid = 1;
419 static struct lov_comp_layout_entry_ops raid0_ops = {
420 .lco_init = lov_init_raid0,
421 .lco_fini = lov_fini_raid0,
422 .lco_getattr = lov_attr_get_raid0,
425 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
426 unsigned int index, struct lov_layout_entry *lle,
427 struct cl_attr **lov_attr)
429 struct lov_layout_dom *dom = &lle->lle_dom;
430 struct lov_oinfo *loi = dom->lo_loi;
431 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
433 if (dom->lo_dom_r0.lo_attr_valid) {
438 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
439 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
441 cl_lvb2attr(attr, &loi->loi_lvb);
443 /* DoM component size can be bigger than stripe size after
444 * client's setattr RPC, so do not count anything beyond
445 * component end. Alternatively, check that limit on server
446 * and do not allow size overflow there. */
447 if (attr->cat_size > lle->lle_extent->e_end)
448 attr->cat_size = lle->lle_extent->e_end;
450 attr->cat_kms = attr->cat_size;
452 dom->lo_dom_r0.lo_attr_valid = 1;
459 * Lookup FLD to get MDS index of the given DOM object FID.
461 * \param[in] ld LOV device
462 * \param[in] fid FID to lookup
463 * \param[out] nr index in MDC array to return back
465 * \retval 0 and \a mds filled with MDS index if successful
466 * \retval negative value on error
468 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
476 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
477 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
479 CERROR("%s: error while looking for mds number. Seq %#llx"
480 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
485 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
488 /* find proper MDC device in the array */
489 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
490 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
491 ld->ld_md_tgts[i].ldm_idx == mds_idx)
495 if (i == ld->ld_md_tgts_nr) {
496 CERROR("%s: cannot find corresponding MDC device for mds #%x "
497 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
507 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
509 * Init the DOM object for the first time. It prepares also RAID0 entry
510 * for it to use in common methods with ordinary RAID0 layout entries.
512 * \param[in] env execution environment
513 * \param[in] dev LOV device
514 * \param[in] lov LOV object
515 * \param[in] index Composite layout entry index in LSM
516 * \param[in] lle Composite LOV layout entry
518 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
519 struct lov_object *lov, unsigned int index,
520 const struct cl_object_conf *conf,
521 struct lov_layout_entry *lle)
523 struct lov_thread_info *lti = lov_env_info(env);
524 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
525 struct cl_object *clo;
526 struct lu_object *o = lov2lu(lov);
527 const struct lu_fid *fid = lu_object_fid(o);
528 struct cl_device *mdcdev;
529 struct lov_oinfo *loi = NULL;
530 struct cl_object_conf *sconf = <i->lti_stripe_conf;
536 /* DOM entry may be not zero index due to FLR but must start from 0 */
537 if (unlikely(lle->lle_extent->e_start != 0)) {
538 CERROR("%s: DOM entry must be the first stripe in a mirror\n",
539 lov2obd(dev->ld_lov)->obd_name);
540 dump_lsm(D_ERROR, lov->lo_lsm);
544 /* find proper MDS device */
545 rc = lov_fld_lookup(dev, fid, &idx);
549 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
550 "LOV md target[%u] is NULL\n", idx);
552 /* check lsm is DOM, more checks are needed */
553 LASSERT(lsme->lsme_stripe_count == 0);
556 * Create lower cl_objects.
558 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
560 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
562 /* DoM object has no oinfo in LSM entry, create it exclusively */
563 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
567 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
569 sconf->u.coc_oinfo = loi;
571 clo = lov_sub_find(env, mdcdev, fid, sconf);
573 GOTO(out, rc = PTR_ERR(clo));
575 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
576 if (rc == -EAGAIN) /* try again */
581 lle->lle_dom.lo_dom = cl2lovsub(clo);
582 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
583 lle->lle_dom.lo_dom_r0.lo_nr = 1;
584 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
585 lle->lle_dom.lo_loi = loi;
587 rc = lov_page_slice_fixup(lov, clo);
592 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
597 * Implementation of lov_layout_operations::llo_fini for DOM object.
599 * Finish the DOM object and free related memory.
601 * \param[in] env execution environment
602 * \param[in] lov LOV object
603 * \param[in] state LOV layout state
605 static void lov_fini_dom(const struct lu_env *env,
606 struct lov_layout_entry *lle)
608 if (lle->lle_dom.lo_dom != NULL)
609 lle->lle_dom.lo_dom = NULL;
610 if (lle->lle_dom.lo_loi != NULL)
611 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
614 static struct lov_comp_layout_entry_ops dom_ops = {
615 .lco_init = lov_init_dom,
616 .lco_fini = lov_fini_dom,
617 .lco_getattr = lov_attr_get_dom,
620 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
621 struct lov_object *lov, struct lov_stripe_md *lsm,
622 const struct cl_object_conf *conf,
623 union lov_layout_state *state)
625 struct lov_layout_composite *comp = &state->composite;
626 struct lov_layout_entry *lle;
627 struct lov_mirror_entry *lre;
628 unsigned int entry_count;
629 unsigned int psz = 0;
630 unsigned int mirror_count;
631 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
634 int i, j, preference;
639 LASSERT(lsm->lsm_entry_count > 0);
640 LASSERT(lov->lo_lsm == NULL);
641 lov->lo_lsm = lsm_addref(lsm);
642 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
644 dump_lsm(D_INODE, lsm);
646 entry_count = lsm->lsm_entry_count;
648 comp->lo_flags = lsm->lsm_flags;
649 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
650 comp->lo_entry_count = lsm->lsm_entry_count;
651 comp->lo_preferred_mirror = -1;
653 if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
656 OBD_ALLOC_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
657 if (comp->lo_mirrors == NULL)
660 OBD_ALLOC_PTR_ARRAY(comp->lo_entries, entry_count);
661 if (comp->lo_entries == NULL)
664 /* Initiate all entry types and extents data at first */
665 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
668 lle = &comp->lo_entries[i];
670 lle->lle_lsme = lsm->lsm_entries[i];
671 lle->lle_type = lov_entry_type(lle->lle_lsme);
672 lle->lle_preference = 0;
673 switch (lle->lle_type) {
674 case LOV_PATTERN_RAID0:
675 lle->lle_comp_ops = &raid0_ops;
677 case LOV_PATTERN_MDT:
678 /* Allowed to have several DOM stripes in different
679 * mirrors with the same DoM size.
682 dom_size = lle->lle_lsme->lsme_extent.e_end;
683 } else if (dom_size !=
684 lle->lle_lsme->lsme_extent.e_end) {
685 CERROR("%s: DOM entries with different sizes\n",
686 lov2obd(dev->ld_lov)->obd_name);
687 dump_lsm(D_ERROR, lsm);
690 lle->lle_comp_ops = &dom_ops;
692 case LOV_PATTERN_FOREIGN:
693 lle->lle_comp_ops = NULL;
696 CERROR("%s: unknown composite layout entry type %i\n",
697 lov2obd(dev->ld_lov)->obd_name,
698 lsm->lsm_entries[i]->lsme_pattern);
699 dump_lsm(D_ERROR, lsm);
703 lle->lle_extent = &lle->lle_lsme->lsme_extent;
704 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
706 if (flr_state != LCM_FL_NONE)
707 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
709 lre = &comp->lo_mirrors[j];
711 if (mirror_id == lre->lre_mirror_id) {
712 lre->lre_valid |= lle->lle_valid;
713 lre->lre_stale |= !lle->lle_valid;
715 lsme_is_foreign(lle->lle_lsme);
720 /* new mirror detected, assume that the mirrors
721 * are shorted in layout */
724 if (j >= comp->lo_mirror_count)
727 lre = &comp->lo_mirrors[j];
730 /* entries must be sorted by mirrors */
731 lre->lre_mirror_id = mirror_id;
732 lre->lre_start = lre->lre_end = i;
733 lre->lre_preference = lle->lle_lsme->lsme_flags &
734 LCME_FL_PREF_RD ? 1000 : 0;
735 lre->lre_valid = lle->lle_valid;
736 lre->lre_stale = !lle->lle_valid;
737 lre->lre_foreign = lsme_is_foreign(lle->lle_lsme);
740 /* sanity check for FLR */
741 if (mirror_count != comp->lo_mirror_count) {
743 " doesn't have the # of mirrors it claims, %u/%u\n",
744 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
745 comp->lo_mirror_count + 1);
747 GOTO(out, result = -EINVAL);
750 lov_foreach_layout_entry(lov, lle) {
751 int index = lov_layout_entry_index(lov, lle);
754 * If the component has not been init-ed on MDS side, for
755 * PFL layout, we'd know that the components beyond this one
756 * will be dynamically init-ed later on file write/trunc ops.
758 if (!lsme_inited(lle->lle_lsme))
761 if (lsme_is_foreign(lle->lle_lsme))
764 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
769 LASSERT(ergo(psz > 0, psz == result));
774 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
776 /* decide the preferred mirror. It uses the hash value of lov_object
777 * so that different clients would use different mirrors for read. */
780 seq = hash_long((unsigned long)lov, 8);
781 for (i = 0; i < comp->lo_mirror_count; i++) {
782 unsigned int idx = (i + seq) % comp->lo_mirror_count;
784 lre = lov_mirror_entry(lov, idx);
788 if (lre->lre_foreign)
791 mirror_count++; /* valid mirror */
793 /* aggregated preference of all involved OSTs */
794 for (j = lre->lre_start; j <= lre->lre_end; j++) {
795 lre->lre_preference +=
796 comp->lo_entries[j].lle_preference;
799 if (lre->lre_preference > preference) {
800 preference = lre->lre_preference;
801 comp->lo_preferred_mirror = idx;
806 " doesn't have any valid mirrors\n",
807 PFID(lu_object_fid(lov2lu(lov))));
809 comp->lo_preferred_mirror = 0;
812 LASSERT(comp->lo_preferred_mirror >= 0);
816 return result > 0 ? 0 : result;
819 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
820 struct lov_object *lov, struct lov_stripe_md *lsm,
821 const struct cl_object_conf *conf,
822 union lov_layout_state *state)
827 static int lov_init_released(const struct lu_env *env,
828 struct lov_device *dev, struct lov_object *lov,
829 struct lov_stripe_md *lsm,
830 const struct cl_object_conf *conf,
831 union lov_layout_state *state)
833 LASSERT(lsm != NULL);
834 LASSERT(lsm->lsm_is_released);
835 LASSERT(lov->lo_lsm == NULL);
837 lov->lo_lsm = lsm_addref(lsm);
841 static int lov_init_foreign(const struct lu_env *env,
842 struct lov_device *dev, struct lov_object *lov,
843 struct lov_stripe_md *lsm,
844 const struct cl_object_conf *conf,
845 union lov_layout_state *state)
847 LASSERT(lsm != NULL);
848 LASSERT(lov->lo_type == LLT_FOREIGN);
849 LASSERT(lov->lo_lsm == NULL);
851 lov->lo_lsm = lsm_addref(lsm);
855 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
856 union lov_layout_state *state)
858 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED ||
859 lov->lo_type == LLT_FOREIGN);
861 lov_layout_wait(env, lov);
865 static int lov_delete_composite(const struct lu_env *env,
866 struct lov_object *lov,
867 union lov_layout_state *state)
869 struct lov_layout_entry *entry;
870 struct lov_layout_composite *comp = &state->composite;
874 dump_lsm(D_INODE, lov->lo_lsm);
876 lov_layout_wait(env, lov);
877 if (comp->lo_entries)
878 lov_foreach_layout_entry(lov, entry) {
879 if (entry->lle_lsme && lsme_is_foreign(entry->lle_lsme))
882 lov_delete_raid0(env, lov, entry);
888 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
889 union lov_layout_state *state)
891 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
894 static void lov_fini_composite(const struct lu_env *env,
895 struct lov_object *lov,
896 union lov_layout_state *state)
898 struct lov_layout_composite *comp = &state->composite;
901 if (comp->lo_entries != NULL) {
902 struct lov_layout_entry *entry;
904 lov_foreach_layout_entry(lov, entry)
905 if (entry->lle_comp_ops)
906 entry->lle_comp_ops->lco_fini(env, entry);
908 OBD_FREE_PTR_ARRAY(comp->lo_entries, comp->lo_entry_count);
909 comp->lo_entries = NULL;
912 if (comp->lo_mirrors != NULL) {
913 OBD_FREE_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
914 comp->lo_mirrors = NULL;
917 memset(comp, 0, sizeof(*comp));
919 dump_lsm(D_INODE, lov->lo_lsm);
920 lov_free_memmd(&lov->lo_lsm);
925 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
926 union lov_layout_state *state)
929 dump_lsm(D_INODE, lov->lo_lsm);
930 lov_free_memmd(&lov->lo_lsm);
934 static int lov_print_empty(const struct lu_env *env, void *cookie,
935 lu_printer_t p, const struct lu_object *o)
937 (*p)(env, cookie, "empty %d\n",
938 test_bit(LO_LAYOUT_INVALID, &lu2lov(o)->lo_obj_flags));
942 static int lov_print_composite(const struct lu_env *env, void *cookie,
943 lu_printer_t p, const struct lu_object *o)
945 struct lov_object *lov = lu2lov(o);
946 struct lov_stripe_md *lsm = lov->lo_lsm;
949 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
950 lsm->lsm_entry_count,
951 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
952 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
953 lsm->lsm_layout_gen);
955 for (i = 0; i < lsm->lsm_entry_count; i++) {
956 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
957 struct lov_layout_entry *lle = lov_entry(lov, i);
960 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
961 PEXT(&lse->lsme_extent), lse->lsme_magic,
962 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
963 lse->lsme_flags, lse->lsme_stripe_count,
964 lse->lsme_stripe_size);
966 if (!lsme_is_foreign(lse))
967 lov_print_raid0(env, cookie, p, lle);
973 static int lov_print_released(const struct lu_env *env, void *cookie,
974 lu_printer_t p, const struct lu_object *o)
976 struct lov_object *lov = lu2lov(o);
977 struct lov_stripe_md *lsm = lov->lo_lsm;
980 "released: %s, lsm{%p 0x%08X %d %u}:\n",
981 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
982 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
983 lsm->lsm_layout_gen);
987 static int lov_print_foreign(const struct lu_env *env, void *cookie,
988 lu_printer_t p, const struct lu_object *o)
990 struct lov_object *lov = lu2lov(o);
991 struct lov_stripe_md *lsm = lov->lo_lsm;
994 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
995 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ?
996 "invalid" : "valid", lsm,
997 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
998 lsm->lsm_layout_gen);
1000 "raw_ea_content '%.*s'\n",
1001 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
1006 * Implements cl_object_operations::coo_attr_get() method for an object
1007 * without stripes (LLT_EMPTY layout type).
1009 * The only attributes this layer is authoritative in this case is
1010 * cl_attr::cat_blocks---it's 0.
1012 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
1013 struct cl_attr *attr)
1015 attr->cat_blocks = 0;
1019 static int lov_attr_get_composite(const struct lu_env *env,
1020 struct cl_object *obj,
1021 struct cl_attr *attr)
1023 struct lov_object *lov = cl2lov(obj);
1024 struct lov_layout_entry *entry;
1030 attr->cat_blocks = 0;
1031 lov_foreach_layout_entry(lov, entry) {
1032 struct cl_attr *lov_attr = NULL;
1033 int index = lov_layout_entry_index(lov, entry);
1035 if (!entry->lle_valid)
1038 /* PFL: This component has not been init-ed. */
1039 if (!lsm_entry_inited(lov->lo_lsm, index))
1042 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1047 if (lov_attr == NULL)
1050 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1051 "b=%llu\n", index - 1, lov_attr->cat_size,
1052 lov_attr->cat_mtime, lov_attr->cat_atime,
1053 lov_attr->cat_ctime, lov_attr->cat_blocks);
1056 attr->cat_blocks += lov_attr->cat_blocks;
1057 if (attr->cat_size < lov_attr->cat_size)
1058 attr->cat_size = lov_attr->cat_size;
1059 if (attr->cat_kms < lov_attr->cat_kms)
1060 attr->cat_kms = lov_attr->cat_kms;
1061 if (attr->cat_atime < lov_attr->cat_atime)
1062 attr->cat_atime = lov_attr->cat_atime;
1063 if (attr->cat_ctime < lov_attr->cat_ctime)
1064 attr->cat_ctime = lov_attr->cat_ctime;
1065 if (attr->cat_mtime < lov_attr->cat_mtime)
1066 attr->cat_mtime = lov_attr->cat_mtime;
1072 static int lov_flush_composite(const struct lu_env *env,
1073 struct cl_object *obj,
1074 struct ldlm_lock *lock)
1076 struct lov_object *lov = cl2lov(obj);
1077 struct lov_layout_entry *lle;
1082 lov_foreach_layout_entry(lov, lle) {
1083 if (!lsme_is_dom(lle->lle_lsme))
1085 rc = cl_object_flush(env, lovsub2cl(lle->lle_dom.lo_dom), lock);
1092 static int lov_flush_empty(const struct lu_env *env, struct cl_object *obj,
1093 struct ldlm_lock *lock)
1098 const static struct lov_layout_operations lov_dispatch[] = {
1100 .llo_init = lov_init_empty,
1101 .llo_delete = lov_delete_empty,
1102 .llo_fini = lov_fini_empty,
1103 .llo_print = lov_print_empty,
1104 .llo_page_init = lov_page_init_empty,
1105 .llo_lock_init = lov_lock_init_empty,
1106 .llo_io_init = lov_io_init_empty,
1107 .llo_getattr = lov_attr_get_empty,
1108 .llo_flush = lov_flush_empty,
1111 .llo_init = lov_init_released,
1112 .llo_delete = lov_delete_empty,
1113 .llo_fini = lov_fini_released,
1114 .llo_print = lov_print_released,
1115 .llo_page_init = lov_page_init_empty,
1116 .llo_lock_init = lov_lock_init_empty,
1117 .llo_io_init = lov_io_init_released,
1118 .llo_getattr = lov_attr_get_empty,
1119 .llo_flush = lov_flush_empty,
1122 .llo_init = lov_init_composite,
1123 .llo_delete = lov_delete_composite,
1124 .llo_fini = lov_fini_composite,
1125 .llo_print = lov_print_composite,
1126 .llo_page_init = lov_page_init_composite,
1127 .llo_lock_init = lov_lock_init_composite,
1128 .llo_io_init = lov_io_init_composite,
1129 .llo_getattr = lov_attr_get_composite,
1130 .llo_flush = lov_flush_composite,
1133 .llo_init = lov_init_foreign,
1134 .llo_delete = lov_delete_empty,
1135 .llo_fini = lov_fini_released,
1136 .llo_print = lov_print_foreign,
1137 .llo_page_init = lov_page_init_foreign,
1138 .llo_lock_init = lov_lock_init_empty,
1139 .llo_io_init = lov_io_init_empty,
1140 .llo_getattr = lov_attr_get_empty,
1141 .llo_flush = lov_flush_empty,
1146 * Performs a double-dispatch based on the layout type of an object.
1148 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1150 struct lov_object *__obj = (obj); \
1151 enum lov_layout_type __llt; \
1153 __llt = __obj->lo_type; \
1154 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1155 lov_dispatch[__llt].op(__VA_ARGS__); \
1159 * Return lov_layout_type associated with a given lsm
1161 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1166 if (lsm->lsm_is_released)
1167 return LLT_RELEASED;
1169 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1170 lsm->lsm_magic == LOV_MAGIC_V3 ||
1171 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1174 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1180 static inline void lov_conf_freeze(struct lov_object *lov)
1182 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1183 lov, lov->lo_owner, current);
1184 if (lov->lo_owner != current)
1185 down_read(&lov->lo_type_guard);
1188 static inline void lov_conf_thaw(struct lov_object *lov)
1190 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1191 lov, lov->lo_owner, current);
1192 if (lov->lo_owner != current)
1193 up_read(&lov->lo_type_guard);
1196 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1198 struct lov_object *__obj = (obj); \
1199 int __lock = !!(lock); \
1200 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1203 lov_conf_freeze(__obj); \
1204 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1206 lov_conf_thaw(__obj); \
1211 * Performs a locked double-dispatch based on the layout type of an object.
1213 #define LOV_2DISPATCH(obj, op, ...) \
1214 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1216 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1218 struct lov_object *__obj = (obj); \
1219 enum lov_layout_type __llt; \
1221 lov_conf_freeze(__obj); \
1222 __llt = __obj->lo_type; \
1223 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1224 lov_dispatch[__llt].op(__VA_ARGS__); \
1225 lov_conf_thaw(__obj); \
1228 static void lov_conf_lock(struct lov_object *lov)
1230 LASSERT(lov->lo_owner != current);
1231 down_write(&lov->lo_type_guard);
1232 LASSERT(lov->lo_owner == NULL);
1233 lov->lo_owner = current;
1234 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1235 lov, lov->lo_owner);
1238 static void lov_conf_unlock(struct lov_object *lov)
1240 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1241 lov, lov->lo_owner);
1242 lov->lo_owner = NULL;
1243 up_write(&lov->lo_type_guard);
1246 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1250 while (atomic_read(&lov->lo_active_ios) > 0) {
1251 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1252 PFID(lu_object_fid(lov2lu(lov))),
1253 atomic_read(&lov->lo_active_ios));
1255 wait_event_idle(lov->lo_waitq,
1256 atomic_read(&lov->lo_active_ios) == 0);
1261 static int lov_layout_change(const struct lu_env *unused,
1262 struct lov_object *lov, struct lov_stripe_md *lsm,
1263 const struct cl_object_conf *conf)
1265 enum lov_layout_type llt = lov_type(lsm);
1266 union lov_layout_state *state = &lov->u;
1267 const struct lov_layout_operations *old_ops;
1268 const struct lov_layout_operations *new_ops;
1269 struct lov_device *lov_dev = lov_object_dev(lov);
1275 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1277 env = cl_env_get(&refcheck);
1279 RETURN(PTR_ERR(env));
1281 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1283 CDEBUG(D_INODE, DFID" from %s to %s\n",
1284 PFID(lu_object_fid(lov2lu(lov))),
1285 llt2str(lov->lo_type), llt2str(llt));
1287 old_ops = &lov_dispatch[lov->lo_type];
1288 new_ops = &lov_dispatch[llt];
1290 rc = cl_object_prune(env, &lov->lo_cl);
1294 rc = old_ops->llo_delete(env, lov, &lov->u);
1298 old_ops->llo_fini(env, lov, &lov->u);
1300 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1302 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1303 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1305 /* page bufsize fixup */
1306 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1307 lov_page_slice_fixup(lov, NULL);
1310 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1312 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1314 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1315 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1316 new_ops->llo_delete(env, lov, state);
1317 new_ops->llo_fini(env, lov, state);
1318 /* this file becomes an EMPTY file. */
1319 lov->lo_type = LLT_EMPTY;
1324 cl_env_put(env, &refcheck);
1328 /*****************************************************************************
1330 * Lov object operations.
1333 static int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1334 const struct lu_object_conf *conf)
1336 struct lov_object *lov = lu2lov(obj);
1337 struct lov_device *dev = lov_object_dev(lov);
1338 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1339 union lov_layout_state *set = &lov->u;
1340 const struct lov_layout_operations *ops;
1341 struct lov_stripe_md *lsm = NULL;
1345 init_rwsem(&lov->lo_type_guard);
1346 atomic_set(&lov->lo_active_ios, 0);
1347 init_waitqueue_head(&lov->lo_waitq);
1348 cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
1350 lov->lo_type = LLT_EMPTY;
1351 if (cconf->u.coc_layout.lb_buf != NULL) {
1352 lsm = lov_unpackmd(dev->ld_lov,
1353 cconf->u.coc_layout.lb_buf,
1354 cconf->u.coc_layout.lb_len);
1356 RETURN(PTR_ERR(lsm));
1358 dump_lsm(D_INODE, lsm);
1361 /* no locking is necessary, as object is being created */
1362 lov->lo_type = lov_type(lsm);
1363 ops = &lov_dispatch[lov->lo_type];
1364 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1374 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1375 const struct cl_object_conf *conf)
1377 struct lov_stripe_md *lsm = NULL;
1378 struct lov_object *lov = cl2lov(obj);
1382 if (conf->coc_opc == OBJECT_CONF_SET &&
1383 conf->u.coc_layout.lb_buf != NULL) {
1384 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1385 conf->u.coc_layout.lb_buf,
1386 conf->u.coc_layout.lb_len);
1388 RETURN(PTR_ERR(lsm));
1389 dump_lsm(D_INODE, lsm);
1392 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1393 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1394 GOTO(out_lsm, result = 0);
1398 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1399 if (test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) &&
1400 atomic_read(&lov->lo_active_ios) > 0) {
1401 lov_conf_unlock(lov);
1402 result = lov_layout_wait(env, lov);
1408 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1410 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1411 ((lsm != NULL && lov->lo_lsm != NULL) &&
1412 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1413 (lov->lo_lsm->lsm_flags == lsm->lsm_flags) &&
1414 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1415 lsm->lsm_entries[0]->lsme_pattern))) {
1416 /* same version of layout */
1417 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1418 GOTO(out, result = 0);
1421 /* will change layout - check if there still exists active IO. */
1422 if (atomic_read(&lov->lo_active_ios) > 0) {
1423 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1424 GOTO(out, result = -EBUSY);
1427 result = lov_layout_change(env, lov, lsm, conf);
1429 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1431 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1435 lov_conf_unlock(lov);
1438 CDEBUG(D_INODE, DFID" lo_layout_invalid=%u\n",
1439 PFID(lu_object_fid(lov2lu(lov))),
1440 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags));
1444 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1446 struct lov_object *lov = lu2lov(obj);
1449 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1453 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1455 struct lov_object *lov = lu2lov(obj);
1458 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1459 lu_object_fini(obj);
1460 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1464 static int lov_object_print(const struct lu_env *env, void *cookie,
1465 lu_printer_t p, const struct lu_object *o)
1467 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1470 static int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1471 struct cl_page *page, pgoff_t index)
1473 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1478 * Implements cl_object_operations::clo_io_init() method for lov
1479 * layer. Dispatches to the appropriate layout io initialization method.
1481 static int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1484 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1486 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1487 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1488 io->ci_ignore_layout, io->ci_verify_layout);
1490 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1491 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1492 * because as long as the OSC object exists, the layout can't be
1494 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1495 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1500 * An implementation of cl_object_operations::clo_attr_get() method for lov
1501 * layer. For raid0 layout this collects and merges attributes of all
1504 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1505 struct cl_attr *attr)
1507 /* do not take lock, as this function is called under a
1508 * spin-lock. Layout is protected from changing by ongoing IO. */
1509 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1512 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1513 const struct cl_attr *attr, unsigned valid)
1516 * No dispatch is required here, as no layout implements this.
1521 static int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1522 struct cl_lock *lock, const struct cl_io *io)
1524 /* No need to lock because we've taken one refcount of layout. */
1525 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1530 * We calculate on which OST the mapping will end. If the length of mapping
1531 * is greater than (stripe_size * stripe_count) then the last_stripe will
1532 * will be one just before start_stripe. Else we check if the mapping
1533 * intersects each OST and find last_stripe.
1534 * This function returns the last_stripe and also sets the stripe_count
1535 * over which the mapping is spread
1537 * \param lsm [in] striping information for the file
1538 * \param index [in] stripe component index
1539 * \param ext [in] logical extent of mapping
1540 * \param start_stripe [in] starting stripe of the mapping
1541 * \param stripe_count [out] the number of stripes across which to map is
1544 * \retval last_stripe return the last stripe of the mapping
1546 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1547 struct lu_extent *ext,
1548 int start_stripe, int *stripe_count)
1550 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1555 init_stripe = lov_stripe_number(lsm, index, ext->e_start);
1557 if (ext->e_end - ext->e_start >
1558 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1559 if (init_stripe == start_stripe) {
1560 last_stripe = (start_stripe < 1) ?
1561 lsme->lsme_stripe_count - 1 : start_stripe - 1;
1562 *stripe_count = lsme->lsme_stripe_count;
1563 } else if (init_stripe < start_stripe) {
1564 last_stripe = (init_stripe < 1) ?
1565 lsme->lsme_stripe_count - 1 : init_stripe - 1;
1566 *stripe_count = lsme->lsme_stripe_count -
1567 (start_stripe - init_stripe);
1569 last_stripe = init_stripe - 1;
1570 *stripe_count = init_stripe - start_stripe;
1573 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1574 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1575 if (!lov_stripe_intersects(lsm, index, i, ext, NULL,
1578 if ((start_stripe != init_stripe) && (i == init_stripe))
1582 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1589 * Set fe_device and copy extents from local buffer into main return buffer.
1591 * \param fiemap [out] fiemap to hold all extents
1592 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1593 * \param ost_index [in] OST index to be written into the fm_device
1594 * field for each extent
1595 * \param ext_count [in] number of extents to be copied
1596 * \param current_extent [in] where to start copying in the extent array
1598 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1599 struct fiemap_extent *lcl_fm_ext,
1600 int ost_index, unsigned int ext_count,
1601 int current_extent, int abs_stripeno)
1606 for (ext = 0; ext < ext_count; ext++) {
1607 set_fe_device_stripenr(&lcl_fm_ext[ext], ost_index,
1609 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1612 /* Copy fm_extent's from fm_local to return buffer */
1613 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1614 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1617 #define FIEMAP_BUFFER_SIZE 4096
1620 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1621 * call. The local end offset and the device are sent in the first
1622 * fm_extent. This function calculates the stripe number from the index.
1623 * This function returns a stripe_no on which mapping is to be restarted.
1625 * This function returns fm_end_offset which is the in-OST offset at which
1626 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1627 * will re-calculate proper offset in next stripe.
1628 * Note that the first extent is passed to lov_get_info via the value field.
1630 * \param fiemap [in] fiemap request header
1631 * \param lsm [in] striping information for the file
1632 * \param index [in] stripe component index
1633 * \param ext [in] logical extent of mapping
1634 * \param start_stripe [out] starting stripe will be returned in this
1636 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1637 struct lov_stripe_md *lsm,
1638 int index, struct lu_extent *ext,
1641 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1642 u64 local_end = fiemap->fm_extents[0].fe_logical;
1647 if (fiemap->fm_extent_count == 0 ||
1648 fiemap->fm_extents[0].fe_logical == 0)
1651 stripe_no = *start_stripe;
1653 if (stripe_no == -1)
1656 /* If we have finished mapping on previous device, shift logical
1657 * offset to start of next device */
1658 if (lov_stripe_intersects(lsm, index, stripe_no, ext, NULL, &lun_end) &&
1659 local_end < lun_end) {
1660 fm_end_offset = local_end;
1662 /* This is a special value to indicate that caller should
1663 * calculate offset in next stripe. */
1665 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1668 return fm_end_offset;
1671 struct fiemap_state {
1672 struct fiemap *fs_fm;
1673 struct lu_extent fs_ext; /* current entry extent */
1675 u64 fs_end_offset; /* last iteration offset */
1676 int fs_cur_extent; /* collected exts so far */
1677 int fs_cnt_need; /* # of extents buf can hold */
1678 int fs_start_stripe;
1680 bool fs_device_done; /* enough for this OST */
1681 bool fs_finish_stripe; /* reached fs_last_stripe */
1682 bool fs_enough; /* enough for this call */
1685 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1686 struct lov_object *lov,
1687 struct lov_stripe_md *lsm,
1690 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1691 struct lov_thread_info *lti = lov_env_info(env);
1692 struct lu_fid *ofid = <i->lti_fid;
1693 struct lov_oinfo *oinfo;
1694 struct cl_device *subdev;
1695 int entry = lov_comp_entry(index);
1696 int stripe = lov_comp_stripe(index);
1699 struct cl_object *result;
1701 if (lov->lo_type != LLT_COMP)
1702 GOTO(out, result = NULL);
1704 if (entry >= lsm->lsm_entry_count ||
1705 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1706 GOTO(out, result = NULL);
1708 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1709 ost_idx = oinfo->loi_ost_idx;
1710 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1712 GOTO(out, result = NULL);
1714 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1715 result = lov_sub_find(env, subdev, ofid, NULL);
1718 result = ERR_PTR(-EINVAL);
1722 static int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1723 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1724 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1725 int index, int stripe_last, int stripeno,
1726 struct fiemap_state *fs)
1728 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1729 struct cl_object *subobj;
1730 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1731 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1732 u64 req_fm_len; /* max requested extent coverage */
1733 u64 len_mapped_single_call;
1736 unsigned int ext_count;
1737 /* EOF for object */
1738 bool ost_eof = false;
1739 /* done with required mapping for this OST? */
1740 bool ost_done = false;
1744 fs->fs_device_done = false;
1745 /* Find out range of mapping on this stripe */
1746 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1747 &obd_start, &obd_end)) == 0)
1750 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1753 /* If this is a continuation FIEMAP call and we are on
1754 * starting stripe then obd_start needs to be set to
1756 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1757 obd_start = fs->fs_end_offset;
1759 if (lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno) ==
1763 req_fm_len = obd_end - obd_start + 1;
1764 fs->fs_fm->fm_length = 0;
1765 len_mapped_single_call = 0;
1767 /* find lobsub object */
1768 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1769 lov_comp_index(index, stripeno));
1771 return PTR_ERR(subobj);
1772 /* If the output buffer is very large and the objects have many
1773 * extents we may need to loop on a single OST repeatedly */
1775 if (fiemap->fm_extent_count > 0) {
1776 /* Don't get too many extents. */
1777 if (fs->fs_cur_extent + fs->fs_cnt_need >
1778 fiemap->fm_extent_count)
1779 fs->fs_cnt_need = fiemap->fm_extent_count -
1783 obd_start += len_mapped_single_call;
1784 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1785 req_fm_len = fs->fs_fm->fm_length;
1787 * If we've collected enough extent map, we'd request 1 more,
1788 * to see whether we coincidentally finished all available
1789 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1791 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1792 1 : fs->fs_cnt_need;
1793 fs->fs_fm->fm_mapped_extents = 0;
1794 fs->fs_fm->fm_flags = fiemap->fm_flags;
1796 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1798 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1799 GOTO(obj_put, rc = -EINVAL);
1800 /* If OST is inactive, return extent with UNKNOWN flag. */
1801 if (!lov->lov_tgts[ost_index]->ltd_active) {
1802 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1803 fs->fs_fm->fm_mapped_extents = 1;
1805 fm_ext[0].fe_logical = obd_start;
1806 fm_ext[0].fe_length = obd_end - obd_start + 1;
1807 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1812 fs->fs_fm->fm_start = obd_start;
1813 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1814 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1815 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1817 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1821 ext_count = fs->fs_fm->fm_mapped_extents;
1822 if (ext_count == 0) {
1824 fs->fs_device_done = true;
1825 /* If last stripe has hold at the end,
1826 * we need to return */
1827 if (stripeno == fs->fs_last_stripe) {
1828 fiemap->fm_mapped_extents = 0;
1829 fs->fs_finish_stripe = true;
1833 } else if (fs->fs_enough) {
1835 * We've collected enough extents and there are
1836 * more extents after it.
1841 /* If we just need num of extents, got to next device */
1842 if (fiemap->fm_extent_count == 0) {
1843 fs->fs_cur_extent += ext_count;
1847 /* prepare to copy retrived map extents */
1848 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1849 fm_ext[ext_count - 1].fe_length -
1852 /* Have we finished mapping on this device? */
1853 if (req_fm_len <= len_mapped_single_call) {
1855 fs->fs_device_done = true;
1858 /* Clear the EXTENT_LAST flag which can be present on
1859 * the last extent */
1860 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1861 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1862 if (lov_stripe_size(lsm, index,
1863 fm_ext[ext_count - 1].fe_logical +
1864 fm_ext[ext_count - 1].fe_length,
1865 stripeno) >= fmkey->lfik_oa.o_size) {
1867 fs->fs_device_done = true;
1870 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1871 ext_count, fs->fs_cur_extent,
1872 stripe_last + stripeno);
1873 fs->fs_cur_extent += ext_count;
1875 /* Ran out of available extents? */
1876 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1877 fs->fs_enough = true;
1878 } while (!ost_done && !ost_eof);
1880 if (stripeno == fs->fs_last_stripe)
1881 fs->fs_finish_stripe = true;
1883 cl_object_put(env, subobj);
1889 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1890 * This also handles the restarting of FIEMAP calls in case mapping overflows
1891 * the available number of extents in single call.
1893 * \param env [in] lustre environment
1894 * \param obj [in] file object
1895 * \param fmkey [in] fiemap request header and other info
1896 * \param fiemap [out] fiemap buffer holding retrived map extents
1897 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1898 * each OST, it is used to limit max map needed
1902 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1903 struct ll_fiemap_info_key *fmkey,
1904 struct fiemap *fiemap, size_t *buflen)
1906 struct lov_stripe_md_entry *lsme;
1907 struct lov_stripe_md *lsm;
1908 struct fiemap *fm_local = NULL;
1912 int start_entry = -1;
1916 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1918 struct fiemap_state fs = { 0 };
1919 struct lu_extent range;
1922 int start_stripe = 0;
1923 bool resume = false;
1926 lsm = lov_lsm_addref(cl2lov(obj));
1928 /* no extent: there is no object for mapping */
1929 fiemap->fm_mapped_extents = 0;
1933 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1935 * If the entry count > 1 or stripe_count > 1 and the
1936 * application does not understand DEVICE_ORDER flag,
1937 * it cannot interpret the extents correctly.
1939 if (lsm->lsm_entry_count > 1 ||
1940 (lsm->lsm_entry_count == 1 &&
1941 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1942 GOTO(out_lsm, rc = -ENOTSUPP);
1945 /* No support for DOM layout yet. */
1946 if (lsme_is_dom(lsm->lsm_entries[0]))
1947 GOTO(out_lsm, rc = -ENOTSUPP);
1949 if (lsm->lsm_is_released) {
1950 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1952 * released file, return a minimal FIEMAP if
1953 * request fits in file-size.
1955 fiemap->fm_mapped_extents = 1;
1956 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1957 if (fiemap->fm_start + fiemap->fm_length <
1958 fmkey->lfik_oa.o_size)
1959 fiemap->fm_extents[0].fe_length =
1962 fiemap->fm_extents[0].fe_length =
1963 fmkey->lfik_oa.o_size -
1965 fiemap->fm_extents[0].fe_flags |=
1966 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1968 GOTO(out_lsm, rc = 0);
1971 /* buffer_size is small to hold fm_extent_count of extents. */
1972 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1973 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1975 OBD_ALLOC_LARGE(fm_local, buffer_size);
1976 if (fm_local == NULL)
1977 GOTO(out_lsm, rc = -ENOMEM);
1980 * Requested extent count exceeds the fiemap buffer size, shrink our
1983 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1984 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1986 fs.fs_enough = false;
1987 fs.fs_cur_extent = 0;
1988 fs.fs_fm = fm_local;
1989 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1991 whole_start = fiemap->fm_start;
1992 /* whole_start is beyond the end of the file */
1993 if (whole_start > fmkey->lfik_oa.o_size)
1994 GOTO(out_fm_local, rc = -EINVAL);
1995 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1996 fmkey->lfik_oa.o_size + 1 :
1997 whole_start + fiemap->fm_length;
1999 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
2002 if (whole_end > fmkey->lfik_oa.o_size + 1)
2003 whole_end = fmkey->lfik_oa.o_size + 1;
2006 * the high 16bits of fe_device remember which stripe the last
2007 * call has been arrived, we'd continue from there in this call.
2009 if (fiemap->fm_extent_count && fiemap->fm_extents[0].fe_logical)
2011 stripe_last = get_fe_stripenr(&fiemap->fm_extents[0]);
2013 * stripe_last records stripe number we've been processed in the last
2016 end_entry = lsm->lsm_entry_count - 1;
2018 for (entry = 0; entry <= end_entry; entry++) {
2019 lsme = lsm->lsm_entries[entry];
2020 if (cur_stripe + lsme->lsme_stripe_count >= stripe_last) {
2021 start_entry = entry;
2022 start_stripe = stripe_last - cur_stripe;
2026 cur_stripe += lsme->lsme_stripe_count;
2028 if (start_entry == -1) {
2029 CERROR(DFID": FIEMAP does not init start entry, cur_stripe=%d, "
2030 "stripe_last=%d\n", PFID(lu_object_fid(&obj->co_lu)),
2031 cur_stripe, stripe_last);
2032 GOTO(out_fm_local, rc = -EINVAL);
2035 * @start_entry & @start_stripe records the position of fiemap
2036 * resumption @stripe_last keeps recording the absolution position
2037 * we'are processing. @resume indicates we'd honor @start_stripe.
2040 range.e_start = whole_start;
2041 range.e_end = whole_end;
2043 for (entry = start_entry; entry <= end_entry; entry++) {
2044 /* remeber to update stripe_last accordingly */
2045 lsme = lsm->lsm_entries[entry];
2047 /* FLR could contain component holes between entries */
2048 if (!lsme_inited(lsme)) {
2049 stripe_last += lsme->lsme_stripe_count;
2054 if (!lu_extent_is_overlapped(&range, &lsme->lsme_extent)) {
2055 stripe_last += lsme->lsme_stripe_count;
2060 /* prepare for a component entry iteration */
2061 if (lsme->lsme_extent.e_start > whole_start)
2062 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
2064 fs.fs_ext.e_start = whole_start;
2065 if (lsme->lsme_extent.e_end > whole_end)
2066 fs.fs_ext.e_end = whole_end;
2068 fs.fs_ext.e_end = lsme->lsme_extent.e_end;
2070 /* Calculate start stripe, last stripe and length of mapping */
2072 fs.fs_start_stripe = start_stripe;
2073 /* put stripe_last to the first stripe of the comp */
2074 stripe_last -= start_stripe;
2077 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
2080 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
2081 &fs.fs_ext, fs.fs_start_stripe,
2084 * A new mirror component is under process, reset
2085 * fs.fs_end_offset and then fiemap_for_stripe() starts from
2086 * the overlapping extent, otherwise starts from
2089 if (entry > start_entry && lsme->lsme_extent.e_start == 0) {
2091 fs.fs_end_offset = 0;
2093 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap,
2094 lsm, entry, &fs.fs_ext,
2095 &fs.fs_start_stripe);
2098 /* Check each stripe */
2099 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
2101 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
2102 /* reset fs_finish_stripe */
2103 fs.fs_finish_stripe = false;
2104 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
2105 fmkey, entry, stripe_last,
2108 GOTO(out_fm_local, rc);
2110 stripe_last += cur_stripe;
2113 if (fs.fs_finish_stripe)
2115 } /* for each stripe */
2116 stripe_last += lsme->lsme_stripe_count;
2117 } /* for covering layout component entry */
2120 if (fs.fs_cur_extent > 0)
2121 cur_ext = fs.fs_cur_extent - 1;
2125 /* done all the processing */
2126 if (entry > end_entry)
2127 fiemap->fm_extents[cur_ext].fe_flags |= FIEMAP_EXTENT_LAST;
2129 /* Indicate that we are returning device offsets unless file just has
2131 if (lsm->lsm_entry_count > 1 ||
2132 (lsm->lsm_entry_count == 1 &&
2133 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2134 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2136 if (fiemap->fm_extent_count == 0)
2137 goto skip_last_device_calc;
2139 skip_last_device_calc:
2140 fiemap->fm_mapped_extents = fs.fs_cur_extent;
2142 OBD_FREE_LARGE(fm_local, buffer_size);
2149 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2150 struct lov_user_md __user *lum, size_t size)
2152 struct lov_object *lov = cl2lov(obj);
2153 struct lov_stripe_md *lsm;
2157 lsm = lov_lsm_addref(lov);
2161 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2166 static int lov_object_layout_get(const struct lu_env *env,
2167 struct cl_object *obj,
2168 struct cl_layout *cl)
2170 struct lov_object *lov = cl2lov(obj);
2171 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2172 struct lu_buf *buf = &cl->cl_buf;
2178 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2183 cl->cl_size = lov_comp_md_size(lsm);
2184 cl->cl_layout_gen = lsm->lsm_layout_gen;
2185 cl->cl_is_released = lsm->lsm_is_released;
2186 cl->cl_is_composite = lsm_is_composite(lsm->lsm_magic);
2188 rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2191 /* return error or number of bytes */
2195 static loff_t lov_object_maxbytes(struct cl_object *obj)
2197 struct lov_object *lov = cl2lov(obj);
2198 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2204 maxbytes = lsm->lsm_maxbytes;
2211 static int lov_object_flush(const struct lu_env *env, struct cl_object *obj,
2212 struct ldlm_lock *lock)
2214 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_flush, true, env, obj,
2218 static const struct cl_object_operations lov_ops = {
2219 .coo_page_init = lov_page_init,
2220 .coo_lock_init = lov_lock_init,
2221 .coo_io_init = lov_io_init,
2222 .coo_attr_get = lov_attr_get,
2223 .coo_attr_update = lov_attr_update,
2224 .coo_conf_set = lov_conf_set,
2225 .coo_getstripe = lov_object_getstripe,
2226 .coo_layout_get = lov_object_layout_get,
2227 .coo_maxbytes = lov_object_maxbytes,
2228 .coo_fiemap = lov_object_fiemap,
2229 .coo_object_flush = lov_object_flush
2232 static const struct lu_object_operations lov_lu_obj_ops = {
2233 .loo_object_init = lov_object_init,
2234 .loo_object_delete = lov_object_delete,
2235 .loo_object_release = NULL,
2236 .loo_object_free = lov_object_free,
2237 .loo_object_print = lov_object_print,
2238 .loo_object_invariant = NULL,
2241 struct lu_object *lov_object_alloc(const struct lu_env *env,
2242 const struct lu_object_header *unused,
2243 struct lu_device *dev)
2245 struct lov_object *lov;
2246 struct lu_object *obj;
2249 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2252 lu_object_init(obj, NULL, dev);
2253 lov->lo_cl.co_ops = &lov_ops;
2254 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2256 * object io operation vector (cl_object::co_iop) is installed
2257 * later in lov_object_init(), as different vectors are used
2258 * for object with different layouts.
2260 obj->lo_ops = &lov_lu_obj_ops;
2266 static struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2268 struct lov_stripe_md *lsm = NULL;
2270 lov_conf_freeze(lov);
2271 if (lov->lo_lsm != NULL) {
2272 lsm = lsm_addref(lov->lo_lsm);
2273 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2274 lsm, atomic_read(&lsm->lsm_refc),
2275 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags),
2282 int lov_read_and_clear_async_rc(struct cl_object *clob)
2284 struct lu_object *luobj;
2288 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2290 if (luobj != NULL) {
2291 struct lov_object *lov = lu2lov(luobj);
2293 lov_conf_freeze(lov);
2294 switch (lov->lo_type) {
2296 struct lov_stripe_md *lsm;
2300 LASSERT(lsm != NULL);
2301 for (i = 0; i < lsm->lsm_entry_count; i++) {
2302 struct lov_stripe_md_entry *lse =
2303 lsm->lsm_entries[i];
2306 if (!lsme_inited(lse))
2309 for (j = 0; j < lse->lsme_stripe_count; j++) {
2310 struct lov_oinfo *loi =
2313 if (lov_oinfo_is_dummy(loi))
2316 if (loi->loi_ar.ar_rc && !rc)
2317 rc = loi->loi_ar.ar_rc;
2318 loi->loi_ar.ar_rc = 0;
2334 EXPORT_SYMBOL(lov_read_and_clear_async_rc);