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;
120 cl_object_for_each(o, stripe)
121 o->co_slice_off += hdr->coh_page_bufsize;
123 return cl_object_header(stripe)->coh_page_bufsize;
126 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
127 struct cl_object *subobj, struct lov_oinfo *oinfo,
130 struct cl_object_header *hdr;
131 struct cl_object_header *subhdr;
132 struct cl_object_header *parent;
133 int entry = lov_comp_entry(idx);
134 int stripe = lov_comp_stripe(idx);
137 if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
138 /* For sanity:test_206.
139 * Do not leave the object in cache to avoid accessing
140 * freed memory. This is because osc_object is referring to
141 * lov_oinfo of lsm_stripe_data which will be freed due to
143 cl_object_kill(env, subobj);
144 cl_object_put(env, subobj);
148 hdr = cl_object_header(lov2cl(lov));
149 subhdr = cl_object_header(subobj);
151 CDEBUG(D_INODE, DFID"@%p[%d:%d] -> "DFID"@%p: ostid: "DOSTID
152 " ost idx: %d gen: %d\n",
153 PFID(lu_object_fid(&subobj->co_lu)), subhdr, entry, stripe,
154 PFID(lu_object_fid(lov2lu(lov))), hdr, POSTID(&oinfo->loi_oi),
155 oinfo->loi_ost_idx, oinfo->loi_ost_gen);
157 /* reuse ->coh_attr_guard to protect coh_parent change */
158 spin_lock(&subhdr->coh_attr_guard);
159 parent = subhdr->coh_parent;
160 if (parent == NULL) {
161 struct lovsub_object *lso = cl2lovsub(subobj);
163 subhdr->coh_parent = hdr;
164 spin_unlock(&subhdr->coh_attr_guard);
165 subhdr->coh_nesting = hdr->coh_nesting + 1;
166 lu_object_ref_add(&subobj->co_lu, "lov-parent", lov);
167 lso->lso_super = lov;
168 lso->lso_index = idx;
171 struct lu_object *old_obj;
172 struct lov_object *old_lov;
173 unsigned int mask = D_INODE;
175 spin_unlock(&subhdr->coh_attr_guard);
176 old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
177 LASSERT(old_obj != NULL);
178 old_lov = cl2lov(lu2cl(old_obj));
179 if (test_bit(LO_LAYOUT_INVALID, &old_lov->lo_obj_flags)) {
180 /* the object's layout has already changed but isn't
182 lu_object_unhash(env, &subobj->co_lu);
189 LU_OBJECT_DEBUG(mask, env, &subobj->co_lu,
190 "stripe %d is already owned.", idx);
191 LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
192 LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
193 cl_object_put(env, subobj);
198 static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
199 struct lov_object *lov, unsigned int index,
200 const struct cl_object_conf *conf,
201 struct lov_layout_entry *lle)
203 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
204 struct lov_thread_info *lti = lov_env_info(env);
205 struct cl_object_conf *subconf = <i->lti_stripe_conf;
206 struct lu_fid *ofid = <i->lti_fid;
207 struct cl_object *stripe;
208 struct lov_stripe_md_entry *lse = lov_lse(lov, index);
215 spin_lock_init(&r0->lo_sub_lock);
216 r0->lo_nr = lse->lsme_stripe_count;
218 OBD_ALLOC_PTR_ARRAY_LARGE(r0->lo_sub, r0->lo_nr);
219 if (r0->lo_sub == NULL)
220 GOTO(out, result = -ENOMEM);
224 memset(subconf, 0, sizeof(*subconf));
227 * Create stripe cl_objects.
229 for (i = 0; i < r0->lo_nr; ++i) {
230 struct cl_device *subdev;
231 struct lov_oinfo *oinfo = lse->lsme_oinfo[i];
232 int ost_idx = oinfo->loi_ost_idx;
233 struct obd_export *exp;
235 if (lov_oinfo_is_dummy(oinfo))
238 result = ostid_to_fid(ofid, &oinfo->loi_oi, oinfo->loi_ost_idx);
242 if (dev->ld_target[ost_idx] == NULL) {
243 CERROR("%s: OST %04x is not initialized\n",
244 lov2obd(dev->ld_lov)->obd_name, ost_idx);
245 GOTO(out, result = -EIO);
248 exp = dev->ld_lov->lov_tgts[ost_idx]->ltd_exp;
250 /* the more fast OSTs the better */
251 if (exp->exp_obd->obd_osfs.os_state & OS_STATFS_NONROT)
252 lle->lle_preference++;
255 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
256 subconf->u.coc_oinfo = oinfo;
257 LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
258 /* In the function below, .hs_keycmp resolves to
259 * lu_obj_hop_keycmp() */
260 /* coverity[overrun-buffer-val] */
261 stripe = lov_sub_find(env, subdev, ofid, subconf);
263 GOTO(out, result = PTR_ERR(stripe));
265 result = lov_init_sub(env, lov, stripe, oinfo,
266 lov_comp_index(index, i));
267 if (result == -EAGAIN) { /* try again */
274 r0->lo_sub[i] = cl2lovsub(stripe);
276 sz = lov_page_slice_fixup(lov, stripe);
277 LASSERT(ergo(psz > 0, psz == sz));
287 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
288 struct lov_layout_raid0 *r0,
289 struct lovsub_object *los, int idx)
291 struct cl_object *sub;
292 struct lu_site *site;
293 wait_queue_head_t *wq;
295 LASSERT(r0->lo_sub[idx] == los);
297 sub = lovsub2cl(los);
298 site = sub->co_lu.lo_dev->ld_site;
299 wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
301 cl_object_kill(env, sub);
302 /* release a reference to the sub-object and ... */
303 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
304 cl_object_put(env, sub);
306 /* ... wait until it is actually destroyed---sub-object clears its
307 * ->lo_sub[] slot in lovsub_object_free() */
308 wait_event(*wq, r0->lo_sub[idx] != los);
309 LASSERT(r0->lo_sub[idx] == NULL);
312 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
313 struct lov_layout_entry *lle)
315 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
319 if (r0->lo_sub != NULL) {
322 for (i = 0; i < r0->lo_nr; ++i) {
323 struct lovsub_object *los = r0->lo_sub[i];
326 cl_object_prune(env, &los->lso_cl);
328 * If top-level object is to be evicted from
329 * the cache, so are its sub-objects.
331 lov_subobject_kill(env, lov, r0, los, i);
339 static void lov_fini_raid0(const struct lu_env *env,
340 struct lov_layout_entry *lle)
342 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
344 if (r0->lo_sub != NULL) {
345 OBD_FREE_PTR_ARRAY_LARGE(r0->lo_sub, r0->lo_nr);
350 static int lov_print_raid0(const struct lu_env *env, void *cookie,
351 lu_printer_t p, const struct lov_layout_entry *lle)
353 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
356 for (i = 0; i < r0->lo_nr; ++i) {
357 struct lu_object *sub;
359 if (r0->lo_sub[i] != NULL) {
360 sub = lovsub2lu(r0->lo_sub[i]);
361 lu_object_print(env, cookie, p, sub);
363 (*p)(env, cookie, "sub %d absent\n", i);
369 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
370 unsigned int index, struct lov_layout_entry *lle,
371 struct cl_attr **lov_attr)
373 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
374 struct lov_stripe_md *lsm = lov->lo_lsm;
375 struct cl_attr *attr = &r0->lo_attr;
378 if (r0->lo_attr_valid) {
384 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
385 * happy. It's not needed, because new code uses
386 * ->coh_attr_guard spin-lock to protect consistency of
387 * sub-object attributes.
389 lov_stripe_lock(lsm);
390 result = lov_merge_lvb_kms(lsm, index, attr);
391 lov_stripe_unlock(lsm);
393 r0->lo_attr_valid = 1;
400 static struct lov_comp_layout_entry_ops raid0_ops = {
401 .lco_init = lov_init_raid0,
402 .lco_fini = lov_fini_raid0,
403 .lco_getattr = lov_attr_get_raid0,
406 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
407 unsigned int index, struct lov_layout_entry *lle,
408 struct cl_attr **lov_attr)
410 struct lov_layout_dom *dom = &lle->lle_dom;
411 struct lov_oinfo *loi = dom->lo_loi;
412 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
414 if (dom->lo_dom_r0.lo_attr_valid) {
419 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
420 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
422 cl_lvb2attr(attr, &loi->loi_lvb);
424 /* DoM component size can be bigger than stripe size after
425 * client's setattr RPC, so do not count anything beyond
426 * component end. Alternatively, check that limit on server
427 * and do not allow size overflow there. */
428 if (attr->cat_size > lle->lle_extent->e_end)
429 attr->cat_size = lle->lle_extent->e_end;
431 attr->cat_kms = attr->cat_size;
433 dom->lo_dom_r0.lo_attr_valid = 1;
440 * Lookup FLD to get MDS index of the given DOM object FID.
442 * \param[in] ld LOV device
443 * \param[in] fid FID to lookup
444 * \param[out] nr index in MDC array to return back
446 * \retval 0 and \a mds filled with MDS index if successful
447 * \retval negative value on error
449 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
457 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
458 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
460 CERROR("%s: error while looking for mds number. Seq %#llx"
461 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
466 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
469 /* find proper MDC device in the array */
470 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
471 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
472 ld->ld_md_tgts[i].ldm_idx == mds_idx)
476 if (i == ld->ld_md_tgts_nr) {
477 CERROR("%s: cannot find corresponding MDC device for mds #%x "
478 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
488 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
490 * Init the DOM object for the first time. It prepares also RAID0 entry
491 * for it to use in common methods with ordinary RAID0 layout entries.
493 * \param[in] env execution environment
494 * \param[in] dev LOV device
495 * \param[in] lov LOV object
496 * \param[in] index Composite layout entry index in LSM
497 * \param[in] lle Composite LOV layout entry
499 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
500 struct lov_object *lov, unsigned int index,
501 const struct cl_object_conf *conf,
502 struct lov_layout_entry *lle)
504 struct lov_thread_info *lti = lov_env_info(env);
505 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
506 struct cl_object *clo;
507 struct lu_object *o = lov2lu(lov);
508 const struct lu_fid *fid = lu_object_fid(o);
509 struct cl_device *mdcdev;
510 struct lov_oinfo *loi = NULL;
511 struct cl_object_conf *sconf = <i->lti_stripe_conf;
517 /* DOM entry may be not zero index due to FLR but must start from 0 */
518 if (unlikely(lle->lle_extent->e_start != 0)) {
519 CERROR("%s: DOM entry must be the first stripe in a mirror\n",
520 lov2obd(dev->ld_lov)->obd_name);
521 dump_lsm(D_ERROR, lov->lo_lsm);
525 /* find proper MDS device */
526 rc = lov_fld_lookup(dev, fid, &idx);
530 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
531 "LOV md target[%u] is NULL\n", idx);
533 /* check lsm is DOM, more checks are needed */
534 LASSERT(lsme->lsme_stripe_count == 0);
537 * Create lower cl_objects.
539 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
541 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
543 /* DoM object has no oinfo in LSM entry, create it exclusively */
544 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
548 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
550 sconf->u.coc_oinfo = loi;
552 clo = lov_sub_find(env, mdcdev, fid, sconf);
554 GOTO(out, rc = PTR_ERR(clo));
556 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
557 if (rc == -EAGAIN) /* try again */
562 lle->lle_dom.lo_dom = cl2lovsub(clo);
563 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
564 lle->lle_dom.lo_dom_r0.lo_nr = 1;
565 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
566 lle->lle_dom.lo_loi = loi;
568 rc = lov_page_slice_fixup(lov, clo);
573 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
578 * Implementation of lov_layout_operations::llo_fini for DOM object.
580 * Finish the DOM object and free related memory.
582 * \param[in] env execution environment
583 * \param[in] lov LOV object
584 * \param[in] state LOV layout state
586 static void lov_fini_dom(const struct lu_env *env,
587 struct lov_layout_entry *lle)
589 if (lle->lle_dom.lo_dom != NULL)
590 lle->lle_dom.lo_dom = NULL;
591 if (lle->lle_dom.lo_loi != NULL)
592 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
595 static struct lov_comp_layout_entry_ops dom_ops = {
596 .lco_init = lov_init_dom,
597 .lco_fini = lov_fini_dom,
598 .lco_getattr = lov_attr_get_dom,
601 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
602 struct lov_object *lov, struct lov_stripe_md *lsm,
603 const struct cl_object_conf *conf,
604 union lov_layout_state *state)
606 struct lov_layout_composite *comp = &state->composite;
607 struct lov_layout_entry *lle;
608 struct lov_mirror_entry *lre;
609 unsigned int entry_count;
610 unsigned int psz = 0;
611 unsigned int mirror_count;
612 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
615 int i, j, preference;
620 LASSERT(lsm->lsm_entry_count > 0);
621 LASSERT(lov->lo_lsm == NULL);
622 lov->lo_lsm = lsm_addref(lsm);
623 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
625 dump_lsm(D_INODE, lsm);
627 entry_count = lsm->lsm_entry_count;
629 comp->lo_flags = lsm->lsm_flags;
630 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
631 comp->lo_entry_count = lsm->lsm_entry_count;
632 comp->lo_preferred_mirror = -1;
634 if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
637 OBD_ALLOC_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
638 if (comp->lo_mirrors == NULL)
641 OBD_ALLOC_PTR_ARRAY(comp->lo_entries, entry_count);
642 if (comp->lo_entries == NULL)
645 /* Initiate all entry types and extents data at first */
646 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
649 lle = &comp->lo_entries[i];
651 lle->lle_lsme = lsm->lsm_entries[i];
652 lle->lle_type = lov_entry_type(lle->lle_lsme);
653 lle->lle_preference = 0;
654 switch (lle->lle_type) {
655 case LOV_PATTERN_RAID0:
656 lle->lle_comp_ops = &raid0_ops;
658 case LOV_PATTERN_MDT:
659 /* Allowed to have several DOM stripes in different
660 * mirrors with the same DoM size.
663 dom_size = lle->lle_lsme->lsme_extent.e_end;
664 } else if (dom_size !=
665 lle->lle_lsme->lsme_extent.e_end) {
666 CERROR("%s: DOM entries with different sizes\n",
667 lov2obd(dev->ld_lov)->obd_name);
668 dump_lsm(D_ERROR, lsm);
671 lle->lle_comp_ops = &dom_ops;
673 case LOV_PATTERN_FOREIGN:
674 lle->lle_comp_ops = NULL;
677 CERROR("%s: unknown composite layout entry type %i\n",
678 lov2obd(dev->ld_lov)->obd_name,
679 lsm->lsm_entries[i]->lsme_pattern);
680 dump_lsm(D_ERROR, lsm);
684 lle->lle_extent = &lle->lle_lsme->lsme_extent;
685 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
687 if (flr_state != LCM_FL_NONE)
688 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
690 lre = &comp->lo_mirrors[j];
692 if (mirror_id == lre->lre_mirror_id) {
693 lre->lre_valid |= lle->lle_valid;
694 lre->lre_stale |= !lle->lle_valid;
696 lsme_is_foreign(lle->lle_lsme);
701 /* new mirror detected, assume that the mirrors
702 * are shorted in layout */
705 if (j >= comp->lo_mirror_count)
708 lre = &comp->lo_mirrors[j];
711 /* entries must be sorted by mirrors */
712 lre->lre_mirror_id = mirror_id;
713 lre->lre_start = lre->lre_end = i;
714 lre->lre_preference = lle->lle_lsme->lsme_flags &
715 LCME_FL_PREF_RD ? 1000 : 0;
716 lre->lre_valid = lle->lle_valid;
717 lre->lre_stale = !lle->lle_valid;
718 lre->lre_foreign = lsme_is_foreign(lle->lle_lsme);
721 /* sanity check for FLR */
722 if (mirror_count != comp->lo_mirror_count) {
724 " doesn't have the # of mirrors it claims, %u/%u\n",
725 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
726 comp->lo_mirror_count + 1);
728 GOTO(out, result = -EINVAL);
731 lov_foreach_layout_entry(lov, lle) {
732 int index = lov_layout_entry_index(lov, lle);
735 * If the component has not been init-ed on MDS side, for
736 * PFL layout, we'd know that the components beyond this one
737 * will be dynamically init-ed later on file write/trunc ops.
739 if (!lsme_inited(lle->lle_lsme))
742 if (lsme_is_foreign(lle->lle_lsme))
745 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
750 LASSERT(ergo(psz > 0, psz == result));
755 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
757 /* decide the preferred mirror. It uses the hash value of lov_object
758 * so that different clients would use different mirrors for read. */
761 seq = hash_long((unsigned long)lov, 8);
762 for (i = 0; i < comp->lo_mirror_count; i++) {
763 unsigned int idx = (i + seq) % comp->lo_mirror_count;
765 lre = lov_mirror_entry(lov, idx);
769 if (lre->lre_foreign)
772 mirror_count++; /* valid mirror */
774 /* aggregated preference of all involved OSTs */
775 for (j = lre->lre_start; j <= lre->lre_end; j++) {
776 lre->lre_preference +=
777 comp->lo_entries[j].lle_preference;
780 if (lre->lre_preference > preference) {
781 preference = lre->lre_preference;
782 comp->lo_preferred_mirror = idx;
787 " doesn't have any valid mirrors\n",
788 PFID(lu_object_fid(lov2lu(lov))));
790 comp->lo_preferred_mirror = 0;
793 LASSERT(comp->lo_preferred_mirror >= 0);
797 return result > 0 ? 0 : result;
800 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
801 struct lov_object *lov, struct lov_stripe_md *lsm,
802 const struct cl_object_conf *conf,
803 union lov_layout_state *state)
808 static int lov_init_released(const struct lu_env *env,
809 struct lov_device *dev, struct lov_object *lov,
810 struct lov_stripe_md *lsm,
811 const struct cl_object_conf *conf,
812 union lov_layout_state *state)
814 LASSERT(lsm != NULL);
815 LASSERT(lsm->lsm_is_released);
816 LASSERT(lov->lo_lsm == NULL);
818 lov->lo_lsm = lsm_addref(lsm);
822 static int lov_init_foreign(const struct lu_env *env,
823 struct lov_device *dev, struct lov_object *lov,
824 struct lov_stripe_md *lsm,
825 const struct cl_object_conf *conf,
826 union lov_layout_state *state)
828 LASSERT(lsm != NULL);
829 LASSERT(lov->lo_type == LLT_FOREIGN);
830 LASSERT(lov->lo_lsm == NULL);
832 lov->lo_lsm = lsm_addref(lsm);
836 static int lov_delete_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 ||
840 lov->lo_type == LLT_FOREIGN);
842 lov_layout_wait(env, lov);
846 static int lov_delete_composite(const struct lu_env *env,
847 struct lov_object *lov,
848 union lov_layout_state *state)
850 struct lov_layout_entry *entry;
854 dump_lsm(D_INODE, lov->lo_lsm);
856 lov_layout_wait(env, lov);
857 lov_foreach_layout_entry(lov, entry) {
858 if (entry->lle_lsme && lsme_is_foreign(entry->lle_lsme))
861 lov_delete_raid0(env, lov, entry);
867 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
868 union lov_layout_state *state)
870 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
873 static void lov_fini_composite(const struct lu_env *env,
874 struct lov_object *lov,
875 union lov_layout_state *state)
877 struct lov_layout_composite *comp = &state->composite;
880 if (comp->lo_entries != NULL) {
881 struct lov_layout_entry *entry;
883 lov_foreach_layout_entry(lov, entry)
884 if (entry->lle_comp_ops)
885 entry->lle_comp_ops->lco_fini(env, entry);
887 OBD_FREE_PTR_ARRAY(comp->lo_entries, comp->lo_entry_count);
888 comp->lo_entries = NULL;
891 if (comp->lo_mirrors != NULL) {
892 OBD_FREE_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
893 comp->lo_mirrors = NULL;
896 memset(comp, 0, sizeof(*comp));
898 dump_lsm(D_INODE, lov->lo_lsm);
899 lov_free_memmd(&lov->lo_lsm);
904 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
905 union lov_layout_state *state)
908 dump_lsm(D_INODE, lov->lo_lsm);
909 lov_free_memmd(&lov->lo_lsm);
913 static int lov_print_empty(const struct lu_env *env, void *cookie,
914 lu_printer_t p, const struct lu_object *o)
916 (*p)(env, cookie, "empty %d\n",
917 test_bit(LO_LAYOUT_INVALID, &lu2lov(o)->lo_obj_flags));
921 static int lov_print_composite(const struct lu_env *env, void *cookie,
922 lu_printer_t p, const struct lu_object *o)
924 struct lov_object *lov = lu2lov(o);
925 struct lov_stripe_md *lsm = lov->lo_lsm;
928 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
929 lsm->lsm_entry_count,
930 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
931 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
932 lsm->lsm_layout_gen);
934 for (i = 0; i < lsm->lsm_entry_count; i++) {
935 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
936 struct lov_layout_entry *lle = lov_entry(lov, i);
939 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
940 PEXT(&lse->lsme_extent), lse->lsme_magic,
941 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
942 lse->lsme_flags, lse->lsme_stripe_count,
943 lse->lsme_stripe_size);
945 if (!lsme_is_foreign(lse))
946 lov_print_raid0(env, cookie, p, lle);
952 static int lov_print_released(const struct lu_env *env, void *cookie,
953 lu_printer_t p, const struct lu_object *o)
955 struct lov_object *lov = lu2lov(o);
956 struct lov_stripe_md *lsm = lov->lo_lsm;
959 "released: %s, lsm{%p 0x%08X %d %u}:\n",
960 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
961 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
962 lsm->lsm_layout_gen);
966 static int lov_print_foreign(const struct lu_env *env, void *cookie,
967 lu_printer_t p, const struct lu_object *o)
969 struct lov_object *lov = lu2lov(o);
970 struct lov_stripe_md *lsm = lov->lo_lsm;
973 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
974 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ?
975 "invalid" : "valid", lsm,
976 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
977 lsm->lsm_layout_gen);
979 "raw_ea_content '%.*s'\n",
980 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
985 * Implements cl_object_operations::coo_attr_get() method for an object
986 * without stripes (LLT_EMPTY layout type).
988 * The only attributes this layer is authoritative in this case is
989 * cl_attr::cat_blocks---it's 0.
991 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
992 struct cl_attr *attr)
994 attr->cat_blocks = 0;
998 static int lov_attr_get_composite(const struct lu_env *env,
999 struct cl_object *obj,
1000 struct cl_attr *attr)
1002 struct lov_object *lov = cl2lov(obj);
1003 struct lov_layout_entry *entry;
1009 attr->cat_blocks = 0;
1012 lov_foreach_layout_entry(lov, entry) {
1013 struct cl_attr *lov_attr = NULL;
1014 int index = lov_layout_entry_index(lov, entry);
1016 if (!entry->lle_valid)
1019 /* PFL: This component has not been init-ed. */
1020 if (!lsm_entry_inited(lov->lo_lsm, index))
1023 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1028 if (lov_attr == NULL)
1031 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1032 "b=%llu\n", index - 1, lov_attr->cat_size,
1033 lov_attr->cat_mtime, lov_attr->cat_atime,
1034 lov_attr->cat_ctime, lov_attr->cat_blocks);
1037 if (lov_attr->cat_kms_valid)
1038 attr->cat_kms_valid = 1;
1039 attr->cat_blocks += lov_attr->cat_blocks;
1040 if (attr->cat_size < lov_attr->cat_size)
1041 attr->cat_size = lov_attr->cat_size;
1042 if (attr->cat_kms < lov_attr->cat_kms)
1043 attr->cat_kms = lov_attr->cat_kms;
1044 if (attr->cat_atime < lov_attr->cat_atime)
1045 attr->cat_atime = lov_attr->cat_atime;
1046 if (attr->cat_ctime < lov_attr->cat_ctime)
1047 attr->cat_ctime = lov_attr->cat_ctime;
1048 if (attr->cat_mtime < lov_attr->cat_mtime)
1049 attr->cat_mtime = lov_attr->cat_mtime;
1055 static int lov_flush_composite(const struct lu_env *env,
1056 struct cl_object *obj,
1057 struct ldlm_lock *lock)
1059 struct lov_object *lov = cl2lov(obj);
1060 struct lov_layout_entry *lle;
1065 lov_foreach_layout_entry(lov, lle) {
1066 if (!lsme_is_dom(lle->lle_lsme))
1068 rc = cl_object_flush(env, lovsub2cl(lle->lle_dom.lo_dom), lock);
1075 static int lov_flush_empty(const struct lu_env *env, struct cl_object *obj,
1076 struct ldlm_lock *lock)
1081 const static struct lov_layout_operations lov_dispatch[] = {
1083 .llo_init = lov_init_empty,
1084 .llo_delete = lov_delete_empty,
1085 .llo_fini = lov_fini_empty,
1086 .llo_print = lov_print_empty,
1087 .llo_page_init = lov_page_init_empty,
1088 .llo_lock_init = lov_lock_init_empty,
1089 .llo_io_init = lov_io_init_empty,
1090 .llo_getattr = lov_attr_get_empty,
1091 .llo_flush = lov_flush_empty,
1094 .llo_init = lov_init_released,
1095 .llo_delete = lov_delete_empty,
1096 .llo_fini = lov_fini_released,
1097 .llo_print = lov_print_released,
1098 .llo_page_init = lov_page_init_empty,
1099 .llo_lock_init = lov_lock_init_empty,
1100 .llo_io_init = lov_io_init_released,
1101 .llo_getattr = lov_attr_get_empty,
1102 .llo_flush = lov_flush_empty,
1105 .llo_init = lov_init_composite,
1106 .llo_delete = lov_delete_composite,
1107 .llo_fini = lov_fini_composite,
1108 .llo_print = lov_print_composite,
1109 .llo_page_init = lov_page_init_composite,
1110 .llo_lock_init = lov_lock_init_composite,
1111 .llo_io_init = lov_io_init_composite,
1112 .llo_getattr = lov_attr_get_composite,
1113 .llo_flush = lov_flush_composite,
1116 .llo_init = lov_init_foreign,
1117 .llo_delete = lov_delete_empty,
1118 .llo_fini = lov_fini_released,
1119 .llo_print = lov_print_foreign,
1120 .llo_page_init = lov_page_init_foreign,
1121 .llo_lock_init = lov_lock_init_empty,
1122 .llo_io_init = lov_io_init_empty,
1123 .llo_getattr = lov_attr_get_empty,
1124 .llo_flush = lov_flush_empty,
1129 * Performs a double-dispatch based on the layout type of an object.
1131 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1133 struct lov_object *__obj = (obj); \
1134 enum lov_layout_type __llt; \
1136 __llt = __obj->lo_type; \
1137 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1138 lov_dispatch[__llt].op(__VA_ARGS__); \
1142 * Return lov_layout_type associated with a given lsm
1144 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1149 if (lsm->lsm_is_released)
1150 return LLT_RELEASED;
1152 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1153 lsm->lsm_magic == LOV_MAGIC_V3 ||
1154 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1157 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1163 static inline void lov_conf_freeze(struct lov_object *lov)
1165 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1166 lov, lov->lo_owner, current);
1167 if (lov->lo_owner != current)
1168 down_read(&lov->lo_type_guard);
1171 static inline void lov_conf_thaw(struct lov_object *lov)
1173 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1174 lov, lov->lo_owner, current);
1175 if (lov->lo_owner != current)
1176 up_read(&lov->lo_type_guard);
1179 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1181 struct lov_object *__obj = (obj); \
1182 int __lock = !!(lock); \
1183 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1186 lov_conf_freeze(__obj); \
1187 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1189 lov_conf_thaw(__obj); \
1194 * Performs a locked double-dispatch based on the layout type of an object.
1196 #define LOV_2DISPATCH(obj, op, ...) \
1197 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1199 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1201 struct lov_object *__obj = (obj); \
1202 enum lov_layout_type __llt; \
1204 lov_conf_freeze(__obj); \
1205 __llt = __obj->lo_type; \
1206 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1207 lov_dispatch[__llt].op(__VA_ARGS__); \
1208 lov_conf_thaw(__obj); \
1211 static void lov_conf_lock(struct lov_object *lov)
1213 LASSERT(lov->lo_owner != current);
1214 down_write(&lov->lo_type_guard);
1215 LASSERT(lov->lo_owner == NULL);
1216 lov->lo_owner = current;
1217 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1218 lov, lov->lo_owner);
1221 static void lov_conf_unlock(struct lov_object *lov)
1223 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1224 lov, lov->lo_owner);
1225 lov->lo_owner = NULL;
1226 up_write(&lov->lo_type_guard);
1229 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1233 while (atomic_read(&lov->lo_active_ios) > 0) {
1234 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1235 PFID(lu_object_fid(lov2lu(lov))),
1236 atomic_read(&lov->lo_active_ios));
1238 wait_event_idle(lov->lo_waitq,
1239 atomic_read(&lov->lo_active_ios) == 0);
1244 static int lov_layout_change(const struct lu_env *unused,
1245 struct lov_object *lov, struct lov_stripe_md *lsm,
1246 const struct cl_object_conf *conf)
1248 enum lov_layout_type llt = lov_type(lsm);
1249 union lov_layout_state *state = &lov->u;
1250 const struct lov_layout_operations *old_ops;
1251 const struct lov_layout_operations *new_ops;
1252 struct lov_device *lov_dev = lov_object_dev(lov);
1258 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1260 env = cl_env_get(&refcheck);
1262 RETURN(PTR_ERR(env));
1264 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1266 CDEBUG(D_INODE, DFID" from %s to %s\n",
1267 PFID(lu_object_fid(lov2lu(lov))),
1268 llt2str(lov->lo_type), llt2str(llt));
1270 old_ops = &lov_dispatch[lov->lo_type];
1271 new_ops = &lov_dispatch[llt];
1273 rc = cl_object_prune(env, &lov->lo_cl);
1277 rc = old_ops->llo_delete(env, lov, &lov->u);
1281 old_ops->llo_fini(env, lov, &lov->u);
1283 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1285 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1286 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1288 /* page bufsize fixup */
1289 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1290 lov_page_slice_fixup(lov, NULL);
1293 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1295 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1297 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1298 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1299 new_ops->llo_delete(env, lov, state);
1300 new_ops->llo_fini(env, lov, state);
1301 /* this file becomes an EMPTY file. */
1302 lov->lo_type = LLT_EMPTY;
1307 cl_env_put(env, &refcheck);
1311 /*****************************************************************************
1313 * Lov object operations.
1316 static int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1317 const struct lu_object_conf *conf)
1319 struct lov_object *lov = lu2lov(obj);
1320 struct lov_device *dev = lov_object_dev(lov);
1321 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1322 union lov_layout_state *set = &lov->u;
1323 const struct lov_layout_operations *ops;
1324 struct lov_stripe_md *lsm = NULL;
1328 init_rwsem(&lov->lo_type_guard);
1329 atomic_set(&lov->lo_active_ios, 0);
1330 init_waitqueue_head(&lov->lo_waitq);
1331 cl_object_page_init(lu2cl(obj), 0);
1333 lov->lo_type = LLT_EMPTY;
1334 if (cconf->u.coc_layout.lb_buf != NULL) {
1335 lsm = lov_unpackmd(dev->ld_lov,
1336 cconf->u.coc_layout.lb_buf,
1337 cconf->u.coc_layout.lb_len);
1339 RETURN(PTR_ERR(lsm));
1341 dump_lsm(D_INODE, lsm);
1344 /* no locking is necessary, as object is being created */
1345 lov->lo_type = lov_type(lsm);
1346 ops = &lov_dispatch[lov->lo_type];
1347 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1357 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1358 const struct cl_object_conf *conf)
1360 struct lov_stripe_md *lsm = NULL;
1361 struct lov_object *lov = cl2lov(obj);
1365 if (conf->coc_opc == OBJECT_CONF_SET &&
1366 conf->u.coc_layout.lb_buf != NULL) {
1367 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1368 conf->u.coc_layout.lb_buf,
1369 conf->u.coc_layout.lb_len);
1371 RETURN(PTR_ERR(lsm));
1372 dump_lsm(D_INODE, lsm);
1375 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1376 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1377 GOTO(out_lsm, result = 0);
1381 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1382 if (test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) &&
1383 atomic_read(&lov->lo_active_ios) > 0) {
1384 lov_conf_unlock(lov);
1385 result = lov_layout_wait(env, lov);
1391 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1393 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1394 ((lsm != NULL && lov->lo_lsm != NULL) &&
1395 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1396 (lov->lo_lsm->lsm_flags == lsm->lsm_flags) &&
1397 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1398 lsm->lsm_entries[0]->lsme_pattern))) {
1399 /* same version of layout */
1400 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1401 GOTO(out, result = 0);
1404 /* will change layout - check if there still exists active IO. */
1405 if (atomic_read(&lov->lo_active_ios) > 0) {
1406 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1407 GOTO(out, result = -EBUSY);
1410 result = lov_layout_change(env, lov, lsm, conf);
1412 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1414 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1418 lov_conf_unlock(lov);
1421 CDEBUG(D_INODE, DFID" lo_layout_invalid=%u\n",
1422 PFID(lu_object_fid(lov2lu(lov))),
1423 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags));
1427 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1429 struct lov_object *lov = lu2lov(obj);
1432 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1436 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1438 struct lov_object *lov = lu2lov(obj);
1441 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1442 lu_object_fini(obj);
1443 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1447 static int lov_object_print(const struct lu_env *env, void *cookie,
1448 lu_printer_t p, const struct lu_object *o)
1450 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1453 static int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1454 struct cl_page *page, pgoff_t index)
1456 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1461 * Implements cl_object_operations::clo_io_init() method for lov
1462 * layer. Dispatches to the appropriate layout io initialization method.
1464 static int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1467 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1469 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1470 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1471 io->ci_ignore_layout, io->ci_verify_layout);
1473 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1474 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1475 * because as long as the OSC object exists, the layout can't be
1477 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1478 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1483 * An implementation of cl_object_operations::clo_attr_get() method for lov
1484 * layer. For raid0 layout this collects and merges attributes of all
1487 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1488 struct cl_attr *attr)
1490 /* do not take lock, as this function is called under a
1491 * spin-lock. Layout is protected from changing by ongoing IO. */
1492 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1495 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1496 const struct cl_attr *attr, unsigned valid)
1499 * No dispatch is required here, as no layout implements this.
1504 static int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1505 struct cl_lock *lock, const struct cl_io *io)
1507 /* No need to lock because we've taken one refcount of layout. */
1508 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1513 * We calculate on which OST the mapping will end. If the length of mapping
1514 * is greater than (stripe_size * stripe_count) then the last_stripe will
1515 * will be one just before start_stripe. Else we check if the mapping
1516 * intersects each OST and find last_stripe.
1517 * This function returns the last_stripe and also sets the stripe_count
1518 * over which the mapping is spread
1520 * \param lsm [in] striping information for the file
1521 * \param index [in] stripe component index
1522 * \param ext [in] logical extent of mapping
1523 * \param start_stripe [in] starting stripe of the mapping
1524 * \param stripe_count [out] the number of stripes across which to map is
1527 * \retval last_stripe return the last stripe of the mapping
1529 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1530 struct lu_extent *ext,
1531 int start_stripe, int *stripe_count)
1533 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1538 init_stripe = lov_stripe_number(lsm, index, ext->e_start);
1540 if (ext->e_end - ext->e_start >
1541 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1542 if (init_stripe == start_stripe) {
1543 last_stripe = (start_stripe < 1) ?
1544 lsme->lsme_stripe_count - 1 : start_stripe - 1;
1545 *stripe_count = lsme->lsme_stripe_count;
1546 } else if (init_stripe < start_stripe) {
1547 last_stripe = (init_stripe < 1) ?
1548 lsme->lsme_stripe_count - 1 : init_stripe - 1;
1549 *stripe_count = lsme->lsme_stripe_count -
1550 (start_stripe - init_stripe);
1552 last_stripe = init_stripe - 1;
1553 *stripe_count = init_stripe - start_stripe;
1556 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1557 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1558 if (!lov_stripe_intersects(lsm, index, i, ext, NULL,
1561 if ((start_stripe != init_stripe) && (i == init_stripe))
1565 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1572 * Set fe_device and copy extents from local buffer into main return buffer.
1574 * \param fiemap [out] fiemap to hold all extents
1575 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1576 * \param ost_index [in] OST index to be written into the fm_device
1577 * field for each extent
1578 * \param ext_count [in] number of extents to be copied
1579 * \param current_extent [in] where to start copying in the extent array
1581 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1582 struct fiemap_extent *lcl_fm_ext,
1583 int ost_index, unsigned int ext_count,
1584 int current_extent, int abs_stripeno)
1589 for (ext = 0; ext < ext_count; ext++) {
1590 set_fe_device_stripenr(&lcl_fm_ext[ext], ost_index,
1592 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1595 /* Copy fm_extent's from fm_local to return buffer */
1596 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1597 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1600 #define FIEMAP_BUFFER_SIZE 4096
1603 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1604 * call. The local end offset and the device are sent in the first
1605 * fm_extent. This function calculates the stripe number from the index.
1606 * This function returns a stripe_no on which mapping is to be restarted.
1608 * This function returns fm_end_offset which is the in-OST offset at which
1609 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1610 * will re-calculate proper offset in next stripe.
1611 * Note that the first extent is passed to lov_get_info via the value field.
1613 * \param fiemap [in] fiemap request header
1614 * \param lsm [in] striping information for the file
1615 * \param index [in] stripe component index
1616 * \param ext [in] logical extent of mapping
1617 * \param start_stripe [out] starting stripe will be returned in this
1619 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1620 struct lov_stripe_md *lsm,
1621 int index, struct lu_extent *ext,
1624 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1625 u64 local_end = fiemap->fm_extents[0].fe_logical;
1630 if (fiemap->fm_extent_count == 0 ||
1631 fiemap->fm_extents[0].fe_logical == 0)
1634 stripe_no = *start_stripe;
1636 if (stripe_no == -1)
1639 /* If we have finished mapping on previous device, shift logical
1640 * offset to start of next device */
1641 if (lov_stripe_intersects(lsm, index, stripe_no, ext, NULL, &lun_end) &&
1642 local_end < lun_end) {
1643 fm_end_offset = local_end;
1645 /* This is a special value to indicate that caller should
1646 * calculate offset in next stripe. */
1648 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1651 return fm_end_offset;
1654 struct fiemap_state {
1655 struct fiemap *fs_fm;
1656 struct lu_extent fs_ext; /* current entry extent */
1658 u64 fs_end_offset; /* last iteration offset */
1659 int fs_cur_extent; /* collected exts so far */
1660 int fs_cnt_need; /* # of extents buf can hold */
1661 int fs_start_stripe;
1663 bool fs_device_done; /* enough for this OST */
1664 bool fs_finish_stripe; /* reached fs_last_stripe */
1665 bool fs_enough; /* enough for this call */
1668 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1669 struct lov_object *lov,
1670 struct lov_stripe_md *lsm,
1673 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1674 struct lov_thread_info *lti = lov_env_info(env);
1675 struct lu_fid *ofid = <i->lti_fid;
1676 struct lov_oinfo *oinfo;
1677 struct cl_device *subdev;
1678 int entry = lov_comp_entry(index);
1679 int stripe = lov_comp_stripe(index);
1682 struct cl_object *result;
1684 if (lov->lo_type != LLT_COMP)
1685 GOTO(out, result = NULL);
1687 if (entry >= lsm->lsm_entry_count ||
1688 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1689 GOTO(out, result = NULL);
1691 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1692 ost_idx = oinfo->loi_ost_idx;
1693 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1695 GOTO(out, result = NULL);
1697 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1698 result = lov_sub_find(env, subdev, ofid, NULL);
1701 result = ERR_PTR(-EINVAL);
1705 static int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1706 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1707 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1708 int index, int stripe_last, int stripeno,
1709 struct fiemap_state *fs)
1711 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1712 struct cl_object *subobj;
1713 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1714 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1715 u64 req_fm_len; /* max requested extent coverage */
1716 u64 len_mapped_single_call;
1719 unsigned int ext_count;
1720 /* EOF for object */
1721 bool ost_eof = false;
1722 /* done with required mapping for this OST? */
1723 bool ost_done = false;
1727 fs->fs_device_done = false;
1728 /* Find out range of mapping on this stripe */
1729 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1730 &obd_start, &obd_end)) == 0)
1733 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1736 /* If this is a continuation FIEMAP call and we are on
1737 * starting stripe then obd_start needs to be set to
1739 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1740 obd_start = fs->fs_end_offset;
1742 if (lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno) ==
1746 req_fm_len = obd_end - obd_start + 1;
1747 fs->fs_fm->fm_length = 0;
1748 len_mapped_single_call = 0;
1750 /* find lobsub object */
1751 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1752 lov_comp_index(index, stripeno));
1754 return PTR_ERR(subobj);
1755 /* If the output buffer is very large and the objects have many
1756 * extents we may need to loop on a single OST repeatedly */
1758 if (fiemap->fm_extent_count > 0) {
1759 /* Don't get too many extents. */
1760 if (fs->fs_cur_extent + fs->fs_cnt_need >
1761 fiemap->fm_extent_count)
1762 fs->fs_cnt_need = fiemap->fm_extent_count -
1766 obd_start += len_mapped_single_call;
1767 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1768 req_fm_len = fs->fs_fm->fm_length;
1770 * If we've collected enough extent map, we'd request 1 more,
1771 * to see whether we coincidentally finished all available
1772 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1774 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1775 1 : fs->fs_cnt_need;
1776 fs->fs_fm->fm_mapped_extents = 0;
1777 fs->fs_fm->fm_flags = fiemap->fm_flags;
1779 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1781 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1782 GOTO(obj_put, rc = -EINVAL);
1783 /* If OST is inactive, return extent with UNKNOWN flag. */
1784 if (!lov->lov_tgts[ost_index]->ltd_active) {
1785 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1786 fs->fs_fm->fm_mapped_extents = 1;
1788 fm_ext[0].fe_logical = obd_start;
1789 fm_ext[0].fe_length = obd_end - obd_start + 1;
1790 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1795 fs->fs_fm->fm_start = obd_start;
1796 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1797 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1798 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1800 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1804 ext_count = fs->fs_fm->fm_mapped_extents;
1805 if (ext_count == 0) {
1807 fs->fs_device_done = true;
1808 /* If last stripe has hold at the end,
1809 * we need to return */
1810 if (stripeno == fs->fs_last_stripe) {
1811 fiemap->fm_mapped_extents = 0;
1812 fs->fs_finish_stripe = true;
1816 } else if (fs->fs_enough) {
1818 * We've collected enough extents and there are
1819 * more extents after it.
1824 /* If we just need num of extents, got to next device */
1825 if (fiemap->fm_extent_count == 0) {
1826 fs->fs_cur_extent += ext_count;
1830 /* prepare to copy retrived map extents */
1831 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1832 fm_ext[ext_count - 1].fe_length -
1835 /* Have we finished mapping on this device? */
1836 if (req_fm_len <= len_mapped_single_call) {
1838 fs->fs_device_done = true;
1841 /* Clear the EXTENT_LAST flag which can be present on
1842 * the last extent */
1843 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1844 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1845 if (lov_stripe_size(lsm, index,
1846 fm_ext[ext_count - 1].fe_logical +
1847 fm_ext[ext_count - 1].fe_length,
1848 stripeno) >= fmkey->lfik_oa.o_size) {
1850 fs->fs_device_done = true;
1853 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1854 ext_count, fs->fs_cur_extent,
1855 stripe_last + stripeno);
1856 fs->fs_cur_extent += ext_count;
1858 /* Ran out of available extents? */
1859 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1860 fs->fs_enough = true;
1861 } while (!ost_done && !ost_eof);
1863 if (stripeno == fs->fs_last_stripe)
1864 fs->fs_finish_stripe = true;
1866 cl_object_put(env, subobj);
1872 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1873 * This also handles the restarting of FIEMAP calls in case mapping overflows
1874 * the available number of extents in single call.
1876 * \param env [in] lustre environment
1877 * \param obj [in] file object
1878 * \param fmkey [in] fiemap request header and other info
1879 * \param fiemap [out] fiemap buffer holding retrived map extents
1880 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1881 * each OST, it is used to limit max map needed
1885 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1886 struct ll_fiemap_info_key *fmkey,
1887 struct fiemap *fiemap, size_t *buflen)
1889 struct lov_stripe_md_entry *lsme;
1890 struct lov_stripe_md *lsm;
1891 struct fiemap *fm_local = NULL;
1895 int start_entry = -1;
1899 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1901 struct fiemap_state fs = { 0 };
1902 struct lu_extent range;
1905 int start_stripe = 0;
1906 bool resume = false;
1909 lsm = lov_lsm_addref(cl2lov(obj));
1911 /* no extent: there is no object for mapping */
1912 fiemap->fm_mapped_extents = 0;
1916 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1918 * If the entry count > 1 or stripe_count > 1 and the
1919 * application does not understand DEVICE_ORDER flag,
1920 * it cannot interpret the extents correctly.
1922 if (lsm->lsm_entry_count > 1 ||
1923 (lsm->lsm_entry_count == 1 &&
1924 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1925 GOTO(out_lsm, rc = -ENOTSUPP);
1928 /* No support for DOM layout yet. */
1929 if (lsme_is_dom(lsm->lsm_entries[0]))
1930 GOTO(out_lsm, rc = -ENOTSUPP);
1932 if (lsm->lsm_is_released) {
1933 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1935 * released file, return a minimal FIEMAP if
1936 * request fits in file-size.
1938 fiemap->fm_mapped_extents = 1;
1939 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1940 if (fiemap->fm_start + fiemap->fm_length <
1941 fmkey->lfik_oa.o_size)
1942 fiemap->fm_extents[0].fe_length =
1945 fiemap->fm_extents[0].fe_length =
1946 fmkey->lfik_oa.o_size -
1948 fiemap->fm_extents[0].fe_flags |=
1949 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1951 GOTO(out_lsm, rc = 0);
1954 /* buffer_size is small to hold fm_extent_count of extents. */
1955 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1956 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1958 OBD_ALLOC_LARGE(fm_local, buffer_size);
1959 if (fm_local == NULL)
1960 GOTO(out_lsm, rc = -ENOMEM);
1963 * Requested extent count exceeds the fiemap buffer size, shrink our
1966 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1967 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1969 fs.fs_enough = false;
1970 fs.fs_cur_extent = 0;
1971 fs.fs_fm = fm_local;
1972 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1974 whole_start = fiemap->fm_start;
1975 /* whole_start is beyond the end of the file */
1976 if (whole_start > fmkey->lfik_oa.o_size)
1977 GOTO(out_fm_local, rc = -EINVAL);
1978 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1979 fmkey->lfik_oa.o_size + 1 :
1980 whole_start + fiemap->fm_length;
1982 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
1985 if (whole_end > fmkey->lfik_oa.o_size + 1)
1986 whole_end = fmkey->lfik_oa.o_size + 1;
1989 * the high 16bits of fe_device remember which stripe the last
1990 * call has been arrived, we'd continue from there in this call.
1992 if (fiemap->fm_extent_count && fiemap->fm_extents[0].fe_logical)
1994 stripe_last = get_fe_stripenr(&fiemap->fm_extents[0]);
1996 * stripe_last records stripe number we've been processed in the last
1999 end_entry = lsm->lsm_entry_count - 1;
2001 for (entry = 0; entry <= end_entry; entry++) {
2002 lsme = lsm->lsm_entries[entry];
2003 if (cur_stripe + lsme->lsme_stripe_count >= stripe_last) {
2004 start_entry = entry;
2005 start_stripe = stripe_last - cur_stripe;
2009 cur_stripe += lsme->lsme_stripe_count;
2011 if (start_entry == -1) {
2012 CERROR(DFID": FIEMAP does not init start entry, cur_stripe=%d, "
2013 "stripe_last=%d\n", PFID(lu_object_fid(&obj->co_lu)),
2014 cur_stripe, stripe_last);
2015 GOTO(out_fm_local, rc = -EINVAL);
2018 * @start_entry & @start_stripe records the position of fiemap
2019 * resumption @stripe_last keeps recording the absolution position
2020 * we'are processing. @resume indicates we'd honor @start_stripe.
2023 range.e_start = whole_start;
2024 range.e_end = whole_end;
2026 for (entry = start_entry; entry <= end_entry; entry++) {
2027 /* remeber to update stripe_last accordingly */
2028 lsme = lsm->lsm_entries[entry];
2030 /* FLR could contain component holes between entries */
2031 if (!lsme_inited(lsme)) {
2032 stripe_last += lsme->lsme_stripe_count;
2037 if (!lu_extent_is_overlapped(&range, &lsme->lsme_extent)) {
2038 stripe_last += lsme->lsme_stripe_count;
2043 /* prepare for a component entry iteration */
2044 if (lsme->lsme_extent.e_start > whole_start)
2045 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
2047 fs.fs_ext.e_start = whole_start;
2048 if (lsme->lsme_extent.e_end > whole_end)
2049 fs.fs_ext.e_end = whole_end;
2051 fs.fs_ext.e_end = lsme->lsme_extent.e_end;
2053 /* Calculate start stripe, last stripe and length of mapping */
2055 fs.fs_start_stripe = start_stripe;
2056 /* put stripe_last to the first stripe of the comp */
2057 stripe_last -= start_stripe;
2060 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
2063 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
2064 &fs.fs_ext, fs.fs_start_stripe,
2067 * A new mirror component is under process, reset
2068 * fs.fs_end_offset and then fiemap_for_stripe() starts from
2069 * the overlapping extent, otherwise starts from
2072 if (entry > start_entry && lsme->lsme_extent.e_start == 0) {
2074 fs.fs_end_offset = 0;
2076 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap,
2077 lsm, entry, &fs.fs_ext,
2078 &fs.fs_start_stripe);
2081 /* Check each stripe */
2082 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
2084 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
2085 /* reset fs_finish_stripe */
2086 fs.fs_finish_stripe = false;
2087 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
2088 fmkey, entry, stripe_last,
2091 GOTO(out_fm_local, rc);
2093 stripe_last += cur_stripe;
2096 if (fs.fs_finish_stripe)
2098 } /* for each stripe */
2099 stripe_last += lsme->lsme_stripe_count;
2100 } /* for covering layout component entry */
2103 if (fs.fs_cur_extent > 0)
2104 cur_ext = fs.fs_cur_extent - 1;
2108 /* done all the processing */
2109 if (entry > end_entry)
2110 fiemap->fm_extents[cur_ext].fe_flags |= FIEMAP_EXTENT_LAST;
2112 /* Indicate that we are returning device offsets unless file just has
2114 if (lsm->lsm_entry_count > 1 ||
2115 (lsm->lsm_entry_count == 1 &&
2116 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2117 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2119 if (fiemap->fm_extent_count == 0)
2120 goto skip_last_device_calc;
2122 skip_last_device_calc:
2123 fiemap->fm_mapped_extents = fs.fs_cur_extent;
2125 OBD_FREE_LARGE(fm_local, buffer_size);
2132 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2133 struct lov_user_md __user *lum, size_t size)
2135 struct lov_object *lov = cl2lov(obj);
2136 struct lov_stripe_md *lsm;
2140 lsm = lov_lsm_addref(lov);
2144 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2149 static int lov_object_layout_get(const struct lu_env *env,
2150 struct cl_object *obj,
2151 struct cl_layout *cl)
2153 struct lov_object *lov = cl2lov(obj);
2154 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2155 struct lu_buf *buf = &cl->cl_buf;
2161 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2166 cl->cl_size = lov_comp_md_size(lsm);
2167 cl->cl_layout_gen = lsm->lsm_layout_gen;
2168 cl->cl_is_released = lsm->lsm_is_released;
2169 cl->cl_is_composite = lsm_is_composite(lsm->lsm_magic);
2171 rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2174 /* return error or number of bytes */
2178 static loff_t lov_object_maxbytes(struct cl_object *obj)
2180 struct lov_object *lov = cl2lov(obj);
2181 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2187 maxbytes = lsm->lsm_maxbytes;
2194 static int lov_object_flush(const struct lu_env *env, struct cl_object *obj,
2195 struct ldlm_lock *lock)
2197 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_flush, true, env, obj,
2201 static const struct cl_object_operations lov_ops = {
2202 .coo_page_init = lov_page_init,
2203 .coo_lock_init = lov_lock_init,
2204 .coo_io_init = lov_io_init,
2205 .coo_attr_get = lov_attr_get,
2206 .coo_attr_update = lov_attr_update,
2207 .coo_conf_set = lov_conf_set,
2208 .coo_getstripe = lov_object_getstripe,
2209 .coo_layout_get = lov_object_layout_get,
2210 .coo_maxbytes = lov_object_maxbytes,
2211 .coo_fiemap = lov_object_fiemap,
2212 .coo_object_flush = lov_object_flush
2215 static const struct lu_object_operations lov_lu_obj_ops = {
2216 .loo_object_init = lov_object_init,
2217 .loo_object_delete = lov_object_delete,
2218 .loo_object_release = NULL,
2219 .loo_object_free = lov_object_free,
2220 .loo_object_print = lov_object_print,
2221 .loo_object_invariant = NULL,
2224 struct lu_object *lov_object_alloc(const struct lu_env *env,
2225 const struct lu_object_header *unused,
2226 struct lu_device *dev)
2228 struct lov_object *lov;
2229 struct lu_object *obj;
2232 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2235 lu_object_init(obj, NULL, dev);
2236 lov->lo_cl.co_ops = &lov_ops;
2237 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2239 * object io operation vector (cl_object::co_iop) is installed
2240 * later in lov_object_init(), as different vectors are used
2241 * for object with different layouts.
2243 obj->lo_ops = &lov_lu_obj_ops;
2249 static struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2251 struct lov_stripe_md *lsm = NULL;
2253 lov_conf_freeze(lov);
2254 if (lov->lo_lsm != NULL) {
2255 lsm = lsm_addref(lov->lo_lsm);
2256 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2257 lsm, atomic_read(&lsm->lsm_refc),
2258 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags),
2265 int lov_read_and_clear_async_rc(struct cl_object *clob)
2267 struct lu_object *luobj;
2271 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2273 if (luobj != NULL) {
2274 struct lov_object *lov = lu2lov(luobj);
2276 lov_conf_freeze(lov);
2277 switch (lov->lo_type) {
2279 struct lov_stripe_md *lsm;
2283 LASSERT(lsm != NULL);
2284 for (i = 0; i < lsm->lsm_entry_count; i++) {
2285 struct lov_stripe_md_entry *lse =
2286 lsm->lsm_entries[i];
2289 if (!lsme_inited(lse))
2292 for (j = 0; j < lse->lsme_stripe_count; j++) {
2293 struct lov_oinfo *loi =
2296 if (lov_oinfo_is_dummy(loi))
2299 if (loi->loi_ar.ar_rc && !rc)
2300 rc = loi->loi_ar.ar_rc;
2301 loi->loi_ar.ar_rc = 0;
2317 EXPORT_SYMBOL(lov_read_and_clear_async_rc);