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 (CFS_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 stripe = lov_sub_find(env, subdev, ofid, subconf);
262 GOTO(out, result = PTR_ERR(stripe));
264 result = lov_init_sub(env, lov, stripe, oinfo,
265 lov_comp_index(index, i));
266 if (result == -EAGAIN) { /* try again */
273 r0->lo_sub[i] = cl2lovsub(stripe);
275 sz = lov_page_slice_fixup(lov, stripe);
276 LASSERT(ergo(psz > 0, psz == sz));
286 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
287 struct lov_layout_raid0 *r0,
288 struct lovsub_object *los, int idx)
290 struct cl_object *sub;
291 struct lu_site *site;
292 wait_queue_head_t *wq;
294 LASSERT(r0->lo_sub[idx] == los);
296 sub = lovsub2cl(los);
297 site = sub->co_lu.lo_dev->ld_site;
298 wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
300 cl_object_kill(env, sub);
301 /* release a reference to the sub-object and ... */
302 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
303 cl_object_put(env, sub);
305 /* ... wait until it is actually destroyed---sub-object clears its
306 * ->lo_sub[] slot in lovsub_object_free() */
307 wait_event(*wq, r0->lo_sub[idx] != los);
308 LASSERT(r0->lo_sub[idx] == NULL);
311 static int lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
312 struct lov_layout_entry *lle)
314 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 rc = cl_object_prune(env, &los->lso_cl);
330 * If top-level object is to be evicted from
331 * the cache, so are its sub-objects.
333 lov_subobject_kill(env, lov, r0, los, i);
341 static void lov_fini_raid0(const struct lu_env *env,
342 struct lov_layout_entry *lle)
344 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
346 if (r0->lo_sub != NULL) {
347 OBD_FREE_PTR_ARRAY_LARGE(r0->lo_sub, r0->lo_nr);
352 static int lov_print_raid0(const struct lu_env *env, void *cookie,
353 lu_printer_t p, const struct lov_layout_entry *lle)
355 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
358 for (i = 0; i < r0->lo_nr; ++i) {
359 struct lu_object *sub;
361 if (r0->lo_sub[i] != NULL) {
362 sub = lovsub2lu(r0->lo_sub[i]);
363 lu_object_print(env, cookie, p, sub);
365 (*p)(env, cookie, "sub %d absent\n", i);
371 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
372 unsigned int index, struct lov_layout_entry *lle,
373 struct cl_attr **lov_attr)
375 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
376 struct lov_stripe_md *lsm = lov->lo_lsm;
377 struct cl_attr *attr = &r0->lo_attr;
380 if (r0->lo_attr_valid) {
386 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
387 * happy. It's not needed, because new code uses
388 * ->coh_attr_guard spin-lock to protect consistency of
389 * sub-object attributes.
391 lov_stripe_lock(lsm);
392 result = lov_merge_lvb_kms(lsm, index, attr);
393 lov_stripe_unlock(lsm);
395 r0->lo_attr_valid = 1;
402 static struct lov_comp_layout_entry_ops raid0_ops = {
403 .lco_init = lov_init_raid0,
404 .lco_fini = lov_fini_raid0,
405 .lco_getattr = lov_attr_get_raid0,
408 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
409 unsigned int index, struct lov_layout_entry *lle,
410 struct cl_attr **lov_attr)
412 struct lov_layout_dom *dom = &lle->lle_dom;
413 struct lov_oinfo *loi = dom->lo_loi;
414 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
416 if (dom->lo_dom_r0.lo_attr_valid) {
421 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
422 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
424 cl_lvb2attr(attr, &loi->loi_lvb);
426 /* DoM component size can be bigger than stripe size after
427 * client's setattr RPC, so do not count anything beyond
428 * component end. Alternatively, check that limit on server
429 * and do not allow size overflow there. */
430 if (attr->cat_size > lle->lle_extent->e_end)
431 attr->cat_size = lle->lle_extent->e_end;
433 attr->cat_kms = attr->cat_size;
435 dom->lo_dom_r0.lo_attr_valid = 1;
442 * Lookup FLD to get MDS index of the given DOM object FID.
444 * \param[in] ld LOV device
445 * \param[in] fid FID to lookup
446 * \param[out] nr index in MDC array to return back
448 * \retval 0 and \a mds filled with MDS index if successful
449 * \retval negative value on error
451 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
459 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
460 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
462 CERROR("%s: error while looking for mds number. Seq %#llx"
463 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
468 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
471 /* find proper MDC device in the array */
472 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
473 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
474 ld->ld_md_tgts[i].ldm_idx == mds_idx)
478 if (i == ld->ld_md_tgts_nr) {
479 CERROR("%s: cannot find corresponding MDC device for mds #%x "
480 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
490 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
492 * Init the DOM object for the first time. It prepares also RAID0 entry
493 * for it to use in common methods with ordinary RAID0 layout entries.
495 * \param[in] env execution environment
496 * \param[in] dev LOV device
497 * \param[in] lov LOV object
498 * \param[in] index Composite layout entry index in LSM
499 * \param[in] lle Composite LOV layout entry
501 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
502 struct lov_object *lov, unsigned int index,
503 const struct cl_object_conf *conf,
504 struct lov_layout_entry *lle)
506 struct lov_thread_info *lti = lov_env_info(env);
507 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
508 struct cl_object *clo;
509 struct lu_object *o = lov2lu(lov);
510 const struct lu_fid *fid = lu_object_fid(o);
511 struct cl_device *mdcdev;
512 struct lov_oinfo *loi = NULL;
513 struct cl_object_conf *sconf = <i->lti_stripe_conf;
519 /* DOM entry may be not zero index due to FLR but must start from 0 */
520 if (unlikely(lle->lle_extent->e_start != 0)) {
521 CERROR("%s: DOM entry must be the first stripe in a mirror\n",
522 lov2obd(dev->ld_lov)->obd_name);
523 dump_lsm(D_ERROR, lov->lo_lsm);
527 /* find proper MDS device */
528 rc = lov_fld_lookup(dev, fid, &idx);
532 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
533 "LOV md target[%u] is NULL\n", idx);
535 /* check lsm is DOM, more checks are needed */
536 LASSERT(lsme->lsme_stripe_count == 0);
539 * Create lower cl_objects.
541 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
543 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
545 /* DoM object has no oinfo in LSM entry, create it exclusively */
546 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
550 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
552 sconf->u.coc_oinfo = loi;
554 clo = lov_sub_find(env, mdcdev, fid, sconf);
556 GOTO(out, rc = PTR_ERR(clo));
558 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
559 if (rc == -EAGAIN) /* try again */
564 lle->lle_dom.lo_dom = cl2lovsub(clo);
565 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
566 lle->lle_dom.lo_dom_r0.lo_nr = 1;
567 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
568 lle->lle_dom.lo_loi = loi;
570 rc = lov_page_slice_fixup(lov, clo);
575 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
580 * Implementation of lov_layout_operations::llo_fini for DOM object.
582 * Finish the DOM object and free related memory.
584 * \param[in] env execution environment
585 * \param[in] lov LOV object
586 * \param[in] state LOV layout state
588 static void lov_fini_dom(const struct lu_env *env,
589 struct lov_layout_entry *lle)
591 if (lle->lle_dom.lo_dom != NULL)
592 lle->lle_dom.lo_dom = NULL;
593 if (lle->lle_dom.lo_loi != NULL)
594 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
597 static struct lov_comp_layout_entry_ops dom_ops = {
598 .lco_init = lov_init_dom,
599 .lco_fini = lov_fini_dom,
600 .lco_getattr = lov_attr_get_dom,
603 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
604 struct lov_object *lov, struct lov_stripe_md *lsm,
605 const struct cl_object_conf *conf,
606 union lov_layout_state *state)
608 struct lov_layout_composite *comp = &state->composite;
609 struct lov_layout_entry *lle;
610 struct lov_mirror_entry *lre;
611 unsigned int entry_count;
612 unsigned int psz = 0;
613 unsigned int mirror_count;
614 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
617 int i, j, preference;
622 LASSERT(lsm->lsm_entry_count > 0);
623 LASSERT(lov->lo_lsm == NULL);
624 lov->lo_lsm = lsm_addref(lsm);
625 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
627 dump_lsm(D_INODE, lsm);
629 entry_count = lsm->lsm_entry_count;
631 comp->lo_flags = lsm->lsm_flags;
632 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
633 comp->lo_entry_count = lsm->lsm_entry_count;
634 comp->lo_preferred_mirror = -1;
636 if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
639 OBD_ALLOC_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
640 if (comp->lo_mirrors == NULL)
643 OBD_ALLOC_PTR_ARRAY(comp->lo_entries, entry_count);
644 if (comp->lo_entries == NULL)
647 /* Initiate all entry types and extents data at first */
648 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
651 lle = &comp->lo_entries[i];
653 lle->lle_lsme = lsm->lsm_entries[i];
654 lle->lle_type = lov_entry_type(lle->lle_lsme);
655 lle->lle_preference = 0;
656 switch (lle->lle_type) {
657 case LOV_PATTERN_RAID0:
658 lle->lle_comp_ops = &raid0_ops;
660 case LOV_PATTERN_MDT:
661 /* Allowed to have several DOM stripes in different
662 * mirrors with the same DoM size.
665 dom_size = lle->lle_lsme->lsme_extent.e_end;
666 } else if (dom_size !=
667 lle->lle_lsme->lsme_extent.e_end) {
668 CERROR("%s: DOM entries with different sizes\n",
669 lov2obd(dev->ld_lov)->obd_name);
670 dump_lsm(D_ERROR, lsm);
673 lle->lle_comp_ops = &dom_ops;
675 case LOV_PATTERN_FOREIGN:
676 lle->lle_comp_ops = NULL;
679 CERROR("%s: unknown composite layout entry type %i\n",
680 lov2obd(dev->ld_lov)->obd_name,
681 lsm->lsm_entries[i]->lsme_pattern);
682 dump_lsm(D_ERROR, lsm);
686 lle->lle_extent = &lle->lle_lsme->lsme_extent;
687 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
689 if (flr_state != LCM_FL_NONE)
690 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
692 lre = &comp->lo_mirrors[j];
694 if (mirror_id == lre->lre_mirror_id) {
695 lre->lre_valid |= lle->lle_valid;
696 lre->lre_stale |= !lle->lle_valid;
698 lsme_is_foreign(lle->lle_lsme);
703 /* new mirror detected, assume that the mirrors
704 * are shorted in layout */
707 if (j >= comp->lo_mirror_count)
710 lre = &comp->lo_mirrors[j];
713 /* entries must be sorted by mirrors */
714 lre->lre_mirror_id = mirror_id;
715 lre->lre_start = lre->lre_end = i;
716 lre->lre_preference = lle->lle_lsme->lsme_flags &
717 LCME_FL_PREF_RD ? 1000 : 0;
718 lre->lre_valid = lle->lle_valid;
719 lre->lre_stale = !lle->lle_valid;
720 lre->lre_foreign = lsme_is_foreign(lle->lle_lsme);
723 /* sanity check for FLR */
724 if (mirror_count != comp->lo_mirror_count) {
726 " doesn't have the # of mirrors it claims, %u/%u\n",
727 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
728 comp->lo_mirror_count + 1);
730 GOTO(out, result = -EINVAL);
733 lov_foreach_layout_entry(lov, lle) {
734 int index = lov_layout_entry_index(lov, lle);
737 * If the component has not been init-ed on MDS side, for
738 * PFL layout, we'd know that the components beyond this one
739 * will be dynamically init-ed later on file write/trunc ops.
741 if (!lsme_inited(lle->lle_lsme))
744 if (lsme_is_foreign(lle->lle_lsme))
747 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
752 LASSERT(ergo(psz > 0, psz == result));
757 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
759 /* decide the preferred mirror. It uses the hash value of lov_object
760 * so that different clients would use different mirrors for read. */
763 seq = cfs_hash_long((unsigned long)lov, 8);
764 for (i = 0; i < comp->lo_mirror_count; i++) {
765 unsigned int idx = (i + seq) % comp->lo_mirror_count;
767 lre = lov_mirror_entry(lov, idx);
771 if (lre->lre_foreign)
774 mirror_count++; /* valid mirror */
776 /* aggregated preference of all involved OSTs */
777 for (j = lre->lre_start; j <= lre->lre_end; j++) {
778 lre->lre_preference +=
779 comp->lo_entries[j].lle_preference;
782 if (lre->lre_preference > preference) {
783 preference = lre->lre_preference;
784 comp->lo_preferred_mirror = idx;
789 " doesn't have any valid mirrors\n",
790 PFID(lu_object_fid(lov2lu(lov))));
792 comp->lo_preferred_mirror = 0;
795 LASSERT(comp->lo_preferred_mirror >= 0);
799 return result > 0 ? 0 : result;
802 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
803 struct lov_object *lov, struct lov_stripe_md *lsm,
804 const struct cl_object_conf *conf,
805 union lov_layout_state *state)
810 static int lov_init_released(const struct lu_env *env,
811 struct lov_device *dev, struct lov_object *lov,
812 struct lov_stripe_md *lsm,
813 const struct cl_object_conf *conf,
814 union lov_layout_state *state)
816 LASSERT(lsm != NULL);
817 LASSERT(lsm->lsm_is_released);
818 LASSERT(lov->lo_lsm == NULL);
820 lov->lo_lsm = lsm_addref(lsm);
824 static int lov_init_foreign(const struct lu_env *env,
825 struct lov_device *dev, struct lov_object *lov,
826 struct lov_stripe_md *lsm,
827 const struct cl_object_conf *conf,
828 union lov_layout_state *state)
830 LASSERT(lsm != NULL);
831 LASSERT(lov->lo_type == LLT_FOREIGN);
832 LASSERT(lov->lo_lsm == NULL);
834 lov->lo_lsm = lsm_addref(lsm);
838 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
839 union lov_layout_state *state)
841 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED ||
842 lov->lo_type == LLT_FOREIGN);
844 lov_layout_wait(env, lov);
848 static int lov_delete_composite(const struct lu_env *env,
849 struct lov_object *lov,
850 union lov_layout_state *state)
852 struct lov_layout_entry *entry;
857 dump_lsm(D_INODE, lov->lo_lsm);
859 lov_layout_wait(env, lov);
860 lov_foreach_layout_entry(lov, entry) {
861 if (entry->lle_lsme && lsme_is_foreign(entry->lle_lsme))
864 rc = lov_delete_raid0(env, lov, entry);
872 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
873 union lov_layout_state *state)
875 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
878 static void lov_fini_composite(const struct lu_env *env,
879 struct lov_object *lov,
880 union lov_layout_state *state)
882 struct lov_layout_composite *comp = &state->composite;
885 if (comp->lo_entries != NULL) {
886 struct lov_layout_entry *entry;
888 lov_foreach_layout_entry(lov, entry)
889 if (entry->lle_comp_ops)
890 entry->lle_comp_ops->lco_fini(env, entry);
892 OBD_FREE_PTR_ARRAY(comp->lo_entries, comp->lo_entry_count);
893 comp->lo_entries = NULL;
896 if (comp->lo_mirrors != NULL) {
897 OBD_FREE_PTR_ARRAY(comp->lo_mirrors, comp->lo_mirror_count);
898 comp->lo_mirrors = NULL;
901 memset(comp, 0, sizeof(*comp));
903 dump_lsm(D_INODE, lov->lo_lsm);
904 lov_free_memmd(&lov->lo_lsm);
909 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
910 union lov_layout_state *state)
913 dump_lsm(D_INODE, lov->lo_lsm);
914 lov_free_memmd(&lov->lo_lsm);
918 static int lov_print_empty(const struct lu_env *env, void *cookie,
919 lu_printer_t p, const struct lu_object *o)
921 (*p)(env, cookie, "empty %d\n",
922 test_bit(LO_LAYOUT_INVALID, &lu2lov(o)->lo_obj_flags));
926 static int lov_print_composite(const struct lu_env *env, void *cookie,
927 lu_printer_t p, const struct lu_object *o)
929 struct lov_object *lov = lu2lov(o);
930 struct lov_stripe_md *lsm = lov->lo_lsm;
933 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
934 lsm->lsm_entry_count,
935 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
936 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
937 lsm->lsm_layout_gen);
939 for (i = 0; i < lsm->lsm_entry_count; i++) {
940 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
941 struct lov_layout_entry *lle = lov_entry(lov, i);
944 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
945 PEXT(&lse->lsme_extent), lse->lsme_magic,
946 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
947 lse->lsme_flags, lse->lsme_stripe_count,
948 lse->lsme_stripe_size);
950 if (!lsme_is_foreign(lse))
951 lov_print_raid0(env, cookie, p, lle);
957 static int lov_print_released(const struct lu_env *env, void *cookie,
958 lu_printer_t p, const struct lu_object *o)
960 struct lov_object *lov = lu2lov(o);
961 struct lov_stripe_md *lsm = lov->lo_lsm;
964 "released: %s, lsm{%p 0x%08X %d %u}:\n",
965 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ? "invalid" :
966 "valid", lsm, lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
967 lsm->lsm_layout_gen);
971 static int lov_print_foreign(const struct lu_env *env, void *cookie,
972 lu_printer_t p, const struct lu_object *o)
974 struct lov_object *lov = lu2lov(o);
975 struct lov_stripe_md *lsm = lov->lo_lsm;
978 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
979 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) ?
980 "invalid" : "valid", lsm,
981 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
982 lsm->lsm_layout_gen);
984 "raw_ea_content '%.*s'\n",
985 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
990 * Implements cl_object_operations::coo_attr_get() method for an object
991 * without stripes (LLT_EMPTY layout type).
993 * The only attributes this layer is authoritative in this case is
994 * cl_attr::cat_blocks---it's 0.
996 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
997 struct cl_attr *attr)
999 attr->cat_blocks = 0;
1003 static int lov_attr_get_composite(const struct lu_env *env,
1004 struct cl_object *obj,
1005 struct cl_attr *attr)
1007 struct lov_object *lov = cl2lov(obj);
1008 struct lov_layout_entry *entry;
1014 attr->cat_blocks = 0;
1017 lov_foreach_layout_entry(lov, entry) {
1018 struct cl_attr *lov_attr = NULL;
1019 int index = lov_layout_entry_index(lov, entry);
1021 if (!entry->lle_valid)
1024 /* PFL: This component has not been init-ed. */
1025 if (!lsm_entry_inited(lov->lo_lsm, index))
1028 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1033 if (lov_attr == NULL)
1036 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1037 "b=%llu\n", index - 1, lov_attr->cat_size,
1038 lov_attr->cat_mtime, lov_attr->cat_atime,
1039 lov_attr->cat_ctime, lov_attr->cat_blocks);
1042 if (lov_attr->cat_kms_valid)
1043 attr->cat_kms_valid = 1;
1044 attr->cat_blocks += lov_attr->cat_blocks;
1045 if (attr->cat_size < lov_attr->cat_size)
1046 attr->cat_size = lov_attr->cat_size;
1047 if (attr->cat_kms < lov_attr->cat_kms)
1048 attr->cat_kms = lov_attr->cat_kms;
1049 if (attr->cat_atime < lov_attr->cat_atime)
1050 attr->cat_atime = lov_attr->cat_atime;
1051 if (attr->cat_ctime < lov_attr->cat_ctime)
1052 attr->cat_ctime = lov_attr->cat_ctime;
1053 if (attr->cat_mtime < lov_attr->cat_mtime)
1054 attr->cat_mtime = lov_attr->cat_mtime;
1060 static int lov_flush_composite(const struct lu_env *env,
1061 struct cl_object *obj,
1062 struct ldlm_lock *lock)
1064 struct lov_object *lov = cl2lov(obj);
1065 struct lov_layout_entry *lle;
1070 lov_foreach_layout_entry(lov, lle) {
1071 if (!lsme_is_dom(lle->lle_lsme))
1073 rc = cl_object_flush(env, lovsub2cl(lle->lle_dom.lo_dom), lock);
1080 static int lov_flush_empty(const struct lu_env *env, struct cl_object *obj,
1081 struct ldlm_lock *lock)
1086 const static struct lov_layout_operations lov_dispatch[] = {
1088 .llo_init = lov_init_empty,
1089 .llo_delete = lov_delete_empty,
1090 .llo_fini = lov_fini_empty,
1091 .llo_print = lov_print_empty,
1092 .llo_page_init = lov_page_init_empty,
1093 .llo_lock_init = lov_lock_init_empty,
1094 .llo_io_init = lov_io_init_empty,
1095 .llo_getattr = lov_attr_get_empty,
1096 .llo_flush = lov_flush_empty,
1099 .llo_init = lov_init_released,
1100 .llo_delete = lov_delete_empty,
1101 .llo_fini = lov_fini_released,
1102 .llo_print = lov_print_released,
1103 .llo_page_init = lov_page_init_empty,
1104 .llo_lock_init = lov_lock_init_empty,
1105 .llo_io_init = lov_io_init_released,
1106 .llo_getattr = lov_attr_get_empty,
1107 .llo_flush = lov_flush_empty,
1110 .llo_init = lov_init_composite,
1111 .llo_delete = lov_delete_composite,
1112 .llo_fini = lov_fini_composite,
1113 .llo_print = lov_print_composite,
1114 .llo_page_init = lov_page_init_composite,
1115 .llo_lock_init = lov_lock_init_composite,
1116 .llo_io_init = lov_io_init_composite,
1117 .llo_getattr = lov_attr_get_composite,
1118 .llo_flush = lov_flush_composite,
1121 .llo_init = lov_init_foreign,
1122 .llo_delete = lov_delete_empty,
1123 .llo_fini = lov_fini_released,
1124 .llo_print = lov_print_foreign,
1125 .llo_page_init = lov_page_init_foreign,
1126 .llo_lock_init = lov_lock_init_empty,
1127 .llo_io_init = lov_io_init_empty,
1128 .llo_getattr = lov_attr_get_empty,
1129 .llo_flush = lov_flush_empty,
1134 * Performs a double-dispatch based on the layout type of an object.
1136 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1138 struct lov_object *__obj = (obj); \
1139 enum lov_layout_type __llt; \
1141 __llt = __obj->lo_type; \
1142 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1143 lov_dispatch[__llt].op(__VA_ARGS__); \
1147 * Return lov_layout_type associated with a given lsm
1149 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1154 if (lsm->lsm_is_released)
1155 return LLT_RELEASED;
1157 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1158 lsm->lsm_magic == LOV_MAGIC_V3 ||
1159 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1162 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1168 static inline void lov_conf_freeze(struct lov_object *lov)
1170 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1171 lov, lov->lo_owner, current);
1172 if (lov->lo_owner != current)
1173 down_read(&lov->lo_type_guard);
1176 static inline void lov_conf_thaw(struct lov_object *lov)
1178 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1179 lov, lov->lo_owner, current);
1180 if (lov->lo_owner != current)
1181 up_read(&lov->lo_type_guard);
1184 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1186 struct lov_object *__obj = (obj); \
1187 int __lock = !!(lock); \
1188 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1191 lov_conf_freeze(__obj); \
1192 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1194 lov_conf_thaw(__obj); \
1199 * Performs a locked double-dispatch based on the layout type of an object.
1201 #define LOV_2DISPATCH(obj, op, ...) \
1202 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1204 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1206 struct lov_object *__obj = (obj); \
1207 enum lov_layout_type __llt; \
1209 lov_conf_freeze(__obj); \
1210 __llt = __obj->lo_type; \
1211 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1212 lov_dispatch[__llt].op(__VA_ARGS__); \
1213 lov_conf_thaw(__obj); \
1216 static void lov_conf_lock(struct lov_object *lov)
1218 LASSERT(lov->lo_owner != current);
1219 down_write(&lov->lo_type_guard);
1220 LASSERT(lov->lo_owner == NULL);
1221 lov->lo_owner = current;
1222 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1223 lov, lov->lo_owner);
1226 static void lov_conf_unlock(struct lov_object *lov)
1228 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1229 lov, lov->lo_owner);
1230 lov->lo_owner = NULL;
1231 up_write(&lov->lo_type_guard);
1234 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1238 while (atomic_read(&lov->lo_active_ios) > 0) {
1239 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1240 PFID(lu_object_fid(lov2lu(lov))),
1241 atomic_read(&lov->lo_active_ios));
1243 wait_event_idle(lov->lo_waitq,
1244 atomic_read(&lov->lo_active_ios) == 0);
1249 static int lov_layout_change(const struct lu_env *unused,
1250 struct lov_object *lov, struct lov_stripe_md *lsm,
1251 const struct cl_object_conf *conf)
1253 enum lov_layout_type llt = lov_type(lsm);
1254 union lov_layout_state *state = &lov->u;
1255 const struct lov_layout_operations *old_ops;
1256 const struct lov_layout_operations *new_ops;
1257 struct lov_device *lov_dev = lov_object_dev(lov);
1263 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1265 env = cl_env_get(&refcheck);
1267 RETURN(PTR_ERR(env));
1269 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1271 CDEBUG(D_INODE, DFID" from %s to %s\n",
1272 PFID(lu_object_fid(lov2lu(lov))),
1273 llt2str(lov->lo_type), llt2str(llt));
1275 old_ops = &lov_dispatch[lov->lo_type];
1276 new_ops = &lov_dispatch[llt];
1278 rc = cl_object_prune(env, &lov->lo_cl);
1282 rc = old_ops->llo_delete(env, lov, &lov->u);
1286 old_ops->llo_fini(env, lov, &lov->u);
1288 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1290 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1291 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1293 /* page bufsize fixup */
1294 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1295 lov_page_slice_fixup(lov, NULL);
1298 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1300 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1302 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1303 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1304 new_ops->llo_delete(env, lov, state);
1305 new_ops->llo_fini(env, lov, state);
1306 /* this file becomes an EMPTY file. */
1307 lov->lo_type = LLT_EMPTY;
1312 cl_env_put(env, &refcheck);
1316 /*****************************************************************************
1318 * Lov object operations.
1321 static int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1322 const struct lu_object_conf *conf)
1324 struct lov_object *lov = lu2lov(obj);
1325 struct lov_device *dev = lov_object_dev(lov);
1326 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1327 union lov_layout_state *set = &lov->u;
1328 const struct lov_layout_operations *ops;
1329 struct lov_stripe_md *lsm = NULL;
1333 init_rwsem(&lov->lo_type_guard);
1334 atomic_set(&lov->lo_active_ios, 0);
1335 init_waitqueue_head(&lov->lo_waitq);
1336 cl_object_page_init(lu2cl(obj), 0);
1338 lov->lo_type = LLT_EMPTY;
1339 if (cconf->u.coc_layout.lb_buf != NULL) {
1340 lsm = lov_unpackmd(dev->ld_lov,
1341 cconf->u.coc_layout.lb_buf,
1342 cconf->u.coc_layout.lb_len);
1344 RETURN(PTR_ERR(lsm));
1346 dump_lsm(D_INODE, lsm);
1349 /* no locking is necessary, as object is being created */
1350 lov->lo_type = lov_type(lsm);
1351 ops = &lov_dispatch[lov->lo_type];
1352 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1362 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1363 const struct cl_object_conf *conf)
1365 struct lov_stripe_md *lsm = NULL;
1366 struct lov_object *lov = cl2lov(obj);
1368 struct cl_object *top = cl_object_top(obj);
1369 bool unlock_inode = false;
1370 bool lock_inode_size = false;
1371 bool lock_layout = false;
1374 if (conf->coc_opc == OBJECT_CONF_SET &&
1375 conf->u.coc_layout.lb_buf != NULL) {
1376 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1377 conf->u.coc_layout.lb_buf,
1378 conf->u.coc_layout.lb_len);
1380 RETURN(PTR_ERR(lsm));
1381 dump_lsm(D_INODE, lsm);
1384 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1385 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1386 GOTO(out_lsm, result = 0);
1391 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1392 if (test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags) &&
1393 atomic_read(&lov->lo_active_ios) > 0) {
1394 lov_conf_unlock(lov);
1395 result = lov_layout_wait(env, lov);
1401 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1404 * don't apply old layouts which can be brought
1405 * if returned w/o ldlm lock.
1406 * XXX: can we rollback in case of recovery?
1408 if (lsm && lov->lo_lsm) {
1409 u32 oldgen = lov->lo_lsm->lsm_layout_gen &= ~LU_LAYOUT_RESYNC;
1410 u32 newgen = lsm->lsm_layout_gen & ~LU_LAYOUT_RESYNC;
1412 if (newgen < oldgen) {
1413 CDEBUG(D_HA, "skip old for "DFID": %d < %d\n",
1414 PFID(lu_object_fid(lov2lu(lov))),
1415 (int)newgen, (int)oldgen);
1416 GOTO(out, result = 0);
1420 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1421 ((lsm != NULL && lov->lo_lsm != NULL) &&
1422 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1423 (lov->lo_lsm->lsm_flags == lsm->lsm_flags) &&
1424 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1425 lsm->lsm_entries[0]->lsme_pattern))) {
1426 /* same version of layout */
1427 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1428 GOTO(out, result = 0);
1431 /* will change layout - check if there still exists active IO. */
1432 if (atomic_read(&lov->lo_active_ios) > 0) {
1433 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1434 GOTO(out, result = -EBUSY);
1437 if (conf->coc_try) {
1438 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1439 GOTO(out, result = -ERESTARTSYS);
1442 result = lov_layout_change(env, lov, lsm, conf);
1444 if (result == -EAGAIN) {
1446 * we need unlocked lov conf and get inode lock.
1447 * It's possible we have already taken inode's size
1448 * mutex and/or layout mutex, so we need keep such lock
1449 * order, lest deadlock happens:
1450 * inode lock (ll_inode_lock())
1451 * inode size lock (ll_inode_size_lock())
1452 * inode layout lock (ll_layout_refresh())
1453 * lov conf lock (lov_conf_lock())
1456 * vfs_setxattr inode locked
1457 * ll_lov_setstripe_ea_info inode size locked
1459 * ll_file_inode_init
1461 * lov_conf_set lov conf locked
1463 * ll_migrate inode locked
1465 * ll_layout_refresh inode layout locked
1468 * lov_conf_set lov conf locked
1470 lov_conf_unlock(lov);
1471 if (cl_object_inode_ops(env, top, COIO_LAYOUT_UNLOCK,
1474 if (cl_object_inode_ops(env, top, COIO_SIZE_UNLOCK,
1476 lock_inode_size = true;
1478 /* take lock in order */
1479 if (cl_object_inode_ops(
1480 env, top, COIO_INODE_LOCK, NULL) == 0)
1481 unlock_inode = true;
1482 if (lock_inode_size)
1483 cl_object_inode_ops(env, top, COIO_SIZE_LOCK,
1486 cl_object_inode_ops(env, top, COIO_LAYOUT_LOCK,
1490 set_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1492 clear_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags);
1497 lov_conf_unlock(lov);
1499 cl_object_inode_ops(env, top, COIO_INODE_UNLOCK, NULL);
1502 CDEBUG(D_INODE, DFID" lo_layout_invalid=%u\n",
1503 PFID(lu_object_fid(lov2lu(lov))),
1504 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags));
1508 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1510 struct lov_object *lov = lu2lov(obj);
1513 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1517 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1519 struct lov_object *lov = lu2lov(obj);
1522 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1523 lu_object_fini(obj);
1524 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1528 static int lov_object_print(const struct lu_env *env, void *cookie,
1529 lu_printer_t p, const struct lu_object *o)
1531 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1534 static int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1535 struct cl_page *page, pgoff_t index)
1537 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1542 * Implements cl_object_operations::clo_io_init() method for lov
1543 * layer. Dispatches to the appropriate layout io initialization method.
1545 static int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1548 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1550 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1551 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1552 io->ci_ignore_layout, io->ci_verify_layout);
1554 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1555 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1556 * because as long as the OSC object exists, the layout can't be
1558 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1559 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1564 * An implementation of cl_object_operations::clo_attr_get() method for lov
1565 * layer. For raid0 layout this collects and merges attributes of all
1568 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1569 struct cl_attr *attr)
1571 /* do not take lock, as this function is called under a
1572 * spin-lock. Layout is protected from changing by ongoing IO. */
1573 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1576 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1577 const struct cl_attr *attr, unsigned valid)
1580 * No dispatch is required here, as no layout implements this.
1585 static int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1586 struct cl_lock *lock, const struct cl_io *io)
1588 /* No need to lock because we've taken one refcount of layout. */
1589 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1594 * We calculate on which OST the mapping will end. If the length of mapping
1595 * is greater than (stripe_size * stripe_count) then the last_stripe will
1596 * will be one just before start_stripe. Else we check if the mapping
1597 * intersects each OST and find last_stripe.
1598 * This function returns the last_stripe and also sets the stripe_count
1599 * over which the mapping is spread
1601 * \param lsm [in] striping information for the file
1602 * \param index [in] stripe component index
1603 * \param ext [in] logical extent of mapping
1604 * \param start_stripe [in] starting stripe of the mapping
1605 * \param stripe_count [out] the number of stripes across which to map is
1608 * \retval last_stripe return the last stripe of the mapping
1610 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1611 struct lu_extent *ext,
1612 int start_stripe, int *stripe_count)
1614 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1619 init_stripe = lov_stripe_number(lsm, index, ext->e_start);
1621 if (ext->e_end - ext->e_start >
1622 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1623 if (init_stripe == start_stripe) {
1624 last_stripe = (start_stripe < 1) ?
1625 lsme->lsme_stripe_count - 1 : start_stripe - 1;
1626 *stripe_count = lsme->lsme_stripe_count;
1627 } else if (init_stripe < start_stripe) {
1628 last_stripe = (init_stripe < 1) ?
1629 lsme->lsme_stripe_count - 1 : init_stripe - 1;
1630 *stripe_count = lsme->lsme_stripe_count -
1631 (start_stripe - init_stripe);
1633 last_stripe = init_stripe - 1;
1634 *stripe_count = init_stripe - start_stripe;
1637 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1638 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1639 if (!lov_stripe_intersects(lsm, index, i, ext, NULL,
1642 if ((start_stripe != init_stripe) && (i == init_stripe))
1646 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1653 * Set fe_device and copy extents from local buffer into main return buffer.
1655 * \param fiemap [out] fiemap to hold all extents
1656 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1657 * \param ost_index [in] OST index to be written into the fm_device
1658 * field for each extent
1659 * \param ext_count [in] number of extents to be copied
1660 * \param current_extent [in] where to start copying in the extent array
1662 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1663 struct fiemap_extent *lcl_fm_ext,
1664 int ost_index, unsigned int ext_count,
1665 int current_extent, int abs_stripeno)
1670 for (ext = 0; ext < ext_count; ext++) {
1671 set_fe_device_stripenr(&lcl_fm_ext[ext], ost_index,
1673 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1676 /* Copy fm_extent's from fm_local to return buffer */
1677 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1678 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1681 #define FIEMAP_BUFFER_SIZE 4096
1684 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1685 * call. The local end offset and the device are sent in the first
1686 * fm_extent. This function calculates the stripe number from the index.
1687 * This function returns a stripe_no on which mapping is to be restarted.
1689 * This function returns fm_end_offset which is the in-OST offset at which
1690 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1691 * will re-calculate proper offset in next stripe.
1692 * Note that the first extent is passed to lov_get_info via the value field.
1694 * \param fiemap [in] fiemap request header
1695 * \param lsm [in] striping information for the file
1696 * \param index [in] stripe component index
1697 * \param ext [in] logical extent of mapping
1698 * \param start_stripe [out] starting stripe will be returned in this
1700 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1701 struct lov_stripe_md *lsm,
1702 int index, struct lu_extent *ext,
1705 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1706 u64 local_end = fiemap->fm_extents[0].fe_logical;
1711 if (fiemap->fm_extent_count == 0 ||
1712 fiemap->fm_extents[0].fe_logical == 0)
1715 stripe_no = *start_stripe;
1717 if (stripe_no == -1)
1720 /* If we have finished mapping on previous device, shift logical
1721 * offset to start of next device */
1722 if (lov_stripe_intersects(lsm, index, stripe_no, ext, NULL, &lun_end) &&
1723 local_end < lun_end) {
1724 fm_end_offset = local_end;
1726 /* This is a special value to indicate that caller should
1727 * calculate offset in next stripe. */
1729 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1732 return fm_end_offset;
1735 struct fiemap_state {
1736 struct fiemap *fs_fm;
1737 struct lu_extent fs_ext; /* current entry extent */
1739 u64 fs_end_offset; /* last iteration offset */
1740 int fs_cur_extent; /* collected exts so far */
1741 int fs_cnt_need; /* # of extents buf can hold */
1742 int fs_start_stripe;
1744 bool fs_device_done; /* enough for this OST */
1745 bool fs_finish_stripe; /* reached fs_last_stripe */
1746 bool fs_enough; /* enough for this call */
1749 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1750 struct lov_object *lov,
1751 struct lov_stripe_md *lsm,
1754 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1755 struct lov_thread_info *lti = lov_env_info(env);
1756 struct lu_fid *ofid = <i->lti_fid;
1757 struct lov_oinfo *oinfo;
1758 struct cl_device *subdev;
1759 int entry = lov_comp_entry(index);
1760 int stripe = lov_comp_stripe(index);
1763 struct cl_object *result;
1765 if (lov->lo_type != LLT_COMP)
1766 GOTO(out, result = NULL);
1768 if (entry >= lsm->lsm_entry_count ||
1769 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1770 GOTO(out, result = NULL);
1772 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1773 ost_idx = oinfo->loi_ost_idx;
1774 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1776 GOTO(out, result = NULL);
1778 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1779 result = lov_sub_find(env, subdev, ofid, NULL);
1782 result = ERR_PTR(-EINVAL);
1786 static int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1787 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1788 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1789 int index, int stripe_last, int stripeno,
1790 struct fiemap_state *fs)
1792 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1793 struct cl_object *subobj;
1794 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1795 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1796 u64 req_fm_len; /* max requested extent coverage */
1797 u64 len_mapped_single_call;
1800 unsigned int ext_count;
1801 /* EOF for object */
1802 bool ost_eof = false;
1803 /* done with required mapping for this OST? */
1804 bool ost_done = false;
1808 fs->fs_device_done = false;
1809 /* Find out range of mapping on this stripe */
1810 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1811 &obd_start, &obd_end)) == 0)
1814 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1817 /* If this is a continuation FIEMAP call and we are on
1818 * starting stripe then obd_start needs to be set to
1820 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1821 obd_start = fs->fs_end_offset;
1823 if (lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno) ==
1827 req_fm_len = obd_end - obd_start + 1;
1828 fs->fs_fm->fm_length = 0;
1829 len_mapped_single_call = 0;
1831 /* find lobsub object */
1832 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1833 lov_comp_index(index, stripeno));
1835 return PTR_ERR(subobj);
1836 /* If the output buffer is very large and the objects have many
1837 * extents we may need to loop on a single OST repeatedly */
1839 if (fiemap->fm_extent_count > 0) {
1840 /* Don't get too many extents. */
1841 if (fs->fs_cur_extent + fs->fs_cnt_need >
1842 fiemap->fm_extent_count)
1843 fs->fs_cnt_need = fiemap->fm_extent_count -
1847 obd_start += len_mapped_single_call;
1848 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1849 req_fm_len = fs->fs_fm->fm_length;
1851 * If we've collected enough extent map, we'd request 1 more,
1852 * to see whether we coincidentally finished all available
1853 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1855 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1856 1 : fs->fs_cnt_need;
1857 fs->fs_fm->fm_mapped_extents = 0;
1858 fs->fs_fm->fm_flags = fiemap->fm_flags;
1860 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1862 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1863 GOTO(obj_put, rc = -EINVAL);
1864 /* If OST is inactive, return extent with UNKNOWN flag. */
1865 if (!lov->lov_tgts[ost_index]->ltd_active) {
1866 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1867 fs->fs_fm->fm_mapped_extents = 1;
1869 fm_ext[0].fe_logical = obd_start;
1870 fm_ext[0].fe_length = obd_end - obd_start + 1;
1871 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1876 fs->fs_fm->fm_start = obd_start;
1877 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1878 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1879 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1881 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1885 ext_count = fs->fs_fm->fm_mapped_extents;
1886 if (ext_count == 0) {
1888 fs->fs_device_done = true;
1889 /* If last stripe has hold at the end,
1890 * we need to return */
1891 if (stripeno == fs->fs_last_stripe) {
1892 fiemap->fm_mapped_extents = 0;
1893 fs->fs_finish_stripe = true;
1897 } else if (fs->fs_enough) {
1899 * We've collected enough extents and there are
1900 * more extents after it.
1905 /* If we just need num of extents, got to next device */
1906 if (fiemap->fm_extent_count == 0) {
1907 fs->fs_cur_extent += ext_count;
1911 /* prepare to copy retrived map extents */
1912 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1913 fm_ext[ext_count - 1].fe_length -
1916 /* Have we finished mapping on this device? */
1917 if (req_fm_len <= len_mapped_single_call) {
1919 fs->fs_device_done = true;
1922 /* Clear the EXTENT_LAST flag which can be present on
1923 * the last extent */
1924 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1925 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1926 if (lov_stripe_size(lsm, index,
1927 fm_ext[ext_count - 1].fe_logical +
1928 fm_ext[ext_count - 1].fe_length,
1929 stripeno) >= fmkey->lfik_oa.o_size) {
1931 fs->fs_device_done = true;
1934 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1935 ext_count, fs->fs_cur_extent,
1936 stripe_last + stripeno);
1937 fs->fs_cur_extent += ext_count;
1939 /* Ran out of available extents? */
1940 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1941 fs->fs_enough = true;
1942 } while (!ost_done && !ost_eof);
1944 if (stripeno == fs->fs_last_stripe)
1945 fs->fs_finish_stripe = true;
1947 cl_object_put(env, subobj);
1953 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1954 * This also handles the restarting of FIEMAP calls in case mapping overflows
1955 * the available number of extents in single call.
1957 * \param env [in] lustre environment
1958 * \param obj [in] file object
1959 * \param fmkey [in] fiemap request header and other info
1960 * \param fiemap [out] fiemap buffer holding retrived map extents
1961 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1962 * each OST, it is used to limit max map needed
1966 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1967 struct ll_fiemap_info_key *fmkey,
1968 struct fiemap *fiemap, size_t *buflen)
1970 struct lov_stripe_md_entry *lsme;
1971 struct lov_stripe_md *lsm;
1972 struct fiemap *fm_local = NULL;
1976 int start_entry = -1;
1980 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1982 struct fiemap_state fs = { 0 };
1983 struct lu_extent range;
1985 int stripe_last = 0;
1986 int start_stripe = 0;
1987 bool resume = false;
1990 lsm = lov_lsm_addref(cl2lov(obj));
1992 /* no extent: there is no object for mapping */
1993 fiemap->fm_mapped_extents = 0;
1997 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1999 * If the entry count > 1 or stripe_count > 1 and the
2000 * application does not understand DEVICE_ORDER flag,
2001 * it cannot interpret the extents correctly.
2003 if (lsm->lsm_entry_count > 1 ||
2004 (lsm->lsm_entry_count == 1 &&
2005 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2006 GOTO(out_lsm, rc = -EOPNOTSUPP);
2009 /* No support for DOM layout yet. */
2010 if (lsme_is_dom(lsm->lsm_entries[0]))
2011 GOTO(out_lsm, rc = -EOPNOTSUPP);
2013 if (lsm->lsm_is_released) {
2014 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
2016 * released file, return a minimal FIEMAP if
2017 * request fits in file-size.
2019 fiemap->fm_mapped_extents = 1;
2020 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
2021 if (fiemap->fm_start + fiemap->fm_length <
2022 fmkey->lfik_oa.o_size)
2023 fiemap->fm_extents[0].fe_length =
2026 fiemap->fm_extents[0].fe_length =
2027 fmkey->lfik_oa.o_size -
2029 fiemap->fm_extents[0].fe_flags |=
2030 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
2032 GOTO(out_lsm, rc = 0);
2035 /* buffer_size is small to hold fm_extent_count of extents. */
2036 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
2037 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
2039 OBD_ALLOC_LARGE(fm_local, buffer_size);
2040 if (fm_local == NULL)
2041 GOTO(out_lsm, rc = -ENOMEM);
2044 * Requested extent count exceeds the fiemap buffer size, shrink our
2047 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
2048 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
2050 fs.fs_enough = false;
2051 fs.fs_cur_extent = 0;
2052 fs.fs_fm = fm_local;
2053 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
2055 whole_start = fiemap->fm_start;
2056 /* whole_start is beyond the end of the file */
2057 if (whole_start > fmkey->lfik_oa.o_size)
2058 GOTO(out_fm_local, rc = -EINVAL);
2059 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
2060 fmkey->lfik_oa.o_size + 1 :
2061 whole_start + fiemap->fm_length;
2063 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
2066 if (whole_end > fmkey->lfik_oa.o_size + 1)
2067 whole_end = fmkey->lfik_oa.o_size + 1;
2070 * the high 16bits of fe_device remember which stripe the last
2071 * call has been arrived, we'd continue from there in this call.
2073 if (fiemap->fm_extent_count && fiemap->fm_extents[0].fe_logical) {
2075 stripe_last = get_fe_stripenr(&fiemap->fm_extents[0]);
2078 * stripe_last records stripe number we've been processed in the last
2081 end_entry = lsm->lsm_entry_count - 1;
2083 for (entry = 0; entry <= end_entry; entry++) {
2084 lsme = lsm->lsm_entries[entry];
2085 if (cur_stripe + lsme->lsme_stripe_count >= stripe_last) {
2086 start_entry = entry;
2087 start_stripe = stripe_last - cur_stripe;
2091 cur_stripe += lsme->lsme_stripe_count;
2093 if (start_entry == -1) {
2094 CERROR(DFID": FIEMAP does not init start entry, cur_stripe=%d, "
2095 "stripe_last=%d\n", PFID(lu_object_fid(&obj->co_lu)),
2096 cur_stripe, stripe_last);
2097 GOTO(out_fm_local, rc = -EINVAL);
2100 * @start_entry & @start_stripe records the position of fiemap
2101 * resumption @stripe_last keeps recording the absolution position
2102 * we'are processing. @resume indicates we'd honor @start_stripe.
2105 range.e_start = whole_start;
2106 range.e_end = whole_end;
2108 for (entry = start_entry; entry <= end_entry; entry++) {
2109 /* remeber to update stripe_last accordingly */
2110 lsme = lsm->lsm_entries[entry];
2112 /* FLR could contain component holes between entries */
2113 if (!lsme_inited(lsme)) {
2114 stripe_last += lsme->lsme_stripe_count;
2119 if (!lu_extent_is_overlapped(&range, &lsme->lsme_extent)) {
2120 stripe_last += lsme->lsme_stripe_count;
2125 /* prepare for a component entry iteration */
2126 if (lsme->lsme_extent.e_start > whole_start)
2127 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
2129 fs.fs_ext.e_start = whole_start;
2130 if (lsme->lsme_extent.e_end > whole_end)
2131 fs.fs_ext.e_end = whole_end;
2133 fs.fs_ext.e_end = lsme->lsme_extent.e_end;
2135 /* Calculate start stripe, last stripe and length of mapping */
2137 fs.fs_start_stripe = start_stripe;
2138 /* put stripe_last to the first stripe of the comp */
2139 stripe_last -= start_stripe;
2142 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
2145 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
2146 &fs.fs_ext, fs.fs_start_stripe,
2149 * A new mirror component is under process, reset
2150 * fs.fs_end_offset and then fiemap_for_stripe() starts from
2151 * the overlapping extent, otherwise starts from
2154 if (entry > start_entry && lsme->lsme_extent.e_start == 0) {
2156 fs.fs_end_offset = 0;
2158 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap,
2159 lsm, entry, &fs.fs_ext,
2160 &fs.fs_start_stripe);
2163 /* Check each stripe */
2164 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
2166 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
2167 /* reset fs_finish_stripe */
2168 fs.fs_finish_stripe = false;
2169 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
2170 fmkey, entry, stripe_last,
2173 GOTO(out_fm_local, rc);
2175 stripe_last += cur_stripe;
2178 if (fs.fs_finish_stripe)
2180 } /* for each stripe */
2181 stripe_last += lsme->lsme_stripe_count;
2182 } /* for covering layout component entry */
2185 if (fs.fs_cur_extent > 0)
2186 cur_ext = fs.fs_cur_extent - 1;
2190 /* done all the processing */
2191 if (entry > end_entry ||
2192 (fs.fs_enough && fs.fs_finish_stripe && entry == end_entry))
2193 fiemap->fm_extents[cur_ext].fe_flags |= FIEMAP_EXTENT_LAST;
2195 /* Indicate that we are returning device offsets unless file just has
2197 if (lsm->lsm_entry_count > 1 ||
2198 (lsm->lsm_entry_count == 1 &&
2199 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2200 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2202 if (fiemap->fm_extent_count == 0)
2203 goto skip_last_device_calc;
2205 skip_last_device_calc:
2206 fiemap->fm_mapped_extents = fs.fs_cur_extent;
2208 OBD_FREE_LARGE(fm_local, buffer_size);
2215 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2216 struct lov_user_md __user *lum, size_t size)
2218 struct lov_object *lov = cl2lov(obj);
2219 struct lov_stripe_md *lsm;
2223 lsm = lov_lsm_addref(lov);
2227 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2232 static int lov_object_layout_get(const struct lu_env *env,
2233 struct cl_object *obj,
2234 struct cl_layout *cl)
2236 struct lov_object *lov = cl2lov(obj);
2237 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2238 struct lu_buf *buf = &cl->cl_buf;
2244 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2249 cl->cl_size = lov_comp_md_size(lsm);
2250 cl->cl_layout_gen = lsm->lsm_layout_gen;
2251 cl->cl_is_released = lsm->lsm_is_released;
2252 cl->cl_is_composite = lsm_is_composite(lsm->lsm_magic);
2254 rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2257 /* return error or number of bytes */
2261 static loff_t lov_object_maxbytes(struct cl_object *obj)
2263 struct lov_object *lov = cl2lov(obj);
2264 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2270 maxbytes = lsm->lsm_maxbytes;
2277 static int lov_object_flush(const struct lu_env *env, struct cl_object *obj,
2278 struct ldlm_lock *lock)
2280 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_flush, true, env, obj,
2284 static const struct cl_object_operations lov_ops = {
2285 .coo_page_init = lov_page_init,
2286 .coo_lock_init = lov_lock_init,
2287 .coo_io_init = lov_io_init,
2288 .coo_attr_get = lov_attr_get,
2289 .coo_attr_update = lov_attr_update,
2290 .coo_conf_set = lov_conf_set,
2291 .coo_getstripe = lov_object_getstripe,
2292 .coo_layout_get = lov_object_layout_get,
2293 .coo_maxbytes = lov_object_maxbytes,
2294 .coo_fiemap = lov_object_fiemap,
2295 .coo_object_flush = lov_object_flush
2298 static const struct lu_object_operations lov_lu_obj_ops = {
2299 .loo_object_init = lov_object_init,
2300 .loo_object_delete = lov_object_delete,
2301 .loo_object_release = NULL,
2302 .loo_object_free = lov_object_free,
2303 .loo_object_print = lov_object_print,
2304 .loo_object_invariant = NULL,
2307 struct lu_object *lov_object_alloc(const struct lu_env *env,
2308 const struct lu_object_header *unused,
2309 struct lu_device *dev)
2311 struct lov_object *lov;
2312 struct lu_object *obj;
2315 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2318 lu_object_init(obj, NULL, dev);
2319 lov->lo_cl.co_ops = &lov_ops;
2320 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2322 * object io operation vector (cl_object::co_iop) is installed
2323 * later in lov_object_init(), as different vectors are used
2324 * for object with different layouts.
2326 obj->lo_ops = &lov_lu_obj_ops;
2332 static struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2334 struct lov_stripe_md *lsm = NULL;
2336 lov_conf_freeze(lov);
2337 if (lov->lo_lsm != NULL) {
2338 lsm = lsm_addref(lov->lo_lsm);
2339 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2340 lsm, atomic_read(&lsm->lsm_refc),
2341 test_bit(LO_LAYOUT_INVALID, &lov->lo_obj_flags),
2348 int lov_read_and_clear_async_rc(struct cl_object *clob)
2350 struct lu_object *luobj;
2354 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2356 if (luobj != NULL) {
2357 struct lov_object *lov = lu2lov(luobj);
2359 lov_conf_freeze(lov);
2360 switch (lov->lo_type) {
2362 struct lov_stripe_md *lsm;
2366 LASSERT(lsm != NULL);
2367 for (i = 0; i < lsm->lsm_entry_count; i++) {
2368 struct lov_stripe_md_entry *lse =
2369 lsm->lsm_entries[i];
2372 if (!lsme_inited(lse))
2375 for (j = 0; j < lse->lsme_stripe_count; j++) {
2376 struct lov_oinfo *loi =
2379 if (lov_oinfo_is_dummy(loi))
2382 if (loi->loi_ar.ar_rc && !rc)
2383 rc = loi->loi_ar.ar_rc;
2384 loi->loi_ar.ar_rc = 0;
2400 EXPORT_SYMBOL(lov_read_and_clear_async_rc);