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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_object for LOV layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_LOV
40 #include <linux/random.h>
42 #include "lov_cl_internal.h"
44 static inline struct lov_device *lov_object_dev(struct lov_object *obj)
46 return lu2lov_dev(obj->lo_cl.co_lu.lo_dev);
53 /*****************************************************************************
59 struct lov_layout_operations {
60 int (*llo_init)(const struct lu_env *env, struct lov_device *dev,
61 struct lov_object *lov, struct lov_stripe_md *lsm,
62 const struct cl_object_conf *conf,
63 union lov_layout_state *state);
64 int (*llo_delete)(const struct lu_env *env, struct lov_object *lov,
65 union lov_layout_state *state);
66 void (*llo_fini)(const struct lu_env *env, struct lov_object *lov,
67 union lov_layout_state *state);
68 int (*llo_print)(const struct lu_env *env, void *cookie,
69 lu_printer_t p, const struct lu_object *o);
70 int (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
71 struct cl_page *page, pgoff_t index);
72 int (*llo_lock_init)(const struct lu_env *env,
73 struct cl_object *obj, struct cl_lock *lock,
74 const struct cl_io *io);
75 int (*llo_io_init)(const struct lu_env *env,
76 struct cl_object *obj, struct cl_io *io);
77 int (*llo_getattr)(const struct lu_env *env, struct cl_object *obj,
78 struct cl_attr *attr);
79 int (*llo_flush)(const struct lu_env *env, struct cl_object *obj,
80 struct ldlm_lock *lock);
83 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov);
85 static void lov_lsm_put(struct lov_stripe_md *lsm)
91 /*****************************************************************************
93 * Lov object layout operations.
97 static struct cl_object *lov_sub_find(const struct lu_env *env,
98 struct cl_device *dev,
99 const struct lu_fid *fid,
100 const struct cl_object_conf *conf)
106 o = lu_object_find_at(env, cl2lu_dev(dev), fid, &conf->coc_lu);
107 LASSERT(ergo(!IS_ERR(o), o->lo_dev->ld_type == &lovsub_device_type));
111 static int lov_page_slice_fixup(struct lov_object *lov,
112 struct cl_object *stripe)
114 struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
118 return hdr->coh_page_bufsize - lov->lo_cl.co_slice_off -
119 cfs_size_round(sizeof(struct lov_page));
121 cl_object_for_each(o, stripe)
122 o->co_slice_off += hdr->coh_page_bufsize;
124 return cl_object_header(stripe)->coh_page_bufsize;
127 static int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
128 struct cl_object *subobj, struct lov_oinfo *oinfo,
131 struct cl_object_header *hdr;
132 struct cl_object_header *subhdr;
133 struct cl_object_header *parent;
134 int entry = lov_comp_entry(idx);
135 int stripe = lov_comp_stripe(idx);
138 if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
139 /* For sanity:test_206.
140 * Do not leave the object in cache to avoid accessing
141 * freed memory. This is because osc_object is referring to
142 * lov_oinfo of lsm_stripe_data which will be freed due to
144 cl_object_kill(env, subobj);
145 cl_object_put(env, subobj);
149 hdr = cl_object_header(lov2cl(lov));
150 subhdr = cl_object_header(subobj);
152 CDEBUG(D_INODE, DFID"@%p[%d:%d] -> "DFID"@%p: ostid: "DOSTID
153 " ost idx: %d gen: %d\n",
154 PFID(lu_object_fid(&subobj->co_lu)), subhdr, entry, stripe,
155 PFID(lu_object_fid(lov2lu(lov))), hdr, POSTID(&oinfo->loi_oi),
156 oinfo->loi_ost_idx, oinfo->loi_ost_gen);
158 /* reuse ->coh_attr_guard to protect coh_parent change */
159 spin_lock(&subhdr->coh_attr_guard);
160 parent = subhdr->coh_parent;
161 if (parent == NULL) {
162 struct lovsub_object *lso = cl2lovsub(subobj);
164 subhdr->coh_parent = hdr;
165 spin_unlock(&subhdr->coh_attr_guard);
166 subhdr->coh_nesting = hdr->coh_nesting + 1;
167 lu_object_ref_add(&subobj->co_lu, "lov-parent", lov);
168 lso->lso_super = lov;
169 lso->lso_index = idx;
172 struct lu_object *old_obj;
173 struct lov_object *old_lov;
174 unsigned int mask = D_INODE;
176 spin_unlock(&subhdr->coh_attr_guard);
177 old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
178 LASSERT(old_obj != NULL);
179 old_lov = cl2lov(lu2cl(old_obj));
180 if (old_lov->lo_layout_invalid) {
181 /* the object's layout has already changed but isn't
183 lu_object_unhash(env, &subobj->co_lu);
190 LU_OBJECT_DEBUG(mask, env, &subobj->co_lu,
191 "stripe %d is already owned.", idx);
192 LU_OBJECT_DEBUG(mask, env, old_obj, "owned.");
193 LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
194 cl_object_put(env, subobj);
199 static int lov_init_raid0(const struct lu_env *env, struct lov_device *dev,
200 struct lov_object *lov, unsigned int index,
201 const struct cl_object_conf *conf,
202 struct lov_layout_entry *lle)
204 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
205 struct lov_thread_info *lti = lov_env_info(env);
206 struct cl_object_conf *subconf = <i->lti_stripe_conf;
207 struct lu_fid *ofid = <i->lti_fid;
208 struct cl_object *stripe;
209 struct lov_stripe_md_entry *lse = lov_lse(lov, index);
216 spin_lock_init(&r0->lo_sub_lock);
217 r0->lo_nr = lse->lsme_stripe_count;
219 OBD_ALLOC_LARGE(r0->lo_sub, r0->lo_nr * sizeof(r0->lo_sub[0]));
220 if (r0->lo_sub == NULL)
221 GOTO(out, result = -ENOMEM);
225 memset(subconf, 0, sizeof(*subconf));
228 * Create stripe cl_objects.
230 for (i = 0; i < r0->lo_nr; ++i) {
231 struct cl_device *subdev;
232 struct lov_oinfo *oinfo = lse->lsme_oinfo[i];
233 int ost_idx = oinfo->loi_ost_idx;
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 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
249 subconf->u.coc_oinfo = oinfo;
250 LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
251 /* In the function below, .hs_keycmp resolves to
252 * lu_obj_hop_keycmp() */
253 /* coverity[overrun-buffer-val] */
254 stripe = lov_sub_find(env, subdev, ofid, subconf);
256 GOTO(out, result = PTR_ERR(stripe));
258 result = lov_init_sub(env, lov, stripe, oinfo,
259 lov_comp_index(index, i));
260 if (result == -EAGAIN) { /* try again */
267 r0->lo_sub[i] = cl2lovsub(stripe);
269 sz = lov_page_slice_fixup(lov, stripe);
270 LASSERT(ergo(psz > 0, psz == sz));
280 static void lov_subobject_kill(const struct lu_env *env, struct lov_object *lov,
281 struct lov_layout_raid0 *r0,
282 struct lovsub_object *los, int idx)
284 struct cl_object *sub;
285 struct lu_site *site;
286 wait_queue_head_t *wq;
287 wait_queue_entry_t *waiter;
289 LASSERT(r0->lo_sub[idx] == los);
291 sub = lovsub2cl(los);
292 site = sub->co_lu.lo_dev->ld_site;
293 wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
295 cl_object_kill(env, sub);
296 /* release a reference to the sub-object and ... */
297 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
298 cl_object_put(env, sub);
300 /* ... wait until it is actually destroyed---sub-object clears its
301 * ->lo_sub[] slot in lovsub_object_free() */
302 if (r0->lo_sub[idx] == los) {
303 waiter = &lov_env_info(env)->lti_waiter;
304 init_waitqueue_entry(waiter, current);
305 add_wait_queue(wq, waiter);
306 set_current_state(TASK_UNINTERRUPTIBLE);
308 /* this wait-queue is signaled at the end of
309 * lu_object_free(). */
310 set_current_state(TASK_UNINTERRUPTIBLE);
311 spin_lock(&r0->lo_sub_lock);
312 if (r0->lo_sub[idx] == los) {
313 spin_unlock(&r0->lo_sub_lock);
316 spin_unlock(&r0->lo_sub_lock);
317 set_current_state(TASK_RUNNING);
321 remove_wait_queue(wq, waiter);
323 LASSERT(r0->lo_sub[idx] == NULL);
326 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
327 struct lov_layout_entry *lle)
329 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
333 if (r0->lo_sub != NULL) {
336 for (i = 0; i < r0->lo_nr; ++i) {
337 struct lovsub_object *los = r0->lo_sub[i];
340 cl_object_prune(env, &los->lso_cl);
342 * If top-level object is to be evicted from
343 * the cache, so are its sub-objects.
345 lov_subobject_kill(env, lov, r0, los, i);
353 static void lov_fini_raid0(const struct lu_env *env,
354 struct lov_layout_entry *lle)
356 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
358 if (r0->lo_sub != NULL) {
359 OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
364 static int lov_print_raid0(const struct lu_env *env, void *cookie,
365 lu_printer_t p, const struct lov_layout_entry *lle)
367 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
370 for (i = 0; i < r0->lo_nr; ++i) {
371 struct lu_object *sub;
373 if (r0->lo_sub[i] != NULL) {
374 sub = lovsub2lu(r0->lo_sub[i]);
375 lu_object_print(env, cookie, p, sub);
377 (*p)(env, cookie, "sub %d absent\n", i);
383 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
384 unsigned int index, struct lov_layout_entry *lle,
385 struct cl_attr **lov_attr)
387 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
388 struct lov_stripe_md *lsm = lov->lo_lsm;
389 struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
390 struct cl_attr *attr = &r0->lo_attr;
394 if (r0->lo_attr_valid) {
399 memset(lvb, 0, sizeof(*lvb));
401 /* XXX: timestamps can be negative by sanity:test_39m,
403 lvb->lvb_atime = LLONG_MIN;
404 lvb->lvb_ctime = LLONG_MIN;
405 lvb->lvb_mtime = LLONG_MIN;
408 * XXX that should be replaced with a loop over sub-objects,
409 * doing cl_object_attr_get() on them. But for now, let's
410 * reuse old lov code.
414 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
415 * happy. It's not needed, because new code uses
416 * ->coh_attr_guard spin-lock to protect consistency of
417 * sub-object attributes.
419 lov_stripe_lock(lsm);
420 result = lov_merge_lvb_kms(lsm, index, lvb, &kms);
421 lov_stripe_unlock(lsm);
423 cl_lvb2attr(attr, lvb);
425 r0->lo_attr_valid = 1;
432 static struct lov_comp_layout_entry_ops raid0_ops = {
433 .lco_init = lov_init_raid0,
434 .lco_fini = lov_fini_raid0,
435 .lco_getattr = lov_attr_get_raid0,
438 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
439 unsigned int index, struct lov_layout_entry *lle,
440 struct cl_attr **lov_attr)
442 struct lov_layout_dom *dom = &lle->lle_dom;
443 struct lov_oinfo *loi = dom->lo_loi;
444 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
446 if (dom->lo_dom_r0.lo_attr_valid) {
451 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
452 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
454 cl_lvb2attr(attr, &loi->loi_lvb);
456 /* DoM component size can be bigger than stripe size after
457 * client's setattr RPC, so do not count anything beyond
458 * component end. Alternatively, check that limit on server
459 * and do not allow size overflow there. */
460 if (attr->cat_size > lle->lle_extent->e_end)
461 attr->cat_size = lle->lle_extent->e_end;
463 attr->cat_kms = attr->cat_size;
465 dom->lo_dom_r0.lo_attr_valid = 1;
472 * Lookup FLD to get MDS index of the given DOM object FID.
474 * \param[in] ld LOV device
475 * \param[in] fid FID to lookup
476 * \param[out] nr index in MDC array to return back
478 * \retval 0 and \a mds filled with MDS index if successful
479 * \retval negative value on error
481 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
489 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
490 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
492 CERROR("%s: error while looking for mds number. Seq %#llx"
493 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
498 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
501 /* find proper MDC device in the array */
502 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
503 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
504 ld->ld_md_tgts[i].ldm_idx == mds_idx)
508 if (i == ld->ld_md_tgts_nr) {
509 CERROR("%s: cannot find corresponding MDC device for mds #%x "
510 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
520 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
522 * Init the DOM object for the first time. It prepares also RAID0 entry
523 * for it to use in common methods with ordinary RAID0 layout entries.
525 * \param[in] env execution environment
526 * \param[in] dev LOV device
527 * \param[in] lov LOV object
528 * \param[in] index Composite layout entry index in LSM
529 * \param[in] lle Composite LOV layout entry
531 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
532 struct lov_object *lov, unsigned int index,
533 const struct cl_object_conf *conf,
534 struct lov_layout_entry *lle)
536 struct lov_thread_info *lti = lov_env_info(env);
537 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
538 struct cl_object *clo;
539 struct lu_object *o = lov2lu(lov);
540 const struct lu_fid *fid = lu_object_fid(o);
541 struct cl_device *mdcdev;
542 struct lov_oinfo *loi = NULL;
543 struct cl_object_conf *sconf = <i->lti_stripe_conf;
552 /* find proper MDS device */
553 rc = lov_fld_lookup(dev, fid, &idx);
557 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
558 "LOV md target[%u] is NULL\n", idx);
560 /* check lsm is DOM, more checks are needed */
561 LASSERT(lsme->lsme_stripe_count == 0);
564 * Create lower cl_objects.
566 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
568 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
570 /* DoM object has no oinfo in LSM entry, create it exclusively */
571 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
575 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
577 sconf->u.coc_oinfo = loi;
579 clo = lov_sub_find(env, mdcdev, fid, sconf);
581 GOTO(out, rc = PTR_ERR(clo));
583 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
584 if (rc == -EAGAIN) /* try again */
589 lle->lle_dom.lo_dom = cl2lovsub(clo);
590 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
591 lle->lle_dom.lo_dom_r0.lo_nr = 1;
592 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
593 lle->lle_dom.lo_loi = loi;
595 rc = lov_page_slice_fixup(lov, clo);
600 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
605 * Implementation of lov_layout_operations::llo_fini for DOM object.
607 * Finish the DOM object and free related memory.
609 * \param[in] env execution environment
610 * \param[in] lov LOV object
611 * \param[in] state LOV layout state
613 static void lov_fini_dom(const struct lu_env *env,
614 struct lov_layout_entry *lle)
616 if (lle->lle_dom.lo_dom != NULL)
617 lle->lle_dom.lo_dom = NULL;
618 if (lle->lle_dom.lo_loi != NULL)
619 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
622 static struct lov_comp_layout_entry_ops dom_ops = {
623 .lco_init = lov_init_dom,
624 .lco_fini = lov_fini_dom,
625 .lco_getattr = lov_attr_get_dom,
628 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
629 struct lov_object *lov, struct lov_stripe_md *lsm,
630 const struct cl_object_conf *conf,
631 union lov_layout_state *state)
633 struct lov_layout_composite *comp = &state->composite;
634 struct lov_layout_entry *lle;
635 struct lov_mirror_entry *lre;
636 unsigned int entry_count;
637 unsigned int psz = 0;
638 unsigned int mirror_count;
639 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
646 LASSERT(lsm->lsm_entry_count > 0);
647 LASSERT(lov->lo_lsm == NULL);
648 lov->lo_lsm = lsm_addref(lsm);
649 lov->lo_layout_invalid = true;
651 dump_lsm(D_INODE, lsm);
653 entry_count = lsm->lsm_entry_count;
655 spin_lock_init(&comp->lo_write_lock);
656 comp->lo_flags = lsm->lsm_flags;
657 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
658 comp->lo_entry_count = lsm->lsm_entry_count;
659 comp->lo_preferred_mirror = -1;
661 if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
664 OBD_ALLOC(comp->lo_mirrors,
665 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
666 if (comp->lo_mirrors == NULL)
669 OBD_ALLOC(comp->lo_entries, entry_count * sizeof(*comp->lo_entries));
670 if (comp->lo_entries == NULL)
673 /* Initiate all entry types and extents data at first */
674 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
677 lle = &comp->lo_entries[i];
679 lle->lle_lsme = lsm->lsm_entries[i];
680 lle->lle_type = lov_entry_type(lle->lle_lsme);
681 switch (lle->lle_type) {
682 case LOV_PATTERN_RAID0:
683 lle->lle_comp_ops = &raid0_ops;
685 case LOV_PATTERN_MDT:
686 lle->lle_comp_ops = &dom_ops;
689 CERROR("%s: unknown composite layout entry type %i\n",
690 lov2obd(dev->ld_lov)->obd_name,
691 lsm->lsm_entries[i]->lsme_pattern);
692 dump_lsm(D_ERROR, lsm);
696 lle->lle_extent = &lle->lle_lsme->lsme_extent;
697 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
699 if (flr_state != LCM_FL_NONE)
700 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
702 lre = &comp->lo_mirrors[j];
704 if (mirror_id == lre->lre_mirror_id) {
705 lre->lre_valid |= lle->lle_valid;
706 lre->lre_stale |= !lle->lle_valid;
711 /* new mirror detected, assume that the mirrors
712 * are shorted in layout */
715 if (j >= comp->lo_mirror_count)
718 lre = &comp->lo_mirrors[j];
721 /* entries must be sorted by mirrors */
722 lre->lre_mirror_id = mirror_id;
723 lre->lre_start = lre->lre_end = i;
724 lre->lre_preferred = !!(lle->lle_lsme->lsme_flags &
726 lre->lre_valid = lle->lle_valid;
727 lre->lre_stale = !lle->lle_valid;
730 /* sanity check for FLR */
731 if (mirror_count != comp->lo_mirror_count) {
733 " doesn't have the # of mirrors it claims, %u/%u\n",
734 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
735 comp->lo_mirror_count + 1);
737 GOTO(out, result = -EINVAL);
740 lov_foreach_layout_entry(lov, lle) {
741 int index = lov_layout_entry_index(lov, lle);
744 * If the component has not been init-ed on MDS side, for
745 * PFL layout, we'd know that the components beyond this one
746 * will be dynamically init-ed later on file write/trunc ops.
748 if (!lsme_inited(lle->lle_lsme))
751 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
756 LASSERT(ergo(psz > 0, psz == result));
761 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
763 /* decide the preferred mirror. It uses the hash value of lov_object
764 * so that different clients would use different mirrors for read. */
766 seq = hash_long((unsigned long)lov, 8);
767 for (i = 0; i < comp->lo_mirror_count; i++) {
768 unsigned int idx = (i + seq) % comp->lo_mirror_count;
770 lre = lov_mirror_entry(lov, idx);
774 mirror_count++; /* valid mirror */
776 if (lre->lre_preferred || comp->lo_preferred_mirror < 0)
777 comp->lo_preferred_mirror = idx;
781 " doesn't have any valid mirrors\n",
782 PFID(lu_object_fid(lov2lu(lov))));
784 comp->lo_preferred_mirror = 0;
787 LASSERT(comp->lo_preferred_mirror >= 0);
791 return result > 0 ? 0 : result;
794 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
795 struct lov_object *lov, struct lov_stripe_md *lsm,
796 const struct cl_object_conf *conf,
797 union lov_layout_state *state)
802 static int lov_init_released(const struct lu_env *env,
803 struct lov_device *dev, struct lov_object *lov,
804 struct lov_stripe_md *lsm,
805 const struct cl_object_conf *conf,
806 union lov_layout_state *state)
808 LASSERT(lsm != NULL);
809 LASSERT(lsm->lsm_is_released);
810 LASSERT(lov->lo_lsm == NULL);
812 lov->lo_lsm = lsm_addref(lsm);
816 static int lov_init_foreign(const struct lu_env *env,
817 struct lov_device *dev, struct lov_object *lov,
818 struct lov_stripe_md *lsm,
819 const struct cl_object_conf *conf,
820 union lov_layout_state *state)
822 LASSERT(lsm != NULL);
823 LASSERT(lov->lo_type == LLT_FOREIGN);
824 LASSERT(lov->lo_lsm == NULL);
826 lov->lo_lsm = lsm_addref(lsm);
830 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
831 union lov_layout_state *state)
833 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED ||
834 lov->lo_type == LLT_FOREIGN);
836 lov_layout_wait(env, lov);
840 static int lov_delete_composite(const struct lu_env *env,
841 struct lov_object *lov,
842 union lov_layout_state *state)
844 struct lov_layout_entry *entry;
845 struct lov_layout_composite *comp = &state->composite;
849 dump_lsm(D_INODE, lov->lo_lsm);
851 lov_layout_wait(env, lov);
852 if (comp->lo_entries)
853 lov_foreach_layout_entry(lov, entry)
854 lov_delete_raid0(env, lov, entry);
859 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
860 union lov_layout_state *state)
862 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
865 static void lov_fini_composite(const struct lu_env *env,
866 struct lov_object *lov,
867 union lov_layout_state *state)
869 struct lov_layout_composite *comp = &state->composite;
872 if (comp->lo_entries != NULL) {
873 struct lov_layout_entry *entry;
875 lov_foreach_layout_entry(lov, entry)
876 entry->lle_comp_ops->lco_fini(env, entry);
878 OBD_FREE(comp->lo_entries,
879 comp->lo_entry_count * sizeof(*comp->lo_entries));
880 comp->lo_entries = NULL;
883 if (comp->lo_mirrors != NULL) {
884 OBD_FREE(comp->lo_mirrors,
885 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
886 comp->lo_mirrors = NULL;
889 memset(comp, 0, sizeof(*comp));
891 dump_lsm(D_INODE, lov->lo_lsm);
892 lov_free_memmd(&lov->lo_lsm);
897 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
898 union lov_layout_state *state)
901 dump_lsm(D_INODE, lov->lo_lsm);
902 lov_free_memmd(&lov->lo_lsm);
906 static int lov_print_empty(const struct lu_env *env, void *cookie,
907 lu_printer_t p, const struct lu_object *o)
909 (*p)(env, cookie, "empty %d\n", lu2lov(o)->lo_layout_invalid);
913 static int lov_print_composite(const struct lu_env *env, void *cookie,
914 lu_printer_t p, const struct lu_object *o)
916 struct lov_object *lov = lu2lov(o);
917 struct lov_stripe_md *lsm = lov->lo_lsm;
920 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
921 lsm->lsm_entry_count,
922 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
923 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
924 lsm->lsm_layout_gen);
926 for (i = 0; i < lsm->lsm_entry_count; i++) {
927 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
928 struct lov_layout_entry *lle = lov_entry(lov, i);
931 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
932 PEXT(&lse->lsme_extent), lse->lsme_magic,
933 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
934 lse->lsme_flags, lse->lsme_stripe_count,
935 lse->lsme_stripe_size);
936 lov_print_raid0(env, cookie, p, lle);
942 static int lov_print_released(const struct lu_env *env, void *cookie,
943 lu_printer_t p, const struct lu_object *o)
945 struct lov_object *lov = lu2lov(o);
946 struct lov_stripe_md *lsm = lov->lo_lsm;
949 "released: %s, lsm{%p 0x%08X %d %u}:\n",
950 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
951 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
952 lsm->lsm_layout_gen);
956 static int lov_print_foreign(const struct lu_env *env, void *cookie,
957 lu_printer_t p, const struct lu_object *o)
959 struct lov_object *lov = lu2lov(o);
960 struct lov_stripe_md *lsm = lov->lo_lsm;
963 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
964 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
965 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
966 lsm->lsm_layout_gen);
968 "raw_ea_content '%.*s'\n",
969 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
974 * Implements cl_object_operations::coo_attr_get() method for an object
975 * without stripes (LLT_EMPTY layout type).
977 * The only attributes this layer is authoritative in this case is
978 * cl_attr::cat_blocks---it's 0.
980 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
981 struct cl_attr *attr)
983 attr->cat_blocks = 0;
987 static int lov_attr_get_composite(const struct lu_env *env,
988 struct cl_object *obj,
989 struct cl_attr *attr)
991 struct lov_object *lov = cl2lov(obj);
992 struct lov_layout_entry *entry;
998 attr->cat_blocks = 0;
999 lov_foreach_layout_entry(lov, entry) {
1000 struct cl_attr *lov_attr = NULL;
1001 int index = lov_layout_entry_index(lov, entry);
1003 if (!entry->lle_valid)
1006 /* PFL: This component has not been init-ed. */
1007 if (!lsm_entry_inited(lov->lo_lsm, index))
1010 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1015 if (lov_attr == NULL)
1018 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1019 "b=%llu\n", index - 1, lov_attr->cat_size,
1020 lov_attr->cat_mtime, lov_attr->cat_atime,
1021 lov_attr->cat_ctime, lov_attr->cat_blocks);
1024 attr->cat_blocks += lov_attr->cat_blocks;
1025 if (attr->cat_size < lov_attr->cat_size)
1026 attr->cat_size = lov_attr->cat_size;
1027 if (attr->cat_kms < lov_attr->cat_kms)
1028 attr->cat_kms = lov_attr->cat_kms;
1029 if (attr->cat_atime < lov_attr->cat_atime)
1030 attr->cat_atime = lov_attr->cat_atime;
1031 if (attr->cat_ctime < lov_attr->cat_ctime)
1032 attr->cat_ctime = lov_attr->cat_ctime;
1033 if (attr->cat_mtime < lov_attr->cat_mtime)
1034 attr->cat_mtime = lov_attr->cat_mtime;
1040 static int lov_flush_composite(const struct lu_env *env,
1041 struct cl_object *obj,
1042 struct ldlm_lock *lock)
1044 struct lov_object *lov = cl2lov(obj);
1045 struct lovsub_object *lovsub;
1049 if (!lsme_is_dom(lov->lo_lsm->lsm_entries[0]))
1052 lovsub = lov->u.composite.lo_entries[0].lle_dom.lo_dom;
1053 RETURN(cl_object_flush(env, lovsub2cl(lovsub), lock));
1056 const static struct lov_layout_operations lov_dispatch[] = {
1058 .llo_init = lov_init_empty,
1059 .llo_delete = lov_delete_empty,
1060 .llo_fini = lov_fini_empty,
1061 .llo_print = lov_print_empty,
1062 .llo_page_init = lov_page_init_empty,
1063 .llo_lock_init = lov_lock_init_empty,
1064 .llo_io_init = lov_io_init_empty,
1065 .llo_getattr = lov_attr_get_empty,
1068 .llo_init = lov_init_released,
1069 .llo_delete = lov_delete_empty,
1070 .llo_fini = lov_fini_released,
1071 .llo_print = lov_print_released,
1072 .llo_page_init = lov_page_init_empty,
1073 .llo_lock_init = lov_lock_init_empty,
1074 .llo_io_init = lov_io_init_released,
1075 .llo_getattr = lov_attr_get_empty,
1078 .llo_init = lov_init_composite,
1079 .llo_delete = lov_delete_composite,
1080 .llo_fini = lov_fini_composite,
1081 .llo_print = lov_print_composite,
1082 .llo_page_init = lov_page_init_composite,
1083 .llo_lock_init = lov_lock_init_composite,
1084 .llo_io_init = lov_io_init_composite,
1085 .llo_getattr = lov_attr_get_composite,
1086 .llo_flush = lov_flush_composite,
1089 .llo_init = lov_init_foreign,
1090 .llo_delete = lov_delete_empty,
1091 .llo_fini = lov_fini_released,
1092 .llo_print = lov_print_foreign,
1093 .llo_page_init = lov_page_init_foreign,
1094 .llo_lock_init = lov_lock_init_empty,
1095 .llo_io_init = lov_io_init_empty,
1096 .llo_getattr = lov_attr_get_empty,
1101 * Performs a double-dispatch based on the layout type of an object.
1103 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1105 struct lov_object *__obj = (obj); \
1106 enum lov_layout_type __llt; \
1108 __llt = __obj->lo_type; \
1109 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1110 lov_dispatch[__llt].op(__VA_ARGS__); \
1114 * Return lov_layout_type associated with a given lsm
1116 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1121 if (lsm->lsm_is_released)
1122 return LLT_RELEASED;
1124 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1125 lsm->lsm_magic == LOV_MAGIC_V3 ||
1126 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1129 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1135 static inline void lov_conf_freeze(struct lov_object *lov)
1137 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1138 lov, lov->lo_owner, current);
1139 if (lov->lo_owner != current)
1140 down_read(&lov->lo_type_guard);
1143 static inline void lov_conf_thaw(struct lov_object *lov)
1145 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1146 lov, lov->lo_owner, current);
1147 if (lov->lo_owner != current)
1148 up_read(&lov->lo_type_guard);
1151 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1153 struct lov_object *__obj = (obj); \
1154 int __lock = !!(lock); \
1155 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1158 lov_conf_freeze(__obj); \
1159 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1161 lov_conf_thaw(__obj); \
1166 * Performs a locked double-dispatch based on the layout type of an object.
1168 #define LOV_2DISPATCH(obj, op, ...) \
1169 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1171 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1173 struct lov_object *__obj = (obj); \
1174 enum lov_layout_type __llt; \
1176 lov_conf_freeze(__obj); \
1177 __llt = __obj->lo_type; \
1178 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1179 lov_dispatch[__llt].op(__VA_ARGS__); \
1180 lov_conf_thaw(__obj); \
1183 static void lov_conf_lock(struct lov_object *lov)
1185 LASSERT(lov->lo_owner != current);
1186 down_write(&lov->lo_type_guard);
1187 LASSERT(lov->lo_owner == NULL);
1188 lov->lo_owner = current;
1189 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1190 lov, lov->lo_owner);
1193 static void lov_conf_unlock(struct lov_object *lov)
1195 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1196 lov, lov->lo_owner);
1197 lov->lo_owner = NULL;
1198 up_write(&lov->lo_type_guard);
1201 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1203 struct l_wait_info lwi = { 0 };
1206 while (atomic_read(&lov->lo_active_ios) > 0) {
1207 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1208 PFID(lu_object_fid(lov2lu(lov))),
1209 atomic_read(&lov->lo_active_ios));
1211 l_wait_event(lov->lo_waitq,
1212 atomic_read(&lov->lo_active_ios) == 0, &lwi);
1217 static int lov_layout_change(const struct lu_env *unused,
1218 struct lov_object *lov, struct lov_stripe_md *lsm,
1219 const struct cl_object_conf *conf)
1221 enum lov_layout_type llt = lov_type(lsm);
1222 union lov_layout_state *state = &lov->u;
1223 const struct lov_layout_operations *old_ops;
1224 const struct lov_layout_operations *new_ops;
1225 struct lov_device *lov_dev = lov_object_dev(lov);
1231 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1233 env = cl_env_get(&refcheck);
1235 RETURN(PTR_ERR(env));
1237 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1239 CDEBUG(D_INODE, DFID" from %s to %s\n",
1240 PFID(lu_object_fid(lov2lu(lov))),
1241 llt2str(lov->lo_type), llt2str(llt));
1243 old_ops = &lov_dispatch[lov->lo_type];
1244 new_ops = &lov_dispatch[llt];
1246 rc = cl_object_prune(env, &lov->lo_cl);
1250 rc = old_ops->llo_delete(env, lov, &lov->u);
1254 old_ops->llo_fini(env, lov, &lov->u);
1256 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1258 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1259 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1261 /* page bufsize fixup */
1262 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1263 lov_page_slice_fixup(lov, NULL);
1266 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1268 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1270 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1271 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1272 new_ops->llo_delete(env, lov, state);
1273 new_ops->llo_fini(env, lov, state);
1274 /* this file becomes an EMPTY file. */
1275 lov->lo_type = LLT_EMPTY;
1280 cl_env_put(env, &refcheck);
1284 /*****************************************************************************
1286 * Lov object operations.
1289 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1290 const struct lu_object_conf *conf)
1292 struct lov_object *lov = lu2lov(obj);
1293 struct lov_device *dev = lov_object_dev(lov);
1294 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1295 union lov_layout_state *set = &lov->u;
1296 const struct lov_layout_operations *ops;
1297 struct lov_stripe_md *lsm = NULL;
1301 init_rwsem(&lov->lo_type_guard);
1302 atomic_set(&lov->lo_active_ios, 0);
1303 init_waitqueue_head(&lov->lo_waitq);
1304 cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
1306 lov->lo_type = LLT_EMPTY;
1307 if (cconf->u.coc_layout.lb_buf != NULL) {
1308 lsm = lov_unpackmd(dev->ld_lov,
1309 cconf->u.coc_layout.lb_buf,
1310 cconf->u.coc_layout.lb_len);
1312 RETURN(PTR_ERR(lsm));
1314 dump_lsm(D_INODE, lsm);
1317 /* no locking is necessary, as object is being created */
1318 lov->lo_type = lov_type(lsm);
1319 ops = &lov_dispatch[lov->lo_type];
1320 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1330 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1331 const struct cl_object_conf *conf)
1333 struct lov_stripe_md *lsm = NULL;
1334 struct lov_object *lov = cl2lov(obj);
1338 if (conf->coc_opc == OBJECT_CONF_SET &&
1339 conf->u.coc_layout.lb_buf != NULL) {
1340 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1341 conf->u.coc_layout.lb_buf,
1342 conf->u.coc_layout.lb_len);
1344 RETURN(PTR_ERR(lsm));
1345 dump_lsm(D_INODE, lsm);
1349 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1350 lov->lo_layout_invalid = true;
1351 GOTO(out, result = 0);
1354 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1355 if (lov->lo_layout_invalid &&
1356 atomic_read(&lov->lo_active_ios) > 0) {
1357 lov_conf_unlock(lov);
1358 result = lov_layout_wait(env, lov);
1364 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1366 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1367 ((lsm != NULL && lov->lo_lsm != NULL) &&
1368 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1369 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1370 lsm->lsm_entries[0]->lsme_pattern))) {
1371 /* same version of layout */
1372 lov->lo_layout_invalid = false;
1373 GOTO(out, result = 0);
1376 /* will change layout - check if there still exists active IO. */
1377 if (atomic_read(&lov->lo_active_ios) > 0) {
1378 lov->lo_layout_invalid = true;
1379 GOTO(out, result = -EBUSY);
1382 result = lov_layout_change(env, lov, lsm, conf);
1383 lov->lo_layout_invalid = result != 0;
1387 lov_conf_unlock(lov);
1389 CDEBUG(D_INODE, DFID" lo_layout_invalid=%d\n",
1390 PFID(lu_object_fid(lov2lu(lov))), lov->lo_layout_invalid);
1394 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1396 struct lov_object *lov = lu2lov(obj);
1399 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1403 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1405 struct lov_object *lov = lu2lov(obj);
1408 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1409 lu_object_fini(obj);
1410 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1414 static int lov_object_print(const struct lu_env *env, void *cookie,
1415 lu_printer_t p, const struct lu_object *o)
1417 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1420 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1421 struct cl_page *page, pgoff_t index)
1423 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1428 * Implements cl_object_operations::clo_io_init() method for lov
1429 * layer. Dispatches to the appropriate layout io initialization method.
1431 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1434 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1436 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1437 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1438 io->ci_ignore_layout, io->ci_verify_layout);
1440 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1441 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1442 * because as long as the OSC object exists, the layout can't be
1444 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1445 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1450 * An implementation of cl_object_operations::clo_attr_get() method for lov
1451 * layer. For raid0 layout this collects and merges attributes of all
1454 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1455 struct cl_attr *attr)
1457 /* do not take lock, as this function is called under a
1458 * spin-lock. Layout is protected from changing by ongoing IO. */
1459 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1462 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1463 const struct cl_attr *attr, unsigned valid)
1466 * No dispatch is required here, as no layout implements this.
1471 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1472 struct cl_lock *lock, const struct cl_io *io)
1474 /* No need to lock because we've taken one refcount of layout. */
1475 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1480 * We calculate on which OST the mapping will end. If the length of mapping
1481 * is greater than (stripe_size * stripe_count) then the last_stripe will
1482 * will be one just before start_stripe. Else we check if the mapping
1483 * intersects each OST and find last_stripe.
1484 * This function returns the last_stripe and also sets the stripe_count
1485 * over which the mapping is spread
1487 * \param lsm [in] striping information for the file
1488 * \param index [in] stripe component index
1489 * \param ext [in] logical extent of mapping
1490 * \param start_stripe [in] starting stripe of the mapping
1491 * \param stripe_count [out] the number of stripes across which to map is
1494 * \retval last_stripe return the last stripe of the mapping
1496 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1497 struct lu_extent *ext,
1498 int start_stripe, int *stripe_count)
1500 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1506 if (ext->e_end - ext->e_start >
1507 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1508 last_stripe = (start_stripe < 1 ? lsme->lsme_stripe_count - 1 :
1510 *stripe_count = lsme->lsme_stripe_count;
1512 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1513 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1514 if ((lov_stripe_intersects(lsm, index, i, ext,
1515 &obd_start, &obd_end)) == 0)
1519 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1526 * Set fe_device and copy extents from local buffer into main return buffer.
1528 * \param fiemap [out] fiemap to hold all extents
1529 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1530 * \param ost_index [in] OST index to be written into the fm_device
1531 * field for each extent
1532 * \param ext_count [in] number of extents to be copied
1533 * \param current_extent [in] where to start copying in the extent array
1535 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1536 struct fiemap_extent *lcl_fm_ext,
1537 int ost_index, unsigned int ext_count,
1543 for (ext = 0; ext < ext_count; ext++) {
1544 lcl_fm_ext[ext].fe_device = ost_index;
1545 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1548 /* Copy fm_extent's from fm_local to return buffer */
1549 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1550 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1553 #define FIEMAP_BUFFER_SIZE 4096
1556 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1557 * call. The local end offset and the device are sent in the first
1558 * fm_extent. This function calculates the stripe number from the index.
1559 * This function returns a stripe_no on which mapping is to be restarted.
1561 * This function returns fm_end_offset which is the in-OST offset at which
1562 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1563 * will re-calculate proper offset in next stripe.
1564 * Note that the first extent is passed to lov_get_info via the value field.
1566 * \param fiemap [in] fiemap request header
1567 * \param lsm [in] striping information for the file
1568 * \param index [in] stripe component index
1569 * \param ext [in] logical extent of mapping
1570 * \param start_stripe [out] starting stripe will be returned in this
1572 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1573 struct lov_stripe_md *lsm,
1574 int index, struct lu_extent *ext,
1577 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1578 u64 local_end = fiemap->fm_extents[0].fe_logical;
1585 if (fiemap->fm_extent_count == 0 ||
1586 fiemap->fm_extents[0].fe_logical == 0)
1589 /* Find out stripe_no from ost_index saved in the fe_device */
1590 for (i = 0; i < lsme->lsme_stripe_count; i++) {
1591 struct lov_oinfo *oinfo = lsme->lsme_oinfo[i];
1593 if (lov_oinfo_is_dummy(oinfo))
1596 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1602 if (stripe_no == -1)
1605 /* If we have finished mapping on previous device, shift logical
1606 * offset to start of next device */
1607 if (lov_stripe_intersects(lsm, index, stripe_no, ext,
1608 &lun_start, &lun_end) != 0 &&
1609 local_end < lun_end) {
1610 fm_end_offset = local_end;
1611 *start_stripe = stripe_no;
1613 /* This is a special value to indicate that caller should
1614 * calculate offset in next stripe. */
1616 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1619 return fm_end_offset;
1622 struct fiemap_state {
1623 struct fiemap *fs_fm;
1624 struct lu_extent fs_ext;
1629 int fs_start_stripe;
1631 bool fs_device_done;
1632 bool fs_finish_stripe;
1636 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1637 struct lov_object *lov,
1638 struct lov_stripe_md *lsm,
1641 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1642 struct lov_thread_info *lti = lov_env_info(env);
1643 struct lu_fid *ofid = <i->lti_fid;
1644 struct lov_oinfo *oinfo;
1645 struct cl_device *subdev;
1646 int entry = lov_comp_entry(index);
1647 int stripe = lov_comp_stripe(index);
1650 struct cl_object *result;
1652 if (lov->lo_type != LLT_COMP)
1653 GOTO(out, result = NULL);
1655 if (entry >= lsm->lsm_entry_count ||
1656 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1657 GOTO(out, result = NULL);
1659 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1660 ost_idx = oinfo->loi_ost_idx;
1661 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1663 GOTO(out, result = NULL);
1665 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1666 result = lov_sub_find(env, subdev, ofid, NULL);
1669 result = ERR_PTR(-EINVAL);
1673 int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1674 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1675 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1676 int index, int stripeno, struct fiemap_state *fs)
1678 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1679 struct cl_object *subobj;
1680 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1681 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1682 u64 req_fm_len; /* Stores length of required mapping */
1683 u64 len_mapped_single_call;
1687 unsigned int ext_count;
1688 /* EOF for object */
1689 bool ost_eof = false;
1690 /* done with required mapping for this OST? */
1691 bool ost_done = false;
1695 fs->fs_device_done = false;
1696 /* Find out range of mapping on this stripe */
1697 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1698 &lun_start, &obd_object_end)) == 0)
1701 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1704 /* If this is a continuation FIEMAP call and we are on
1705 * starting stripe then lun_start needs to be set to
1707 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1708 lun_start = fs->fs_end_offset;
1709 lun_end = lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno);
1710 if (lun_start == lun_end)
1713 req_fm_len = obd_object_end - lun_start + 1;
1714 fs->fs_fm->fm_length = 0;
1715 len_mapped_single_call = 0;
1717 /* find lobsub object */
1718 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1719 lov_comp_index(index, stripeno));
1721 return PTR_ERR(subobj);
1722 /* If the output buffer is very large and the objects have many
1723 * extents we may need to loop on a single OST repeatedly */
1725 if (fiemap->fm_extent_count > 0) {
1726 /* Don't get too many extents. */
1727 if (fs->fs_cur_extent + fs->fs_cnt_need >
1728 fiemap->fm_extent_count)
1729 fs->fs_cnt_need = fiemap->fm_extent_count -
1733 lun_start += len_mapped_single_call;
1734 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1735 req_fm_len = fs->fs_fm->fm_length;
1737 * If we've collected enough extent map, we'd request 1 more,
1738 * to see whether we coincidentally finished all available
1739 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1741 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1742 1 : fs->fs_cnt_need;
1743 fs->fs_fm->fm_mapped_extents = 0;
1744 fs->fs_fm->fm_flags = fiemap->fm_flags;
1746 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1748 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1749 GOTO(obj_put, rc = -EINVAL);
1750 /* If OST is inactive, return extent with UNKNOWN flag. */
1751 if (!lov->lov_tgts[ost_index]->ltd_active) {
1752 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1753 fs->fs_fm->fm_mapped_extents = 1;
1755 fm_ext[0].fe_logical = lun_start;
1756 fm_ext[0].fe_length = obd_object_end - lun_start + 1;
1757 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1762 fs->fs_fm->fm_start = lun_start;
1763 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1764 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1765 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1767 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1771 ext_count = fs->fs_fm->fm_mapped_extents;
1772 if (ext_count == 0) {
1774 fs->fs_device_done = true;
1775 /* If last stripe has hold at the end,
1776 * we need to return */
1777 if (stripeno == fs->fs_last_stripe) {
1778 fiemap->fm_mapped_extents = 0;
1779 fs->fs_finish_stripe = true;
1783 } else if (fs->fs_enough) {
1785 * We've collected enough extents and there are
1786 * more extents after it.
1791 /* If we just need num of extents, got to next device */
1792 if (fiemap->fm_extent_count == 0) {
1793 fs->fs_cur_extent += ext_count;
1797 /* prepare to copy retrived map extents */
1798 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1799 fm_ext[ext_count - 1].fe_length -
1802 /* Have we finished mapping on this device? */
1803 if (req_fm_len <= len_mapped_single_call) {
1805 fs->fs_device_done = true;
1808 /* Clear the EXTENT_LAST flag which can be present on
1809 * the last extent */
1810 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1811 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1812 if (lov_stripe_size(lsm, index,
1813 fm_ext[ext_count - 1].fe_logical +
1814 fm_ext[ext_count - 1].fe_length,
1815 stripeno) >= fmkey->lfik_oa.o_size) {
1817 fs->fs_device_done = true;
1820 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1821 ext_count, fs->fs_cur_extent);
1822 fs->fs_cur_extent += ext_count;
1824 /* Ran out of available extents? */
1825 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1826 fs->fs_enough = true;
1827 } while (!ost_done && !ost_eof);
1829 if (stripeno == fs->fs_last_stripe)
1830 fs->fs_finish_stripe = true;
1832 cl_object_put(env, subobj);
1838 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1839 * This also handles the restarting of FIEMAP calls in case mapping overflows
1840 * the available number of extents in single call.
1842 * \param env [in] lustre environment
1843 * \param obj [in] file object
1844 * \param fmkey [in] fiemap request header and other info
1845 * \param fiemap [out] fiemap buffer holding retrived map extents
1846 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1847 * each OST, it is used to limit max map needed
1851 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1852 struct ll_fiemap_info_key *fmkey,
1853 struct fiemap *fiemap, size_t *buflen)
1855 struct lov_stripe_md_entry *lsme;
1856 struct lov_stripe_md *lsm;
1857 struct fiemap *fm_local = NULL;
1865 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1867 struct fiemap_state fs = { 0 };
1870 lsm = lov_lsm_addref(cl2lov(obj));
1872 /* no extent: there is no object for mapping */
1873 fiemap->fm_mapped_extents = 0;
1877 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1879 * If the entry count > 1 or stripe_count > 1 and the
1880 * application does not understand DEVICE_ORDER flag,
1881 * it cannot interpret the extents correctly.
1883 if (lsm->lsm_entry_count > 1 ||
1884 (lsm->lsm_entry_count == 1 &&
1885 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1886 GOTO(out_lsm, rc = -ENOTSUPP);
1889 /* No support for DOM layout yet. */
1890 if (lsme_is_dom(lsm->lsm_entries[0]))
1891 GOTO(out_lsm, rc = -ENOTSUPP);
1893 if (lsm->lsm_is_released) {
1894 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1896 * released file, return a minimal FIEMAP if
1897 * request fits in file-size.
1899 fiemap->fm_mapped_extents = 1;
1900 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1901 if (fiemap->fm_start + fiemap->fm_length <
1902 fmkey->lfik_oa.o_size)
1903 fiemap->fm_extents[0].fe_length =
1906 fiemap->fm_extents[0].fe_length =
1907 fmkey->lfik_oa.o_size -
1909 fiemap->fm_extents[0].fe_flags |=
1910 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1912 GOTO(out_lsm, rc = 0);
1915 /* buffer_size is small to hold fm_extent_count of extents. */
1916 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1917 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1919 OBD_ALLOC_LARGE(fm_local, buffer_size);
1920 if (fm_local == NULL)
1921 GOTO(out_lsm, rc = -ENOMEM);
1924 * Requested extent count exceeds the fiemap buffer size, shrink our
1927 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1928 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1929 if (fiemap->fm_extent_count == 0)
1932 fs.fs_enough = false;
1933 fs.fs_cur_extent = 0;
1934 fs.fs_fm = fm_local;
1935 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1937 whole_start = fiemap->fm_start;
1938 /* whole_start is beyond the end of the file */
1939 if (whole_start > fmkey->lfik_oa.o_size)
1940 GOTO(out_fm_local, rc = -EINVAL);
1941 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1942 fmkey->lfik_oa.o_size :
1943 whole_start + fiemap->fm_length - 1;
1945 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
1948 if (whole_end > fmkey->lfik_oa.o_size)
1949 whole_end = fmkey->lfik_oa.o_size;
1951 start_entry = lov_lsm_entry(lsm, whole_start);
1952 end_entry = lov_lsm_entry(lsm, whole_end);
1953 if (end_entry == -1)
1954 end_entry = lsm->lsm_entry_count - 1;
1956 if (start_entry == -1 || end_entry == -1)
1957 GOTO(out_fm_local, rc = -EINVAL);
1959 /* TODO: rewrite it with lov_foreach_io_layout() */
1960 for (entry = start_entry; entry <= end_entry; entry++) {
1961 lsme = lsm->lsm_entries[entry];
1963 if (!lsme_inited(lsme))
1966 if (entry == start_entry)
1967 fs.fs_ext.e_start = whole_start;
1969 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
1970 if (entry == end_entry)
1971 fs.fs_ext.e_end = whole_end;
1973 fs.fs_ext.e_end = lsme->lsme_extent.e_end - 1;
1974 fs.fs_length = fs.fs_ext.e_end - fs.fs_ext.e_start + 1;
1976 /* Calculate start stripe, last stripe and length of mapping */
1977 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
1979 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
1980 &fs.fs_ext, fs.fs_start_stripe,
1982 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, entry,
1983 &fs.fs_ext, &fs.fs_start_stripe);
1984 /* Check each stripe */
1985 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
1987 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
1988 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
1989 fmkey, entry, cur_stripe, &fs);
1991 GOTO(out_fm_local, rc);
1994 if (fs.fs_finish_stripe)
1996 } /* for each stripe */
1997 } /* for covering layout component */
1999 * We've traversed all components, set @entry to the last component
2000 * entry, it's for the last stripe check.
2004 /* Indicate that we are returning device offsets unless file just has
2006 if (lsm->lsm_entry_count > 1 ||
2007 (lsm->lsm_entry_count == 1 &&
2008 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2009 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2011 if (fiemap->fm_extent_count == 0)
2012 goto skip_last_device_calc;
2014 /* Check if we have reached the last stripe and whether mapping for that
2015 * stripe is done. */
2016 if ((cur_stripe == fs.fs_last_stripe) && fs.fs_device_done)
2017 fiemap->fm_extents[fs.fs_cur_extent - 1].fe_flags |=
2019 skip_last_device_calc:
2020 fiemap->fm_mapped_extents = fs.fs_cur_extent;
2022 OBD_FREE_LARGE(fm_local, buffer_size);
2029 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2030 struct lov_user_md __user *lum, size_t size)
2032 struct lov_object *lov = cl2lov(obj);
2033 struct lov_stripe_md *lsm;
2037 lsm = lov_lsm_addref(lov);
2041 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2046 static int lov_object_layout_get(const struct lu_env *env,
2047 struct cl_object *obj,
2048 struct cl_layout *cl)
2050 struct lov_object *lov = cl2lov(obj);
2051 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2052 struct lu_buf *buf = &cl->cl_buf;
2058 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2063 cl->cl_size = lov_comp_md_size(lsm);
2064 cl->cl_layout_gen = lsm->lsm_layout_gen;
2065 cl->cl_dom_comp_size = 0;
2066 cl->cl_is_released = lsm->lsm_is_released;
2067 if (lsm_is_composite(lsm->lsm_magic)) {
2068 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[0];
2070 cl->cl_is_composite = true;
2072 if (lsme_is_dom(lsme))
2073 cl->cl_dom_comp_size = lsme->lsme_extent.e_end;
2075 cl->cl_is_composite = false;
2078 rc = lov_lsm_pack(lsm, buf->lb_buf, buf->lb_len);
2081 RETURN(rc < 0 ? rc : 0);
2084 static loff_t lov_object_maxbytes(struct cl_object *obj)
2086 struct lov_object *lov = cl2lov(obj);
2087 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2093 maxbytes = lsm->lsm_maxbytes;
2100 static int lov_object_flush(const struct lu_env *env, struct cl_object *obj,
2101 struct ldlm_lock *lock)
2103 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_flush, env, obj, lock);
2106 static const struct cl_object_operations lov_ops = {
2107 .coo_page_init = lov_page_init,
2108 .coo_lock_init = lov_lock_init,
2109 .coo_io_init = lov_io_init,
2110 .coo_attr_get = lov_attr_get,
2111 .coo_attr_update = lov_attr_update,
2112 .coo_conf_set = lov_conf_set,
2113 .coo_getstripe = lov_object_getstripe,
2114 .coo_layout_get = lov_object_layout_get,
2115 .coo_maxbytes = lov_object_maxbytes,
2116 .coo_fiemap = lov_object_fiemap,
2117 .coo_object_flush = lov_object_flush
2120 static const struct lu_object_operations lov_lu_obj_ops = {
2121 .loo_object_init = lov_object_init,
2122 .loo_object_delete = lov_object_delete,
2123 .loo_object_release = NULL,
2124 .loo_object_free = lov_object_free,
2125 .loo_object_print = lov_object_print,
2126 .loo_object_invariant = NULL
2129 struct lu_object *lov_object_alloc(const struct lu_env *env,
2130 const struct lu_object_header *unused,
2131 struct lu_device *dev)
2133 struct lov_object *lov;
2134 struct lu_object *obj;
2137 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2140 lu_object_init(obj, NULL, dev);
2141 lov->lo_cl.co_ops = &lov_ops;
2142 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2144 * object io operation vector (cl_object::co_iop) is installed
2145 * later in lov_object_init(), as different vectors are used
2146 * for object with different layouts.
2148 obj->lo_ops = &lov_lu_obj_ops;
2154 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2156 struct lov_stripe_md *lsm = NULL;
2158 lov_conf_freeze(lov);
2159 if (lov->lo_lsm != NULL) {
2160 lsm = lsm_addref(lov->lo_lsm);
2161 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2162 lsm, atomic_read(&lsm->lsm_refc),
2163 lov->lo_layout_invalid, current);
2169 int lov_read_and_clear_async_rc(struct cl_object *clob)
2171 struct lu_object *luobj;
2175 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2177 if (luobj != NULL) {
2178 struct lov_object *lov = lu2lov(luobj);
2180 lov_conf_freeze(lov);
2181 switch (lov->lo_type) {
2183 struct lov_stripe_md *lsm;
2187 LASSERT(lsm != NULL);
2188 for (i = 0; i < lsm->lsm_entry_count; i++) {
2189 struct lov_stripe_md_entry *lse =
2190 lsm->lsm_entries[i];
2193 if (!lsme_inited(lse))
2196 for (j = 0; j < lse->lsme_stripe_count; j++) {
2197 struct lov_oinfo *loi =
2200 if (lov_oinfo_is_dummy(loi))
2203 if (loi->loi_ar.ar_rc && !rc)
2204 rc = loi->loi_ar.ar_rc;
2205 loi->loi_ar.ar_rc = 0;
2221 EXPORT_SYMBOL(lov_read_and_clear_async_rc);