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;
288 LASSERT(r0->lo_sub[idx] == los);
290 sub = lovsub2cl(los);
291 site = sub->co_lu.lo_dev->ld_site;
292 wq = lu_site_wq_from_fid(site, &sub->co_lu.lo_header->loh_fid);
294 cl_object_kill(env, sub);
295 /* release a reference to the sub-object and ... */
296 lu_object_ref_del(&sub->co_lu, "lov-parent", lov);
297 cl_object_put(env, sub);
299 /* ... wait until it is actually destroyed---sub-object clears its
300 * ->lo_sub[] slot in lovsub_object_free() */
301 wait_event(*wq, r0->lo_sub[idx] != los);
302 LASSERT(r0->lo_sub[idx] == NULL);
305 static void lov_delete_raid0(const struct lu_env *env, struct lov_object *lov,
306 struct lov_layout_entry *lle)
308 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
312 if (r0->lo_sub != NULL) {
315 for (i = 0; i < r0->lo_nr; ++i) {
316 struct lovsub_object *los = r0->lo_sub[i];
319 cl_object_prune(env, &los->lso_cl);
321 * If top-level object is to be evicted from
322 * the cache, so are its sub-objects.
324 lov_subobject_kill(env, lov, r0, los, i);
332 static void lov_fini_raid0(const struct lu_env *env,
333 struct lov_layout_entry *lle)
335 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
337 if (r0->lo_sub != NULL) {
338 OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
343 static int lov_print_raid0(const struct lu_env *env, void *cookie,
344 lu_printer_t p, const struct lov_layout_entry *lle)
346 const struct lov_layout_raid0 *r0 = &lle->lle_raid0;
349 for (i = 0; i < r0->lo_nr; ++i) {
350 struct lu_object *sub;
352 if (r0->lo_sub[i] != NULL) {
353 sub = lovsub2lu(r0->lo_sub[i]);
354 lu_object_print(env, cookie, p, sub);
356 (*p)(env, cookie, "sub %d absent\n", i);
362 static int lov_attr_get_raid0(const struct lu_env *env, struct lov_object *lov,
363 unsigned int index, struct lov_layout_entry *lle,
364 struct cl_attr **lov_attr)
366 struct lov_layout_raid0 *r0 = &lle->lle_raid0;
367 struct lov_stripe_md *lsm = lov->lo_lsm;
368 struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
369 struct cl_attr *attr = &r0->lo_attr;
373 if (r0->lo_attr_valid) {
378 memset(lvb, 0, sizeof(*lvb));
380 /* XXX: timestamps can be negative by sanity:test_39m,
382 lvb->lvb_atime = LLONG_MIN;
383 lvb->lvb_ctime = LLONG_MIN;
384 lvb->lvb_mtime = LLONG_MIN;
387 * XXX that should be replaced with a loop over sub-objects,
388 * doing cl_object_attr_get() on them. But for now, let's
389 * reuse old lov code.
393 * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
394 * happy. It's not needed, because new code uses
395 * ->coh_attr_guard spin-lock to protect consistency of
396 * sub-object attributes.
398 lov_stripe_lock(lsm);
399 result = lov_merge_lvb_kms(lsm, index, lvb, &kms);
400 lov_stripe_unlock(lsm);
402 cl_lvb2attr(attr, lvb);
404 r0->lo_attr_valid = 1;
411 static struct lov_comp_layout_entry_ops raid0_ops = {
412 .lco_init = lov_init_raid0,
413 .lco_fini = lov_fini_raid0,
414 .lco_getattr = lov_attr_get_raid0,
417 static int lov_attr_get_dom(const struct lu_env *env, struct lov_object *lov,
418 unsigned int index, struct lov_layout_entry *lle,
419 struct cl_attr **lov_attr)
421 struct lov_layout_dom *dom = &lle->lle_dom;
422 struct lov_oinfo *loi = dom->lo_loi;
423 struct cl_attr *attr = &dom->lo_dom_r0.lo_attr;
425 if (dom->lo_dom_r0.lo_attr_valid) {
430 if (OST_LVB_IS_ERR(loi->loi_lvb.lvb_blocks))
431 return OST_LVB_GET_ERR(loi->loi_lvb.lvb_blocks);
433 cl_lvb2attr(attr, &loi->loi_lvb);
435 /* DoM component size can be bigger than stripe size after
436 * client's setattr RPC, so do not count anything beyond
437 * component end. Alternatively, check that limit on server
438 * and do not allow size overflow there. */
439 if (attr->cat_size > lle->lle_extent->e_end)
440 attr->cat_size = lle->lle_extent->e_end;
442 attr->cat_kms = attr->cat_size;
444 dom->lo_dom_r0.lo_attr_valid = 1;
451 * Lookup FLD to get MDS index of the given DOM object FID.
453 * \param[in] ld LOV device
454 * \param[in] fid FID to lookup
455 * \param[out] nr index in MDC array to return back
457 * \retval 0 and \a mds filled with MDS index if successful
458 * \retval negative value on error
460 static int lov_fld_lookup(struct lov_device *ld, const struct lu_fid *fid,
468 rc = fld_client_lookup(&ld->ld_lmv->u.lmv.lmv_fld, fid_seq(fid),
469 &mds_idx, LU_SEQ_RANGE_MDT, NULL);
471 CERROR("%s: error while looking for mds number. Seq %#llx"
472 ", err = %d\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
477 CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
480 /* find proper MDC device in the array */
481 for (i = 0; i < ld->ld_md_tgts_nr; i++) {
482 if (ld->ld_md_tgts[i].ldm_mdc != NULL &&
483 ld->ld_md_tgts[i].ldm_idx == mds_idx)
487 if (i == ld->ld_md_tgts_nr) {
488 CERROR("%s: cannot find corresponding MDC device for mds #%x "
489 "for fid="DFID"\n", lu_dev_name(cl2lu_dev(&ld->ld_cl)),
499 * Implementation of lov_comp_layout_entry_ops::lco_init for DOM object.
501 * Init the DOM object for the first time. It prepares also RAID0 entry
502 * for it to use in common methods with ordinary RAID0 layout entries.
504 * \param[in] env execution environment
505 * \param[in] dev LOV device
506 * \param[in] lov LOV object
507 * \param[in] index Composite layout entry index in LSM
508 * \param[in] lle Composite LOV layout entry
510 static int lov_init_dom(const struct lu_env *env, struct lov_device *dev,
511 struct lov_object *lov, unsigned int index,
512 const struct cl_object_conf *conf,
513 struct lov_layout_entry *lle)
515 struct lov_thread_info *lti = lov_env_info(env);
516 struct lov_stripe_md_entry *lsme = lov_lse(lov, index);
517 struct cl_object *clo;
518 struct lu_object *o = lov2lu(lov);
519 const struct lu_fid *fid = lu_object_fid(o);
520 struct cl_device *mdcdev;
521 struct lov_oinfo *loi = NULL;
522 struct cl_object_conf *sconf = <i->lti_stripe_conf;
528 /* DOM entry may be not zero index due to FLR but must start from 0 */
529 if (unlikely(lle->lle_extent->e_start != 0)) {
530 CERROR("%s: DOM entry must be the first stripe in a mirror\n",
531 lov2obd(dev->ld_lov)->obd_name);
532 dump_lsm(D_ERROR, lov->lo_lsm);
536 /* find proper MDS device */
537 rc = lov_fld_lookup(dev, fid, &idx);
541 LASSERTF(dev->ld_md_tgts[idx].ldm_mdc != NULL,
542 "LOV md target[%u] is NULL\n", idx);
544 /* check lsm is DOM, more checks are needed */
545 LASSERT(lsme->lsme_stripe_count == 0);
548 * Create lower cl_objects.
550 mdcdev = dev->ld_md_tgts[idx].ldm_mdc;
552 LASSERTF(mdcdev != NULL, "non-initialized mdc subdev\n");
554 /* DoM object has no oinfo in LSM entry, create it exclusively */
555 OBD_SLAB_ALLOC_PTR_GFP(loi, lov_oinfo_slab, GFP_NOFS);
559 fid_to_ostid(lu_object_fid(lov2lu(lov)), &loi->loi_oi);
561 sconf->u.coc_oinfo = loi;
563 clo = lov_sub_find(env, mdcdev, fid, sconf);
565 GOTO(out, rc = PTR_ERR(clo));
567 rc = lov_init_sub(env, lov, clo, loi, lov_comp_index(index, 0));
568 if (rc == -EAGAIN) /* try again */
573 lle->lle_dom.lo_dom = cl2lovsub(clo);
574 spin_lock_init(&lle->lle_dom.lo_dom_r0.lo_sub_lock);
575 lle->lle_dom.lo_dom_r0.lo_nr = 1;
576 lle->lle_dom.lo_dom_r0.lo_sub = &lle->lle_dom.lo_dom;
577 lle->lle_dom.lo_loi = loi;
579 rc = lov_page_slice_fixup(lov, clo);
584 OBD_SLAB_FREE_PTR(loi, lov_oinfo_slab);
589 * Implementation of lov_layout_operations::llo_fini for DOM object.
591 * Finish the DOM object and free related memory.
593 * \param[in] env execution environment
594 * \param[in] lov LOV object
595 * \param[in] state LOV layout state
597 static void lov_fini_dom(const struct lu_env *env,
598 struct lov_layout_entry *lle)
600 if (lle->lle_dom.lo_dom != NULL)
601 lle->lle_dom.lo_dom = NULL;
602 if (lle->lle_dom.lo_loi != NULL)
603 OBD_SLAB_FREE_PTR(lle->lle_dom.lo_loi, lov_oinfo_slab);
606 static struct lov_comp_layout_entry_ops dom_ops = {
607 .lco_init = lov_init_dom,
608 .lco_fini = lov_fini_dom,
609 .lco_getattr = lov_attr_get_dom,
612 static int lov_init_composite(const struct lu_env *env, struct lov_device *dev,
613 struct lov_object *lov, struct lov_stripe_md *lsm,
614 const struct cl_object_conf *conf,
615 union lov_layout_state *state)
617 struct lov_layout_composite *comp = &state->composite;
618 struct lov_layout_entry *lle;
619 struct lov_mirror_entry *lre;
620 unsigned int entry_count;
621 unsigned int psz = 0;
622 unsigned int mirror_count;
623 int flr_state = lsm->lsm_flags & LCM_FL_FLR_MASK;
631 LASSERT(lsm->lsm_entry_count > 0);
632 LASSERT(lov->lo_lsm == NULL);
633 lov->lo_lsm = lsm_addref(lsm);
634 lov->lo_layout_invalid = true;
636 dump_lsm(D_INODE, lsm);
638 entry_count = lsm->lsm_entry_count;
640 spin_lock_init(&comp->lo_write_lock);
641 comp->lo_flags = lsm->lsm_flags;
642 comp->lo_mirror_count = lsm->lsm_mirror_count + 1;
643 comp->lo_entry_count = lsm->lsm_entry_count;
644 comp->lo_preferred_mirror = -1;
646 if (equi(flr_state == LCM_FL_NONE, comp->lo_mirror_count > 1))
649 OBD_ALLOC(comp->lo_mirrors,
650 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
651 if (comp->lo_mirrors == NULL)
654 OBD_ALLOC(comp->lo_entries, entry_count * sizeof(*comp->lo_entries));
655 if (comp->lo_entries == NULL)
658 /* Initiate all entry types and extents data at first */
659 for (i = 0, j = 0, mirror_count = 1; i < entry_count; i++) {
662 lle = &comp->lo_entries[i];
664 lle->lle_lsme = lsm->lsm_entries[i];
665 lle->lle_type = lov_entry_type(lle->lle_lsme);
666 switch (lle->lle_type) {
667 case LOV_PATTERN_RAID0:
668 lle->lle_comp_ops = &raid0_ops;
670 case LOV_PATTERN_MDT:
671 /* Allowed to have several DOM stripes in different
672 * mirrors with the same DoM size.
675 dom_size = lle->lle_lsme->lsme_extent.e_end;
676 } else if (dom_size !=
677 lle->lle_lsme->lsme_extent.e_end) {
678 CERROR("%s: DOM entries with different sizes\n",
679 lov2obd(dev->ld_lov)->obd_name);
680 dump_lsm(D_ERROR, lsm);
683 lle->lle_comp_ops = &dom_ops;
686 CERROR("%s: unknown composite layout entry type %i\n",
687 lov2obd(dev->ld_lov)->obd_name,
688 lsm->lsm_entries[i]->lsme_pattern);
689 dump_lsm(D_ERROR, lsm);
693 lle->lle_extent = &lle->lle_lsme->lsme_extent;
694 lle->lle_valid = !(lle->lle_lsme->lsme_flags & LCME_FL_STALE);
696 if (flr_state != LCM_FL_NONE)
697 mirror_id = mirror_id_of(lle->lle_lsme->lsme_id);
699 lre = &comp->lo_mirrors[j];
701 if (mirror_id == lre->lre_mirror_id) {
702 lre->lre_valid |= lle->lle_valid;
703 lre->lre_stale |= !lle->lle_valid;
708 /* new mirror detected, assume that the mirrors
709 * are shorted in layout */
712 if (j >= comp->lo_mirror_count)
715 lre = &comp->lo_mirrors[j];
718 /* entries must be sorted by mirrors */
719 lre->lre_mirror_id = mirror_id;
720 lre->lre_start = lre->lre_end = i;
721 lre->lre_preferred = !!(lle->lle_lsme->lsme_flags &
723 lre->lre_valid = lle->lle_valid;
724 lre->lre_stale = !lle->lle_valid;
727 /* sanity check for FLR */
728 if (mirror_count != comp->lo_mirror_count) {
730 " doesn't have the # of mirrors it claims, %u/%u\n",
731 PFID(lu_object_fid(lov2lu(lov))), mirror_count,
732 comp->lo_mirror_count + 1);
734 GOTO(out, result = -EINVAL);
737 lov_foreach_layout_entry(lov, lle) {
738 int index = lov_layout_entry_index(lov, lle);
741 * If the component has not been init-ed on MDS side, for
742 * PFL layout, we'd know that the components beyond this one
743 * will be dynamically init-ed later on file write/trunc ops.
745 if (!lsme_inited(lle->lle_lsme))
748 result = lle->lle_comp_ops->lco_init(env, dev, lov, index,
753 LASSERT(ergo(psz > 0, psz == result));
758 cl_object_header(&lov->lo_cl)->coh_page_bufsize += psz;
760 /* decide the preferred mirror. It uses the hash value of lov_object
761 * so that different clients would use different mirrors for read. */
763 seq = 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 mirror_count++; /* valid mirror */
773 if (lre->lre_preferred || comp->lo_preferred_mirror < 0)
774 comp->lo_preferred_mirror = idx;
778 " doesn't have any valid mirrors\n",
779 PFID(lu_object_fid(lov2lu(lov))));
781 comp->lo_preferred_mirror = 0;
784 LASSERT(comp->lo_preferred_mirror >= 0);
788 return result > 0 ? 0 : result;
791 static int lov_init_empty(const struct lu_env *env, struct lov_device *dev,
792 struct lov_object *lov, struct lov_stripe_md *lsm,
793 const struct cl_object_conf *conf,
794 union lov_layout_state *state)
799 static int lov_init_released(const struct lu_env *env,
800 struct lov_device *dev, struct lov_object *lov,
801 struct lov_stripe_md *lsm,
802 const struct cl_object_conf *conf,
803 union lov_layout_state *state)
805 LASSERT(lsm != NULL);
806 LASSERT(lsm->lsm_is_released);
807 LASSERT(lov->lo_lsm == NULL);
809 lov->lo_lsm = lsm_addref(lsm);
813 static int lov_init_foreign(const struct lu_env *env,
814 struct lov_device *dev, struct lov_object *lov,
815 struct lov_stripe_md *lsm,
816 const struct cl_object_conf *conf,
817 union lov_layout_state *state)
819 LASSERT(lsm != NULL);
820 LASSERT(lov->lo_type == LLT_FOREIGN);
821 LASSERT(lov->lo_lsm == NULL);
823 lov->lo_lsm = lsm_addref(lsm);
827 static int lov_delete_empty(const struct lu_env *env, struct lov_object *lov,
828 union lov_layout_state *state)
830 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED ||
831 lov->lo_type == LLT_FOREIGN);
833 lov_layout_wait(env, lov);
837 static int lov_delete_composite(const struct lu_env *env,
838 struct lov_object *lov,
839 union lov_layout_state *state)
841 struct lov_layout_entry *entry;
842 struct lov_layout_composite *comp = &state->composite;
846 dump_lsm(D_INODE, lov->lo_lsm);
848 lov_layout_wait(env, lov);
849 if (comp->lo_entries)
850 lov_foreach_layout_entry(lov, entry)
851 lov_delete_raid0(env, lov, entry);
856 static void lov_fini_empty(const struct lu_env *env, struct lov_object *lov,
857 union lov_layout_state *state)
859 LASSERT(lov->lo_type == LLT_EMPTY || lov->lo_type == LLT_RELEASED);
862 static void lov_fini_composite(const struct lu_env *env,
863 struct lov_object *lov,
864 union lov_layout_state *state)
866 struct lov_layout_composite *comp = &state->composite;
869 if (comp->lo_entries != NULL) {
870 struct lov_layout_entry *entry;
872 lov_foreach_layout_entry(lov, entry)
873 if (entry->lle_comp_ops)
874 entry->lle_comp_ops->lco_fini(env, entry);
876 OBD_FREE(comp->lo_entries,
877 comp->lo_entry_count * sizeof(*comp->lo_entries));
878 comp->lo_entries = NULL;
881 if (comp->lo_mirrors != NULL) {
882 OBD_FREE(comp->lo_mirrors,
883 comp->lo_mirror_count * sizeof(*comp->lo_mirrors));
884 comp->lo_mirrors = NULL;
887 memset(comp, 0, sizeof(*comp));
889 dump_lsm(D_INODE, lov->lo_lsm);
890 lov_free_memmd(&lov->lo_lsm);
895 static void lov_fini_released(const struct lu_env *env, struct lov_object *lov,
896 union lov_layout_state *state)
899 dump_lsm(D_INODE, lov->lo_lsm);
900 lov_free_memmd(&lov->lo_lsm);
904 static int lov_print_empty(const struct lu_env *env, void *cookie,
905 lu_printer_t p, const struct lu_object *o)
907 (*p)(env, cookie, "empty %d\n", lu2lov(o)->lo_layout_invalid);
911 static int lov_print_composite(const struct lu_env *env, void *cookie,
912 lu_printer_t p, const struct lu_object *o)
914 struct lov_object *lov = lu2lov(o);
915 struct lov_stripe_md *lsm = lov->lo_lsm;
918 (*p)(env, cookie, "entries: %d, %s, lsm{%p 0x%08X %d %u}:\n",
919 lsm->lsm_entry_count,
920 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
921 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
922 lsm->lsm_layout_gen);
924 for (i = 0; i < lsm->lsm_entry_count; i++) {
925 struct lov_stripe_md_entry *lse = lsm->lsm_entries[i];
926 struct lov_layout_entry *lle = lov_entry(lov, i);
929 DEXT ": { 0x%08X, %u, %#x, %u, %#x, %u, %u }\n",
930 PEXT(&lse->lsme_extent), lse->lsme_magic,
931 lse->lsme_id, lse->lsme_pattern, lse->lsme_layout_gen,
932 lse->lsme_flags, lse->lsme_stripe_count,
933 lse->lsme_stripe_size);
934 lov_print_raid0(env, cookie, p, lle);
940 static int lov_print_released(const struct lu_env *env, void *cookie,
941 lu_printer_t p, const struct lu_object *o)
943 struct lov_object *lov = lu2lov(o);
944 struct lov_stripe_md *lsm = lov->lo_lsm;
947 "released: %s, lsm{%p 0x%08X %d %u}:\n",
948 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
949 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
950 lsm->lsm_layout_gen);
954 static int lov_print_foreign(const struct lu_env *env, void *cookie,
955 lu_printer_t p, const struct lu_object *o)
957 struct lov_object *lov = lu2lov(o);
958 struct lov_stripe_md *lsm = lov->lo_lsm;
961 "foreign: %s, lsm{%p 0x%08X %d %u}:\n",
962 lov->lo_layout_invalid ? "invalid" : "valid", lsm,
963 lsm->lsm_magic, atomic_read(&lsm->lsm_refc),
964 lsm->lsm_layout_gen);
966 "raw_ea_content '%.*s'\n",
967 (int)lsm->lsm_foreign_size, (char *)lsm_foreign(lsm));
972 * Implements cl_object_operations::coo_attr_get() method for an object
973 * without stripes (LLT_EMPTY layout type).
975 * The only attributes this layer is authoritative in this case is
976 * cl_attr::cat_blocks---it's 0.
978 static int lov_attr_get_empty(const struct lu_env *env, struct cl_object *obj,
979 struct cl_attr *attr)
981 attr->cat_blocks = 0;
985 static int lov_attr_get_composite(const struct lu_env *env,
986 struct cl_object *obj,
987 struct cl_attr *attr)
989 struct lov_object *lov = cl2lov(obj);
990 struct lov_layout_entry *entry;
996 attr->cat_blocks = 0;
997 lov_foreach_layout_entry(lov, entry) {
998 struct cl_attr *lov_attr = NULL;
999 int index = lov_layout_entry_index(lov, entry);
1001 if (!entry->lle_valid)
1004 /* PFL: This component has not been init-ed. */
1005 if (!lsm_entry_inited(lov->lo_lsm, index))
1008 result = entry->lle_comp_ops->lco_getattr(env, lov, index,
1013 if (lov_attr == NULL)
1016 CDEBUG(D_INODE, "COMP ID #%i: s=%llu m=%llu a=%llu c=%llu "
1017 "b=%llu\n", index - 1, lov_attr->cat_size,
1018 lov_attr->cat_mtime, lov_attr->cat_atime,
1019 lov_attr->cat_ctime, lov_attr->cat_blocks);
1022 attr->cat_blocks += lov_attr->cat_blocks;
1023 if (attr->cat_size < lov_attr->cat_size)
1024 attr->cat_size = lov_attr->cat_size;
1025 if (attr->cat_kms < lov_attr->cat_kms)
1026 attr->cat_kms = lov_attr->cat_kms;
1027 if (attr->cat_atime < lov_attr->cat_atime)
1028 attr->cat_atime = lov_attr->cat_atime;
1029 if (attr->cat_ctime < lov_attr->cat_ctime)
1030 attr->cat_ctime = lov_attr->cat_ctime;
1031 if (attr->cat_mtime < lov_attr->cat_mtime)
1032 attr->cat_mtime = lov_attr->cat_mtime;
1038 static int lov_flush_composite(const struct lu_env *env,
1039 struct cl_object *obj,
1040 struct ldlm_lock *lock)
1042 struct lov_object *lov = cl2lov(obj);
1043 struct lov_layout_entry *lle;
1048 lov_foreach_layout_entry(lov, lle) {
1049 if (!lsme_is_dom(lle->lle_lsme))
1051 rc = cl_object_flush(env, lovsub2cl(lle->lle_dom.lo_dom), lock);
1058 static int lov_flush_empty(const struct lu_env *env, struct cl_object *obj,
1059 struct ldlm_lock *lock)
1064 const static struct lov_layout_operations lov_dispatch[] = {
1066 .llo_init = lov_init_empty,
1067 .llo_delete = lov_delete_empty,
1068 .llo_fini = lov_fini_empty,
1069 .llo_print = lov_print_empty,
1070 .llo_page_init = lov_page_init_empty,
1071 .llo_lock_init = lov_lock_init_empty,
1072 .llo_io_init = lov_io_init_empty,
1073 .llo_getattr = lov_attr_get_empty,
1074 .llo_flush = lov_flush_empty,
1077 .llo_init = lov_init_released,
1078 .llo_delete = lov_delete_empty,
1079 .llo_fini = lov_fini_released,
1080 .llo_print = lov_print_released,
1081 .llo_page_init = lov_page_init_empty,
1082 .llo_lock_init = lov_lock_init_empty,
1083 .llo_io_init = lov_io_init_released,
1084 .llo_getattr = lov_attr_get_empty,
1085 .llo_flush = lov_flush_empty,
1088 .llo_init = lov_init_composite,
1089 .llo_delete = lov_delete_composite,
1090 .llo_fini = lov_fini_composite,
1091 .llo_print = lov_print_composite,
1092 .llo_page_init = lov_page_init_composite,
1093 .llo_lock_init = lov_lock_init_composite,
1094 .llo_io_init = lov_io_init_composite,
1095 .llo_getattr = lov_attr_get_composite,
1096 .llo_flush = lov_flush_composite,
1099 .llo_init = lov_init_foreign,
1100 .llo_delete = lov_delete_empty,
1101 .llo_fini = lov_fini_released,
1102 .llo_print = lov_print_foreign,
1103 .llo_page_init = lov_page_init_foreign,
1104 .llo_lock_init = lov_lock_init_empty,
1105 .llo_io_init = lov_io_init_empty,
1106 .llo_getattr = lov_attr_get_empty,
1107 .llo_flush = lov_flush_empty,
1112 * Performs a double-dispatch based on the layout type of an object.
1114 #define LOV_2DISPATCH_NOLOCK(obj, op, ...) \
1116 struct lov_object *__obj = (obj); \
1117 enum lov_layout_type __llt; \
1119 __llt = __obj->lo_type; \
1120 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1121 lov_dispatch[__llt].op(__VA_ARGS__); \
1125 * Return lov_layout_type associated with a given lsm
1127 static enum lov_layout_type lov_type(struct lov_stripe_md *lsm)
1132 if (lsm->lsm_is_released)
1133 return LLT_RELEASED;
1135 if (lsm->lsm_magic == LOV_MAGIC_V1 ||
1136 lsm->lsm_magic == LOV_MAGIC_V3 ||
1137 lsm->lsm_magic == LOV_MAGIC_COMP_V1)
1140 if (lsm->lsm_magic == LOV_MAGIC_FOREIGN)
1146 static inline void lov_conf_freeze(struct lov_object *lov)
1148 CDEBUG(D_INODE, "To take share lov(%p) owner %p/%p\n",
1149 lov, lov->lo_owner, current);
1150 if (lov->lo_owner != current)
1151 down_read(&lov->lo_type_guard);
1154 static inline void lov_conf_thaw(struct lov_object *lov)
1156 CDEBUG(D_INODE, "To release share lov(%p) owner %p/%p\n",
1157 lov, lov->lo_owner, current);
1158 if (lov->lo_owner != current)
1159 up_read(&lov->lo_type_guard);
1162 #define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
1164 struct lov_object *__obj = (obj); \
1165 int __lock = !!(lock); \
1166 typeof(lov_dispatch[0].op(__VA_ARGS__)) __result; \
1169 lov_conf_freeze(__obj); \
1170 __result = LOV_2DISPATCH_NOLOCK(obj, op, __VA_ARGS__); \
1172 lov_conf_thaw(__obj); \
1177 * Performs a locked double-dispatch based on the layout type of an object.
1179 #define LOV_2DISPATCH(obj, op, ...) \
1180 LOV_2DISPATCH_MAYLOCK(obj, op, 1, __VA_ARGS__)
1182 #define LOV_2DISPATCH_VOID(obj, op, ...) \
1184 struct lov_object *__obj = (obj); \
1185 enum lov_layout_type __llt; \
1187 lov_conf_freeze(__obj); \
1188 __llt = __obj->lo_type; \
1189 LASSERT(__llt < ARRAY_SIZE(lov_dispatch)); \
1190 lov_dispatch[__llt].op(__VA_ARGS__); \
1191 lov_conf_thaw(__obj); \
1194 static void lov_conf_lock(struct lov_object *lov)
1196 LASSERT(lov->lo_owner != current);
1197 down_write(&lov->lo_type_guard);
1198 LASSERT(lov->lo_owner == NULL);
1199 lov->lo_owner = current;
1200 CDEBUG(D_INODE, "Took exclusive lov(%p) owner %p\n",
1201 lov, lov->lo_owner);
1204 static void lov_conf_unlock(struct lov_object *lov)
1206 CDEBUG(D_INODE, "To release exclusive lov(%p) owner %p\n",
1207 lov, lov->lo_owner);
1208 lov->lo_owner = NULL;
1209 up_write(&lov->lo_type_guard);
1212 static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
1216 while (atomic_read(&lov->lo_active_ios) > 0) {
1217 CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
1218 PFID(lu_object_fid(lov2lu(lov))),
1219 atomic_read(&lov->lo_active_ios));
1221 wait_event_idle(lov->lo_waitq,
1222 atomic_read(&lov->lo_active_ios) == 0);
1227 static int lov_layout_change(const struct lu_env *unused,
1228 struct lov_object *lov, struct lov_stripe_md *lsm,
1229 const struct cl_object_conf *conf)
1231 enum lov_layout_type llt = lov_type(lsm);
1232 union lov_layout_state *state = &lov->u;
1233 const struct lov_layout_operations *old_ops;
1234 const struct lov_layout_operations *new_ops;
1235 struct lov_device *lov_dev = lov_object_dev(lov);
1241 LASSERT(lov->lo_type < ARRAY_SIZE(lov_dispatch));
1243 env = cl_env_get(&refcheck);
1245 RETURN(PTR_ERR(env));
1247 LASSERT(llt < ARRAY_SIZE(lov_dispatch));
1249 CDEBUG(D_INODE, DFID" from %s to %s\n",
1250 PFID(lu_object_fid(lov2lu(lov))),
1251 llt2str(lov->lo_type), llt2str(llt));
1253 old_ops = &lov_dispatch[lov->lo_type];
1254 new_ops = &lov_dispatch[llt];
1256 rc = cl_object_prune(env, &lov->lo_cl);
1260 rc = old_ops->llo_delete(env, lov, &lov->u);
1264 old_ops->llo_fini(env, lov, &lov->u);
1266 LASSERT(atomic_read(&lov->lo_active_ios) == 0);
1268 CDEBUG(D_INODE, DFID "Apply new layout lov %p, type %d\n",
1269 PFID(lu_object_fid(lov2lu(lov))), lov, llt);
1271 /* page bufsize fixup */
1272 cl_object_header(&lov->lo_cl)->coh_page_bufsize -=
1273 lov_page_slice_fixup(lov, NULL);
1276 rc = new_ops->llo_init(env, lov_dev, lov, lsm, conf, state);
1278 struct obd_device *obd = lov2obd(lov_dev->ld_lov);
1280 CERROR("%s: cannot apply new layout on "DFID" : rc = %d\n",
1281 obd->obd_name, PFID(lu_object_fid(lov2lu(lov))), rc);
1282 new_ops->llo_delete(env, lov, state);
1283 new_ops->llo_fini(env, lov, state);
1284 /* this file becomes an EMPTY file. */
1285 lov->lo_type = LLT_EMPTY;
1290 cl_env_put(env, &refcheck);
1294 /*****************************************************************************
1296 * Lov object operations.
1299 int lov_object_init(const struct lu_env *env, struct lu_object *obj,
1300 const struct lu_object_conf *conf)
1302 struct lov_object *lov = lu2lov(obj);
1303 struct lov_device *dev = lov_object_dev(lov);
1304 const struct cl_object_conf *cconf = lu2cl_conf(conf);
1305 union lov_layout_state *set = &lov->u;
1306 const struct lov_layout_operations *ops;
1307 struct lov_stripe_md *lsm = NULL;
1311 init_rwsem(&lov->lo_type_guard);
1312 atomic_set(&lov->lo_active_ios, 0);
1313 init_waitqueue_head(&lov->lo_waitq);
1314 cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
1316 lov->lo_type = LLT_EMPTY;
1317 if (cconf->u.coc_layout.lb_buf != NULL) {
1318 lsm = lov_unpackmd(dev->ld_lov,
1319 cconf->u.coc_layout.lb_buf,
1320 cconf->u.coc_layout.lb_len);
1322 RETURN(PTR_ERR(lsm));
1324 dump_lsm(D_INODE, lsm);
1327 /* no locking is necessary, as object is being created */
1328 lov->lo_type = lov_type(lsm);
1329 ops = &lov_dispatch[lov->lo_type];
1330 rc = ops->llo_init(env, dev, lov, lsm, cconf, set);
1340 static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
1341 const struct cl_object_conf *conf)
1343 struct lov_stripe_md *lsm = NULL;
1344 struct lov_object *lov = cl2lov(obj);
1348 if (conf->coc_opc == OBJECT_CONF_SET &&
1349 conf->u.coc_layout.lb_buf != NULL) {
1350 lsm = lov_unpackmd(lov_object_dev(lov)->ld_lov,
1351 conf->u.coc_layout.lb_buf,
1352 conf->u.coc_layout.lb_len);
1354 RETURN(PTR_ERR(lsm));
1355 dump_lsm(D_INODE, lsm);
1359 if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
1360 lov->lo_layout_invalid = true;
1361 GOTO(out, result = 0);
1364 if (conf->coc_opc == OBJECT_CONF_WAIT) {
1365 if (lov->lo_layout_invalid &&
1366 atomic_read(&lov->lo_active_ios) > 0) {
1367 lov_conf_unlock(lov);
1368 result = lov_layout_wait(env, lov);
1374 LASSERT(conf->coc_opc == OBJECT_CONF_SET);
1376 if ((lsm == NULL && lov->lo_lsm == NULL) ||
1377 ((lsm != NULL && lov->lo_lsm != NULL) &&
1378 (lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen) &&
1379 (lov->lo_lsm->lsm_entries[0]->lsme_pattern ==
1380 lsm->lsm_entries[0]->lsme_pattern))) {
1381 /* same version of layout */
1382 lov->lo_layout_invalid = false;
1383 GOTO(out, result = 0);
1386 /* will change layout - check if there still exists active IO. */
1387 if (atomic_read(&lov->lo_active_ios) > 0) {
1388 lov->lo_layout_invalid = true;
1389 GOTO(out, result = -EBUSY);
1392 result = lov_layout_change(env, lov, lsm, conf);
1393 lov->lo_layout_invalid = result != 0;
1397 lov_conf_unlock(lov);
1399 CDEBUG(D_INODE, DFID" lo_layout_invalid=%d\n",
1400 PFID(lu_object_fid(lov2lu(lov))), lov->lo_layout_invalid);
1404 static void lov_object_delete(const struct lu_env *env, struct lu_object *obj)
1406 struct lov_object *lov = lu2lov(obj);
1409 LOV_2DISPATCH_VOID(lov, llo_delete, env, lov, &lov->u);
1413 static void lov_object_free(const struct lu_env *env, struct lu_object *obj)
1415 struct lov_object *lov = lu2lov(obj);
1418 LOV_2DISPATCH_VOID(lov, llo_fini, env, lov, &lov->u);
1419 lu_object_fini(obj);
1420 OBD_SLAB_FREE_PTR(lov, lov_object_kmem);
1424 static int lov_object_print(const struct lu_env *env, void *cookie,
1425 lu_printer_t p, const struct lu_object *o)
1427 return LOV_2DISPATCH_NOLOCK(lu2lov(o), llo_print, env, cookie, p, o);
1430 int lov_page_init(const struct lu_env *env, struct cl_object *obj,
1431 struct cl_page *page, pgoff_t index)
1433 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_page_init, env, obj, page,
1438 * Implements cl_object_operations::clo_io_init() method for lov
1439 * layer. Dispatches to the appropriate layout io initialization method.
1441 int lov_io_init(const struct lu_env *env, struct cl_object *obj,
1444 CL_IO_SLICE_CLEAN(lov_env_io(env), lis_preserved);
1446 CDEBUG(D_INODE, DFID "io %p type %d ignore/verify layout %d/%d\n",
1447 PFID(lu_object_fid(&obj->co_lu)), io, io->ci_type,
1448 io->ci_ignore_layout, io->ci_verify_layout);
1450 /* IO type CIT_MISC with ci_ignore_layout set are usually invoked from
1451 * the OSC layer. It shouldn't take lov layout conf lock in that case,
1452 * because as long as the OSC object exists, the layout can't be
1454 return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
1455 !(io->ci_ignore_layout && io->ci_type == CIT_MISC),
1460 * An implementation of cl_object_operations::clo_attr_get() method for lov
1461 * layer. For raid0 layout this collects and merges attributes of all
1464 static int lov_attr_get(const struct lu_env *env, struct cl_object *obj,
1465 struct cl_attr *attr)
1467 /* do not take lock, as this function is called under a
1468 * spin-lock. Layout is protected from changing by ongoing IO. */
1469 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_getattr, env, obj, attr);
1472 static int lov_attr_update(const struct lu_env *env, struct cl_object *obj,
1473 const struct cl_attr *attr, unsigned valid)
1476 * No dispatch is required here, as no layout implements this.
1481 int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
1482 struct cl_lock *lock, const struct cl_io *io)
1484 /* No need to lock because we've taken one refcount of layout. */
1485 return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
1490 * We calculate on which OST the mapping will end. If the length of mapping
1491 * is greater than (stripe_size * stripe_count) then the last_stripe will
1492 * will be one just before start_stripe. Else we check if the mapping
1493 * intersects each OST and find last_stripe.
1494 * This function returns the last_stripe and also sets the stripe_count
1495 * over which the mapping is spread
1497 * \param lsm [in] striping information for the file
1498 * \param index [in] stripe component index
1499 * \param ext [in] logical extent of mapping
1500 * \param start_stripe [in] starting stripe of the mapping
1501 * \param stripe_count [out] the number of stripes across which to map is
1504 * \retval last_stripe return the last stripe of the mapping
1506 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, int index,
1507 struct lu_extent *ext,
1508 int start_stripe, int *stripe_count)
1510 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1516 if (ext->e_end - ext->e_start >
1517 lsme->lsme_stripe_size * lsme->lsme_stripe_count) {
1518 last_stripe = (start_stripe < 1 ? lsme->lsme_stripe_count - 1 :
1520 *stripe_count = lsme->lsme_stripe_count;
1522 for (j = 0, i = start_stripe; j < lsme->lsme_stripe_count;
1523 i = (i + 1) % lsme->lsme_stripe_count, j++) {
1524 if ((lov_stripe_intersects(lsm, index, i, ext,
1525 &obd_start, &obd_end)) == 0)
1529 last_stripe = (start_stripe + j - 1) % lsme->lsme_stripe_count;
1536 * Set fe_device and copy extents from local buffer into main return buffer.
1538 * \param fiemap [out] fiemap to hold all extents
1539 * \param lcl_fm_ext [in] array of fiemap extents get from OSC layer
1540 * \param ost_index [in] OST index to be written into the fm_device
1541 * field for each extent
1542 * \param ext_count [in] number of extents to be copied
1543 * \param current_extent [in] where to start copying in the extent array
1545 static void fiemap_prepare_and_copy_exts(struct fiemap *fiemap,
1546 struct fiemap_extent *lcl_fm_ext,
1547 int ost_index, unsigned int ext_count,
1553 for (ext = 0; ext < ext_count; ext++) {
1554 lcl_fm_ext[ext].fe_device = ost_index;
1555 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1558 /* Copy fm_extent's from fm_local to return buffer */
1559 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1560 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct fiemap_extent));
1563 #define FIEMAP_BUFFER_SIZE 4096
1566 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1567 * call. The local end offset and the device are sent in the first
1568 * fm_extent. This function calculates the stripe number from the index.
1569 * This function returns a stripe_no on which mapping is to be restarted.
1571 * This function returns fm_end_offset which is the in-OST offset at which
1572 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1573 * will re-calculate proper offset in next stripe.
1574 * Note that the first extent is passed to lov_get_info via the value field.
1576 * \param fiemap [in] fiemap request header
1577 * \param lsm [in] striping information for the file
1578 * \param index [in] stripe component index
1579 * \param ext [in] logical extent of mapping
1580 * \param start_stripe [out] starting stripe will be returned in this
1582 static u64 fiemap_calc_fm_end_offset(struct fiemap *fiemap,
1583 struct lov_stripe_md *lsm,
1584 int index, struct lu_extent *ext,
1587 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1588 u64 local_end = fiemap->fm_extents[0].fe_logical;
1595 if (fiemap->fm_extent_count == 0 ||
1596 fiemap->fm_extents[0].fe_logical == 0)
1599 /* Find out stripe_no from ost_index saved in the fe_device */
1600 for (i = 0; i < lsme->lsme_stripe_count; i++) {
1601 struct lov_oinfo *oinfo = lsme->lsme_oinfo[i];
1603 if (lov_oinfo_is_dummy(oinfo))
1606 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1612 if (stripe_no == -1)
1615 /* If we have finished mapping on previous device, shift logical
1616 * offset to start of next device */
1617 if (lov_stripe_intersects(lsm, index, stripe_no, ext,
1618 &lun_start, &lun_end) != 0 &&
1619 local_end < lun_end) {
1620 fm_end_offset = local_end;
1621 *start_stripe = stripe_no;
1623 /* This is a special value to indicate that caller should
1624 * calculate offset in next stripe. */
1626 *start_stripe = (stripe_no + 1) % lsme->lsme_stripe_count;
1629 return fm_end_offset;
1632 struct fiemap_state {
1633 struct fiemap *fs_fm;
1634 struct lu_extent fs_ext;
1639 int fs_start_stripe;
1641 bool fs_device_done;
1642 bool fs_finish_stripe;
1646 static struct cl_object *lov_find_subobj(const struct lu_env *env,
1647 struct lov_object *lov,
1648 struct lov_stripe_md *lsm,
1651 struct lov_device *dev = lu2lov_dev(lov2lu(lov)->lo_dev);
1652 struct lov_thread_info *lti = lov_env_info(env);
1653 struct lu_fid *ofid = <i->lti_fid;
1654 struct lov_oinfo *oinfo;
1655 struct cl_device *subdev;
1656 int entry = lov_comp_entry(index);
1657 int stripe = lov_comp_stripe(index);
1660 struct cl_object *result;
1662 if (lov->lo_type != LLT_COMP)
1663 GOTO(out, result = NULL);
1665 if (entry >= lsm->lsm_entry_count ||
1666 stripe >= lsm->lsm_entries[entry]->lsme_stripe_count)
1667 GOTO(out, result = NULL);
1669 oinfo = lsm->lsm_entries[entry]->lsme_oinfo[stripe];
1670 ost_idx = oinfo->loi_ost_idx;
1671 rc = ostid_to_fid(ofid, &oinfo->loi_oi, ost_idx);
1673 GOTO(out, result = NULL);
1675 subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
1676 result = lov_sub_find(env, subdev, ofid, NULL);
1679 result = ERR_PTR(-EINVAL);
1683 int fiemap_for_stripe(const struct lu_env *env, struct cl_object *obj,
1684 struct lov_stripe_md *lsm, struct fiemap *fiemap,
1685 size_t *buflen, struct ll_fiemap_info_key *fmkey,
1686 int index, int stripeno, struct fiemap_state *fs)
1688 struct lov_stripe_md_entry *lsme = lsm->lsm_entries[index];
1689 struct cl_object *subobj;
1690 struct lov_obd *lov = lu2lov_dev(obj->co_lu.lo_dev)->ld_lov;
1691 struct fiemap_extent *fm_ext = &fs->fs_fm->fm_extents[0];
1692 u64 req_fm_len; /* Stores length of required mapping */
1693 u64 len_mapped_single_call;
1697 unsigned int ext_count;
1698 /* EOF for object */
1699 bool ost_eof = false;
1700 /* done with required mapping for this OST? */
1701 bool ost_done = false;
1705 fs->fs_device_done = false;
1706 /* Find out range of mapping on this stripe */
1707 if ((lov_stripe_intersects(lsm, index, stripeno, &fs->fs_ext,
1708 &lun_start, &obd_object_end)) == 0)
1711 if (lov_oinfo_is_dummy(lsme->lsme_oinfo[stripeno]))
1714 /* If this is a continuation FIEMAP call and we are on
1715 * starting stripe then lun_start needs to be set to
1717 if (fs->fs_end_offset != 0 && stripeno == fs->fs_start_stripe)
1718 lun_start = fs->fs_end_offset;
1719 lun_end = lov_size_to_stripe(lsm, index, fs->fs_ext.e_end, stripeno);
1720 if (lun_start == lun_end)
1723 req_fm_len = obd_object_end - lun_start + 1;
1724 fs->fs_fm->fm_length = 0;
1725 len_mapped_single_call = 0;
1727 /* find lobsub object */
1728 subobj = lov_find_subobj(env, cl2lov(obj), lsm,
1729 lov_comp_index(index, stripeno));
1731 return PTR_ERR(subobj);
1732 /* If the output buffer is very large and the objects have many
1733 * extents we may need to loop on a single OST repeatedly */
1735 if (fiemap->fm_extent_count > 0) {
1736 /* Don't get too many extents. */
1737 if (fs->fs_cur_extent + fs->fs_cnt_need >
1738 fiemap->fm_extent_count)
1739 fs->fs_cnt_need = fiemap->fm_extent_count -
1743 lun_start += len_mapped_single_call;
1744 fs->fs_fm->fm_length = req_fm_len - len_mapped_single_call;
1745 req_fm_len = fs->fs_fm->fm_length;
1747 * If we've collected enough extent map, we'd request 1 more,
1748 * to see whether we coincidentally finished all available
1749 * extent map, so that FIEMAP_EXTENT_LAST would be set.
1751 fs->fs_fm->fm_extent_count = fs->fs_enough ?
1752 1 : fs->fs_cnt_need;
1753 fs->fs_fm->fm_mapped_extents = 0;
1754 fs->fs_fm->fm_flags = fiemap->fm_flags;
1756 ost_index = lsme->lsme_oinfo[stripeno]->loi_ost_idx;
1758 if (ost_index < 0 || ost_index >= lov->desc.ld_tgt_count)
1759 GOTO(obj_put, rc = -EINVAL);
1760 /* If OST is inactive, return extent with UNKNOWN flag. */
1761 if (!lov->lov_tgts[ost_index]->ltd_active) {
1762 fs->fs_fm->fm_flags |= FIEMAP_EXTENT_LAST;
1763 fs->fs_fm->fm_mapped_extents = 1;
1765 fm_ext[0].fe_logical = lun_start;
1766 fm_ext[0].fe_length = obd_object_end - lun_start + 1;
1767 fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1772 fs->fs_fm->fm_start = lun_start;
1773 fs->fs_fm->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1774 memcpy(&fmkey->lfik_fiemap, fs->fs_fm, sizeof(*fs->fs_fm));
1775 *buflen = fiemap_count_to_size(fs->fs_fm->fm_extent_count);
1777 rc = cl_object_fiemap(env, subobj, fmkey, fs->fs_fm, buflen);
1781 ext_count = fs->fs_fm->fm_mapped_extents;
1782 if (ext_count == 0) {
1784 fs->fs_device_done = true;
1785 /* If last stripe has hold at the end,
1786 * we need to return */
1787 if (stripeno == fs->fs_last_stripe) {
1788 fiemap->fm_mapped_extents = 0;
1789 fs->fs_finish_stripe = true;
1793 } else if (fs->fs_enough) {
1795 * We've collected enough extents and there are
1796 * more extents after it.
1801 /* If we just need num of extents, got to next device */
1802 if (fiemap->fm_extent_count == 0) {
1803 fs->fs_cur_extent += ext_count;
1807 /* prepare to copy retrived map extents */
1808 len_mapped_single_call = fm_ext[ext_count - 1].fe_logical +
1809 fm_ext[ext_count - 1].fe_length -
1812 /* Have we finished mapping on this device? */
1813 if (req_fm_len <= len_mapped_single_call) {
1815 fs->fs_device_done = true;
1818 /* Clear the EXTENT_LAST flag which can be present on
1819 * the last extent */
1820 if (fm_ext[ext_count - 1].fe_flags & FIEMAP_EXTENT_LAST)
1821 fm_ext[ext_count - 1].fe_flags &= ~FIEMAP_EXTENT_LAST;
1822 if (lov_stripe_size(lsm, index,
1823 fm_ext[ext_count - 1].fe_logical +
1824 fm_ext[ext_count - 1].fe_length,
1825 stripeno) >= fmkey->lfik_oa.o_size) {
1827 fs->fs_device_done = true;
1830 fiemap_prepare_and_copy_exts(fiemap, fm_ext, ost_index,
1831 ext_count, fs->fs_cur_extent);
1832 fs->fs_cur_extent += ext_count;
1834 /* Ran out of available extents? */
1835 if (fs->fs_cur_extent >= fiemap->fm_extent_count)
1836 fs->fs_enough = true;
1837 } while (!ost_done && !ost_eof);
1839 if (stripeno == fs->fs_last_stripe)
1840 fs->fs_finish_stripe = true;
1842 cl_object_put(env, subobj);
1848 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1849 * This also handles the restarting of FIEMAP calls in case mapping overflows
1850 * the available number of extents in single call.
1852 * \param env [in] lustre environment
1853 * \param obj [in] file object
1854 * \param fmkey [in] fiemap request header and other info
1855 * \param fiemap [out] fiemap buffer holding retrived map extents
1856 * \param buflen [in/out] max buffer length of @fiemap, when iterate
1857 * each OST, it is used to limit max map needed
1861 static int lov_object_fiemap(const struct lu_env *env, struct cl_object *obj,
1862 struct ll_fiemap_info_key *fmkey,
1863 struct fiemap *fiemap, size_t *buflen)
1865 struct lov_stripe_md_entry *lsme;
1866 struct lov_stripe_md *lsm;
1867 struct fiemap *fm_local = NULL;
1875 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1877 struct fiemap_state fs = { 0 };
1880 lsm = lov_lsm_addref(cl2lov(obj));
1882 /* no extent: there is no object for mapping */
1883 fiemap->fm_mapped_extents = 0;
1887 if (!(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1889 * If the entry count > 1 or stripe_count > 1 and the
1890 * application does not understand DEVICE_ORDER flag,
1891 * it cannot interpret the extents correctly.
1893 if (lsm->lsm_entry_count > 1 ||
1894 (lsm->lsm_entry_count == 1 &&
1895 lsm->lsm_entries[0]->lsme_stripe_count > 1))
1896 GOTO(out_lsm, rc = -ENOTSUPP);
1899 /* No support for DOM layout yet. */
1900 if (lsme_is_dom(lsm->lsm_entries[0]))
1901 GOTO(out_lsm, rc = -ENOTSUPP);
1903 if (lsm->lsm_is_released) {
1904 if (fiemap->fm_start < fmkey->lfik_oa.o_size) {
1906 * released file, return a minimal FIEMAP if
1907 * request fits in file-size.
1909 fiemap->fm_mapped_extents = 1;
1910 fiemap->fm_extents[0].fe_logical = fiemap->fm_start;
1911 if (fiemap->fm_start + fiemap->fm_length <
1912 fmkey->lfik_oa.o_size)
1913 fiemap->fm_extents[0].fe_length =
1916 fiemap->fm_extents[0].fe_length =
1917 fmkey->lfik_oa.o_size -
1919 fiemap->fm_extents[0].fe_flags |=
1920 FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_LAST;
1922 GOTO(out_lsm, rc = 0);
1925 /* buffer_size is small to hold fm_extent_count of extents. */
1926 if (fiemap_count_to_size(fiemap->fm_extent_count) < buffer_size)
1927 buffer_size = fiemap_count_to_size(fiemap->fm_extent_count);
1929 OBD_ALLOC_LARGE(fm_local, buffer_size);
1930 if (fm_local == NULL)
1931 GOTO(out_lsm, rc = -ENOMEM);
1934 * Requested extent count exceeds the fiemap buffer size, shrink our
1937 if (fiemap_count_to_size(fiemap->fm_extent_count) > *buflen)
1938 fiemap->fm_extent_count = fiemap_size_to_count(*buflen);
1939 if (fiemap->fm_extent_count == 0)
1942 fs.fs_enough = false;
1943 fs.fs_cur_extent = 0;
1944 fs.fs_fm = fm_local;
1945 fs.fs_cnt_need = fiemap_size_to_count(buffer_size);
1947 whole_start = fiemap->fm_start;
1948 /* whole_start is beyond the end of the file */
1949 if (whole_start > fmkey->lfik_oa.o_size)
1950 GOTO(out_fm_local, rc = -EINVAL);
1951 whole_end = (fiemap->fm_length == OBD_OBJECT_EOF) ?
1952 fmkey->lfik_oa.o_size :
1953 whole_start + fiemap->fm_length - 1;
1955 * If fiemap->fm_length != OBD_OBJECT_EOF but whole_end exceeds file
1958 if (whole_end > fmkey->lfik_oa.o_size)
1959 whole_end = fmkey->lfik_oa.o_size;
1961 start_entry = lov_lsm_entry(lsm, whole_start);
1962 end_entry = lov_lsm_entry(lsm, whole_end);
1963 if (end_entry == -1)
1964 end_entry = lsm->lsm_entry_count - 1;
1966 if (start_entry == -1 || end_entry == -1)
1967 GOTO(out_fm_local, rc = -EINVAL);
1969 /* TODO: rewrite it with lov_foreach_io_layout() */
1970 for (entry = start_entry; entry <= end_entry; entry++) {
1971 lsme = lsm->lsm_entries[entry];
1973 if (!lsme_inited(lsme))
1976 if (entry == start_entry)
1977 fs.fs_ext.e_start = whole_start;
1979 fs.fs_ext.e_start = lsme->lsme_extent.e_start;
1980 if (entry == end_entry)
1981 fs.fs_ext.e_end = whole_end;
1983 fs.fs_ext.e_end = lsme->lsme_extent.e_end - 1;
1984 fs.fs_length = fs.fs_ext.e_end - fs.fs_ext.e_start + 1;
1986 /* Calculate start stripe, last stripe and length of mapping */
1987 fs.fs_start_stripe = lov_stripe_number(lsm, entry,
1989 fs.fs_last_stripe = fiemap_calc_last_stripe(lsm, entry,
1990 &fs.fs_ext, fs.fs_start_stripe,
1992 fs.fs_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, entry,
1993 &fs.fs_ext, &fs.fs_start_stripe);
1994 /* Check each stripe */
1995 for (cur_stripe = fs.fs_start_stripe; stripe_count > 0;
1997 cur_stripe = (cur_stripe + 1) % lsme->lsme_stripe_count) {
1998 rc = fiemap_for_stripe(env, obj, lsm, fiemap, buflen,
1999 fmkey, entry, cur_stripe, &fs);
2001 GOTO(out_fm_local, rc);
2004 if (fs.fs_finish_stripe)
2006 } /* for each stripe */
2007 } /* for covering layout component */
2009 * We've traversed all components, set @entry to the last component
2010 * entry, it's for the last stripe check.
2014 /* Indicate that we are returning device offsets unless file just has
2016 if (lsm->lsm_entry_count > 1 ||
2017 (lsm->lsm_entry_count == 1 &&
2018 lsm->lsm_entries[0]->lsme_stripe_count > 1))
2019 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2021 if (fiemap->fm_extent_count == 0)
2022 goto skip_last_device_calc;
2024 /* Check if we have reached the last stripe and whether mapping for that
2025 * stripe is done. */
2026 if ((cur_stripe == fs.fs_last_stripe) && fs.fs_device_done)
2027 fiemap->fm_extents[fs.fs_cur_extent - 1].fe_flags |=
2029 skip_last_device_calc:
2030 fiemap->fm_mapped_extents = fs.fs_cur_extent;
2032 OBD_FREE_LARGE(fm_local, buffer_size);
2039 static int lov_object_getstripe(const struct lu_env *env, struct cl_object *obj,
2040 struct lov_user_md __user *lum, size_t size)
2042 struct lov_object *lov = cl2lov(obj);
2043 struct lov_stripe_md *lsm;
2047 lsm = lov_lsm_addref(lov);
2051 rc = lov_getstripe(env, cl2lov(obj), lsm, lum, size);
2056 static int lov_object_layout_get(const struct lu_env *env,
2057 struct cl_object *obj,
2058 struct cl_layout *cl)
2060 struct lov_object *lov = cl2lov(obj);
2061 struct lov_stripe_md *lsm = lov_lsm_addref(lov);
2062 struct lu_buf *buf = &cl->cl_buf;
2068 cl->cl_layout_gen = CL_LAYOUT_GEN_EMPTY;
2073 cl->cl_size = lov_comp_md_size(lsm);
2074 cl->cl_layout_gen = lsm->lsm_layout_gen;
2075 cl->cl_is_released = lsm->lsm_is_released;
2076 cl->cl_is_composite = lsm_is_composite(lsm->lsm_magic);
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_MAYLOCK(cl2lov(obj), llo_flush, true, env, obj,
2107 static const struct cl_object_operations lov_ops = {
2108 .coo_page_init = lov_page_init,
2109 .coo_lock_init = lov_lock_init,
2110 .coo_io_init = lov_io_init,
2111 .coo_attr_get = lov_attr_get,
2112 .coo_attr_update = lov_attr_update,
2113 .coo_conf_set = lov_conf_set,
2114 .coo_getstripe = lov_object_getstripe,
2115 .coo_layout_get = lov_object_layout_get,
2116 .coo_maxbytes = lov_object_maxbytes,
2117 .coo_fiemap = lov_object_fiemap,
2118 .coo_object_flush = lov_object_flush
2121 static const struct lu_object_operations lov_lu_obj_ops = {
2122 .loo_object_init = lov_object_init,
2123 .loo_object_delete = lov_object_delete,
2124 .loo_object_release = NULL,
2125 .loo_object_free = lov_object_free,
2126 .loo_object_print = lov_object_print,
2127 .loo_object_invariant = NULL
2130 struct lu_object *lov_object_alloc(const struct lu_env *env,
2131 const struct lu_object_header *unused,
2132 struct lu_device *dev)
2134 struct lov_object *lov;
2135 struct lu_object *obj;
2138 OBD_SLAB_ALLOC_PTR_GFP(lov, lov_object_kmem, GFP_NOFS);
2141 lu_object_init(obj, NULL, dev);
2142 lov->lo_cl.co_ops = &lov_ops;
2143 lov->lo_type = -1; /* invalid, to catch uninitialized type */
2145 * object io operation vector (cl_object::co_iop) is installed
2146 * later in lov_object_init(), as different vectors are used
2147 * for object with different layouts.
2149 obj->lo_ops = &lov_lu_obj_ops;
2155 struct lov_stripe_md *lov_lsm_addref(struct lov_object *lov)
2157 struct lov_stripe_md *lsm = NULL;
2159 lov_conf_freeze(lov);
2160 if (lov->lo_lsm != NULL) {
2161 lsm = lsm_addref(lov->lo_lsm);
2162 CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
2163 lsm, atomic_read(&lsm->lsm_refc),
2164 lov->lo_layout_invalid, current);
2170 int lov_read_and_clear_async_rc(struct cl_object *clob)
2172 struct lu_object *luobj;
2176 luobj = lu_object_locate(&cl_object_header(clob)->coh_lu,
2178 if (luobj != NULL) {
2179 struct lov_object *lov = lu2lov(luobj);
2181 lov_conf_freeze(lov);
2182 switch (lov->lo_type) {
2184 struct lov_stripe_md *lsm;
2188 LASSERT(lsm != NULL);
2189 for (i = 0; i < lsm->lsm_entry_count; i++) {
2190 struct lov_stripe_md_entry *lse =
2191 lsm->lsm_entries[i];
2194 if (!lsme_inited(lse))
2197 for (j = 0; j < lse->lsme_stripe_count; j++) {
2198 struct lov_oinfo *loi =
2201 if (lov_oinfo_is_dummy(loi))
2204 if (loi->loi_ar.ar_rc && !rc)
2205 rc = loi->loi_ar.ar_rc;
2206 loi->loi_ar.ar_rc = 0;
2222 EXPORT_SYMBOL(lov_read_and_clear_async_rc);