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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Author: Nikita Danilov <nikita@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <linux/module.h>
44 * struct OBD_{ALLOC,FREE}*()
48 #include <obd_support.h>
51 #include <lustre_fid.h>
52 #include <dt_object.h>
55 /* osd_lookup(), struct osd_thread_info */
56 #include "osd_internal.h"
57 #include "osd_scrub.h"
59 static unsigned int osd_oi_count = OSD_OI_FID_NR;
60 module_param(osd_oi_count, int, 0444);
61 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
63 /** to serialize concurrent OI index initialization */
64 static struct mutex oi_init_lock;
66 static struct dt_index_features oi_feat = {
67 .dif_flags = DT_IND_UPDATE,
68 .dif_recsize_min = sizeof(struct osd_inode_id),
69 .dif_recsize_max = sizeof(struct osd_inode_id),
73 #define OSD_OI_NAME_BASE "oi.16"
75 static void osd_oi_table_put(struct osd_thread_info *info,
76 struct osd_oi **oi_table, unsigned oi_count)
78 struct iam_container *bag;
81 for (i = 0; i < oi_count; i++) {
82 if (oi_table[i] == NULL)
85 LASSERT(oi_table[i]->oi_inode != NULL);
87 bag = &(oi_table[i]->oi_dir.od_container);
88 if (bag->ic_object == oi_table[i]->oi_inode)
89 iam_container_fini(bag);
90 iput(oi_table[i]->oi_inode);
91 oi_table[i]->oi_inode = NULL;
92 OBD_FREE_PTR(oi_table[i]);
97 static int osd_oi_index_create_one(struct osd_thread_info *info,
98 struct osd_device *osd, const char *name,
99 struct dt_index_features *feat)
101 const struct lu_env *env = info->oti_env;
102 struct osd_inode_id *id = &info->oti_id;
103 struct buffer_head *bh;
105 struct ldiskfs_dir_entry_2 *de;
106 struct dentry *dentry;
107 struct super_block *sb = osd_sb(osd);
108 struct inode *dir = sb->s_root->d_inode;
112 dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
113 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
115 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
117 inode = osd_iget(info, osd, id);
118 if (!IS_ERR(inode)) {
120 inode = ERR_PTR(-EEXIST);
122 return PTR_ERR(inode);
125 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
129 inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR));
131 ldiskfs_journal_stop(jh);
132 return PTR_ERR(inode);
135 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
136 unlock_new_inode(inode);
138 if (feat->dif_flags & DT_IND_VARKEY)
139 rc = iam_lvar_create(inode, feat->dif_keysize_max,
140 feat->dif_ptrsize, feat->dif_recsize_max,
143 rc = iam_lfix_create(inode, feat->dif_keysize_max,
144 feat->dif_ptrsize, feat->dif_recsize_max,
146 dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
147 rc = osd_ldiskfs_add_entry(info, osd, jh, dentry, inode, NULL);
148 ldiskfs_journal_stop(jh);
153 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
154 struct osd_device *osd,
156 struct dt_index_features *f,
159 struct dentry *dentry;
163 dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
165 return (void *) dentry;
167 if (dentry->d_inode) {
168 LASSERT(!is_bad_inode(dentry->d_inode));
169 inode = dentry->d_inode;
170 atomic_inc(&inode->i_count);
177 shrink_dcache_parent(osd_sb(osd)->s_root);
179 return ERR_PTR(-ENOENT);
181 rc = osd_oi_index_create_one(info, osd, name, f);
185 dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
187 return (void *) dentry;
189 if (dentry->d_inode) {
190 LASSERT(!is_bad_inode(dentry->d_inode));
191 inode = dentry->d_inode;
192 atomic_inc(&inode->i_count);
197 return ERR_PTR(-ENOENT);
201 * Open an OI(Ojbect Index) container.
203 * \param name Name of OI container
204 * \param objp Pointer of returned OI
207 * \retval -ve failure
209 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
210 char *name, struct osd_oi **oi_slot, bool create)
212 struct osd_directory *dir;
213 struct iam_container *bag;
220 oi_feat.dif_keysize_min = sizeof(struct lu_fid);
221 oi_feat.dif_keysize_max = sizeof(struct lu_fid);
223 inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
225 RETURN(PTR_ERR(inode));
227 /* 'What the @fid is' is not imporatant, because these objects
228 * have no OI mappings, and only are visible inside the OSD.*/
229 lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
230 rc = osd_ea_fid_set(info, inode, &info->oti_fid, LMAC_NOT_IN_OI, 0);
236 GOTO(out_inode, rc = -ENOMEM);
238 oi->oi_inode = inode;
241 bag = &dir->od_container;
242 rc = iam_container_init(bag, &dir->od_descr, inode);
246 rc = iam_container_setup(bag);
248 GOTO(out_container, rc);
254 iam_container_fini(bag);
263 * Open OI(Object Index) table.
264 * If \a oi_count is zero, which means caller doesn't know how many OIs there
265 * will be, this function can either return 0 for new filesystem, or number
266 * of OIs on existed filesystem.
268 * If \a oi_count is non-zero, which means caller does know number of OIs on
269 * filesystem, this function should return the exactly same number on
270 * success, or error code in failure.
272 * \param oi_count Number of expected OI containers
273 * \param create Create OIs if doesn't exist
275 * \retval +ve number of opened OI containers
276 * \retval 0 no OI containers found
277 * \retval -ve failure
280 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
281 struct osd_oi **oi_table, unsigned oi_count, bool create)
283 struct scrub_file *sf = &osd->od_scrub.os_file;
289 /* NB: oi_count != 0 means that we have already created/known all OIs
290 * and have known exact number of OIs. */
291 LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
293 for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
296 if (oi_table[i] != NULL) {
301 sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
302 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
308 if (rc == -ENOENT && create == false) {
313 ldiskfs_set_bit(i, sf->sf_oi_bitmap);
317 CERROR("%s: can't open %s: rc = %d\n",
318 osd_dev2name(osd), name, rc);
320 CERROR("%s: expect to open total %d OI files.\n",
321 osd_dev2name(osd), oi_count);
326 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
333 static int osd_remove_oi_one(struct dentry *parent, const char *name,
336 struct dentry *child;
339 child = ll_lookup_one_len(name, parent, namelen);
343 rc = ll_vfs_unlink(parent->d_inode, child);
347 return rc == -ENOENT ? 0 : rc;
350 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
357 for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
358 namelen = snprintf(name, sizeof(name), "%s.%d",
359 OSD_OI_NAME_BASE, i);
360 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
362 CERROR("%s: fail to remove the stale OI file %s: "
363 "rc = %d\n", osd_dev2name(osd), name, rc);
368 namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
369 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
371 CERROR("%s: fail to remove the stale OI file %s: rc = %d\n",
372 osd_dev2name(osd), name, rc);
377 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
380 struct osd_scrub *scrub = &osd->od_scrub;
381 struct scrub_file *sf = &scrub->os_file;
387 rc = osd_remove_ois(info, osd);
392 OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
396 mutex_lock(&oi_init_lock);
397 /* try to open existing multiple OIs first */
398 rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
403 if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
406 osd_scrub_file_reset(scrub,
407 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
409 osd_oi_count = sf->sf_oi_count;
413 /* if previous failed then try found single OI from old filesystem */
414 rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
415 if (rc == 0) { /* found single OI from old filesystem */
416 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
417 if (sf->sf_success_count == 0)
418 /* XXX: There is one corner case that if the OI_scrub
419 * file crashed or lost and we regard it upgrade,
420 * then we allow IGIF lookup to bypass OI files.
422 * The risk is that osd_fid_lookup() may found
423 * a wrong inode with the given IGIF especially
424 * when the MDT has performed file-level backup
425 * and restored after former upgrading from 1.8
426 * to 2.x. Fortunately, the osd_fid_lookup()can
427 * verify the inode to decrease the risk. */
428 osd_scrub_file_reset(scrub,
429 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
432 } else if (rc != -ENOENT) {
433 CERROR("%s: can't open %s: rc = %d\n",
434 osd_dev2name(osd), OSD_OI_NAME_BASE, rc);
438 if (sf->sf_oi_count > 0) {
441 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
442 for (i = 0; i < osd_oi_count; i++)
443 ldiskfs_set_bit(i, sf->sf_oi_bitmap);
444 osd_scrub_file_reset(scrub,
445 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
448 sf->sf_oi_count = osd_oi_count;
451 rc = osd_scrub_file_store(scrub);
453 osd_oi_table_put(info, oi, sf->sf_oi_count);
457 /* No OIs exist, new filesystem, create OI objects */
458 rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
459 LASSERT(ergo(rc >= 0, rc == osd_oi_count));
465 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
467 LASSERT((rc & (rc - 1)) == 0);
468 osd->od_oi_table = oi;
469 osd->od_oi_count = rc;
470 if (sf->sf_oi_count != rc) {
471 sf->sf_oi_count = rc;
472 rc = osd_scrub_file_store(scrub);
474 osd_oi_table_put(info, oi, sf->sf_oi_count);
475 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
482 mutex_unlock(&oi_init_lock);
486 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
488 if (unlikely(osd->od_oi_table == NULL))
491 osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
493 OBD_FREE(osd->od_oi_table,
494 sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
495 osd->od_oi_table = NULL;
498 static inline int fid_is_fs_root(const struct lu_fid *fid)
500 /* Map root inode to special local object FID */
501 return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
502 fid_oid(fid) == OSD_FS_ROOT_OID));
505 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
506 struct osd_oi *oi, struct dt_rec *rec,
507 const struct dt_key *key)
509 struct iam_container *bag;
510 struct iam_iterator *it = &oti->oti_idx_it;
511 struct iam_path_descr *ipd;
516 LASSERT(oi->oi_inode);
518 bag = &oi->oi_dir.od_container;
519 ipd = osd_idx_ipd_get(oti->oti_env, bag);
523 /* got ipd now we can start iterator. */
524 iam_it_init(it, bag, 0, ipd);
526 rc = iam_it_get(it, (struct iam_key *)key);
528 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
531 osd_ipd_put(oti->oti_env, bag, ipd);
533 LINVRNT(osd_invariant(obj));
538 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
539 const struct lu_fid *fid, enum oi_check_flags flags)
541 struct lu_seq_range *range = &info->oti_seq_range;
545 if (flags & OI_KNOWN_ON_OST)
548 if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
549 fid_is_llog(fid)) || fid_is_name_llog(fid) ||
553 if (fid_is_idif(fid) || fid_is_last_id(fid))
556 if (!(flags & OI_CHECK_FLD))
559 if (osd_seq_site(osd)->ss_server_fld == NULL)
562 rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
564 /* During upgrade, OST FLDB might not be loaded because
565 * OST FLDB is not created until 2.6, so if some DNE
566 * filesystem upgrade from 2.5 to 2.7/2.8, they will
567 * not be able to find the sequence from local FLDB
568 * cache see fld_index_init(). */
569 if (rc == -ENOENT && osd->od_is_ost)
573 CERROR("%s: lookup FLD "DFID": rc = %d\n",
574 osd_name(osd), PFID(fid), rc);
578 if (fld_range_is_ost(range))
584 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
585 const struct lu_fid *fid, struct osd_inode_id *id)
587 struct lu_fid *oi_fid = &info->oti_fid2;
590 fid_cpu_to_be(oi_fid, fid);
591 rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
592 (const struct dt_key *)oi_fid);
594 osd_id_unpack(id, id);
596 } else if (rc == 0) {
602 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
603 const struct lu_fid *fid, struct osd_inode_id *id,
604 enum oi_check_flags flags)
606 if (unlikely(fid_is_last_id(fid)))
607 return osd_obj_spec_lookup(info, osd, fid, id);
609 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
610 return osd_obj_map_lookup(info, osd, fid, id);
613 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
615 if (fid_is_fs_root(fid)) {
616 osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
617 osd_sb(osd)->s_root->d_inode->i_generation);
620 if (unlikely(fid_is_acct(fid)))
621 return osd_acct_obj_lookup(info, osd, fid, id);
623 /* For other special FIDs, try OI first, then do spec lookup */
624 rc = __osd_oi_lookup(info, osd, fid, id);
626 return osd_obj_spec_lookup(info, osd, fid, id);
630 if (!osd->od_igif_inoi && fid_is_igif(fid)) {
631 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
635 return __osd_oi_lookup(info, osd, fid, id);
638 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
639 const struct dt_rec *rec, const struct dt_key *key,
640 handle_t *th, bool insert)
642 struct iam_container *bag;
643 struct iam_path_descr *ipd;
648 LASSERT(oi->oi_inode);
649 ll_vfs_dq_init(oi->oi_inode);
651 bag = &oi->oi_dir.od_container;
652 ipd = osd_idx_ipd_get(oti->oti_env, bag);
653 if (unlikely(ipd == NULL))
657 LASSERT(th->h_transaction != NULL);
659 rc = iam_insert(th, bag, (const struct iam_key *)key,
660 (const struct iam_rec *)rec, ipd);
662 rc = iam_update(th, bag, (const struct iam_key *)key,
663 (const struct iam_rec *)rec, ipd);
664 osd_ipd_put(oti->oti_env, bag, ipd);
665 LINVRNT(osd_invariant(obj));
669 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
670 const struct lu_fid *fid, const struct osd_inode_id *id,
671 handle_t *th, enum oi_check_flags flags, bool *exist)
673 struct lu_fid *oi_fid = &info->oti_fid2;
674 struct osd_inode_id *oi_id = &info->oti_id2;
677 if (unlikely(fid_is_last_id(fid)))
678 return osd_obj_spec_insert(info, osd, fid, id, th);
680 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
681 return osd_obj_map_insert(info, osd, fid, id, th);
683 fid_cpu_to_be(oi_fid, fid);
684 osd_id_pack(oi_id, id);
685 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
686 (const struct dt_rec *)oi_id,
687 (const struct dt_key *)oi_fid, th, true);
690 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
695 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
699 if (unlikely(osd_id_eq(id, oi_id)))
702 /* Check whether the mapping for oi_id is valid or not. */
703 inode = osd_iget(info, osd, oi_id);
706 if (rc == -ENOENT || rc == -ESTALE)
711 /* The EA inode should NOT be in OI, old OI scrub may added
712 * such OI mapping by wrong, replace it. */
713 if (unlikely(osd_is_ea_inode(inode))) {
718 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
719 &info->oti_ost_attrs);
727 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
728 lu_fid_eq(fid, &lma->lma_self_fid)) {
729 CERROR("%s: the FID "DFID" is used by two objects: "
730 "%u/%u %u/%u\n", osd_dev2name(osd),
731 PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
732 id->oii_ino, id->oii_gen);
737 osd_id_pack(oi_id, id);
738 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
739 (const struct dt_rec *)oi_id,
740 (const struct dt_key *)oi_fid, th, false);
748 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
749 rc = osd_obj_spec_insert(info, osd, fid, id, th);
753 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
754 const struct dt_key *key, handle_t *th)
756 struct iam_container *bag;
757 struct iam_path_descr *ipd;
762 LASSERT(oi->oi_inode);
763 ll_vfs_dq_init(oi->oi_inode);
765 bag = &oi->oi_dir.od_container;
766 ipd = osd_idx_ipd_get(oti->oti_env, bag);
767 if (unlikely(ipd == NULL))
771 LASSERT(th->h_transaction != NULL);
773 rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
774 osd_ipd_put(oti->oti_env, bag, ipd);
775 LINVRNT(osd_invariant(obj));
779 int osd_oi_delete(struct osd_thread_info *info,
780 struct osd_device *osd, const struct lu_fid *fid,
781 handle_t *th, enum oi_check_flags flags)
783 struct lu_fid *oi_fid = &info->oti_fid2;
785 /* clear idmap cache */
786 if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
787 fid_zero(&info->oti_cache.oic_fid);
789 if (fid_is_last_id(fid))
792 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
793 return osd_obj_map_delete(info, osd, fid, th);
795 fid_cpu_to_be(oi_fid, fid);
796 return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
797 (const struct dt_key *)oi_fid, th);
800 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
801 const struct lu_fid *fid, const struct osd_inode_id *id,
802 handle_t *th, enum oi_check_flags flags)
804 struct lu_fid *oi_fid = &info->oti_fid2;
805 struct osd_inode_id *oi_id = &info->oti_id2;
808 if (unlikely(fid_is_last_id(fid)))
809 return osd_obj_spec_update(info, osd, fid, id, th);
811 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
812 return osd_obj_map_update(info, osd, fid, id, th);
814 fid_cpu_to_be(oi_fid, fid);
815 osd_id_pack(oi_id, id);
816 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
817 (const struct dt_rec *)oi_id,
818 (const struct dt_key *)oi_fid, th, false);
822 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
823 rc = osd_obj_spec_update(info, osd, fid, id, th);
827 int osd_oi_mod_init(void)
829 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
830 osd_oi_count = OSD_OI_FID_NR;
832 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
833 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
834 osd_oi_count, size_roundup_power2(osd_oi_count));
835 osd_oi_count = size_roundup_power2(osd_oi_count);
838 mutex_init(&oi_init_lock);