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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 * Author: Nikita Danilov <nikita@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_OSD
45 #include <linux/module.h>
47 /* LUSTRE_VERSION_CODE */
48 #include <lustre_ver.h>
50 * struct OBD_{ALLOC,FREE}*()
54 #include <obd_support.h>
57 #include <lustre_fid.h>
58 #include <dt_object.h>
61 /* osd_lookup(), struct osd_thread_info */
62 #include "osd_internal.h"
63 #include "osd_scrub.h"
65 static unsigned int osd_oi_count = OSD_OI_FID_NR;
66 module_param(osd_oi_count, int, 0444);
67 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
69 /** to serialize concurrent OI index initialization */
70 static struct mutex oi_init_lock;
72 static struct dt_index_features oi_feat = {
73 .dif_flags = DT_IND_UPDATE,
74 .dif_recsize_min = sizeof(struct osd_inode_id),
75 .dif_recsize_max = sizeof(struct osd_inode_id),
79 #define OSD_OI_NAME_BASE "oi.16"
81 static void osd_oi_table_put(struct osd_thread_info *info,
82 struct osd_oi **oi_table, unsigned oi_count)
84 struct iam_container *bag;
87 for (i = 0; i < oi_count; i++) {
88 if (oi_table[i] == NULL)
91 LASSERT(oi_table[i]->oi_inode != NULL);
93 bag = &(oi_table[i]->oi_dir.od_container);
94 if (bag->ic_object == oi_table[i]->oi_inode)
95 iam_container_fini(bag);
96 iput(oi_table[i]->oi_inode);
97 oi_table[i]->oi_inode = NULL;
98 OBD_FREE_PTR(oi_table[i]);
103 static int osd_oi_index_create_one(struct osd_thread_info *info,
104 struct osd_device *osd, const char *name,
105 struct dt_index_features *feat)
107 const struct lu_env *env = info->oti_env;
108 struct osd_inode_id *id = &info->oti_id;
109 struct buffer_head *bh;
111 struct ldiskfs_dir_entry_2 *de;
112 struct dentry *dentry;
113 struct super_block *sb = osd_sb(osd);
114 struct inode *dir = sb->s_root->d_inode;
118 dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
119 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, NULL);
121 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
123 inode = osd_iget(info, osd, id);
124 if (!IS_ERR(inode)) {
126 inode = ERR_PTR(-EEXIST);
128 return PTR_ERR(inode);
131 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, 100);
135 inode = ldiskfs_create_inode(jh, dir, (S_IFREG | S_IRUGO | S_IWUSR));
137 ldiskfs_journal_stop(jh);
138 return PTR_ERR(inode);
141 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
142 unlock_new_inode(inode);
144 if (feat->dif_flags & DT_IND_VARKEY)
145 rc = iam_lvar_create(inode, feat->dif_keysize_max,
146 feat->dif_ptrsize, feat->dif_recsize_max,
149 rc = iam_lfix_create(inode, feat->dif_keysize_max,
150 feat->dif_ptrsize, feat->dif_recsize_max,
152 dentry = osd_child_dentry_by_inode(env, dir, name, strlen(name));
153 rc = osd_ldiskfs_add_entry(info, jh, dentry, inode, NULL);
154 ldiskfs_journal_stop(jh);
159 static struct inode *osd_oi_index_open(struct osd_thread_info *info,
160 struct osd_device *osd,
162 struct dt_index_features *f,
165 struct dentry *dentry;
169 dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
171 return (void *) dentry;
173 if (dentry->d_inode) {
174 LASSERT(!is_bad_inode(dentry->d_inode));
175 inode = dentry->d_inode;
176 atomic_inc(&inode->i_count);
183 shrink_dcache_parent(osd_sb(osd)->s_root);
185 return ERR_PTR(-ENOENT);
187 rc = osd_oi_index_create_one(info, osd, name, f);
191 dentry = ll_lookup_one_len(name, osd_sb(osd)->s_root, strlen(name));
193 return (void *) dentry;
195 if (dentry->d_inode) {
196 LASSERT(!is_bad_inode(dentry->d_inode));
197 inode = dentry->d_inode;
198 atomic_inc(&inode->i_count);
203 return ERR_PTR(-ENOENT);
207 * Open an OI(Ojbect Index) container.
209 * \param name Name of OI container
210 * \param objp Pointer of returned OI
213 * \retval -ve failure
215 static int osd_oi_open(struct osd_thread_info *info, struct osd_device *osd,
216 char *name, struct osd_oi **oi_slot, bool create)
218 struct osd_directory *dir;
219 struct iam_container *bag;
226 oi_feat.dif_keysize_min = sizeof(struct lu_fid);
227 oi_feat.dif_keysize_max = sizeof(struct lu_fid);
229 inode = osd_oi_index_open(info, osd, name, &oi_feat, create);
231 RETURN(PTR_ERR(inode));
233 /* 'What the @fid is' is not imporatant, because these objects
234 * have no OI mappings, and only are visible inside the OSD.*/
235 lu_igif_build(&info->oti_fid, inode->i_ino, inode->i_generation);
236 rc = osd_ea_fid_set(info, inode, &info->oti_fid, LMAC_NOT_IN_OI, 0);
242 GOTO(out_inode, rc = -ENOMEM);
244 oi->oi_inode = inode;
247 bag = &dir->od_container;
248 rc = iam_container_init(bag, &dir->od_descr, inode);
252 rc = iam_container_setup(bag);
254 GOTO(out_container, rc);
260 iam_container_fini(bag);
269 * Open OI(Object Index) table.
270 * If \a oi_count is zero, which means caller doesn't know how many OIs there
271 * will be, this function can either return 0 for new filesystem, or number
272 * of OIs on existed filesystem.
274 * If \a oi_count is non-zero, which means caller does know number of OIs on
275 * filesystem, this function should return the exactly same number on
276 * success, or error code in failure.
278 * \param oi_count Number of expected OI containers
279 * \param create Create OIs if doesn't exist
281 * \retval +ve number of opened OI containers
282 * \retval 0 no OI containers found
283 * \retval -ve failure
286 osd_oi_table_open(struct osd_thread_info *info, struct osd_device *osd,
287 struct osd_oi **oi_table, unsigned oi_count, bool create)
289 struct scrub_file *sf = &osd->od_scrub.os_file;
295 /* NB: oi_count != 0 means that we have already created/known all OIs
296 * and have known exact number of OIs. */
297 LASSERT(oi_count <= OSD_OI_FID_NR_MAX);
299 for (i = 0; i < (oi_count != 0 ? oi_count : OSD_OI_FID_NR_MAX); i++) {
302 if (oi_table[i] != NULL) {
307 sprintf(name, "%s.%d", OSD_OI_NAME_BASE, i);
308 rc = osd_oi_open(info, osd, name, &oi_table[i], create);
314 if (rc == -ENOENT && create == false) {
319 ldiskfs_set_bit(i, sf->sf_oi_bitmap);
323 CERROR("%.16s: can't open %s: rc = %d\n",
324 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
326 CERROR("%.16s: expect to open total %d OI files.\n",
327 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
333 osd_oi_table_put(info, oi_table, oi_count > 0 ? oi_count : i);
340 static int osd_remove_oi_one(struct dentry *parent, const char *name,
343 struct dentry *child;
346 child = ll_lookup_one_len(name, parent, namelen);
350 rc = ll_vfs_unlink(parent->d_inode, child);
354 return rc == -ENOENT ? 0 : rc;
357 static int osd_remove_ois(struct osd_thread_info *info, struct osd_device *osd)
364 for (i = 0; i < osd->od_scrub.os_file.sf_oi_count; i++) {
365 namelen = snprintf(name, sizeof(name), "%s.%d",
366 OSD_OI_NAME_BASE, i);
367 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
369 CERROR("%.16s: fail to remove the stale OI file %s: "
371 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
377 namelen = snprintf(name, sizeof(name), "%s", OSD_OI_NAME_BASE);
378 rc = osd_remove_oi_one(osd_sb(osd)->s_root, name, namelen);
380 CERROR("%.16s: fail to remove the stale OI file %s: rc = %d\n",
381 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name, name, rc);
386 int osd_oi_init(struct osd_thread_info *info, struct osd_device *osd,
389 struct osd_scrub *scrub = &osd->od_scrub;
390 struct scrub_file *sf = &scrub->os_file;
396 rc = osd_remove_ois(info, osd);
401 OBD_ALLOC(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
405 mutex_lock(&oi_init_lock);
406 /* try to open existing multiple OIs first */
407 rc = osd_oi_table_open(info, osd, oi, sf->sf_oi_count, false);
412 if (rc == sf->sf_oi_count || sf->sf_oi_count == 0)
415 osd_scrub_file_reset(scrub,
416 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
418 osd_oi_count = sf->sf_oi_count;
422 /* if previous failed then try found single OI from old filesystem */
423 rc = osd_oi_open(info, osd, OSD_OI_NAME_BASE, &oi[0], false);
424 if (rc == 0) { /* found single OI from old filesystem */
425 ldiskfs_clear_bit(0, sf->sf_oi_bitmap);
426 if (sf->sf_success_count == 0)
427 /* XXX: There is one corner case that if the OI_scrub
428 * file crashed or lost and we regard it upgrade,
429 * then we allow IGIF lookup to bypass OI files.
431 * The risk is that osd_fid_lookup() may found
432 * a wrong inode with the given IGIF especially
433 * when the MDT has performed file-level backup
434 * and restored after former upgrading from 1.8
435 * to 2.x. Fortunately, the osd_fid_lookup()can
436 * verify the inode to decrease the risk. */
437 osd_scrub_file_reset(scrub,
438 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
441 } else if (rc != -ENOENT) {
442 CERROR("%.16s: can't open %s: rc = %d\n",
443 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
444 OSD_OI_NAME_BASE, rc);
448 if (sf->sf_oi_count > 0) {
451 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
452 for (i = 0; i < osd_oi_count; i++)
453 ldiskfs_set_bit(i, sf->sf_oi_bitmap);
454 osd_scrub_file_reset(scrub,
455 LDISKFS_SB(osd_sb(osd))->s_es->s_uuid,
458 sf->sf_oi_count = osd_oi_count;
461 rc = osd_scrub_file_store(scrub);
463 osd_oi_table_put(info, oi, sf->sf_oi_count);
467 /* No OIs exist, new filesystem, create OI objects */
468 rc = osd_oi_table_open(info, osd, oi, osd_oi_count, true);
469 LASSERT(ergo(rc >= 0, rc == osd_oi_count));
475 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
477 LASSERT((rc & (rc - 1)) == 0);
478 osd->od_oi_table = oi;
479 osd->od_oi_count = rc;
480 if (sf->sf_oi_count != rc) {
481 sf->sf_oi_count = rc;
482 rc = osd_scrub_file_store(scrub);
484 osd_oi_table_put(info, oi, sf->sf_oi_count);
485 OBD_FREE(oi, sizeof(*oi) * OSD_OI_FID_NR_MAX);
492 mutex_unlock(&oi_init_lock);
496 void osd_oi_fini(struct osd_thread_info *info, struct osd_device *osd)
498 if (unlikely(osd->od_oi_table == NULL))
501 osd_oi_table_put(info, osd->od_oi_table, osd->od_oi_count);
503 OBD_FREE(osd->od_oi_table,
504 sizeof(*(osd->od_oi_table)) * OSD_OI_FID_NR_MAX);
505 osd->od_oi_table = NULL;
508 static inline int fid_is_fs_root(const struct lu_fid *fid)
510 /* Map root inode to special local object FID */
511 return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
512 fid_oid(fid) == OSD_FS_ROOT_OID));
515 static int osd_oi_iam_lookup(struct osd_thread_info *oti,
516 struct osd_oi *oi, struct dt_rec *rec,
517 const struct dt_key *key)
519 struct iam_container *bag;
520 struct iam_iterator *it = &oti->oti_idx_it;
521 struct iam_path_descr *ipd;
526 LASSERT(oi->oi_inode);
528 bag = &oi->oi_dir.od_container;
529 ipd = osd_idx_ipd_get(oti->oti_env, bag);
533 /* got ipd now we can start iterator. */
534 iam_it_init(it, bag, 0, ipd);
536 rc = iam_it_get(it, (struct iam_key *)key);
538 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
541 osd_ipd_put(oti->oti_env, bag, ipd);
543 LINVRNT(osd_invariant(obj));
548 int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
549 const struct lu_fid *fid, enum oi_check_flags flags)
551 struct lu_seq_range *range = &info->oti_seq_range;
555 if (flags & OI_KNOWN_ON_OST)
558 if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
559 fid_is_llog(fid)) || fid_is_name_llog(fid) ||
563 if (fid_is_idif(fid) || fid_is_last_id(fid))
566 if (!(flags & OI_CHECK_FLD))
569 if (osd_seq_site(osd)->ss_server_fld == NULL)
572 rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
575 CERROR("%s: lookup FLD "DFID": rc = %d\n",
576 osd_name(osd), PFID(fid), rc);
580 if (fld_range_is_ost(range))
586 static int __osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
587 const struct lu_fid *fid, struct osd_inode_id *id)
589 struct lu_fid *oi_fid = &info->oti_fid2;
592 fid_cpu_to_be(oi_fid, fid);
593 rc = osd_oi_iam_lookup(info, osd_fid2oi(osd, fid), (struct dt_rec *)id,
594 (const struct dt_key *)oi_fid);
596 osd_id_unpack(id, id);
598 } else if (rc == 0) {
604 int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
605 const struct lu_fid *fid, struct osd_inode_id *id,
606 enum oi_check_flags flags)
608 if (unlikely(fid_is_last_id(fid)))
609 return osd_obj_spec_lookup(info, osd, fid, id);
611 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
612 return osd_obj_map_lookup(info, osd, fid, id);
615 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
617 if (fid_is_fs_root(fid)) {
618 osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
619 osd_sb(osd)->s_root->d_inode->i_generation);
622 if (unlikely(fid_is_acct(fid)))
623 return osd_acct_obj_lookup(info, osd, fid, id);
625 /* For other special FIDs, try OI first, then do spec lookup */
626 rc = __osd_oi_lookup(info, osd, fid, id);
628 return osd_obj_spec_lookup(info, osd, fid, id);
632 if (!osd->od_igif_inoi && fid_is_igif(fid)) {
633 osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
637 return __osd_oi_lookup(info, osd, fid, id);
640 static int osd_oi_iam_refresh(struct osd_thread_info *oti, struct osd_oi *oi,
641 const struct dt_rec *rec, const struct dt_key *key,
642 handle_t *th, bool insert)
644 struct iam_container *bag;
645 struct iam_path_descr *ipd;
650 LASSERT(oi->oi_inode);
651 ll_vfs_dq_init(oi->oi_inode);
653 bag = &oi->oi_dir.od_container;
654 ipd = osd_idx_ipd_get(oti->oti_env, bag);
655 if (unlikely(ipd == NULL))
659 LASSERT(th->h_transaction != NULL);
661 rc = iam_insert(th, bag, (const struct iam_key *)key,
662 (const struct iam_rec *)rec, ipd);
664 rc = iam_update(th, bag, (const struct iam_key *)key,
665 (const struct iam_rec *)rec, ipd);
666 osd_ipd_put(oti->oti_env, bag, ipd);
667 LINVRNT(osd_invariant(obj));
671 int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
672 const struct lu_fid *fid, const struct osd_inode_id *id,
673 handle_t *th, enum oi_check_flags flags, bool *exist)
675 struct lu_fid *oi_fid = &info->oti_fid2;
676 struct osd_inode_id *oi_id = &info->oti_id2;
679 if (unlikely(fid_is_last_id(fid)))
680 return osd_obj_spec_insert(info, osd, fid, id, th);
682 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
683 return osd_obj_map_insert(info, osd, fid, id, th);
685 fid_cpu_to_be(oi_fid, fid);
686 osd_id_pack(oi_id, id);
687 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
688 (const struct dt_rec *)oi_id,
689 (const struct dt_key *)oi_fid, th, true);
692 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
697 rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
701 if (unlikely(osd_id_eq(id, oi_id)))
704 /* Check whether the mapping for oi_id is valid or not. */
705 inode = osd_iget(info, osd, oi_id);
708 if (rc == -ENOENT || rc == -ESTALE)
713 /* The EA inode should NOT be in OI, old OI scrub may added
714 * such OI mapping by wrong, replace it. */
715 if (unlikely(osd_is_ea_inode(inode))) {
720 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
728 if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
729 lu_fid_eq(fid, &lma->lma_self_fid)) {
730 CERROR("%.16s: the FID "DFID" is used by two objects: "
732 LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
733 PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
734 id->oii_ino, id->oii_gen);
739 osd_id_pack(oi_id, id);
740 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
741 (const struct dt_rec *)oi_id,
742 (const struct dt_key *)oi_fid, th, false);
750 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
751 rc = osd_obj_spec_insert(info, osd, fid, id, th);
755 static int osd_oi_iam_delete(struct osd_thread_info *oti, struct osd_oi *oi,
756 const struct dt_key *key, handle_t *th)
758 struct iam_container *bag;
759 struct iam_path_descr *ipd;
764 LASSERT(oi->oi_inode);
765 ll_vfs_dq_init(oi->oi_inode);
767 bag = &oi->oi_dir.od_container;
768 ipd = osd_idx_ipd_get(oti->oti_env, bag);
769 if (unlikely(ipd == NULL))
773 LASSERT(th->h_transaction != NULL);
775 rc = iam_delete(th, bag, (const struct iam_key *)key, ipd);
776 osd_ipd_put(oti->oti_env, bag, ipd);
777 LINVRNT(osd_invariant(obj));
781 int osd_oi_delete(struct osd_thread_info *info,
782 struct osd_device *osd, const struct lu_fid *fid,
783 handle_t *th, enum oi_check_flags flags)
785 struct lu_fid *oi_fid = &info->oti_fid2;
787 /* clear idmap cache */
788 if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
789 fid_zero(&info->oti_cache.oic_fid);
791 if (fid_is_last_id(fid))
794 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
795 return osd_obj_map_delete(info, osd, fid, th);
797 fid_cpu_to_be(oi_fid, fid);
798 return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
799 (const struct dt_key *)oi_fid, th);
802 int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
803 const struct lu_fid *fid, const struct osd_inode_id *id,
804 handle_t *th, enum oi_check_flags flags)
806 struct lu_fid *oi_fid = &info->oti_fid2;
807 struct osd_inode_id *oi_id = &info->oti_id2;
810 if (unlikely(fid_is_last_id(fid)))
811 return osd_obj_spec_update(info, osd, fid, id, th);
813 if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
814 return osd_obj_map_update(info, osd, fid, id, th);
816 fid_cpu_to_be(oi_fid, fid);
817 osd_id_pack(oi_id, id);
818 rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
819 (const struct dt_rec *)oi_id,
820 (const struct dt_key *)oi_fid, th, false);
824 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
825 rc = osd_obj_spec_update(info, osd, fid, id, th);
829 int osd_oi_mod_init(void)
831 if (osd_oi_count == 0 || osd_oi_count > OSD_OI_FID_NR_MAX)
832 osd_oi_count = OSD_OI_FID_NR;
834 if ((osd_oi_count & (osd_oi_count - 1)) != 0) {
835 LCONSOLE_WARN("Round up oi_count %d to power2 %d\n",
836 osd_oi_count, size_roundup_power2(osd_oi_count));
837 osd_oi_count = size_roundup_power2(osd_oi_count);
840 mutex_init(&oi_init_lock);