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) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
60 * struct OBD_{ALLOC,FREE}*()
63 #include <obd_support.h>
64 /* struct ptlrpc_thread */
65 #include <lustre_net.h>
68 #include <lustre_fid.h>
70 #include "osd_internal.h"
73 /* llo_* api support */
74 #include <md_object.h>
75 #include <lustre_quota.h>
77 #ifdef HAVE_LDISKFS_PDO
79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
80 "ldiskfs with parallel directory operations");
85 static const char dot[] = ".";
86 static const char dotdot[] = "..";
87 static const char remote_obj_dir[] = "REM_OBJ_DIR";
89 static const struct lu_object_operations osd_lu_obj_ops;
90 static const struct dt_object_operations osd_obj_ops;
91 static const struct dt_object_operations osd_obj_ea_ops;
92 static const struct dt_object_operations osd_obj_otable_it_ops;
93 static const struct dt_index_operations osd_index_iam_ops;
94 static const struct dt_index_operations osd_index_ea_ops;
96 static int osd_has_index(const struct osd_object *obj)
98 return obj->oo_dt.do_index_ops != NULL;
101 static int osd_object_invariant(const struct lu_object *l)
103 return osd_invariant(osd_obj(l));
107 * Concurrency: doesn't matter
109 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
111 return osd_oti_get(env)->oti_r_locks > 0;
115 * Concurrency: doesn't matter
117 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
119 struct osd_thread_info *oti = osd_oti_get(env);
120 return oti->oti_w_locks > 0 && o->oo_owner == env;
124 * Concurrency: doesn't access mutable data
126 static int osd_root_get(const struct lu_env *env,
127 struct dt_device *dev, struct lu_fid *f)
129 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
134 * OSD object methods.
138 * Concurrency: no concurrent access is possible that early in object
141 static struct lu_object *osd_object_alloc(const struct lu_env *env,
142 const struct lu_object_header *hdr,
145 struct osd_object *mo;
151 l = &mo->oo_dt.do_lu;
152 dt_object_init(&mo->oo_dt, NULL, d);
153 mo->oo_dt.do_ops = &osd_obj_ea_ops;
154 l->lo_ops = &osd_lu_obj_ops;
155 init_rwsem(&mo->oo_sem);
156 init_rwsem(&mo->oo_ext_idx_sem);
157 spin_lock_init(&mo->oo_guard);
164 static int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
165 struct dentry *dentry, struct lustre_mdt_attrs *lma)
169 dentry->d_inode = inode;
170 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
173 /* try with old lma size */
174 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
175 info->oti_mdt_attrs_old,
178 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
181 /* Check LMA compatibility */
182 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
183 CWARN("%.16s: unsupported incompat LMA feature(s) "
185 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
186 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
190 lustre_lma_swab(lma);
193 } else if (rc == 0) {
201 * retrieve object from backend ext fs.
203 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
204 struct osd_inode_id *id)
206 struct inode *inode = NULL;
208 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
210 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
211 id->oii_ino, PTR_ERR(inode));
212 } else if (id->oii_gen != OSD_OII_NOGEN &&
213 inode->i_generation != id->oii_gen) {
214 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
216 id->oii_ino, id->oii_gen, inode->i_generation);
218 inode = ERR_PTR(-ESTALE);
219 } else if (inode->i_nlink == 0) {
220 /* due to parallel readdir and unlink,
221 * we can have dead inode here. */
222 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
223 make_bad_inode(inode);
225 inode = ERR_PTR(-ESTALE);
226 } else if (is_bad_inode(inode)) {
227 CWARN("%.16s: bad inode: ino = %u\n",
228 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
230 inode = ERR_PTR(-ENOENT);
232 if (id->oii_gen == OSD_OII_NOGEN)
233 osd_id_gen(id, inode->i_ino, inode->i_generation);
235 /* Do not update file c/mtime in ldiskfs.
236 * NB: we don't have any lock to protect this because we don't
237 * have reference on osd_object now, but contention with
238 * another lookup + attr_set can't happen in the tiny window
239 * between if (...) and set S_NOCMTIME. */
240 if (!(inode->i_flags & S_NOCMTIME))
241 inode->i_flags |= S_NOCMTIME;
246 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
247 struct osd_inode_id *id, struct lu_fid *fid)
249 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
253 inode = osd_iget(info, dev, id);
257 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
259 *fid = lma->lma_self_fid;
260 } else if (rc == -ENODATA) {
261 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
269 static struct inode *
270 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
271 struct osd_inode_id *id, const struct lu_fid *fid)
273 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
277 inode = osd_iget(info, dev, id);
281 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
290 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
291 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
292 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
294 return ERR_PTR(-EREMCHG);
300 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
301 const struct lu_fid *fid,
302 const struct lu_object_conf *conf)
304 struct osd_thread_info *info;
305 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
306 struct osd_device *dev;
307 struct osd_idmap_cache *oic;
308 struct osd_inode_id *id;
310 struct osd_scrub *scrub;
311 struct scrub_file *sf;
316 LINVRNT(osd_invariant(obj));
317 LASSERT(obj->oo_inode == NULL);
318 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
321 scrub = &dev->od_scrub;
322 sf = &scrub->os_file;
323 info = osd_oti_get(env);
325 oic = &info->oti_cache;
327 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
330 /* Search order: 1. per-thread cache. */
331 if (lu_fid_eq(fid, &oic->oic_fid)) {
337 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
338 /* Search order: 2. OI scrub pending list. */
339 result = osd_oii_lookup(dev, fid, id);
344 if (sf->sf_flags & SF_INCONSISTENT)
348 * Objects are created as locking anchors or place holders for objects
349 * yet to be created. No need to osd_oi_lookup() at here because FID
350 * shouldn't never be re-used, if it's really a duplicate FID from
351 * unexpected reason, we should be able to detect it later by calling
352 * do_create->osd_oi_insert()
354 if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
355 GOTO(out, result = 0);
357 /* Search order: 3. OI files. */
358 result = osd_oi_lookup(info, dev, fid, id);
359 if (result == -ENOENT) {
360 if (!fid_is_norm(fid) ||
361 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
363 GOTO(out, result = 0);
373 inode = osd_iget(info, dev, id);
375 inode = osd_iget_verify(info, dev, id, fid);
377 result = PTR_ERR(inode);
378 if (result == -ENOENT || result == -ESTALE) {
379 fid_zero(&oic->oic_fid);
381 } else if (result == -EREMCHG) {
384 if (thread_is_running(&scrub->os_thread)) {
385 result = -EINPROGRESS;
386 } else if (!dev->od_noscrub) {
387 result = osd_scrub_start(dev);
388 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
389 "for "DFID", rc = %d [1]\n",
390 LDISKFS_SB(osd_sb(dev))->s_es->\
391 s_volume_name,PFID(fid), result);
392 if (result == 0 || result == -EALREADY)
393 result = -EINPROGRESS;
402 obj->oo_inode = inode;
403 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
405 obj->oo_compat_dot_created = 1;
406 obj->oo_compat_dotdot_created = 1;
408 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
409 GOTO(out, result = 0);
411 LASSERT(obj->oo_hl_head == NULL);
412 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
413 if (obj->oo_hl_head == NULL) {
414 obj->oo_inode = NULL;
416 GOTO(out, result = -ENOMEM);
418 GOTO(out, result = 0);
421 LINVRNT(osd_invariant(obj));
426 * Concurrency: shouldn't matter.
428 static void osd_object_init0(struct osd_object *obj)
430 LASSERT(obj->oo_inode != NULL);
431 obj->oo_dt.do_body_ops = &osd_body_ops;
432 obj->oo_dt.do_lu.lo_header->loh_attr |=
433 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
437 * Concurrency: no concurrent access is possible that early in object
440 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
441 const struct lu_object_conf *conf)
443 struct osd_object *obj = osd_obj(l);
446 LINVRNT(osd_invariant(obj));
448 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
449 obj->oo_dt.do_body_ops = &osd_body_ops_new;
451 if (obj->oo_inode != NULL) {
452 osd_object_init0(obj);
453 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
454 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
455 /* LFSCK iterator object is special without inode */
456 l->lo_header->loh_attr |= LOHA_EXISTS;
459 LINVRNT(osd_invariant(obj));
464 * Concurrency: no concurrent access is possible that late in object
467 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
469 struct osd_object *obj = osd_obj(l);
471 LINVRNT(osd_invariant(obj));
473 dt_object_fini(&obj->oo_dt);
474 if (obj->oo_hl_head != NULL)
475 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
480 * Concurrency: no concurrent access is possible that late in object
483 static void osd_index_fini(struct osd_object *o)
485 struct iam_container *bag;
487 if (o->oo_dir != NULL) {
488 bag = &o->oo_dir->od_container;
489 if (o->oo_inode != NULL) {
490 if (bag->ic_object == o->oo_inode)
491 iam_container_fini(bag);
493 OBD_FREE_PTR(o->oo_dir);
499 * Concurrency: no concurrent access is possible that late in object
500 * life-cycle (for all existing callers, that is. New callers have to provide
501 * their own locking.)
503 static int osd_inode_unlinked(const struct inode *inode)
505 return inode->i_nlink == 0;
509 OSD_TXN_OI_DELETE_CREDITS = 20,
510 OSD_TXN_INODE_DELETE_CREDITS = 20
517 #if OSD_THANDLE_STATS
519 * Set time when the handle is allocated
521 static void osd_th_alloced(struct osd_thandle *oth)
523 oth->oth_alloced = cfs_time_current();
527 * Set time when the handle started
529 static void osd_th_started(struct osd_thandle *oth)
531 oth->oth_started = cfs_time_current();
535 * Helper function to convert time interval to microseconds packed in
536 * long int (default time units for the counter in "stats" initialized
537 * by lu_time_init() )
539 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
543 cfs_duration_usec(cfs_time_sub(end, start), &val);
544 return val.tv_sec * 1000000 + val.tv_usec;
548 * Check whether the we deal with this handle for too long.
550 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
551 cfs_time_t alloced, cfs_time_t started,
554 cfs_time_t now = cfs_time_current();
556 LASSERT(dev != NULL);
558 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
559 interval_to_usec(alloced, started));
560 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
561 interval_to_usec(started, closed));
562 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
563 interval_to_usec(closed, now));
565 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
566 CWARN("transaction handle %p was open for too long: "
568 "alloced "CFS_TIME_T" ,"
569 "started "CFS_TIME_T" ,"
570 "closed "CFS_TIME_T"\n",
571 oth, now, alloced, started, closed);
572 libcfs_debug_dumpstack(NULL);
576 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
578 cfs_time_t __closed = cfs_time_current(); \
579 cfs_time_t __alloced = oth->oth_alloced; \
580 cfs_time_t __started = oth->oth_started; \
583 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
586 #else /* OSD_THANDLE_STATS */
588 #define osd_th_alloced(h) do {} while(0)
589 #define osd_th_started(h) do {} while(0)
590 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
592 #endif /* OSD_THANDLE_STATS */
595 * Concurrency: doesn't access mutable data.
597 static int osd_param_is_not_sane(const struct osd_device *dev,
598 const struct thandle *th)
600 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
602 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
606 * Concurrency: shouldn't matter.
608 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
609 static void osd_trans_commit_cb(struct super_block *sb,
610 struct journal_callback *jcb, int error)
612 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
615 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
616 struct thandle *th = &oh->ot_super;
617 struct lu_device *lud = &th->th_dev->dd_lu_dev;
618 struct dt_txn_commit_cb *dcb, *tmp;
620 LASSERT(oh->ot_handle == NULL);
623 CERROR("transaction @0x%p commit error: %d\n", th, error);
625 dt_txn_hook_commit(th);
627 /* call per-transaction callbacks if any */
628 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
629 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
630 "commit callback entry: magic=%x name='%s'\n",
631 dcb->dcb_magic, dcb->dcb_name);
632 cfs_list_del_init(&dcb->dcb_linkage);
633 dcb->dcb_func(NULL, th, dcb, error);
636 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
640 lu_context_exit(&th->th_ctx);
641 lu_context_fini(&th->th_ctx);
645 static struct thandle *osd_trans_create(const struct lu_env *env,
648 struct osd_thread_info *oti = osd_oti_get(env);
649 struct osd_iobuf *iobuf = &oti->oti_iobuf;
650 struct osd_thandle *oh;
654 /* on pending IO in this thread should left from prev. request */
655 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
657 th = ERR_PTR(-ENOMEM);
658 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
660 oh->ot_quota_trans = &oti->oti_quota_trans;
661 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
665 th->th_tags = LCT_TX_HANDLE;
667 oti->oti_dev = osd_dt_dev(d);
668 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
675 * Concurrency: shouldn't matter.
677 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
680 struct osd_thread_info *oti = osd_oti_get(env);
681 struct osd_device *dev = osd_dt_dev(d);
683 struct osd_thandle *oh;
688 LASSERT(current->journal_info == NULL);
690 oh = container_of0(th, struct osd_thandle, ot_super);
692 LASSERT(oh->ot_handle == NULL);
694 rc = dt_txn_hook_start(env, d, th);
698 if (unlikely(osd_param_is_not_sane(dev, th))) {
699 static unsigned long last_printed;
700 static int last_credits;
702 CWARN("%.16s: too many transaction credits (%d > %d)\n",
703 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
705 osd_journal(dev)->j_max_transaction_buffers);
706 #ifdef OSD_TRACK_DECLARES
707 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
708 oh->ot_declare_create, oh->ot_declare_create_cred,
709 oh->ot_declare_delete, oh->ot_declare_delete_cred,
710 oh->ot_declare_destroy, oh->ot_declare_destroy_cred);
711 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
712 oh->ot_declare_attr_set, oh->ot_declare_attr_set_cred,
713 oh->ot_declare_xattr_set, oh->ot_declare_xattr_set_cred);
714 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
715 oh->ot_declare_write, oh->ot_declare_write_cred,
716 oh->ot_declare_punch, oh->ot_declare_punch_cred,
717 oh->ot_declare_quota, oh->ot_declare_quota_cred);
718 CWARN(" insert: %u/%u, delete: %u/%u\n",
719 oh->ot_declare_insert, oh->ot_declare_insert_cred,
720 oh->ot_declare_delete, oh->ot_declare_destroy_cred);
721 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
722 oh->ot_declare_ref_add, oh->ot_declare_ref_add_cred,
723 oh->ot_declare_ref_del, oh->ot_declare_ref_del_cred);
725 if (last_credits != oh->ot_credits &&
726 time_after(jiffies, last_printed + 60 * HZ)) {
727 libcfs_debug_dumpstack(NULL);
728 last_credits = oh->ot_credits;
729 last_printed = jiffies;
732 /* XXX Limit the credits to 'max_transaction_buffers', and
733 * let the underlying filesystem to catch the error if
734 * we really need so many credits.
736 * This should be removed when we can calculate the
737 * credits precisely. */
738 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
742 * XXX temporary stuff. Some abstraction layer should
745 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
749 LASSERT(oti->oti_txns == 0);
750 lu_context_init(&th->th_ctx, th->th_tags);
751 lu_context_enter(&th->th_ctx);
753 lu_device_get(&d->dd_lu_dev);
754 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
766 * Concurrency: shouldn't matter.
768 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
771 struct osd_thandle *oh;
772 struct osd_thread_info *oti = osd_oti_get(env);
773 struct osd_iobuf *iobuf = &oti->oti_iobuf;
774 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
777 oh = container_of0(th, struct osd_thandle, ot_super);
780 /* inform the quota slave device that the transaction is
782 qsd_op_end(env, qsd, oh->ot_quota_trans);
783 oh->ot_quota_trans = NULL;
785 if (oh->ot_handle != NULL) {
786 handle_t *hdl = oh->ot_handle;
789 * add commit callback
790 * notice we don't do this in osd_trans_start()
791 * as underlying transaction can change during truncate
793 osd_journal_callback_set(hdl, osd_trans_commit_cb,
796 LASSERT(oti->oti_txns == 1);
798 rc = dt_txn_hook_stop(env, th);
800 CERROR("Failure in transaction hook: %d\n", rc);
802 /* hook functions might modify th_sync */
803 hdl->h_sync = th->th_sync;
805 oh->ot_handle = NULL;
806 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
807 rc = ldiskfs_journal_stop(hdl));
809 CERROR("Failure to stop transaction: %d\n", rc);
814 /* as we want IO to journal and data IO be concurrent, we don't block
815 * awaiting data IO completion in osd_do_bio(), instead we wait here
816 * once transaction is submitted to the journal. all reqular requests
817 * don't do direct IO (except read/write), thus this wait_event becomes
820 * IMPORTANT: we have to wait till any IO submited by the thread is
821 * completed otherwise iobuf may be corrupted by different request
823 cfs_wait_event(iobuf->dr_wait,
824 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
826 rc = iobuf->dr_error;
831 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
833 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
836 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
837 LASSERT(&dcb->dcb_func != NULL);
838 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
844 * Called just before object is freed. Releases all resources except for
845 * object itself (that is released by osd_object_free()).
847 * Concurrency: no concurrent access is possible that late in object
850 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
852 struct osd_object *obj = osd_obj(l);
853 struct inode *inode = obj->oo_inode;
855 LINVRNT(osd_invariant(obj));
858 * If object is unlinked remove fid->ino mapping from object index.
863 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
864 qid_t uid = inode->i_uid;
865 qid_t gid = inode->i_gid;
868 obj->oo_inode = NULL;
871 struct osd_thread_info *info = osd_oti_get(env);
872 struct lquota_id_info *qi = &info->oti_qi;
874 /* Release granted quota to master if necessary */
875 qi->lqi_id.qid_uid = uid;
876 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
878 qi->lqi_id.qid_uid = gid;
879 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
885 * Concurrency: ->loo_object_release() is called under site spin-lock.
887 static void osd_object_release(const struct lu_env *env,
893 * Concurrency: shouldn't matter.
895 static int osd_object_print(const struct lu_env *env, void *cookie,
896 lu_printer_t p, const struct lu_object *l)
898 struct osd_object *o = osd_obj(l);
901 if (o->oo_dir != NULL)
902 d = o->oo_dir->od_container.ic_descr;
905 return (*p)(env, cookie,
906 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
908 o->oo_inode ? o->oo_inode->i_ino : 0UL,
909 o->oo_inode ? o->oo_inode->i_generation : 0,
910 d ? d->id_ops->id_name : "plain");
914 * Concurrency: shouldn't matter.
916 int osd_statfs(const struct lu_env *env, struct dt_device *d,
917 struct obd_statfs *sfs)
919 struct osd_device *osd = osd_dt_dev(d);
920 struct super_block *sb = osd_sb(osd);
921 struct kstatfs *ksfs;
924 if (unlikely(osd->od_mnt == NULL))
927 /* osd_lproc.c call this without env, allocate ksfs for that case */
928 if (unlikely(env == NULL)) {
933 ksfs = &osd_oti_get(env)->oti_ksfs;
936 spin_lock(&osd->od_osfs_lock);
938 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
939 result = sb->s_op->statfs(sb->s_root, ksfs);
940 if (likely(result == 0)) { /* N.B. statfs can't really fail */
941 osd->od_osfs_age = cfs_time_current_64();
942 statfs_pack(&osd->od_statfs, ksfs);
943 if (sb->s_flags & MS_RDONLY)
944 sfs->os_state = OS_STATE_READONLY;
948 if (likely(result == 0))
949 *sfs = osd->od_statfs;
950 spin_unlock(&osd->od_osfs_lock);
952 if (unlikely(env == NULL))
959 * Estimate space needed for file creations. We assume the largest filename
960 * which is 2^64 - 1, hence a filename of 20 chars.
961 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
963 #ifdef __LDISKFS_DIR_REC_LEN
964 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
966 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
970 * Concurrency: doesn't access mutable data.
972 static void osd_conf_get(const struct lu_env *env,
973 const struct dt_device *dev,
974 struct dt_device_param *param)
976 struct super_block *sb = osd_sb(osd_dt_dev(dev));
979 * XXX should be taken from not-yet-existing fs abstraction layer.
981 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
982 param->ddp_max_name_len = LDISKFS_NAME_LEN;
983 param->ddp_max_nlink = LDISKFS_LINK_MAX;
984 param->ddp_block_shift = sb->s_blocksize_bits;
985 param->ddp_mount_type = LDD_MT_LDISKFS;
986 param->ddp_maxbytes = sb->s_maxbytes;
987 /* Overhead estimate should be fairly accurate, so we really take a tiny
988 * error margin which also avoids fragmenting the filesystem too much */
989 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
990 /* inode are statically allocated, so per-inode space consumption
991 * is the space consumed by the directory entry */
992 param->ddp_inodespace = PER_OBJ_USAGE;
993 /* per-fragment overhead to be used by the client code */
994 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
995 param->ddp_mntopts = 0;
996 if (test_opt(sb, XATTR_USER))
997 param->ddp_mntopts |= MNTOPT_USERXATTR;
998 if (test_opt(sb, POSIX_ACL))
999 param->ddp_mntopts |= MNTOPT_ACL;
1001 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1002 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1003 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1006 param->ddp_max_ea_size = sb->s_blocksize;
1011 * Helper function to get and fill the buffer with input values.
1013 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1017 buf = &osd_oti_get(env)->oti_buf;
1024 * Concurrency: shouldn't matter.
1026 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1028 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1029 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1033 * Start commit for OSD device.
1035 * An implementation of dt_commit_async method for OSD device.
1036 * Asychronously starts underlayng fs sync and thereby a transaction
1039 * \param env environment
1040 * \param d dt device
1042 * \see dt_device_operations
1044 static int osd_commit_async(const struct lu_env *env,
1045 struct dt_device *d)
1047 struct super_block *s = osd_sb(osd_dt_dev(d));
1050 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1051 RETURN(s->s_op->sync_fs(s, 0));
1055 * Concurrency: shouldn't matter.
1058 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1060 struct super_block *sb = osd_sb(osd_dt_dev(d));
1064 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1066 rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1071 * Concurrency: serialization provided by callers.
1073 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1074 int mode, unsigned long timeout, __u32 alg,
1075 struct lustre_capa_key *keys)
1077 struct osd_device *dev = osd_dt_dev(d);
1080 dev->od_fl_capa = mode;
1081 dev->od_capa_timeout = timeout;
1082 dev->od_capa_alg = alg;
1083 dev->od_capa_keys = keys;
1088 * Note: we do not count into QUOTA here.
1089 * If we mount with --data_journal we may need more.
1091 const int osd_dto_credits_noquota[DTO_NR] = {
1094 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1095 * SINGLEDATA_TRANS_BLOCKS(8)
1096 * XXX Note: maybe iam need more, since iam have more level than
1099 [DTO_INDEX_INSERT] = 16,
1100 [DTO_INDEX_DELETE] = 16,
1104 [DTO_INDEX_UPDATE] = 16,
1106 * Create a object. The same as create object in EXT3.
1107 * DATA_TRANS_BLOCKS(14) +
1108 * INDEX_EXTRA_BLOCKS(8) +
1109 * 3(inode bits, groups, GDT)
1111 [DTO_OBJECT_CREATE] = 25,
1113 * XXX: real credits to be fixed
1115 [DTO_OBJECT_DELETE] = 25,
1117 * Attr set credits (inode)
1119 [DTO_ATTR_SET_BASE] = 1,
1121 * Xattr set. The same as xattr of EXT3.
1122 * DATA_TRANS_BLOCKS(14)
1123 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1124 * are also counted in. Do not know why?
1126 [DTO_XATTR_SET] = 14,
1129 * credits for inode change during write.
1131 [DTO_WRITE_BASE] = 3,
1133 * credits for single block write.
1135 [DTO_WRITE_BLOCK] = 14,
1137 * Attr set credits for chown.
1138 * This is extra credits for setattr, and it is null without quota
1140 [DTO_ATTR_SET_CHOWN]= 0
1143 static const struct dt_device_operations osd_dt_ops = {
1144 .dt_root_get = osd_root_get,
1145 .dt_statfs = osd_statfs,
1146 .dt_trans_create = osd_trans_create,
1147 .dt_trans_start = osd_trans_start,
1148 .dt_trans_stop = osd_trans_stop,
1149 .dt_trans_cb_add = osd_trans_cb_add,
1150 .dt_conf_get = osd_conf_get,
1151 .dt_sync = osd_sync,
1153 .dt_commit_async = osd_commit_async,
1154 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1157 static void osd_object_read_lock(const struct lu_env *env,
1158 struct dt_object *dt, unsigned role)
1160 struct osd_object *obj = osd_dt_obj(dt);
1161 struct osd_thread_info *oti = osd_oti_get(env);
1163 LINVRNT(osd_invariant(obj));
1165 LASSERT(obj->oo_owner != env);
1166 down_read_nested(&obj->oo_sem, role);
1168 LASSERT(obj->oo_owner == NULL);
1172 static void osd_object_write_lock(const struct lu_env *env,
1173 struct dt_object *dt, unsigned role)
1175 struct osd_object *obj = osd_dt_obj(dt);
1176 struct osd_thread_info *oti = osd_oti_get(env);
1178 LINVRNT(osd_invariant(obj));
1180 LASSERT(obj->oo_owner != env);
1181 down_write_nested(&obj->oo_sem, role);
1183 LASSERT(obj->oo_owner == NULL);
1184 obj->oo_owner = env;
1188 static void osd_object_read_unlock(const struct lu_env *env,
1189 struct dt_object *dt)
1191 struct osd_object *obj = osd_dt_obj(dt);
1192 struct osd_thread_info *oti = osd_oti_get(env);
1194 LINVRNT(osd_invariant(obj));
1196 LASSERT(oti->oti_r_locks > 0);
1198 up_read(&obj->oo_sem);
1201 static void osd_object_write_unlock(const struct lu_env *env,
1202 struct dt_object *dt)
1204 struct osd_object *obj = osd_dt_obj(dt);
1205 struct osd_thread_info *oti = osd_oti_get(env);
1207 LINVRNT(osd_invariant(obj));
1209 LASSERT(obj->oo_owner == env);
1210 LASSERT(oti->oti_w_locks > 0);
1212 obj->oo_owner = NULL;
1213 up_write(&obj->oo_sem);
1216 static int osd_object_write_locked(const struct lu_env *env,
1217 struct dt_object *dt)
1219 struct osd_object *obj = osd_dt_obj(dt);
1221 LINVRNT(osd_invariant(obj));
1223 return obj->oo_owner == env;
1226 static int capa_is_sane(const struct lu_env *env,
1227 struct osd_device *dev,
1228 struct lustre_capa *capa,
1229 struct lustre_capa_key *keys)
1231 struct osd_thread_info *oti = osd_oti_get(env);
1232 struct lustre_capa *tcapa = &oti->oti_capa;
1233 struct obd_capa *oc;
1237 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1239 if (capa_is_expired(oc)) {
1240 DEBUG_CAPA(D_ERROR, capa, "expired");
1247 if (capa_is_expired_sec(capa)) {
1248 DEBUG_CAPA(D_ERROR, capa, "expired");
1252 spin_lock(&capa_lock);
1253 for (i = 0; i < 2; i++) {
1254 if (keys[i].lk_keyid == capa->lc_keyid) {
1255 oti->oti_capa_key = keys[i];
1259 spin_unlock(&capa_lock);
1262 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1266 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1270 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1271 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1275 oc = capa_add(dev->od_capa_hash, capa);
1281 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1282 struct lustre_capa *capa, __u64 opc)
1284 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1285 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1286 struct md_capainfo *ci;
1289 if (!dev->od_fl_capa)
1292 if (capa == BYPASS_CAPA)
1295 ci = md_capainfo(env);
1299 if (ci->mc_auth == LC_ID_NONE)
1303 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1307 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1308 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1313 if (!capa_opc_supported(capa, opc)) {
1314 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1318 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1319 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1326 static struct timespec *osd_inode_time(const struct lu_env *env,
1327 struct inode *inode, __u64 seconds)
1329 struct osd_thread_info *oti = osd_oti_get(env);
1330 struct timespec *t = &oti->oti_time;
1332 t->tv_sec = seconds;
1334 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1339 static void osd_inode_getattr(const struct lu_env *env,
1340 struct inode *inode, struct lu_attr *attr)
1342 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1343 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1344 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1346 attr->la_atime = LTIME_S(inode->i_atime);
1347 attr->la_mtime = LTIME_S(inode->i_mtime);
1348 attr->la_ctime = LTIME_S(inode->i_ctime);
1349 attr->la_mode = inode->i_mode;
1350 attr->la_size = i_size_read(inode);
1351 attr->la_blocks = inode->i_blocks;
1352 attr->la_uid = inode->i_uid;
1353 attr->la_gid = inode->i_gid;
1354 attr->la_flags = LDISKFS_I(inode)->i_flags;
1355 attr->la_nlink = inode->i_nlink;
1356 attr->la_rdev = inode->i_rdev;
1357 attr->la_blksize = 1 << inode->i_blkbits;
1358 attr->la_blkbits = inode->i_blkbits;
1361 static int osd_attr_get(const struct lu_env *env,
1362 struct dt_object *dt,
1363 struct lu_attr *attr,
1364 struct lustre_capa *capa)
1366 struct osd_object *obj = osd_dt_obj(dt);
1368 LASSERT(dt_object_exists(dt));
1369 LINVRNT(osd_invariant(obj));
1371 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1374 spin_lock(&obj->oo_guard);
1375 osd_inode_getattr(env, obj->oo_inode, attr);
1376 spin_unlock(&obj->oo_guard);
1380 static int osd_declare_attr_set(const struct lu_env *env,
1381 struct dt_object *dt,
1382 const struct lu_attr *attr,
1383 struct thandle *handle)
1385 struct osd_thandle *oh;
1386 struct osd_object *obj;
1387 struct osd_thread_info *info = osd_oti_get(env);
1388 struct lquota_id_info *qi = &info->oti_qi;
1394 LASSERT(dt != NULL);
1395 LASSERT(handle != NULL);
1397 obj = osd_dt_obj(dt);
1398 LASSERT(osd_invariant(obj));
1400 oh = container_of0(handle, struct osd_thandle, ot_super);
1401 LASSERT(oh->ot_handle == NULL);
1403 OSD_DECLARE_OP(oh, attr_set,
1404 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1406 if (attr == NULL || obj->oo_inode == NULL)
1409 bspace = obj->oo_inode->i_blocks;
1410 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1411 bspace = toqb(bspace);
1413 /* Changing ownership is always preformed by super user, it should not
1416 * We still need to call the osd_declare_qid() to calculate the journal
1417 * credits for updating quota accounting files and to trigger quota
1418 * space adjustment once the operation is completed.*/
1419 if ((attr->la_valid & LA_UID) != 0 &&
1420 attr->la_uid != obj->oo_inode->i_uid) {
1421 qi->lqi_type = USRQUOTA;
1423 /* inode accounting */
1424 qi->lqi_is_blk = false;
1426 /* one more inode for the new owner ... */
1427 qi->lqi_id.qid_uid = attr->la_uid;
1429 allocated = (attr->la_uid == 0) ? true : false;
1430 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1431 if (rc == -EDQUOT || rc == -EINPROGRESS)
1436 /* and one less inode for the current uid */
1437 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1439 rc = osd_declare_qid(env, oh, qi, true, NULL);
1440 if (rc == -EDQUOT || rc == -EINPROGRESS)
1445 /* block accounting */
1446 qi->lqi_is_blk = true;
1448 /* more blocks for the new owner ... */
1449 qi->lqi_id.qid_uid = attr->la_uid;
1450 qi->lqi_space = bspace;
1451 allocated = (attr->la_uid == 0) ? true : false;
1452 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1453 if (rc == -EDQUOT || rc == -EINPROGRESS)
1458 /* and finally less blocks for the current owner */
1459 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1460 qi->lqi_space = -bspace;
1461 rc = osd_declare_qid(env, oh, qi, true, NULL);
1462 if (rc == -EDQUOT || rc == -EINPROGRESS)
1468 if (attr->la_valid & LA_GID &&
1469 attr->la_gid != obj->oo_inode->i_gid) {
1470 qi->lqi_type = GRPQUOTA;
1472 /* inode accounting */
1473 qi->lqi_is_blk = false;
1475 /* one more inode for the new group owner ... */
1476 qi->lqi_id.qid_gid = attr->la_gid;
1478 allocated = (attr->la_gid == 0) ? true : false;
1479 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1480 if (rc == -EDQUOT || rc == -EINPROGRESS)
1485 /* and one less inode for the current gid */
1486 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1488 rc = osd_declare_qid(env, oh, qi, true, NULL);
1489 if (rc == -EDQUOT || rc == -EINPROGRESS)
1494 /* block accounting */
1495 qi->lqi_is_blk = true;
1497 /* more blocks for the new owner ... */
1498 qi->lqi_id.qid_gid = attr->la_gid;
1499 qi->lqi_space = bspace;
1500 allocated = (attr->la_gid == 0) ? true : false;
1501 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1502 if (rc == -EDQUOT || rc == -EINPROGRESS)
1507 /* and finally less blocks for the current owner */
1508 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1509 qi->lqi_space = -bspace;
1510 rc = osd_declare_qid(env, oh, qi, true, NULL);
1511 if (rc == -EDQUOT || rc == -EINPROGRESS)
1520 static int osd_inode_setattr(const struct lu_env *env,
1521 struct inode *inode, const struct lu_attr *attr)
1525 bits = attr->la_valid;
1527 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1529 if (bits & LA_ATIME)
1530 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1531 if (bits & LA_CTIME)
1532 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1533 if (bits & LA_MTIME)
1534 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1535 if (bits & LA_SIZE) {
1536 LDISKFS_I(inode)->i_disksize = attr->la_size;
1537 i_size_write(inode, attr->la_size);
1541 /* OSD should not change "i_blocks" which is used by quota.
1542 * "i_blocks" should be changed by ldiskfs only. */
1543 if (bits & LA_BLOCKS)
1544 inode->i_blocks = attr->la_blocks;
1547 inode->i_mode = (inode->i_mode & S_IFMT) |
1548 (attr->la_mode & ~S_IFMT);
1550 inode->i_uid = attr->la_uid;
1552 inode->i_gid = attr->la_gid;
1553 if (bits & LA_NLINK)
1554 set_nlink(inode, attr->la_nlink);
1556 inode->i_rdev = attr->la_rdev;
1558 if (bits & LA_FLAGS) {
1559 /* always keep S_NOCMTIME */
1560 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1566 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1568 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1569 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1574 if (attr->la_valid & LA_UID)
1575 iattr.ia_valid |= ATTR_UID;
1576 if (attr->la_valid & LA_GID)
1577 iattr.ia_valid |= ATTR_GID;
1578 iattr.ia_uid = attr->la_uid;
1579 iattr.ia_gid = attr->la_gid;
1581 rc = ll_vfs_dq_transfer(inode, &iattr);
1583 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1584 "enforcement enabled on the ldiskfs filesystem?",
1585 inode->i_sb->s_id, rc);
1592 static int osd_attr_set(const struct lu_env *env,
1593 struct dt_object *dt,
1594 const struct lu_attr *attr,
1595 struct thandle *handle,
1596 struct lustre_capa *capa)
1598 struct osd_object *obj = osd_dt_obj(dt);
1599 struct inode *inode;
1602 LASSERT(handle != NULL);
1603 LASSERT(dt_object_exists(dt));
1604 LASSERT(osd_invariant(obj));
1606 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1609 OSD_EXEC_OP(handle, attr_set);
1611 inode = obj->oo_inode;
1612 ll_vfs_dq_init(inode);
1614 rc = osd_quota_transfer(inode, attr);
1618 spin_lock(&obj->oo_guard);
1619 rc = osd_inode_setattr(env, inode, attr);
1620 spin_unlock(&obj->oo_guard);
1623 inode->i_sb->s_op->dirty_inode(inode);
1627 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1628 struct osd_object *obj,
1629 const char *name, const int namelen)
1631 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1634 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1636 struct dt_allocation_hint *hint,
1640 struct osd_device *osd = osd_obj2dev(obj);
1641 struct osd_thandle *oth;
1642 struct dt_object *parent = NULL;
1643 struct inode *inode;
1645 LINVRNT(osd_invariant(obj));
1646 LASSERT(obj->oo_inode == NULL);
1647 LASSERT(obj->oo_hl_head == NULL);
1649 if (S_ISDIR(mode) && ldiskfs_pdo) {
1650 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1651 if (obj->oo_hl_head == NULL)
1655 oth = container_of(th, struct osd_thandle, ot_super);
1656 LASSERT(oth->ot_handle->h_transaction != NULL);
1658 if (hint && hint->dah_parent)
1659 parent = hint->dah_parent;
1661 inode = ldiskfs_create_inode(oth->ot_handle,
1662 parent ? osd_dt_obj(parent)->oo_inode :
1663 osd_sb(osd)->s_root->d_inode,
1665 if (!IS_ERR(inode)) {
1666 /* Do not update file c/mtime in ldiskfs.
1667 * NB: don't need any lock because no contention at this
1669 inode->i_flags |= S_NOCMTIME;
1670 inode->i_state |= I_LUSTRE_NOSCRUB;
1671 obj->oo_inode = inode;
1674 if (obj->oo_hl_head != NULL) {
1675 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1676 obj->oo_hl_head = NULL;
1678 result = PTR_ERR(inode);
1680 LINVRNT(osd_invariant(obj));
1688 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1689 struct lu_attr *attr,
1690 struct dt_allocation_hint *hint,
1691 struct dt_object_format *dof,
1695 struct osd_thandle *oth;
1696 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1698 LASSERT(S_ISDIR(attr->la_mode));
1700 oth = container_of(th, struct osd_thandle, ot_super);
1701 LASSERT(oth->ot_handle->h_transaction != NULL);
1702 result = osd_mkfile(info, obj, mode, hint, th);
1707 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1708 struct lu_attr *attr,
1709 struct dt_allocation_hint *hint,
1710 struct dt_object_format *dof,
1714 struct osd_thandle *oth;
1715 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1717 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1719 LASSERT(S_ISREG(attr->la_mode));
1721 oth = container_of(th, struct osd_thandle, ot_super);
1722 LASSERT(oth->ot_handle->h_transaction != NULL);
1724 result = osd_mkfile(info, obj, mode, hint, th);
1726 LASSERT(obj->oo_inode != NULL);
1727 if (feat->dif_flags & DT_IND_VARKEY)
1728 result = iam_lvar_create(obj->oo_inode,
1729 feat->dif_keysize_max,
1731 feat->dif_recsize_max,
1734 result = iam_lfix_create(obj->oo_inode,
1735 feat->dif_keysize_max,
1737 feat->dif_recsize_max,
1744 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1745 struct lu_attr *attr,
1746 struct dt_allocation_hint *hint,
1747 struct dt_object_format *dof,
1750 LASSERT(S_ISREG(attr->la_mode));
1751 return osd_mkfile(info, obj, (attr->la_mode &
1752 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1755 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1756 struct lu_attr *attr,
1757 struct dt_allocation_hint *hint,
1758 struct dt_object_format *dof,
1761 LASSERT(S_ISLNK(attr->la_mode));
1762 return osd_mkfile(info, obj, (attr->la_mode &
1763 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1766 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1767 struct lu_attr *attr,
1768 struct dt_allocation_hint *hint,
1769 struct dt_object_format *dof,
1772 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1775 LINVRNT(osd_invariant(obj));
1776 LASSERT(obj->oo_inode == NULL);
1777 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1778 S_ISFIFO(mode) || S_ISSOCK(mode));
1780 result = osd_mkfile(info, obj, mode, hint, th);
1782 LASSERT(obj->oo_inode != NULL);
1784 * This inode should be marked dirty for i_rdev. Currently
1785 * that is done in the osd_attr_init().
1787 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1789 LINVRNT(osd_invariant(obj));
1793 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1795 struct dt_allocation_hint *hint,
1796 struct dt_object_format *dof,
1799 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1801 osd_obj_type_f result;
1817 result = osd_mk_index;
1828 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1829 struct dt_object *parent, struct dt_object *child,
1830 cfs_umode_t child_mode)
1834 memset(ah, 0, sizeof(*ah));
1835 ah->dah_parent = parent;
1836 ah->dah_mode = child_mode;
1839 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1840 struct lu_attr *attr, struct dt_object_format *dof)
1842 struct inode *inode = obj->oo_inode;
1843 __u64 valid = attr->la_valid;
1846 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1848 if (dof->dof_type != DFT_NODE)
1849 attr->la_valid &= ~LA_RDEV;
1850 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1851 attr->la_valid &= ~LA_ATIME;
1852 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1853 attr->la_valid &= ~LA_CTIME;
1854 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1855 attr->la_valid &= ~LA_MTIME;
1857 result = osd_quota_transfer(inode, attr);
1861 if (attr->la_valid != 0) {
1862 result = osd_inode_setattr(info->oti_env, inode, attr);
1864 * The osd_inode_setattr() should always succeed here. The
1865 * only error that could be returned is EDQUOT when we are
1866 * trying to change the UID or GID of the inode. However, this
1867 * should not happen since quota enforcement is no longer
1868 * enabled on ldiskfs (lquota takes care of it).
1870 LASSERTF(result == 0, "%d", result);
1871 inode->i_sb->s_op->dirty_inode(inode);
1874 attr->la_valid = valid;
1878 * Helper function for osd_object_create()
1880 * \retval 0, on success
1882 static int __osd_object_create(struct osd_thread_info *info,
1883 struct osd_object *obj, struct lu_attr *attr,
1884 struct dt_allocation_hint *hint,
1885 struct dt_object_format *dof,
1890 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1893 osd_attr_init(info, obj, attr, dof);
1894 osd_object_init0(obj);
1896 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1897 unlock_new_inode(obj->oo_inode);
1904 * Helper function for osd_object_create()
1906 * \retval 0, on success
1908 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1909 const struct lu_fid *fid, struct thandle *th)
1911 struct osd_thread_info *info = osd_oti_get(env);
1912 struct osd_inode_id *id = &info->oti_id;
1913 struct osd_device *osd = osd_obj2dev(obj);
1915 LASSERT(obj->oo_inode != NULL);
1917 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1918 return osd_oi_insert(info, osd, fid, id, th);
1921 static int osd_declare_object_create(const struct lu_env *env,
1922 struct dt_object *dt,
1923 struct lu_attr *attr,
1924 struct dt_allocation_hint *hint,
1925 struct dt_object_format *dof,
1926 struct thandle *handle)
1928 struct osd_thandle *oh;
1932 LASSERT(handle != NULL);
1934 oh = container_of0(handle, struct osd_thandle, ot_super);
1935 LASSERT(oh->ot_handle == NULL);
1937 OSD_DECLARE_OP(oh, create, osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
1938 /* XXX: So far, only normal fid needs be inserted into the oi,
1939 * things could be changed later. Revise following code then. */
1940 if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1941 /* Reuse idle OI block may cause additional one OI block
1943 OSD_DECLARE_OP(oh, insert,
1944 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
1946 /* If this is directory, then we expect . and .. to be inserted as
1947 * well. The one directory block always needs to be created for the
1948 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1949 * block), there is no danger of needing a tree for the first block.
1951 if (attr && S_ISDIR(attr->la_mode)) {
1952 OSD_DECLARE_OP(oh, insert,
1953 osd_dto_credits_noquota[DTO_WRITE_BASE]);
1954 OSD_DECLARE_OP(oh, insert, 0);
1960 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
1961 false, false, NULL, false);
1965 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1966 struct lu_attr *attr,
1967 struct dt_allocation_hint *hint,
1968 struct dt_object_format *dof,
1971 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1972 struct osd_object *obj = osd_dt_obj(dt);
1973 struct osd_thread_info *info = osd_oti_get(env);
1978 LINVRNT(osd_invariant(obj));
1979 LASSERT(!dt_object_exists(dt));
1980 LASSERT(osd_write_locked(env, obj));
1981 LASSERT(th != NULL);
1983 if (unlikely(fid_is_acct(fid)))
1984 /* Quota files can't be created from the kernel any more,
1985 * 'tune2fs -O quota' will take care of creating them */
1988 OSD_EXEC_OP(th, create);
1990 result = __osd_object_create(info, obj, attr, hint, dof, th);
1992 result = __osd_oi_insert(env, obj, fid, th);
1994 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1995 LASSERT(osd_invariant(obj));
2000 * Called to destroy on-disk representation of the object
2002 * Concurrency: must be locked
2004 static int osd_declare_object_destroy(const struct lu_env *env,
2005 struct dt_object *dt,
2008 struct osd_object *obj = osd_dt_obj(dt);
2009 struct inode *inode = obj->oo_inode;
2010 struct osd_thandle *oh;
2014 oh = container_of0(th, struct osd_thandle, ot_super);
2015 LASSERT(oh->ot_handle == NULL);
2018 OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2019 /* XXX: So far, only normal fid needs to be inserted into the OI,
2020 * so only normal fid needs to be removed from the OI also.
2021 * Recycle idle OI leaf may cause additional three OI blocks
2023 OSD_DECLARE_OP(oh, destroy, fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2024 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
2026 /* one less inode */
2027 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2028 false, true, NULL, false);
2031 /* data to be truncated */
2032 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2033 true, true, NULL, false);
2037 static int osd_object_destroy(const struct lu_env *env,
2038 struct dt_object *dt,
2041 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2042 struct osd_object *obj = osd_dt_obj(dt);
2043 struct inode *inode = obj->oo_inode;
2044 struct osd_device *osd = osd_obj2dev(obj);
2045 struct osd_thandle *oh;
2049 oh = container_of0(th, struct osd_thandle, ot_super);
2050 LASSERT(oh->ot_handle);
2052 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2054 if (unlikely(fid_is_acct(fid)))
2057 /* Parallel control for OI scrub. For most of cases, there is no
2058 * lock contention. So it will not affect unlink performance. */
2059 mutex_lock(&inode->i_mutex);
2060 if (S_ISDIR(inode->i_mode)) {
2061 LASSERT(osd_inode_unlinked(inode) ||
2062 inode->i_nlink == 1);
2063 spin_lock(&obj->oo_guard);
2065 spin_unlock(&obj->oo_guard);
2066 inode->i_sb->s_op->dirty_inode(inode);
2068 LASSERT(osd_inode_unlinked(inode));
2071 OSD_EXEC_OP(th, destroy);
2073 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2074 mutex_unlock(&inode->i_mutex);
2076 /* XXX: add to ext3 orphan list */
2077 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2079 /* not needed in the cache anymore */
2080 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2086 * Helper function for osd_xattr_set()
2088 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2089 const struct lu_buf *buf, const char *name, int fl)
2091 struct osd_object *obj = osd_dt_obj(dt);
2092 struct inode *inode = obj->oo_inode;
2093 struct osd_thread_info *info = osd_oti_get(env);
2094 struct dentry *dentry = &info->oti_child_dentry;
2098 LASSERT(dt_object_exists(dt));
2099 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2101 if (fl & LU_XATTR_REPLACE)
2102 fs_flags |= XATTR_REPLACE;
2104 if (fl & LU_XATTR_CREATE)
2105 fs_flags |= XATTR_CREATE;
2107 ll_vfs_dq_init(inode);
2108 dentry->d_inode = inode;
2109 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2110 buf->lb_len, fs_flags);
2115 * Put the fid into lustre_mdt_attrs, and then place the structure
2116 * inode's ea. This fid should not be altered during the life time
2119 * \retval +ve, on success
2120 * \retval -ve, on error
2122 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2124 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2125 const struct lu_fid *fid)
2127 struct osd_thread_info *info = osd_oti_get(env);
2128 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2130 lustre_lma_init(mdt_attrs, fid);
2131 lustre_lma_swab(mdt_attrs);
2132 return __osd_xattr_set(env, dt,
2133 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2134 XATTR_NAME_LMA, LU_XATTR_CREATE);
2139 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2140 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2141 * To have compatilibility with 1.8 ldiskfs driver we need to have
2142 * magic number at start of fid data.
2143 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2146 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2147 const struct dt_rec *fid)
2149 param->edp_magic = LDISKFS_LUFID_MAGIC;
2150 param->edp_len = sizeof(struct lu_fid) + 1;
2152 fid_cpu_to_be((struct lu_fid *)param->edp_data,
2153 (struct lu_fid *)fid);
2157 * Try to read the fid from inode ea into dt_rec, if return value
2158 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2160 * \param fid object fid.
2162 * \retval 0 on success
2164 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2165 __u32 ino, struct lu_fid *fid,
2166 struct osd_inode_id *id)
2168 struct osd_thread_info *info = osd_oti_get(env);
2169 struct inode *inode;
2172 osd_id_gen(id, ino, OSD_OII_NOGEN);
2173 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2175 RETURN(PTR_ERR(inode));
2182 * OSD layer object create function for interoperability mode (b11826).
2183 * This is mostly similar to osd_object_create(). Only difference being, fid is
2184 * inserted into inode ea here.
2186 * \retval 0, on success
2187 * \retval -ve, on error
2189 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2190 struct lu_attr *attr,
2191 struct dt_allocation_hint *hint,
2192 struct dt_object_format *dof,
2195 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2196 struct osd_object *obj = osd_dt_obj(dt);
2197 struct osd_thread_info *info = osd_oti_get(env);
2202 LASSERT(osd_invariant(obj));
2203 LASSERT(!dt_object_exists(dt));
2204 LASSERT(osd_write_locked(env, obj));
2205 LASSERT(th != NULL);
2207 if (unlikely(fid_is_acct(fid)))
2208 /* Quota files can't be created from the kernel any more,
2209 * 'tune2fs -O quota' will take care of creating them */
2212 OSD_EXEC_OP(th, create);
2214 result = __osd_object_create(info, obj, attr, hint, dof, th);
2215 /* objects under osd root shld have igif fid, so dont add fid EA */
2216 if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2217 result = osd_ea_fid_set(env, dt, fid);
2220 result = __osd_oi_insert(env, obj, fid, th);
2222 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2223 LINVRNT(osd_invariant(obj));
2227 static int osd_declare_object_ref_add(const struct lu_env *env,
2228 struct dt_object *dt,
2229 struct thandle *handle)
2231 struct osd_thandle *oh;
2233 /* it's possible that object doesn't exist yet */
2234 LASSERT(handle != NULL);
2236 oh = container_of0(handle, struct osd_thandle, ot_super);
2237 LASSERT(oh->ot_handle == NULL);
2239 OSD_DECLARE_OP(oh, ref_add, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2245 * Concurrency: @dt is write locked.
2247 static int osd_object_ref_add(const struct lu_env *env,
2248 struct dt_object *dt, struct thandle *th)
2250 struct osd_object *obj = osd_dt_obj(dt);
2251 struct inode *inode = obj->oo_inode;
2253 LINVRNT(osd_invariant(obj));
2254 LASSERT(dt_object_exists(dt));
2255 LASSERT(osd_write_locked(env, obj));
2256 LASSERT(th != NULL);
2258 OSD_EXEC_OP(th, ref_add);
2261 * DIR_NLINK feature is set for compatibility reasons if:
2262 * 1) nlinks > LDISKFS_LINK_MAX, or
2263 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2265 * It is easier to always set this flag (rather than check and set),
2266 * since it has less overhead, and the superblock will be dirtied
2267 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2268 * do not actually care whether this flag is set or not.
2270 spin_lock(&obj->oo_guard);
2271 /* inc_nlink from 0 may cause WARN_ON */
2272 if(inode->i_nlink == 0)
2273 set_nlink(inode, 1);
2276 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2277 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2278 inode->i_nlink == 2)
2279 set_nlink(inode, 1);
2281 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2282 spin_unlock(&obj->oo_guard);
2283 inode->i_sb->s_op->dirty_inode(inode);
2284 LINVRNT(osd_invariant(obj));
2289 static int osd_declare_object_ref_del(const struct lu_env *env,
2290 struct dt_object *dt,
2291 struct thandle *handle)
2293 struct osd_thandle *oh;
2295 LASSERT(dt_object_exists(dt));
2296 LASSERT(handle != NULL);
2298 oh = container_of0(handle, struct osd_thandle, ot_super);
2299 LASSERT(oh->ot_handle == NULL);
2301 OSD_DECLARE_OP(oh, ref_del, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2307 * Concurrency: @dt is write locked.
2309 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2312 struct osd_object *obj = osd_dt_obj(dt);
2313 struct inode *inode = obj->oo_inode;
2315 LINVRNT(osd_invariant(obj));
2316 LASSERT(dt_object_exists(dt));
2317 LASSERT(osd_write_locked(env, obj));
2318 LASSERT(th != NULL);
2320 OSD_EXEC_OP(th, ref_del);
2322 spin_lock(&obj->oo_guard);
2323 LASSERT(inode->i_nlink > 0);
2325 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2326 * then the nlink count is 1. Don't let it be set to 0 or the directory
2327 * inode will be deleted incorrectly. */
2328 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2329 set_nlink(inode, 1);
2330 spin_unlock(&obj->oo_guard);
2331 inode->i_sb->s_op->dirty_inode(inode);
2332 LINVRNT(osd_invariant(obj));
2338 * Get the 64-bit version for an inode.
2340 static int osd_object_version_get(const struct lu_env *env,
2341 struct dt_object *dt, dt_obj_version_t *ver)
2343 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2345 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2346 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2347 *ver = LDISKFS_I(inode)->i_fs_version;
2352 * Concurrency: @dt is read locked.
2354 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2355 struct lu_buf *buf, const char *name,
2356 struct lustre_capa *capa)
2358 struct osd_object *obj = osd_dt_obj(dt);
2359 struct inode *inode = obj->oo_inode;
2360 struct osd_thread_info *info = osd_oti_get(env);
2361 struct dentry *dentry = &info->oti_obj_dentry;
2363 /* version get is not real XATTR but uses xattr API */
2364 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2365 /* for version we are just using xattr API but change inode
2367 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2368 osd_object_version_get(env, dt, buf->lb_buf);
2369 return sizeof(dt_obj_version_t);
2372 LASSERT(dt_object_exists(dt));
2373 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2375 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2378 dentry->d_inode = inode;
2379 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2383 static int osd_declare_xattr_set(const struct lu_env *env,
2384 struct dt_object *dt,
2385 const struct lu_buf *buf, const char *name,
2386 int fl, struct thandle *handle)
2388 struct osd_thandle *oh;
2390 LASSERT(handle != NULL);
2392 oh = container_of0(handle, struct osd_thandle, ot_super);
2393 LASSERT(oh->ot_handle == NULL);
2395 OSD_DECLARE_OP(oh, xattr_set, strcmp(name, XATTR_NAME_VERSION) == 0 ?
2396 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2397 osd_dto_credits_noquota[DTO_XATTR_SET]);
2403 * Set the 64-bit version for object
2405 static void osd_object_version_set(const struct lu_env *env,
2406 struct dt_object *dt,
2407 dt_obj_version_t *new_version)
2409 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2411 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2412 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2414 LDISKFS_I(inode)->i_fs_version = *new_version;
2415 /** Version is set after all inode operations are finished,
2416 * so we should mark it dirty here */
2417 inode->i_sb->s_op->dirty_inode(inode);
2421 * Concurrency: @dt is write locked.
2423 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2424 const struct lu_buf *buf, const char *name, int fl,
2425 struct thandle *handle, struct lustre_capa *capa)
2427 LASSERT(handle != NULL);
2429 /* version set is not real XATTR */
2430 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2431 /* for version we are just using xattr API but change inode
2433 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2434 osd_object_version_set(env, dt, buf->lb_buf);
2435 return sizeof(dt_obj_version_t);
2438 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2441 OSD_EXEC_OP(handle, xattr_set);
2442 return __osd_xattr_set(env, dt, buf, name, fl);
2446 * Concurrency: @dt is read locked.
2448 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2449 struct lu_buf *buf, struct lustre_capa *capa)
2451 struct osd_object *obj = osd_dt_obj(dt);
2452 struct inode *inode = obj->oo_inode;
2453 struct osd_thread_info *info = osd_oti_get(env);
2454 struct dentry *dentry = &info->oti_obj_dentry;
2456 LASSERT(dt_object_exists(dt));
2457 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2458 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2460 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2463 dentry->d_inode = inode;
2464 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2467 static int osd_declare_xattr_del(const struct lu_env *env,
2468 struct dt_object *dt, const char *name,
2469 struct thandle *handle)
2471 struct osd_thandle *oh;
2473 LASSERT(dt_object_exists(dt));
2474 LASSERT(handle != NULL);
2476 oh = container_of0(handle, struct osd_thandle, ot_super);
2477 LASSERT(oh->ot_handle == NULL);
2479 OSD_DECLARE_OP(oh, xattr_set, osd_dto_credits_noquota[DTO_XATTR_SET]);
2485 * Concurrency: @dt is write locked.
2487 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2488 const char *name, struct thandle *handle,
2489 struct lustre_capa *capa)
2491 struct osd_object *obj = osd_dt_obj(dt);
2492 struct inode *inode = obj->oo_inode;
2493 struct osd_thread_info *info = osd_oti_get(env);
2494 struct dentry *dentry = &info->oti_obj_dentry;
2497 LASSERT(dt_object_exists(dt));
2498 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2499 LASSERT(osd_write_locked(env, obj));
2500 LASSERT(handle != NULL);
2502 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2505 OSD_EXEC_OP(handle, xattr_set);
2507 ll_vfs_dq_init(inode);
2508 dentry->d_inode = inode;
2509 rc = inode->i_op->removexattr(dentry, name);
2513 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2514 struct dt_object *dt,
2515 struct lustre_capa *old,
2518 struct osd_thread_info *info = osd_oti_get(env);
2519 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2520 struct osd_object *obj = osd_dt_obj(dt);
2521 struct osd_device *dev = osd_obj2dev(obj);
2522 struct lustre_capa_key *key = &info->oti_capa_key;
2523 struct lustre_capa *capa = &info->oti_capa;
2524 struct obd_capa *oc;
2525 struct md_capainfo *ci;
2529 if (!dev->od_fl_capa)
2530 RETURN(ERR_PTR(-ENOENT));
2532 LASSERT(dt_object_exists(dt));
2533 LINVRNT(osd_invariant(obj));
2535 /* renewal sanity check */
2536 if (old && osd_object_auth(env, dt, old, opc))
2537 RETURN(ERR_PTR(-EACCES));
2539 ci = md_capainfo(env);
2541 RETURN(ERR_PTR(-ENOENT));
2543 switch (ci->mc_auth) {
2547 capa->lc_uid = obj->oo_inode->i_uid;
2548 capa->lc_gid = obj->oo_inode->i_gid;
2549 capa->lc_flags = LC_ID_PLAIN;
2551 case LC_ID_CONVERT: {
2554 s[0] = obj->oo_inode->i_uid;
2555 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2556 s[2] = obj->oo_inode->i_gid;
2557 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2558 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2560 RETURN(ERR_PTR(rc));
2562 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2563 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2564 capa->lc_flags = LC_ID_CONVERT;
2568 RETURN(ERR_PTR(-EINVAL));
2571 capa->lc_fid = *fid;
2573 capa->lc_flags |= dev->od_capa_alg << 24;
2574 capa->lc_timeout = dev->od_capa_timeout;
2575 capa->lc_expiry = 0;
2577 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2579 LASSERT(!capa_is_expired(oc));
2583 spin_lock(&capa_lock);
2584 *key = dev->od_capa_keys[1];
2585 spin_unlock(&capa_lock);
2587 capa->lc_keyid = key->lk_keyid;
2588 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2590 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2592 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2593 RETURN(ERR_PTR(rc));
2596 oc = capa_add(dev->od_capa_hash, capa);
2600 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2602 struct osd_object *obj = osd_dt_obj(dt);
2603 struct inode *inode = obj->oo_inode;
2604 struct osd_thread_info *info = osd_oti_get(env);
2605 struct dentry *dentry = &info->oti_obj_dentry;
2606 struct file *file = &info->oti_file;
2611 dentry->d_inode = inode;
2612 file->f_dentry = dentry;
2613 file->f_mapping = inode->i_mapping;
2614 file->f_op = inode->i_fop;
2615 mutex_lock(&inode->i_mutex);
2616 rc = file->f_op->fsync(file, dentry, 0);
2617 mutex_unlock(&inode->i_mutex);
2621 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2624 struct osd_object *obj = osd_dt_obj(dt);
2627 *data = (void *)obj->oo_inode;
2635 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2636 const struct dt_index_features *feat)
2638 struct iam_descr *descr;
2640 if (osd_object_is_root(o))
2641 return feat == &dt_directory_features;
2643 LASSERT(o->oo_dir != NULL);
2645 descr = o->oo_dir->od_container.ic_descr;
2646 if (feat == &dt_directory_features) {
2647 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2653 feat->dif_keysize_min <= descr->id_key_size &&
2654 descr->id_key_size <= feat->dif_keysize_max &&
2655 feat->dif_recsize_min <= descr->id_rec_size &&
2656 descr->id_rec_size <= feat->dif_recsize_max &&
2657 !(feat->dif_flags & (DT_IND_VARKEY |
2658 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2659 ergo(feat->dif_flags & DT_IND_UPDATE,
2660 1 /* XXX check that object (and file system) is
2665 static int osd_iam_container_init(const struct lu_env *env,
2666 struct osd_object *obj,
2667 struct osd_directory *dir)
2669 struct iam_container *bag = &dir->od_container;
2672 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2676 result = iam_container_setup(bag);
2678 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2680 iam_container_fini(bag);
2687 * Concurrency: no external locking is necessary.
2689 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2690 const struct dt_index_features *feat)
2694 struct osd_object *obj = osd_dt_obj(dt);
2696 LINVRNT(osd_invariant(obj));
2697 LASSERT(dt_object_exists(dt));
2699 if (osd_object_is_root(obj)) {
2700 dt->do_index_ops = &osd_index_ea_ops;
2702 } else if (feat == &dt_directory_features) {
2703 dt->do_index_ops = &osd_index_ea_ops;
2704 if (S_ISDIR(obj->oo_inode->i_mode))
2709 } else if (unlikely(feat == &dt_otable_features)) {
2710 dt->do_index_ops = &osd_otable_ops;
2712 } else if (feat == &dt_acct_features) {
2713 dt->do_index_ops = &osd_acct_index_ops;
2716 } else if (!osd_has_index(obj)) {
2717 struct osd_directory *dir;
2722 spin_lock(&obj->oo_guard);
2723 if (obj->oo_dir == NULL)
2727 * Concurrent thread allocated container data.
2730 spin_unlock(&obj->oo_guard);
2732 * Now, that we have container data, serialize its
2735 down_write(&obj->oo_ext_idx_sem);
2737 * recheck under lock.
2739 if (!osd_has_index(obj))
2740 result = osd_iam_container_init(env, obj, dir);
2743 up_write(&obj->oo_ext_idx_sem);
2751 if (result == 0 && skip_iam == 0) {
2752 if (!osd_iam_index_probe(env, obj, feat))
2755 LINVRNT(osd_invariant(obj));
2757 if (is_quota_glb_feat(feat))
2758 result = osd_quota_migration(env, dt, feat);
2763 static int osd_otable_it_attr_get(const struct lu_env *env,
2764 struct dt_object *dt,
2765 struct lu_attr *attr,
2766 struct lustre_capa *capa)
2772 static const struct dt_object_operations osd_obj_ops = {
2773 .do_read_lock = osd_object_read_lock,
2774 .do_write_lock = osd_object_write_lock,
2775 .do_read_unlock = osd_object_read_unlock,
2776 .do_write_unlock = osd_object_write_unlock,
2777 .do_write_locked = osd_object_write_locked,
2778 .do_attr_get = osd_attr_get,
2779 .do_declare_attr_set = osd_declare_attr_set,
2780 .do_attr_set = osd_attr_set,
2781 .do_ah_init = osd_ah_init,
2782 .do_declare_create = osd_declare_object_create,
2783 .do_create = osd_object_create,
2784 .do_declare_destroy = osd_declare_object_destroy,
2785 .do_destroy = osd_object_destroy,
2786 .do_index_try = osd_index_try,
2787 .do_declare_ref_add = osd_declare_object_ref_add,
2788 .do_ref_add = osd_object_ref_add,
2789 .do_declare_ref_del = osd_declare_object_ref_del,
2790 .do_ref_del = osd_object_ref_del,
2791 .do_xattr_get = osd_xattr_get,
2792 .do_declare_xattr_set = osd_declare_xattr_set,
2793 .do_xattr_set = osd_xattr_set,
2794 .do_declare_xattr_del = osd_declare_xattr_del,
2795 .do_xattr_del = osd_xattr_del,
2796 .do_xattr_list = osd_xattr_list,
2797 .do_capa_get = osd_capa_get,
2798 .do_object_sync = osd_object_sync,
2799 .do_data_get = osd_data_get,
2803 * dt_object_operations for interoperability mode
2804 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2806 static const struct dt_object_operations osd_obj_ea_ops = {
2807 .do_read_lock = osd_object_read_lock,
2808 .do_write_lock = osd_object_write_lock,
2809 .do_read_unlock = osd_object_read_unlock,
2810 .do_write_unlock = osd_object_write_unlock,
2811 .do_write_locked = osd_object_write_locked,
2812 .do_attr_get = osd_attr_get,
2813 .do_declare_attr_set = osd_declare_attr_set,
2814 .do_attr_set = osd_attr_set,
2815 .do_ah_init = osd_ah_init,
2816 .do_declare_create = osd_declare_object_create,
2817 .do_create = osd_object_ea_create,
2818 .do_declare_destroy = osd_declare_object_destroy,
2819 .do_destroy = osd_object_destroy,
2820 .do_index_try = osd_index_try,
2821 .do_declare_ref_add = osd_declare_object_ref_add,
2822 .do_ref_add = osd_object_ref_add,
2823 .do_declare_ref_del = osd_declare_object_ref_del,
2824 .do_ref_del = osd_object_ref_del,
2825 .do_xattr_get = osd_xattr_get,
2826 .do_declare_xattr_set = osd_declare_xattr_set,
2827 .do_xattr_set = osd_xattr_set,
2828 .do_declare_xattr_del = osd_declare_xattr_del,
2829 .do_xattr_del = osd_xattr_del,
2830 .do_xattr_list = osd_xattr_list,
2831 .do_capa_get = osd_capa_get,
2832 .do_object_sync = osd_object_sync,
2833 .do_data_get = osd_data_get,
2836 static const struct dt_object_operations osd_obj_otable_it_ops = {
2837 .do_attr_get = osd_otable_it_attr_get,
2838 .do_index_try = osd_index_try,
2841 static int osd_index_declare_iam_delete(const struct lu_env *env,
2842 struct dt_object *dt,
2843 const struct dt_key *key,
2844 struct thandle *handle)
2846 struct osd_thandle *oh;
2848 oh = container_of0(handle, struct osd_thandle, ot_super);
2849 LASSERT(oh->ot_handle == NULL);
2851 OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2857 * delete a (key, value) pair from index \a dt specified by \a key
2859 * \param dt osd index object
2860 * \param key key for index
2861 * \param rec record reference
2862 * \param handle transaction handler
2865 * \retval -ve failure
2868 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2869 const struct dt_key *key,
2870 struct thandle *handle,
2871 struct lustre_capa *capa)
2873 struct osd_thread_info *oti = osd_oti_get(env);
2874 struct osd_object *obj = osd_dt_obj(dt);
2875 struct osd_thandle *oh;
2876 struct iam_path_descr *ipd;
2877 struct iam_container *bag = &obj->oo_dir->od_container;
2882 LINVRNT(osd_invariant(obj));
2883 LASSERT(dt_object_exists(dt));
2884 LASSERT(bag->ic_object == obj->oo_inode);
2885 LASSERT(handle != NULL);
2887 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2890 OSD_EXEC_OP(handle, delete);
2892 ipd = osd_idx_ipd_get(env, bag);
2893 if (unlikely(ipd == NULL))
2896 oh = container_of0(handle, struct osd_thandle, ot_super);
2897 LASSERT(oh->ot_handle != NULL);
2898 LASSERT(oh->ot_handle->h_transaction != NULL);
2900 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
2901 /* swab quota uid/gid provided by caller */
2902 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
2903 key = (const struct dt_key *)&oti->oti_quota_id;
2906 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2907 osd_ipd_put(env, bag, ipd);
2908 LINVRNT(osd_invariant(obj));
2912 static int osd_index_declare_ea_delete(const struct lu_env *env,
2913 struct dt_object *dt,
2914 const struct dt_key *key,
2915 struct thandle *handle)
2917 struct osd_thandle *oh;
2918 struct inode *inode;
2922 LASSERT(dt_object_exists(dt));
2923 LASSERT(handle != NULL);
2925 oh = container_of0(handle, struct osd_thandle, ot_super);
2926 LASSERT(oh->ot_handle == NULL);
2928 OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2930 inode = osd_dt_obj(dt)->oo_inode;
2933 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2934 true, true, NULL, false);
2938 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2941 struct osd_fid_pack *rec;
2944 if (de->file_type & LDISKFS_DIRENT_LUFID) {
2945 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2946 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2952 * Index delete function for interoperability mode (b11826).
2953 * It will remove the directory entry added by osd_index_ea_insert().
2954 * This entry is needed to maintain name->fid mapping.
2956 * \param key, key i.e. file entry to be deleted
2958 * \retval 0, on success
2959 * \retval -ve, on error
2961 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2962 const struct dt_key *key,
2963 struct thandle *handle,
2964 struct lustre_capa *capa)
2966 struct osd_object *obj = osd_dt_obj(dt);
2967 struct inode *dir = obj->oo_inode;
2968 struct dentry *dentry;
2969 struct osd_thandle *oh;
2970 struct ldiskfs_dir_entry_2 *de;
2971 struct buffer_head *bh;
2972 struct htree_lock *hlock = NULL;
2977 LINVRNT(osd_invariant(obj));
2978 LASSERT(dt_object_exists(dt));
2979 LASSERT(handle != NULL);
2981 OSD_EXEC_OP(handle, delete);
2983 oh = container_of(handle, struct osd_thandle, ot_super);
2984 LASSERT(oh->ot_handle != NULL);
2985 LASSERT(oh->ot_handle->h_transaction != NULL);
2987 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2990 ll_vfs_dq_init(dir);
2991 dentry = osd_child_dentry_get(env, obj,
2992 (char *)key, strlen((char *)key));
2994 if (obj->oo_hl_head != NULL) {
2995 hlock = osd_oti_get(env)->oti_hlock;
2996 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
2997 dir, LDISKFS_HLOCK_DEL);
2999 down_write(&obj->oo_ext_idx_sem);
3002 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3004 rc = ldiskfs_delete_entry(oh->ot_handle,
3011 ldiskfs_htree_unlock(hlock);
3013 up_write(&obj->oo_ext_idx_sem);
3015 LASSERT(osd_invariant(obj));
3020 * Lookup index for \a key and copy record to \a rec.
3022 * \param dt osd index object
3023 * \param key key for index
3024 * \param rec record reference
3026 * \retval +ve success : exact mach
3027 * \retval 0 return record with key not greater than \a key
3028 * \retval -ve failure
3030 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3031 struct dt_rec *rec, const struct dt_key *key,
3032 struct lustre_capa *capa)
3034 struct osd_object *obj = osd_dt_obj(dt);
3035 struct iam_path_descr *ipd;
3036 struct iam_container *bag = &obj->oo_dir->od_container;
3037 struct osd_thread_info *oti = osd_oti_get(env);
3038 struct iam_iterator *it = &oti->oti_idx_it;
3039 struct iam_rec *iam_rec;
3044 LASSERT(osd_invariant(obj));
3045 LASSERT(dt_object_exists(dt));
3046 LASSERT(bag->ic_object == obj->oo_inode);
3048 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3051 ipd = osd_idx_ipd_get(env, bag);
3055 /* got ipd now we can start iterator. */
3056 iam_it_init(it, bag, 0, ipd);
3058 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3059 /* swab quota uid/gid provided by caller */
3060 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3061 key = (const struct dt_key *)&oti->oti_quota_id;
3064 rc = iam_it_get(it, (struct iam_key *)key);
3066 if (S_ISDIR(obj->oo_inode->i_mode))
3067 iam_rec = (struct iam_rec *)oti->oti_ldp;
3069 iam_rec = (struct iam_rec *) rec;
3071 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3073 if (S_ISDIR(obj->oo_inode->i_mode))
3074 osd_fid_unpack((struct lu_fid *) rec,
3075 (struct osd_fid_pack *)iam_rec);
3076 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3077 osd_quota_unpack(obj, rec);
3082 osd_ipd_put(env, bag, ipd);
3084 LINVRNT(osd_invariant(obj));
3089 static int osd_index_declare_iam_insert(const struct lu_env *env,
3090 struct dt_object *dt,
3091 const struct dt_rec *rec,
3092 const struct dt_key *key,
3093 struct thandle *handle)
3095 struct osd_thandle *oh;
3097 LASSERT(dt_object_exists(dt));
3098 LASSERT(handle != NULL);
3100 oh = container_of0(handle, struct osd_thandle, ot_super);
3101 LASSERT(oh->ot_handle == NULL);
3103 OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3109 * Inserts (key, value) pair in \a dt index object.
3111 * \param dt osd index object
3112 * \param key key for index
3113 * \param rec record reference
3114 * \param th transaction handler
3117 * \retval -ve failure
3119 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3120 const struct dt_rec *rec,
3121 const struct dt_key *key, struct thandle *th,
3122 struct lustre_capa *capa, int ignore_quota)
3124 struct osd_object *obj = osd_dt_obj(dt);
3125 struct iam_path_descr *ipd;
3126 struct osd_thandle *oh;
3127 struct iam_container *bag = &obj->oo_dir->od_container;
3128 struct osd_thread_info *oti = osd_oti_get(env);
3129 struct iam_rec *iam_rec;
3134 LINVRNT(osd_invariant(obj));
3135 LASSERT(dt_object_exists(dt));
3136 LASSERT(bag->ic_object == obj->oo_inode);
3137 LASSERT(th != NULL);
3139 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3142 OSD_EXEC_OP(th, insert);
3144 ipd = osd_idx_ipd_get(env, bag);
3145 if (unlikely(ipd == NULL))
3148 oh = container_of0(th, struct osd_thandle, ot_super);
3149 LASSERT(oh->ot_handle != NULL);
3150 LASSERT(oh->ot_handle->h_transaction != NULL);
3151 if (S_ISDIR(obj->oo_inode->i_mode)) {
3152 iam_rec = (struct iam_rec *)oti->oti_ldp;
3153 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3154 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3155 /* pack quota uid/gid */
3156 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3157 key = (const struct dt_key *)&oti->oti_quota_id;
3158 /* pack quota record */
3159 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3160 iam_rec = (struct iam_rec *)rec;
3162 iam_rec = (struct iam_rec *)rec;
3165 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3167 osd_ipd_put(env, bag, ipd);
3168 LINVRNT(osd_invariant(obj));
3173 * Calls ldiskfs_add_entry() to add directory entry
3174 * into the directory. This is required for
3175 * interoperability mode (b11826)
3177 * \retval 0, on success
3178 * \retval -ve, on error
3180 static int __osd_ea_add_rec(struct osd_thread_info *info,
3181 struct osd_object *pobj, struct inode *cinode,
3182 const char *name, const struct dt_rec *fid,
3183 struct htree_lock *hlock, struct thandle *th)
3185 struct ldiskfs_dentry_param *ldp;
3186 struct dentry *child;
3187 struct osd_thandle *oth;
3190 oth = container_of(th, struct osd_thandle, ot_super);
3191 LASSERT(oth->ot_handle != NULL);
3192 LASSERT(oth->ot_handle->h_transaction != NULL);
3194 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3196 /* XXX: remove fid_is_igif() check here.
3197 * IGIF check is just to handle insertion of .. when it is 'ROOT',
3198 * it is IGIF now but needs FID in dir entry as well for readdir
3200 * LU-838 should fix that and remove fid_is_igif() check */
3201 if (fid_is_igif((struct lu_fid *)fid) ||
3202 fid_is_norm((struct lu_fid *)fid)) {
3203 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3204 osd_get_ldiskfs_dirent_param(ldp, fid);
3205 child->d_fsdata = (void *)ldp;
3207 child->d_fsdata = NULL;
3209 LASSERT(pobj->oo_inode);
3210 ll_vfs_dq_init(pobj->oo_inode);
3211 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3217 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3218 * into the directory.Also sets flags into osd object to
3219 * indicate dot and dotdot are created. This is required for
3220 * interoperability mode (b11826)
3222 * \param dir directory for dot and dotdot fixup.
3223 * \param obj child object for linking
3225 * \retval 0, on success
3226 * \retval -ve, on error
3228 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3229 struct osd_object *dir,
3230 struct inode *parent_dir, const char *name,
3231 const struct dt_rec *dot_fid,
3232 const struct dt_rec *dot_dot_fid,
3235 struct inode *inode = dir->oo_inode;
3236 struct ldiskfs_dentry_param *dot_ldp;
3237 struct ldiskfs_dentry_param *dot_dot_ldp;
3238 struct osd_thandle *oth;
3241 oth = container_of(th, struct osd_thandle, ot_super);
3242 LASSERT(oth->ot_handle->h_transaction != NULL);
3243 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3245 if (strcmp(name, dot) == 0) {
3246 if (dir->oo_compat_dot_created) {
3249 LASSERT(inode == parent_dir);
3250 dir->oo_compat_dot_created = 1;
3253 } else if(strcmp(name, dotdot) == 0) {
3254 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3255 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3257 if (!dir->oo_compat_dot_created)
3259 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3260 osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3261 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3266 /* in case of rename, dotdot is already created */
3267 if (dir->oo_compat_dotdot_created) {
3268 return __osd_ea_add_rec(info, dir, parent_dir, name,
3269 dot_dot_fid, NULL, th);
3272 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3273 inode, dot_ldp, dot_dot_ldp);
3275 dir->oo_compat_dotdot_created = 1;
3283 * It will call the appropriate osd_add* function and return the
3284 * value, return by respective functions.
3286 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3287 struct inode *cinode, const char *name,
3288 const struct dt_rec *fid, struct thandle *th)
3290 struct osd_thread_info *info = osd_oti_get(env);
3291 struct htree_lock *hlock;
3294 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3296 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3298 if (hlock != NULL) {
3299 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3302 down_write(&pobj->oo_ext_idx_sem);
3304 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3305 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3308 if (hlock != NULL) {
3309 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3310 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3312 down_write(&pobj->oo_ext_idx_sem);
3315 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3319 ldiskfs_htree_unlock(hlock);
3321 up_write(&pobj->oo_ext_idx_sem);
3327 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3328 struct osd_idmap_cache *oic)
3330 struct osd_scrub *scrub = &dev->od_scrub;
3331 struct lu_fid *fid = &oic->oic_fid;
3332 struct osd_inode_id *id = &oti->oti_id;
3337 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3341 rc = osd_oi_lookup(oti, dev, fid, id);
3342 if (rc != 0 && rc != -ENOENT)
3345 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3348 if (thread_is_running(&scrub->os_thread)) {
3349 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3350 /* There is race condition between osd_oi_lookup and OI scrub.
3351 * The OI scrub finished just after osd_oi_lookup() failure.
3352 * Under such case, it is unnecessary to trigger OI scrub again,
3353 * but try to call osd_oi_lookup() again. */
3354 if (unlikely(rc == -EAGAIN))
3360 if (!dev->od_noscrub && ++once == 1) {
3361 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3363 rc = osd_scrub_start(dev);
3364 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3366 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3376 * Calls ->lookup() to find dentry. From dentry get inode and
3377 * read inode's ea to get fid. This is required for interoperability
3380 * \retval 0, on success
3381 * \retval -ve, on error
3383 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3384 struct dt_rec *rec, const struct dt_key *key)
3386 struct inode *dir = obj->oo_inode;
3387 struct dentry *dentry;
3388 struct ldiskfs_dir_entry_2 *de;
3389 struct buffer_head *bh;
3390 struct lu_fid *fid = (struct lu_fid *) rec;
3391 struct htree_lock *hlock = NULL;
3395 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3397 dentry = osd_child_dentry_get(env, obj,
3398 (char *)key, strlen((char *)key));
3400 if (obj->oo_hl_head != NULL) {
3401 hlock = osd_oti_get(env)->oti_hlock;
3402 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3403 dir, LDISKFS_HLOCK_LOOKUP);
3405 down_read(&obj->oo_ext_idx_sem);
3408 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3410 struct osd_thread_info *oti = osd_oti_get(env);
3411 struct osd_idmap_cache *oic = &oti->oti_cache;
3412 struct osd_device *dev = osd_obj2dev(obj);
3413 struct osd_scrub *scrub = &dev->od_scrub;
3414 struct scrub_file *sf = &scrub->os_file;
3416 ino = le32_to_cpu(de->inode);
3417 rc = osd_get_fid_from_dentry(de, rec);
3419 /* done with de, release bh */
3422 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3424 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3426 fid_zero(&oic->oic_fid);
3430 oic->oic_fid = *fid;
3431 if ((scrub->os_pos_current <= ino) &&
3432 (sf->sf_flags & SF_INCONSISTENT ||
3433 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3435 osd_consistency_check(oti, dev, oic);
3444 ldiskfs_htree_unlock(hlock);
3446 up_read(&obj->oo_ext_idx_sem);
3451 * Find the osd object for given fid.
3453 * \param fid need to find the osd object having this fid
3455 * \retval osd_object on success
3456 * \retval -ve on error
3458 struct osd_object *osd_object_find(const struct lu_env *env,
3459 struct dt_object *dt,
3460 const struct lu_fid *fid)
3462 struct lu_device *ludev = dt->do_lu.lo_dev;
3463 struct osd_object *child = NULL;
3464 struct lu_object *luch;
3465 struct lu_object *lo;
3468 * at this point topdev might not exist yet
3469 * (i.e. MGS is preparing profiles). so we can
3470 * not rely on topdev and instead lookup with
3471 * our device passed as topdev. this can't work
3472 * if the object isn't cached yet (as osd doesn't
3473 * allocate lu_header). IOW, the object must be
3474 * in the cache, otherwise lu_object_alloc() crashes
3477 luch = lu_object_find_at(env, ludev, fid, NULL);
3478 if (!IS_ERR(luch)) {
3479 if (lu_object_exists(luch)) {
3480 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3482 child = osd_obj(lo);
3484 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3485 "lu_object can't be located"
3486 DFID"\n", PFID(fid));
3488 if (child == NULL) {
3489 lu_object_put(env, luch);
3490 CERROR("Unable to get osd_object\n");
3491 child = ERR_PTR(-ENOENT);
3494 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3495 "lu_object does not exists "DFID"\n",
3497 lu_object_put(env, luch);
3498 child = ERR_PTR(-ENOENT);
3501 child = (void *)luch;
3507 * Put the osd object once done with it.
3509 * \param obj osd object that needs to be put
3511 static inline void osd_object_put(const struct lu_env *env,
3512 struct osd_object *obj)
3514 lu_object_put(env, &obj->oo_dt.do_lu);
3517 static int osd_index_declare_ea_insert(const struct lu_env *env,
3518 struct dt_object *dt,
3519 const struct dt_rec *rec,
3520 const struct dt_key *key,
3521 struct thandle *handle)
3523 struct osd_thandle *oh;
3524 struct inode *inode;
3528 LASSERT(dt_object_exists(dt));
3529 LASSERT(handle != NULL);
3531 oh = container_of0(handle, struct osd_thandle, ot_super);
3532 LASSERT(oh->ot_handle == NULL);
3534 OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3536 inode = osd_dt_obj(dt)->oo_inode;
3539 /* We ignore block quota on meta pool (MDTs), so needn't
3540 * calculate how many blocks will be consumed by this index
3542 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3543 true, true, NULL, false);
3548 * Index add function for interoperability mode (b11826).
3549 * It will add the directory entry.This entry is needed to
3550 * maintain name->fid mapping.
3552 * \param key it is key i.e. file entry to be inserted
3553 * \param rec it is value of given key i.e. fid
3555 * \retval 0, on success
3556 * \retval -ve, on error
3558 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3559 const struct dt_rec *rec,
3560 const struct dt_key *key, struct thandle *th,
3561 struct lustre_capa *capa, int ignore_quota)
3563 struct osd_object *obj = osd_dt_obj(dt);
3564 struct lu_fid *fid = (struct lu_fid *) rec;
3565 const char *name = (const char *)key;
3566 struct osd_object *child;
3571 LASSERT(osd_invariant(obj));
3572 LASSERT(dt_object_exists(dt));
3573 LASSERT(th != NULL);
3575 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3578 child = osd_object_find(env, dt, fid);
3579 if (!IS_ERR(child)) {
3580 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3581 osd_object_put(env, child);
3583 rc = PTR_ERR(child);
3586 LASSERT(osd_invariant(obj));
3591 * Initialize osd Iterator for given osd index object.
3593 * \param dt osd index object
3596 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3597 struct dt_object *dt,
3599 struct lustre_capa *capa)
3601 struct osd_it_iam *it;
3602 struct osd_thread_info *oti = osd_oti_get(env);
3603 struct osd_object *obj = osd_dt_obj(dt);
3604 struct lu_object *lo = &dt->do_lu;
3605 struct iam_path_descr *ipd;
3606 struct iam_container *bag = &obj->oo_dir->od_container;
3608 LASSERT(lu_object_exists(lo));
3610 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3611 return ERR_PTR(-EACCES);
3614 ipd = osd_it_ipd_get(env, bag);
3615 if (likely(ipd != NULL)) {
3619 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3620 return (struct dt_it *)it;
3622 return ERR_PTR(-ENOMEM);
3626 * free given Iterator.
3629 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3631 struct osd_it_iam *it = (struct osd_it_iam *)di;
3632 struct osd_object *obj = it->oi_obj;
3634 iam_it_fini(&it->oi_it);
3635 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3636 lu_object_put(env, &obj->oo_dt.do_lu);
3640 * Move Iterator to record specified by \a key
3642 * \param di osd iterator
3643 * \param key key for index
3645 * \retval +ve di points to record with least key not larger than key
3646 * \retval 0 di points to exact matched key
3647 * \retval -ve failure
3650 static int osd_it_iam_get(const struct lu_env *env,
3651 struct dt_it *di, const struct dt_key *key)
3653 struct osd_thread_info *oti = osd_oti_get(env);
3654 struct osd_it_iam *it = (struct osd_it_iam *)di;
3656 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3657 /* swab quota uid/gid */
3658 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3659 key = (struct dt_key *)&oti->oti_quota_id;
3662 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3668 * \param di osd iterator
3671 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3673 struct osd_it_iam *it = (struct osd_it_iam *)di;
3675 iam_it_put(&it->oi_it);
3679 * Move iterator by one record
3681 * \param di osd iterator
3683 * \retval +1 end of container reached
3685 * \retval -ve failure
3688 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3690 struct osd_it_iam *it = (struct osd_it_iam *)di;
3692 return iam_it_next(&it->oi_it);
3696 * Return pointer to the key under iterator.
3699 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3700 const struct dt_it *di)
3702 struct osd_thread_info *oti = osd_oti_get(env);
3703 struct osd_it_iam *it = (struct osd_it_iam *)di;
3704 struct osd_object *obj = it->oi_obj;
3707 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3709 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3710 /* swab quota uid/gid */
3711 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3712 key = (struct dt_key *)&oti->oti_quota_id;
3719 * Return size of key under iterator (in bytes)
3722 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3724 struct osd_it_iam *it = (struct osd_it_iam *)di;
3726 return iam_it_key_size(&it->oi_it);
3729 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3730 int len, __u16 type)
3732 struct luda_type *lt;
3733 const unsigned align = sizeof(struct luda_type) - 1;
3735 /* check if file type is required */
3736 if (attr & LUDA_TYPE) {
3737 len = (len + align) & ~align;
3739 lt = (void *) ent->lde_name + len;
3740 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3741 ent->lde_attrs |= LUDA_TYPE;
3744 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3748 * build lu direct from backend fs dirent.
3751 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3752 struct lu_fid *fid, __u64 offset,
3753 char *name, __u16 namelen,
3754 __u16 type, __u32 attr)
3756 fid_cpu_to_le(&ent->lde_fid, fid);
3757 ent->lde_attrs = LUDA_FID;
3759 ent->lde_hash = cpu_to_le64(offset);
3760 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3762 strncpy(ent->lde_name, name, namelen);
3763 ent->lde_namelen = cpu_to_le16(namelen);
3765 /* append lustre attributes */
3766 osd_it_append_attrs(ent, attr, namelen, type);
3770 * Return pointer to the record under iterator.
3772 static int osd_it_iam_rec(const struct lu_env *env,
3773 const struct dt_it *di,
3774 struct dt_rec *dtrec, __u32 attr)
3776 struct osd_it_iam *it = (struct osd_it_iam *)di;
3777 struct osd_thread_info *info = osd_oti_get(env);
3780 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3781 const struct osd_fid_pack *rec;
3782 struct lu_fid *fid = &info->oti_fid;
3783 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3789 name = (char *)iam_it_key_get(&it->oi_it);
3791 RETURN(PTR_ERR(name));
3793 namelen = iam_it_key_size(&it->oi_it);
3795 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3797 RETURN(PTR_ERR(rec));
3799 rc = osd_fid_unpack(fid, rec);
3803 hash = iam_it_store(&it->oi_it);
3805 /* IAM does not store object type in IAM index (dir) */
3806 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3808 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3809 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3810 (struct iam_rec *)dtrec);
3811 osd_quota_unpack(it->oi_obj, dtrec);
3813 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3814 (struct iam_rec *)dtrec);
3821 * Returns cookie for current Iterator position.
3823 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3825 struct osd_it_iam *it = (struct osd_it_iam *)di;
3827 return iam_it_store(&it->oi_it);
3831 * Restore iterator from cookie.
3833 * \param di osd iterator
3834 * \param hash Iterator location cookie
3836 * \retval +ve di points to record with least key not larger than key.
3837 * \retval 0 di points to exact matched key
3838 * \retval -ve failure
3841 static int osd_it_iam_load(const struct lu_env *env,
3842 const struct dt_it *di, __u64 hash)
3844 struct osd_it_iam *it = (struct osd_it_iam *)di;
3846 return iam_it_load(&it->oi_it, hash);
3849 static const struct dt_index_operations osd_index_iam_ops = {
3850 .dio_lookup = osd_index_iam_lookup,
3851 .dio_declare_insert = osd_index_declare_iam_insert,
3852 .dio_insert = osd_index_iam_insert,
3853 .dio_declare_delete = osd_index_declare_iam_delete,
3854 .dio_delete = osd_index_iam_delete,
3856 .init = osd_it_iam_init,
3857 .fini = osd_it_iam_fini,
3858 .get = osd_it_iam_get,
3859 .put = osd_it_iam_put,
3860 .next = osd_it_iam_next,
3861 .key = osd_it_iam_key,
3862 .key_size = osd_it_iam_key_size,
3863 .rec = osd_it_iam_rec,
3864 .store = osd_it_iam_store,
3865 .load = osd_it_iam_load
3871 * Creates or initializes iterator context.
3873 * \retval struct osd_it_ea, iterator structure on success
3876 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3877 struct dt_object *dt,
3879 struct lustre_capa *capa)
3881 struct osd_object *obj = osd_dt_obj(dt);
3882 struct osd_thread_info *info = osd_oti_get(env);
3883 struct osd_it_ea *it = &info->oti_it_ea;
3884 struct lu_object *lo = &dt->do_lu;
3885 struct dentry *obj_dentry = &info->oti_it_dentry;
3887 LASSERT(lu_object_exists(lo));
3889 obj_dentry->d_inode = obj->oo_inode;
3890 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3891 obj_dentry->d_name.hash = 0;
3893 it->oie_rd_dirent = 0;
3894 it->oie_it_dirent = 0;
3895 it->oie_dirent = NULL;
3896 it->oie_buf = info->oti_it_ea_buf;
3898 it->oie_file.f_pos = 0;
3899 it->oie_file.f_dentry = obj_dentry;
3900 if (attr & LUDA_64BITHASH)
3901 it->oie_file.f_mode |= FMODE_64BITHASH;
3903 it->oie_file.f_mode |= FMODE_32BITHASH;
3904 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
3905 it->oie_file.f_op = obj->oo_inode->i_fop;
3906 it->oie_file.private_data = NULL;
3908 RETURN((struct dt_it *) it);
3912 * Destroy or finishes iterator context.
3914 * \param di iterator structure to be destroyed
3916 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3918 struct osd_it_ea *it = (struct osd_it_ea *)di;
3919 struct osd_object *obj = it->oie_obj;
3920 struct inode *inode = obj->oo_inode;
3923 it->oie_file.f_op->release(inode, &it->oie_file);
3924 lu_object_put(env, &obj->oo_dt.do_lu);
3929 * It position the iterator at given key, so that next lookup continues from
3930 * that key Or it is similar to dio_it->load() but based on a key,
3931 * rather than file position.
3933 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3936 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3938 static int osd_it_ea_get(const struct lu_env *env,
3939 struct dt_it *di, const struct dt_key *key)
3941 struct osd_it_ea *it = (struct osd_it_ea *)di;
3944 LASSERT(((const char *)key)[0] == '\0');
3945 it->oie_file.f_pos = 0;
3946 it->oie_rd_dirent = 0;
3947 it->oie_it_dirent = 0;
3948 it->oie_dirent = NULL;
3956 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3961 * It is called internally by ->readdir(). It fills the
3962 * iterator's in-memory data structure with required
3963 * information i.e. name, namelen, rec_size etc.
3965 * \param buf in which information to be filled in.
3966 * \param name name of the file in given dir
3968 * \retval 0 on success
3969 * \retval 1 on buffer full
3971 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3972 loff_t offset, __u64 ino,
3975 struct osd_it_ea *it = (struct osd_it_ea *)buf;
3976 struct osd_it_ea_dirent *ent = it->oie_dirent;
3977 struct lu_fid *fid = &ent->oied_fid;
3978 struct osd_fid_pack *rec;
3981 /* this should never happen */
3982 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3983 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3987 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3991 if (d_type & LDISKFS_DIRENT_LUFID) {
3992 rec = (struct osd_fid_pack*) (name + namelen + 1);
3994 if (osd_fid_unpack(fid, rec) != 0)
3997 d_type &= ~LDISKFS_DIRENT_LUFID;
4002 ent->oied_ino = ino;
4003 ent->oied_off = offset;
4004 ent->oied_namelen = namelen;
4005 ent->oied_type = d_type;
4007 memcpy(ent->oied_name, name, namelen);
4009 it->oie_rd_dirent++;
4010 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4015 * Calls ->readdir() to load a directory entry at a time
4016 * and stored it in iterator's in-memory data structure.
4018 * \param di iterator's in memory structure
4020 * \retval 0 on success
4021 * \retval -ve on error
4023 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4024 const struct dt_it *di)
4026 struct osd_it_ea *it = (struct osd_it_ea *)di;
4027 struct osd_object *obj = it->oie_obj;
4028 struct inode *inode = obj->oo_inode;
4029 struct htree_lock *hlock = NULL;
4033 it->oie_dirent = it->oie_buf;
4034 it->oie_rd_dirent = 0;
4036 if (obj->oo_hl_head != NULL) {
4037 hlock = osd_oti_get(env)->oti_hlock;
4038 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4039 inode, LDISKFS_HLOCK_READDIR);
4041 down_read(&obj->oo_ext_idx_sem);
4044 result = inode->i_fop->readdir(&it->oie_file, it,
4045 (filldir_t) osd_ldiskfs_filldir);
4048 ldiskfs_htree_unlock(hlock);
4050 up_read(&obj->oo_ext_idx_sem);
4052 if (it->oie_rd_dirent == 0) {
4055 it->oie_dirent = it->oie_buf;
4056 it->oie_it_dirent = 1;
4063 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4064 * to load a directory entry at a time and stored it in
4065 * iterator's in-memory data structure.
4067 * \param di iterator's in memory structure
4069 * \retval +ve iterator reached to end
4070 * \retval 0 iterator not reached to end
4071 * \retval -ve on error
4073 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4075 struct osd_it_ea *it = (struct osd_it_ea *)di;
4080 if (it->oie_it_dirent < it->oie_rd_dirent) {
4082 (void *) it->oie_dirent +
4083 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4084 it->oie_dirent->oied_namelen);
4085 it->oie_it_dirent++;
4088 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4091 rc = osd_ldiskfs_it_fill(env, di);
4098 * Returns the key at current position from iterator's in memory structure.
4100 * \param di iterator's in memory structure
4102 * \retval key i.e. struct dt_key on success
4104 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4105 const struct dt_it *di)
4107 struct osd_it_ea *it = (struct osd_it_ea *)di;
4109 return (struct dt_key *)it->oie_dirent->oied_name;
4113 * Returns the key's size at current position from iterator's in memory structure.
4115 * \param di iterator's in memory structure
4117 * \retval key_size i.e. struct dt_key on success
4119 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4121 struct osd_it_ea *it = (struct osd_it_ea *)di;
4123 return it->oie_dirent->oied_namelen;
4128 * Returns the value (i.e. fid/igif) at current position from iterator's
4129 * in memory structure.
4131 * \param di struct osd_it_ea, iterator's in memory structure
4132 * \param attr attr requested for dirent.
4133 * \param lde lustre dirent
4135 * \retval 0 no error and \param lde has correct lustre dirent.
4136 * \retval -ve on error
4138 static inline int osd_it_ea_rec(const struct lu_env *env,
4139 const struct dt_it *di,
4140 struct dt_rec *dtrec, __u32 attr)
4142 struct osd_it_ea *it = (struct osd_it_ea *)di;
4143 struct osd_object *obj = it->oie_obj;
4144 struct osd_device *dev = osd_obj2dev(obj);
4145 struct osd_scrub *scrub = &dev->od_scrub;
4146 struct scrub_file *sf = &scrub->os_file;
4147 struct osd_thread_info *oti = osd_oti_get(env);
4148 struct osd_idmap_cache *oic = &oti->oti_cache;
4149 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4150 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4151 __u32 ino = it->oie_dirent->oied_ino;
4155 if (!fid_is_sane(fid)) {
4156 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4158 fid_zero(&oic->oic_fid);
4162 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4165 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4166 it->oie_dirent->oied_name,
4167 it->oie_dirent->oied_namelen,
4168 it->oie_dirent->oied_type, attr);
4169 oic->oic_fid = *fid;
4170 if ((scrub->os_pos_current <= ino) &&
4171 (sf->sf_flags & SF_INCONSISTENT ||
4172 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4173 osd_consistency_check(oti, dev, oic);
4179 * Returns a cookie for current position of the iterator head, so that
4180 * user can use this cookie to load/start the iterator next time.
4182 * \param di iterator's in memory structure
4184 * \retval cookie for current position, on success
4186 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4188 struct osd_it_ea *it = (struct osd_it_ea *)di;
4190 return it->oie_dirent->oied_off;
4194 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4195 * to load a directory entry at a time and stored it i inn,
4196 * in iterator's in-memory data structure.
4198 * \param di struct osd_it_ea, iterator's in memory structure
4200 * \retval +ve on success
4201 * \retval -ve on error
4203 static int osd_it_ea_load(const struct lu_env *env,
4204 const struct dt_it *di, __u64 hash)
4206 struct osd_it_ea *it = (struct osd_it_ea *)di;
4210 it->oie_file.f_pos = hash;
4212 rc = osd_ldiskfs_it_fill(env, di);
4220 * Index lookup function for interoperability mode (b11826).
4222 * \param key, key i.e. file name to be searched
4224 * \retval +ve, on success
4225 * \retval -ve, on error
4227 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4228 struct dt_rec *rec, const struct dt_key *key,
4229 struct lustre_capa *capa)
4231 struct osd_object *obj = osd_dt_obj(dt);
4236 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4237 LINVRNT(osd_invariant(obj));
4239 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4242 rc = osd_ea_lookup_rec(env, obj, rec, key);
4249 * Index and Iterator operations for interoperability
4250 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4252 static const struct dt_index_operations osd_index_ea_ops = {
4253 .dio_lookup = osd_index_ea_lookup,
4254 .dio_declare_insert = osd_index_declare_ea_insert,
4255 .dio_insert = osd_index_ea_insert,
4256 .dio_declare_delete = osd_index_declare_ea_delete,
4257 .dio_delete = osd_index_ea_delete,
4259 .init = osd_it_ea_init,
4260 .fini = osd_it_ea_fini,
4261 .get = osd_it_ea_get,
4262 .put = osd_it_ea_put,
4263 .next = osd_it_ea_next,
4264 .key = osd_it_ea_key,
4265 .key_size = osd_it_ea_key_size,
4266 .rec = osd_it_ea_rec,
4267 .store = osd_it_ea_store,
4268 .load = osd_it_ea_load
4272 static void *osd_key_init(const struct lu_context *ctx,
4273 struct lu_context_key *key)
4275 struct osd_thread_info *info;
4277 OBD_ALLOC_PTR(info);
4279 return ERR_PTR(-ENOMEM);
4281 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4282 if (info->oti_it_ea_buf == NULL)
4285 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4287 info->oti_hlock = ldiskfs_htree_lock_alloc();
4288 if (info->oti_hlock == NULL)
4294 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4297 return ERR_PTR(-ENOMEM);
4300 static void osd_key_fini(const struct lu_context *ctx,
4301 struct lu_context_key *key, void* data)
4303 struct osd_thread_info *info = data;
4305 if (info->oti_hlock != NULL)
4306 ldiskfs_htree_lock_free(info->oti_hlock);
4307 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4311 static void osd_key_exit(const struct lu_context *ctx,
4312 struct lu_context_key *key, void *data)
4314 struct osd_thread_info *info = data;
4316 LASSERT(info->oti_r_locks == 0);
4317 LASSERT(info->oti_w_locks == 0);
4318 LASSERT(info->oti_txns == 0);
4321 /* type constructor/destructor: osd_type_init, osd_type_fini */
4322 LU_TYPE_INIT_FINI(osd, &osd_key);
4324 struct lu_context_key osd_key = {
4325 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4326 .lct_init = osd_key_init,
4327 .lct_fini = osd_key_fini,
4328 .lct_exit = osd_key_exit
4332 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4333 const char *name, struct lu_device *next)
4335 struct osd_device *osd = osd_dev(d);
4337 strncpy(osd->od_svname, name, MAX_OBD_NAME);
4338 return osd_procfs_init(osd, name);
4341 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4345 osd_scrub_cleanup(env, o);
4348 fsfilt_put_ops(o->od_fsops);
4352 /* shutdown quota slave instance associated with the device */
4353 if (o->od_quota_slave != NULL) {
4354 qsd_fini(env, o->od_quota_slave);
4355 o->od_quota_slave = NULL;
4361 static int osd_mount(const struct lu_env *env,
4362 struct osd_device *o, struct lustre_cfg *cfg)
4364 const char *name = lustre_cfg_string(cfg, 0);
4365 const char *dev = lustre_cfg_string(cfg, 1);
4367 unsigned long page, s_flags, lmd_flags = 0;
4368 struct page *__page;
4369 struct file_system_type *type;
4370 char *options = NULL;
4375 if (o->od_mnt != NULL)
4378 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4379 if (o->od_fsops == NULL) {
4380 CERROR("Can't find fsfilt_ldiskfs\n");
4384 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4386 GOTO(out, rc = -ENOMEM);
4388 str = lustre_cfg_string(cfg, 2);
4389 s_flags = simple_strtoul(str, NULL, 0);
4390 str = strstr(str, ":");
4392 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4393 opts = lustre_cfg_string(cfg, 3);
4394 page = (unsigned long)cfs_page_address(__page);
4395 options = (char *)page;
4398 strcat(options, "user_xattr,acl");
4400 strcat(options, opts);
4402 /* Glom up mount options */
4403 if (*options != '\0')
4404 strcat(options, ",");
4405 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4407 type = get_fs_type("ldiskfs");
4409 CERROR("%s: cannot find ldiskfs module\n", name);
4410 GOTO(out, rc = -ENODEV);
4413 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4414 cfs_module_put(type->owner);
4416 if (IS_ERR(o->od_mnt)) {
4417 rc = PTR_ERR(o->od_mnt);
4418 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4423 if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4424 CERROR("%s: underlying device %s is marked as read-only. "
4425 "Setup failed\n", name, dev);
4428 GOTO(out, rc = -EROFS);
4431 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4432 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4433 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4436 GOTO(out, rc = -EINVAL);
4439 if (lmd_flags & LMD_FLG_NOSCRUB)
4444 OBD_PAGE_FREE(__page);
4446 fsfilt_put_ops(o->od_fsops);
4451 static struct lu_device *osd_device_fini(const struct lu_env *env,
4452 struct lu_device *d)
4457 rc = osd_shutdown(env, osd_dev(d));
4459 osd_compat_fini(osd_dev(d));
4461 shrink_dcache_sb(osd_sb(osd_dev(d)));
4462 osd_sync(env, lu2dt_dev(d));
4464 rc = osd_procfs_fini(osd_dev(d));
4466 CERROR("proc fini error %d \n", rc);
4467 RETURN (ERR_PTR(rc));
4470 if (osd_dev(d)->od_mnt) {
4471 mntput(osd_dev(d)->od_mnt);
4472 osd_dev(d)->od_mnt = NULL;
4478 static int osd_device_init0(const struct lu_env *env,
4479 struct osd_device *o,
4480 struct lustre_cfg *cfg)
4482 struct lu_device *l = osd2lu_dev(o);
4483 struct osd_thread_info *info;
4486 /* if the module was re-loaded, env can loose its keys */
4487 rc = lu_env_refill((struct lu_env *) env);
4490 info = osd_oti_get(env);
4493 l->ld_ops = &osd_lu_ops;
4494 o->od_dt_dev.dd_ops = &osd_dt_ops;
4496 spin_lock_init(&o->od_osfs_lock);
4497 mutex_init(&o->od_otable_mutex);
4498 o->od_osfs_age = cfs_time_shift_64(-1000);
4500 o->od_capa_hash = init_capa_hash();
4501 if (o->od_capa_hash == NULL)
4502 GOTO(out, rc = -ENOMEM);
4504 o->od_read_cache = 1;
4505 o->od_writethrough_cache = 1;
4506 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4508 rc = osd_mount(env, o, cfg);
4512 /* setup scrub, including OI files initialization */
4513 rc = osd_scrub_setup(env, o);
4517 strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4518 sizeof(o->od_svname) - 1);
4520 rc = osd_compat_init(o);
4522 GOTO(out_scrub, rc);
4524 rc = lu_site_init(&o->od_site, l);
4526 GOTO(out_compat, rc);
4527 o->od_site.ls_bottom_dev = l;
4529 rc = lu_site_init_finish(&o->od_site);
4533 rc = osd_procfs_init(o, o->od_svname);
4535 CERROR("%s: can't initialize procfs: rc = %d\n",
4540 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4542 /* initialize quota slave instance */
4543 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
4545 if (IS_ERR(o->od_quota_slave)) {
4546 rc = PTR_ERR(o->od_quota_slave);
4547 o->od_quota_slave = NULL;
4548 GOTO(out_procfs, rc);
4555 lu_site_fini(&o->od_site);
4559 osd_scrub_cleanup(env, o);
4561 osd_oi_fini(info, o);
4562 osd_shutdown(env, o);
4566 cleanup_capa_hash(o->od_capa_hash);
4571 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4572 struct lu_device_type *t,
4573 struct lustre_cfg *cfg)
4575 struct osd_device *o;
4580 return ERR_PTR(-ENOMEM);
4582 rc = dt_device_init(&o->od_dt_dev, t);
4584 rc = osd_device_init0(env, o, cfg);
4586 dt_device_fini(&o->od_dt_dev);
4589 if (unlikely(rc != 0))
4592 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4595 static struct lu_device *osd_device_free(const struct lu_env *env,
4596 struct lu_device *d)
4598 struct osd_device *o = osd_dev(d);
4601 cleanup_capa_hash(o->od_capa_hash);
4602 /* XXX: make osd top device in order to release reference */
4603 d->ld_site->ls_top_dev = d;
4604 lu_site_purge(env, d->ld_site, -1);
4605 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4606 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4607 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4609 lu_site_fini(&o->od_site);
4610 dt_device_fini(&o->od_dt_dev);
4615 static int osd_process_config(const struct lu_env *env,
4616 struct lu_device *d, struct lustre_cfg *cfg)
4618 struct osd_device *o = osd_dev(d);
4622 switch(cfg->lcfg_command) {
4624 err = osd_mount(env, o, cfg);
4627 lu_dev_del_linkage(d->ld_site, d);
4628 err = osd_shutdown(env, o);
4637 static int osd_recovery_complete(const struct lu_env *env,
4638 struct lu_device *d)
4640 struct osd_device *osd = osd_dev(d);
4644 if (osd->od_quota_slave == NULL)
4647 /* start qsd instance on recovery completion, this notifies the quota
4648 * slave code that we are about to process new requests now */
4649 rc = qsd_start(env, osd->od_quota_slave);
4654 * we use exports to track all osd users
4656 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4657 struct obd_device *obd, struct obd_uuid *cluuid,
4658 struct obd_connect_data *data, void *localdata)
4660 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4661 struct lustre_handle conn;
4665 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4667 rc = class_connect(&conn, obd, cluuid);
4671 *exp = class_conn2export(&conn);
4673 spin_lock(&osd->od_osfs_lock);
4675 spin_unlock(&osd->od_osfs_lock);
4681 * once last export (we don't count self-export) disappeared
4682 * osd can be released
4684 static int osd_obd_disconnect(struct obd_export *exp)
4686 struct obd_device *obd = exp->exp_obd;
4687 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4688 int rc, release = 0;
4691 /* Only disconnect the underlying layers on the final disconnect. */
4692 spin_lock(&osd->od_osfs_lock);
4694 if (osd->od_connects == 0)
4696 spin_unlock(&osd->od_osfs_lock);
4698 rc = class_disconnect(exp); /* bz 9811 */
4700 if (rc == 0 && release)
4701 class_manual_cleanup(obd);
4705 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4706 struct lu_device *dev)
4708 struct osd_device *osd = osd_dev(dev);
4712 if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
4713 /* MDT/MDD still use old infrastructure to create
4715 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4721 if (osd->od_quota_slave != NULL)
4722 /* set up quota slave objects */
4723 result = qsd_prepare(env, osd->od_quota_slave);
4728 static const struct lu_object_operations osd_lu_obj_ops = {
4729 .loo_object_init = osd_object_init,
4730 .loo_object_delete = osd_object_delete,
4731 .loo_object_release = osd_object_release,
4732 .loo_object_free = osd_object_free,
4733 .loo_object_print = osd_object_print,
4734 .loo_object_invariant = osd_object_invariant
4737 const struct lu_device_operations osd_lu_ops = {
4738 .ldo_object_alloc = osd_object_alloc,
4739 .ldo_process_config = osd_process_config,
4740 .ldo_recovery_complete = osd_recovery_complete,
4741 .ldo_prepare = osd_prepare,
4744 static const struct lu_device_type_operations osd_device_type_ops = {
4745 .ldto_init = osd_type_init,
4746 .ldto_fini = osd_type_fini,
4748 .ldto_start = osd_type_start,
4749 .ldto_stop = osd_type_stop,
4751 .ldto_device_alloc = osd_device_alloc,
4752 .ldto_device_free = osd_device_free,
4754 .ldto_device_init = osd_device_init,
4755 .ldto_device_fini = osd_device_fini
4758 struct lu_device_type osd_device_type = {
4759 .ldt_tags = LU_DEVICE_DT,
4760 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
4761 .ldt_ops = &osd_device_type_ops,
4762 .ldt_ctx_tags = LCT_LOCAL,
4766 * lprocfs legacy support.
4768 static struct obd_ops osd_obd_device_ops = {
4769 .o_owner = THIS_MODULE,
4770 .o_connect = osd_obd_connect,
4771 .o_disconnect = osd_obd_disconnect
4774 static int __init osd_mod_init(void)
4776 struct lprocfs_static_vars lvars;
4779 lprocfs_osd_init_vars(&lvars);
4780 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4781 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4784 static void __exit osd_mod_exit(void)
4786 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4789 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4790 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4791 MODULE_LICENSE("GPL");
4793 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);