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 /* dt_acct_features */
78 #ifdef HAVE_LDISKFS_PDO
80 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
81 "ldiskfs with parallel directory operations");
86 static const char dot[] = ".";
87 static const char dotdot[] = "..";
88 static const char remote_obj_dir[] = "REM_OBJ_DIR";
90 static const struct lu_object_operations osd_lu_obj_ops;
91 static const struct dt_object_operations osd_obj_ops;
92 static const struct dt_object_operations osd_obj_ea_ops;
93 static const struct dt_object_operations osd_obj_otable_it_ops;
94 static const struct dt_index_operations osd_index_iam_ops;
95 static const struct dt_index_operations osd_index_ea_ops;
97 static int osd_has_index(const struct osd_object *obj)
99 return obj->oo_dt.do_index_ops != NULL;
102 static int osd_object_invariant(const struct lu_object *l)
104 return osd_invariant(osd_obj(l));
108 * Concurrency: doesn't matter
110 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
112 return osd_oti_get(env)->oti_r_locks > 0;
116 * Concurrency: doesn't matter
118 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
120 struct osd_thread_info *oti = osd_oti_get(env);
121 return oti->oti_w_locks > 0 && o->oo_owner == env;
125 * Concurrency: doesn't access mutable data
127 static int osd_root_get(const struct lu_env *env,
128 struct dt_device *dev, struct lu_fid *f)
130 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
135 * OSD object methods.
139 * Concurrency: no concurrent access is possible that early in object
142 static struct lu_object *osd_object_alloc(const struct lu_env *env,
143 const struct lu_object_header *hdr,
146 struct osd_object *mo;
152 l = &mo->oo_dt.do_lu;
153 dt_object_init(&mo->oo_dt, NULL, d);
154 if (osd_dev(d)->od_iop_mode)
155 mo->oo_dt.do_ops = &osd_obj_ea_ops;
157 mo->oo_dt.do_ops = &osd_obj_ops;
159 l->lo_ops = &osd_lu_obj_ops;
160 cfs_init_rwsem(&mo->oo_sem);
161 cfs_init_rwsem(&mo->oo_ext_idx_sem);
162 cfs_spin_lock_init(&mo->oo_guard);
169 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
170 struct lustre_mdt_attrs *lma)
174 dentry->d_inode = inode;
175 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
178 /* Check LMA compatibility */
179 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
180 CWARN("%.16s: unsupported incompat LMA feature(s) "
182 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
183 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
187 lustre_lma_swab(lma);
190 } else if (rc == 0) {
198 * retrieve object from backend ext fs.
200 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
201 struct osd_inode_id *id)
203 struct inode *inode = NULL;
205 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
207 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
208 id->oii_ino, PTR_ERR(inode));
209 } else if (id->oii_gen != OSD_OII_NOGEN &&
210 inode->i_generation != id->oii_gen) {
211 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
213 id->oii_ino, id->oii_gen, inode->i_generation);
215 inode = ERR_PTR(-ESTALE);
216 } else if (inode->i_nlink == 0) {
217 /* due to parallel readdir and unlink,
218 * we can have dead inode here. */
219 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
220 make_bad_inode(inode);
222 inode = ERR_PTR(-ESTALE);
223 } else if (is_bad_inode(inode)) {
224 CWARN("%.16s: bad inode: ino = %u\n",
225 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
227 inode = ERR_PTR(-ENOENT);
229 if (id->oii_gen == OSD_OII_NOGEN)
230 osd_id_gen(id, inode->i_ino, inode->i_generation);
232 /* Do not update file c/mtime in ldiskfs.
233 * NB: we don't have any lock to protect this because we don't
234 * have reference on osd_object now, but contention with
235 * another lookup + attr_set can't happen in the tiny window
236 * between if (...) and set S_NOCMTIME. */
237 if (!(inode->i_flags & S_NOCMTIME))
238 inode->i_flags |= S_NOCMTIME;
243 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
244 struct osd_inode_id *id, struct lu_fid *fid)
246 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
250 inode = osd_iget(info, dev, id);
254 rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
256 *fid = lma->lma_self_fid;
257 } else if (rc == -ENODATA) {
258 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
266 static struct inode *
267 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
268 struct osd_inode_id *id, const struct lu_fid *fid)
270 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
274 inode = osd_iget(info, dev, id);
278 rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
280 if (rc == -ENODATA) {
281 CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
282 inode->i_ino, PFID(fid));
289 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
290 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
291 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
293 return ERR_PTR(EREMCHG);
298 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
299 const struct lu_fid *fid,
300 const struct lu_object_conf *conf)
302 struct osd_thread_info *info;
303 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
304 struct osd_device *dev;
305 struct osd_idmap_cache *oic;
306 struct osd_inode_id *id;
308 struct osd_scrub *scrub;
309 struct scrub_file *sf;
314 LINVRNT(osd_invariant(obj));
315 LASSERT(obj->oo_inode == NULL);
316 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
319 scrub = &dev->od_scrub;
320 sf = &scrub->os_file;
321 info = osd_oti_get(env);
323 oic = &info->oti_cache;
326 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
329 if (fid_is_norm(fid)) {
330 /* Search order: 1. per-thread cache. */
331 if (lu_fid_eq(fid, &oic->oic_fid)) {
333 } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
334 /* Search order: 2. OI scrub pending list. */
335 result = osd_oii_lookup(dev, fid, id);
340 if (sf->sf_flags & SF_INCONSISTENT)
345 * Objects are created as locking anchors or place holders for objects
346 * yet to be created. No need to osd_oi_lookup() at here because FID
347 * shouldn't never be re-used, if it's really a duplicate FID from
348 * unexpected reason, we should be able to detect it later by calling
349 * do_create->osd_oi_insert()
351 if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
352 GOTO(out, result = 0);
354 /* Search order: 3. OI files. */
355 result = osd_oi_lookup(info, dev, fid, id);
356 if (result == -ENOENT) {
357 if (!fid_is_norm(fid) ||
358 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
360 GOTO(out, result = 0);
370 inode = osd_iget(info, dev, id);
372 inode = osd_iget_verify(info, dev, id, fid);
374 result = PTR_ERR(inode);
375 if (result == -ENOENT || result == -ESTALE) {
376 fid_zero(&oic->oic_fid);
378 } else if (result == -EREMCHG) {
381 if (thread_is_running(&scrub->os_thread)) {
382 result = -EINPROGRESS;
383 } else if (!scrub->os_no_scrub) {
384 result = osd_scrub_start(dev);
385 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
386 "for "DFID", rc = %d [1]\n",
387 LDISKFS_SB(osd_sb(dev))->s_es->\
388 s_volume_name,PFID(fid), result);
389 if (result == 0 || result == -EALREADY)
390 result = -EINPROGRESS;
399 obj->oo_inode = inode;
400 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
401 if (dev->od_iop_mode) {
402 obj->oo_compat_dot_created = 1;
403 obj->oo_compat_dotdot_created = 1;
406 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
407 GOTO(out, result = 0);
409 LASSERT(obj->oo_hl_head == NULL);
410 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
411 if (obj->oo_hl_head == NULL) {
412 obj->oo_inode = NULL;
414 GOTO(out, result = -ENOMEM);
416 GOTO(out, result = 0);
419 LINVRNT(osd_invariant(obj));
424 * Concurrency: shouldn't matter.
426 static void osd_object_init0(struct osd_object *obj)
428 LASSERT(obj->oo_inode != NULL);
429 obj->oo_dt.do_body_ops = &osd_body_ops;
430 obj->oo_dt.do_lu.lo_header->loh_attr |=
431 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
435 * Concurrency: no concurrent access is possible that early in object
438 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
439 const struct lu_object_conf *conf)
441 struct osd_object *obj = osd_obj(l);
444 LINVRNT(osd_invariant(obj));
446 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
447 obj->oo_dt.do_body_ops = &osd_body_ops_new;
449 if (obj->oo_inode != NULL) {
450 osd_object_init0(obj);
451 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
452 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
453 /* LFSCK iterator object is special without inode */
454 l->lo_header->loh_attr |= LOHA_EXISTS;
457 LINVRNT(osd_invariant(obj));
462 * Concurrency: no concurrent access is possible that late in object
465 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
467 struct osd_object *obj = osd_obj(l);
469 LINVRNT(osd_invariant(obj));
471 dt_object_fini(&obj->oo_dt);
472 if (obj->oo_hl_head != NULL)
473 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
478 * Concurrency: no concurrent access is possible that late in object
481 static void osd_index_fini(struct osd_object *o)
483 struct iam_container *bag;
485 if (o->oo_dir != NULL) {
486 bag = &o->oo_dir->od_container;
487 if (o->oo_inode != NULL) {
488 if (bag->ic_object == o->oo_inode)
489 iam_container_fini(bag);
491 OBD_FREE_PTR(o->oo_dir);
497 * Concurrency: no concurrent access is possible that late in object
498 * life-cycle (for all existing callers, that is. New callers have to provide
499 * their own locking.)
501 static int osd_inode_unlinked(const struct inode *inode)
503 return inode->i_nlink == 0;
507 OSD_TXN_OI_DELETE_CREDITS = 20,
508 OSD_TXN_INODE_DELETE_CREDITS = 20
515 #if OSD_THANDLE_STATS
517 * Set time when the handle is allocated
519 static void osd_th_alloced(struct osd_thandle *oth)
521 oth->oth_alloced = cfs_time_current();
525 * Set time when the handle started
527 static void osd_th_started(struct osd_thandle *oth)
529 oth->oth_started = cfs_time_current();
533 * Helper function to convert time interval to microseconds packed in
534 * long int (default time units for the counter in "stats" initialized
535 * by lu_time_init() )
537 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
541 cfs_duration_usec(cfs_time_sub(end, start), &val);
542 return val.tv_sec * 1000000 + val.tv_usec;
546 * Check whether the we deal with this handle for too long.
548 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
549 cfs_time_t alloced, cfs_time_t started,
552 cfs_time_t now = cfs_time_current();
554 LASSERT(dev != NULL);
556 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
557 interval_to_usec(alloced, started));
558 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
559 interval_to_usec(started, closed));
560 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
561 interval_to_usec(closed, now));
563 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
564 CWARN("transaction handle %p was open for too long: "
566 "alloced "CFS_TIME_T" ,"
567 "started "CFS_TIME_T" ,"
568 "closed "CFS_TIME_T"\n",
569 oth, now, alloced, started, closed);
570 libcfs_debug_dumpstack(NULL);
574 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
576 cfs_time_t __closed = cfs_time_current(); \
577 cfs_time_t __alloced = oth->oth_alloced; \
578 cfs_time_t __started = oth->oth_started; \
581 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
584 #else /* OSD_THANDLE_STATS */
586 #define osd_th_alloced(h) do {} while(0)
587 #define osd_th_started(h) do {} while(0)
588 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
590 #endif /* OSD_THANDLE_STATS */
593 * Concurrency: doesn't access mutable data.
595 static int osd_param_is_sane(const struct osd_device *dev,
596 const struct thandle *th)
598 struct osd_thandle *oh;
599 oh = container_of0(th, struct osd_thandle, ot_super);
600 return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
604 * Concurrency: shouldn't matter.
606 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
607 static void osd_trans_commit_cb(struct super_block *sb,
608 struct journal_callback *jcb, int error)
610 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
613 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
614 struct thandle *th = &oh->ot_super;
615 struct lu_device *lud = &th->th_dev->dd_lu_dev;
616 struct dt_txn_commit_cb *dcb, *tmp;
618 LASSERT(oh->ot_handle == NULL);
621 CERROR("transaction @0x%p commit error: %d\n", th, error);
623 dt_txn_hook_commit(th);
625 /* call per-transaction callbacks if any */
626 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
627 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
628 "commit callback entry: magic=%x name='%s'\n",
629 dcb->dcb_magic, dcb->dcb_name);
630 cfs_list_del_init(&dcb->dcb_linkage);
631 dcb->dcb_func(NULL, th, dcb, error);
634 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
638 lu_context_exit(&th->th_ctx);
639 lu_context_fini(&th->th_ctx);
643 static struct thandle *osd_trans_create(const struct lu_env *env,
646 struct osd_thread_info *oti = osd_oti_get(env);
647 struct osd_iobuf *iobuf = &oti->oti_iobuf;
648 struct osd_thandle *oh;
652 /* on pending IO in this thread should left from prev. request */
653 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
655 th = ERR_PTR(-ENOMEM);
656 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
658 oh->ot_quota_trans = &oti->oti_quota_trans;
659 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
663 th->th_tags = LCT_TX_HANDLE;
665 oti->oti_dev = osd_dt_dev(d);
666 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
673 * Concurrency: shouldn't matter.
675 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
678 struct osd_thread_info *oti = osd_oti_get(env);
679 struct osd_device *dev = osd_dt_dev(d);
681 struct osd_thandle *oh;
686 LASSERT(current->journal_info == NULL);
688 oh = container_of0(th, struct osd_thandle, ot_super);
690 LASSERT(oh->ot_handle == NULL);
692 rc = dt_txn_hook_start(env, d, th);
696 if (!osd_param_is_sane(dev, th)) {
697 CWARN("%.16s: too many transaction credits (%d > %d)\n",
698 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
700 osd_journal(dev)->j_max_transaction_buffers);
701 /* XXX Limit the credits to 'max_transaction_buffers', and
702 * let the underlying filesystem to catch the error if
703 * we really need so many credits.
705 * This should be removed when we can calculate the
706 * credits precisely. */
707 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
708 #ifdef OSD_TRACK_DECLARES
709 CERROR(" attr_set: %d, punch: %d, xattr_set: %d,\n",
710 oh->ot_declare_attr_set, oh->ot_declare_punch,
711 oh->ot_declare_xattr_set);
712 CERROR(" create: %d, ref_add: %d, ref_del: %d, write: %d\n",
713 oh->ot_declare_create, oh->ot_declare_ref_add,
714 oh->ot_declare_ref_del, oh->ot_declare_write);
715 CERROR(" insert: %d, delete: %d, destroy: %d\n",
716 oh->ot_declare_insert, oh->ot_declare_delete,
717 oh->ot_declare_destroy);
722 * XXX temporary stuff. Some abstraction layer should
725 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
729 LASSERT(oti->oti_txns == 0);
730 lu_context_init(&th->th_ctx, th->th_tags);
731 lu_context_enter(&th->th_ctx);
733 lu_device_get(&d->dd_lu_dev);
734 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
746 * Concurrency: shouldn't matter.
748 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
751 struct osd_thandle *oh;
752 struct osd_thread_info *oti = osd_oti_get(env);
753 struct osd_iobuf *iobuf = &oti->oti_iobuf;
754 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
757 oh = container_of0(th, struct osd_thandle, ot_super);
760 /* inform the quota slave device that the transaction is
762 qsd_op_end(env, qsd, oh->ot_quota_trans);
763 oh->ot_quota_trans = NULL;
765 if (oh->ot_handle != NULL) {
766 handle_t *hdl = oh->ot_handle;
769 * add commit callback
770 * notice we don't do this in osd_trans_start()
771 * as underlying transaction can change during truncate
773 osd_journal_callback_set(hdl, osd_trans_commit_cb,
776 LASSERT(oti->oti_txns == 1);
778 rc = dt_txn_hook_stop(env, th);
780 CERROR("Failure in transaction hook: %d\n", rc);
782 /* hook functions might modify th_sync */
783 hdl->h_sync = th->th_sync;
785 oh->ot_handle = NULL;
786 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
787 rc = ldiskfs_journal_stop(hdl));
789 CERROR("Failure to stop transaction: %d\n", rc);
794 /* as we want IO to journal and data IO be concurrent, we don't block
795 * awaiting data IO completion in osd_do_bio(), instead we wait here
796 * once transaction is submitted to the journal. all reqular requests
797 * don't do direct IO (except read/write), thus this wait_event becomes
800 * IMPORTANT: we have to wait till any IO submited by the thread is
801 * completed otherwise iobuf may be corrupted by different request
803 cfs_wait_event(iobuf->dr_wait,
804 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
806 rc = iobuf->dr_error;
811 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
813 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
816 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
817 LASSERT(&dcb->dcb_func != NULL);
818 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
824 * Called just before object is freed. Releases all resources except for
825 * object itself (that is released by osd_object_free()).
827 * Concurrency: no concurrent access is possible that late in object
830 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
832 struct osd_object *obj = osd_obj(l);
833 struct inode *inode = obj->oo_inode;
835 LINVRNT(osd_invariant(obj));
838 * If object is unlinked remove fid->ino mapping from object index.
844 obj->oo_inode = NULL;
849 * Concurrency: ->loo_object_release() is called under site spin-lock.
851 static void osd_object_release(const struct lu_env *env,
857 * Concurrency: shouldn't matter.
859 static int osd_object_print(const struct lu_env *env, void *cookie,
860 lu_printer_t p, const struct lu_object *l)
862 struct osd_object *o = osd_obj(l);
865 if (o->oo_dir != NULL)
866 d = o->oo_dir->od_container.ic_descr;
869 return (*p)(env, cookie,
870 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
872 o->oo_inode ? o->oo_inode->i_ino : 0UL,
873 o->oo_inode ? o->oo_inode->i_generation : 0,
874 d ? d->id_ops->id_name : "plain");
878 * Concurrency: shouldn't matter.
880 int osd_statfs(const struct lu_env *env, struct dt_device *d,
881 struct obd_statfs *sfs)
883 struct osd_device *osd = osd_dt_dev(d);
884 struct super_block *sb = osd_sb(osd);
885 struct kstatfs *ksfs;
888 if (unlikely(osd->od_mnt == NULL))
891 /* osd_lproc.c call this without env, allocate ksfs for that case */
892 if (unlikely(env == NULL)) {
897 ksfs = &osd_oti_get(env)->oti_ksfs;
900 cfs_spin_lock(&osd->od_osfs_lock);
902 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
903 result = sb->s_op->statfs(sb->s_root, ksfs);
904 if (likely(result == 0)) { /* N.B. statfs can't really fail */
905 osd->od_osfs_age = cfs_time_current_64();
906 statfs_pack(&osd->od_statfs, ksfs);
907 if (sb->s_flags & MS_RDONLY)
908 sfs->os_state = OS_STATE_READONLY;
912 if (likely(result == 0))
913 *sfs = osd->od_statfs;
914 cfs_spin_unlock(&osd->od_osfs_lock);
916 if (unlikely(env == NULL))
923 * Concurrency: doesn't access mutable data.
925 static void osd_conf_get(const struct lu_env *env,
926 const struct dt_device *dev,
927 struct dt_device_param *param)
929 struct super_block *sb = osd_sb(osd_dt_dev(dev));
932 * XXX should be taken from not-yet-existing fs abstraction layer.
934 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
935 param->ddp_max_name_len = LDISKFS_NAME_LEN;
936 param->ddp_max_nlink = LDISKFS_LINK_MAX;
937 param->ddp_block_shift = sb->s_blocksize_bits;
938 param->ddp_mount_type = LDD_MT_LDISKFS;
939 param->ddp_mntopts = 0;
940 if (test_opt(sb, XATTR_USER))
941 param->ddp_mntopts |= MNTOPT_USERXATTR;
942 if (test_opt(sb, POSIX_ACL))
943 param->ddp_mntopts |= MNTOPT_ACL;
945 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
946 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
947 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
950 param->ddp_max_ea_size = sb->s_blocksize;
955 * Helper function to get and fill the buffer with input values.
957 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
961 buf = &osd_oti_get(env)->oti_buf;
968 * Concurrency: shouldn't matter.
970 static int osd_sync(const struct lu_env *env, struct dt_device *d)
972 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
973 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
977 * Start commit for OSD device.
979 * An implementation of dt_commit_async method for OSD device.
980 * Asychronously starts underlayng fs sync and thereby a transaction
983 * \param env environment
986 * \see dt_device_operations
988 static int osd_commit_async(const struct lu_env *env,
991 struct super_block *s = osd_sb(osd_dt_dev(d));
994 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
995 RETURN(s->s_op->sync_fs(s, 0));
999 * Concurrency: shouldn't matter.
1002 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1004 struct super_block *sb = osd_sb(osd_dt_dev(d));
1008 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1010 rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1015 * Concurrency: serialization provided by callers.
1017 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1018 int mode, unsigned long timeout, __u32 alg,
1019 struct lustre_capa_key *keys)
1021 struct osd_device *dev = osd_dt_dev(d);
1024 dev->od_fl_capa = mode;
1025 dev->od_capa_timeout = timeout;
1026 dev->od_capa_alg = alg;
1027 dev->od_capa_keys = keys;
1032 * Note: we do not count into QUOTA here.
1033 * If we mount with --data_journal we may need more.
1035 const int osd_dto_credits_noquota[DTO_NR] = {
1038 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1039 * SINGLEDATA_TRANS_BLOCKS(8)
1040 * XXX Note: maybe iam need more, since iam have more level than
1043 [DTO_INDEX_INSERT] = 16,
1044 [DTO_INDEX_DELETE] = 16,
1048 [DTO_INDEX_UPDATE] = 16,
1050 * Create a object. The same as create object in EXT3.
1051 * DATA_TRANS_BLOCKS(14) +
1052 * INDEX_EXTRA_BLOCKS(8) +
1053 * 3(inode bits, groups, GDT)
1055 [DTO_OBJECT_CREATE] = 25,
1057 * XXX: real credits to be fixed
1059 [DTO_OBJECT_DELETE] = 25,
1061 * Attr set credits (inode)
1063 [DTO_ATTR_SET_BASE] = 1,
1065 * Xattr set. The same as xattr of EXT3.
1066 * DATA_TRANS_BLOCKS(14)
1067 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1068 * are also counted in. Do not know why?
1070 [DTO_XATTR_SET] = 14,
1073 * credits for inode change during write.
1075 [DTO_WRITE_BASE] = 3,
1077 * credits for single block write.
1079 [DTO_WRITE_BLOCK] = 14,
1081 * Attr set credits for chown.
1082 * This is extra credits for setattr, and it is null without quota
1084 [DTO_ATTR_SET_CHOWN]= 0
1087 static const struct dt_device_operations osd_dt_ops = {
1088 .dt_root_get = osd_root_get,
1089 .dt_statfs = osd_statfs,
1090 .dt_trans_create = osd_trans_create,
1091 .dt_trans_start = osd_trans_start,
1092 .dt_trans_stop = osd_trans_stop,
1093 .dt_trans_cb_add = osd_trans_cb_add,
1094 .dt_conf_get = osd_conf_get,
1095 .dt_sync = osd_sync,
1097 .dt_commit_async = osd_commit_async,
1098 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1101 static void osd_object_read_lock(const struct lu_env *env,
1102 struct dt_object *dt, unsigned role)
1104 struct osd_object *obj = osd_dt_obj(dt);
1105 struct osd_thread_info *oti = osd_oti_get(env);
1107 LINVRNT(osd_invariant(obj));
1109 LASSERT(obj->oo_owner != env);
1110 cfs_down_read_nested(&obj->oo_sem, role);
1112 LASSERT(obj->oo_owner == NULL);
1116 static void osd_object_write_lock(const struct lu_env *env,
1117 struct dt_object *dt, unsigned role)
1119 struct osd_object *obj = osd_dt_obj(dt);
1120 struct osd_thread_info *oti = osd_oti_get(env);
1122 LINVRNT(osd_invariant(obj));
1124 LASSERT(obj->oo_owner != env);
1125 cfs_down_write_nested(&obj->oo_sem, role);
1127 LASSERT(obj->oo_owner == NULL);
1128 obj->oo_owner = env;
1132 static void osd_object_read_unlock(const struct lu_env *env,
1133 struct dt_object *dt)
1135 struct osd_object *obj = osd_dt_obj(dt);
1136 struct osd_thread_info *oti = osd_oti_get(env);
1138 LINVRNT(osd_invariant(obj));
1140 LASSERT(oti->oti_r_locks > 0);
1142 cfs_up_read(&obj->oo_sem);
1145 static void osd_object_write_unlock(const struct lu_env *env,
1146 struct dt_object *dt)
1148 struct osd_object *obj = osd_dt_obj(dt);
1149 struct osd_thread_info *oti = osd_oti_get(env);
1151 LINVRNT(osd_invariant(obj));
1153 LASSERT(obj->oo_owner == env);
1154 LASSERT(oti->oti_w_locks > 0);
1156 obj->oo_owner = NULL;
1157 cfs_up_write(&obj->oo_sem);
1160 static int osd_object_write_locked(const struct lu_env *env,
1161 struct dt_object *dt)
1163 struct osd_object *obj = osd_dt_obj(dt);
1165 LINVRNT(osd_invariant(obj));
1167 return obj->oo_owner == env;
1170 static int capa_is_sane(const struct lu_env *env,
1171 struct osd_device *dev,
1172 struct lustre_capa *capa,
1173 struct lustre_capa_key *keys)
1175 struct osd_thread_info *oti = osd_oti_get(env);
1176 struct lustre_capa *tcapa = &oti->oti_capa;
1177 struct obd_capa *oc;
1181 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1183 if (capa_is_expired(oc)) {
1184 DEBUG_CAPA(D_ERROR, capa, "expired");
1191 if (capa_is_expired_sec(capa)) {
1192 DEBUG_CAPA(D_ERROR, capa, "expired");
1196 cfs_spin_lock(&capa_lock);
1197 for (i = 0; i < 2; i++) {
1198 if (keys[i].lk_keyid == capa->lc_keyid) {
1199 oti->oti_capa_key = keys[i];
1203 cfs_spin_unlock(&capa_lock);
1206 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1210 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1214 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1215 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1219 oc = capa_add(dev->od_capa_hash, capa);
1225 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1226 struct lustre_capa *capa, __u64 opc)
1228 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1229 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1230 struct md_capainfo *ci;
1233 if (!dev->od_fl_capa)
1236 if (capa == BYPASS_CAPA)
1239 ci = md_capainfo(env);
1243 if (ci->mc_auth == LC_ID_NONE)
1247 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1251 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1252 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1257 if (!capa_opc_supported(capa, opc)) {
1258 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1262 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1263 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1270 static struct timespec *osd_inode_time(const struct lu_env *env,
1271 struct inode *inode, __u64 seconds)
1273 struct osd_thread_info *oti = osd_oti_get(env);
1274 struct timespec *t = &oti->oti_time;
1276 t->tv_sec = seconds;
1278 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1283 static void osd_inode_getattr(const struct lu_env *env,
1284 struct inode *inode, struct lu_attr *attr)
1286 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1287 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1288 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1290 attr->la_atime = LTIME_S(inode->i_atime);
1291 attr->la_mtime = LTIME_S(inode->i_mtime);
1292 attr->la_ctime = LTIME_S(inode->i_ctime);
1293 attr->la_mode = inode->i_mode;
1294 attr->la_size = i_size_read(inode);
1295 attr->la_blocks = inode->i_blocks;
1296 attr->la_uid = inode->i_uid;
1297 attr->la_gid = inode->i_gid;
1298 attr->la_flags = LDISKFS_I(inode)->i_flags;
1299 attr->la_nlink = inode->i_nlink;
1300 attr->la_rdev = inode->i_rdev;
1301 attr->la_blksize = 1 << inode->i_blkbits;
1302 attr->la_blkbits = inode->i_blkbits;
1305 static int osd_attr_get(const struct lu_env *env,
1306 struct dt_object *dt,
1307 struct lu_attr *attr,
1308 struct lustre_capa *capa)
1310 struct osd_object *obj = osd_dt_obj(dt);
1312 LASSERT(dt_object_exists(dt));
1313 LINVRNT(osd_invariant(obj));
1315 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1318 cfs_spin_lock(&obj->oo_guard);
1319 osd_inode_getattr(env, obj->oo_inode, attr);
1320 cfs_spin_unlock(&obj->oo_guard);
1324 static int osd_declare_attr_set(const struct lu_env *env,
1325 struct dt_object *dt,
1326 const struct lu_attr *attr,
1327 struct thandle *handle)
1329 struct osd_thandle *oh;
1330 struct osd_object *obj;
1331 struct osd_thread_info *info = osd_oti_get(env);
1332 struct lquota_id_info *qi = &info->oti_qi;
1338 LASSERT(dt != NULL);
1339 LASSERT(handle != NULL);
1341 obj = osd_dt_obj(dt);
1342 LASSERT(osd_invariant(obj));
1344 oh = container_of0(handle, struct osd_thandle, ot_super);
1345 LASSERT(oh->ot_handle == NULL);
1347 OSD_DECLARE_OP(oh, attr_set);
1348 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1350 if (attr == NULL || obj->oo_inode == NULL)
1353 bspace = obj->oo_inode->i_blocks;
1354 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1355 bspace = toqb(bspace);
1357 /* Changing ownership is always preformed by super user, it should not
1360 * We still need to call the osd_declare_qid() to calculate the journal
1361 * credits for updating quota accounting files and to trigger quota
1362 * space adjustment once the operation is completed.*/
1363 if ((attr->la_valid & LA_UID) != 0 &&
1364 attr->la_uid != obj->oo_inode->i_uid) {
1365 qi->lqi_type = USRQUOTA;
1367 /* inode accounting */
1368 qi->lqi_is_blk = false;
1370 /* one more inode for the new owner ... */
1371 qi->lqi_id.qid_uid = attr->la_uid;
1373 allocated = (attr->la_uid == 0) ? true : false;
1374 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1375 if (rc == -EDQUOT || rc == -EINPROGRESS)
1380 /* and one less inode for the current uid */
1381 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1383 rc = osd_declare_qid(env, oh, qi, true, NULL);
1384 if (rc == -EDQUOT || rc == -EINPROGRESS)
1389 /* block accounting */
1390 qi->lqi_is_blk = true;
1392 /* more blocks for the new owner ... */
1393 qi->lqi_id.qid_uid = attr->la_uid;
1394 qi->lqi_space = bspace;
1395 allocated = (attr->la_uid == 0) ? true : false;
1396 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1397 if (rc == -EDQUOT || rc == -EINPROGRESS)
1402 /* and finally less blocks for the current owner */
1403 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1404 qi->lqi_space = -bspace;
1405 rc = osd_declare_qid(env, oh, qi, true, NULL);
1406 if (rc == -EDQUOT || rc == -EINPROGRESS)
1412 if (attr->la_valid & LA_GID &&
1413 attr->la_gid != obj->oo_inode->i_gid) {
1414 qi->lqi_type = GRPQUOTA;
1416 /* inode accounting */
1417 qi->lqi_is_blk = false;
1419 /* one more inode for the new group owner ... */
1420 qi->lqi_id.qid_gid = attr->la_gid;
1422 allocated = (attr->la_gid == 0) ? true : false;
1423 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1424 if (rc == -EDQUOT || rc == -EINPROGRESS)
1429 /* and one less inode for the current gid */
1430 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1432 rc = osd_declare_qid(env, oh, qi, true, NULL);
1433 if (rc == -EDQUOT || rc == -EINPROGRESS)
1438 /* block accounting */
1439 qi->lqi_is_blk = true;
1441 /* more blocks for the new owner ... */
1442 qi->lqi_id.qid_gid = attr->la_gid;
1443 qi->lqi_space = bspace;
1444 allocated = (attr->la_gid == 0) ? true : false;
1445 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1446 if (rc == -EDQUOT || rc == -EINPROGRESS)
1451 /* and finally less blocks for the current owner */
1452 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1453 qi->lqi_space = -bspace;
1454 rc = osd_declare_qid(env, oh, qi, true, NULL);
1455 if (rc == -EDQUOT || rc == -EINPROGRESS)
1464 static int osd_inode_setattr(const struct lu_env *env,
1465 struct inode *inode, const struct lu_attr *attr)
1469 bits = attr->la_valid;
1471 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1473 if (bits & LA_ATIME)
1474 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1475 if (bits & LA_CTIME)
1476 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1477 if (bits & LA_MTIME)
1478 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1479 if (bits & LA_SIZE) {
1480 LDISKFS_I(inode)->i_disksize = attr->la_size;
1481 i_size_write(inode, attr->la_size);
1485 /* OSD should not change "i_blocks" which is used by quota.
1486 * "i_blocks" should be changed by ldiskfs only. */
1487 if (bits & LA_BLOCKS)
1488 inode->i_blocks = attr->la_blocks;
1491 inode->i_mode = (inode->i_mode & S_IFMT) |
1492 (attr->la_mode & ~S_IFMT);
1494 inode->i_uid = attr->la_uid;
1496 inode->i_gid = attr->la_gid;
1497 if (bits & LA_NLINK)
1498 inode->i_nlink = attr->la_nlink;
1500 inode->i_rdev = attr->la_rdev;
1502 if (bits & LA_FLAGS) {
1503 /* always keep S_NOCMTIME */
1504 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1510 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1512 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1513 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1518 if (attr->la_valid & LA_UID)
1519 iattr.ia_valid |= ATTR_UID;
1520 if (attr->la_valid & LA_GID)
1521 iattr.ia_valid |= ATTR_GID;
1522 iattr.ia_uid = attr->la_uid;
1523 iattr.ia_gid = attr->la_gid;
1525 rc = ll_vfs_dq_transfer(inode, &iattr);
1527 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1528 "enforcement enabled on the ldiskfs filesystem?",
1529 inode->i_sb->s_id, rc);
1536 static int osd_attr_set(const struct lu_env *env,
1537 struct dt_object *dt,
1538 const struct lu_attr *attr,
1539 struct thandle *handle,
1540 struct lustre_capa *capa)
1542 struct osd_object *obj = osd_dt_obj(dt);
1543 struct inode *inode;
1546 LASSERT(handle != NULL);
1547 LASSERT(dt_object_exists(dt));
1548 LASSERT(osd_invariant(obj));
1550 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1553 OSD_EXEC_OP(handle, attr_set);
1555 inode = obj->oo_inode;
1557 rc = osd_quota_transfer(inode, attr);
1561 cfs_spin_lock(&obj->oo_guard);
1562 rc = osd_inode_setattr(env, inode, attr);
1563 cfs_spin_unlock(&obj->oo_guard);
1566 inode->i_sb->s_op->dirty_inode(inode);
1570 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1571 struct osd_object *obj,
1572 const char *name, const int namelen)
1574 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1577 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1579 struct dt_allocation_hint *hint,
1583 struct osd_device *osd = osd_obj2dev(obj);
1584 struct osd_thandle *oth;
1585 struct dt_object *parent = NULL;
1586 struct inode *inode;
1588 LINVRNT(osd_invariant(obj));
1589 LASSERT(obj->oo_inode == NULL);
1590 LASSERT(obj->oo_hl_head == NULL);
1592 if (S_ISDIR(mode) && ldiskfs_pdo) {
1593 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1594 if (obj->oo_hl_head == NULL)
1598 oth = container_of(th, struct osd_thandle, ot_super);
1599 LASSERT(oth->ot_handle->h_transaction != NULL);
1601 if (hint && hint->dah_parent)
1602 parent = hint->dah_parent;
1604 inode = ldiskfs_create_inode(oth->ot_handle,
1605 parent ? osd_dt_obj(parent)->oo_inode :
1606 osd_sb(osd)->s_root->d_inode,
1608 if (!IS_ERR(inode)) {
1609 /* Do not update file c/mtime in ldiskfs.
1610 * NB: don't need any lock because no contention at this
1612 inode->i_flags |= S_NOCMTIME;
1613 inode->i_state |= I_LUSTRE_NOSCRUB;
1614 obj->oo_inode = inode;
1617 if (obj->oo_hl_head != NULL) {
1618 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1619 obj->oo_hl_head = NULL;
1621 result = PTR_ERR(inode);
1623 LINVRNT(osd_invariant(obj));
1631 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1632 struct lu_attr *attr,
1633 struct dt_allocation_hint *hint,
1634 struct dt_object_format *dof,
1638 struct osd_thandle *oth;
1639 struct osd_device *osd = osd_obj2dev(obj);
1640 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1642 LASSERT(S_ISDIR(attr->la_mode));
1644 oth = container_of(th, struct osd_thandle, ot_super);
1645 LASSERT(oth->ot_handle->h_transaction != NULL);
1646 result = osd_mkfile(info, obj, mode, hint, th);
1647 if (result == 0 && osd->od_iop_mode == 0) {
1648 LASSERT(obj->oo_inode != NULL);
1650 * XXX uh-oh... call low-level iam function directly.
1653 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1654 sizeof (struct osd_fid_pack),
1660 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1661 struct lu_attr *attr,
1662 struct dt_allocation_hint *hint,
1663 struct dt_object_format *dof,
1667 struct osd_thandle *oth;
1668 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1670 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1672 LASSERT(S_ISREG(attr->la_mode));
1674 oth = container_of(th, struct osd_thandle, ot_super);
1675 LASSERT(oth->ot_handle->h_transaction != NULL);
1677 result = osd_mkfile(info, obj, mode, hint, th);
1679 LASSERT(obj->oo_inode != NULL);
1680 if (feat->dif_flags & DT_IND_VARKEY)
1681 result = iam_lvar_create(obj->oo_inode,
1682 feat->dif_keysize_max,
1684 feat->dif_recsize_max,
1687 result = iam_lfix_create(obj->oo_inode,
1688 feat->dif_keysize_max,
1690 feat->dif_recsize_max,
1697 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1698 struct lu_attr *attr,
1699 struct dt_allocation_hint *hint,
1700 struct dt_object_format *dof,
1703 LASSERT(S_ISREG(attr->la_mode));
1704 return osd_mkfile(info, obj, (attr->la_mode &
1705 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1708 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1709 struct lu_attr *attr,
1710 struct dt_allocation_hint *hint,
1711 struct dt_object_format *dof,
1714 LASSERT(S_ISLNK(attr->la_mode));
1715 return osd_mkfile(info, obj, (attr->la_mode &
1716 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1719 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1720 struct lu_attr *attr,
1721 struct dt_allocation_hint *hint,
1722 struct dt_object_format *dof,
1725 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1728 LINVRNT(osd_invariant(obj));
1729 LASSERT(obj->oo_inode == NULL);
1730 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1731 S_ISFIFO(mode) || S_ISSOCK(mode));
1733 result = osd_mkfile(info, obj, mode, hint, th);
1735 LASSERT(obj->oo_inode != NULL);
1737 * This inode should be marked dirty for i_rdev. Currently
1738 * that is done in the osd_attr_init().
1740 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1742 LINVRNT(osd_invariant(obj));
1746 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1748 struct dt_allocation_hint *hint,
1749 struct dt_object_format *dof,
1752 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1754 osd_obj_type_f result;
1770 result = osd_mk_index;
1781 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1782 struct dt_object *parent, struct dt_object *child,
1783 cfs_umode_t child_mode)
1787 memset(ah, 0, sizeof(*ah));
1788 ah->dah_parent = parent;
1789 ah->dah_mode = child_mode;
1792 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1793 struct lu_attr *attr, struct dt_object_format *dof)
1795 struct inode *inode = obj->oo_inode;
1796 __u64 valid = attr->la_valid;
1799 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1801 if (dof->dof_type != DFT_NODE)
1802 attr->la_valid &= ~LA_RDEV;
1803 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1804 attr->la_valid &= ~LA_ATIME;
1805 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1806 attr->la_valid &= ~LA_CTIME;
1807 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1808 attr->la_valid &= ~LA_MTIME;
1810 result = osd_quota_transfer(inode, attr);
1814 if (attr->la_valid != 0) {
1815 result = osd_inode_setattr(info->oti_env, inode, attr);
1817 * The osd_inode_setattr() should always succeed here. The
1818 * only error that could be returned is EDQUOT when we are
1819 * trying to change the UID or GID of the inode. However, this
1820 * should not happen since quota enforcement is no longer
1821 * enabled on ldiskfs (lquota takes care of it).
1823 LASSERTF(result == 0, "%d", result);
1824 inode->i_sb->s_op->dirty_inode(inode);
1827 attr->la_valid = valid;
1831 * Helper function for osd_object_create()
1833 * \retval 0, on success
1835 static int __osd_object_create(struct osd_thread_info *info,
1836 struct osd_object *obj, struct lu_attr *attr,
1837 struct dt_allocation_hint *hint,
1838 struct dt_object_format *dof,
1844 /* we drop umask so that permissions we pass are not affected */
1845 umask = current->fs->umask;
1846 current->fs->umask = 0;
1848 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1851 osd_attr_init(info, obj, attr, dof);
1852 osd_object_init0(obj);
1854 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1855 unlock_new_inode(obj->oo_inode);
1858 /* restore previous umask value */
1859 current->fs->umask = umask;
1865 * Helper function for osd_object_create()
1867 * \retval 0, on success
1869 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1870 const struct lu_fid *fid, struct thandle *th)
1872 struct osd_thread_info *info = osd_oti_get(env);
1873 struct osd_inode_id *id = &info->oti_id;
1874 struct osd_device *osd = osd_obj2dev(obj);
1876 LASSERT(obj->oo_inode != NULL);
1878 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1879 return osd_oi_insert(info, osd, fid, id, th);
1882 static int osd_declare_object_create(const struct lu_env *env,
1883 struct dt_object *dt,
1884 struct lu_attr *attr,
1885 struct dt_allocation_hint *hint,
1886 struct dt_object_format *dof,
1887 struct thandle *handle)
1889 struct osd_thandle *oh;
1893 LASSERT(handle != NULL);
1895 oh = container_of0(handle, struct osd_thandle, ot_super);
1896 LASSERT(oh->ot_handle == NULL);
1898 OSD_DECLARE_OP(oh, create);
1899 oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1900 /* XXX: So far, only normal fid needs be inserted into the oi,
1901 * things could be changed later. Revise following code then. */
1902 if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1903 OSD_DECLARE_OP(oh, insert);
1904 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1905 /* Reuse idle OI block may cause additional one OI block
1907 oh->ot_credits += 1;
1909 /* If this is directory, then we expect . and .. to be inserted as
1910 * well. The one directory block always needs to be created for the
1911 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1912 * block), there is no danger of needing a tree for the first block.
1914 if (attr && S_ISDIR(attr->la_mode)) {
1915 OSD_DECLARE_OP(oh, insert);
1916 OSD_DECLARE_OP(oh, insert);
1917 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1923 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
1924 false, false, NULL, false);
1928 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1929 struct lu_attr *attr,
1930 struct dt_allocation_hint *hint,
1931 struct dt_object_format *dof,
1934 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1935 struct osd_object *obj = osd_dt_obj(dt);
1936 struct osd_thread_info *info = osd_oti_get(env);
1941 LINVRNT(osd_invariant(obj));
1942 LASSERT(!dt_object_exists(dt));
1943 LASSERT(osd_write_locked(env, obj));
1944 LASSERT(th != NULL);
1946 if (unlikely(fid_is_acct(fid)))
1947 /* Quota files can't be created from the kernel any more,
1948 * 'tune2fs -O quota' will take care of creating them */
1951 OSD_EXEC_OP(th, create);
1953 result = __osd_object_create(info, obj, attr, hint, dof, th);
1955 result = __osd_oi_insert(env, obj, fid, th);
1957 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1958 LASSERT(osd_invariant(obj));
1963 * Called to destroy on-disk representation of the object
1965 * Concurrency: must be locked
1967 static int osd_declare_object_destroy(const struct lu_env *env,
1968 struct dt_object *dt,
1971 struct osd_object *obj = osd_dt_obj(dt);
1972 struct inode *inode = obj->oo_inode;
1973 struct osd_thandle *oh;
1977 oh = container_of0(th, struct osd_thandle, ot_super);
1978 LASSERT(oh->ot_handle == NULL);
1981 OSD_DECLARE_OP(oh, destroy);
1982 OSD_DECLARE_OP(oh, delete);
1983 oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
1984 /* XXX: So far, only normal fid needs to be inserted into the OI,
1985 * so only normal fid needs to be removed from the OI also. */
1986 if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1987 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
1988 /* Recycle idle OI leaf may cause additional three OI blocks
1990 oh->ot_credits += 3;
1993 /* one less inode */
1994 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
1995 false, true, NULL, false);
1998 /* data to be truncated */
1999 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true,
2004 static int osd_object_destroy(const struct lu_env *env,
2005 struct dt_object *dt,
2008 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2009 struct osd_object *obj = osd_dt_obj(dt);
2010 struct inode *inode = obj->oo_inode;
2011 struct osd_device *osd = osd_obj2dev(obj);
2012 struct osd_thandle *oh;
2016 oh = container_of0(th, struct osd_thandle, ot_super);
2017 LASSERT(oh->ot_handle);
2019 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2021 if (unlikely(fid_is_acct(fid)))
2024 /* Parallel control for OI scrub. For most of cases, there is no
2025 * lock contention. So it will not affect unlink performance. */
2026 cfs_mutex_lock(&inode->i_mutex);
2027 if (S_ISDIR(inode->i_mode)) {
2028 LASSERT(osd_inode_unlinked(inode) ||
2029 inode->i_nlink == 1);
2030 cfs_spin_lock(&obj->oo_guard);
2032 cfs_spin_unlock(&obj->oo_guard);
2033 inode->i_sb->s_op->dirty_inode(inode);
2035 LASSERT(osd_inode_unlinked(inode));
2038 OSD_EXEC_OP(th, destroy);
2040 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2041 cfs_mutex_unlock(&inode->i_mutex);
2043 /* XXX: add to ext3 orphan list */
2044 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2046 /* not needed in the cache anymore */
2047 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2053 * Helper function for osd_xattr_set()
2055 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2056 const struct lu_buf *buf, const char *name, int fl)
2058 struct osd_object *obj = osd_dt_obj(dt);
2059 struct inode *inode = obj->oo_inode;
2060 struct osd_thread_info *info = osd_oti_get(env);
2061 struct dentry *dentry = &info->oti_child_dentry;
2065 LASSERT(dt_object_exists(dt));
2066 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2068 if (fl & LU_XATTR_REPLACE)
2069 fs_flags |= XATTR_REPLACE;
2071 if (fl & LU_XATTR_CREATE)
2072 fs_flags |= XATTR_CREATE;
2074 dentry->d_inode = inode;
2075 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2076 buf->lb_len, fs_flags);
2081 * Put the fid into lustre_mdt_attrs, and then place the structure
2082 * inode's ea. This fid should not be altered during the life time
2085 * \retval +ve, on success
2086 * \retval -ve, on error
2088 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2090 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2091 const struct lu_fid *fid)
2093 struct osd_thread_info *info = osd_oti_get(env);
2094 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2096 lustre_lma_init(mdt_attrs, fid);
2097 lustre_lma_swab(mdt_attrs);
2098 return __osd_xattr_set(env, dt,
2099 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2100 XATTR_NAME_LMA, LU_XATTR_CREATE);
2105 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2106 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2107 * To have compatilibility with 1.8 ldiskfs driver we need to have
2108 * magic number at start of fid data.
2109 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2112 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2113 const struct dt_rec *fid)
2115 param->edp_magic = LDISKFS_LUFID_MAGIC;
2116 param->edp_len = sizeof(struct lu_fid) + 1;
2118 fid_cpu_to_be((struct lu_fid *)param->edp_data,
2119 (struct lu_fid *)fid);
2123 * Try to read the fid from inode ea into dt_rec, if return value
2124 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2126 * \param fid object fid.
2128 * \retval 0 on success
2130 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2131 __u32 ino, struct lu_fid *fid,
2132 struct osd_inode_id *id)
2134 struct osd_thread_info *info = osd_oti_get(env);
2135 struct inode *inode;
2138 osd_id_gen(id, ino, OSD_OII_NOGEN);
2139 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2141 RETURN(PTR_ERR(inode));
2148 * OSD layer object create function for interoperability mode (b11826).
2149 * This is mostly similar to osd_object_create(). Only difference being, fid is
2150 * inserted into inode ea here.
2152 * \retval 0, on success
2153 * \retval -ve, on error
2155 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2156 struct lu_attr *attr,
2157 struct dt_allocation_hint *hint,
2158 struct dt_object_format *dof,
2161 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2162 struct osd_object *obj = osd_dt_obj(dt);
2163 struct osd_thread_info *info = osd_oti_get(env);
2168 LASSERT(osd_invariant(obj));
2169 LASSERT(!dt_object_exists(dt));
2170 LASSERT(osd_write_locked(env, obj));
2171 LASSERT(th != NULL);
2173 if (unlikely(fid_is_acct(fid)))
2174 /* Quota files can't be created from the kernel any more,
2175 * 'tune2fs -O quota' will take care of creating them */
2178 OSD_EXEC_OP(th, create);
2180 result = __osd_object_create(info, obj, attr, hint, dof, th);
2181 /* objects under osd root shld have igif fid, so dont add fid EA */
2182 if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2183 result = osd_ea_fid_set(env, dt, fid);
2186 result = __osd_oi_insert(env, obj, fid, th);
2188 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2189 LINVRNT(osd_invariant(obj));
2193 static int osd_declare_object_ref_add(const struct lu_env *env,
2194 struct dt_object *dt,
2195 struct thandle *handle)
2197 struct osd_thandle *oh;
2199 /* it's possible that object doesn't exist yet */
2200 LASSERT(handle != NULL);
2202 oh = container_of0(handle, struct osd_thandle, ot_super);
2203 LASSERT(oh->ot_handle == NULL);
2205 OSD_DECLARE_OP(oh, ref_add);
2206 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2212 * Concurrency: @dt is write locked.
2214 static int osd_object_ref_add(const struct lu_env *env,
2215 struct dt_object *dt, struct thandle *th)
2217 struct osd_object *obj = osd_dt_obj(dt);
2218 struct inode *inode = obj->oo_inode;
2220 LINVRNT(osd_invariant(obj));
2221 LASSERT(dt_object_exists(dt));
2222 LASSERT(osd_write_locked(env, obj));
2223 LASSERT(th != NULL);
2225 OSD_EXEC_OP(th, ref_add);
2228 * DIR_NLINK feature is set for compatibility reasons if:
2229 * 1) nlinks > LDISKFS_LINK_MAX, or
2230 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2232 * It is easier to always set this flag (rather than check and set),
2233 * since it has less overhead, and the superblock will be dirtied
2234 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2235 * do not actually care whether this flag is set or not.
2237 cfs_spin_lock(&obj->oo_guard);
2239 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2240 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2241 inode->i_nlink == 2)
2244 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2245 cfs_spin_unlock(&obj->oo_guard);
2246 inode->i_sb->s_op->dirty_inode(inode);
2247 LINVRNT(osd_invariant(obj));
2252 static int osd_declare_object_ref_del(const struct lu_env *env,
2253 struct dt_object *dt,
2254 struct thandle *handle)
2256 struct osd_thandle *oh;
2258 LASSERT(dt_object_exists(dt));
2259 LASSERT(handle != NULL);
2261 oh = container_of0(handle, struct osd_thandle, ot_super);
2262 LASSERT(oh->ot_handle == NULL);
2264 OSD_DECLARE_OP(oh, ref_del);
2265 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2271 * Concurrency: @dt is write locked.
2273 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2276 struct osd_object *obj = osd_dt_obj(dt);
2277 struct inode *inode = obj->oo_inode;
2279 LINVRNT(osd_invariant(obj));
2280 LASSERT(dt_object_exists(dt));
2281 LASSERT(osd_write_locked(env, obj));
2282 LASSERT(th != NULL);
2284 OSD_EXEC_OP(th, ref_del);
2286 cfs_spin_lock(&obj->oo_guard);
2287 LASSERT(inode->i_nlink > 0);
2289 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2290 * then the nlink count is 1. Don't let it be set to 0 or the directory
2291 * inode will be deleted incorrectly. */
2292 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2294 cfs_spin_unlock(&obj->oo_guard);
2295 inode->i_sb->s_op->dirty_inode(inode);
2296 LINVRNT(osd_invariant(obj));
2302 * Get the 64-bit version for an inode.
2304 static int osd_object_version_get(const struct lu_env *env,
2305 struct dt_object *dt, dt_obj_version_t *ver)
2307 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2309 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2310 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2311 *ver = LDISKFS_I(inode)->i_fs_version;
2316 * Concurrency: @dt is read locked.
2318 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2319 struct lu_buf *buf, const char *name,
2320 struct lustre_capa *capa)
2322 struct osd_object *obj = osd_dt_obj(dt);
2323 struct inode *inode = obj->oo_inode;
2324 struct osd_thread_info *info = osd_oti_get(env);
2325 struct dentry *dentry = &info->oti_obj_dentry;
2327 /* version get is not real XATTR but uses xattr API */
2328 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2329 /* for version we are just using xattr API but change inode
2331 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2332 osd_object_version_get(env, dt, buf->lb_buf);
2333 return sizeof(dt_obj_version_t);
2336 LASSERT(dt_object_exists(dt));
2337 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2339 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2342 dentry->d_inode = inode;
2343 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2347 static int osd_declare_xattr_set(const struct lu_env *env,
2348 struct dt_object *dt,
2349 const struct lu_buf *buf, const char *name,
2350 int fl, struct thandle *handle)
2352 struct osd_thandle *oh;
2354 LASSERT(handle != NULL);
2356 oh = container_of0(handle, struct osd_thandle, ot_super);
2357 LASSERT(oh->ot_handle == NULL);
2359 OSD_DECLARE_OP(oh, xattr_set);
2360 if (strcmp(name, XATTR_NAME_VERSION) == 0)
2361 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2363 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2369 * Set the 64-bit version for object
2371 static void osd_object_version_set(const struct lu_env *env,
2372 struct dt_object *dt,
2373 dt_obj_version_t *new_version)
2375 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2377 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2378 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2380 LDISKFS_I(inode)->i_fs_version = *new_version;
2381 /** Version is set after all inode operations are finished,
2382 * so we should mark it dirty here */
2383 inode->i_sb->s_op->dirty_inode(inode);
2387 * Concurrency: @dt is write locked.
2389 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2390 const struct lu_buf *buf, const char *name, int fl,
2391 struct thandle *handle, struct lustre_capa *capa)
2393 LASSERT(handle != NULL);
2395 /* version set is not real XATTR */
2396 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2397 /* for version we are just using xattr API but change inode
2399 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2400 osd_object_version_set(env, dt, buf->lb_buf);
2401 return sizeof(dt_obj_version_t);
2404 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2407 OSD_EXEC_OP(handle, xattr_set);
2408 return __osd_xattr_set(env, dt, buf, name, fl);
2412 * Concurrency: @dt is read locked.
2414 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2415 struct lu_buf *buf, struct lustre_capa *capa)
2417 struct osd_object *obj = osd_dt_obj(dt);
2418 struct inode *inode = obj->oo_inode;
2419 struct osd_thread_info *info = osd_oti_get(env);
2420 struct dentry *dentry = &info->oti_obj_dentry;
2422 LASSERT(dt_object_exists(dt));
2423 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2424 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2426 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2429 dentry->d_inode = inode;
2430 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2433 static int osd_declare_xattr_del(const struct lu_env *env,
2434 struct dt_object *dt, const char *name,
2435 struct thandle *handle)
2437 struct osd_thandle *oh;
2439 LASSERT(dt_object_exists(dt));
2440 LASSERT(handle != NULL);
2442 oh = container_of0(handle, struct osd_thandle, ot_super);
2443 LASSERT(oh->ot_handle == NULL);
2445 OSD_DECLARE_OP(oh, xattr_set);
2446 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2452 * Concurrency: @dt is write locked.
2454 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2455 const char *name, struct thandle *handle,
2456 struct lustre_capa *capa)
2458 struct osd_object *obj = osd_dt_obj(dt);
2459 struct inode *inode = obj->oo_inode;
2460 struct osd_thread_info *info = osd_oti_get(env);
2461 struct dentry *dentry = &info->oti_obj_dentry;
2464 LASSERT(dt_object_exists(dt));
2465 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2466 LASSERT(osd_write_locked(env, obj));
2467 LASSERT(handle != NULL);
2469 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2472 OSD_EXEC_OP(handle, xattr_set);
2474 dentry->d_inode = inode;
2475 rc = inode->i_op->removexattr(dentry, name);
2479 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2480 struct dt_object *dt,
2481 struct lustre_capa *old,
2484 struct osd_thread_info *info = osd_oti_get(env);
2485 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2486 struct osd_object *obj = osd_dt_obj(dt);
2487 struct osd_device *dev = osd_obj2dev(obj);
2488 struct lustre_capa_key *key = &info->oti_capa_key;
2489 struct lustre_capa *capa = &info->oti_capa;
2490 struct obd_capa *oc;
2491 struct md_capainfo *ci;
2495 if (!dev->od_fl_capa)
2496 RETURN(ERR_PTR(-ENOENT));
2498 LASSERT(dt_object_exists(dt));
2499 LINVRNT(osd_invariant(obj));
2501 /* renewal sanity check */
2502 if (old && osd_object_auth(env, dt, old, opc))
2503 RETURN(ERR_PTR(-EACCES));
2505 ci = md_capainfo(env);
2507 RETURN(ERR_PTR(-ENOENT));
2509 switch (ci->mc_auth) {
2513 capa->lc_uid = obj->oo_inode->i_uid;
2514 capa->lc_gid = obj->oo_inode->i_gid;
2515 capa->lc_flags = LC_ID_PLAIN;
2517 case LC_ID_CONVERT: {
2520 s[0] = obj->oo_inode->i_uid;
2521 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2522 s[2] = obj->oo_inode->i_gid;
2523 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2524 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2526 RETURN(ERR_PTR(rc));
2528 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2529 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2530 capa->lc_flags = LC_ID_CONVERT;
2534 RETURN(ERR_PTR(-EINVAL));
2537 capa->lc_fid = *fid;
2539 capa->lc_flags |= dev->od_capa_alg << 24;
2540 capa->lc_timeout = dev->od_capa_timeout;
2541 capa->lc_expiry = 0;
2543 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2545 LASSERT(!capa_is_expired(oc));
2549 cfs_spin_lock(&capa_lock);
2550 *key = dev->od_capa_keys[1];
2551 cfs_spin_unlock(&capa_lock);
2553 capa->lc_keyid = key->lk_keyid;
2554 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2556 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2558 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2559 RETURN(ERR_PTR(rc));
2562 oc = capa_add(dev->od_capa_hash, capa);
2566 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2568 struct osd_object *obj = osd_dt_obj(dt);
2569 struct inode *inode = obj->oo_inode;
2570 struct osd_thread_info *info = osd_oti_get(env);
2571 struct dentry *dentry = &info->oti_obj_dentry;
2572 struct file *file = &info->oti_file;
2577 dentry->d_inode = inode;
2578 file->f_dentry = dentry;
2579 file->f_mapping = inode->i_mapping;
2580 file->f_op = inode->i_fop;
2581 mutex_lock(&inode->i_mutex);
2582 rc = file->f_op->fsync(file, dentry, 0);
2583 mutex_unlock(&inode->i_mutex);
2587 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2590 struct osd_object *obj = osd_dt_obj(dt);
2593 *data = (void *)obj->oo_inode;
2601 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2602 const struct dt_index_features *feat)
2604 struct iam_descr *descr;
2606 if (osd_object_is_root(o))
2607 return feat == &dt_directory_features;
2609 LASSERT(o->oo_dir != NULL);
2611 descr = o->oo_dir->od_container.ic_descr;
2612 if (feat == &dt_directory_features) {
2613 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2619 feat->dif_keysize_min <= descr->id_key_size &&
2620 descr->id_key_size <= feat->dif_keysize_max &&
2621 feat->dif_recsize_min <= descr->id_rec_size &&
2622 descr->id_rec_size <= feat->dif_recsize_max &&
2623 !(feat->dif_flags & (DT_IND_VARKEY |
2624 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2625 ergo(feat->dif_flags & DT_IND_UPDATE,
2626 1 /* XXX check that object (and file system) is
2631 static int osd_iam_container_init(const struct lu_env *env,
2632 struct osd_object *obj,
2633 struct osd_directory *dir)
2635 struct iam_container *bag = &dir->od_container;
2638 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2642 result = iam_container_setup(bag);
2644 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2646 iam_container_fini(bag);
2653 * Concurrency: no external locking is necessary.
2655 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2656 const struct dt_index_features *feat)
2660 struct osd_object *obj = osd_dt_obj(dt);
2661 struct osd_device *osd = osd_obj2dev(obj);
2663 LINVRNT(osd_invariant(obj));
2664 LASSERT(dt_object_exists(dt));
2666 if (osd_object_is_root(obj)) {
2667 dt->do_index_ops = &osd_index_ea_ops;
2669 } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2670 dt->do_index_ops = &osd_index_ea_ops;
2671 if (S_ISDIR(obj->oo_inode->i_mode))
2676 } else if (unlikely(feat == &dt_otable_features)) {
2677 dt->do_index_ops = &osd_otable_ops;
2679 } else if (feat == &dt_acct_features) {
2680 dt->do_index_ops = &osd_acct_index_ops;
2683 } else if (!osd_has_index(obj)) {
2684 struct osd_directory *dir;
2689 cfs_spin_lock(&obj->oo_guard);
2690 if (obj->oo_dir == NULL)
2694 * Concurrent thread allocated container data.
2697 cfs_spin_unlock(&obj->oo_guard);
2699 * Now, that we have container data, serialize its
2702 cfs_down_write(&obj->oo_ext_idx_sem);
2704 * recheck under lock.
2706 if (!osd_has_index(obj))
2707 result = osd_iam_container_init(env, obj, dir);
2710 cfs_up_write(&obj->oo_ext_idx_sem);
2718 if (result == 0 && skip_iam == 0) {
2719 if (!osd_iam_index_probe(env, obj, feat))
2722 LINVRNT(osd_invariant(obj));
2724 if (is_quota_glb_feat(feat))
2725 result = osd_quota_migration(env, dt, feat);
2730 static int osd_otable_it_attr_get(const struct lu_env *env,
2731 struct dt_object *dt,
2732 struct lu_attr *attr,
2733 struct lustre_capa *capa)
2739 static const struct dt_object_operations osd_obj_ops = {
2740 .do_read_lock = osd_object_read_lock,
2741 .do_write_lock = osd_object_write_lock,
2742 .do_read_unlock = osd_object_read_unlock,
2743 .do_write_unlock = osd_object_write_unlock,
2744 .do_write_locked = osd_object_write_locked,
2745 .do_attr_get = osd_attr_get,
2746 .do_declare_attr_set = osd_declare_attr_set,
2747 .do_attr_set = osd_attr_set,
2748 .do_ah_init = osd_ah_init,
2749 .do_declare_create = osd_declare_object_create,
2750 .do_create = osd_object_create,
2751 .do_declare_destroy = osd_declare_object_destroy,
2752 .do_destroy = osd_object_destroy,
2753 .do_index_try = osd_index_try,
2754 .do_declare_ref_add = osd_declare_object_ref_add,
2755 .do_ref_add = osd_object_ref_add,
2756 .do_declare_ref_del = osd_declare_object_ref_del,
2757 .do_ref_del = osd_object_ref_del,
2758 .do_xattr_get = osd_xattr_get,
2759 .do_declare_xattr_set = osd_declare_xattr_set,
2760 .do_xattr_set = osd_xattr_set,
2761 .do_declare_xattr_del = osd_declare_xattr_del,
2762 .do_xattr_del = osd_xattr_del,
2763 .do_xattr_list = osd_xattr_list,
2764 .do_capa_get = osd_capa_get,
2765 .do_object_sync = osd_object_sync,
2766 .do_data_get = osd_data_get,
2770 * dt_object_operations for interoperability mode
2771 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2773 static const struct dt_object_operations osd_obj_ea_ops = {
2774 .do_read_lock = osd_object_read_lock,
2775 .do_write_lock = osd_object_write_lock,
2776 .do_read_unlock = osd_object_read_unlock,
2777 .do_write_unlock = osd_object_write_unlock,
2778 .do_write_locked = osd_object_write_locked,
2779 .do_attr_get = osd_attr_get,
2780 .do_declare_attr_set = osd_declare_attr_set,
2781 .do_attr_set = osd_attr_set,
2782 .do_ah_init = osd_ah_init,
2783 .do_declare_create = osd_declare_object_create,
2784 .do_create = osd_object_ea_create,
2785 .do_declare_destroy = osd_declare_object_destroy,
2786 .do_destroy = osd_object_destroy,
2787 .do_index_try = osd_index_try,
2788 .do_declare_ref_add = osd_declare_object_ref_add,
2789 .do_ref_add = osd_object_ref_add,
2790 .do_declare_ref_del = osd_declare_object_ref_del,
2791 .do_ref_del = osd_object_ref_del,
2792 .do_xattr_get = osd_xattr_get,
2793 .do_declare_xattr_set = osd_declare_xattr_set,
2794 .do_xattr_set = osd_xattr_set,
2795 .do_declare_xattr_del = osd_declare_xattr_del,
2796 .do_xattr_del = osd_xattr_del,
2797 .do_xattr_list = osd_xattr_list,
2798 .do_capa_get = osd_capa_get,
2799 .do_object_sync = osd_object_sync,
2800 .do_data_get = osd_data_get,
2803 static const struct dt_object_operations osd_obj_otable_it_ops = {
2804 .do_attr_get = osd_otable_it_attr_get,
2805 .do_index_try = osd_index_try,
2808 static int osd_index_declare_iam_delete(const struct lu_env *env,
2809 struct dt_object *dt,
2810 const struct dt_key *key,
2811 struct thandle *handle)
2813 struct osd_thandle *oh;
2815 oh = container_of0(handle, struct osd_thandle, ot_super);
2816 LASSERT(oh->ot_handle == NULL);
2818 OSD_DECLARE_OP(oh, delete);
2819 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2825 * delete a (key, value) pair from index \a dt specified by \a key
2827 * \param dt osd index object
2828 * \param key key for index
2829 * \param rec record reference
2830 * \param handle transaction handler
2833 * \retval -ve failure
2836 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2837 const struct dt_key *key,
2838 struct thandle *handle,
2839 struct lustre_capa *capa)
2841 struct osd_thread_info *oti = osd_oti_get(env);
2842 struct osd_object *obj = osd_dt_obj(dt);
2843 struct osd_thandle *oh;
2844 struct iam_path_descr *ipd;
2845 struct iam_container *bag = &obj->oo_dir->od_container;
2850 LINVRNT(osd_invariant(obj));
2851 LASSERT(dt_object_exists(dt));
2852 LASSERT(bag->ic_object == obj->oo_inode);
2853 LASSERT(handle != NULL);
2855 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2858 OSD_EXEC_OP(handle, delete);
2860 ipd = osd_idx_ipd_get(env, bag);
2861 if (unlikely(ipd == NULL))
2864 oh = container_of0(handle, struct osd_thandle, ot_super);
2865 LASSERT(oh->ot_handle != NULL);
2866 LASSERT(oh->ot_handle->h_transaction != NULL);
2868 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
2869 /* swab quota uid/gid provided by caller */
2870 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
2871 key = (const struct dt_key *)&oti->oti_quota_id;
2874 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2875 osd_ipd_put(env, bag, ipd);
2876 LINVRNT(osd_invariant(obj));
2880 static int osd_index_declare_ea_delete(const struct lu_env *env,
2881 struct dt_object *dt,
2882 const struct dt_key *key,
2883 struct thandle *handle)
2885 struct osd_thandle *oh;
2886 struct inode *inode;
2890 LASSERT(dt_object_exists(dt));
2891 LASSERT(handle != NULL);
2893 oh = container_of0(handle, struct osd_thandle, ot_super);
2894 LASSERT(oh->ot_handle == NULL);
2896 OSD_DECLARE_OP(oh, delete);
2897 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2899 inode = osd_dt_obj(dt)->oo_inode;
2902 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2903 true, true, NULL, false);
2907 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2910 struct osd_fid_pack *rec;
2913 if (de->file_type & LDISKFS_DIRENT_LUFID) {
2914 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2915 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2921 * Index delete function for interoperability mode (b11826).
2922 * It will remove the directory entry added by osd_index_ea_insert().
2923 * This entry is needed to maintain name->fid mapping.
2925 * \param key, key i.e. file entry to be deleted
2927 * \retval 0, on success
2928 * \retval -ve, on error
2930 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2931 const struct dt_key *key,
2932 struct thandle *handle,
2933 struct lustre_capa *capa)
2935 struct osd_object *obj = osd_dt_obj(dt);
2936 struct inode *dir = obj->oo_inode;
2937 struct dentry *dentry;
2938 struct osd_thandle *oh;
2939 struct ldiskfs_dir_entry_2 *de;
2940 struct buffer_head *bh;
2941 struct htree_lock *hlock = NULL;
2946 LINVRNT(osd_invariant(obj));
2947 LASSERT(dt_object_exists(dt));
2948 LASSERT(handle != NULL);
2950 OSD_EXEC_OP(handle, delete);
2952 oh = container_of(handle, struct osd_thandle, ot_super);
2953 LASSERT(oh->ot_handle != NULL);
2954 LASSERT(oh->ot_handle->h_transaction != NULL);
2956 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2959 dentry = osd_child_dentry_get(env, obj,
2960 (char *)key, strlen((char *)key));
2962 if (obj->oo_hl_head != NULL) {
2963 hlock = osd_oti_get(env)->oti_hlock;
2964 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
2965 dir, LDISKFS_HLOCK_DEL);
2967 cfs_down_write(&obj->oo_ext_idx_sem);
2970 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
2972 rc = ldiskfs_delete_entry(oh->ot_handle,
2979 ldiskfs_htree_unlock(hlock);
2981 cfs_up_write(&obj->oo_ext_idx_sem);
2983 LASSERT(osd_invariant(obj));
2988 * Lookup index for \a key and copy record to \a rec.
2990 * \param dt osd index object
2991 * \param key key for index
2992 * \param rec record reference
2994 * \retval +ve success : exact mach
2995 * \retval 0 return record with key not greater than \a key
2996 * \retval -ve failure
2998 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2999 struct dt_rec *rec, const struct dt_key *key,
3000 struct lustre_capa *capa)
3002 struct osd_object *obj = osd_dt_obj(dt);
3003 struct iam_path_descr *ipd;
3004 struct iam_container *bag = &obj->oo_dir->od_container;
3005 struct osd_thread_info *oti = osd_oti_get(env);
3006 struct iam_iterator *it = &oti->oti_idx_it;
3007 struct iam_rec *iam_rec;
3012 LASSERT(osd_invariant(obj));
3013 LASSERT(dt_object_exists(dt));
3014 LASSERT(bag->ic_object == obj->oo_inode);
3016 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3019 ipd = osd_idx_ipd_get(env, bag);
3023 /* got ipd now we can start iterator. */
3024 iam_it_init(it, bag, 0, ipd);
3026 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3027 /* swab quota uid/gid provided by caller */
3028 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3029 key = (const struct dt_key *)&oti->oti_quota_id;
3032 rc = iam_it_get(it, (struct iam_key *)key);
3034 if (S_ISDIR(obj->oo_inode->i_mode))
3035 iam_rec = (struct iam_rec *)oti->oti_ldp;
3037 iam_rec = (struct iam_rec *) rec;
3039 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3041 if (S_ISDIR(obj->oo_inode->i_mode))
3042 osd_fid_unpack((struct lu_fid *) rec,
3043 (struct osd_fid_pack *)iam_rec);
3044 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3045 osd_quota_unpack(obj, rec);
3050 osd_ipd_put(env, bag, ipd);
3052 LINVRNT(osd_invariant(obj));
3057 static int osd_index_declare_iam_insert(const struct lu_env *env,
3058 struct dt_object *dt,
3059 const struct dt_rec *rec,
3060 const struct dt_key *key,
3061 struct thandle *handle)
3063 struct osd_thandle *oh;
3065 LASSERT(dt_object_exists(dt));
3066 LASSERT(handle != NULL);
3068 oh = container_of0(handle, struct osd_thandle, ot_super);
3069 LASSERT(oh->ot_handle == NULL);
3071 OSD_DECLARE_OP(oh, insert);
3072 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3078 * Inserts (key, value) pair in \a dt index object.
3080 * \param dt osd index object
3081 * \param key key for index
3082 * \param rec record reference
3083 * \param th transaction handler
3086 * \retval -ve failure
3088 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3089 const struct dt_rec *rec,
3090 const struct dt_key *key, struct thandle *th,
3091 struct lustre_capa *capa, int ignore_quota)
3093 struct osd_object *obj = osd_dt_obj(dt);
3094 struct iam_path_descr *ipd;
3095 struct osd_thandle *oh;
3096 struct iam_container *bag = &obj->oo_dir->od_container;
3097 struct osd_thread_info *oti = osd_oti_get(env);
3098 struct iam_rec *iam_rec;
3103 LINVRNT(osd_invariant(obj));
3104 LASSERT(dt_object_exists(dt));
3105 LASSERT(bag->ic_object == obj->oo_inode);
3106 LASSERT(th != NULL);
3108 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3111 OSD_EXEC_OP(th, insert);
3113 ipd = osd_idx_ipd_get(env, bag);
3114 if (unlikely(ipd == NULL))
3117 oh = container_of0(th, struct osd_thandle, ot_super);
3118 LASSERT(oh->ot_handle != NULL);
3119 LASSERT(oh->ot_handle->h_transaction != NULL);
3120 if (S_ISDIR(obj->oo_inode->i_mode)) {
3121 iam_rec = (struct iam_rec *)oti->oti_ldp;
3122 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3123 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3124 /* pack quota uid/gid */
3125 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3126 key = (const struct dt_key *)&oti->oti_quota_id;
3127 /* pack quota record */
3128 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3129 iam_rec = (struct iam_rec *)rec;
3131 iam_rec = (struct iam_rec *)rec;
3134 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3136 osd_ipd_put(env, bag, ipd);
3137 LINVRNT(osd_invariant(obj));
3142 * Calls ldiskfs_add_entry() to add directory entry
3143 * into the directory. This is required for
3144 * interoperability mode (b11826)
3146 * \retval 0, on success
3147 * \retval -ve, on error
3149 static int __osd_ea_add_rec(struct osd_thread_info *info,
3150 struct osd_object *pobj, struct inode *cinode,
3151 const char *name, const struct dt_rec *fid,
3152 struct htree_lock *hlock, struct thandle *th)
3154 struct ldiskfs_dentry_param *ldp;
3155 struct dentry *child;
3156 struct osd_thandle *oth;
3159 oth = container_of(th, struct osd_thandle, ot_super);
3160 LASSERT(oth->ot_handle != NULL);
3161 LASSERT(oth->ot_handle->h_transaction != NULL);
3163 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3165 /* XXX: remove fid_is_igif() check here.
3166 * IGIF check is just to handle insertion of .. when it is 'ROOT',
3167 * it is IGIF now but needs FID in dir entry as well for readdir
3169 * LU-838 should fix that and remove fid_is_igif() check */
3170 if (fid_is_igif((struct lu_fid *)fid) ||
3171 fid_is_norm((struct lu_fid *)fid)) {
3172 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3173 osd_get_ldiskfs_dirent_param(ldp, fid);
3174 child->d_fsdata = (void *)ldp;
3176 child->d_fsdata = NULL;
3178 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3184 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3185 * into the directory.Also sets flags into osd object to
3186 * indicate dot and dotdot are created. This is required for
3187 * interoperability mode (b11826)
3189 * \param dir directory for dot and dotdot fixup.
3190 * \param obj child object for linking
3192 * \retval 0, on success
3193 * \retval -ve, on error
3195 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3196 struct osd_object *dir,
3197 struct inode *parent_dir, const char *name,
3198 const struct dt_rec *dot_fid,
3199 const struct dt_rec *dot_dot_fid,
3202 struct inode *inode = dir->oo_inode;
3203 struct ldiskfs_dentry_param *dot_ldp;
3204 struct ldiskfs_dentry_param *dot_dot_ldp;
3205 struct osd_thandle *oth;
3208 oth = container_of(th, struct osd_thandle, ot_super);
3209 LASSERT(oth->ot_handle->h_transaction != NULL);
3210 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3212 if (strcmp(name, dot) == 0) {
3213 if (dir->oo_compat_dot_created) {
3216 LASSERT(inode == parent_dir);
3217 dir->oo_compat_dot_created = 1;
3220 } else if(strcmp(name, dotdot) == 0) {
3221 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3222 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3224 if (!dir->oo_compat_dot_created)
3226 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3227 osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3228 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3233 /* in case of rename, dotdot is already created */
3234 if (dir->oo_compat_dotdot_created) {
3235 return __osd_ea_add_rec(info, dir, parent_dir, name,
3236 dot_dot_fid, NULL, th);
3239 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3240 inode, dot_ldp, dot_dot_ldp);
3242 dir->oo_compat_dotdot_created = 1;
3250 * It will call the appropriate osd_add* function and return the
3251 * value, return by respective functions.
3253 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3254 struct inode *cinode, const char *name,
3255 const struct dt_rec *fid, struct thandle *th)
3257 struct osd_thread_info *info = osd_oti_get(env);
3258 struct htree_lock *hlock;
3261 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3263 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3265 if (hlock != NULL) {
3266 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3269 cfs_down_write(&pobj->oo_ext_idx_sem);
3271 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3272 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3275 if (hlock != NULL) {
3276 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3277 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3279 cfs_down_write(&pobj->oo_ext_idx_sem);
3282 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3286 ldiskfs_htree_unlock(hlock);
3288 cfs_up_write(&pobj->oo_ext_idx_sem);
3294 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3295 struct osd_idmap_cache *oic)
3297 struct osd_scrub *scrub = &dev->od_scrub;
3298 struct lu_fid *fid = &oic->oic_fid;
3299 struct osd_inode_id *id = &oti->oti_id;
3305 rc = osd_oi_lookup(oti, dev, fid, id);
3306 if (rc != 0 && rc != -ENOENT)
3309 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3312 if (thread_is_running(&scrub->os_thread)) {
3313 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3314 /* There is race condition between osd_oi_lookup and OI scrub.
3315 * The OI scrub finished just after osd_oi_lookup() failure.
3316 * Under such case, it is unnecessary to trigger OI scrub again,
3317 * but try to call osd_oi_lookup() again. */
3318 if (unlikely(rc == -EAGAIN))
3324 if (!scrub->os_no_scrub && ++once == 1) {
3325 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3327 rc = osd_scrub_start(dev);
3328 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3330 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3336 RETURN(rc = -EREMCHG);
3340 * Calls ->lookup() to find dentry. From dentry get inode and
3341 * read inode's ea to get fid. This is required for interoperability
3344 * \retval 0, on success
3345 * \retval -ve, on error
3347 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3348 struct dt_rec *rec, const struct dt_key *key)
3350 struct inode *dir = obj->oo_inode;
3351 struct dentry *dentry;
3352 struct ldiskfs_dir_entry_2 *de;
3353 struct buffer_head *bh;
3354 struct lu_fid *fid = (struct lu_fid *) rec;
3355 struct htree_lock *hlock = NULL;
3359 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3361 dentry = osd_child_dentry_get(env, obj,
3362 (char *)key, strlen((char *)key));
3364 if (obj->oo_hl_head != NULL) {
3365 hlock = osd_oti_get(env)->oti_hlock;
3366 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3367 dir, LDISKFS_HLOCK_LOOKUP);
3369 cfs_down_read(&obj->oo_ext_idx_sem);
3372 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3374 struct osd_thread_info *oti = osd_oti_get(env);
3375 struct osd_idmap_cache *oic = &oti->oti_cache;
3376 struct osd_device *dev = osd_obj2dev(obj);
3377 struct osd_scrub *scrub = &dev->od_scrub;
3378 struct scrub_file *sf = &scrub->os_file;
3380 ino = le32_to_cpu(de->inode);
3381 rc = osd_get_fid_from_dentry(de, rec);
3383 /* done with de, release bh */
3386 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3388 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3389 if (rc != 0 || !fid_is_norm(fid)) {
3390 fid_zero(&oic->oic_fid);
3394 oic->oic_fid = *fid;
3395 if ((scrub->os_pos_current <= ino) &&
3396 (sf->sf_flags & SF_INCONSISTENT ||
3397 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3399 rc = osd_consistency_check(oti, dev, oic);
3408 ldiskfs_htree_unlock(hlock);
3410 cfs_up_read(&obj->oo_ext_idx_sem);
3415 * Find the osd object for given fid.
3417 * \param fid need to find the osd object having this fid
3419 * \retval osd_object on success
3420 * \retval -ve on error
3422 struct osd_object *osd_object_find(const struct lu_env *env,
3423 struct dt_object *dt,
3424 const struct lu_fid *fid)
3426 struct lu_device *ludev = dt->do_lu.lo_dev;
3427 struct osd_object *child = NULL;
3428 struct lu_object *luch;
3429 struct lu_object *lo;
3432 * at this point topdev might not exist yet
3433 * (i.e. MGS is preparing profiles). so we can
3434 * not rely on topdev and instead lookup with
3435 * our device passed as topdev. this can't work
3436 * if the object isn't cached yet (as osd doesn't
3437 * allocate lu_header). IOW, the object must be
3438 * in the cache, otherwise lu_object_alloc() crashes
3441 luch = lu_object_find_at(env, ludev, fid, NULL);
3442 if (!IS_ERR(luch)) {
3443 if (lu_object_exists(luch)) {
3444 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3446 child = osd_obj(lo);
3448 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3449 "lu_object can't be located"
3450 DFID"\n", PFID(fid));
3452 if (child == NULL) {
3453 lu_object_put(env, luch);
3454 CERROR("Unable to get osd_object\n");
3455 child = ERR_PTR(-ENOENT);
3458 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3459 "lu_object does not exists "DFID"\n",
3461 lu_object_put(env, luch);
3462 child = ERR_PTR(-ENOENT);
3465 child = (void *)luch;
3471 * Put the osd object once done with it.
3473 * \param obj osd object that needs to be put
3475 static inline void osd_object_put(const struct lu_env *env,
3476 struct osd_object *obj)
3478 lu_object_put(env, &obj->oo_dt.do_lu);
3481 static int osd_index_declare_ea_insert(const struct lu_env *env,
3482 struct dt_object *dt,
3483 const struct dt_rec *rec,
3484 const struct dt_key *key,
3485 struct thandle *handle)
3487 struct osd_thandle *oh;
3488 struct inode *inode;
3492 LASSERT(dt_object_exists(dt));
3493 LASSERT(handle != NULL);
3495 oh = container_of0(handle, struct osd_thandle, ot_super);
3496 LASSERT(oh->ot_handle == NULL);
3498 OSD_DECLARE_OP(oh, insert);
3499 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3501 inode = osd_dt_obj(dt)->oo_inode;
3504 /* We ignore block quota on meta pool (MDTs), so needn't
3505 * calculate how many blocks will be consumed by this index
3507 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3508 true, true, NULL, false);
3513 * Index add function for interoperability mode (b11826).
3514 * It will add the directory entry.This entry is needed to
3515 * maintain name->fid mapping.
3517 * \param key it is key i.e. file entry to be inserted
3518 * \param rec it is value of given key i.e. fid
3520 * \retval 0, on success
3521 * \retval -ve, on error
3523 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3524 const struct dt_rec *rec,
3525 const struct dt_key *key, struct thandle *th,
3526 struct lustre_capa *capa, int ignore_quota)
3528 struct osd_object *obj = osd_dt_obj(dt);
3529 struct lu_fid *fid = (struct lu_fid *) rec;
3530 const char *name = (const char *)key;
3531 struct osd_object *child;
3536 LASSERT(osd_invariant(obj));
3537 LASSERT(dt_object_exists(dt));
3538 LASSERT(th != NULL);
3540 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3543 child = osd_object_find(env, dt, fid);
3544 if (!IS_ERR(child)) {
3545 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3546 osd_object_put(env, child);
3548 rc = PTR_ERR(child);
3551 LASSERT(osd_invariant(obj));
3556 * Initialize osd Iterator for given osd index object.
3558 * \param dt osd index object
3561 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3562 struct dt_object *dt,
3564 struct lustre_capa *capa)
3566 struct osd_it_iam *it;
3567 struct osd_thread_info *oti = osd_oti_get(env);
3568 struct osd_object *obj = osd_dt_obj(dt);
3569 struct lu_object *lo = &dt->do_lu;
3570 struct iam_path_descr *ipd;
3571 struct iam_container *bag = &obj->oo_dir->od_container;
3573 LASSERT(lu_object_exists(lo));
3575 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3576 return ERR_PTR(-EACCES);
3579 ipd = osd_it_ipd_get(env, bag);
3580 if (likely(ipd != NULL)) {
3584 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3585 return (struct dt_it *)it;
3587 return ERR_PTR(-ENOMEM);
3591 * free given Iterator.
3594 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3596 struct osd_it_iam *it = (struct osd_it_iam *)di;
3597 struct osd_object *obj = it->oi_obj;
3599 iam_it_fini(&it->oi_it);
3600 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3601 lu_object_put(env, &obj->oo_dt.do_lu);
3605 * Move Iterator to record specified by \a key
3607 * \param di osd iterator
3608 * \param key key for index
3610 * \retval +ve di points to record with least key not larger than key
3611 * \retval 0 di points to exact matched key
3612 * \retval -ve failure
3615 static int osd_it_iam_get(const struct lu_env *env,
3616 struct dt_it *di, const struct dt_key *key)
3618 struct osd_thread_info *oti = osd_oti_get(env);
3619 struct osd_it_iam *it = (struct osd_it_iam *)di;
3621 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3622 /* swab quota uid/gid */
3623 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3624 key = (struct dt_key *)&oti->oti_quota_id;
3627 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3633 * \param di osd iterator
3636 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3638 struct osd_it_iam *it = (struct osd_it_iam *)di;
3640 iam_it_put(&it->oi_it);
3644 * Move iterator by one record
3646 * \param di osd iterator
3648 * \retval +1 end of container reached
3650 * \retval -ve failure
3653 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3655 struct osd_it_iam *it = (struct osd_it_iam *)di;
3657 return iam_it_next(&it->oi_it);
3661 * Return pointer to the key under iterator.
3664 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3665 const struct dt_it *di)
3667 struct osd_thread_info *oti = osd_oti_get(env);
3668 struct osd_it_iam *it = (struct osd_it_iam *)di;
3669 struct osd_object *obj = it->oi_obj;
3672 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3674 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3675 /* swab quota uid/gid */
3676 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3677 key = (struct dt_key *)&oti->oti_quota_id;
3684 * Return size of key under iterator (in bytes)
3687 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3689 struct osd_it_iam *it = (struct osd_it_iam *)di;
3691 return iam_it_key_size(&it->oi_it);
3694 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3695 int len, __u16 type)
3697 struct luda_type *lt;
3698 const unsigned align = sizeof(struct luda_type) - 1;
3700 /* check if file type is required */
3701 if (attr & LUDA_TYPE) {
3702 len = (len + align) & ~align;
3704 lt = (void *) ent->lde_name + len;
3705 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3706 ent->lde_attrs |= LUDA_TYPE;
3709 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3713 * build lu direct from backend fs dirent.
3716 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3717 struct lu_fid *fid, __u64 offset,
3718 char *name, __u16 namelen,
3719 __u16 type, __u32 attr)
3721 fid_cpu_to_le(&ent->lde_fid, fid);
3722 ent->lde_attrs = LUDA_FID;
3724 ent->lde_hash = cpu_to_le64(offset);
3725 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3727 strncpy(ent->lde_name, name, namelen);
3728 ent->lde_namelen = cpu_to_le16(namelen);
3730 /* append lustre attributes */
3731 osd_it_append_attrs(ent, attr, namelen, type);
3735 * Return pointer to the record under iterator.
3737 static int osd_it_iam_rec(const struct lu_env *env,
3738 const struct dt_it *di,
3739 struct dt_rec *dtrec, __u32 attr)
3741 struct osd_it_iam *it = (struct osd_it_iam *)di;
3742 struct osd_thread_info *info = osd_oti_get(env);
3745 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3746 const struct osd_fid_pack *rec;
3747 struct lu_fid *fid = &info->oti_fid;
3748 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3754 name = (char *)iam_it_key_get(&it->oi_it);
3756 RETURN(PTR_ERR(name));
3758 namelen = iam_it_key_size(&it->oi_it);
3760 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3762 RETURN(PTR_ERR(rec));
3764 rc = osd_fid_unpack(fid, rec);
3768 hash = iam_it_store(&it->oi_it);
3770 /* IAM does not store object type in IAM index (dir) */
3771 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3773 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3774 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3775 (struct iam_rec *)dtrec);
3776 osd_quota_unpack(it->oi_obj, dtrec);
3778 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3779 (struct iam_rec *)dtrec);
3786 * Returns cookie for current Iterator position.
3788 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3790 struct osd_it_iam *it = (struct osd_it_iam *)di;
3792 return iam_it_store(&it->oi_it);
3796 * Restore iterator from cookie.
3798 * \param di osd iterator
3799 * \param hash Iterator location cookie
3801 * \retval +ve di points to record with least key not larger than key.
3802 * \retval 0 di points to exact matched key
3803 * \retval -ve failure
3806 static int osd_it_iam_load(const struct lu_env *env,
3807 const struct dt_it *di, __u64 hash)
3809 struct osd_it_iam *it = (struct osd_it_iam *)di;
3811 return iam_it_load(&it->oi_it, hash);
3814 static const struct dt_index_operations osd_index_iam_ops = {
3815 .dio_lookup = osd_index_iam_lookup,
3816 .dio_declare_insert = osd_index_declare_iam_insert,
3817 .dio_insert = osd_index_iam_insert,
3818 .dio_declare_delete = osd_index_declare_iam_delete,
3819 .dio_delete = osd_index_iam_delete,
3821 .init = osd_it_iam_init,
3822 .fini = osd_it_iam_fini,
3823 .get = osd_it_iam_get,
3824 .put = osd_it_iam_put,
3825 .next = osd_it_iam_next,
3826 .key = osd_it_iam_key,
3827 .key_size = osd_it_iam_key_size,
3828 .rec = osd_it_iam_rec,
3829 .store = osd_it_iam_store,
3830 .load = osd_it_iam_load
3835 * Creates or initializes iterator context.
3837 * \retval struct osd_it_ea, iterator structure on success
3840 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3841 struct dt_object *dt,
3843 struct lustre_capa *capa)
3845 struct osd_object *obj = osd_dt_obj(dt);
3846 struct osd_thread_info *info = osd_oti_get(env);
3847 struct osd_it_ea *it = &info->oti_it_ea;
3848 struct lu_object *lo = &dt->do_lu;
3849 struct dentry *obj_dentry = &info->oti_it_dentry;
3851 LASSERT(lu_object_exists(lo));
3853 obj_dentry->d_inode = obj->oo_inode;
3854 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3855 obj_dentry->d_name.hash = 0;
3857 it->oie_rd_dirent = 0;
3858 it->oie_it_dirent = 0;
3859 it->oie_dirent = NULL;
3860 it->oie_buf = info->oti_it_ea_buf;
3862 it->oie_file.f_pos = 0;
3863 it->oie_file.f_dentry = obj_dentry;
3864 if (attr & LUDA_64BITHASH)
3865 it->oie_file.f_mode |= FMODE_64BITHASH;
3867 it->oie_file.f_mode |= FMODE_32BITHASH;
3868 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
3869 it->oie_file.f_op = obj->oo_inode->i_fop;
3870 it->oie_file.private_data = NULL;
3872 RETURN((struct dt_it *) it);
3876 * Destroy or finishes iterator context.
3878 * \param di iterator structure to be destroyed
3880 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3882 struct osd_it_ea *it = (struct osd_it_ea *)di;
3883 struct osd_object *obj = it->oie_obj;
3884 struct inode *inode = obj->oo_inode;
3887 it->oie_file.f_op->release(inode, &it->oie_file);
3888 lu_object_put(env, &obj->oo_dt.do_lu);
3893 * It position the iterator at given key, so that next lookup continues from
3894 * that key Or it is similar to dio_it->load() but based on a key,
3895 * rather than file position.
3897 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3900 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3902 static int osd_it_ea_get(const struct lu_env *env,
3903 struct dt_it *di, const struct dt_key *key)
3905 struct osd_it_ea *it = (struct osd_it_ea *)di;
3908 LASSERT(((const char *)key)[0] == '\0');
3909 it->oie_file.f_pos = 0;
3910 it->oie_rd_dirent = 0;
3911 it->oie_it_dirent = 0;
3912 it->oie_dirent = NULL;
3920 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3925 * It is called internally by ->readdir(). It fills the
3926 * iterator's in-memory data structure with required
3927 * information i.e. name, namelen, rec_size etc.
3929 * \param buf in which information to be filled in.
3930 * \param name name of the file in given dir
3932 * \retval 0 on success
3933 * \retval 1 on buffer full
3935 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3936 loff_t offset, __u64 ino,
3939 struct osd_it_ea *it = (struct osd_it_ea *)buf;
3940 struct osd_it_ea_dirent *ent = it->oie_dirent;
3941 struct lu_fid *fid = &ent->oied_fid;
3942 struct osd_fid_pack *rec;
3945 /* this should never happen */
3946 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3947 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3951 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3955 if (d_type & LDISKFS_DIRENT_LUFID) {
3956 rec = (struct osd_fid_pack*) (name + namelen + 1);
3958 if (osd_fid_unpack(fid, rec) != 0)
3961 d_type &= ~LDISKFS_DIRENT_LUFID;
3966 ent->oied_ino = ino;
3967 ent->oied_off = offset;
3968 ent->oied_namelen = namelen;
3969 ent->oied_type = d_type;
3971 memcpy(ent->oied_name, name, namelen);
3973 it->oie_rd_dirent++;
3974 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3979 * Calls ->readdir() to load a directory entry at a time
3980 * and stored it in iterator's in-memory data structure.
3982 * \param di iterator's in memory structure
3984 * \retval 0 on success
3985 * \retval -ve on error
3987 static int osd_ldiskfs_it_fill(const struct lu_env *env,
3988 const struct dt_it *di)
3990 struct osd_it_ea *it = (struct osd_it_ea *)di;
3991 struct osd_object *obj = it->oie_obj;
3992 struct inode *inode = obj->oo_inode;
3993 struct htree_lock *hlock = NULL;
3997 it->oie_dirent = it->oie_buf;
3998 it->oie_rd_dirent = 0;
4000 if (obj->oo_hl_head != NULL) {
4001 hlock = osd_oti_get(env)->oti_hlock;
4002 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4003 inode, LDISKFS_HLOCK_READDIR);
4005 cfs_down_read(&obj->oo_ext_idx_sem);
4008 result = inode->i_fop->readdir(&it->oie_file, it,
4009 (filldir_t) osd_ldiskfs_filldir);
4012 ldiskfs_htree_unlock(hlock);
4014 cfs_up_read(&obj->oo_ext_idx_sem);
4016 if (it->oie_rd_dirent == 0) {
4019 it->oie_dirent = it->oie_buf;
4020 it->oie_it_dirent = 1;
4027 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4028 * to load a directory entry at a time and stored it in
4029 * iterator's in-memory data structure.
4031 * \param di iterator's in memory structure
4033 * \retval +ve iterator reached to end
4034 * \retval 0 iterator not reached to end
4035 * \retval -ve on error
4037 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4039 struct osd_it_ea *it = (struct osd_it_ea *)di;
4044 if (it->oie_it_dirent < it->oie_rd_dirent) {
4046 (void *) it->oie_dirent +
4047 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4048 it->oie_dirent->oied_namelen);
4049 it->oie_it_dirent++;
4052 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4055 rc = osd_ldiskfs_it_fill(env, di);
4062 * Returns the key at current position from iterator's in memory structure.
4064 * \param di iterator's in memory structure
4066 * \retval key i.e. struct dt_key on success
4068 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4069 const struct dt_it *di)
4071 struct osd_it_ea *it = (struct osd_it_ea *)di;
4073 return (struct dt_key *)it->oie_dirent->oied_name;
4077 * Returns the key's size at current position from iterator's in memory structure.
4079 * \param di iterator's in memory structure
4081 * \retval key_size i.e. struct dt_key on success
4083 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4085 struct osd_it_ea *it = (struct osd_it_ea *)di;
4087 return it->oie_dirent->oied_namelen;
4092 * Returns the value (i.e. fid/igif) at current position from iterator's
4093 * in memory structure.
4095 * \param di struct osd_it_ea, iterator's in memory structure
4096 * \param attr attr requested for dirent.
4097 * \param lde lustre dirent
4099 * \retval 0 no error and \param lde has correct lustre dirent.
4100 * \retval -ve on error
4102 static inline int osd_it_ea_rec(const struct lu_env *env,
4103 const struct dt_it *di,
4104 struct dt_rec *dtrec, __u32 attr)
4106 struct osd_it_ea *it = (struct osd_it_ea *)di;
4107 struct osd_object *obj = it->oie_obj;
4108 struct osd_device *dev = osd_obj2dev(obj);
4109 struct osd_scrub *scrub = &dev->od_scrub;
4110 struct scrub_file *sf = &scrub->os_file;
4111 struct osd_thread_info *oti = osd_oti_get(env);
4112 struct osd_idmap_cache *oic = &oti->oti_cache;
4113 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4114 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4115 __u32 ino = it->oie_dirent->oied_ino;
4119 if (!fid_is_sane(fid)) {
4120 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4124 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4127 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4128 it->oie_dirent->oied_name,
4129 it->oie_dirent->oied_namelen,
4130 it->oie_dirent->oied_type, attr);
4131 if (!fid_is_norm(fid)) {
4132 fid_zero(&oic->oic_fid);
4136 oic->oic_fid = *fid;
4137 if ((scrub->os_pos_current <= ino) &&
4138 (sf->sf_flags & SF_INCONSISTENT ||
4139 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4140 rc = osd_consistency_check(oti, dev, oic);
4146 * Returns a cookie for current position of the iterator head, so that
4147 * user can use this cookie to load/start the iterator next time.
4149 * \param di iterator's in memory structure
4151 * \retval cookie for current position, on success
4153 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4155 struct osd_it_ea *it = (struct osd_it_ea *)di;
4157 return it->oie_dirent->oied_off;
4161 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4162 * to load a directory entry at a time and stored it i inn,
4163 * in iterator's in-memory data structure.
4165 * \param di struct osd_it_ea, iterator's in memory structure
4167 * \retval +ve on success
4168 * \retval -ve on error
4170 static int osd_it_ea_load(const struct lu_env *env,
4171 const struct dt_it *di, __u64 hash)
4173 struct osd_it_ea *it = (struct osd_it_ea *)di;
4177 it->oie_file.f_pos = hash;
4179 rc = osd_ldiskfs_it_fill(env, di);
4187 * Index lookup function for interoperability mode (b11826).
4189 * \param key, key i.e. file name to be searched
4191 * \retval +ve, on success
4192 * \retval -ve, on error
4194 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4195 struct dt_rec *rec, const struct dt_key *key,
4196 struct lustre_capa *capa)
4198 struct osd_object *obj = osd_dt_obj(dt);
4203 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4204 LINVRNT(osd_invariant(obj));
4206 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4209 rc = osd_ea_lookup_rec(env, obj, rec, key);
4217 * Index and Iterator operations for interoperability
4218 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4220 static const struct dt_index_operations osd_index_ea_ops = {
4221 .dio_lookup = osd_index_ea_lookup,
4222 .dio_declare_insert = osd_index_declare_ea_insert,
4223 .dio_insert = osd_index_ea_insert,
4224 .dio_declare_delete = osd_index_declare_ea_delete,
4225 .dio_delete = osd_index_ea_delete,
4227 .init = osd_it_ea_init,
4228 .fini = osd_it_ea_fini,
4229 .get = osd_it_ea_get,
4230 .put = osd_it_ea_put,
4231 .next = osd_it_ea_next,
4232 .key = osd_it_ea_key,
4233 .key_size = osd_it_ea_key_size,
4234 .rec = osd_it_ea_rec,
4235 .store = osd_it_ea_store,
4236 .load = osd_it_ea_load
4240 static void *osd_key_init(const struct lu_context *ctx,
4241 struct lu_context_key *key)
4243 struct osd_thread_info *info;
4245 OBD_ALLOC_PTR(info);
4247 return ERR_PTR(-ENOMEM);
4249 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4250 if (info->oti_it_ea_buf == NULL)
4253 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4255 info->oti_hlock = ldiskfs_htree_lock_alloc();
4256 if (info->oti_hlock == NULL)
4262 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4265 return ERR_PTR(-ENOMEM);
4268 static void osd_key_fini(const struct lu_context *ctx,
4269 struct lu_context_key *key, void* data)
4271 struct osd_thread_info *info = data;
4273 if (info->oti_hlock != NULL)
4274 ldiskfs_htree_lock_free(info->oti_hlock);
4275 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4279 static void osd_key_exit(const struct lu_context *ctx,
4280 struct lu_context_key *key, void *data)
4282 struct osd_thread_info *info = data;
4284 LASSERT(info->oti_r_locks == 0);
4285 LASSERT(info->oti_w_locks == 0);
4286 LASSERT(info->oti_txns == 0);
4289 /* type constructor/destructor: osd_type_init, osd_type_fini */
4290 LU_TYPE_INIT_FINI(osd, &osd_key);
4292 struct lu_context_key osd_key = {
4293 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4294 .lct_init = osd_key_init,
4295 .lct_fini = osd_key_fini,
4296 .lct_exit = osd_key_exit
4300 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4301 const char *name, struct lu_device *next)
4303 struct osd_device *osd = osd_dev(d);
4305 strncpy(osd->od_svname, name, MAX_OBD_NAME);
4306 return osd_procfs_init(osd, name);
4309 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4313 osd_scrub_cleanup(env, o);
4316 fsfilt_put_ops(o->od_fsops);
4320 /* shutdown quota slave instance associated with the device */
4321 if (o->od_quota_slave != NULL) {
4322 qsd_fini(env, o->od_quota_slave);
4323 o->od_quota_slave = NULL;
4329 static int osd_mount(const struct lu_env *env,
4330 struct osd_device *o, struct lustre_cfg *cfg)
4332 const char *name = lustre_cfg_string(cfg, 0);
4333 const char *dev = lustre_cfg_string(cfg, 1);
4335 unsigned long page, s_flags, lmd_flags = 0;
4336 struct page *__page;
4337 struct file_system_type *type;
4338 char *options = NULL;
4343 if (o->od_mnt != NULL)
4346 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4347 if (o->od_fsops == NULL) {
4348 CERROR("Can't find fsfilt_ldiskfs\n");
4352 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4354 GOTO(out, rc = -ENOMEM);
4356 str = lustre_cfg_string(cfg, 2);
4357 s_flags = simple_strtoul(str, NULL, 0);
4358 str = strstr(str, ":");
4360 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4361 opts = lustre_cfg_string(cfg, 3);
4362 page = (unsigned long)cfs_page_address(__page);
4363 options = (char *)page;
4366 strcat(options, "user_xattr,acl");
4368 strcat(options, opts);
4370 /* Glom up mount options */
4371 if (*options != '\0')
4372 strcat(options, ",");
4373 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4375 type = get_fs_type("ldiskfs");
4377 CERROR("%s: cannot find ldiskfs module\n", name);
4378 GOTO(out, rc = -ENODEV);
4381 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4382 cfs_module_put(type->owner);
4384 if (IS_ERR(o->od_mnt)) {
4385 rc = PTR_ERR(o->od_mnt);
4386 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4391 if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4392 CERROR("%s: underlying device %s is marked as read-only. "
4393 "Setup failed\n", name, dev);
4396 GOTO(out, rc = -EROFS);
4399 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4400 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4401 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4404 GOTO(out, rc = -EINVAL);
4407 if (lmd_flags & LMD_FLG_IAM) {
4409 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name);
4412 if (lmd_flags & LMD_FLG_NOSCRUB)
4413 o->od_scrub.os_no_scrub = 1;
4417 OBD_PAGE_FREE(__page);
4419 fsfilt_put_ops(o->od_fsops);
4424 static struct lu_device *osd_device_fini(const struct lu_env *env,
4425 struct lu_device *d)
4430 rc = osd_shutdown(env, osd_dev(d));
4432 osd_compat_fini(osd_dev(d));
4434 shrink_dcache_sb(osd_sb(osd_dev(d)));
4435 osd_sync(env, lu2dt_dev(d));
4437 rc = osd_procfs_fini(osd_dev(d));
4439 CERROR("proc fini error %d \n", rc);
4440 RETURN (ERR_PTR(rc));
4443 if (osd_dev(d)->od_mnt) {
4444 mntput(osd_dev(d)->od_mnt);
4445 osd_dev(d)->od_mnt = NULL;
4451 static int osd_device_init0(const struct lu_env *env,
4452 struct osd_device *o,
4453 struct lustre_cfg *cfg)
4455 struct lu_device *l = osd2lu_dev(o);
4456 struct osd_thread_info *info;
4459 /* if the module was re-loaded, env can loose its keys */
4460 rc = lu_env_refill((struct lu_env *) env);
4463 info = osd_oti_get(env);
4466 l->ld_ops = &osd_lu_ops;
4467 o->od_dt_dev.dd_ops = &osd_dt_ops;
4469 cfs_spin_lock_init(&o->od_osfs_lock);
4470 cfs_mutex_init(&o->od_otable_mutex);
4471 o->od_osfs_age = cfs_time_shift_64(-1000);
4473 o->od_capa_hash = init_capa_hash();
4474 if (o->od_capa_hash == NULL)
4475 GOTO(out, rc = -ENOMEM);
4477 o->od_read_cache = 1;
4478 o->od_writethrough_cache = 1;
4479 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4481 rc = osd_mount(env, o, cfg);
4485 /* setup scrub, including OI files initialization */
4486 rc = osd_scrub_setup(env, o);
4490 strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4491 sizeof(o->od_svname) - 1);
4493 rc = osd_compat_init(o);
4495 GOTO(out_scrub, rc);
4497 rc = lu_site_init(&o->od_site, l);
4499 GOTO(out_compat, rc);
4500 o->od_site.ls_bottom_dev = l;
4502 rc = lu_site_init_finish(&o->od_site);
4506 rc = osd_procfs_init(o, o->od_svname);
4508 CERROR("%s: can't initialize procfs: rc = %d\n",
4513 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4517 lu_site_fini(&o->od_site);
4521 osd_scrub_cleanup(env, o);
4523 osd_oi_fini(info, o);
4524 osd_shutdown(env, o);
4528 cleanup_capa_hash(o->od_capa_hash);
4533 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4534 struct lu_device_type *t,
4535 struct lustre_cfg *cfg)
4537 struct osd_device *o;
4542 return ERR_PTR(-ENOMEM);
4544 rc = dt_device_init(&o->od_dt_dev, t);
4546 rc = osd_device_init0(env, o, cfg);
4548 dt_device_fini(&o->od_dt_dev);
4551 if (unlikely(rc != 0))
4554 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4557 static struct lu_device *osd_device_free(const struct lu_env *env,
4558 struct lu_device *d)
4560 struct osd_device *o = osd_dev(d);
4563 cleanup_capa_hash(o->od_capa_hash);
4564 /* XXX: make osd top device in order to release reference */
4565 d->ld_site->ls_top_dev = d;
4566 lu_site_purge(env, d->ld_site, -1);
4567 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4568 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4569 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4571 lu_site_fini(&o->od_site);
4572 dt_device_fini(&o->od_dt_dev);
4577 static int osd_process_config(const struct lu_env *env,
4578 struct lu_device *d, struct lustre_cfg *cfg)
4580 struct osd_device *o = osd_dev(d);
4584 switch(cfg->lcfg_command) {
4586 err = osd_mount(env, o, cfg);
4589 lu_dev_del_linkage(d->ld_site, d);
4590 err = osd_shutdown(env, o);
4599 static int osd_recovery_complete(const struct lu_env *env,
4600 struct lu_device *d)
4606 * we use exports to track all osd users
4608 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4609 struct obd_device *obd, struct obd_uuid *cluuid,
4610 struct obd_connect_data *data, void *localdata)
4612 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4613 struct lustre_handle conn;
4617 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4619 rc = class_connect(&conn, obd, cluuid);
4623 *exp = class_conn2export(&conn);
4625 cfs_spin_lock(&osd->od_osfs_lock);
4627 cfs_spin_unlock(&osd->od_osfs_lock);
4633 * once last export (we don't count self-export) disappeared
4634 * osd can be released
4636 static int osd_obd_disconnect(struct obd_export *exp)
4638 struct obd_device *obd = exp->exp_obd;
4639 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4640 int rc, release = 0;
4643 /* Only disconnect the underlying layers on the final disconnect. */
4644 cfs_spin_lock(&osd->od_osfs_lock);
4646 if (osd->od_connects == 0)
4648 cfs_spin_unlock(&osd->od_osfs_lock);
4650 rc = class_disconnect(exp); /* bz 9811 */
4652 if (rc == 0 && release)
4653 class_manual_cleanup(obd);
4657 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4658 struct lu_device *dev)
4660 struct osd_device *osd = osd_dev(dev);
4664 if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
4665 /* MDT/MDD still use old infrastructure to create
4667 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4673 /* 2. setup quota slave instance */
4674 osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
4675 osd->od_proc_entry);
4676 if (IS_ERR(osd->od_quota_slave)) {
4677 result = PTR_ERR(osd->od_quota_slave);
4678 osd->od_quota_slave = NULL;
4684 static const struct lu_object_operations osd_lu_obj_ops = {
4685 .loo_object_init = osd_object_init,
4686 .loo_object_delete = osd_object_delete,
4687 .loo_object_release = osd_object_release,
4688 .loo_object_free = osd_object_free,
4689 .loo_object_print = osd_object_print,
4690 .loo_object_invariant = osd_object_invariant
4693 const struct lu_device_operations osd_lu_ops = {
4694 .ldo_object_alloc = osd_object_alloc,
4695 .ldo_process_config = osd_process_config,
4696 .ldo_recovery_complete = osd_recovery_complete,
4697 .ldo_prepare = osd_prepare,
4700 static const struct lu_device_type_operations osd_device_type_ops = {
4701 .ldto_init = osd_type_init,
4702 .ldto_fini = osd_type_fini,
4704 .ldto_start = osd_type_start,
4705 .ldto_stop = osd_type_stop,
4707 .ldto_device_alloc = osd_device_alloc,
4708 .ldto_device_free = osd_device_free,
4710 .ldto_device_init = osd_device_init,
4711 .ldto_device_fini = osd_device_fini
4714 struct lu_device_type osd_device_type = {
4715 .ldt_tags = LU_DEVICE_DT,
4716 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
4717 .ldt_ops = &osd_device_type_ops,
4718 .ldt_ctx_tags = LCT_LOCAL,
4722 * lprocfs legacy support.
4724 static struct obd_ops osd_obd_device_ops = {
4725 .o_owner = THIS_MODULE,
4726 .o_connect = osd_obd_connect,
4727 .o_disconnect = osd_obd_disconnect
4730 static int __init osd_mod_init(void)
4732 struct lprocfs_static_vars lvars;
4735 lprocfs_osd_init_vars(&lvars);
4736 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4737 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4740 static void __exit osd_mod_exit(void)
4742 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4745 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4746 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4747 MODULE_LICENSE("GPL");
4749 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);