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, Intel Corporation.
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>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include "osd_internal.h"
68 /* llo_* api support */
69 #include <md_object.h>
70 #include <lustre_quota.h>
73 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
74 "ldiskfs with parallel directory operations");
76 static const char dot[] = ".";
77 static const char dotdot[] = "..";
78 static const char remote_obj_dir[] = "REM_OBJ_DIR";
80 static const struct lu_object_operations osd_lu_obj_ops;
81 static const struct dt_object_operations osd_obj_ops;
82 static const struct dt_object_operations osd_obj_ea_ops;
83 static const struct dt_object_operations osd_obj_otable_it_ops;
84 static const struct dt_index_operations osd_index_iam_ops;
85 static const struct dt_index_operations osd_index_ea_ops;
87 #ifdef OSD_TRACK_DECLARES
88 int osd_trans_declare_op2rb[] = {
89 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
90 [OSD_OT_PUNCH] = OSD_OT_MAX,
91 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
92 [OSD_OT_CREATE] = OSD_OT_DESTROY,
93 [OSD_OT_DESTROY] = OSD_OT_CREATE,
94 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
95 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
96 [OSD_OT_WRITE] = OSD_OT_WRITE,
97 [OSD_OT_INSERT] = OSD_OT_DELETE,
98 [OSD_OT_DELETE] = OSD_OT_INSERT,
99 [OSD_OT_QUOTA] = OSD_OT_MAX,
103 static int osd_has_index(const struct osd_object *obj)
105 return obj->oo_dt.do_index_ops != NULL;
108 static int osd_object_invariant(const struct lu_object *l)
110 return osd_invariant(osd_obj(l));
114 * Concurrency: doesn't matter
116 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
118 return osd_oti_get(env)->oti_r_locks > 0;
122 * Concurrency: doesn't matter
124 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
126 struct osd_thread_info *oti = osd_oti_get(env);
127 return oti->oti_w_locks > 0 && o->oo_owner == env;
131 * Concurrency: doesn't access mutable data
133 static int osd_root_get(const struct lu_env *env,
134 struct dt_device *dev, struct lu_fid *f)
136 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
141 * OSD object methods.
145 * Concurrency: no concurrent access is possible that early in object
148 static struct lu_object *osd_object_alloc(const struct lu_env *env,
149 const struct lu_object_header *hdr,
152 struct osd_object *mo;
158 l = &mo->oo_dt.do_lu;
159 dt_object_init(&mo->oo_dt, NULL, d);
160 mo->oo_dt.do_ops = &osd_obj_ea_ops;
161 l->lo_ops = &osd_lu_obj_ops;
162 init_rwsem(&mo->oo_sem);
163 init_rwsem(&mo->oo_ext_idx_sem);
164 spin_lock_init(&mo->oo_guard);
171 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
172 const char *name, void *buf, int len)
174 dentry->d_inode = inode;
175 return inode->i_op->getxattr(dentry, name, buf, len);
178 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
179 struct dentry *dentry, struct lustre_mdt_attrs *lma)
183 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
186 /* try with old lma size */
187 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
188 info->oti_mdt_attrs_old,
191 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
194 /* Check LMA compatibility */
195 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
196 CWARN("%.16s: unsupported incompat LMA feature(s) "
198 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
199 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
203 lustre_lma_swab(lma);
206 } else if (rc == 0) {
214 * retrieve object from backend ext fs.
216 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
217 struct osd_inode_id *id)
219 struct inode *inode = NULL;
221 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
223 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
224 id->oii_ino, PTR_ERR(inode));
225 } else if (id->oii_gen != OSD_OII_NOGEN &&
226 inode->i_generation != id->oii_gen) {
227 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
229 id->oii_ino, id->oii_gen, inode->i_generation);
231 inode = ERR_PTR(-ESTALE);
232 } else if (inode->i_nlink == 0) {
233 /* due to parallel readdir and unlink,
234 * we can have dead inode here. */
235 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
236 make_bad_inode(inode);
238 inode = ERR_PTR(-ESTALE);
239 } else if (is_bad_inode(inode)) {
240 CWARN("%.16s: bad inode: ino = %u\n",
241 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
243 inode = ERR_PTR(-ENOENT);
245 if (id->oii_gen == OSD_OII_NOGEN)
246 osd_id_gen(id, inode->i_ino, inode->i_generation);
248 /* Do not update file c/mtime in ldiskfs.
249 * NB: we don't have any lock to protect this because we don't
250 * have reference on osd_object now, but contention with
251 * another lookup + attr_set can't happen in the tiny window
252 * between if (...) and set S_NOCMTIME. */
253 if (!(inode->i_flags & S_NOCMTIME))
254 inode->i_flags |= S_NOCMTIME;
259 static struct inode *
260 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
261 struct osd_inode_id *id, struct lu_fid *fid)
263 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
267 inode = osd_iget(info, dev, id);
271 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
273 *fid = lma->lma_self_fid;
274 } else if (rc == -ENODATA) {
275 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
276 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
278 lu_igif_build(fid, inode->i_ino, inode->i_generation);
286 static struct inode *
287 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
288 struct osd_inode_id *id, const struct lu_fid *fid)
290 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
294 inode = osd_iget(info, dev, id);
298 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
307 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
308 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
309 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
311 return ERR_PTR(-EREMCHG);
317 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
318 const struct lu_fid *fid,
319 const struct lu_object_conf *conf)
321 struct osd_thread_info *info;
322 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
323 struct osd_device *dev;
324 struct osd_idmap_cache *oic;
325 struct osd_inode_id *id;
327 struct osd_scrub *scrub;
328 struct scrub_file *sf;
333 LINVRNT(osd_invariant(obj));
334 LASSERT(obj->oo_inode == NULL);
335 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
338 scrub = &dev->od_scrub;
339 sf = &scrub->os_file;
340 info = osd_oti_get(env);
342 oic = &info->oti_cache;
344 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
347 /* Search order: 1. per-thread cache. */
348 if (lu_fid_eq(fid, &oic->oic_fid)) {
354 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
355 /* Search order: 2. OI scrub pending list. */
356 result = osd_oii_lookup(dev, fid, id);
361 if (sf->sf_flags & SF_INCONSISTENT)
365 * Objects are created as locking anchors or place holders for objects
366 * yet to be created. No need to osd_oi_lookup() at here because FID
367 * shouldn't never be re-used, if it's really a duplicate FID from
368 * unexpected reason, we should be able to detect it later by calling
369 * do_create->osd_oi_insert()
371 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
372 GOTO(out, result = 0);
374 /* Search order: 3. OI files. */
375 result = osd_oi_lookup(info, dev, fid, id, true);
376 if (result == -ENOENT) {
377 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
378 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
380 GOTO(out, result = 0);
390 inode = osd_iget(info, dev, id);
392 inode = osd_iget_verify(info, dev, id, fid);
394 result = PTR_ERR(inode);
395 if (result == -ENOENT || result == -ESTALE) {
396 fid_zero(&oic->oic_fid);
398 } else if (result == -EREMCHG) {
401 if (thread_is_running(&scrub->os_thread)) {
402 result = -EINPROGRESS;
403 } else if (!dev->od_noscrub) {
404 result = osd_scrub_start(dev);
405 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
406 "for "DFID", rc = %d [1]\n",
407 LDISKFS_SB(osd_sb(dev))->s_es->\
408 s_volume_name,PFID(fid), result);
409 if (result == 0 || result == -EALREADY)
410 result = -EINPROGRESS;
419 obj->oo_inode = inode;
420 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
422 obj->oo_compat_dot_created = 1;
423 obj->oo_compat_dotdot_created = 1;
425 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
426 GOTO(out, result = 0);
428 LASSERT(obj->oo_hl_head == NULL);
429 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
430 if (obj->oo_hl_head == NULL) {
431 obj->oo_inode = NULL;
433 GOTO(out, result = -ENOMEM);
435 GOTO(out, result = 0);
438 LINVRNT(osd_invariant(obj));
443 * Concurrency: shouldn't matter.
445 static void osd_object_init0(struct osd_object *obj)
447 LASSERT(obj->oo_inode != NULL);
448 obj->oo_dt.do_body_ops = &osd_body_ops;
449 obj->oo_dt.do_lu.lo_header->loh_attr |=
450 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
454 * Concurrency: no concurrent access is possible that early in object
457 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
458 const struct lu_object_conf *conf)
460 struct osd_object *obj = osd_obj(l);
463 LINVRNT(osd_invariant(obj));
465 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
466 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
467 l->lo_header->loh_attr |= LOHA_EXISTS;
471 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
472 obj->oo_dt.do_body_ops = &osd_body_ops_new;
473 if (result == 0 && obj->oo_inode != NULL)
474 osd_object_init0(obj);
476 LINVRNT(osd_invariant(obj));
481 * Concurrency: no concurrent access is possible that late in object
484 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
486 struct osd_object *obj = osd_obj(l);
488 LINVRNT(osd_invariant(obj));
490 dt_object_fini(&obj->oo_dt);
491 if (obj->oo_hl_head != NULL)
492 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
497 * Concurrency: no concurrent access is possible that late in object
500 static void osd_index_fini(struct osd_object *o)
502 struct iam_container *bag;
504 if (o->oo_dir != NULL) {
505 bag = &o->oo_dir->od_container;
506 if (o->oo_inode != NULL) {
507 if (bag->ic_object == o->oo_inode)
508 iam_container_fini(bag);
510 OBD_FREE_PTR(o->oo_dir);
516 * Concurrency: no concurrent access is possible that late in object
517 * life-cycle (for all existing callers, that is. New callers have to provide
518 * their own locking.)
520 static int osd_inode_unlinked(const struct inode *inode)
522 return inode->i_nlink == 0;
526 OSD_TXN_OI_DELETE_CREDITS = 20,
527 OSD_TXN_INODE_DELETE_CREDITS = 20
534 #if OSD_THANDLE_STATS
536 * Set time when the handle is allocated
538 static void osd_th_alloced(struct osd_thandle *oth)
540 oth->oth_alloced = cfs_time_current();
544 * Set time when the handle started
546 static void osd_th_started(struct osd_thandle *oth)
548 oth->oth_started = cfs_time_current();
552 * Helper function to convert time interval to microseconds packed in
555 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
559 cfs_duration_usec(cfs_time_sub(end, start), &val);
560 return val.tv_sec * 1000000 + val.tv_usec;
564 * Check whether the we deal with this handle for too long.
566 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
567 cfs_time_t alloced, cfs_time_t started,
570 cfs_time_t now = cfs_time_current();
572 LASSERT(dev != NULL);
574 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
575 interval_to_usec(alloced, started));
576 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
577 interval_to_usec(started, closed));
578 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
579 interval_to_usec(closed, now));
581 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
582 CWARN("transaction handle %p was open for too long: "
584 "alloced "CFS_TIME_T" ,"
585 "started "CFS_TIME_T" ,"
586 "closed "CFS_TIME_T"\n",
587 oth, now, alloced, started, closed);
588 libcfs_debug_dumpstack(NULL);
592 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
594 cfs_time_t __closed = cfs_time_current(); \
595 cfs_time_t __alloced = oth->oth_alloced; \
596 cfs_time_t __started = oth->oth_started; \
599 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
602 #else /* OSD_THANDLE_STATS */
604 #define osd_th_alloced(h) do {} while(0)
605 #define osd_th_started(h) do {} while(0)
606 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
608 #endif /* OSD_THANDLE_STATS */
611 * Concurrency: doesn't access mutable data.
613 static int osd_param_is_not_sane(const struct osd_device *dev,
614 const struct thandle *th)
616 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
618 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
622 * Concurrency: shouldn't matter.
624 static void osd_trans_commit_cb(struct super_block *sb,
625 struct ldiskfs_journal_cb_entry *jcb, int error)
627 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
628 struct thandle *th = &oh->ot_super;
629 struct lu_device *lud = &th->th_dev->dd_lu_dev;
630 struct dt_txn_commit_cb *dcb, *tmp;
632 LASSERT(oh->ot_handle == NULL);
635 CERROR("transaction @0x%p commit error: %d\n", th, error);
637 dt_txn_hook_commit(th);
639 /* call per-transaction callbacks if any */
640 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
641 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
642 "commit callback entry: magic=%x name='%s'\n",
643 dcb->dcb_magic, dcb->dcb_name);
644 cfs_list_del_init(&dcb->dcb_linkage);
645 dcb->dcb_func(NULL, th, dcb, error);
648 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
652 lu_context_exit(&th->th_ctx);
653 lu_context_fini(&th->th_ctx);
657 static struct thandle *osd_trans_create(const struct lu_env *env,
660 struct osd_thread_info *oti = osd_oti_get(env);
661 struct osd_iobuf *iobuf = &oti->oti_iobuf;
662 struct osd_thandle *oh;
666 /* on pending IO in this thread should left from prev. request */
667 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
669 th = ERR_PTR(-ENOMEM);
670 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
672 oh->ot_quota_trans = &oti->oti_quota_trans;
673 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
677 th->th_tags = LCT_TX_HANDLE;
679 oti->oti_dev = osd_dt_dev(d);
680 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
683 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
684 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
685 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
686 oti->oti_rollback = false;
692 * Concurrency: shouldn't matter.
694 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
697 struct osd_thread_info *oti = osd_oti_get(env);
698 struct osd_device *dev = osd_dt_dev(d);
700 struct osd_thandle *oh;
705 LASSERT(current->journal_info == NULL);
707 oh = container_of0(th, struct osd_thandle, ot_super);
709 LASSERT(oh->ot_handle == NULL);
711 rc = dt_txn_hook_start(env, d, th);
715 if (unlikely(osd_param_is_not_sane(dev, th))) {
716 static unsigned long last_printed;
717 static int last_credits;
719 CWARN("%.16s: too many transaction credits (%d > %d)\n",
720 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
722 osd_journal(dev)->j_max_transaction_buffers);
723 #ifdef OSD_TRACK_DECLARES
724 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
725 oti->oti_declare_ops[OSD_OT_CREATE],
726 oti->oti_declare_ops_cred[OSD_OT_CREATE],
727 oti->oti_declare_ops[OSD_OT_DELETE],
728 oti->oti_declare_ops_cred[OSD_OT_DELETE],
729 oti->oti_declare_ops[OSD_OT_DESTROY],
730 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
731 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
732 oti->oti_declare_ops[OSD_OT_ATTR_SET],
733 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
734 oti->oti_declare_ops[OSD_OT_XATTR_SET],
735 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
736 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
737 oti->oti_declare_ops[OSD_OT_WRITE],
738 oti->oti_declare_ops_cred[OSD_OT_WRITE],
739 oti->oti_declare_ops[OSD_OT_PUNCH],
740 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
741 oti->oti_declare_ops[OSD_OT_QUOTA],
742 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
743 CWARN(" insert: %u/%u, delete: %u/%u\n",
744 oti->oti_declare_ops[OSD_OT_INSERT],
745 oti->oti_declare_ops_cred[OSD_OT_INSERT],
746 oti->oti_declare_ops[OSD_OT_DESTROY],
747 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
748 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
749 oti->oti_declare_ops[OSD_OT_REF_ADD],
750 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
751 oti->oti_declare_ops[OSD_OT_REF_DEL],
752 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
754 if (last_credits != oh->ot_credits &&
755 time_after(jiffies, last_printed + 60 * HZ)) {
756 libcfs_debug_dumpstack(NULL);
757 last_credits = oh->ot_credits;
758 last_printed = jiffies;
761 /* XXX Limit the credits to 'max_transaction_buffers', and
762 * let the underlying filesystem to catch the error if
763 * we really need so many credits.
765 * This should be removed when we can calculate the
766 * credits precisely. */
767 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
771 * XXX temporary stuff. Some abstraction layer should
774 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
778 LASSERT(oti->oti_txns == 0);
779 lu_context_init(&th->th_ctx, th->th_tags);
780 lu_context_enter(&th->th_ctx);
782 lu_device_get(&d->dd_lu_dev);
783 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
795 * Concurrency: shouldn't matter.
797 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
800 struct osd_thandle *oh;
801 struct osd_thread_info *oti = osd_oti_get(env);
802 struct osd_iobuf *iobuf = &oti->oti_iobuf;
803 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
806 oh = container_of0(th, struct osd_thandle, ot_super);
809 /* inform the quota slave device that the transaction is
811 qsd_op_end(env, qsd, oh->ot_quota_trans);
812 oh->ot_quota_trans = NULL;
814 if (oh->ot_handle != NULL) {
815 handle_t *hdl = oh->ot_handle;
818 * add commit callback
819 * notice we don't do this in osd_trans_start()
820 * as underlying transaction can change during truncate
822 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
825 LASSERT(oti->oti_txns == 1);
827 rc = dt_txn_hook_stop(env, th);
829 CERROR("Failure in transaction hook: %d\n", rc);
831 /* hook functions might modify th_sync */
832 hdl->h_sync = th->th_sync;
834 oh->ot_handle = NULL;
835 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
836 rc = ldiskfs_journal_stop(hdl));
838 CERROR("Failure to stop transaction: %d\n", rc);
843 /* as we want IO to journal and data IO be concurrent, we don't block
844 * awaiting data IO completion in osd_do_bio(), instead we wait here
845 * once transaction is submitted to the journal. all reqular requests
846 * don't do direct IO (except read/write), thus this wait_event becomes
849 * IMPORTANT: we have to wait till any IO submited by the thread is
850 * completed otherwise iobuf may be corrupted by different request
852 cfs_wait_event(iobuf->dr_wait,
853 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
855 rc = iobuf->dr_error;
860 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
862 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
865 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
866 LASSERT(&dcb->dcb_func != NULL);
867 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
873 * Called just before object is freed. Releases all resources except for
874 * object itself (that is released by osd_object_free()).
876 * Concurrency: no concurrent access is possible that late in object
879 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
881 struct osd_object *obj = osd_obj(l);
882 struct inode *inode = obj->oo_inode;
884 LINVRNT(osd_invariant(obj));
887 * If object is unlinked remove fid->ino mapping from object index.
892 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
893 qid_t uid = inode->i_uid;
894 qid_t gid = inode->i_gid;
897 obj->oo_inode = NULL;
900 struct osd_thread_info *info = osd_oti_get(env);
901 struct lquota_id_info *qi = &info->oti_qi;
903 /* Release granted quota to master if necessary */
904 qi->lqi_id.qid_uid = uid;
905 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
907 qi->lqi_id.qid_uid = gid;
908 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
914 * Concurrency: ->loo_object_release() is called under site spin-lock.
916 static void osd_object_release(const struct lu_env *env,
922 * Concurrency: shouldn't matter.
924 static int osd_object_print(const struct lu_env *env, void *cookie,
925 lu_printer_t p, const struct lu_object *l)
927 struct osd_object *o = osd_obj(l);
930 if (o->oo_dir != NULL)
931 d = o->oo_dir->od_container.ic_descr;
934 return (*p)(env, cookie,
935 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
937 o->oo_inode ? o->oo_inode->i_ino : 0UL,
938 o->oo_inode ? o->oo_inode->i_generation : 0,
939 d ? d->id_ops->id_name : "plain");
943 * Concurrency: shouldn't matter.
945 int osd_statfs(const struct lu_env *env, struct dt_device *d,
946 struct obd_statfs *sfs)
948 struct osd_device *osd = osd_dt_dev(d);
949 struct super_block *sb = osd_sb(osd);
950 struct kstatfs *ksfs;
953 if (unlikely(osd->od_mnt == NULL))
956 /* osd_lproc.c call this without env, allocate ksfs for that case */
957 if (unlikely(env == NULL)) {
962 ksfs = &osd_oti_get(env)->oti_ksfs;
965 spin_lock(&osd->od_osfs_lock);
967 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
968 result = sb->s_op->statfs(sb->s_root, ksfs);
969 if (likely(result == 0)) { /* N.B. statfs can't really fail */
970 osd->od_osfs_age = cfs_time_current_64();
971 statfs_pack(&osd->od_statfs, ksfs);
972 if (sb->s_flags & MS_RDONLY)
973 sfs->os_state = OS_STATE_READONLY;
977 if (likely(result == 0))
978 *sfs = osd->od_statfs;
979 spin_unlock(&osd->od_osfs_lock);
981 if (unlikely(env == NULL))
988 * Estimate space needed for file creations. We assume the largest filename
989 * which is 2^64 - 1, hence a filename of 20 chars.
990 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
992 #ifdef __LDISKFS_DIR_REC_LEN
993 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
995 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
999 * Concurrency: doesn't access mutable data.
1001 static void osd_conf_get(const struct lu_env *env,
1002 const struct dt_device *dev,
1003 struct dt_device_param *param)
1005 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1008 * XXX should be taken from not-yet-existing fs abstraction layer.
1010 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1011 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1012 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1013 param->ddp_block_shift = sb->s_blocksize_bits;
1014 param->ddp_mount_type = LDD_MT_LDISKFS;
1015 param->ddp_maxbytes = sb->s_maxbytes;
1016 /* Overhead estimate should be fairly accurate, so we really take a tiny
1017 * error margin which also avoids fragmenting the filesystem too much */
1018 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1019 /* inode are statically allocated, so per-inode space consumption
1020 * is the space consumed by the directory entry */
1021 param->ddp_inodespace = PER_OBJ_USAGE;
1022 /* per-fragment overhead to be used by the client code */
1023 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1024 param->ddp_mntopts = 0;
1025 if (test_opt(sb, XATTR_USER))
1026 param->ddp_mntopts |= MNTOPT_USERXATTR;
1027 if (test_opt(sb, POSIX_ACL))
1028 param->ddp_mntopts |= MNTOPT_ACL;
1030 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1031 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1032 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1035 param->ddp_max_ea_size = sb->s_blocksize;
1040 * Concurrency: shouldn't matter.
1042 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1044 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1045 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1049 * Start commit for OSD device.
1051 * An implementation of dt_commit_async method for OSD device.
1052 * Asychronously starts underlayng fs sync and thereby a transaction
1055 * \param env environment
1056 * \param d dt device
1058 * \see dt_device_operations
1060 static int osd_commit_async(const struct lu_env *env,
1061 struct dt_device *d)
1063 struct super_block *s = osd_sb(osd_dt_dev(d));
1066 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1067 RETURN(s->s_op->sync_fs(s, 0));
1071 * Concurrency: shouldn't matter.
1074 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1076 struct super_block *sb = osd_sb(osd_dt_dev(d));
1077 struct block_device *dev = sb->s_bdev;
1078 #ifdef HAVE_DEV_SET_RDONLY
1079 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1082 int rc = -EOPNOTSUPP;
1086 #ifdef HAVE_DEV_SET_RDONLY
1087 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1089 if (jdev && (jdev != dev)) {
1090 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1092 dev_set_rdonly(jdev);
1094 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1095 dev_set_rdonly(dev);
1097 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1098 osd_dt_dev(d)->od_svname, (long)dev, rc);
1104 * Concurrency: serialization provided by callers.
1106 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1107 int mode, unsigned long timeout, __u32 alg,
1108 struct lustre_capa_key *keys)
1110 struct osd_device *dev = osd_dt_dev(d);
1113 dev->od_fl_capa = mode;
1114 dev->od_capa_timeout = timeout;
1115 dev->od_capa_alg = alg;
1116 dev->od_capa_keys = keys;
1121 * Note: we do not count into QUOTA here.
1122 * If we mount with --data_journal we may need more.
1124 const int osd_dto_credits_noquota[DTO_NR] = {
1127 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1128 * SINGLEDATA_TRANS_BLOCKS(8)
1129 * XXX Note: maybe iam need more, since iam have more level than
1132 [DTO_INDEX_INSERT] = 16,
1133 [DTO_INDEX_DELETE] = 16,
1137 [DTO_INDEX_UPDATE] = 16,
1139 * Create a object. The same as create object in EXT3.
1140 * DATA_TRANS_BLOCKS(14) +
1141 * INDEX_EXTRA_BLOCKS(8) +
1142 * 3(inode bits, groups, GDT)
1144 [DTO_OBJECT_CREATE] = 25,
1146 * XXX: real credits to be fixed
1148 [DTO_OBJECT_DELETE] = 25,
1150 * Attr set credits (inode)
1152 [DTO_ATTR_SET_BASE] = 1,
1154 * Xattr set. The same as xattr of EXT3.
1155 * DATA_TRANS_BLOCKS(14)
1156 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1157 * are also counted in. Do not know why?
1159 [DTO_XATTR_SET] = 14,
1162 * credits for inode change during write.
1164 [DTO_WRITE_BASE] = 3,
1166 * credits for single block write.
1168 [DTO_WRITE_BLOCK] = 14,
1170 * Attr set credits for chown.
1171 * This is extra credits for setattr, and it is null without quota
1173 [DTO_ATTR_SET_CHOWN]= 0
1176 static const struct dt_device_operations osd_dt_ops = {
1177 .dt_root_get = osd_root_get,
1178 .dt_statfs = osd_statfs,
1179 .dt_trans_create = osd_trans_create,
1180 .dt_trans_start = osd_trans_start,
1181 .dt_trans_stop = osd_trans_stop,
1182 .dt_trans_cb_add = osd_trans_cb_add,
1183 .dt_conf_get = osd_conf_get,
1184 .dt_sync = osd_sync,
1186 .dt_commit_async = osd_commit_async,
1187 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1190 static void osd_object_read_lock(const struct lu_env *env,
1191 struct dt_object *dt, unsigned role)
1193 struct osd_object *obj = osd_dt_obj(dt);
1194 struct osd_thread_info *oti = osd_oti_get(env);
1196 LINVRNT(osd_invariant(obj));
1198 LASSERT(obj->oo_owner != env);
1199 down_read_nested(&obj->oo_sem, role);
1201 LASSERT(obj->oo_owner == NULL);
1205 static void osd_object_write_lock(const struct lu_env *env,
1206 struct dt_object *dt, unsigned role)
1208 struct osd_object *obj = osd_dt_obj(dt);
1209 struct osd_thread_info *oti = osd_oti_get(env);
1211 LINVRNT(osd_invariant(obj));
1213 LASSERT(obj->oo_owner != env);
1214 down_write_nested(&obj->oo_sem, role);
1216 LASSERT(obj->oo_owner == NULL);
1217 obj->oo_owner = env;
1221 static void osd_object_read_unlock(const struct lu_env *env,
1222 struct dt_object *dt)
1224 struct osd_object *obj = osd_dt_obj(dt);
1225 struct osd_thread_info *oti = osd_oti_get(env);
1227 LINVRNT(osd_invariant(obj));
1229 LASSERT(oti->oti_r_locks > 0);
1231 up_read(&obj->oo_sem);
1234 static void osd_object_write_unlock(const struct lu_env *env,
1235 struct dt_object *dt)
1237 struct osd_object *obj = osd_dt_obj(dt);
1238 struct osd_thread_info *oti = osd_oti_get(env);
1240 LINVRNT(osd_invariant(obj));
1242 LASSERT(obj->oo_owner == env);
1243 LASSERT(oti->oti_w_locks > 0);
1245 obj->oo_owner = NULL;
1246 up_write(&obj->oo_sem);
1249 static int osd_object_write_locked(const struct lu_env *env,
1250 struct dt_object *dt)
1252 struct osd_object *obj = osd_dt_obj(dt);
1254 LINVRNT(osd_invariant(obj));
1256 return obj->oo_owner == env;
1259 static int capa_is_sane(const struct lu_env *env,
1260 struct osd_device *dev,
1261 struct lustre_capa *capa,
1262 struct lustre_capa_key *keys)
1264 struct osd_thread_info *oti = osd_oti_get(env);
1265 struct lustre_capa *tcapa = &oti->oti_capa;
1266 struct obd_capa *oc;
1270 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1272 if (capa_is_expired(oc)) {
1273 DEBUG_CAPA(D_ERROR, capa, "expired");
1280 if (capa_is_expired_sec(capa)) {
1281 DEBUG_CAPA(D_ERROR, capa, "expired");
1285 spin_lock(&capa_lock);
1286 for (i = 0; i < 2; i++) {
1287 if (keys[i].lk_keyid == capa->lc_keyid) {
1288 oti->oti_capa_key = keys[i];
1292 spin_unlock(&capa_lock);
1295 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1299 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1303 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1304 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1308 oc = capa_add(dev->od_capa_hash, capa);
1314 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1315 struct lustre_capa *capa, __u64 opc)
1317 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1318 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1319 struct md_capainfo *ci;
1322 if (!dev->od_fl_capa)
1325 if (capa == BYPASS_CAPA)
1328 ci = md_capainfo(env);
1332 if (ci->mc_auth == LC_ID_NONE)
1336 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1340 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1341 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1346 if (!capa_opc_supported(capa, opc)) {
1347 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1351 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1352 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1359 static struct timespec *osd_inode_time(const struct lu_env *env,
1360 struct inode *inode, __u64 seconds)
1362 struct osd_thread_info *oti = osd_oti_get(env);
1363 struct timespec *t = &oti->oti_time;
1365 t->tv_sec = seconds;
1367 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1372 static void osd_inode_getattr(const struct lu_env *env,
1373 struct inode *inode, struct lu_attr *attr)
1375 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1376 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1377 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1380 attr->la_atime = LTIME_S(inode->i_atime);
1381 attr->la_mtime = LTIME_S(inode->i_mtime);
1382 attr->la_ctime = LTIME_S(inode->i_ctime);
1383 attr->la_mode = inode->i_mode;
1384 attr->la_size = i_size_read(inode);
1385 attr->la_blocks = inode->i_blocks;
1386 attr->la_uid = inode->i_uid;
1387 attr->la_gid = inode->i_gid;
1388 attr->la_flags = LDISKFS_I(inode)->i_flags;
1389 attr->la_nlink = inode->i_nlink;
1390 attr->la_rdev = inode->i_rdev;
1391 attr->la_blksize = 1 << inode->i_blkbits;
1392 attr->la_blkbits = inode->i_blkbits;
1395 static int osd_attr_get(const struct lu_env *env,
1396 struct dt_object *dt,
1397 struct lu_attr *attr,
1398 struct lustre_capa *capa)
1400 struct osd_object *obj = osd_dt_obj(dt);
1402 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1403 LINVRNT(osd_invariant(obj));
1405 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1408 spin_lock(&obj->oo_guard);
1409 osd_inode_getattr(env, obj->oo_inode, attr);
1410 spin_unlock(&obj->oo_guard);
1414 static int osd_declare_attr_set(const struct lu_env *env,
1415 struct dt_object *dt,
1416 const struct lu_attr *attr,
1417 struct thandle *handle)
1419 struct osd_thandle *oh;
1420 struct osd_object *obj;
1421 struct osd_thread_info *info = osd_oti_get(env);
1422 struct lquota_id_info *qi = &info->oti_qi;
1428 LASSERT(dt != NULL);
1429 LASSERT(handle != NULL);
1431 obj = osd_dt_obj(dt);
1432 LASSERT(osd_invariant(obj));
1434 oh = container_of0(handle, struct osd_thandle, ot_super);
1435 LASSERT(oh->ot_handle == NULL);
1437 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1438 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1440 if (attr == NULL || obj->oo_inode == NULL)
1443 bspace = obj->oo_inode->i_blocks;
1444 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1445 bspace = toqb(bspace);
1447 /* Changing ownership is always preformed by super user, it should not
1450 * We still need to call the osd_declare_qid() to calculate the journal
1451 * credits for updating quota accounting files and to trigger quota
1452 * space adjustment once the operation is completed.*/
1453 if ((attr->la_valid & LA_UID) != 0 &&
1454 attr->la_uid != obj->oo_inode->i_uid) {
1455 qi->lqi_type = USRQUOTA;
1457 /* inode accounting */
1458 qi->lqi_is_blk = false;
1460 /* one more inode for the new owner ... */
1461 qi->lqi_id.qid_uid = attr->la_uid;
1463 allocated = (attr->la_uid == 0) ? true : false;
1464 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1465 if (rc == -EDQUOT || rc == -EINPROGRESS)
1470 /* and one less inode for the current uid */
1471 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1473 rc = osd_declare_qid(env, oh, qi, true, NULL);
1474 if (rc == -EDQUOT || rc == -EINPROGRESS)
1479 /* block accounting */
1480 qi->lqi_is_blk = true;
1482 /* more blocks for the new owner ... */
1483 qi->lqi_id.qid_uid = attr->la_uid;
1484 qi->lqi_space = bspace;
1485 allocated = (attr->la_uid == 0) ? true : false;
1486 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1487 if (rc == -EDQUOT || rc == -EINPROGRESS)
1492 /* and finally less blocks for the current owner */
1493 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1494 qi->lqi_space = -bspace;
1495 rc = osd_declare_qid(env, oh, qi, true, NULL);
1496 if (rc == -EDQUOT || rc == -EINPROGRESS)
1502 if (attr->la_valid & LA_GID &&
1503 attr->la_gid != obj->oo_inode->i_gid) {
1504 qi->lqi_type = GRPQUOTA;
1506 /* inode accounting */
1507 qi->lqi_is_blk = false;
1509 /* one more inode for the new group owner ... */
1510 qi->lqi_id.qid_gid = attr->la_gid;
1512 allocated = (attr->la_gid == 0) ? true : false;
1513 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1514 if (rc == -EDQUOT || rc == -EINPROGRESS)
1519 /* and one less inode for the current gid */
1520 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1522 rc = osd_declare_qid(env, oh, qi, true, NULL);
1523 if (rc == -EDQUOT || rc == -EINPROGRESS)
1528 /* block accounting */
1529 qi->lqi_is_blk = true;
1531 /* more blocks for the new owner ... */
1532 qi->lqi_id.qid_gid = attr->la_gid;
1533 qi->lqi_space = bspace;
1534 allocated = (attr->la_gid == 0) ? true : false;
1535 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1536 if (rc == -EDQUOT || rc == -EINPROGRESS)
1541 /* and finally less blocks for the current owner */
1542 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1543 qi->lqi_space = -bspace;
1544 rc = osd_declare_qid(env, oh, qi, true, NULL);
1545 if (rc == -EDQUOT || rc == -EINPROGRESS)
1554 static int osd_inode_setattr(const struct lu_env *env,
1555 struct inode *inode, const struct lu_attr *attr)
1559 bits = attr->la_valid;
1561 if (bits & LA_ATIME)
1562 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1563 if (bits & LA_CTIME)
1564 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1565 if (bits & LA_MTIME)
1566 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1567 if (bits & LA_SIZE) {
1568 LDISKFS_I(inode)->i_disksize = attr->la_size;
1569 i_size_write(inode, attr->la_size);
1573 /* OSD should not change "i_blocks" which is used by quota.
1574 * "i_blocks" should be changed by ldiskfs only. */
1575 if (bits & LA_BLOCKS)
1576 inode->i_blocks = attr->la_blocks;
1579 inode->i_mode = (inode->i_mode & S_IFMT) |
1580 (attr->la_mode & ~S_IFMT);
1582 inode->i_uid = attr->la_uid;
1584 inode->i_gid = attr->la_gid;
1585 if (bits & LA_NLINK)
1586 set_nlink(inode, attr->la_nlink);
1588 inode->i_rdev = attr->la_rdev;
1590 if (bits & LA_FLAGS) {
1591 /* always keep S_NOCMTIME */
1592 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1598 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1600 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1601 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1606 if (attr->la_valid & LA_UID)
1607 iattr.ia_valid |= ATTR_UID;
1608 if (attr->la_valid & LA_GID)
1609 iattr.ia_valid |= ATTR_GID;
1610 iattr.ia_uid = attr->la_uid;
1611 iattr.ia_gid = attr->la_gid;
1613 rc = ll_vfs_dq_transfer(inode, &iattr);
1615 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1616 "enforcement enabled on the ldiskfs filesystem?",
1617 inode->i_sb->s_id, rc);
1624 static int osd_attr_set(const struct lu_env *env,
1625 struct dt_object *dt,
1626 const struct lu_attr *attr,
1627 struct thandle *handle,
1628 struct lustre_capa *capa)
1630 struct osd_object *obj = osd_dt_obj(dt);
1631 struct inode *inode;
1634 LASSERT(handle != NULL);
1635 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1636 LASSERT(osd_invariant(obj));
1638 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1641 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1643 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1644 struct osd_thread_info *oti = osd_oti_get(env);
1645 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1646 struct lu_fid *fid1 = &oti->oti_fid;
1647 struct osd_inode_id *id = &oti->oti_id;
1648 struct iam_path_descr *ipd;
1649 struct iam_container *bag;
1650 struct osd_thandle *oh;
1653 fid_cpu_to_be(fid1, fid0);
1654 memset(id, 1, sizeof(*id));
1655 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1656 fid0)->oi_dir.od_container;
1657 ipd = osd_idx_ipd_get(env, bag);
1658 if (unlikely(ipd == NULL))
1661 oh = container_of0(handle, struct osd_thandle, ot_super);
1662 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1663 (const struct iam_rec *)id, ipd);
1664 osd_ipd_put(env, bag, ipd);
1665 return(rc > 0 ? 0 : rc);
1668 inode = obj->oo_inode;
1669 ll_vfs_dq_init(inode);
1671 rc = osd_quota_transfer(inode, attr);
1675 spin_lock(&obj->oo_guard);
1676 rc = osd_inode_setattr(env, inode, attr);
1677 spin_unlock(&obj->oo_guard);
1680 inode->i_sb->s_op->dirty_inode(inode);
1684 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1685 struct osd_object *obj,
1686 const char *name, const int namelen)
1688 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1691 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1693 struct dt_allocation_hint *hint,
1697 struct osd_device *osd = osd_obj2dev(obj);
1698 struct osd_thandle *oth;
1699 struct dt_object *parent = NULL;
1700 struct inode *inode;
1702 LINVRNT(osd_invariant(obj));
1703 LASSERT(obj->oo_inode == NULL);
1704 LASSERT(obj->oo_hl_head == NULL);
1706 if (S_ISDIR(mode) && ldiskfs_pdo) {
1707 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1708 if (obj->oo_hl_head == NULL)
1712 oth = container_of(th, struct osd_thandle, ot_super);
1713 LASSERT(oth->ot_handle->h_transaction != NULL);
1715 if (hint && hint->dah_parent)
1716 parent = hint->dah_parent;
1718 inode = ldiskfs_create_inode(oth->ot_handle,
1719 parent ? osd_dt_obj(parent)->oo_inode :
1720 osd_sb(osd)->s_root->d_inode,
1722 if (!IS_ERR(inode)) {
1723 /* Do not update file c/mtime in ldiskfs.
1724 * NB: don't need any lock because no contention at this
1726 inode->i_flags |= S_NOCMTIME;
1728 /* For new created object, it must be consistent,
1729 * and it is unnecessary to scrub against it. */
1730 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1731 obj->oo_inode = inode;
1734 if (obj->oo_hl_head != NULL) {
1735 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1736 obj->oo_hl_head = NULL;
1738 result = PTR_ERR(inode);
1740 LINVRNT(osd_invariant(obj));
1748 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1749 struct lu_attr *attr,
1750 struct dt_allocation_hint *hint,
1751 struct dt_object_format *dof,
1755 struct osd_thandle *oth;
1756 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1758 LASSERT(S_ISDIR(attr->la_mode));
1760 oth = container_of(th, struct osd_thandle, ot_super);
1761 LASSERT(oth->ot_handle->h_transaction != NULL);
1762 result = osd_mkfile(info, obj, mode, hint, th);
1767 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1768 struct lu_attr *attr,
1769 struct dt_allocation_hint *hint,
1770 struct dt_object_format *dof,
1774 struct osd_thandle *oth;
1775 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1777 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1779 LASSERT(S_ISREG(attr->la_mode));
1781 oth = container_of(th, struct osd_thandle, ot_super);
1782 LASSERT(oth->ot_handle->h_transaction != NULL);
1784 result = osd_mkfile(info, obj, mode, hint, th);
1786 LASSERT(obj->oo_inode != NULL);
1787 if (feat->dif_flags & DT_IND_VARKEY)
1788 result = iam_lvar_create(obj->oo_inode,
1789 feat->dif_keysize_max,
1791 feat->dif_recsize_max,
1794 result = iam_lfix_create(obj->oo_inode,
1795 feat->dif_keysize_max,
1797 feat->dif_recsize_max,
1804 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1805 struct lu_attr *attr,
1806 struct dt_allocation_hint *hint,
1807 struct dt_object_format *dof,
1810 LASSERT(S_ISREG(attr->la_mode));
1811 return osd_mkfile(info, obj, (attr->la_mode &
1812 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1815 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1816 struct lu_attr *attr,
1817 struct dt_allocation_hint *hint,
1818 struct dt_object_format *dof,
1821 LASSERT(S_ISLNK(attr->la_mode));
1822 return osd_mkfile(info, obj, (attr->la_mode &
1823 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1826 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1827 struct lu_attr *attr,
1828 struct dt_allocation_hint *hint,
1829 struct dt_object_format *dof,
1832 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1835 LINVRNT(osd_invariant(obj));
1836 LASSERT(obj->oo_inode == NULL);
1837 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1838 S_ISFIFO(mode) || S_ISSOCK(mode));
1840 result = osd_mkfile(info, obj, mode, hint, th);
1842 LASSERT(obj->oo_inode != NULL);
1844 * This inode should be marked dirty for i_rdev. Currently
1845 * that is done in the osd_attr_init().
1847 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1850 LINVRNT(osd_invariant(obj));
1854 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1856 struct dt_allocation_hint *hint,
1857 struct dt_object_format *dof,
1860 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1862 osd_obj_type_f result;
1878 result = osd_mk_index;
1889 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1890 struct dt_object *parent, struct dt_object *child,
1891 cfs_umode_t child_mode)
1895 memset(ah, 0, sizeof(*ah));
1896 ah->dah_parent = parent;
1897 ah->dah_mode = child_mode;
1900 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1901 struct lu_attr *attr, struct dt_object_format *dof)
1903 struct inode *inode = obj->oo_inode;
1904 __u64 valid = attr->la_valid;
1907 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1909 if (dof->dof_type != DFT_NODE)
1910 attr->la_valid &= ~LA_RDEV;
1911 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1912 attr->la_valid &= ~LA_ATIME;
1913 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1914 attr->la_valid &= ~LA_CTIME;
1915 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1916 attr->la_valid &= ~LA_MTIME;
1918 result = osd_quota_transfer(inode, attr);
1922 if (attr->la_valid != 0) {
1923 result = osd_inode_setattr(info->oti_env, inode, attr);
1925 * The osd_inode_setattr() should always succeed here. The
1926 * only error that could be returned is EDQUOT when we are
1927 * trying to change the UID or GID of the inode. However, this
1928 * should not happen since quota enforcement is no longer
1929 * enabled on ldiskfs (lquota takes care of it).
1931 LASSERTF(result == 0, "%d", result);
1932 inode->i_sb->s_op->dirty_inode(inode);
1935 attr->la_valid = valid;
1939 * Helper function for osd_object_create()
1941 * \retval 0, on success
1943 static int __osd_object_create(struct osd_thread_info *info,
1944 struct osd_object *obj, struct lu_attr *attr,
1945 struct dt_allocation_hint *hint,
1946 struct dt_object_format *dof,
1951 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1954 osd_attr_init(info, obj, attr, dof);
1955 osd_object_init0(obj);
1957 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1958 unlock_new_inode(obj->oo_inode);
1965 * Helper function for osd_object_create()
1967 * \retval 0, on success
1969 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1970 const struct lu_fid *fid, struct thandle *th)
1972 struct osd_thread_info *info = osd_oti_get(env);
1973 struct osd_inode_id *id = &info->oti_id;
1974 struct osd_device *osd = osd_obj2dev(obj);
1976 LASSERT(obj->oo_inode != NULL);
1978 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1979 return osd_oi_insert(info, osd, fid, id, th);
1982 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1983 const struct lu_fid *fid, struct lu_seq_range *range)
1985 struct seq_server_site *ss = osd_seq_site(osd);
1988 if (fid_is_idif(fid)) {
1989 range->lsr_flags = LU_SEQ_RANGE_OST;
1990 range->lsr_index = fid_idif_ost_idx(fid);
1994 if (!fid_seq_in_fldb(fid_seq(fid))) {
1995 range->lsr_flags = LU_SEQ_RANGE_MDT;
1997 /* FIXME: If ss is NULL, it suppose not get lsr_index
1999 range->lsr_index = ss->ss_node_id;
2003 LASSERT(ss != NULL);
2004 range->lsr_flags = -1;
2005 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2007 CERROR("%s can not find "DFID": rc = %d\n",
2008 osd_name(osd), PFID(fid), rc);
2014 * Concurrency: no external locking is necessary.
2016 static int osd_declare_object_create(const struct lu_env *env,
2017 struct dt_object *dt,
2018 struct lu_attr *attr,
2019 struct dt_allocation_hint *hint,
2020 struct dt_object_format *dof,
2021 struct thandle *handle)
2023 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2024 struct osd_thandle *oh;
2028 LASSERT(handle != NULL);
2030 oh = container_of0(handle, struct osd_thandle, ot_super);
2031 LASSERT(oh->ot_handle == NULL);
2033 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2034 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2035 if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2036 lu_object_fid(&dt->do_lu)))
2037 /* Reuse idle OI block may cause additional one OI block
2039 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2040 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2042 /* If this is directory, then we expect . and .. to be inserted as
2043 * well. The one directory block always needs to be created for the
2044 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2045 * block), there is no danger of needing a tree for the first block.
2047 if (attr && S_ISDIR(attr->la_mode)) {
2048 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2049 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2050 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2056 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2057 false, false, NULL, false);
2061 /* It does fld look up inside declare, and the result will be
2062 * added to fld cache, so the following fld lookup inside insert
2063 * does not need send RPC anymore, so avoid send rpc with holding
2065 if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2066 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2067 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2068 lu_object_fid(&dt->do_lu), range);
2074 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2075 struct lu_attr *attr,
2076 struct dt_allocation_hint *hint,
2077 struct dt_object_format *dof,
2080 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2081 struct osd_object *obj = osd_dt_obj(dt);
2082 struct osd_thread_info *info = osd_oti_get(env);
2087 LINVRNT(osd_invariant(obj));
2088 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2089 LASSERT(osd_write_locked(env, obj));
2090 LASSERT(th != NULL);
2092 if (unlikely(fid_is_acct(fid)))
2093 /* Quota files can't be created from the kernel any more,
2094 * 'tune2fs -O quota' will take care of creating them */
2097 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2098 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2100 result = __osd_object_create(info, obj, attr, hint, dof, th);
2102 result = __osd_oi_insert(env, obj, fid, th);
2104 LASSERT(ergo(result == 0,
2105 dt_object_exists(dt) && !dt_object_remote(dt)));
2107 LASSERT(osd_invariant(obj));
2112 * Called to destroy on-disk representation of the object
2114 * Concurrency: must be locked
2116 static int osd_declare_object_destroy(const struct lu_env *env,
2117 struct dt_object *dt,
2120 struct osd_object *obj = osd_dt_obj(dt);
2121 struct inode *inode = obj->oo_inode;
2122 struct osd_thandle *oh;
2126 oh = container_of0(th, struct osd_thandle, ot_super);
2127 LASSERT(oh->ot_handle == NULL);
2130 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2131 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2132 /* Recycle idle OI leaf may cause additional three OI blocks
2134 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2135 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2137 /* one less inode */
2138 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2139 false, true, NULL, false);
2142 /* data to be truncated */
2143 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2144 true, true, NULL, false);
2148 static int osd_object_destroy(const struct lu_env *env,
2149 struct dt_object *dt,
2152 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2153 struct osd_object *obj = osd_dt_obj(dt);
2154 struct inode *inode = obj->oo_inode;
2155 struct osd_device *osd = osd_obj2dev(obj);
2156 struct osd_thandle *oh;
2160 oh = container_of0(th, struct osd_thandle, ot_super);
2161 LASSERT(oh->ot_handle);
2163 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2165 if (unlikely(fid_is_acct(fid)))
2168 /* Parallel control for OI scrub. For most of cases, there is no
2169 * lock contention. So it will not affect unlink performance. */
2170 mutex_lock(&inode->i_mutex);
2171 if (S_ISDIR(inode->i_mode)) {
2172 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2173 /* it will check/delete the agent inode for every dir
2174 * destory, how to optimize it? unlink performance
2176 result = osd_delete_from_agent(env, osd, obj, oh);
2177 if (result != 0 && result != -ENOENT) {
2178 CERROR("%s: delete agent inode "DFID": rc = %d\n",
2179 osd_name(osd), PFID(fid), result);
2181 spin_lock(&obj->oo_guard);
2183 spin_unlock(&obj->oo_guard);
2184 inode->i_sb->s_op->dirty_inode(inode);
2187 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2189 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2190 mutex_unlock(&inode->i_mutex);
2192 /* XXX: add to ext3 orphan list */
2193 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2195 /* not needed in the cache anymore */
2196 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2201 static inline int __osd_xattr_set(struct osd_thread_info *info,
2202 struct inode *inode, const char *name,
2203 const void *buf, int buflen, int fl)
2205 struct dentry *dentry = &info->oti_child_dentry;
2207 ll_vfs_dq_init(inode);
2208 dentry->d_inode = inode;
2209 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
2213 * Put the fid into lustre_mdt_attrs, and then place the structure
2214 * inode's ea. This fid should not be altered during the life time
2217 * \retval +ve, on success
2218 * \retval -ve, on error
2220 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2222 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2223 const struct lu_fid *fid)
2225 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2228 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2231 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid))
2234 lustre_lma_init(lma, fid);
2235 lustre_lma_swab(lma);
2237 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2239 /* Someone may created the EA by race. */
2240 if (unlikely(rc == -EEXIST))
2246 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2247 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2248 * To have compatilibility with 1.8 ldiskfs driver we need to have
2249 * magic number at start of fid data.
2250 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2253 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2254 const struct dt_rec *fid)
2256 if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
2257 param->edp_magic = 0;
2261 param->edp_magic = LDISKFS_LUFID_MAGIC;
2262 param->edp_len = sizeof(struct lu_fid) + 1;
2263 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2267 * Try to read the fid from inode ea into dt_rec.
2269 * \param fid object fid.
2271 * \retval 0 on success
2273 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2274 __u32 ino, struct lu_fid *fid,
2275 struct osd_inode_id *id)
2277 struct osd_thread_info *info = osd_oti_get(env);
2278 struct inode *inode;
2281 osd_id_gen(id, ino, OSD_OII_NOGEN);
2282 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2284 RETURN(PTR_ERR(inode));
2290 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2292 struct inode *parent_dir,
2293 const struct dt_rec *dot_fid,
2294 const struct dt_rec *dot_dot_fid,
2295 struct osd_thandle *oth)
2297 struct ldiskfs_dentry_param *dot_ldp;
2298 struct ldiskfs_dentry_param *dot_dot_ldp;
2300 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2301 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2303 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2304 dot_ldp->edp_magic = 0;
2305 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2306 dir, dot_ldp, dot_dot_ldp);
2310 * Create an local inode for remote entry
2312 static struct inode *osd_create_remote_inode(const struct lu_env *env,
2313 struct osd_device *osd,
2314 struct osd_object *pobj,
2315 const struct lu_fid *fid,
2318 struct osd_thread_info *info = osd_oti_get(env);
2319 struct inode *local;
2320 struct osd_thandle *oh;
2325 oh = container_of(th, struct osd_thandle, ot_super);
2326 LASSERT(oh->ot_handle->h_transaction != NULL);
2328 /* FIXME: Insert index api needs to know the mode of
2329 * the remote object. Just use S_IFDIR for now */
2330 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2331 if (IS_ERR(local)) {
2332 CERROR("%s: create local error %d\n", osd_name(osd),
2333 (int)PTR_ERR(local));
2337 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2338 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2339 (const struct dt_rec *)fid, oh);
2341 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2342 osd_name(osd), PFID(fid), rc);
2343 RETURN(ERR_PTR(rc));
2350 * Delete local inode for remote entry
2352 static int osd_delete_remote_inode(const struct lu_env *env,
2353 struct osd_device *osd,
2354 const struct lu_fid *fid,
2355 __u32 ino, struct osd_thandle *oh)
2357 struct osd_thread_info *oti = osd_oti_get(env);
2358 struct osd_inode_id *id = &oti->oti_id;
2359 struct inode *inode;
2362 id->oii_ino = le32_to_cpu(ino);
2363 id->oii_gen = OSD_OII_NOGEN;
2364 inode = osd_iget(oti, osd, id);
2365 if (IS_ERR(inode)) {
2366 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2367 PFID(fid), id->oii_ino, id->oii_gen);
2368 RETURN(PTR_ERR(inode));
2372 mark_inode_dirty(inode);
2373 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2374 osd_name(osd), PFID(fid), inode->i_ino);
2380 * OSD layer object create function for interoperability mode (b11826).
2381 * This is mostly similar to osd_object_create(). Only difference being, fid is
2382 * inserted into inode ea here.
2384 * \retval 0, on success
2385 * \retval -ve, on error
2387 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2388 struct lu_attr *attr,
2389 struct dt_allocation_hint *hint,
2390 struct dt_object_format *dof,
2393 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2394 struct osd_object *obj = osd_dt_obj(dt);
2395 struct osd_thread_info *info = osd_oti_get(env);
2400 LASSERT(osd_invariant(obj));
2401 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2402 LASSERT(osd_write_locked(env, obj));
2403 LASSERT(th != NULL);
2405 if (unlikely(fid_is_acct(fid)))
2406 /* Quota files can't be created from the kernel any more,
2407 * 'tune2fs -O quota' will take care of creating them */
2410 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2411 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2413 result = __osd_object_create(info, obj, attr, hint, dof, th);
2414 if ((result == 0) &&
2415 (fid_is_last_id(fid) ||
2416 !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
2417 result = osd_ea_fid_set(info, obj->oo_inode, fid);
2420 result = __osd_oi_insert(env, obj, fid, th);
2422 LASSERT(ergo(result == 0,
2423 dt_object_exists(dt) && !dt_object_remote(dt)));
2424 LINVRNT(osd_invariant(obj));
2428 static int osd_declare_object_ref_add(const struct lu_env *env,
2429 struct dt_object *dt,
2430 struct thandle *handle)
2432 struct osd_thandle *oh;
2434 /* it's possible that object doesn't exist yet */
2435 LASSERT(handle != NULL);
2437 oh = container_of0(handle, struct osd_thandle, ot_super);
2438 LASSERT(oh->ot_handle == NULL);
2440 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2441 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2447 * Concurrency: @dt is write locked.
2449 static int osd_object_ref_add(const struct lu_env *env,
2450 struct dt_object *dt, struct thandle *th)
2452 struct osd_object *obj = osd_dt_obj(dt);
2453 struct inode *inode = obj->oo_inode;
2455 LINVRNT(osd_invariant(obj));
2456 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2457 LASSERT(osd_write_locked(env, obj));
2458 LASSERT(th != NULL);
2460 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2463 * DIR_NLINK feature is set for compatibility reasons if:
2464 * 1) nlinks > LDISKFS_LINK_MAX, or
2465 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2467 * It is easier to always set this flag (rather than check and set),
2468 * since it has less overhead, and the superblock will be dirtied
2469 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2470 * do not actually care whether this flag is set or not.
2472 spin_lock(&obj->oo_guard);
2473 /* inc_nlink from 0 may cause WARN_ON */
2474 if(inode->i_nlink == 0)
2475 set_nlink(inode, 1);
2478 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2479 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2480 inode->i_nlink == 2)
2481 set_nlink(inode, 1);
2483 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2484 spin_unlock(&obj->oo_guard);
2485 inode->i_sb->s_op->dirty_inode(inode);
2486 LINVRNT(osd_invariant(obj));
2491 static int osd_declare_object_ref_del(const struct lu_env *env,
2492 struct dt_object *dt,
2493 struct thandle *handle)
2495 struct osd_thandle *oh;
2497 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2498 LASSERT(handle != NULL);
2500 oh = container_of0(handle, struct osd_thandle, ot_super);
2501 LASSERT(oh->ot_handle == NULL);
2503 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2504 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2510 * Concurrency: @dt is write locked.
2512 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2515 struct osd_object *obj = osd_dt_obj(dt);
2516 struct inode *inode = obj->oo_inode;
2518 LINVRNT(osd_invariant(obj));
2519 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2520 LASSERT(osd_write_locked(env, obj));
2521 LASSERT(th != NULL);
2523 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2525 spin_lock(&obj->oo_guard);
2526 LASSERT(inode->i_nlink > 0);
2528 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2529 * then the nlink count is 1. Don't let it be set to 0 or the directory
2530 * inode will be deleted incorrectly. */
2531 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2532 set_nlink(inode, 1);
2533 spin_unlock(&obj->oo_guard);
2534 inode->i_sb->s_op->dirty_inode(inode);
2535 LINVRNT(osd_invariant(obj));
2541 * Get the 64-bit version for an inode.
2543 static int osd_object_version_get(const struct lu_env *env,
2544 struct dt_object *dt, dt_obj_version_t *ver)
2546 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2548 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2549 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2550 *ver = LDISKFS_I(inode)->i_fs_version;
2555 * Concurrency: @dt is read locked.
2557 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2558 struct lu_buf *buf, const char *name,
2559 struct lustre_capa *capa)
2561 struct osd_object *obj = osd_dt_obj(dt);
2562 struct inode *inode = obj->oo_inode;
2563 struct osd_thread_info *info = osd_oti_get(env);
2564 struct dentry *dentry = &info->oti_obj_dentry;
2566 /* version get is not real XATTR but uses xattr API */
2567 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2568 /* for version we are just using xattr API but change inode
2570 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2571 osd_object_version_get(env, dt, buf->lb_buf);
2572 return sizeof(dt_obj_version_t);
2575 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2576 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2578 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2581 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2585 static int osd_declare_xattr_set(const struct lu_env *env,
2586 struct dt_object *dt,
2587 const struct lu_buf *buf, const char *name,
2588 int fl, struct thandle *handle)
2590 struct osd_thandle *oh;
2592 LASSERT(handle != NULL);
2594 oh = container_of0(handle, struct osd_thandle, ot_super);
2595 LASSERT(oh->ot_handle == NULL);
2597 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2598 strcmp(name, XATTR_NAME_VERSION) == 0 ?
2599 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2600 osd_dto_credits_noquota[DTO_XATTR_SET]);
2606 * Set the 64-bit version for object
2608 static void osd_object_version_set(const struct lu_env *env,
2609 struct dt_object *dt,
2610 dt_obj_version_t *new_version)
2612 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2614 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2615 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2617 LDISKFS_I(inode)->i_fs_version = *new_version;
2618 /** Version is set after all inode operations are finished,
2619 * so we should mark it dirty here */
2620 inode->i_sb->s_op->dirty_inode(inode);
2624 * Concurrency: @dt is write locked.
2626 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2627 const struct lu_buf *buf, const char *name, int fl,
2628 struct thandle *handle, struct lustre_capa *capa)
2630 struct osd_object *obj = osd_dt_obj(dt);
2631 struct inode *inode = obj->oo_inode;
2632 struct osd_thread_info *info = osd_oti_get(env);
2635 LASSERT(handle != NULL);
2637 /* version set is not real XATTR */
2638 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2639 /* for version we are just using xattr API but change inode
2641 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2642 osd_object_version_set(env, dt, buf->lb_buf);
2643 return sizeof(dt_obj_version_t);
2646 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2649 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2650 if (fl & LU_XATTR_REPLACE)
2651 fs_flags |= XATTR_REPLACE;
2653 if (fl & LU_XATTR_CREATE)
2654 fs_flags |= XATTR_CREATE;
2656 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2661 * Concurrency: @dt is read locked.
2663 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2664 struct lu_buf *buf, struct lustre_capa *capa)
2666 struct osd_object *obj = osd_dt_obj(dt);
2667 struct inode *inode = obj->oo_inode;
2668 struct osd_thread_info *info = osd_oti_get(env);
2669 struct dentry *dentry = &info->oti_obj_dentry;
2671 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2672 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2673 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2675 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2678 dentry->d_inode = inode;
2679 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2682 static int osd_declare_xattr_del(const struct lu_env *env,
2683 struct dt_object *dt, const char *name,
2684 struct thandle *handle)
2686 struct osd_thandle *oh;
2688 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2689 LASSERT(handle != NULL);
2691 oh = container_of0(handle, struct osd_thandle, ot_super);
2692 LASSERT(oh->ot_handle == NULL);
2694 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2695 osd_dto_credits_noquota[DTO_XATTR_SET]);
2701 * Concurrency: @dt is write locked.
2703 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2704 const char *name, struct thandle *handle,
2705 struct lustre_capa *capa)
2707 struct osd_object *obj = osd_dt_obj(dt);
2708 struct inode *inode = obj->oo_inode;
2709 struct osd_thread_info *info = osd_oti_get(env);
2710 struct dentry *dentry = &info->oti_obj_dentry;
2713 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2714 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2715 LASSERT(handle != NULL);
2717 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2720 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2722 ll_vfs_dq_init(inode);
2723 dentry->d_inode = inode;
2724 rc = inode->i_op->removexattr(dentry, name);
2728 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2729 struct dt_object *dt,
2730 struct lustre_capa *old,
2733 struct osd_thread_info *info = osd_oti_get(env);
2734 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2735 struct osd_object *obj = osd_dt_obj(dt);
2736 struct osd_device *dev = osd_obj2dev(obj);
2737 struct lustre_capa_key *key = &info->oti_capa_key;
2738 struct lustre_capa *capa = &info->oti_capa;
2739 struct obd_capa *oc;
2740 struct md_capainfo *ci;
2744 if (!dev->od_fl_capa)
2745 RETURN(ERR_PTR(-ENOENT));
2747 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2748 LINVRNT(osd_invariant(obj));
2750 /* renewal sanity check */
2751 if (old && osd_object_auth(env, dt, old, opc))
2752 RETURN(ERR_PTR(-EACCES));
2754 ci = md_capainfo(env);
2756 RETURN(ERR_PTR(-ENOENT));
2758 switch (ci->mc_auth) {
2762 capa->lc_uid = obj->oo_inode->i_uid;
2763 capa->lc_gid = obj->oo_inode->i_gid;
2764 capa->lc_flags = LC_ID_PLAIN;
2766 case LC_ID_CONVERT: {
2769 s[0] = obj->oo_inode->i_uid;
2770 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2771 s[2] = obj->oo_inode->i_gid;
2772 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2773 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2775 RETURN(ERR_PTR(rc));
2777 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2778 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2779 capa->lc_flags = LC_ID_CONVERT;
2783 RETURN(ERR_PTR(-EINVAL));
2786 capa->lc_fid = *fid;
2788 capa->lc_flags |= dev->od_capa_alg << 24;
2789 capa->lc_timeout = dev->od_capa_timeout;
2790 capa->lc_expiry = 0;
2792 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2794 LASSERT(!capa_is_expired(oc));
2798 spin_lock(&capa_lock);
2799 *key = dev->od_capa_keys[1];
2800 spin_unlock(&capa_lock);
2802 capa->lc_keyid = key->lk_keyid;
2803 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2805 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2807 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2808 RETURN(ERR_PTR(rc));
2811 oc = capa_add(dev->od_capa_hash, capa);
2815 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2817 struct osd_object *obj = osd_dt_obj(dt);
2818 struct inode *inode = obj->oo_inode;
2819 struct osd_thread_info *info = osd_oti_get(env);
2820 struct dentry *dentry = &info->oti_obj_dentry;
2821 struct file *file = &info->oti_file;
2826 dentry->d_inode = inode;
2827 file->f_dentry = dentry;
2828 file->f_mapping = inode->i_mapping;
2829 file->f_op = inode->i_fop;
2830 #ifndef HAVE_FILE_FSYNC_4ARGS
2831 mutex_lock(&inode->i_mutex);
2833 rc = cfs_do_fsync(file, 0);
2834 #ifndef HAVE_FILE_FSYNC_4ARGS
2835 mutex_unlock(&inode->i_mutex);
2840 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2843 struct osd_object *obj = osd_dt_obj(dt);
2846 *data = (void *)obj->oo_inode;
2854 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2855 const struct dt_index_features *feat)
2857 struct iam_descr *descr;
2859 if (osd_object_is_root(o))
2860 return feat == &dt_directory_features;
2862 LASSERT(o->oo_dir != NULL);
2864 descr = o->oo_dir->od_container.ic_descr;
2865 if (feat == &dt_directory_features) {
2866 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2872 feat->dif_keysize_min <= descr->id_key_size &&
2873 descr->id_key_size <= feat->dif_keysize_max &&
2874 feat->dif_recsize_min <= descr->id_rec_size &&
2875 descr->id_rec_size <= feat->dif_recsize_max &&
2876 !(feat->dif_flags & (DT_IND_VARKEY |
2877 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2878 ergo(feat->dif_flags & DT_IND_UPDATE,
2879 1 /* XXX check that object (and file system) is
2884 static int osd_iam_container_init(const struct lu_env *env,
2885 struct osd_object *obj,
2886 struct osd_directory *dir)
2888 struct iam_container *bag = &dir->od_container;
2891 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2895 result = iam_container_setup(bag);
2897 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2899 iam_container_fini(bag);
2906 * Concurrency: no external locking is necessary.
2908 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2909 const struct dt_index_features *feat)
2913 struct osd_object *obj = osd_dt_obj(dt);
2915 LINVRNT(osd_invariant(obj));
2917 if (osd_object_is_root(obj)) {
2918 dt->do_index_ops = &osd_index_ea_ops;
2920 } else if (feat == &dt_directory_features) {
2921 dt->do_index_ops = &osd_index_ea_ops;
2922 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2927 } else if (unlikely(feat == &dt_otable_features)) {
2928 dt->do_index_ops = &osd_otable_ops;
2930 } else if (unlikely(feat == &dt_acct_features)) {
2931 dt->do_index_ops = &osd_acct_index_ops;
2934 } else if (!osd_has_index(obj)) {
2935 struct osd_directory *dir;
2940 spin_lock(&obj->oo_guard);
2941 if (obj->oo_dir == NULL)
2945 * Concurrent thread allocated container data.
2948 spin_unlock(&obj->oo_guard);
2950 * Now, that we have container data, serialize its
2953 down_write(&obj->oo_ext_idx_sem);
2955 * recheck under lock.
2957 if (!osd_has_index(obj))
2958 result = osd_iam_container_init(env, obj, dir);
2961 up_write(&obj->oo_ext_idx_sem);
2969 if (result == 0 && skip_iam == 0) {
2970 if (!osd_iam_index_probe(env, obj, feat))
2973 LINVRNT(osd_invariant(obj));
2975 if (result == 0 && is_quota_glb_feat(feat) &&
2976 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
2977 result = osd_quota_migration(env, dt, feat);
2982 static int osd_otable_it_attr_get(const struct lu_env *env,
2983 struct dt_object *dt,
2984 struct lu_attr *attr,
2985 struct lustre_capa *capa)
2991 static const struct dt_object_operations osd_obj_ops = {
2992 .do_read_lock = osd_object_read_lock,
2993 .do_write_lock = osd_object_write_lock,
2994 .do_read_unlock = osd_object_read_unlock,
2995 .do_write_unlock = osd_object_write_unlock,
2996 .do_write_locked = osd_object_write_locked,
2997 .do_attr_get = osd_attr_get,
2998 .do_declare_attr_set = osd_declare_attr_set,
2999 .do_attr_set = osd_attr_set,
3000 .do_ah_init = osd_ah_init,
3001 .do_declare_create = osd_declare_object_create,
3002 .do_create = osd_object_create,
3003 .do_declare_destroy = osd_declare_object_destroy,
3004 .do_destroy = osd_object_destroy,
3005 .do_index_try = osd_index_try,
3006 .do_declare_ref_add = osd_declare_object_ref_add,
3007 .do_ref_add = osd_object_ref_add,
3008 .do_declare_ref_del = osd_declare_object_ref_del,
3009 .do_ref_del = osd_object_ref_del,
3010 .do_xattr_get = osd_xattr_get,
3011 .do_declare_xattr_set = osd_declare_xattr_set,
3012 .do_xattr_set = osd_xattr_set,
3013 .do_declare_xattr_del = osd_declare_xattr_del,
3014 .do_xattr_del = osd_xattr_del,
3015 .do_xattr_list = osd_xattr_list,
3016 .do_capa_get = osd_capa_get,
3017 .do_object_sync = osd_object_sync,
3018 .do_data_get = osd_data_get,
3022 * dt_object_operations for interoperability mode
3023 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3025 static const struct dt_object_operations osd_obj_ea_ops = {
3026 .do_read_lock = osd_object_read_lock,
3027 .do_write_lock = osd_object_write_lock,
3028 .do_read_unlock = osd_object_read_unlock,
3029 .do_write_unlock = osd_object_write_unlock,
3030 .do_write_locked = osd_object_write_locked,
3031 .do_attr_get = osd_attr_get,
3032 .do_declare_attr_set = osd_declare_attr_set,
3033 .do_attr_set = osd_attr_set,
3034 .do_ah_init = osd_ah_init,
3035 .do_declare_create = osd_declare_object_create,
3036 .do_create = osd_object_ea_create,
3037 .do_declare_destroy = osd_declare_object_destroy,
3038 .do_destroy = osd_object_destroy,
3039 .do_index_try = osd_index_try,
3040 .do_declare_ref_add = osd_declare_object_ref_add,
3041 .do_ref_add = osd_object_ref_add,
3042 .do_declare_ref_del = osd_declare_object_ref_del,
3043 .do_ref_del = osd_object_ref_del,
3044 .do_xattr_get = osd_xattr_get,
3045 .do_declare_xattr_set = osd_declare_xattr_set,
3046 .do_xattr_set = osd_xattr_set,
3047 .do_declare_xattr_del = osd_declare_xattr_del,
3048 .do_xattr_del = osd_xattr_del,
3049 .do_xattr_list = osd_xattr_list,
3050 .do_capa_get = osd_capa_get,
3051 .do_object_sync = osd_object_sync,
3052 .do_data_get = osd_data_get,
3055 static const struct dt_object_operations osd_obj_otable_it_ops = {
3056 .do_attr_get = osd_otable_it_attr_get,
3057 .do_index_try = osd_index_try,
3060 static int osd_index_declare_iam_delete(const struct lu_env *env,
3061 struct dt_object *dt,
3062 const struct dt_key *key,
3063 struct thandle *handle)
3065 struct osd_thandle *oh;
3067 oh = container_of0(handle, struct osd_thandle, ot_super);
3068 LASSERT(oh->ot_handle == NULL);
3070 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3071 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3077 * delete a (key, value) pair from index \a dt specified by \a key
3079 * \param dt osd index object
3080 * \param key key for index
3081 * \param rec record reference
3082 * \param handle transaction handler
3085 * \retval -ve failure
3088 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3089 const struct dt_key *key,
3090 struct thandle *handle,
3091 struct lustre_capa *capa)
3093 struct osd_thread_info *oti = osd_oti_get(env);
3094 struct osd_object *obj = osd_dt_obj(dt);
3095 struct osd_thandle *oh;
3096 struct iam_path_descr *ipd;
3097 struct iam_container *bag = &obj->oo_dir->od_container;
3102 LINVRNT(osd_invariant(obj));
3103 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3104 LASSERT(bag->ic_object == obj->oo_inode);
3105 LASSERT(handle != NULL);
3107 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3110 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3112 ipd = osd_idx_ipd_get(env, bag);
3113 if (unlikely(ipd == NULL))
3116 oh = container_of0(handle, struct osd_thandle, ot_super);
3117 LASSERT(oh->ot_handle != NULL);
3118 LASSERT(oh->ot_handle->h_transaction != NULL);
3120 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3121 /* swab quota uid/gid provided by caller */
3122 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3123 key = (const struct dt_key *)&oti->oti_quota_id;
3126 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3127 osd_ipd_put(env, bag, ipd);
3128 LINVRNT(osd_invariant(obj));
3132 static int osd_index_declare_ea_delete(const struct lu_env *env,
3133 struct dt_object *dt,
3134 const struct dt_key *key,
3135 struct thandle *handle)
3137 struct osd_thandle *oh;
3138 struct inode *inode;
3142 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3143 LASSERT(handle != NULL);
3145 oh = container_of0(handle, struct osd_thandle, ot_super);
3146 LASSERT(oh->ot_handle == NULL);
3148 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3149 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3151 inode = osd_dt_obj(dt)->oo_inode;
3154 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3155 true, true, NULL, false);
3159 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3162 struct osd_fid_pack *rec;
3165 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3166 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3167 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3172 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3175 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
3176 struct seq_server_site *ss = osd_seq_site(osd);
3180 /* Those FID seqs, which are not in FLDB, must be local seq */
3181 if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3184 rc = osd_fld_lookup(env, osd, fid, range);
3186 CERROR("%s: Can not lookup fld for "DFID"\n",
3187 osd_name(osd), PFID(fid));
3191 RETURN(ss->ss_node_id != range->lsr_index);
3195 * Index delete function for interoperability mode (b11826).
3196 * It will remove the directory entry added by osd_index_ea_insert().
3197 * This entry is needed to maintain name->fid mapping.
3199 * \param key, key i.e. file entry to be deleted
3201 * \retval 0, on success
3202 * \retval -ve, on error
3204 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3205 const struct dt_key *key,
3206 struct thandle *handle,
3207 struct lustre_capa *capa)
3209 struct osd_object *obj = osd_dt_obj(dt);
3210 struct inode *dir = obj->oo_inode;
3211 struct dentry *dentry;
3212 struct osd_thandle *oh;
3213 struct ldiskfs_dir_entry_2 *de = NULL;
3214 struct buffer_head *bh;
3215 struct htree_lock *hlock = NULL;
3216 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3217 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3221 LINVRNT(osd_invariant(obj));
3222 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3223 LASSERT(handle != NULL);
3225 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3227 oh = container_of(handle, struct osd_thandle, ot_super);
3228 LASSERT(oh->ot_handle != NULL);
3229 LASSERT(oh->ot_handle->h_transaction != NULL);
3231 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3234 ll_vfs_dq_init(dir);
3235 dentry = osd_child_dentry_get(env, obj,
3236 (char *)key, strlen((char *)key));
3238 if (obj->oo_hl_head != NULL) {
3239 hlock = osd_oti_get(env)->oti_hlock;
3240 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3241 dir, LDISKFS_HLOCK_DEL);
3243 down_write(&obj->oo_ext_idx_sem);
3246 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3248 rc = ldiskfs_delete_entry(oh->ot_handle,
3255 ldiskfs_htree_unlock(hlock);
3257 up_write(&obj->oo_ext_idx_sem);
3262 /* For inode on the remote MDT, .. will point to
3263 * /Agent directory. So do not try to lookup/delete
3264 * remote inode for .. */
3265 if (strcmp((char *)key, dotdot) == 0)
3268 LASSERT(de != NULL);
3269 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3270 if (rc == 0 && osd_remote_fid(env, osd, fid)) {
3271 __u32 ino = le32_to_cpu(de->inode);
3273 rc = osd_delete_remote_inode(env, osd, fid, ino, oh);
3275 CERROR("%s: del local inode "DFID": rc = %d\n",
3276 osd_name(osd), PFID(fid), rc);
3283 LASSERT(osd_invariant(obj));
3288 * Lookup index for \a key and copy record to \a rec.
3290 * \param dt osd index object
3291 * \param key key for index
3292 * \param rec record reference
3294 * \retval +ve success : exact mach
3295 * \retval 0 return record with key not greater than \a key
3296 * \retval -ve failure
3298 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3299 struct dt_rec *rec, const struct dt_key *key,
3300 struct lustre_capa *capa)
3302 struct osd_object *obj = osd_dt_obj(dt);
3303 struct iam_path_descr *ipd;
3304 struct iam_container *bag = &obj->oo_dir->od_container;
3305 struct osd_thread_info *oti = osd_oti_get(env);
3306 struct iam_iterator *it = &oti->oti_idx_it;
3307 struct iam_rec *iam_rec;
3312 LASSERT(osd_invariant(obj));
3313 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3314 LASSERT(bag->ic_object == obj->oo_inode);
3316 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3319 ipd = osd_idx_ipd_get(env, bag);
3323 /* got ipd now we can start iterator. */
3324 iam_it_init(it, bag, 0, ipd);
3326 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3327 /* swab quota uid/gid provided by caller */
3328 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3329 key = (const struct dt_key *)&oti->oti_quota_id;
3332 rc = iam_it_get(it, (struct iam_key *)key);
3334 if (S_ISDIR(obj->oo_inode->i_mode))
3335 iam_rec = (struct iam_rec *)oti->oti_ldp;
3337 iam_rec = (struct iam_rec *) rec;
3339 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3341 if (S_ISDIR(obj->oo_inode->i_mode))
3342 osd_fid_unpack((struct lu_fid *) rec,
3343 (struct osd_fid_pack *)iam_rec);
3344 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3345 osd_quota_unpack(obj, rec);
3350 osd_ipd_put(env, bag, ipd);
3352 LINVRNT(osd_invariant(obj));
3357 static int osd_index_declare_iam_insert(const struct lu_env *env,
3358 struct dt_object *dt,
3359 const struct dt_rec *rec,
3360 const struct dt_key *key,
3361 struct thandle *handle)
3363 struct osd_thandle *oh;
3365 LASSERT(handle != NULL);
3367 oh = container_of0(handle, struct osd_thandle, ot_super);
3368 LASSERT(oh->ot_handle == NULL);
3370 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3371 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3377 * Inserts (key, value) pair in \a dt index object.
3379 * \param dt osd index object
3380 * \param key key for index
3381 * \param rec record reference
3382 * \param th transaction handler
3385 * \retval -ve failure
3387 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3388 const struct dt_rec *rec,
3389 const struct dt_key *key, struct thandle *th,
3390 struct lustre_capa *capa, int ignore_quota)
3392 struct osd_object *obj = osd_dt_obj(dt);
3393 struct iam_path_descr *ipd;
3394 struct osd_thandle *oh;
3395 struct iam_container *bag = &obj->oo_dir->od_container;
3396 struct osd_thread_info *oti = osd_oti_get(env);
3397 struct iam_rec *iam_rec;
3402 LINVRNT(osd_invariant(obj));
3403 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3404 LASSERT(bag->ic_object == obj->oo_inode);
3405 LASSERT(th != NULL);
3407 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3410 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3412 ipd = osd_idx_ipd_get(env, bag);
3413 if (unlikely(ipd == NULL))
3416 oh = container_of0(th, struct osd_thandle, ot_super);
3417 LASSERT(oh->ot_handle != NULL);
3418 LASSERT(oh->ot_handle->h_transaction != NULL);
3419 if (S_ISDIR(obj->oo_inode->i_mode)) {
3420 iam_rec = (struct iam_rec *)oti->oti_ldp;
3421 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3422 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3423 /* pack quota uid/gid */
3424 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3425 key = (const struct dt_key *)&oti->oti_quota_id;
3426 /* pack quota record */
3427 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3428 iam_rec = (struct iam_rec *)rec;
3430 iam_rec = (struct iam_rec *)rec;
3433 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3435 osd_ipd_put(env, bag, ipd);
3436 LINVRNT(osd_invariant(obj));
3441 * Calls ldiskfs_add_entry() to add directory entry
3442 * into the directory. This is required for
3443 * interoperability mode (b11826)
3445 * \retval 0, on success
3446 * \retval -ve, on error
3448 static int __osd_ea_add_rec(struct osd_thread_info *info,
3449 struct osd_object *pobj, struct inode *cinode,
3450 const char *name, const struct dt_rec *fid,
3451 struct htree_lock *hlock, struct thandle *th)
3453 struct ldiskfs_dentry_param *ldp;
3454 struct dentry *child;
3455 struct osd_thandle *oth;
3458 oth = container_of(th, struct osd_thandle, ot_super);
3459 LASSERT(oth->ot_handle != NULL);
3460 LASSERT(oth->ot_handle->h_transaction != NULL);
3461 LASSERT(pobj->oo_inode);
3463 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3464 if (unlikely(pobj->oo_inode ==
3465 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3468 osd_get_ldiskfs_dirent_param(ldp, fid);
3469 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3470 child->d_fsdata = (void *)ldp;
3471 ll_vfs_dq_init(pobj->oo_inode);
3472 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3478 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3479 * into the directory.Also sets flags into osd object to
3480 * indicate dot and dotdot are created. This is required for
3481 * interoperability mode (b11826)
3483 * \param dir directory for dot and dotdot fixup.
3484 * \param obj child object for linking
3486 * \retval 0, on success
3487 * \retval -ve, on error
3489 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3490 struct osd_object *dir,
3491 struct inode *parent_dir, const char *name,
3492 const struct dt_rec *dot_fid,
3493 const struct dt_rec *dot_dot_fid,
3496 struct inode *inode = dir->oo_inode;
3497 struct osd_thandle *oth;
3500 oth = container_of(th, struct osd_thandle, ot_super);
3501 LASSERT(oth->ot_handle->h_transaction != NULL);
3502 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3504 if (strcmp(name, dot) == 0) {
3505 if (dir->oo_compat_dot_created) {
3508 LASSERT(inode == parent_dir);
3509 dir->oo_compat_dot_created = 1;
3512 } else if (strcmp(name, dotdot) == 0) {
3513 if (!dir->oo_compat_dot_created)
3515 /* in case of rename, dotdot is already created */
3516 if (dir->oo_compat_dotdot_created) {
3517 return __osd_ea_add_rec(info, dir, parent_dir, name,
3518 dot_dot_fid, NULL, th);
3521 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3522 parent_dir, dot_fid,
3525 dir->oo_compat_dotdot_created = 1;
3533 * It will call the appropriate osd_add* function and return the
3534 * value, return by respective functions.
3536 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3537 struct inode *cinode, const char *name,
3538 const struct dt_rec *fid, struct thandle *th)
3540 struct osd_thread_info *info = osd_oti_get(env);
3541 struct htree_lock *hlock;
3544 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3546 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3548 if (hlock != NULL) {
3549 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3552 down_write(&pobj->oo_ext_idx_sem);
3554 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3555 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3558 if (hlock != NULL) {
3559 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3560 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3562 down_write(&pobj->oo_ext_idx_sem);
3565 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3566 struct lu_fid *tfid = &info->oti_fid;
3568 *tfid = *(const struct lu_fid *)fid;
3570 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3571 (const struct dt_rec *)tfid,
3574 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3579 ldiskfs_htree_unlock(hlock);
3581 up_write(&pobj->oo_ext_idx_sem);
3587 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3588 struct osd_idmap_cache *oic)
3590 struct osd_scrub *scrub = &dev->od_scrub;
3591 struct lu_fid *fid = &oic->oic_fid;
3592 struct osd_inode_id *id = &oti->oti_id;
3597 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3601 rc = osd_oi_lookup(oti, dev, fid, id, true);
3602 if (rc != 0 && rc != -ENOENT)
3605 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3608 if (thread_is_running(&scrub->os_thread)) {
3609 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3610 /* There is race condition between osd_oi_lookup and OI scrub.
3611 * The OI scrub finished just after osd_oi_lookup() failure.
3612 * Under such case, it is unnecessary to trigger OI scrub again,
3613 * but try to call osd_oi_lookup() again. */
3614 if (unlikely(rc == -EAGAIN))
3620 if (!dev->od_noscrub && ++once == 1) {
3621 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3623 rc = osd_scrub_start(dev);
3624 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3626 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3635 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3636 struct osd_device *dev,
3637 struct osd_idmap_cache *oic,
3638 struct lu_fid *fid, __u32 ino)
3640 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
3641 struct inode *inode;
3644 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3645 inode = osd_iget(oti, dev, &oic->oic_lid);
3646 if (IS_ERR(inode)) {
3647 fid_zero(&oic->oic_fid);
3648 return PTR_ERR(inode);
3651 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3654 fid_zero(&oic->oic_fid);
3656 *fid = oic->oic_fid = lma->lma_self_fid;
3661 * Calls ->lookup() to find dentry. From dentry get inode and
3662 * read inode's ea to get fid. This is required for interoperability
3665 * \retval 0, on success
3666 * \retval -ve, on error
3668 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3669 struct dt_rec *rec, const struct dt_key *key)
3671 struct inode *dir = obj->oo_inode;
3672 struct dentry *dentry;
3673 struct ldiskfs_dir_entry_2 *de;
3674 struct buffer_head *bh;
3675 struct lu_fid *fid = (struct lu_fid *) rec;
3676 struct htree_lock *hlock = NULL;
3681 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3683 dentry = osd_child_dentry_get(env, obj,
3684 (char *)key, strlen((char *)key));
3686 if (obj->oo_hl_head != NULL) {
3687 hlock = osd_oti_get(env)->oti_hlock;
3688 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3689 dir, LDISKFS_HLOCK_LOOKUP);
3691 down_read(&obj->oo_ext_idx_sem);
3694 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3696 struct osd_thread_info *oti = osd_oti_get(env);
3697 struct osd_inode_id *id = &oti->oti_id;
3698 struct osd_idmap_cache *oic = &oti->oti_cache;
3699 struct osd_device *dev = osd_obj2dev(obj);
3700 struct osd_scrub *scrub = &dev->od_scrub;