4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/ofd/ofd_objects.c
33 * This file contains OSD API methods related to OBD Filter Device (OFD)
36 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
37 * Author: Mikhail Pershin <mike.pershin@intel.com>
40 #define DEBUG_SUBSYSTEM S_FILTER
42 #include <dt_object.h>
43 #include <lustre_lfsck.h>
45 #include "ofd_internal.h"
48 * Get object version from disk and check it.
50 * This function checks object version from disk with
51 * ofd_thread_info::fti_pre_version filled from incoming RPC. This is part of
52 * VBR (Version-Based Recovery) and ensures that object has the same version
53 * upon replay as it has during original modification.
55 * \param[in] info execution thread OFD private data
56 * \param[in] fo OFD object
58 * \retval 0 if version matches
59 * \retval -EOVERFLOW on version mismatch
61 static int ofd_version_get_check(struct ofd_thread_info *info,
62 struct ofd_object *fo)
64 dt_obj_version_t curr_version;
66 if (info->fti_exp == NULL)
69 curr_version = dt_version_get(info->fti_env, ofd_object_child(fo));
70 if ((__s64)curr_version == -EOPNOTSUPP)
72 /* VBR: version is checked always because costs nothing */
73 if (info->fti_pre_version != 0 &&
74 info->fti_pre_version != curr_version) {
75 CDEBUG(D_INODE, "Version mismatch %#llx != %#llx\n",
76 info->fti_pre_version, curr_version);
77 spin_lock(&info->fti_exp->exp_lock);
78 info->fti_exp->exp_vbr_failed = 1;
79 spin_unlock(&info->fti_exp->exp_lock);
82 info->fti_pre_version = curr_version;
87 * Get OFD object by FID.
89 * This function finds OFD slice of compound object with the given FID.
91 * \param[in] env execution environment
92 * \param[in] ofd OFD device
93 * \param[in] fid FID of the object
95 * \retval pointer to the found ofd_object
96 * \retval ERR_PTR(errno) in case of error
98 struct ofd_object *ofd_object_find(const struct lu_env *env,
99 struct ofd_device *ofd,
100 const struct lu_fid *fid)
102 struct ofd_object *fo;
107 o = lu_object_find(env, &ofd->ofd_dt_dev.dd_lu_dev, fid, NULL);
108 if (likely(!IS_ERR(o)))
111 fo = ERR_CAST(o); /* return error */
117 * Get FID of parent MDT object.
119 * This function reads extended attribute XATTR_NAME_FID of OFD object which
120 * contains the MDT parent object FID and saves it in ofd_object::ofo_ff.
122 * The filter_fid::ff_parent::f_ver field currently holds
123 * the OST-object index in the parent MDT-object's layout EA,
124 * not the actual FID::f_ver of the parent. We therefore access
125 * it via the macro f_stripe_idx.
127 * \param[in] env execution environment
128 * \param[in] fo OFD object
130 * \retval 0 if successful
131 * \retval -ENODATA if there is no such xattr
132 * \retval negative value on error
134 int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo)
136 struct ofd_thread_info *info = ofd_info(env);
137 struct filter_fid *ff = &fo->ofo_ff;
138 struct lu_buf *buf = &info->fti_buf;
141 if (fid_is_sane(&ff->ff_parent))
145 buf->lb_len = sizeof(*ff);
146 rc = dt_xattr_get(env, ofd_object_child(fo), buf, XATTR_NAME_FID);
150 if (unlikely(rc < sizeof(struct lu_fid))) {
151 fid_zero(&ff->ff_parent);
155 filter_fid_le_to_cpu(ff, ff, rc);
160 struct ofd_precreate_cb {
161 struct dt_txn_commit_cb opc_cb;
162 struct ofd_seq *opc_oseq;
166 static void ofd_cb_precreate(struct lu_env *env, struct thandle *th,
167 struct dt_txn_commit_cb *cb, int err)
169 struct ofd_precreate_cb *opc;
170 struct ofd_seq *oseq;
172 opc = container_of(cb, struct ofd_precreate_cb, opc_cb);
173 oseq = opc->opc_oseq;
175 CDEBUG(D_OTHER, "Sub %d from %d for "DFID", th_sync %d\n",
176 opc->opc_objects, atomic_read(&oseq->os_precreate_in_progress),
177 PFID(&oseq->os_oi.oi_fid), th->th_sync);
178 atomic_sub(opc->opc_objects, &oseq->os_precreate_in_progress);
179 ofd_seq_put(env, opc->opc_oseq);
183 static int ofd_precreate_cb_add(const struct lu_env *env, struct thandle *th,
184 struct ofd_seq *oseq, int objects)
186 struct ofd_precreate_cb *opc;
187 struct dt_txn_commit_cb *dcb;
194 precreate = atomic_read(&oseq->os_precreate_in_progress);
195 atomic_inc(&oseq->os_refc);
196 opc->opc_oseq = oseq;
197 opc->opc_objects = objects;
198 CDEBUG(D_OTHER, "Add %d to %d for "DFID", th_sync %d\n",
199 opc->opc_objects, precreate,
200 PFID(&oseq->os_oi.oi_fid), th->th_sync);
202 if ((precreate + objects) >= (5 * OST_MAX_PRECREATE))
206 dcb->dcb_func = ofd_cb_precreate;
207 INIT_LIST_HEAD(&dcb->dcb_linkage);
208 strlcpy(dcb->dcb_name, "ofd_cb_precreate", sizeof(dcb->dcb_name));
210 rc = dt_trans_cb_add(th, dcb);
212 ofd_seq_put(env, oseq);
217 atomic_add(objects, &oseq->os_precreate_in_progress);
223 * Precreate the given number \a nr of objects in the given sequence \a oseq.
225 * This function precreates new OST objects in the given sequence.
226 * The precreation starts from \a id and creates \a nr objects sequentially.
229 * This function may create fewer objects than requested.
231 * We mark object SUID+SGID to flag it for accepting UID+GID from client on
232 * first write. Currently the permission bits on the OST are never used,
235 * Initialize a/c/m time so any client timestamp will always be newer and
236 * update the inode. The ctime = 0 case is also handled specially in
237 * osd_inode_setattr(). See LU-221, LU-1042 for details.
239 * \param[in] env execution environment
240 * \param[in] ofd OFD device
241 * \param[in] id object ID to start precreation from
242 * \param[in] oseq object sequence
243 * \param[in] nr number of objects to precreate
244 * \param[in] sync synchronous precreation flag
246 * \retval 0 if successful
247 * \retval negative value on error
249 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
250 u64 id, struct ofd_seq *oseq, int nr, int sync)
252 struct ofd_thread_info *info = ofd_info(env);
253 struct ofd_object *fo = NULL;
254 struct dt_object *next;
256 struct ofd_object **batch;
257 struct lu_fid *fid = &info->fti_fid;
267 /* Don't create objects beyond the valid range for this SEQ
268 * Last object to create is (id + nr - 1), but we move -1 on LHS
269 * to +1 on RHS to evaluate constant at compile time. LU-11186
271 if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
272 id + nr > IDIF_MAX_OID + 1)) {
273 CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
274 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
275 RETURN(rc = -ENOSPC);
276 } else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
277 id + nr > OBIF_MAX_OID + 1)) {
278 CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
279 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
280 RETURN(rc = -ENOSPC);
283 OBD_ALLOC_PTR_ARRAY(batch, nr_saved);
287 info->fti_attr.la_valid = LA_TYPE | LA_MODE;
288 info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666;
289 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
291 info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
292 info->fti_attr.la_atime = 0;
293 info->fti_attr.la_mtime = 0;
294 info->fti_attr.la_ctime = 0;
298 /* prepare objects */
299 *fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
300 for (i = 0; i < nr; i++) {
301 rc = fid_set_id(fid, id + i);
310 fo = ofd_object_find(env, ofd, fid);
313 GOTO(out, rc = PTR_ERR(fo));
321 info->fti_buf.lb_buf = &tmp;
322 info->fti_buf.lb_len = sizeof(tmp);
325 th = ofd_trans_create(env, ofd);
327 GOTO(out, rc = PTR_ERR(th));
331 rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
334 GOTO(trans_stop, rc);
336 for (i = 0; i < nr; i++) {
340 if (unlikely(ofd_object_exists(fo))) {
341 /* object may exist being re-created by write replay */
342 CDEBUG(D_INODE, "object %#llx/%#llx exists: "
343 DFID"\n", ostid_seq(&oseq->os_oi), id,
344 PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
348 next = ofd_object_child(fo);
349 LASSERT(next != NULL);
351 rc = dt_declare_create(env, next, &info->fti_attr, NULL,
355 GOTO(trans_stop, rc);
362 rc = dt_trans_start(env, ofd->ofd_osd, th);
364 GOTO(trans_stop, rc);
366 CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
367 ofd_name(ofd), PFID(fid), nr);
369 /* When the LFSCK scanning the whole device to verify the LAST_ID file
370 * consistency, it will load the last_id into RAM firstly, and compare
371 * the last_id with each OST-object's ID. If the later one is larger,
372 * then it will regard the LAST_ID file crashed. But during the LFSCK
373 * scanning, the OFD may continue to create new OST-objects. Those new
374 * created OST-objects will have larger IDs than the LFSCK known ones.
375 * So from the LFSCK view, it needs to re-load the last_id from disk
376 * file, and if the latest last_id is still smaller than the object's
377 * ID, then the LAST_ID file is real crashed.
379 * To make above mechanism to work, before OFD pre-create OST-objects,
380 * it needs to update the LAST_ID file firstly, otherwise, the LFSCK
381 * may cannot get latest last_id although new OST-object created. */
382 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
383 tmp = cpu_to_le64(id + nr - 1);
384 dt_write_lock(env, oseq->os_lastid_obj, DT_LASTID);
385 rc = dt_record_write(env, oseq->os_lastid_obj,
386 &info->fti_buf, &info->fti_off, th);
387 dt_write_unlock(env, oseq->os_lastid_obj);
389 GOTO(trans_stop, rc);
392 for (i = 0; i < nr; i++) {
396 ofd_write_lock(env, fo);
398 /* Only the new created objects need to be recorded. */
399 if (ofd->ofd_osd->dd_record_fid_accessed) {
400 struct lfsck_req_local *lrl = &ofd_info(env)->fti_lrl;
402 lfsck_pack_rfa(lrl, lu_object_fid(&fo->ofo_obj.do_lu),
403 LEL_FID_ACCESSED, LFSCK_TYPE_LAYOUT);
404 lfsck_in_notify_local(env, ofd->ofd_osd, lrl, NULL);
407 if (likely(!ofd_object_exists(fo) &&
408 !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
409 next = ofd_object_child(fo);
410 LASSERT(next != NULL);
412 rc = dt_create(env, next, &info->fti_attr, NULL,
414 ofd_write_unlock(env, fo);
417 GOTO(trans_stop, rc);
422 LASSERT(ofd_object_exists(fo));
424 ofd_write_unlock(env, fo);
427 ofd_seq_last_oid_set(oseq, id + i);
431 /* NOT all the wanted objects have been created,
432 * set the LAST_ID as the real created. */
433 if (unlikely(objects < nr)) {
437 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
438 dt_write_lock(env, oseq->os_lastid_obj, DT_LASTID);
439 rc1 = dt_record_write(env, oseq->os_lastid_obj,
440 &info->fti_buf, &info->fti_off, th);
441 dt_write_unlock(env, oseq->os_lastid_obj);
443 CERROR("%s: fail to reset the LAST_ID for seq (%#llx"
444 ") from %llu to %llu\n", ofd_name(ofd),
445 ostid_seq(&oseq->os_oi), id + nr - 1,
446 ofd_seq_last_oid(oseq));
450 ofd_precreate_cb_add(env, th, oseq, objects);
452 rc2 = ofd_trans_stop(env, ofd, th, rc);
454 CERROR("%s: failed to stop transaction: rc = %d\n",
459 for (i = 0; i < nr_saved; i++) {
463 ofd_object_put(env, fo);
465 OBD_FREE_PTR_ARRAY(batch, nr_saved);
467 CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
468 "created %d/%d objects: %d\n", objects, nr_saved, rc);
470 LASSERT(ergo(objects == 0, rc < 0));
471 RETURN(objects > 0 ? objects : rc);
475 * Fix the OFD object ownership.
477 * If the object still has SUID+SGID bits set, meaning that it was precreated
478 * by the MDT before it was assigned to any file, (see ofd_precreate_objects())
479 * then we will accept the UID/GID/PROJID if sent by the client for initializing
480 * the ownership of this object. We only allow this to happen once (so clear
481 * these bits) and later only allow setattr.
483 * \param[in] env execution environment
484 * \param[in] fo OFD object
485 * \param[in] la object attributes
486 * \param[in] is_setattr was this function called from setattr or not
488 * \retval 0 if successful
489 * \retval negative value on error
491 int ofd_attr_handle_id(const struct lu_env *env, struct ofd_object *fo,
492 struct lu_attr *la, int is_setattr)
494 struct ofd_thread_info *info = ofd_info(env);
495 struct lu_attr *ln = &info->fti_attr2;
501 if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID) &&
502 !(la->la_valid & LA_PROJID))
505 rc = dt_attr_get(env, ofd_object_child(fo), ln);
509 LASSERT(ln->la_valid & LA_MODE);
512 * Only allow setattr to change UID/GID/PROJID, if
513 * SUID+SGID is not set which means this is not
514 * initialization of this objects.
517 if (!(ln->la_mode & S_ISUID))
518 la->la_valid &= ~LA_UID;
519 if (!(ln->la_mode & S_ISGID))
520 la->la_valid &= ~LA_GID;
521 if (!(ln->la_mode & S_ISVTX))
522 la->la_valid &= ~LA_PROJID;
525 /* Initialize ownership of this object, clear SUID+SGID bits*/
526 if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID))
528 if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID))
530 if ((la->la_valid & LA_PROJID) && (ln->la_mode & S_ISVTX))
533 if (!(la->la_valid & LA_MODE) || !is_setattr) {
534 la->la_mode = ln->la_mode;
535 la->la_valid |= LA_MODE;
537 la->la_mode &= ~mask;
544 * Check if it needs to update filter_fid by the value of @oa.
547 * \param[in] fo ofd object
548 * \param[in] oa obdo from client or MDT
549 * \param[out] ff if filter_fid needs updating, this field is used to
550 * return the new buffer
552 * \retval < 0 error occurred
553 * \retval 0 doesn't need to update filter_fid
554 * \retval FL_XATTR_{CREATE,REPLACE} flag for xattr update
556 int ofd_object_ff_update(const struct lu_env *env, struct ofd_object *fo,
557 const struct obdo *oa, struct filter_fid *ff)
563 (OBD_MD_FLFID | OBD_MD_FLOSTLAYOUT | OBD_MD_LAYOUT_VERSION)))
566 rc = ofd_object_ff_load(env, fo);
567 if (rc < 0 && rc != -ENODATA)
570 LASSERT(ff != &fo->ofo_ff);
571 if (rc == -ENODATA) {
572 rc = LU_XATTR_CREATE;
573 memset(ff, 0, sizeof(*ff));
575 rc = LU_XATTR_REPLACE;
576 memcpy(ff, &fo->ofo_ff, sizeof(*ff));
579 if (oa->o_valid & OBD_MD_FLFID) {
580 /* packing fid and converting it to LE for storing into EA.
581 * Here ->o_stripe_idx should be filled by LOV and rest of
582 * fields - by client. */
583 ff->ff_parent.f_seq = oa->o_parent_seq;
584 ff->ff_parent.f_oid = oa->o_parent_oid;
585 /* XXX: we are ignoring o_parent_ver here, since this should
586 * be the same for all objects in this fileset. */
587 ff->ff_parent.f_ver = oa->o_stripe_idx;
589 if (oa->o_valid & OBD_MD_FLOSTLAYOUT)
590 ff->ff_layout = oa->o_layout;
592 if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
593 CDEBUG(D_INODE, DFID": OST("DFID") layout version %u -> %u\n",
594 PFID(&fo->ofo_ff.ff_parent),
595 PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
596 ff->ff_layout_version, oa->o_layout_version);
599 * resync write from client on non-primary objects and
600 * resync start from MDS on primary objects will contain
601 * LU_LAYOUT_RESYNC flag in the @oa.
603 * The layout version checking for write/punch from client
604 * happens in ofd_verify_layout_version() before coming to
605 * here, so that resync with smaller layout version client
606 * will be rejected there, the biggest resync version will
607 * be recorded in the OFD objects.
609 if (ff->ff_layout_version & LU_LAYOUT_RESYNC) {
610 /* this opens a new era of writing */
611 ff->ff_layout_version = 0;
615 /* it's not allowed to change it to a smaller value */
616 if (ofd_layout_version_less(oa->o_layout_version,
617 ff->ff_layout_version))
620 if (ff->ff_layout_version == 0 ||
621 oa->o_layout_version & LU_LAYOUT_RESYNC) {
622 /* if LU_LAYOUT_RESYNC is set, it closes the era of
623 * writing. Only mirror I/O can write this object. */
624 ff->ff_layout_version = oa->o_layout_version;
626 } else if (oa->o_layout_version > ff->ff_layout_version) {
627 ff->ff_range = max_t(__u32, ff->ff_range,
628 oa->o_layout_version -
629 ff->ff_layout_version);
633 if (memcmp(ff, &fo->ofo_ff, sizeof(*ff)))
634 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
642 * Set OFD object attributes.
644 * This function sets OFD object attributes taken from incoming request.
645 * It sets not only regular attributes but also XATTR_NAME_FID extended
646 * attribute if needed. The "fid" xattr allows the object's MDT parent inode
647 * to be found and verified by LFSCK and other tools in case of inconsistency.
649 * \param[in] env execution environment
650 * \param[in] fo OFD object
651 * \param[in] la object attributes
652 * \param[in] oa obdo carries fid, ost_layout, layout version
654 * \retval 0 if successful
655 * \retval negative value on error
657 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
658 struct lu_attr *la, struct obdo *oa)
660 struct ofd_thread_info *info = ofd_info(env);
661 struct ofd_device *ofd = ofd_obj2dev(fo);
662 struct filter_fid *ff = &info->fti_mds_fid;
668 if (!ofd_object_exists(fo))
669 GOTO(out, rc = -ENOENT);
671 /* VBR: version recovery check */
672 rc = ofd_version_get_check(info, fo);
676 rc = ofd_attr_handle_id(env, fo, la, 1 /* is_setattr */);
680 th = ofd_trans_create(env, ofd);
682 GOTO(out, rc = PTR_ERR(th));
684 rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
688 info->fti_buf.lb_buf = ff;
689 info->fti_buf.lb_len = sizeof(*ff);
690 rc = dt_declare_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
691 XATTR_NAME_FID, 0, th);
695 rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
699 ofd_write_lock(env, fo);
700 if (!ofd_object_exists(fo))
701 GOTO(unlock, rc = -ENOENT);
703 /* serialize vs ofd_commitrw_write() */
704 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
705 tgt_fmd_update(info->fti_exp, &fo->ofo_header.loh_fid,
708 rc = dt_attr_set(env, ofd_object_child(fo), la, th);
712 fl = ofd_object_ff_update(env, fo, oa, ff);
714 GOTO(unlock, rc = fl);
717 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
718 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
719 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
720 le32_add_cpu(&ff->ff_parent.f_oid, -1);
721 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
724 info->fti_buf.lb_buf = ff;
725 info->fti_buf.lb_len = sizeof(*ff);
726 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
727 XATTR_NAME_FID, fl, th);
729 filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff));
735 ofd_write_unlock(env, fo);
737 rc2 = ofd_trans_stop(env, ofd, th, rc);
739 CERROR("%s: failed to stop transaction: rc = %d\n",
748 * Fallocate(Preallocate) space for OFD object.
750 * This function allocates space for the object from the \a start
751 * offset to the \a end offset.
753 * \param[in] env execution environment
754 * \param[in] fo OFD object
755 * \param[in] start start offset to allocate from
756 * \param[in] end end of allocate
757 * \param[in] mode fallocate mode
758 * \param[in] la object attributes
759 * \param[in] ff filter_fid structure
761 * \retval 0 if successful
762 * \retval negative value on error
764 int ofd_object_fallocate(const struct lu_env *env, struct ofd_object *fo,
765 __u64 start, __u64 end, int mode, struct lu_attr *la,
768 struct ofd_thread_info *info = ofd_info(env);
769 struct ofd_device *ofd = ofd_obj2dev(fo);
770 struct dt_object *dob = ofd_object_child(fo);
772 struct filter_fid *ff = &info->fti_mds_fid;
773 bool ff_needed = false;
778 if (!ofd_object_exists(fo))
781 /* VBR: version recovery check */
782 rc = ofd_version_get_check(info, fo);
787 rc = ofd_object_ff_load(env, fo);
794 if (oa->o_valid & OBD_MD_FLFID) {
795 ff->ff_parent.f_seq = oa->o_parent_seq;
796 ff->ff_parent.f_oid = oa->o_parent_oid;
797 ff->ff_parent.f_ver = oa->o_stripe_idx;
799 if (oa->o_valid & OBD_MD_FLOSTLAYOUT)
800 ff->ff_layout = oa->o_layout;
801 if (oa->o_valid & OBD_MD_LAYOUT_VERSION)
802 ff->ff_layout_version = oa->o_layout_version;
803 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
807 th = ofd_trans_create(env, ofd);
811 rc = dt_declare_attr_set(env, dob, la, th);
815 rc = dt_declare_fallocate(env, dob, start, end, mode, th);
820 info->fti_buf.lb_buf = ff;
821 info->fti_buf.lb_len = sizeof(*ff);
822 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
823 &info->fti_buf, XATTR_NAME_FID, 0,
829 rc = ofd_trans_start(env, ofd, fo, th);
833 ofd_read_lock(env, fo);
834 if (!ofd_object_exists(fo))
835 GOTO(unlock, rc = -ENOENT);
837 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
838 tgt_fmd_update(info->fti_exp, &fo->ofo_header.loh_fid,
841 rc = dt_falloc(env, dob, start, end, mode, th);
845 rc = dt_attr_set(env, dob, la, th);
850 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
851 XATTR_NAME_FID, 0, th);
853 filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff));
856 ofd_read_unlock(env, fo);
858 ofd_trans_stop(env, ofd, th, rc);
863 * Truncate/punch OFD object.
865 * This function frees all of the allocated object's space from the \a start
866 * offset to the \a end offset. For truncate() operations the \a end offset
867 * is OBD_OBJECT_EOF. The functionality to punch holes in an object via
868 * fallocate(FALLOC_FL_PUNCH_HOLE) is not yet implemented (see LU-3606).
870 * \param[in] env execution environment
871 * \param[in] fo OFD object
872 * \param[in] start start offset to punch from
873 * \param[in] end end of punch
874 * \param[in] la object attributes
875 * \param[in] oa obdo struct from incoming request
877 * \retval 0 if successful
878 * \retval negative value on error
880 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
881 __u64 start, __u64 end, struct lu_attr *la,
884 struct ofd_thread_info *info = ofd_info(env);
885 struct ofd_device *ofd = ofd_obj2dev(fo);
886 struct dt_object *dob = ofd_object_child(fo);
887 struct filter_fid *ff = &info->fti_mds_fid;
893 /* we support truncate, not punch yet */
894 LASSERT(end == OBD_OBJECT_EOF);
896 if (!ofd_object_exists(fo))
897 GOTO(out, rc = -ENOENT);
899 if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
900 rc = ofd_verify_ff(env, fo, oa);
905 /* VBR: version recovery check */
906 rc = ofd_version_get_check(info, fo);
910 rc = ofd_attr_handle_id(env, fo, la, 0 /* !is_setattr */);
914 th = ofd_trans_create(env, ofd);
916 GOTO(out, rc = PTR_ERR(th));
918 if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & LUSTRE_ENCRYPT_FL) {
919 /* punch must be aware we are dealing with an encrypted file */
920 la->la_valid |= LA_FLAGS;
921 la->la_flags |= LUSTRE_ENCRYPT_FL;
923 rc = dt_declare_attr_set(env, dob, la, th);
927 rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
931 info->fti_buf.lb_buf = ff;
932 info->fti_buf.lb_len = sizeof(*ff);
933 rc = dt_declare_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
934 XATTR_NAME_FID, 0, th);
938 rc = ofd_trans_start(env, ofd, fo, th);
942 ofd_write_lock(env, fo);
944 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
945 tgt_fmd_update(info->fti_exp, &fo->ofo_header.loh_fid,
948 if (!ofd_object_exists(fo))
949 GOTO(unlock, rc = -ENOENT);
951 /* need to verify layout version */
952 if (oa->o_valid & OBD_MD_LAYOUT_VERSION) {
953 rc = ofd_verify_layout_version(env, fo, oa);
958 rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th);
962 fl = ofd_object_ff_update(env, fo, oa, ff);
964 GOTO(unlock, rc = fl);
966 rc = dt_attr_set(env, dob, la, th);
971 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
972 ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
973 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
974 le32_add_cpu(&ff->ff_parent.f_oid, -1);
975 else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
978 info->fti_buf.lb_buf = ff;
979 info->fti_buf.lb_len = sizeof(*ff);
980 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
981 XATTR_NAME_FID, fl, th);
983 filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff));
989 ofd_write_unlock(env, fo);
991 rc2 = ofd_trans_stop(env, ofd, th, rc);
993 CERROR("%s: failed to stop transaction: rc = %d\n",
1002 * Destroy OFD object.
1004 * This function destroys OFD object. If object wasn't used at all (orphan)
1005 * then local transaction is used, which means the transaction data is not
1006 * returned back in reply.
1008 * \param[in] env execution environment
1009 * \param[in] fo OFD object
1010 * \param[in] orphan flag to indicate that object is orphaned
1012 * \retval 0 if successful
1013 * \retval negative value on error
1015 int ofd_destroy(const struct lu_env *env, struct ofd_object *fo,
1018 struct ofd_device *ofd = ofd_obj2dev(fo);
1025 if (!ofd_object_exists(fo))
1026 GOTO(out, rc = -ENOENT);
1028 th = ofd_trans_create(env, ofd);
1030 GOTO(out, rc = PTR_ERR(th));
1032 rc = dt_declare_ref_del(env, ofd_object_child(fo), th);
1036 rc = dt_declare_destroy(env, ofd_object_child(fo), th);
1041 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
1043 rc = ofd_trans_start(env, ofd, NULL, th);
1047 ofd_write_lock(env, fo);
1048 if (!ofd_object_exists(fo))
1049 GOTO(unlock, rc = -ENOENT);
1051 tgt_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
1053 dt_ref_del(env, ofd_object_child(fo), th);
1054 dt_destroy(env, ofd_object_child(fo), th);
1056 ofd_write_unlock(env, fo);
1058 rc2 = ofd_trans_stop(env, ofd, th, rc);
1060 CERROR("%s failed to stop transaction: %d\n",
1061 ofd_name(ofd), rc2);
1069 * Get OFD object attributes.
1071 * This function gets OFD object regular attributes. It is used to serve
1072 * incoming request as well as for local OFD purposes.
1074 * \param[in] env execution environment
1075 * \param[in] fo OFD object
1076 * \param[in] la object attributes
1078 * \retval 0 if successful
1079 * \retval negative value on error
1081 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
1088 if (ofd_object_exists(fo)) {
1089 rc = dt_attr_get(env, ofd_object_child(fo), la);