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) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ofd/ofd_objects.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mikhail Pershin <tappro@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_FILTER
44 #include <dt_object.h>
45 #include <lustre/lustre_idl.h>
46 #include <lustre_lfsck.h>
48 #include "ofd_internal.h"
50 int ofd_version_get_check(struct ofd_thread_info *info,
51 struct ofd_object *fo)
53 dt_obj_version_t curr_version;
55 LASSERT(ofd_object_exists(fo));
56 LASSERT(info->fti_exp);
58 curr_version = dt_version_get(info->fti_env, ofd_object_child(fo));
59 if ((__s64)curr_version == -EOPNOTSUPP)
61 /* VBR: version is checked always because costs nothing */
62 if (info->fti_pre_version != 0 &&
63 info->fti_pre_version != curr_version) {
64 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
65 info->fti_pre_version, curr_version);
66 spin_lock(&info->fti_exp->exp_lock);
67 info->fti_exp->exp_vbr_failed = 1;
68 spin_unlock(&info->fti_exp->exp_lock);
71 info->fti_pre_version = curr_version;
75 struct ofd_object *ofd_object_find(const struct lu_env *env,
76 struct ofd_device *ofd,
77 const struct lu_fid *fid)
79 struct ofd_object *fo;
84 o = lu_object_find(env, &ofd->ofd_dt_dev.dd_lu_dev, fid, NULL);
85 if (likely(!IS_ERR(o)))
88 fo = ERR_CAST(o); /* return error */
93 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
94 struct ofd_device *ofd,
95 const struct lu_fid *fid,
98 struct ofd_thread_info *info = ofd_info(env);
99 struct lu_object *fo_obj;
100 struct dt_object *dto;
104 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
106 dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr);
108 RETURN(ERR_CAST(dto));
110 fo_obj = lu_object_locate(dto->do_lu.lo_header,
111 ofd->ofd_dt_dev.dd_lu_dev.ld_type);
112 RETURN(ofd_obj(fo_obj));
115 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo)
121 if (!fo->ofo_ff_exists) {
123 * This actually means that we don't know whether the object
124 * has the "fid" EA or not.
126 rc = dt_xattr_get(env, ofd_object_child(fo), &LU_BUF_NULL,
127 XATTR_NAME_FID, BYPASS_CAPA);
128 if (rc >= 0 || rc == -ENODATA) {
130 * Here we assume that, if the object doesn't have the
131 * "fid" EA, the caller will add one, unless a fatal
132 * error (e.g., a memory or disk failure) prevents it
135 fo->ofo_ff_exists = 1;
143 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo)
145 lu_object_put(env, &fo->ofo_obj.do_lu);
148 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
149 obd_id id, struct ofd_seq *oseq, int nr, int sync)
151 struct ofd_thread_info *info = ofd_info(env);
152 struct ofd_object *fo = NULL;
153 struct dt_object *next;
155 struct ofd_object **batch;
156 struct lu_fid *fid = &info->fti_fid;
165 /* Don't create objects beyond the valid range for this SEQ */
166 if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
167 (id + nr) >= IDIF_MAX_OID)) {
168 CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
169 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
170 RETURN(rc = -ENOSPC);
171 } else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
172 (id + nr) >= OBIF_MAX_OID)) {
173 CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
174 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
175 RETURN(rc = -ENOSPC);
178 OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
182 info->fti_attr.la_valid = LA_TYPE | LA_MODE;
184 * We mark object SUID+SGID to flag it for accepting UID+GID from
185 * client on first write. Currently the permission bits on the OST are
186 * never used, so this is OK.
188 info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
189 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
191 /* Initialize a/c/m time so any client timestamp will always
192 * be newer and update the inode. ctime = 0 is also handled
193 * specially in osd_inode_setattr(). See LU-221, LU-1042 */
194 info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
195 info->fti_attr.la_atime = 0;
196 info->fti_attr.la_mtime = 0;
197 info->fti_attr.la_ctime = 0;
201 /* prepare objects */
202 *fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
203 for (i = 0; i < nr; i++) {
204 rc = fid_set_id(fid, id + i);
213 fo = ofd_object_find(env, ofd, fid);
216 GOTO(out, rc = PTR_ERR(fo));
222 ofd_write_lock(env, fo);
225 info->fti_buf.lb_buf = &tmp;
226 info->fti_buf.lb_len = sizeof(tmp);
229 th = ofd_trans_create(env, ofd);
231 GOTO(out, rc = PTR_ERR(th));
235 rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
238 GOTO(trans_stop, rc);
240 for (i = 0; i < nr; i++) {
244 if (unlikely(ofd_object_exists(fo))) {
245 /* object may exist being re-created by write replay */
246 CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
247 DFID"\n", ostid_seq(&oseq->os_oi), id,
248 PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
252 next = ofd_object_child(fo);
253 LASSERT(next != NULL);
255 rc = dt_declare_create(env, next, &info->fti_attr, NULL,
263 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
265 GOTO(trans_stop, rc);
267 CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
268 ofd_name(ofd), PFID(fid), nr);
272 /* When the LFSCK scanning the whole device to verify the LAST_ID file
273 * consistency, it will load the last_id into RAM firstly, and compare
274 * the last_id with each OST-object's ID. If the later one is larger,
275 * then it will regard the LAST_ID file crashed. But during the LFSCK
276 * scanning, the OFD may continue to create new OST-objects. Those new
277 * created OST-objects will have larger IDs than the LFSCK known ones.
278 * So from the LFSCK view, it needs to re-load the last_id from disk
279 * file, and if the latest last_id is still smaller than the object's
280 * ID, then the LAST_ID file is real crashed.
282 * To make above mechanism to work, before OFD pre-create OST-objects,
283 * it needs to update the LAST_ID file firstly, otherwise, the LFSCK
284 * may cannot get latest last_id although new OST-object created. */
285 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
286 tmp = cpu_to_le64(id + nr - 1);
287 dt_write_lock(env, oseq->os_lastid_obj, 0);
288 rc = dt_record_write(env, oseq->os_lastid_obj,
289 &info->fti_buf, &info->fti_off, th);
290 dt_write_unlock(env, oseq->os_lastid_obj);
292 GOTO(trans_stop, rc);
295 for (i = 0; i < nr; i++) {
299 /* Only the new created objects need to be recorded. */
300 if (ofd->ofd_osd->dd_record_fid_accessed) {
301 lfsck_pack_rfa(&ofd_info(env)->fti_lr,
302 lu_object_fid(&fo->ofo_obj.do_lu));
303 lfsck_in_notify(env, ofd->ofd_osd,
304 &ofd_info(env)->fti_lr);
307 if (likely(!ofd_object_exists(fo) &&
308 !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
309 next = ofd_object_child(fo);
310 LASSERT(next != NULL);
312 rc = dt_create(env, next, &info->fti_attr, NULL,
316 LASSERT(ofd_object_exists(fo));
318 ofd_seq_last_oid_set(oseq, id + i);
322 /* NOT all the wanted objects have been created,
323 * set the LAST_ID as the real created. */
324 if (unlikely(objects < nr)) {
328 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
329 dt_write_lock(env, oseq->os_lastid_obj, 0);
330 rc1 = dt_record_write(env, oseq->os_lastid_obj,
331 &info->fti_buf, &info->fti_off, th);
332 dt_write_unlock(env, oseq->os_lastid_obj);
334 CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
335 ") from "LPU64" to "LPU64"\n", ofd_name(ofd),
336 ostid_seq(&oseq->os_oi), id + nr - 1,
337 ofd_seq_last_oid(oseq));
341 ofd_trans_stop(env, ofd, th, rc);
343 for (i = 0; i < nr_saved; i++) {
346 ofd_write_unlock(env, fo);
347 ofd_object_put(env, fo);
350 OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
352 CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
353 "created %d/%d objects: %d\n", objects, nr_saved, rc);
355 LASSERT(ergo(objects == 0, rc < 0));
356 RETURN(objects > 0 ? objects : rc);
360 * If the object still has SUID+SGID bits set (see ofd_precreate_object()) then
361 * we will accept the UID+GID if sent by the client for initializing the
362 * ownership of this object. We only allow this to happen once (so clear these
363 * bits) and later only allow setattr.
365 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
366 struct lu_attr *la, int is_setattr)
368 struct ofd_thread_info *info = ofd_info(env);
369 struct lu_attr *ln = &info->fti_attr2;
375 if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID))
378 rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA);
382 LASSERT(ln->la_valid & LA_MODE);
385 if (!(ln->la_mode & S_ISUID))
386 la->la_valid &= ~LA_UID;
387 if (!(ln->la_mode & S_ISGID))
388 la->la_valid &= ~LA_GID;
391 if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID))
393 if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID))
396 if (!(la->la_valid & LA_MODE) || !is_setattr) {
397 la->la_mode = ln->la_mode;
398 la->la_valid |= LA_MODE;
400 la->la_mode &= ~mask;
406 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
407 struct lu_attr *la, struct filter_fid *ff)
409 struct ofd_thread_info *info = ofd_info(env);
410 struct ofd_device *ofd = ofd_obj2dev(fo);
412 struct ofd_mod_data *fmd;
417 ofd_write_lock(env, fo);
418 if (!ofd_object_exists(fo))
419 GOTO(unlock, rc = -ENOENT);
421 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
422 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
423 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
424 fmd->fmd_mactime_xid = info->fti_xid;
425 ofd_fmd_put(info->fti_exp, fmd);
428 /* VBR: version recovery check */
429 rc = ofd_version_get_check(info, fo);
433 rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
438 rc = ofd_object_ff_check(env, fo);
445 th = ofd_trans_create(env, ofd);
447 GOTO(unlock, rc = PTR_ERR(th));
449 rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
454 info->fti_buf.lb_buf = ff;
455 info->fti_buf.lb_len = sizeof(*ff);
456 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
457 &info->fti_buf, XATTR_NAME_FID, 0,
463 rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
467 rc = dt_attr_set(env, ofd_object_child(fo), la, th,
468 ofd_object_capa(env, fo));
473 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
474 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
477 ofd_trans_stop(env, ofd, th, rc);
479 ofd_write_unlock(env, fo);
483 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
484 __u64 start, __u64 end, struct lu_attr *la,
485 struct filter_fid *ff)
487 struct ofd_thread_info *info = ofd_info(env);
488 struct ofd_device *ofd = ofd_obj2dev(fo);
489 struct ofd_mod_data *fmd;
490 struct dt_object *dob = ofd_object_child(fo);
497 /* we support truncate, not punch yet */
498 LASSERT(end == OBD_OBJECT_EOF);
500 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
501 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
502 fmd->fmd_mactime_xid = info->fti_xid;
503 ofd_fmd_put(info->fti_exp, fmd);
505 ofd_write_lock(env, fo);
506 if (!ofd_object_exists(fo))
507 GOTO(unlock, rc = -ENOENT);
509 /* VBR: version recovery check */
510 rc = ofd_version_get_check(info, fo);
514 rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
519 rc = ofd_object_ff_check(env, fo);
526 th = ofd_trans_create(env, ofd);
528 GOTO(unlock, rc = PTR_ERR(th));
530 rc = dt_declare_attr_set(env, dob, la, th);
534 rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
539 info->fti_buf.lb_buf = ff;
540 info->fti_buf.lb_len = sizeof(*ff);
541 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
542 &info->fti_buf, XATTR_NAME_FID, 0,
548 rc = ofd_trans_start(env, ofd, fo, th);
552 rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
553 ofd_object_capa(env, fo));
557 rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
562 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
563 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
566 ofd_trans_stop(env, ofd, th, rc);
568 ofd_write_unlock(env, fo);
572 int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,
575 struct ofd_device *ofd = ofd_obj2dev(fo);
581 ofd_write_lock(env, fo);
582 if (!ofd_object_exists(fo))
583 GOTO(unlock, rc = -ENOENT);
585 th = ofd_trans_create(env, ofd);
587 GOTO(unlock, rc = PTR_ERR(th));
589 dt_declare_ref_del(env, ofd_object_child(fo), th);
590 dt_declare_destroy(env, ofd_object_child(fo), th);
592 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
594 rc = ofd_trans_start(env, ofd, NULL, th);
598 ofd_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
600 dt_ref_del(env, ofd_object_child(fo), th);
601 dt_destroy(env, ofd_object_child(fo), th);
603 ofd_trans_stop(env, ofd, th, rc);
605 ofd_write_unlock(env, fo);
609 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
616 if (ofd_object_exists(fo)) {
617 rc = dt_attr_get(env, ofd_object_child(fo), la,
618 ofd_object_capa(env, fo));
620 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
621 /* Try to correct for a bug in 2.1.0 (LU-221) that caused
622 * negative timestamps to appear to be in the far future,
623 * due old timestamp being stored on disk as an unsigned value.
624 * This fixes up any bad values stored on disk before
625 * returning them to the client, and ensures any timestamp
626 * updates are correct. LU-1042 */
627 if (unlikely(la->la_atime == LU221_BAD_TIME))
629 if (unlikely(la->la_mtime == LU221_BAD_TIME))
631 if (unlikely(la->la_ctime == LU221_BAD_TIME))
634 #warning "remove old LU-221/LU-1042 workaround code"