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>
47 #include "ofd_internal.h"
49 int ofd_version_get_check(struct ofd_thread_info *info,
50 struct ofd_object *fo)
52 dt_obj_version_t curr_version;
54 LASSERT(ofd_object_exists(fo));
55 LASSERT(info->fti_exp);
57 curr_version = dt_version_get(info->fti_env, ofd_object_child(fo));
58 if ((__s64)curr_version == -EOPNOTSUPP)
60 /* VBR: version is checked always because costs nothing */
61 if (info->fti_pre_version != 0 &&
62 info->fti_pre_version != curr_version) {
63 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
64 info->fti_pre_version, curr_version);
65 spin_lock(&info->fti_exp->exp_lock);
66 info->fti_exp->exp_vbr_failed = 1;
67 spin_unlock(&info->fti_exp->exp_lock);
70 info->fti_pre_version = curr_version;
74 struct ofd_object *ofd_object_find(const struct lu_env *env,
75 struct ofd_device *ofd,
76 const struct lu_fid *fid)
78 struct ofd_object *fo;
83 o = lu_object_find(env, &ofd->ofd_dt_dev.dd_lu_dev, fid, NULL);
84 if (likely(!IS_ERR(o)))
87 fo = ERR_CAST(o); /* return error */
92 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
93 struct ofd_device *ofd,
94 const struct lu_fid *fid,
97 struct ofd_thread_info *info = ofd_info(env);
98 struct lu_object *fo_obj;
99 struct dt_object *dto;
103 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
105 dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr);
107 RETURN(ERR_CAST(dto));
109 fo_obj = lu_object_locate(dto->do_lu.lo_header,
110 ofd->ofd_dt_dev.dd_lu_dev.ld_type);
111 RETURN(ofd_obj(fo_obj));
114 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo)
120 if (!fo->ofo_ff_exists) {
122 * This actually means that we don't know whether the object
123 * has the "fid" EA or not.
125 rc = dt_xattr_get(env, ofd_object_child(fo), &LU_BUF_NULL,
126 XATTR_NAME_FID, BYPASS_CAPA);
127 if (rc >= 0 || rc == -ENODATA) {
129 * Here we assume that, if the object doesn't have the
130 * "fid" EA, the caller will add one, unless a fatal
131 * error (e.g., a memory or disk failure) prevents it
134 fo->ofo_ff_exists = 1;
142 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo)
144 lu_object_put(env, &fo->ofo_obj.do_lu);
147 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
148 obd_id id, struct ofd_seq *oseq, int nr, int sync)
150 struct ofd_thread_info *info = ofd_info(env);
151 struct ofd_object *fo = NULL;
152 struct dt_object *next;
154 struct ofd_object **batch;
155 struct lu_fid *fid = &info->fti_fid;
164 /* Don't create objects beyond the valid range for this SEQ */
165 if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
166 (id + nr) >= IDIF_MAX_OID)) {
167 CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
168 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
169 RETURN(rc = -ENOSPC);
170 } else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
171 (id + nr) >= OBIF_MAX_OID)) {
172 CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
173 ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
174 RETURN(rc = -ENOSPC);
177 OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
181 info->fti_attr.la_valid = LA_TYPE | LA_MODE;
183 * We mark object SUID+SGID to flag it for accepting UID+GID from
184 * client on first write. Currently the permission bits on the OST are
185 * never used, so this is OK.
187 info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
188 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
190 /* Initialize a/c/m time so any client timestamp will always
191 * be newer and update the inode. ctime = 0 is also handled
192 * specially in osd_inode_setattr(). See LU-221, LU-1042 */
193 info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
194 info->fti_attr.la_atime = 0;
195 info->fti_attr.la_mtime = 0;
196 info->fti_attr.la_ctime = 0;
200 /* prepare objects */
201 *fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
202 for (i = 0; i < nr; i++) {
203 rc = fid_set_id(fid, id + i);
212 fo = ofd_object_find(env, ofd, fid);
215 GOTO(out, rc = PTR_ERR(fo));
221 ofd_write_lock(env, fo);
224 info->fti_buf.lb_buf = &tmp;
225 info->fti_buf.lb_len = sizeof(tmp);
228 th = ofd_trans_create(env, ofd);
230 GOTO(out, rc = PTR_ERR(th));
234 rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
237 GOTO(trans_stop, rc);
239 for (i = 0; i < nr; i++) {
243 if (unlikely(ofd_object_exists(fo))) {
244 /* object may exist being re-created by write replay */
245 CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
246 DFID"\n", ostid_seq(&oseq->os_oi), id,
247 PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
251 next = ofd_object_child(fo);
252 LASSERT(next != NULL);
254 rc = dt_declare_create(env, next, &info->fti_attr, NULL,
262 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
264 GOTO(trans_stop, rc);
266 CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
267 ofd_name(ofd), PFID(fid), nr);
271 /* When the LFSCK scanning the whole device to verify the LAST_ID file
272 * consistency, it will load the last_id into RAM firstly, and compare
273 * the last_id with echo OST-object's ID. If the later one is larger,
274 * then it will regard the LAST_ID file crashed. But during the LFSCK
275 * scanning, the OFD may continue to create new OST-objects. Those new
276 * created OST-objects will have larger IDs than the LFSCK known ones.
277 * So from the LFSCK view, it needs to re-load the last_id from disk
278 * file, and if the latest last_id is still smaller than the object's
279 * ID, then the LAST_ID file is real crashed.
281 * To make above mechanism to work, before OFD pre-create OST-objects,
282 * it needs to update the LAST_ID file firstly, otherwise, the LFSCK
283 * may cannot get latest last_id although new OST-object created. */
284 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
285 tmp = cpu_to_le64(id + nr - 1);
286 dt_write_lock(env, oseq->os_lastid_obj, 0);
287 rc = dt_record_write(env, oseq->os_lastid_obj,
288 &info->fti_buf, &info->fti_off, th);
289 dt_write_unlock(env, oseq->os_lastid_obj);
291 GOTO(trans_stop, rc);
294 for (i = 0; i < nr; i++) {
298 if (likely(!ofd_object_exists(fo))) {
299 next = ofd_object_child(fo);
300 LASSERT(next != NULL);
302 rc = dt_create(env, next, &info->fti_attr, NULL,
306 LASSERT(ofd_object_exists(fo));
308 ofd_seq_last_oid_set(oseq, id + i);
312 /* NOT all the wanted objects have been created,
313 * set the LAST_ID as the real created. */
314 if (unlikely(objects < nr)) {
318 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
319 dt_write_lock(env, oseq->os_lastid_obj, 0);
320 rc1 = dt_record_write(env, oseq->os_lastid_obj,
321 &info->fti_buf, &info->fti_off, th);
322 dt_write_unlock(env, oseq->os_lastid_obj);
324 CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
325 ") from "LPU64" to "LPU64"\n", ofd_name(ofd),
326 ostid_seq(&oseq->os_oi), id + nr - 1,
327 ofd_seq_last_oid(oseq));
331 ofd_trans_stop(env, ofd, th, rc);
333 for (i = 0; i < nr_saved; i++) {
336 ofd_write_unlock(env, fo);
337 ofd_object_put(env, fo);
340 OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
342 CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
343 "created %d/%d objects: %d\n", objects, nr_saved, rc);
345 LASSERT(ergo(objects == 0, rc < 0));
346 RETURN(objects > 0 ? objects : rc);
350 * If the object still has SUID+SGID bits set (see ofd_precreate_object()) then
351 * we will accept the UID+GID if sent by the client for initializing the
352 * ownership of this object. We only allow this to happen once (so clear these
353 * bits) and later only allow setattr.
355 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
356 struct lu_attr *la, int is_setattr)
358 struct ofd_thread_info *info = ofd_info(env);
359 struct lu_attr *ln = &info->fti_attr2;
365 if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID))
368 rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA);
372 LASSERT(ln->la_valid & LA_MODE);
375 if (!(ln->la_mode & S_ISUID))
376 la->la_valid &= ~LA_UID;
377 if (!(ln->la_mode & S_ISGID))
378 la->la_valid &= ~LA_GID;
381 if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID))
383 if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID))
386 if (!(la->la_valid & LA_MODE) || !is_setattr) {
387 la->la_mode = ln->la_mode;
388 la->la_valid |= LA_MODE;
390 la->la_mode &= ~mask;
396 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
397 struct lu_attr *la, struct filter_fid *ff)
399 struct ofd_thread_info *info = ofd_info(env);
400 struct ofd_device *ofd = ofd_obj2dev(fo);
402 struct ofd_mod_data *fmd;
407 ofd_write_lock(env, fo);
408 if (!ofd_object_exists(fo))
409 GOTO(unlock, rc = -ENOENT);
411 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
412 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
413 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
414 fmd->fmd_mactime_xid = info->fti_xid;
415 ofd_fmd_put(info->fti_exp, fmd);
418 /* VBR: version recovery check */
419 rc = ofd_version_get_check(info, fo);
423 rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
428 rc = ofd_object_ff_check(env, fo);
435 th = ofd_trans_create(env, ofd);
437 GOTO(unlock, rc = PTR_ERR(th));
439 rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
444 info->fti_buf.lb_buf = ff;
445 info->fti_buf.lb_len = sizeof(*ff);
446 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
447 &info->fti_buf, XATTR_NAME_FID, 0,
453 rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
457 rc = dt_attr_set(env, ofd_object_child(fo), la, th,
458 ofd_object_capa(env, fo));
463 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
464 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
467 ofd_trans_stop(env, ofd, th, rc);
469 ofd_write_unlock(env, fo);
473 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
474 __u64 start, __u64 end, struct lu_attr *la,
475 struct filter_fid *ff)
477 struct ofd_thread_info *info = ofd_info(env);
478 struct ofd_device *ofd = ofd_obj2dev(fo);
479 struct ofd_mod_data *fmd;
480 struct dt_object *dob = ofd_object_child(fo);
487 /* we support truncate, not punch yet */
488 LASSERT(end == OBD_OBJECT_EOF);
490 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
491 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
492 fmd->fmd_mactime_xid = info->fti_xid;
493 ofd_fmd_put(info->fti_exp, fmd);
495 ofd_write_lock(env, fo);
496 if (!ofd_object_exists(fo))
497 GOTO(unlock, rc = -ENOENT);
499 /* VBR: version recovery check */
500 rc = ofd_version_get_check(info, fo);
504 rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
509 rc = ofd_object_ff_check(env, fo);
516 th = ofd_trans_create(env, ofd);
518 GOTO(unlock, rc = PTR_ERR(th));
520 rc = dt_declare_attr_set(env, dob, la, th);
524 rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
529 info->fti_buf.lb_buf = ff;
530 info->fti_buf.lb_len = sizeof(*ff);
531 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
532 &info->fti_buf, XATTR_NAME_FID, 0,
538 rc = ofd_trans_start(env, ofd, fo, th);
542 rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
543 ofd_object_capa(env, fo));
547 rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
552 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
553 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
556 ofd_trans_stop(env, ofd, th, rc);
558 ofd_write_unlock(env, fo);
562 int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,
565 struct ofd_device *ofd = ofd_obj2dev(fo);
571 ofd_write_lock(env, fo);
572 if (!ofd_object_exists(fo))
573 GOTO(unlock, rc = -ENOENT);
575 th = ofd_trans_create(env, ofd);
577 GOTO(unlock, rc = PTR_ERR(th));
579 dt_declare_ref_del(env, ofd_object_child(fo), th);
580 dt_declare_destroy(env, ofd_object_child(fo), th);
582 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
584 rc = ofd_trans_start(env, ofd, NULL, th);
588 ofd_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
590 dt_ref_del(env, ofd_object_child(fo), th);
591 dt_destroy(env, ofd_object_child(fo), th);
593 ofd_trans_stop(env, ofd, th, rc);
595 ofd_write_unlock(env, fo);
599 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
606 if (ofd_object_exists(fo)) {
607 rc = dt_attr_get(env, ofd_object_child(fo), la,
608 ofd_object_capa(env, fo));
610 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
611 /* Try to correct for a bug in 2.1.0 (LU-221) that caused
612 * negative timestamps to appear to be in the far future,
613 * due old timestamp being stored on disk as an unsigned value.
614 * This fixes up any bad values stored on disk before
615 * returning them to the client, and ensures any timestamp
616 * updates are correct. LU-1042 */
617 if (unlikely(la->la_atime == LU221_BAD_TIME))
619 if (unlikely(la->la_mtime == LU221_BAD_TIME))
621 if (unlikely(la->la_ctime == LU221_BAD_TIME))
624 #warning "remove old LU-221/LU-1042 workaround code"