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, 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 = (struct ofd_object *)o; /* return error */
91 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
92 struct ofd_device *ofd,
93 const struct lu_fid *fid,
96 struct ofd_thread_info *info = ofd_info(env);
97 struct lu_object *fo_obj;
98 struct dt_object *dto;
102 info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
104 dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr);
106 RETURN((struct ofd_object *)dto);
108 fo_obj = lu_object_locate(dto->do_lu.lo_header,
109 ofd->ofd_dt_dev.dd_lu_dev.ld_type);
110 RETURN(ofd_obj(fo_obj));
113 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo)
115 struct ofd_thread_info *info = ofd_info(env);
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 info->fti_buf.lb_buf = &info->fti_mds_fid2;
126 info->fti_buf.lb_len = sizeof(info->fti_mds_fid2);
127 rc = dt_xattr_get(env, ofd_object_child(fo), &info->fti_buf,
128 XATTR_NAME_FID, BYPASS_CAPA);
129 if (rc >= 0 || rc == -ENODATA) {
131 * Here we assume that, if the object doesn't have the
132 * "fid" EA, the caller will add one, unless a fatal
133 * error (e.g., a memory or disk failure) prevents it
136 fo->ofo_ff_exists = 1;
144 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo)
146 lu_object_put(env, &fo->ofo_obj.do_lu);
149 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
150 obd_id id, struct ofd_seq *oseq, int nr, int sync)
152 struct ofd_thread_info *info = ofd_info(env);
153 struct ofd_object *fo = NULL;
154 struct dt_object *next;
156 struct ofd_object **batch;
165 /* Don't create objects beyond the valid range for this SEQ */
166 if (unlikely(fid_seq_is_mdt0(oseq->os_seq) && (id + nr) >= IDIF_MAX_OID)) {
167 CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
168 ofd_name(ofd), id, oseq->os_seq);
169 RETURN(rc = -ENOSPC);
170 } else if (unlikely(!fid_seq_is_mdt0(oseq->os_seq) &&
171 (id + nr) >= OBIF_MAX_OID)) {
172 CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
173 ofd_name(ofd), id, oseq->os_seq);
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;
198 /* prepare objects */
199 for (i = 0; i < nr; i++) {
200 info->fti_ostid.oi_id = id + i;
201 info->fti_ostid.oi_seq = oseq->os_seq;
203 rc = fid_ostid_unpack(&info->fti_fid, &info->fti_ostid, 0);
206 GOTO(out, rc = PTR_ERR(fo));
212 fo = ofd_object_find(env, ofd, &info->fti_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", oseq->os_seq, id,
247 PFID(&info->fti_fid));
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, "create new object "DFID"\n", PFID(&info->fti_fid));
268 for (i = 0; i < nr; i++) {
272 if (likely(!ofd_object_exists(fo))) {
273 next = ofd_object_child(fo);
274 LASSERT(next != NULL);
276 rc = dt_create(env, next, &info->fti_attr, NULL,
280 LASSERT(ofd_object_exists(fo));
282 ofd_seq_last_oid_set(oseq, id + i);
287 tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
288 rc = dt_record_write(env, oseq->os_lastid_obj,
289 &info->fti_buf, &info->fti_off, th);
292 ofd_trans_stop(env, ofd, th, rc);
294 for (i = 0; i < nr_saved; i++) {
297 ofd_write_unlock(env, fo);
298 ofd_object_put(env, fo);
301 OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
303 CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
304 "created %d/%d objects: %d\n", objects, nr_saved, rc);
306 LASSERT(ergo(objects == 0, rc < 0));
307 RETURN(objects > 0 ? objects : rc);
311 * If the object still has SUID+SGID bits set (see ofd_precreate_object()) then
312 * we will accept the UID+GID if sent by the client for initializing the
313 * ownership of this object. We only allow this to happen once (so clear these
314 * bits) and later only allow setattr.
316 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
317 struct lu_attr *la, int is_setattr)
319 struct ofd_thread_info *info = ofd_info(env);
320 struct lu_attr *ln = &info->fti_attr2;
326 if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID))
329 rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA);
333 LASSERT(ln->la_valid & LA_MODE);
336 if (!(ln->la_mode & S_ISUID))
337 la->la_valid &= ~LA_UID;
338 if (!(ln->la_mode & S_ISGID))
339 la->la_valid &= ~LA_GID;
342 if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID))
344 if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID))
347 if (!(la->la_valid & LA_MODE) || !is_setattr) {
348 la->la_mode = ln->la_mode;
349 la->la_valid |= LA_MODE;
351 la->la_mode &= ~mask;
357 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
358 struct lu_attr *la, struct filter_fid *ff)
360 struct ofd_thread_info *info = ofd_info(env);
361 struct ofd_device *ofd = ofd_obj2dev(fo);
363 struct ofd_mod_data *fmd;
368 ofd_write_lock(env, fo);
369 if (!ofd_object_exists(fo))
370 GOTO(unlock, rc = -ENOENT);
372 if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
373 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
374 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
375 fmd->fmd_mactime_xid = info->fti_xid;
376 ofd_fmd_put(info->fti_exp, fmd);
379 /* VBR: version recovery check */
380 rc = ofd_version_get_check(info, fo);
384 rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
389 rc = ofd_object_ff_check(env, fo);
396 th = ofd_trans_create(env, ofd);
398 GOTO(unlock, rc = PTR_ERR(th));
400 rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
405 info->fti_buf.lb_buf = ff;
406 info->fti_buf.lb_len = sizeof(*ff);
407 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
408 &info->fti_buf, XATTR_NAME_FID, 0,
414 rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
418 rc = dt_attr_set(env, ofd_object_child(fo), la, th,
419 ofd_object_capa(env, fo));
424 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
425 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
428 ofd_trans_stop(env, ofd, th, rc);
430 ofd_write_unlock(env, fo);
434 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
435 __u64 start, __u64 end, struct lu_attr *la,
436 struct filter_fid *ff)
438 struct ofd_thread_info *info = ofd_info(env);
439 struct ofd_device *ofd = ofd_obj2dev(fo);
440 struct ofd_mod_data *fmd;
441 struct dt_object *dob = ofd_object_child(fo);
448 /* we support truncate, not punch yet */
449 LASSERT(end == OBD_OBJECT_EOF);
451 fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
452 if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
453 fmd->fmd_mactime_xid = info->fti_xid;
454 ofd_fmd_put(info->fti_exp, fmd);
456 ofd_write_lock(env, fo);
457 if (!ofd_object_exists(fo))
458 GOTO(unlock, rc = -ENOENT);
460 /* VBR: version recovery check */
461 rc = ofd_version_get_check(info, fo);
465 rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
470 rc = ofd_object_ff_check(env, fo);
477 th = ofd_trans_create(env, ofd);
479 GOTO(unlock, rc = PTR_ERR(th));
481 rc = dt_declare_attr_set(env, dob, la, th);
485 rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
490 info->fti_buf.lb_buf = ff;
491 info->fti_buf.lb_len = sizeof(*ff);
492 rc = dt_declare_xattr_set(env, ofd_object_child(fo),
493 &info->fti_buf, XATTR_NAME_FID, 0,
499 rc = ofd_trans_start(env, ofd, fo, th);
503 rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
504 ofd_object_capa(env, fo));
508 rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
513 rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
514 XATTR_NAME_FID, 0, th, BYPASS_CAPA);
517 ofd_trans_stop(env, ofd, th, rc);
519 ofd_write_unlock(env, fo);
523 int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo,
526 struct ofd_device *ofd = ofd_obj2dev(fo);
532 ofd_write_lock(env, fo);
533 if (!ofd_object_exists(fo))
534 GOTO(unlock, rc = -ENOENT);
536 th = ofd_trans_create(env, ofd);
538 GOTO(unlock, rc = PTR_ERR(th));
540 dt_declare_ref_del(env, ofd_object_child(fo), th);
541 dt_declare_destroy(env, ofd_object_child(fo), th);
543 rc = dt_trans_start_local(env, ofd->ofd_osd, th);
545 rc = ofd_trans_start(env, ofd, NULL, th);
549 ofd_fmd_drop(ofd_info(env)->fti_exp, &fo->ofo_header.loh_fid);
551 dt_ref_del(env, ofd_object_child(fo), th);
552 dt_destroy(env, ofd_object_child(fo), th);
554 ofd_trans_stop(env, ofd, th, rc);
556 ofd_write_unlock(env, fo);
560 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
567 if (ofd_object_exists(fo)) {
568 rc = dt_attr_get(env, ofd_object_child(fo), la,
569 ofd_object_capa(env, fo));
571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
572 /* Try to correct for a bug in 2.1.0 (LU-221) that caused
573 * negative timestamps to appear to be in the far future,
574 * due old timestamp being stored on disk as an unsigned value.
575 * This fixes up any bad values stored on disk before
576 * returning them to the client, and ensures any timestamp
577 * updates are correct. LU-1042 */
578 if (unlikely(la->la_atime == LU221_BAD_TIME))
580 if (unlikely(la->la_mtime == LU221_BAD_TIME))
582 if (unlikely(la->la_ctime == LU221_BAD_TIME))
585 #warning "remove old LU-221/LU-1042 workaround code"