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) 2010, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 #include <dt_object.h>
54 #include <lustre_acl.h>
55 #include <lustre_export.h>
56 #include <lustre_ioctl.h>
57 #include <lustre_lfsck.h>
58 #include <lustre_log.h>
59 #include <lustre_net.h>
60 #include <lustre_nodemap.h>
61 #include <lustre_mds.h>
62 #include <lustre_param.h>
63 #include <lustre_quota.h>
64 #include <lustre_swab.h>
66 #include <obd_support.h>
68 #include <llog_swab.h>
70 #include "mdt_internal.h"
73 static unsigned int max_mod_rpcs_per_client = 8;
74 module_param(max_mod_rpcs_per_client, uint, 0644);
75 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78 [LCK_MINMODE] = MDL_MINMODE,
85 [LCK_GROUP] = MDL_GROUP
88 enum ldlm_mode mdt_dlm_lock_modes[] = {
89 [MDL_MINMODE] = LCK_MINMODE,
96 [MDL_GROUP] = LCK_GROUP
99 static struct mdt_device *mdt_dev(struct lu_device *d);
100 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
102 static const struct lu_object_operations mdt_obj_ops;
104 /* Slab for MDT object allocation */
105 static struct kmem_cache *mdt_object_kmem;
107 /* For HSM restore handles */
108 struct kmem_cache *mdt_hsm_cdt_kmem;
110 /* For HSM request handles */
111 struct kmem_cache *mdt_hsm_car_kmem;
113 static struct lu_kmem_descr mdt_caches[] = {
115 .ckd_cache = &mdt_object_kmem,
116 .ckd_name = "mdt_obj",
117 .ckd_size = sizeof(struct mdt_object)
120 .ckd_cache = &mdt_hsm_cdt_kmem,
121 .ckd_name = "mdt_cdt_restore_handle",
122 .ckd_size = sizeof(struct cdt_restore_handle)
125 .ckd_cache = &mdt_hsm_car_kmem,
126 .ckd_name = "mdt_cdt_agent_req",
127 .ckd_size = sizeof(struct cdt_agent_req)
134 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
138 return rep->lock_policy_res1 & op_flag;
141 void mdt_clear_disposition(struct mdt_thread_info *info,
142 struct ldlm_reply *rep, __u64 op_flag)
145 info->mti_opdata &= ~op_flag;
146 tgt_opdata_clear(info->mti_env, op_flag);
149 rep->lock_policy_res1 &= ~op_flag;
152 void mdt_set_disposition(struct mdt_thread_info *info,
153 struct ldlm_reply *rep, __u64 op_flag)
156 info->mti_opdata |= op_flag;
157 tgt_opdata_set(info->mti_env, op_flag);
160 rep->lock_policy_res1 |= op_flag;
163 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
165 lh->mlh_pdo_hash = 0;
166 lh->mlh_reg_mode = lm;
167 lh->mlh_rreg_mode = lm;
168 lh->mlh_type = MDT_REG_LOCK;
171 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
172 const struct lu_name *lname)
174 lh->mlh_reg_mode = lock_mode;
175 lh->mlh_pdo_mode = LCK_MINMODE;
176 lh->mlh_rreg_mode = lock_mode;
177 lh->mlh_type = MDT_PDO_LOCK;
179 if (lu_name_is_valid(lname)) {
180 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
182 /* XXX Workaround for LU-2856
184 * Zero is a valid return value of full_name_hash, but
185 * several users of mlh_pdo_hash assume a non-zero
186 * hash value. We therefore map zero onto an
187 * arbitrary, but consistent value (1) to avoid
188 * problems further down the road. */
189 if (unlikely(lh->mlh_pdo_hash == 0))
190 lh->mlh_pdo_hash = 1;
192 lh->mlh_pdo_hash = 0;
196 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
197 struct mdt_lock_handle *lh)
203 * Any dir access needs couple of locks:
205 * 1) on part of dir we gonna take lookup/modify;
207 * 2) on whole dir to protect it from concurrent splitting and/or to
208 * flush client's cache for readdir().
210 * so, for a given mode and object this routine decides what lock mode
211 * to use for lock #2:
213 * 1) if caller's gonna lookup in dir then we need to protect dir from
214 * being splitted only - LCK_CR
216 * 2) if caller's gonna modify dir then we need to protect dir from
217 * being splitted and to flush cache - LCK_CW
219 * 3) if caller's gonna modify dir and that dir seems ready for
220 * splitting then we need to protect it from any type of access
221 * (lookup/modify/split) - LCK_EX --bzzz
224 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
225 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
228 * Ask underlaying level its opinion about preferable PDO lock mode
229 * having access type passed as regular lock mode:
231 * - MDL_MINMODE means that lower layer does not want to specify lock
234 * - MDL_NL means that no PDO lock should be taken. This is used in some
235 * cases. Say, for non-splittable directories no need to use PDO locks
238 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
239 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
241 if (mode != MDL_MINMODE) {
242 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
245 * Lower layer does not want to specify locking mode. We do it
246 * our selves. No special protection is needed, just flush
247 * client's cache on modification and allow concurrent
250 switch (lh->mlh_reg_mode) {
252 lh->mlh_pdo_mode = LCK_EX;
255 lh->mlh_pdo_mode = LCK_CR;
258 lh->mlh_pdo_mode = LCK_CW;
261 CERROR("Not expected lock type (0x%x)\n",
262 (int)lh->mlh_reg_mode);
267 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
271 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
274 struct mdt_device *mdt = info->mti_mdt;
275 struct lu_name *lname = &info->mti_name;
277 struct mdt_object *parent;
281 LASSERT(!info->mti_cross_ref);
283 OBD_ALLOC(name, NAME_MAX + 1);
286 lname->ln_name = name;
289 * We may want to allow this to mount a completely separate
290 * fileset from the MDT in the future, but keeping it to
291 * ROOT/ only for now avoid potential security issues.
293 *fid = mdt->mdt_md_root_fid;
295 while (rc == 0 && fileset != NULL && *fileset != '\0') {
296 const char *s1 = fileset;
302 while (*s2 != '/' && *s2 != '\0')
310 lname->ln_namelen = s2 - s1;
311 if (lname->ln_namelen > NAME_MAX) {
316 /* reject .. as a path component */
317 if (lname->ln_namelen == 2 &&
318 strncmp(s1, "..", 2) == 0) {
323 strncpy(name, s1, lname->ln_namelen);
324 name[lname->ln_namelen] = '\0';
326 parent = mdt_object_find(info->mti_env, mdt, fid);
327 if (IS_ERR(parent)) {
328 rc = PTR_ERR(parent);
331 /* Only got the fid of this obj by name */
333 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
334 fid, &info->mti_spec);
335 mdt_object_put(info->mti_env, parent);
338 parent = mdt_object_find(info->mti_env, mdt, fid);
340 rc = PTR_ERR(parent);
342 mode = lu_object_attr(&parent->mot_obj);
343 mdt_object_put(info->mti_env, parent);
349 OBD_FREE(name, NAME_MAX + 1);
354 static int mdt_get_root(struct tgt_session_info *tsi)
356 struct mdt_thread_info *info = tsi2mdt_info(tsi);
357 struct mdt_device *mdt = info->mti_mdt;
358 struct mdt_body *repbody;
359 char *fileset = NULL, *buffer = NULL;
361 struct obd_export *exp = info->mti_exp;
362 char *nodemap_fileset;
366 rc = mdt_check_ucred(info);
368 GOTO(out, rc = err_serious(rc));
370 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
371 GOTO(out, rc = err_serious(-ENOMEM));
373 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
374 if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
375 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
377 GOTO(out, rc = err_serious(-EFAULT));
380 nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
381 if (nodemap_fileset && nodemap_fileset[0]) {
383 /* consider fileset from client as a sub-fileset
384 * of the nodemap one */
385 OBD_ALLOC(buffer, PATH_MAX + 1);
387 GOTO(out, rc = err_serious(-ENOMEM));
388 if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
389 nodemap_fileset, fileset) >= PATH_MAX + 1)
390 GOTO(out, rc = err_serious(-EINVAL));
393 /* enforce fileset as specified in the nodemap */
394 fileset = nodemap_fileset;
399 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
401 GOTO(out, rc = err_serious(rc));
403 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
405 repbody->mbo_valid |= OBD_MD_FLID;
409 mdt_thread_info_fini(info);
411 OBD_FREE(buffer, PATH_MAX+1);
415 static int mdt_statfs(struct tgt_session_info *tsi)
417 struct ptlrpc_request *req = tgt_ses_req(tsi);
418 struct mdt_thread_info *info = tsi2mdt_info(tsi);
419 struct md_device *next = info->mti_mdt->mdt_child;
420 struct ptlrpc_service_part *svcpt;
421 struct obd_statfs *osfs;
426 svcpt = req->rq_rqbd->rqbd_svcpt;
428 /* This will trigger a watchdog timeout */
429 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
430 (MDT_SERVICE_WATCHDOG_FACTOR *
431 at_get(&svcpt->scp_at_estimate)) + 1);
433 rc = mdt_check_ucred(info);
435 GOTO(out, rc = err_serious(rc));
437 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
438 GOTO(out, rc = err_serious(-ENOMEM));
440 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
442 GOTO(out, rc = -EPROTO);
444 /** statfs information are cached in the mdt_device */
445 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
446 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
447 /** statfs data is too old, get up-to-date one */
448 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
451 spin_lock(&info->mti_mdt->mdt_osfs_lock);
452 info->mti_mdt->mdt_osfs = *osfs;
453 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
454 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
456 /** use cached statfs data */
457 spin_lock(&info->mti_mdt->mdt_osfs_lock);
458 *osfs = info->mti_mdt->mdt_osfs;
459 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
463 mdt_counter_incr(req, LPROC_MDT_STATFS);
465 mdt_thread_info_fini(info);
469 #ifdef CONFIG_FS_POSIX_ACL
471 * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
473 * \param info thread info object
474 * \param repbody reply to pack ACLs into
475 * \param o mdt object of file to examine
476 * \param nodemap nodemap of client to reply to
478 * \retval -errno error getting or parsing ACL from disk
480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
481 struct mdt_object *o, struct lu_nodemap *nodemap)
483 const struct lu_env *env = info->mti_env;
484 struct md_object *next = mdt_object_child(o);
485 struct lu_buf *buf = &info->mti_buf;
488 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
489 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
491 if (buf->lb_len == 0)
494 rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
496 if (rc == -ENODATA) {
497 repbody->mbo_aclsize = 0;
498 repbody->mbo_valid |= OBD_MD_FLACL;
500 } else if (rc == -EOPNOTSUPP) {
503 CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
504 mdt_obd_name(info->mti_mdt),
505 PFID(mdt_object_fid(o)), rc);
508 rc = nodemap_map_acl(nodemap, buf->lb_buf,
509 rc, NODEMAP_FS_TO_CLIENT);
510 /* if all ACLs mapped out, rc is still >= 0 */
512 CERROR("%s: nodemap_map_acl unable to parse "DFID
513 " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
514 PFID(mdt_object_fid(o)), rc);
516 repbody->mbo_aclsize = rc;
517 repbody->mbo_valid |= OBD_MD_FLACL;
525 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
526 const struct lu_attr *attr, const struct lu_fid *fid)
528 struct md_attr *ma = &info->mti_attr;
529 struct obd_export *exp = info->mti_exp;
530 struct lu_nodemap *nodemap = NULL;
532 LASSERT(ma->ma_valid & MA_INODE);
534 if (attr->la_valid & LA_ATIME) {
535 b->mbo_atime = attr->la_atime;
536 b->mbo_valid |= OBD_MD_FLATIME;
538 if (attr->la_valid & LA_MTIME) {
539 b->mbo_mtime = attr->la_mtime;
540 b->mbo_valid |= OBD_MD_FLMTIME;
542 if (attr->la_valid & LA_CTIME) {
543 b->mbo_ctime = attr->la_ctime;
544 b->mbo_valid |= OBD_MD_FLCTIME;
546 if (attr->la_valid & LA_FLAGS) {
547 b->mbo_flags = attr->la_flags;
548 b->mbo_valid |= OBD_MD_FLFLAGS;
550 if (attr->la_valid & LA_NLINK) {
551 b->mbo_nlink = attr->la_nlink;
552 b->mbo_valid |= OBD_MD_FLNLINK;
554 if (attr->la_valid & (LA_UID|LA_GID)) {
555 nodemap = nodemap_get_from_exp(exp);
559 if (attr->la_valid & LA_UID) {
560 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
561 NODEMAP_FS_TO_CLIENT,
563 b->mbo_valid |= OBD_MD_FLUID;
565 if (attr->la_valid & LA_GID) {
566 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
567 NODEMAP_FS_TO_CLIENT,
569 b->mbo_valid |= OBD_MD_FLGID;
572 b->mbo_mode = attr->la_mode;
573 if (attr->la_valid & LA_MODE)
574 b->mbo_valid |= OBD_MD_FLMODE;
575 if (attr->la_valid & LA_TYPE)
576 b->mbo_valid |= OBD_MD_FLTYPE;
580 b->mbo_valid |= OBD_MD_FLID;
581 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
582 PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
586 mdt_body_reverse_idmap(info, b);
588 if (!(attr->la_valid & LA_TYPE))
591 b->mbo_rdev = attr->la_rdev;
592 b->mbo_size = attr->la_size;
593 b->mbo_blocks = attr->la_blocks;
595 if (!S_ISREG(attr->la_mode)) {
596 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
597 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
598 /* means no objects are allocated on osts. */
599 LASSERT(!(ma->ma_valid & MA_LOV));
600 /* just ignore blocks occupied by extend attributes on MDS */
602 /* if no object is allocated on osts, the size on mds is valid.
604 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
605 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
606 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
607 /* A released file stores its size on MDS. */
608 /* But return 1 block for released file, unless tools like tar
609 * will consider it fully sparse. (LU-3864)
611 if (unlikely(b->mbo_size == 0))
615 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
618 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
619 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
620 PFID(fid), (unsigned long long)b->mbo_size);
623 if (!IS_ERR_OR_NULL(nodemap))
624 nodemap_putref(nodemap);
627 static inline int mdt_body_has_lov(const struct lu_attr *la,
628 const struct mdt_body *body)
630 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
631 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
634 void mdt_client_compatibility(struct mdt_thread_info *info)
636 struct mdt_body *body;
637 struct ptlrpc_request *req = mdt_info_req(info);
638 struct obd_export *exp = req->rq_export;
639 struct md_attr *ma = &info->mti_attr;
640 struct lu_attr *la = &ma->ma_attr;
643 if (exp_connect_layout(exp))
644 /* the client can deal with 16-bit lmm_stripe_count */
647 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
649 if (!mdt_body_has_lov(la, body))
652 /* now we have a reply with a lov for a client not compatible with the
653 * layout lock so we have to clean the layout generation number */
654 if (S_ISREG(la->la_mode))
655 ma->ma_lmm->lmm_layout_gen = 0;
659 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
660 struct mdt_object *o)
662 const struct lu_env *env = info->mti_env;
665 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
674 /* Is it a directory? Let's check for the LMV as well */
675 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
676 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
680 rc2 = mo_xattr_get(env, mdt_object_child(o),
682 XATTR_NAME_DEFAULT_LMV);
684 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
692 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
695 const struct lu_env *env = info->mti_env;
699 LASSERT(info->mti_big_lmm_used == 0);
700 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
704 /* big_lmm may need to be grown */
705 if (info->mti_big_lmmsize < rc) {
706 int size = size_roundup_power2(rc);
708 if (info->mti_big_lmmsize > 0) {
709 /* free old buffer */
710 LASSERT(info->mti_big_lmm);
711 OBD_FREE_LARGE(info->mti_big_lmm,
712 info->mti_big_lmmsize);
713 info->mti_big_lmm = NULL;
714 info->mti_big_lmmsize = 0;
717 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
718 if (info->mti_big_lmm == NULL)
720 info->mti_big_lmmsize = size;
722 LASSERT(info->mti_big_lmmsize >= rc);
724 info->mti_buf.lb_buf = info->mti_big_lmm;
725 info->mti_buf.lb_len = info->mti_big_lmmsize;
726 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
731 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
732 struct md_attr *ma, const char *name)
734 struct md_object *next = mdt_object_child(o);
735 struct lu_buf *buf = &info->mti_buf;
738 if (strcmp(name, XATTR_NAME_LOV) == 0) {
739 buf->lb_buf = ma->ma_lmm;
740 buf->lb_len = ma->ma_lmm_size;
741 LASSERT(!(ma->ma_valid & MA_LOV));
742 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
743 buf->lb_buf = ma->ma_lmv;
744 buf->lb_len = ma->ma_lmv_size;
745 LASSERT(!(ma->ma_valid & MA_LMV));
746 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
747 buf->lb_buf = ma->ma_lmv;
748 buf->lb_len = ma->ma_lmv_size;
749 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
754 rc = mo_xattr_get(info->mti_env, next, buf, name);
758 if (strcmp(name, XATTR_NAME_LOV) == 0) {
759 if (info->mti_big_lmm_used)
760 ma->ma_lmm = info->mti_big_lmm;
762 /* NOT return LOV EA with hole to old client. */
763 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
764 LOV_PATTERN_F_HOLE) &&
765 !(exp_connect_flags(info->mti_exp) &
766 OBD_CONNECT_LFSCK)) {
769 ma->ma_lmm_size = rc;
770 ma->ma_valid |= MA_LOV;
772 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
773 if (info->mti_big_lmm_used)
774 ma->ma_lmv = info->mti_big_lmm;
776 ma->ma_lmv_size = rc;
777 ma->ma_valid |= MA_LMV;
778 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
779 ma->ma_lmv_size = rc;
780 ma->ma_valid |= MA_LMV_DEF;
783 /* Update mdt_max_mdsize so all clients will be aware that */
784 if (info->mti_mdt->mdt_max_mdsize < rc)
785 info->mti_mdt->mdt_max_mdsize = rc;
788 } else if (rc == -ENODATA) {
791 } else if (rc == -ERANGE) {
792 /* Default LMV has fixed size, so it must be able to fit
793 * in the original buffer */
794 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
796 rc = mdt_big_xattr_get(info, o, name);
798 info->mti_big_lmm_used = 1;
806 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
807 struct mdt_object *o, struct lu_fid *pfid)
809 struct lu_buf *buf = &info->mti_buf;
810 struct link_ea_header *leh;
811 struct link_ea_entry *lee;
815 buf->lb_buf = info->mti_big_lmm;
816 buf->lb_len = info->mti_big_lmmsize;
817 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
818 buf, XATTR_NAME_LINK);
819 /* ignore errors, MA_PFID won't be set and it is
820 * up to the caller to treat this as an error */
821 if (rc == -ERANGE || buf->lb_len == 0) {
822 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
823 buf->lb_buf = info->mti_big_lmm;
824 buf->lb_len = info->mti_big_lmmsize;
829 if (rc < sizeof(*leh)) {
830 CERROR("short LinkEA on "DFID": rc = %d\n",
831 PFID(mdt_object_fid(o)), rc);
835 leh = (struct link_ea_header *) buf->lb_buf;
836 lee = (struct link_ea_entry *)(leh + 1);
837 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
838 leh->leh_magic = LINK_EA_MAGIC;
839 leh->leh_reccount = __swab32(leh->leh_reccount);
840 leh->leh_len = __swab64(leh->leh_len);
842 if (leh->leh_magic != LINK_EA_MAGIC)
844 if (leh->leh_reccount == 0)
847 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
848 fid_be_to_cpu(pfid, pfid);
853 int mdt_attr_get_complex(struct mdt_thread_info *info,
854 struct mdt_object *o, struct md_attr *ma)
856 const struct lu_env *env = info->mti_env;
857 struct md_object *next = mdt_object_child(o);
858 struct lu_buf *buf = &info->mti_buf;
859 int need = ma->ma_need;
866 if (mdt_object_exists(o) == 0)
867 GOTO(out, rc = -ENOENT);
868 mode = lu_object_attr(&next->mo_lu);
870 if (need & MA_INODE) {
871 ma->ma_need = MA_INODE;
872 rc = mo_attr_get(env, next, ma);
875 ma->ma_valid |= MA_INODE;
878 if (need & MA_PFID) {
879 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
881 ma->ma_valid |= MA_PFID;
882 /* ignore this error, parent fid is not mandatory */
886 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
887 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
892 if (need & MA_LMV && S_ISDIR(mode)) {
893 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
898 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
899 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
904 if (need & MA_HSM && S_ISREG(mode)) {
905 buf->lb_buf = info->mti_xattr_buf;
906 buf->lb_len = sizeof(info->mti_xattr_buf);
907 CLASSERT(sizeof(struct hsm_attrs) <=
908 sizeof(info->mti_xattr_buf));
909 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
910 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
912 ma->ma_valid |= MA_HSM;
913 else if (rc2 < 0 && rc2 != -ENODATA)
917 #ifdef CONFIG_FS_POSIX_ACL
918 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
919 buf->lb_buf = ma->ma_acl;
920 buf->lb_len = ma->ma_acl_size;
921 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
923 ma->ma_acl_size = rc2;
924 ma->ma_valid |= MA_ACL_DEF;
925 } else if (rc2 == -ENODATA) {
934 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
935 rc, ma->ma_valid, ma->ma_lmm);
939 static int mdt_getattr_internal(struct mdt_thread_info *info,
940 struct mdt_object *o, int ma_need)
942 struct md_object *next = mdt_object_child(o);
943 const struct mdt_body *reqbody = info->mti_body;
944 struct ptlrpc_request *req = mdt_info_req(info);
945 struct md_attr *ma = &info->mti_attr;
946 struct lu_attr *la = &ma->ma_attr;
947 struct req_capsule *pill = info->mti_pill;
948 const struct lu_env *env = info->mti_env;
949 struct mdt_body *repbody;
950 struct lu_buf *buffer = &info->mti_buf;
951 struct obd_export *exp = info->mti_exp;
956 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
957 RETURN(err_serious(-ENOMEM));
959 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
963 if (mdt_object_remote(o)) {
964 /* This object is located on remote node.*/
965 /* Return -ENOTSUPP for old client */
966 if (!mdt_is_dne_client(req->rq_export))
967 GOTO(out, rc = -ENOTSUPP);
969 repbody->mbo_fid1 = *mdt_object_fid(o);
970 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
974 if (reqbody->mbo_eadatasize > 0) {
975 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
976 if (buffer->lb_buf == NULL)
977 GOTO(out, rc = -EPROTO);
978 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
981 buffer->lb_buf = NULL;
983 ma_need &= ~(MA_LOV | MA_LMV);
984 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
985 mdt_obd_name(info->mti_mdt),
986 req->rq_export->exp_client_uuid.uuid);
989 /* If it is dir object and client require MEA, then we got MEA */
990 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
991 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
992 /* Assumption: MDT_MD size is enough for lmv size. */
993 ma->ma_lmv = buffer->lb_buf;
994 ma->ma_lmv_size = buffer->lb_len;
995 ma->ma_need = MA_INODE;
996 if (ma->ma_lmv_size > 0) {
997 if (reqbody->mbo_valid & OBD_MD_MEA)
998 ma->ma_need |= MA_LMV;
999 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1000 ma->ma_need |= MA_LMV_DEF;
1003 ma->ma_lmm = buffer->lb_buf;
1004 ma->ma_lmm_size = buffer->lb_len;
1005 ma->ma_need = MA_INODE | MA_HSM;
1006 if (ma->ma_lmm_size > 0)
1007 ma->ma_need |= MA_LOV;
1010 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1011 reqbody->mbo_valid & OBD_MD_FLDIREA &&
1012 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1013 /* get default stripe info for this dir. */
1014 ma->ma_need |= MA_LOV_DEF;
1016 ma->ma_need |= ma_need;
1018 rc = mdt_attr_get_complex(info, o, ma);
1020 CERROR("%s: getattr error for "DFID": rc = %d\n",
1021 mdt_obd_name(info->mti_mdt),
1022 PFID(mdt_object_fid(o)), rc);
1026 /* if file is released, check if a restore is running */
1027 if (ma->ma_valid & MA_HSM) {
1028 repbody->mbo_valid |= OBD_MD_TSTATE;
1029 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1030 mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1031 repbody->mbo_t_state = MS_RESTORE;
1034 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
1036 /* the Lustre protocol supposes to return default striping
1037 * on the user-visible root if explicitly requested */
1038 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
1039 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
1040 struct lu_fid rootfid;
1041 struct mdt_object *root;
1042 struct mdt_device *mdt = info->mti_mdt;
1044 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1047 root = mdt_object_find(env, mdt, &rootfid);
1049 RETURN(PTR_ERR(root));
1050 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1051 mdt_object_put(info->mti_env, root);
1053 CERROR("%s: getattr error for "DFID": rc = %d\n",
1054 mdt_obd_name(info->mti_mdt),
1055 PFID(mdt_object_fid(o)), rc);
1060 if (likely(ma->ma_valid & MA_INODE))
1061 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1065 if (mdt_body_has_lov(la, reqbody)) {
1066 if (ma->ma_valid & MA_LOV) {
1067 LASSERT(ma->ma_lmm_size);
1068 repbody->mbo_eadatasize = ma->ma_lmm_size;
1069 if (S_ISDIR(la->la_mode))
1070 repbody->mbo_valid |= OBD_MD_FLDIREA;
1072 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1073 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1075 if (ma->ma_valid & MA_LMV) {
1076 /* Return -ENOTSUPP for old client */
1077 if (!mdt_is_striped_client(req->rq_export))
1080 LASSERT(S_ISDIR(la->la_mode));
1081 mdt_dump_lmv(D_INFO, ma->ma_lmv);
1082 repbody->mbo_eadatasize = ma->ma_lmv_size;
1083 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1085 if (ma->ma_valid & MA_LMV_DEF) {
1086 /* Return -ENOTSUPP for old client */
1087 if (!mdt_is_striped_client(req->rq_export))
1089 LASSERT(S_ISDIR(la->la_mode));
1090 repbody->mbo_eadatasize = ma->ma_lmv_size;
1091 repbody->mbo_valid |= (OBD_MD_FLDIREA |
1092 OBD_MD_DEFAULT_MEA);
1094 } else if (S_ISLNK(la->la_mode) &&
1095 reqbody->mbo_valid & OBD_MD_LINKNAME) {
1096 buffer->lb_buf = ma->ma_lmm;
1097 /* eadatasize from client includes NULL-terminator, so
1098 * there is no need to read it */
1099 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1100 rc = mo_readlink(env, next, buffer);
1101 if (unlikely(rc <= 0)) {
1102 CERROR("%s: readlink failed for "DFID": rc = %d\n",
1103 mdt_obd_name(info->mti_mdt),
1104 PFID(mdt_object_fid(o)), rc);
1107 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1109 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1111 repbody->mbo_valid |= OBD_MD_LINKNAME;
1112 /* we need to report back size with NULL-terminator
1113 * because client expects that */
1114 repbody->mbo_eadatasize = rc + 1;
1115 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1116 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1117 "on "DFID ", expected %d\n",
1118 mdt_obd_name(info->mti_mdt),
1119 rc, PFID(mdt_object_fid(o)),
1120 reqbody->mbo_eadatasize - 1);
1121 /* NULL terminate */
1122 ((char *)ma->ma_lmm)[rc] = 0;
1124 /* If the total CDEBUG() size is larger than a page, it
1125 * will print a warning to the console, avoid this by
1126 * printing just the last part of the symlink. */
1127 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1128 print_limit < rc ? "..." : "", print_limit,
1129 (char *)ma->ma_lmm + rc - print_limit, rc);
1134 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1135 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1136 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1137 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1138 repbody->mbo_max_mdsize);
1141 if (exp_connect_rmtclient(info->mti_exp) &&
1142 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1143 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1145 /* mdt_getattr_lock only */
1146 rc = mdt_pack_remote_perm(info, o, buf);
1148 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1149 repbody->mbo_aclsize = 0;
1152 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1153 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1156 #ifdef CONFIG_FS_POSIX_ACL
1157 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1158 (reqbody->mbo_valid & OBD_MD_FLACL)) {
1159 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1160 if (IS_ERR(nodemap))
1161 RETURN(PTR_ERR(nodemap));
1163 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1164 nodemap_putref(nodemap);
1170 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1175 static int mdt_getattr(struct tgt_session_info *tsi)
1177 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1178 struct mdt_object *obj = info->mti_object;
1179 struct req_capsule *pill = info->mti_pill;
1180 struct mdt_body *reqbody;
1181 struct mdt_body *repbody;
1185 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1187 LASSERT(obj != NULL);
1188 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1190 /* Unlike intent case where we need to pre-fill out buffers early on
1191 * in intent policy for ldlm reasons, here we can have a much better
1192 * guess at EA size by just reading it from disk.
1193 * Exceptions are readdir and (missing) directory striping */
1195 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1196 /* No easy way to know how long is the symlink, but it cannot
1197 * be more than PATH_MAX, so we allocate +1 */
1200 /* A special case for fs ROOT: getattr there might fetch
1201 * default EA for entire fs, not just for this dir!
1203 } else if (lu_fid_eq(mdt_object_fid(obj),
1204 &info->mti_mdt->mdt_md_root_fid) &&
1205 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1206 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1208 /* Should the default strping be bigger, mdt_fix_reply
1209 * will reallocate */
1210 rc = DEF_REP_MD_SIZE;
1212 /* Read the actual EA size from disk */
1213 rc = mdt_attr_get_eabuf_size(info, obj);
1217 GOTO(out, rc = err_serious(rc));
1219 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1221 rc = req_capsule_server_pack(pill);
1222 if (unlikely(rc != 0))
1223 GOTO(out, rc = err_serious(rc));
1225 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1226 LASSERT(repbody != NULL);
1227 repbody->mbo_eadatasize = 0;
1228 repbody->mbo_aclsize = 0;
1230 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1231 rc = mdt_init_ucred(info, reqbody);
1233 rc = mdt_check_ucred(info);
1235 GOTO(out_shrink, rc);
1237 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1239 rc = mdt_getattr_internal(info, obj, 0);
1240 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1241 mdt_exit_ucred(info);
1244 mdt_client_compatibility(info);
1245 rc2 = mdt_fix_reply(info);
1249 mdt_thread_info_fini(info);
1254 * Exchange MOF_LOV_CREATED flags between two objects after a
1255 * layout swap. No assumption is made on whether o1 or o2 have
1256 * created objects or not.
1258 * \param[in,out] o1 First swap layout object
1259 * \param[in,out] o2 Second swap layout object
1261 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1265 mutex_lock(&o1->mot_lov_mutex);
1266 mutex_lock(&o2->mot_lov_mutex);
1268 o1_flags = o1->mot_flags;
1269 o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1270 (o2->mot_flags & MOF_LOV_CREATED);
1272 o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1273 (o1_flags & MOF_LOV_CREATED);
1275 mutex_unlock(&o2->mot_lov_mutex);
1276 mutex_unlock(&o1->mot_lov_mutex);
1279 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1281 struct mdt_thread_info *info;
1282 struct ptlrpc_request *req = tgt_ses_req(tsi);
1283 struct obd_export *exp = req->rq_export;
1284 struct mdt_object *o1, *o2, *o;
1285 struct mdt_lock_handle *lh1, *lh2;
1286 struct mdc_swap_layouts *msl;
1290 /* client does not support layout lock, so layout swaping
1292 * FIXME: there is a problem for old clients which don't support
1293 * layout lock yet. If those clients have already opened the file
1294 * they won't be notified at all so that old layout may still be
1295 * used to do IO. This can be fixed after file release is landed by
1296 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1297 if (!exp_connect_layout(exp))
1298 RETURN(-EOPNOTSUPP);
1300 info = tsi2mdt_info(tsi);
1302 if (info->mti_dlm_req != NULL)
1303 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1305 o1 = info->mti_object;
1306 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1307 &info->mti_body->mbo_fid2);
1309 GOTO(out, rc = PTR_ERR(o));
1311 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1312 GOTO(put, rc = -ENOENT);
1314 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1315 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1321 /* permission check. Make sure the calling process having permission
1322 * to write both files. */
1323 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1328 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1333 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1335 GOTO(put, rc = -EPROTO);
1337 lh1 = &info->mti_lh[MDT_LH_NEW];
1338 mdt_lock_reg_init(lh1, LCK_EX);
1339 lh2 = &info->mti_lh[MDT_LH_OLD];
1340 mdt_lock_reg_init(lh2, LCK_EX);
1342 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1343 MDS_INODELOCK_XATTR);
1347 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1348 MDS_INODELOCK_XATTR);
1352 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1353 mdt_object_child(o2), msl->msl_flags);
1357 mdt_swap_lov_flag(o1, o2);
1360 mdt_object_unlock(info, o2, lh2, rc);
1362 mdt_object_unlock(info, o1, lh1, rc);
1364 mdt_object_put(info->mti_env, o);
1366 mdt_thread_info_fini(info);
1370 static int mdt_raw_lookup(struct mdt_thread_info *info,
1371 struct mdt_object *parent,
1372 const struct lu_name *lname,
1373 struct ldlm_reply *ldlm_rep)
1375 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1379 LASSERT(!info->mti_cross_ref);
1381 /* Only got the fid of this obj by name */
1382 fid_zero(child_fid);
1383 rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1384 lname, child_fid, &info->mti_spec);
1386 struct mdt_body *repbody;
1388 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1389 repbody->mbo_fid1 = *child_fid;
1390 repbody->mbo_valid = OBD_MD_FLID;
1391 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1392 } else if (rc == -ENOENT) {
1393 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1400 * UPDATE lock should be taken against parent, and be released before exit;
1401 * child_bits lock should be taken against child, and be returned back:
1402 * (1)normal request should release the child lock;
1403 * (2)intent request will grant the lock to client.
1405 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1406 struct mdt_lock_handle *lhc,
1408 struct ldlm_reply *ldlm_rep)
1410 struct ptlrpc_request *req = mdt_info_req(info);
1411 struct mdt_body *reqbody = NULL;
1412 struct mdt_object *parent = info->mti_object;
1413 struct mdt_object *child;
1414 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1415 struct lu_name *lname = NULL;
1416 struct mdt_lock_handle *lhp = NULL;
1417 struct ldlm_lock *lock;
1424 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1425 LASSERT(ergo(is_resent,
1426 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1431 if (info->mti_cross_ref) {
1432 /* Only getattr on the child. Parent is on another node. */
1433 mdt_set_disposition(info, ldlm_rep,
1434 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1436 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1438 PFID(mdt_object_fid(child)), ldlm_rep);
1440 rc = mdt_check_resent_lock(info, child, lhc);
1443 } else if (rc > 0) {
1444 mdt_lock_handle_init(lhc);
1445 mdt_lock_reg_init(lhc, LCK_PR);
1448 * Object's name is on another MDS, no lookup or layout
1449 * lock is needed here but update lock is.
1451 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1452 MDS_INODELOCK_LAYOUT);
1453 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1455 rc = mdt_object_lock(info, child, lhc, child_bits);
1460 /* Finally, we can get attr for child. */
1461 if (!mdt_object_exists(child)) {
1462 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1464 "remote object doesn't exist.");
1465 mdt_object_unlock(info, child, lhc, 1);
1469 rc = mdt_getattr_internal(info, child, 0);
1470 if (unlikely(rc != 0))
1471 mdt_object_unlock(info, child, lhc, 1);
1476 lname = &info->mti_name;
1477 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1479 if (lu_name_is_valid(lname)) {
1480 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1481 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1482 PNAME(lname), ldlm_rep);
1484 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1485 if (unlikely(reqbody == NULL))
1486 RETURN(err_serious(-EPROTO));
1488 *child_fid = reqbody->mbo_fid2;
1490 if (unlikely(!fid_is_sane(child_fid)))
1491 RETURN(err_serious(-EINVAL));
1493 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1495 PFID(mdt_object_fid(parent)),
1496 PFID(&reqbody->mbo_fid2), ldlm_rep);
1499 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1501 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1502 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1504 "Parent doesn't exist!");
1508 if (mdt_object_remote(parent)) {
1509 CERROR("%s: parent "DFID" is on remote target\n",
1510 mdt_obd_name(info->mti_mdt),
1511 PFID(mdt_object_fid(parent)));
1515 if (lu_name_is_valid(lname)) {
1516 /* Always allow to lookup ".." */
1517 if (unlikely(lname->ln_namelen == 2 &&
1518 lname->ln_name[0] == '.' &&
1519 lname->ln_name[1] == '.'))
1520 info->mti_spec.sp_permitted = 1;
1522 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1523 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1528 /* step 1: lock parent only if parent is a directory */
1529 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1530 lhp = &info->mti_lh[MDT_LH_PARENT];
1531 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1532 rc = mdt_object_lock(info, parent, lhp,
1533 MDS_INODELOCK_UPDATE);
1534 if (unlikely(rc != 0))
1538 /* step 2: lookup child's fid by name */
1539 fid_zero(child_fid);
1540 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1541 child_fid, &info->mti_spec);
1543 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1546 GOTO(out_parent, rc);
1549 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1552 *step 3: find the child object by fid & lock it.
1553 * regardless if it is local or remote.
1555 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1556 * set parent dir fid the same as child fid in getattr by fid case
1557 * we should not lu_object_find() the object again, could lead
1558 * to hung if there is a concurrent unlink destroyed the object.
1560 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1561 mdt_object_get(info->mti_env, parent);
1564 child = mdt_object_find(info->mti_env, info->mti_mdt,
1568 if (unlikely(IS_ERR(child)))
1569 GOTO(out_parent, rc = PTR_ERR(child));
1571 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1572 if (!mdt_object_exists(child)) {
1573 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1575 "Object doesn't exist!");
1576 GOTO(out_child, rc = -ENOENT);
1579 rc = mdt_check_resent_lock(info, child, lhc);
1581 GOTO(out_child, rc);
1582 } else if (rc > 0) {
1583 mdt_lock_handle_init(lhc);
1584 mdt_lock_reg_init(lhc, LCK_PR);
1587 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1588 mdt_object_exists(child) && !mdt_object_remote(child)) {
1589 struct md_attr *ma = &info->mti_attr;
1592 ma->ma_need = MA_INODE;
1593 rc = mdt_attr_get_complex(info, child, ma);
1594 if (unlikely(rc != 0))
1595 GOTO(out_child, rc);
1597 /* If the file has not been changed for some time, we
1598 * return not only a LOOKUP lock, but also an UPDATE
1599 * lock and this might save us RPC on later STAT. For
1600 * directories, it also let negative dentry cache start
1601 * working for this dir. */
1602 if (ma->ma_valid & MA_INODE &&
1603 ma->ma_attr.la_valid & LA_CTIME &&
1604 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1605 ma->ma_attr.la_ctime < cfs_time_current_sec())
1606 child_bits |= MDS_INODELOCK_UPDATE;
1609 /* layout lock must be granted in a best-effort way
1610 * for IT operations */
1611 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1612 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1613 exp_connect_layout(info->mti_exp) &&
1614 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1615 !mdt_object_remote(child) && ldlm_rep != NULL) {
1616 /* try to grant layout lock for regular file. */
1622 child_bits |= MDS_INODELOCK_LAYOUT;
1623 /* try layout lock, it may fail to be granted due to
1624 * contention at LOOKUP or UPDATE */
1625 if (!mdt_object_lock_try(info, child, lhc,
1627 child_bits &= ~MDS_INODELOCK_LAYOUT;
1628 LASSERT(child_bits != 0);
1629 rc = mdt_object_lock(info, child, lhc,
1635 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1636 * client will enqueue the lock to the remote MDT */
1637 if (mdt_object_remote(child))
1638 child_bits &= ~MDS_INODELOCK_UPDATE;
1639 rc = mdt_object_lock(info, child, lhc, child_bits);
1641 if (unlikely(rc != 0))
1642 GOTO(out_child, rc);
1645 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1647 /* finally, we can get attr for child. */
1648 rc = mdt_getattr_internal(info, child, ma_need);
1649 if (unlikely(rc != 0)) {
1650 mdt_object_unlock(info, child, lhc, 1);
1652 /* Debugging code. */
1653 LDLM_DEBUG(lock, "Returning lock to client");
1654 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1655 &lock->l_resource->lr_name),
1656 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1657 PLDLMRES(lock->l_resource),
1658 PFID(mdt_object_fid(child)));
1661 LDLM_LOCK_PUT(lock);
1665 mdt_object_put(info->mti_env, child);
1668 mdt_object_unlock(info, parent, lhp, 1);
1672 /* normal handler: should release the child lock */
1673 static int mdt_getattr_name(struct tgt_session_info *tsi)
1675 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1676 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1677 struct mdt_body *reqbody;
1678 struct mdt_body *repbody;
1682 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1683 LASSERT(reqbody != NULL);
1684 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1685 LASSERT(repbody != NULL);
1687 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1688 repbody->mbo_eadatasize = 0;
1689 repbody->mbo_aclsize = 0;
1691 rc = mdt_init_ucred_intent_getattr(info, reqbody);
1693 GOTO(out_shrink, rc);
1695 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1696 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1697 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1698 lhc->mlh_reg_lh.cookie = 0;
1700 mdt_exit_ucred(info);
1703 mdt_client_compatibility(info);
1704 rc2 = mdt_fix_reply(info);
1707 mdt_thread_info_fini(info);
1711 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1712 void *karg, void __user *uarg);
1714 static int mdt_set_info(struct tgt_session_info *tsi)
1716 struct ptlrpc_request *req = tgt_ses_req(tsi);
1719 int keylen, vallen, rc = 0;
1723 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1725 DEBUG_REQ(D_HA, req, "no set_info key");
1726 RETURN(err_serious(-EFAULT));
1729 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1732 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1734 DEBUG_REQ(D_HA, req, "no set_info val");
1735 RETURN(err_serious(-EFAULT));
1738 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1741 /* Swab any part of val you need to here */
1742 if (KEY_IS(KEY_READ_ONLY)) {
1743 spin_lock(&req->rq_export->exp_lock);
1745 *exp_connect_flags_ptr(req->rq_export) |=
1748 *exp_connect_flags_ptr(req->rq_export) &=
1749 ~OBD_CONNECT_RDONLY;
1750 spin_unlock(&req->rq_export->exp_lock);
1751 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1752 struct changelog_setinfo *cs = val;
1754 if (vallen != sizeof(*cs)) {
1755 CERROR("%s: bad changelog_clear setinfo size %d\n",
1756 tgt_name(tsi->tsi_tgt), vallen);
1759 if (ptlrpc_req_need_swab(req)) {
1760 __swab64s(&cs->cs_recno);
1761 __swab32s(&cs->cs_id);
1764 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1766 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1768 obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1775 static int mdt_readpage(struct tgt_session_info *tsi)
1777 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1778 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1779 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1780 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1781 struct mdt_body *repbody;
1787 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1788 RETURN(err_serious(-ENOMEM));
1790 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1791 if (repbody == NULL || reqbody == NULL)
1792 RETURN(err_serious(-EFAULT));
1795 * prepare @rdpg before calling lower layers and transfer itself. Here
1796 * reqbody->size contains offset of where to start to read and
1797 * reqbody->nlink contains number bytes to read.
1799 rdpg->rp_hash = reqbody->mbo_size;
1800 if (rdpg->rp_hash != reqbody->mbo_size) {
1801 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1802 rdpg->rp_hash, reqbody->mbo_size);
1806 rdpg->rp_attrs = reqbody->mbo_mode;
1807 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1808 rdpg->rp_attrs |= LUDA_64BITHASH;
1809 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1810 exp_max_brw_size(tsi->tsi_exp));
1811 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1813 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1814 if (rdpg->rp_pages == NULL)
1817 for (i = 0; i < rdpg->rp_npages; ++i) {
1818 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1819 if (rdpg->rp_pages[i] == NULL)
1820 GOTO(free_rdpg, rc = -ENOMEM);
1823 /* call lower layers to fill allocated pages with directory data */
1824 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1826 GOTO(free_rdpg, rc);
1828 /* send pages to client */
1829 rc = tgt_sendpage(tsi, rdpg, rc);
1834 for (i = 0; i < rdpg->rp_npages; i++)
1835 if (rdpg->rp_pages[i] != NULL)
1836 __free_page(rdpg->rp_pages[i]);
1837 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1839 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1845 static int mdt_reint_internal(struct mdt_thread_info *info,
1846 struct mdt_lock_handle *lhc,
1849 struct req_capsule *pill = info->mti_pill;
1850 struct mdt_body *repbody;
1855 rc = mdt_reint_unpack(info, op);
1857 CERROR("Can't unpack reint, rc %d\n", rc);
1858 RETURN(err_serious(rc));
1861 /* for replay (no_create) lmm is not needed, client has it already */
1862 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1863 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1866 /* llog cookies are always 0, the field is kept for compatibility */
1867 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1868 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1870 rc = req_capsule_server_pack(pill);
1872 CERROR("Can't pack response, rc %d\n", rc);
1873 RETURN(err_serious(rc));
1876 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1877 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1879 repbody->mbo_eadatasize = 0;
1880 repbody->mbo_aclsize = 0;
1883 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1885 /* for replay no cookkie / lmm need, because client have this already */
1886 if (info->mti_spec.no_create)
1887 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1888 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1890 rc = mdt_init_ucred_reint(info);
1892 GOTO(out_shrink, rc);
1894 rc = mdt_fix_attr_ucred(info, op);
1896 GOTO(out_ucred, rc = err_serious(rc));
1898 rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1900 GOTO(out_ucred, rc);
1901 } else if (rc == 1) {
1902 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1903 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1904 GOTO(out_ucred, rc);
1906 rc = mdt_reint_rec(info, lhc);
1909 mdt_exit_ucred(info);
1911 mdt_client_compatibility(info);
1912 rc2 = mdt_fix_reply(info);
1918 static long mdt_reint_opcode(struct ptlrpc_request *req,
1919 const struct req_format **fmt)
1921 struct mdt_device *mdt;
1922 struct mdt_rec_reint *rec;
1925 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1927 opc = rec->rr_opcode;
1928 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1929 if (opc < REINT_MAX && fmt[opc] != NULL)
1930 req_capsule_extend(&req->rq_pill, fmt[opc]);
1932 mdt = mdt_exp2dev(req->rq_export);
1933 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1934 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1935 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1936 opc = err_serious(-EFAULT);
1939 opc = err_serious(-EFAULT);
1944 static int mdt_reint(struct tgt_session_info *tsi)
1948 static const struct req_format *reint_fmts[REINT_MAX] = {
1949 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1950 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1951 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1952 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1953 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1954 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1955 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1956 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1957 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1962 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1964 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1966 * No lock possible here from client to pass it to reint code
1969 rc = mdt_reint_internal(info, NULL, opc);
1970 mdt_thread_info_fini(info);
1975 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1979 /* this should sync the whole device */
1980 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1982 struct dt_device *dt = mdt->mdt_bottom;
1986 rc = dt->dd_ops->dt_sync(env, dt);
1990 /* this should sync this object */
1991 static int mdt_object_sync(struct mdt_thread_info *info)
1993 struct md_object *next;
1997 if (!mdt_object_exists(info->mti_object)) {
1998 CWARN("%s: non existing object "DFID": rc = %d\n",
1999 mdt_obd_name(info->mti_mdt),
2000 PFID(mdt_object_fid(info->mti_object)), -ESTALE);
2003 next = mdt_object_child(info->mti_object);
2004 rc = mo_object_sync(info->mti_env, next);
2009 static int mdt_sync(struct tgt_session_info *tsi)
2011 struct ptlrpc_request *req = tgt_ses_req(tsi);
2012 struct req_capsule *pill = tsi->tsi_pill;
2013 struct mdt_body *body;
2018 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2019 RETURN(err_serious(-ENOMEM));
2021 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2022 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2024 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2026 /* sync an object */
2027 rc = mdt_object_sync(info);
2029 const struct lu_fid *fid;
2030 struct lu_attr *la = &info->mti_attr.ma_attr;
2032 info->mti_attr.ma_need = MA_INODE;
2033 info->mti_attr.ma_valid = 0;
2034 rc = mdt_attr_get_complex(info, info->mti_object,
2037 body = req_capsule_server_get(pill,
2039 fid = mdt_object_fid(info->mti_object);
2040 mdt_pack_attr2body(info, body, la, fid);
2043 mdt_thread_info_fini(info);
2046 mdt_counter_incr(req, LPROC_MDT_SYNC);
2052 * Handle quota control requests to consult current usage/limit, but also
2053 * to configure quota enforcement
2055 static int mdt_quotactl(struct tgt_session_info *tsi)
2057 struct obd_export *exp = tsi->tsi_exp;
2058 struct req_capsule *pill = tsi->tsi_pill;
2059 struct obd_quotactl *oqctl, *repoqc;
2061 struct mdt_device *mdt = mdt_exp2dev(exp);
2062 struct lu_device *qmt = mdt->mdt_qmt_dev;
2063 struct lu_nodemap *nodemap;
2066 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2068 RETURN(err_serious(-EPROTO));
2070 rc = req_capsule_server_pack(pill);
2072 RETURN(err_serious(rc));
2074 nodemap = nodemap_get_from_exp(exp);
2075 if (IS_ERR(nodemap))
2076 RETURN(PTR_ERR(nodemap));
2078 switch (oqctl->qc_cmd) {
2079 /* master quotactl */
2082 if (!nodemap_can_setquota(nodemap))
2083 GOTO(out_nodemap, rc = -EPERM);
2087 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2088 /* slave quotactl */
2093 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2094 GOTO(out_nodemap, rc = -EFAULT);
2097 /* map uid/gid for remote client */
2099 if (exp_connect_rmtclient(exp)) {
2100 struct lustre_idmap_table *idmap;
2102 idmap = exp->exp_mdt_data.med_idmap;
2104 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2105 oqctl->qc_cmd != Q_GETINFO))
2106 GOTO(out_nodemap, rc = -EPERM);
2108 if (oqctl->qc_type == USRQUOTA)
2109 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2111 else if (oqctl->qc_type == GRPQUOTA)
2112 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2115 GOTO(out_nodemap, rc = -EINVAL);
2117 if (id == CFS_IDMAP_NOTFOUND) {
2118 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2119 GOTO(out_nodemap, rc = -EACCES);
2123 if (oqctl->qc_type == USRQUOTA)
2124 id = nodemap_map_id(nodemap, NODEMAP_UID,
2125 NODEMAP_CLIENT_TO_FS, id);
2126 else if (oqctl->qc_type == GRPQUOTA)
2127 id = nodemap_map_id(nodemap, NODEMAP_UID,
2128 NODEMAP_CLIENT_TO_FS, id);
2130 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2132 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2134 if (oqctl->qc_id != id)
2135 swap(oqctl->qc_id, id);
2137 switch (oqctl->qc_cmd) {
2143 /* forward quotactl request to QMT */
2144 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2149 /* slave quotactl */
2150 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2155 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2156 GOTO(out_nodemap, rc = -EFAULT);
2159 if (oqctl->qc_id != id)
2160 swap(oqctl->qc_id, id);
2167 nodemap_putref(nodemap);
2172 /** clone llog ctxt from child (mdd)
2173 * This allows remote llog (replicator) access.
2174 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2175 * context was originally set up, or we can handle them directly.
2176 * I choose the latter, but that means I need any llog
2177 * contexts set up by child to be accessable by the mdt. So we clone the
2178 * context into our context list here.
2180 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2183 struct md_device *next = mdt->mdt_child;
2184 struct llog_ctxt *ctxt;
2187 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2190 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2191 if (rc || ctxt == NULL) {
2195 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2197 CERROR("Can't set mdt ctxt %d\n", rc);
2202 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2203 struct mdt_device *mdt, int idx)
2205 struct llog_ctxt *ctxt;
2207 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2210 /* Put once for the get we just did, and once for the clone */
2211 llog_ctxt_put(ctxt);
2212 llog_ctxt_put(ctxt);
2217 * sec context handlers
2219 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2223 rc = mdt_handle_idmap(tsi);
2225 struct ptlrpc_request *req = tgt_ses_req(tsi);
2228 opc = lustre_msg_get_opc(req->rq_reqmsg);
2229 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2230 sptlrpc_svc_ctx_invalidate(req);
2233 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2239 * quota request handlers
2241 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2243 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2244 struct lu_device *qmt = mdt->mdt_qmt_dev;
2249 RETURN(err_serious(-EOPNOTSUPP));
2251 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2255 struct mdt_object *mdt_object_new(const struct lu_env *env,
2256 struct mdt_device *d,
2257 const struct lu_fid *f)
2259 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2260 struct lu_object *o;
2261 struct mdt_object *m;
2264 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2265 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2266 if (unlikely(IS_ERR(o)))
2267 m = (struct mdt_object *)o;
2273 struct mdt_object *mdt_object_find(const struct lu_env *env,
2274 struct mdt_device *d,
2275 const struct lu_fid *f)
2277 struct lu_object *o;
2278 struct mdt_object *m;
2281 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2282 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2283 if (unlikely(IS_ERR(o)))
2284 m = (struct mdt_object *)o;
2292 * Asyncronous commit for mdt device.
2294 * Pass asynchonous commit call down the MDS stack.
2296 * \param env environment
2297 * \param mdt the mdt device
2299 static void mdt_device_commit_async(const struct lu_env *env,
2300 struct mdt_device *mdt)
2302 struct dt_device *dt = mdt->mdt_bottom;
2306 rc = dt->dd_ops->dt_commit_async(env, dt);
2307 if (unlikely(rc != 0))
2308 CWARN("%s: async commit start failed: rc = %d\n",
2309 mdt_obd_name(mdt), rc);
2310 atomic_inc(&mdt->mdt_async_commit_count);
2315 * Mark the lock as "synchonous".
2317 * Mark the lock to deffer transaction commit to the unlock time.
2319 * \param lock the lock to mark as "synchonous"
2321 * \see mdt_is_lock_sync
2322 * \see mdt_save_lock
2324 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2326 lock->l_ast_data = (void*)1;
2330 * Check whehter the lock "synchonous" or not.
2332 * \param lock the lock to check
2333 * \retval 1 the lock is "synchonous"
2334 * \retval 0 the lock isn't "synchronous"
2336 * \see mdt_set_lock_sync
2337 * \see mdt_save_lock
2339 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2341 return lock->l_ast_data != NULL;
2345 * Blocking AST for mdt locks.
2347 * Starts transaction commit if in case of COS lock conflict or
2348 * deffers such a commit to the mdt_save_lock.
2350 * \param lock the lock which blocks a request or cancelling lock
2351 * \param desc unused
2352 * \param data unused
2353 * \param flag indicates whether this cancelling or blocking callback
2355 * \see ldlm_blocking_ast_nocheck
2357 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2358 void *data, int flag)
2360 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2361 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2365 if (flag == LDLM_CB_CANCELING)
2368 lock_res_and_lock(lock);
2369 if (lock->l_blocking_ast != mdt_blocking_ast) {
2370 unlock_res_and_lock(lock);
2373 if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2374 lock->l_blocking_lock != NULL) {
2375 if (mdt_cos_is_enabled(mdt) &&
2376 lock->l_client_cookie !=
2377 lock->l_blocking_lock->l_client_cookie)
2378 mdt_set_lock_sync(lock);
2379 else if (mdt_slc_is_enabled(mdt) &&
2380 ldlm_is_cos_incompat(lock->l_blocking_lock))
2381 mdt_set_lock_sync(lock);
2383 rc = ldlm_blocking_ast_nocheck(lock);
2385 /* There is no lock conflict if l_blocking_lock == NULL,
2386 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2387 * when the last reference to a local lock was released */
2388 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2391 rc = lu_env_init(&env, LCT_LOCAL);
2392 if (unlikely(rc != 0))
2393 CWARN("%s: lu_env initialization failed, cannot "
2394 "start asynchronous commit: rc = %d\n",
2397 mdt_device_commit_async(&env, mdt);
2404 * Blocking AST for cross-MDT lock
2406 * Discard lock from uncommitted_slc_locks and cancel it.
2408 * \param lock the lock which blocks a request or cancelling lock
2409 * \param desc unused
2410 * \param data unused
2411 * \param flag indicates whether this cancelling or blocking callback
2412 * \retval 0 on success
2413 * \retval negative number on error
2415 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2416 void *data, int flag)
2418 struct lustre_handle lockh;
2423 case LDLM_CB_BLOCKING:
2424 ldlm_lock2handle(lock, &lockh);
2425 rc = ldlm_cli_cancel(&lockh,
2426 ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2428 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2432 case LDLM_CB_CANCELING:
2433 LDLM_DEBUG(lock, "Revoke remote lock");
2434 /* discard slc lock here so that it can be cleaned anytime,
2435 * especially for cleanup_resource() */
2436 tgt_discard_slc_lock(lock);
2445 int mdt_check_resent_lock(struct mdt_thread_info *info,
2446 struct mdt_object *mo,
2447 struct mdt_lock_handle *lhc)
2449 /* the lock might already be gotten in ldlm_handle_enqueue() */
2450 if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2451 struct ptlrpc_request *req = mdt_info_req(info);
2452 struct ldlm_lock *lock;
2454 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2455 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2457 /* Lock is pinned by ldlm_handle_enqueue0() as it is
2458 * a resend case, however, it could be already destroyed
2459 * due to client eviction or a raced cancel RPC. */
2460 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64,
2461 lhc->mlh_reg_lh.cookie);
2465 if (!fid_res_name_eq(mdt_object_fid(mo),
2466 &lock->l_resource->lr_name)) {
2467 CWARN("%s: Although resent, but still not "
2468 "get child lock:"DFID"\n",
2469 info->mti_exp->exp_obd->obd_name,
2470 PFID(mdt_object_fid(mo)));
2471 LDLM_LOCK_PUT(lock);
2474 LDLM_LOCK_PUT(lock);
2480 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2481 const struct lu_fid *fid, struct lustre_handle *lh,
2482 enum ldlm_mode mode, __u64 ibits, bool nonblock)
2484 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2485 union ldlm_policy_data *policy = &mti->mti_policy;
2486 struct ldlm_res_id *res_id = &mti->mti_res_id;
2490 LASSERT(mdt_object_remote(o));
2492 fid_build_reg_res_name(fid, res_id);
2494 memset(einfo, 0, sizeof(*einfo));
2495 einfo->ei_type = LDLM_IBITS;
2496 einfo->ei_mode = mode;
2497 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2498 einfo->ei_cb_cp = ldlm_completion_ast;
2499 einfo->ei_enq_slave = 0;
2500 einfo->ei_res_id = res_id;
2502 einfo->ei_nonblock = 1;
2504 memset(policy, 0, sizeof(*policy));
2505 policy->l_inodebits.bits = ibits;
2507 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2512 static int mdt_object_local_lock(struct mdt_thread_info *info,
2513 struct mdt_object *o,
2514 struct mdt_lock_handle *lh, __u64 ibits,
2515 bool nonblock, bool cos_incompat)
2517 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2518 union ldlm_policy_data *policy = &info->mti_policy;
2519 struct ldlm_res_id *res_id = &info->mti_res_id;
2524 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2525 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2526 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2527 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2530 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2531 lh->mlh_reg_mode == LCK_EX);
2532 dlmflags |= LDLM_FL_COS_INCOMPAT;
2533 } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2534 dlmflags |= LDLM_FL_COS_ENABLED;
2537 /* Only enqueue LOOKUP lock for remote object */
2538 if (mdt_object_remote(o))
2539 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2541 if (lh->mlh_type == MDT_PDO_LOCK) {
2542 /* check for exists after object is locked */
2543 if (mdt_object_exists(o) == 0) {
2544 /* Non-existent object shouldn't have PDO lock */
2547 /* Non-dir object shouldn't have PDO lock */
2548 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2553 memset(policy, 0, sizeof(*policy));
2554 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2556 dlmflags |= LDLM_FL_ATOMIC_CB;
2558 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2561 * Take PDO lock on whole directory and build correct @res_id for lock
2562 * on part of directory.
2564 if (lh->mlh_pdo_hash != 0) {
2565 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2566 mdt_lock_pdo_mode(info, o, lh);
2567 if (lh->mlh_pdo_mode != LCK_NL) {
2569 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2570 * is never going to be sent to client and we do not
2571 * want it slowed down due to possible cancels.
2573 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2574 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2575 policy, res_id, dlmflags,
2576 info->mti_exp == NULL ? NULL :
2577 &info->mti_exp->exp_handle.h_cookie);
2578 if (unlikely(rc != 0))
2579 GOTO(out_unlock, rc);
2583 * Finish res_id initializing by name hash marking part of
2584 * directory which is taking modification.
2586 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2589 policy->l_inodebits.bits = ibits;
2592 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2593 * going to be sent to client. If it is - mdt_intent_policy() path will
2594 * fix it up and turn FL_LOCAL flag off.
2596 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2597 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2598 info->mti_exp == NULL ? NULL :
2599 &info->mti_exp->exp_handle.h_cookie);
2602 mdt_object_unlock(info, o, lh, 1);
2603 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2604 lh->mlh_pdo_hash != 0 &&
2605 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2606 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2612 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2613 struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2616 struct mdt_lock_handle *local_lh = NULL;
2620 if (!mdt_object_remote(o)) {
2621 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2626 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2627 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2628 MDS_INODELOCK_XATTR);
2630 /* Only enqueue LOOKUP lock for remote object */
2631 if (ibits & MDS_INODELOCK_LOOKUP) {
2632 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2633 nonblock, cos_incompat);
2640 if (ibits & MDS_INODELOCK_UPDATE) {
2641 /* Sigh, PDO needs to enqueue 2 locks right now, but
2642 * enqueue RPC can only request 1 lock, to avoid extra
2643 * RPC, so it will instead enqueue EX lock for remote
2644 * object anyway XXX*/
2645 if (lh->mlh_type == MDT_PDO_LOCK &&
2646 lh->mlh_pdo_hash != 0) {
2647 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2648 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2649 PFID(mdt_object_fid(o)));
2650 lh->mlh_pdo_hash = 0;
2651 lh->mlh_rreg_mode = LCK_EX;
2652 lh->mlh_type = MDT_REG_LOCK;
2654 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2657 MDS_INODELOCK_UPDATE, nonblock);
2658 if (rc != ELDLM_OK) {
2659 if (local_lh != NULL)
2660 mdt_object_unlock(info, o, local_lh, rc);
2668 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2669 struct mdt_lock_handle *lh, __u64 ibits)
2671 return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2674 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2675 struct mdt_lock_handle *lh, __u64 ibits,
2678 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2679 return mdt_object_lock_internal(info, o, lh, ibits, false,
2683 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2684 struct mdt_lock_handle *lh, __u64 ibits)
2686 struct mdt_lock_handle tmp = *lh;
2689 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2696 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2697 struct mdt_object *o, struct mdt_lock_handle *lh,
2698 __u64 ibits, bool cos_incompat)
2700 struct mdt_lock_handle tmp = *lh;
2703 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2704 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2712 * Save a lock within request object.
2714 * Keep the lock referenced until whether client ACK or transaction
2715 * commit happens or release the lock immediately depending on input
2716 * parameters. If COS is ON, a write lock is converted to COS lock
2719 * \param info thead info object
2720 * \param h lock handle
2721 * \param mode lock mode
2722 * \param decref force immediate lock releasing
2724 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2725 enum ldlm_mode mode, int decref)
2729 if (lustre_handle_is_used(h)) {
2730 if (decref || !info->mti_has_trans ||
2731 !(mode & (LCK_PW | LCK_EX))) {
2732 mdt_fid_unlock(h, mode);
2734 struct mdt_device *mdt = info->mti_mdt;
2735 struct ldlm_lock *lock = ldlm_handle2lock(h);
2736 struct ptlrpc_request *req = mdt_info_req(info);
2739 cos = (mdt_cos_is_enabled(mdt) ||
2740 mdt_slc_is_enabled(mdt));
2742 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2745 /* there is no request if mdt_object_unlock() is called
2746 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2747 if (likely(req != NULL)) {
2748 CDEBUG(D_HA, "request = %p reply state = %p"
2749 " transno = "LPD64"\n", req,
2750 req->rq_reply_state, req->rq_transno);
2752 ldlm_lock_downgrade(lock, LCK_COS);
2755 ptlrpc_save_lock(req, h, mode, cos);
2757 mdt_fid_unlock(h, mode);
2759 if (mdt_is_lock_sync(lock)) {
2760 CDEBUG(D_HA, "found sync-lock,"
2761 " async commit started\n");
2762 mdt_device_commit_async(info->mti_env,
2765 LDLM_LOCK_PUT(lock);
2774 * Save cross-MDT lock in uncommitted_slc_locks
2776 * Keep the lock referenced until transaction commit happens or release the lock
2777 * immediately depending on input parameters.
2779 * \param info thead info object
2780 * \param h lock handle
2781 * \param mode lock mode
2782 * \param decref force immediate lock releasing
2784 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2785 struct lustre_handle *h, enum ldlm_mode mode,
2790 if (lustre_handle_is_used(h)) {
2791 if (decref || !info->mti_has_trans ||
2792 !(mode & (LCK_PW | LCK_EX))) {
2793 ldlm_lock_decref_and_cancel(h, mode);
2795 struct ldlm_lock *lock = ldlm_handle2lock(h);
2796 struct ptlrpc_request *req = mdt_info_req(info);
2798 LASSERT(req != NULL);
2799 tgt_save_slc_lock(lock, req->rq_transno);
2800 ldlm_lock_decref(h, mode);
2809 * Unlock mdt object.
2811 * Immeditely release the regular lock and the PDO lock or save the
2812 * lock in request and keep them referenced until client ACK or
2813 * transaction commit.
2815 * \param info thread info object
2816 * \param o mdt object
2817 * \param lh mdt lock handle referencing regular and PDO locks
2818 * \param decref force immediate lock releasing
2820 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2821 struct mdt_lock_handle *lh, int decref)
2825 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2826 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2827 mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
2832 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2833 const struct lu_fid *f,
2834 struct mdt_lock_handle *lh,
2837 struct mdt_object *o;
2839 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2843 rc = mdt_object_lock(info, o, lh, ibits);
2845 mdt_object_put(info->mti_env, o);
2852 void mdt_object_unlock_put(struct mdt_thread_info * info,
2853 struct mdt_object * o,
2854 struct mdt_lock_handle *lh,
2857 mdt_object_unlock(info, o, lh, decref);
2858 mdt_object_put(info->mti_env, o);
2862 * Generic code handling requests that have struct mdt_body passed in:
2864 * - extract mdt_body from request and save it in @info, if present;
2866 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2869 * - if HABEO_CORPUS flag is set for this request type check whether object
2870 * actually exists on storage (lu_object_exists()).
2873 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2875 const struct mdt_body *body;
2876 struct mdt_object *obj;
2877 const struct lu_env *env;
2878 struct req_capsule *pill;
2882 env = info->mti_env;
2883 pill = info->mti_pill;
2885 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2889 if (!(body->mbo_valid & OBD_MD_FLID))
2892 if (!fid_is_sane(&body->mbo_fid1)) {
2893 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2897 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2899 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2900 mdt_object_put(env, obj);
2903 info->mti_object = obj;
2912 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2914 struct req_capsule *pill = info->mti_pill;
2918 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2919 rc = mdt_body_unpack(info, flags);
2923 if (rc == 0 && (flags & HABEO_REFERO)) {
2925 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2926 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2928 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2929 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2932 rc = req_capsule_server_pack(pill);
2937 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2939 lh->mlh_type = MDT_NUL_LOCK;
2940 lh->mlh_reg_lh.cookie = 0ull;
2941 lh->mlh_reg_mode = LCK_MINMODE;
2942 lh->mlh_pdo_lh.cookie = 0ull;
2943 lh->mlh_pdo_mode = LCK_MINMODE;
2944 lh->mlh_rreg_lh.cookie = 0ull;
2945 lh->mlh_rreg_mode = LCK_MINMODE;
2948 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2950 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2951 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2955 * Initialize fields of struct mdt_thread_info. Other fields are left in
2956 * uninitialized state, because it's too expensive to zero out whole
2957 * mdt_thread_info (> 1K) on each request arrival.
2959 void mdt_thread_info_init(struct ptlrpc_request *req,
2960 struct mdt_thread_info *info)
2964 info->mti_pill = &req->rq_pill;
2967 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2968 mdt_lock_handle_init(&info->mti_lh[i]);
2970 /* mdt device: it can be NULL while CONNECT */
2971 if (req->rq_export) {
2972 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2973 info->mti_exp = req->rq_export;
2975 info->mti_mdt = NULL;
2976 info->mti_env = req->rq_svc_thread->t_env;
2977 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2979 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2980 info->mti_big_buf = LU_BUF_NULL;
2981 info->mti_body = NULL;
2982 info->mti_object = NULL;
2983 info->mti_dlm_req = NULL;
2984 info->mti_has_trans = 0;
2985 info->mti_cross_ref = 0;
2986 info->mti_opdata = 0;
2987 info->mti_big_lmm_used = 0;
2989 info->mti_spec.no_create = 0;
2990 info->mti_spec.sp_rm_entry = 0;
2991 info->mti_spec.sp_permitted = 0;
2992 info->mti_spec.sp_migrate_close = 0;
2994 info->mti_spec.u.sp_ea.eadata = NULL;
2995 info->mti_spec.u.sp_ea.eadatalen = 0;
2998 void mdt_thread_info_fini(struct mdt_thread_info *info)
3002 if (info->mti_object != NULL) {
3003 mdt_object_put(info->mti_env, info->mti_object);
3004 info->mti_object = NULL;
3007 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3008 mdt_lock_handle_fini(&info->mti_lh[i]);
3009 info->mti_env = NULL;
3010 info->mti_pill = NULL;
3011 info->mti_exp = NULL;
3013 if (unlikely(info->mti_big_buf.lb_buf != NULL))
3014 lu_buf_free(&info->mti_big_buf);
3017 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3019 struct mdt_thread_info *mti;
3021 mti = mdt_th_info(tsi->tsi_env);
3022 LASSERT(mti != NULL);
3024 mdt_thread_info_init(tgt_ses_req(tsi), mti);
3025 if (tsi->tsi_corpus != NULL) {
3026 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3027 lu_object_get(tsi->tsi_corpus);
3029 mti->mti_body = tsi->tsi_mdt_body;
3030 mti->mti_dlm_req = tsi->tsi_dlm_req;
3035 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3037 struct ptlrpc_request *req = tgt_ses_req(tsi);
3042 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3044 tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3045 set_current_state(TASK_UNINTERRUPTIBLE);
3046 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3049 rc = tgt_connect(tsi);
3053 rc = mdt_init_idmap(tsi);
3058 obd_disconnect(class_export_get(req->rq_export));
3077 static int mdt_intent_getattr(enum mdt_it_code opcode,
3078 struct mdt_thread_info *info,
3079 struct ldlm_lock **,
3082 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3083 struct mdt_thread_info *info,
3084 struct ldlm_lock **lockp,
3087 static int mdt_intent_layout(enum mdt_it_code opcode,
3088 struct mdt_thread_info *info,
3089 struct ldlm_lock **,
3091 static int mdt_intent_reint(enum mdt_it_code opcode,
3092 struct mdt_thread_info *info,
3093 struct ldlm_lock **,
3096 static struct mdt_it_flavor {
3097 const struct req_format *it_fmt;
3099 int (*it_act)(enum mdt_it_code ,
3100 struct mdt_thread_info *,
3101 struct ldlm_lock **,
3104 } mdt_it_flavor[] = {
3106 .it_fmt = &RQF_LDLM_INTENT,
3107 /*.it_flags = HABEO_REFERO,*/
3109 .it_act = mdt_intent_reint,
3110 .it_reint = REINT_OPEN
3113 .it_fmt = &RQF_LDLM_INTENT,
3115 * OCREAT is not a MUTABOR request as if the file
3117 * We do the extra check of OBD_CONNECT_RDONLY in
3118 * mdt_reint_open() when we really need to create
3122 .it_act = mdt_intent_reint,
3123 .it_reint = REINT_OPEN
3126 .it_fmt = &RQF_LDLM_INTENT,
3127 .it_flags = MUTABOR,
3128 .it_act = mdt_intent_reint,
3129 .it_reint = REINT_CREATE
3131 [MDT_IT_GETATTR] = {
3132 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3133 .it_flags = HABEO_REFERO,
3134 .it_act = mdt_intent_getattr
3136 [MDT_IT_READDIR] = {
3142 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3143 .it_flags = HABEO_REFERO,
3144 .it_act = mdt_intent_getattr
3147 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3148 .it_flags = MUTABOR,
3150 .it_reint = REINT_UNLINK
3154 .it_flags = MUTABOR,
3157 [MDT_IT_GETXATTR] = {
3158 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
3159 .it_flags = HABEO_CORPUS,
3160 .it_act = mdt_intent_getxattr
3163 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3165 .it_act = mdt_intent_layout
3170 mdt_intent_lock_replace(struct mdt_thread_info *info,
3171 struct ldlm_lock **lockp,
3172 struct mdt_lock_handle *lh,
3173 __u64 flags, int result)
3175 struct ptlrpc_request *req = mdt_info_req(info);
3176 struct ldlm_lock *lock = *lockp;
3177 struct ldlm_lock *new_lock;
3179 /* If possible resent found a lock, @lh is set to its handle */
3180 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3182 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3183 lh->mlh_reg_lh.cookie = 0;
3187 if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3188 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3189 * a resend case, however, it could be already destroyed
3190 * due to client eviction or a raced cancel RPC. */
3191 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
3192 lh->mlh_reg_lh.cookie);
3193 lh->mlh_reg_lh.cookie = 0;
3197 LASSERTF(new_lock != NULL,
3198 "lockh "LPX64" flags "LPX64" rc %d\n",
3199 lh->mlh_reg_lh.cookie, flags, result);
3202 * If we've already given this lock to a client once, then we should
3203 * have no readers or writers. Otherwise, we should have one reader
3204 * _or_ writer ref (which will be zeroed below) before returning the
3207 if (new_lock->l_export == req->rq_export) {
3208 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3210 LASSERT(new_lock->l_export == NULL);
3211 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3216 if (new_lock->l_export == req->rq_export) {
3218 * Already gave this to the client, which means that we
3219 * reconstructed a reply.
3221 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3224 LDLM_LOCK_RELEASE(new_lock);
3225 lh->mlh_reg_lh.cookie = 0;
3226 RETURN(ELDLM_LOCK_REPLACED);
3230 * Fixup the lock to be given to the client.
3232 lock_res_and_lock(new_lock);
3233 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3234 * possible blocking AST. */
3235 while (new_lock->l_readers > 0) {
3236 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3237 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3238 new_lock->l_readers--;
3240 while (new_lock->l_writers > 0) {
3241 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3242 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3243 new_lock->l_writers--;
3246 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3247 new_lock->l_blocking_ast = lock->l_blocking_ast;
3248 new_lock->l_completion_ast = lock->l_completion_ast;
3249 new_lock->l_remote_handle = lock->l_remote_handle;
3250 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3252 unlock_res_and_lock(new_lock);
3254 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3255 &new_lock->l_remote_handle,
3256 &new_lock->l_exp_hash);
3258 LDLM_LOCK_RELEASE(new_lock);
3259 lh->mlh_reg_lh.cookie = 0;
3261 RETURN(ELDLM_LOCK_REPLACED);
3264 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3265 struct ldlm_lock *new_lock,
3266 struct mdt_lock_handle *lh,
3269 struct ptlrpc_request *req = mdt_info_req(info);
3270 struct ldlm_request *dlmreq;
3272 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3275 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3277 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3278 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3280 if (flags & LDLM_FL_RESENT) {
3281 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3282 lh->mlh_reg_mode = new_lock->l_granted_mode;
3284 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3285 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3286 lh->mlh_reg_lh.cookie);
3291 * If the xid matches, then we know this is a resent request, and allow
3292 * it. (It's probably an OPEN, for which we don't send a lock.
3294 if (req_can_reconstruct(req, NULL))
3298 * This remote handle isn't enqueued, so we never received or processed
3299 * this request. Clear MSG_RESENT, because it can be handled like any
3300 * normal request now.
3302 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3304 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3305 dlmreq->lock_handle[0].cookie);
3308 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3309 struct mdt_thread_info *info,
3310 struct ldlm_lock **lockp,
3313 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3314 struct ldlm_reply *ldlm_rep = NULL;
3318 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3319 * (for the resend case) or a new lock. Below we will use it to
3320 * replace the original lock.
3322 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3323 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3324 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3325 rc = mdt_object_lock(info, info->mti_object, lhc,
3326 MDS_INODELOCK_XATTR);
3331 grc = mdt_getxattr(info);
3333 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, 0);
3335 if (mdt_info_req(info)->rq_repmsg != NULL)
3336 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3337 if (ldlm_rep == NULL)
3338 RETURN(err_serious(-EFAULT));
3340 ldlm_rep->lock_policy_res2 = grc;
3345 static int mdt_intent_getattr(enum mdt_it_code opcode,
3346 struct mdt_thread_info *info,
3347 struct ldlm_lock **lockp,
3350 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3352 struct ldlm_reply *ldlm_rep;
3353 struct mdt_body *reqbody;
3354 struct mdt_body *repbody;
3358 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3361 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3364 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3365 repbody->mbo_eadatasize = 0;
3366 repbody->mbo_aclsize = 0;
3370 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3372 case MDT_IT_GETATTR:
3373 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3377 CERROR("Unsupported intent (%d)\n", opcode);
3378 GOTO(out_shrink, rc = -EINVAL);
3381 rc = mdt_init_ucred_intent_getattr(info, reqbody);
3383 GOTO(out_shrink, rc);
3385 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3386 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3388 /* Get lock from request for possible resent case. */
3389 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3391 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3392 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3394 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3395 ldlm_rep->lock_policy_res2 = 0;
3396 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3397 ldlm_rep->lock_policy_res2) {
3398 lhc->mlh_reg_lh.cookie = 0ull;
3399 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3402 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3405 mdt_exit_ucred(info);
3407 mdt_client_compatibility(info);
3408 rc2 = mdt_fix_reply(info);
3414 static int mdt_intent_layout(enum mdt_it_code opcode,
3415 struct mdt_thread_info *info,
3416 struct ldlm_lock **lockp,
3419 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3420 struct layout_intent *layout;
3422 struct mdt_object *obj = NULL;
3423 int layout_size = 0;
3427 if (opcode != MDT_IT_LAYOUT) {
3428 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3433 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3437 if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3438 CERROR("%s: Unsupported layout intent opc %d\n",
3439 mdt_obd_name(info->mti_mdt), layout->li_opc);
3443 fid = &info->mti_tmp_fid2;
3444 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3446 /* Get lock from request for possible resent case. */
3447 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3449 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3451 GOTO(out, rc = PTR_ERR(obj));
3453 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3454 layout_size = mdt_attr_get_eabuf_size(info, obj);
3455 if (layout_size < 0)
3456 GOTO(out_obj, rc = layout_size);
3458 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3459 info->mti_mdt->mdt_max_mdsize = layout_size;
3462 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3463 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3465 rc = req_capsule_server_pack(info->mti_pill);
3469 mdt_object_put(info->mti_env, obj);
3471 if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3472 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3475 lhc->mlh_reg_lh.cookie = 0;
3480 static int mdt_intent_reint(enum mdt_it_code opcode,
3481 struct mdt_thread_info *info,
3482 struct ldlm_lock **lockp,
3485 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3486 struct ldlm_reply *rep = NULL;
3490 static const struct req_format *intent_fmts[REINT_MAX] = {
3491 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3492 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3497 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3501 if (mdt_it_flavor[opcode].it_reint != opc) {
3502 CERROR("Reint code %ld doesn't match intent: %d\n",
3504 RETURN(err_serious(-EPROTO));
3507 /* Get lock from request for possible resent case. */
3508 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3510 rc = mdt_reint_internal(info, lhc, opc);
3512 /* Check whether the reply has been packed successfully. */
3513 if (mdt_info_req(info)->rq_repmsg != NULL)
3514 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3516 RETURN(err_serious(-EFAULT));
3518 /* MDC expects this in any case */
3520 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3522 /* the open lock or the lock for cross-ref object should be
3523 * returned to the client */
3524 if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3525 (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3526 rep->lock_policy_res2 = 0;
3527 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3531 rep->lock_policy_res2 = clear_serious(rc);
3533 if (rep->lock_policy_res2 == -ENOENT &&
3534 mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3535 !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3536 rep->lock_policy_res2 = 0;
3538 lhc->mlh_reg_lh.cookie = 0ull;
3539 if (rc == -ENOTCONN || rc == -ENODEV ||
3540 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3542 * If it is the disconnect error (ENODEV & ENOCONN), the error
3543 * will be returned by rq_status, and client at ptlrpc layer
3544 * will detect this, then disconnect, reconnect the import
3545 * immediately, instead of impacting the following the rpc.
3550 * For other cases, the error will be returned by intent, and client
3551 * will retrieve the result from intent.
3553 RETURN(ELDLM_LOCK_ABORTED);
3556 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3564 case IT_OPEN|IT_CREAT:
3571 rc = MDT_IT_READDIR;
3574 rc = MDT_IT_GETATTR;
3586 rc = MDT_IT_GETXATTR;
3591 case IT_QUOTA_DQACQ:
3596 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3603 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3604 struct mdt_thread_info *info,
3605 struct ldlm_lock **lockp, __u64 flags)
3607 struct req_capsule *pill = info->mti_pill;
3608 struct ptlrpc_request *req = mdt_info_req(info);
3609 struct mdt_it_flavor *flv;
3614 opc = mdt_intent_code(itopc);
3618 if (opc == MDT_IT_QUOTA) {
3619 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3622 RETURN(-EOPNOTSUPP);
3624 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3625 /* pass the request to quota master */
3626 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3627 mdt_info_req(info), lockp,
3632 flv = &mdt_it_flavor[opc];
3633 if (flv->it_fmt != NULL)
3634 req_capsule_extend(pill, flv->it_fmt);
3636 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3640 if (flv->it_flags & MUTABOR &&
3641 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3644 if (flv->it_act != NULL) {
3645 struct ldlm_reply *rep;
3647 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3649 /* execute policy */
3650 rc = flv->it_act(opc, info, lockp, flags);
3652 /* Check whether the reply has been packed successfully. */
3653 if (req->rq_repmsg != NULL) {
3654 rep = req_capsule_server_get(info->mti_pill,
3656 rep->lock_policy_res2 =
3657 ptlrpc_status_hton(rep->lock_policy_res2);
3664 static int mdt_intent_policy(struct ldlm_namespace *ns,
3665 struct ldlm_lock **lockp, void *req_cookie,
3666 enum ldlm_mode mode, __u64 flags, void *data)
3668 struct tgt_session_info *tsi;
3669 struct mdt_thread_info *info;
3670 struct ptlrpc_request *req = req_cookie;
3671 struct ldlm_intent *it;
3672 struct req_capsule *pill;
3677 LASSERT(req != NULL);
3679 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3681 info = tsi2mdt_info(tsi);
3682 LASSERT(info != NULL);
3683 pill = info->mti_pill;
3684 LASSERT(pill->rc_req == req);
3686 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3687 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3688 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3690 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3694 /* Lock without inodebits makes no sense and will oops
3695 * later in ldlm. Let's check it now to see if we have
3696 * ibits corrupted somewhere in mdt_intent_opc().
3697 * The case for client miss to set ibits has been
3698 * processed by others. */
3699 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3700 lr_type == LDLM_IBITS,
3701 info->mti_dlm_req->lock_desc.\
3702 l_policy_data.l_inodebits.bits != 0));
3704 rc = err_serious(-EFAULT);
3706 /* No intent was provided */
3707 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3708 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3709 rc = req_capsule_server_pack(pill);
3711 rc = err_serious(rc);
3713 mdt_thread_info_fini(info);
3717 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3719 struct seq_server_site *ss = mdt_seq_site(mdt);
3721 if (ss->ss_node_id == 0)
3724 if (ss->ss_client_seq != NULL) {
3725 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3726 ss->ss_client_seq->lcs_exp = NULL;
3729 if (ss->ss_server_fld != NULL) {
3730 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3731 ss->ss_server_fld->lsf_control_exp = NULL;
3735 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3737 struct seq_server_site *ss = mdt_seq_site(mdt);
3742 if (ss->ss_server_seq != NULL)
3743 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3746 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3748 mdt_seq_fini_cli(mdt);
3749 mdt_deregister_seq_exp(mdt);
3751 return seq_site_fini(env, mdt_seq_site(mdt));
3755 * It will retrieve its FLDB entries from MDT0, and it only happens
3756 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3757 * and it needs to refresh FLDB from the MDT0.
3759 static int mdt_register_lwp_callback(void *data)
3762 struct mdt_device *mdt = data;
3763 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3767 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3769 rc = lu_env_init(&env, LCT_MD_THREAD);
3771 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3775 /* Allocate new sequence now to avoid creating local transaction
3776 * in the normal transaction process */
3777 rc = seq_server_check_and_alloc_super(&env,
3778 mdt_seq_site(mdt)->ss_server_seq);
3783 rc = fld_update_from_controller(&env, fld);
3785 CERROR("%s: cannot update controller: rc = %d\n",
3786 mdt_obd_name(mdt), rc);
3795 static int mdt_register_seq_exp(struct mdt_device *mdt)
3797 struct seq_server_site *ss = mdt_seq_site(mdt);
3798 char *lwp_name = NULL;
3801 if (ss->ss_node_id == 0)
3804 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3805 if (lwp_name == NULL)
3806 GOTO(out_free, rc = -ENOMEM);
3808 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3812 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3817 rc = lustre_register_lwp_item(lwp_name,
3818 &ss->ss_server_fld->lsf_control_exp,
3819 mdt_register_lwp_callback, mdt);
3821 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3822 ss->ss_client_seq->lcs_exp = NULL;
3826 if (lwp_name != NULL)
3827 OBD_FREE(lwp_name, MAX_OBD_NAME);
3833 * Init client sequence manager which is used by local MDS to talk to sequence
3834 * controller on remote node.
3836 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3838 struct seq_server_site *ss = mdt_seq_site(mdt);
3843 /* check if this is adding the first MDC and controller is not yet
3845 OBD_ALLOC_PTR(ss->ss_client_seq);
3846 if (ss->ss_client_seq == NULL)
3849 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3850 if (prefix == NULL) {
3851 OBD_FREE_PTR(ss->ss_client_seq);
3852 ss->ss_client_seq = NULL;
3856 /* Note: seq_client_fini will be called in seq_site_fini */
3857 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3858 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3859 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3861 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3863 OBD_FREE_PTR(ss->ss_client_seq);
3864 ss->ss_client_seq = NULL;
3868 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3873 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3875 struct seq_server_site *ss;
3879 ss = mdt_seq_site(mdt);
3880 /* init sequence controller server(MDT0) */
3881 if (ss->ss_node_id == 0) {
3882 OBD_ALLOC_PTR(ss->ss_control_seq);
3883 if (ss->ss_control_seq == NULL)
3886 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3887 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3890 GOTO(out_seq_fini, rc);
3893 /* Init normal sequence server */
3894 OBD_ALLOC_PTR(ss->ss_server_seq);
3895 if (ss->ss_server_seq == NULL)
3896 GOTO(out_seq_fini, rc = -ENOMEM);
3898 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3899 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3901 GOTO(out_seq_fini, rc);
3903 /* init seq client for seq server to talk to seq controller(MDT0) */
3904 rc = mdt_seq_init_cli(env, mdt);
3906 GOTO(out_seq_fini, rc);
3908 if (ss->ss_node_id != 0)
3909 /* register controller export through lwp */
3910 rc = mdt_register_seq_exp(mdt);
3915 mdt_seq_fini(env, mdt);
3923 static int mdt_fld_fini(const struct lu_env *env,
3924 struct mdt_device *m)
3926 struct seq_server_site *ss = mdt_seq_site(m);
3929 if (ss && ss->ss_server_fld) {
3930 fld_server_fini(env, ss->ss_server_fld);
3931 OBD_FREE_PTR(ss->ss_server_fld);
3932 ss->ss_server_fld = NULL;
3938 static int mdt_fld_init(const struct lu_env *env,
3940 struct mdt_device *m)
3942 struct seq_server_site *ss;
3946 ss = mdt_seq_site(m);
3948 OBD_ALLOC_PTR(ss->ss_server_fld);
3949 if (ss->ss_server_fld == NULL)
3950 RETURN(rc = -ENOMEM);
3952 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3955 OBD_FREE_PTR(ss->ss_server_fld);
3956 ss->ss_server_fld = NULL;
3963 static void mdt_stack_pre_fini(const struct lu_env *env,
3964 struct mdt_device *m, struct lu_device *top)
3966 struct lustre_cfg_bufs *bufs;
3967 struct lustre_cfg *lcfg;
3968 struct mdt_thread_info *info;
3973 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3974 LASSERT(info != NULL);
3976 bufs = &info->mti_u.bufs;
3978 LASSERT(m->mdt_child_exp);
3979 LASSERT(m->mdt_child_exp->exp_obd);
3981 /* process cleanup, pass mdt obd name to get obd umount flags */
3982 /* XXX: this is needed because all layers are referenced by
3983 * objects (some of them are pinned by osd, for example *
3984 * the proper solution should be a model where object used
3985 * by osd only doesn't have mdt/mdd slices -bzzz */
3986 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3987 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3988 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3992 top->ld_ops->ldo_process_config(env, top, lcfg);
3993 lustre_cfg_free(lcfg);
3997 static void mdt_stack_fini(const struct lu_env *env,
3998 struct mdt_device *m, struct lu_device *top)
4000 struct obd_device *obd = mdt2obd_dev(m);
4001 struct lustre_cfg_bufs *bufs;
4002 struct lustre_cfg *lcfg;
4003 struct mdt_thread_info *info;
4007 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4008 LASSERT(info != NULL);
4010 lu_dev_del_linkage(top->ld_site, top);
4012 lu_site_purge(env, top->ld_site, -1);
4014 bufs = &info->mti_u.bufs;
4015 /* process cleanup, pass mdt obd name to get obd umount flags */
4016 /* another purpose is to let all layers to release their objects */
4017 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4022 lustre_cfg_bufs_set_string(bufs, 1, flags);
4023 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4028 top->ld_ops->ldo_process_config(env, top, lcfg);
4029 lustre_cfg_free(lcfg);
4031 lu_site_purge(env, top->ld_site, -1);
4033 m->mdt_child = NULL;
4034 m->mdt_bottom = NULL;
4036 obd_disconnect(m->mdt_child_exp);
4037 m->mdt_child_exp = NULL;
4039 obd_disconnect(m->mdt_bottom_exp);
4040 m->mdt_child_exp = NULL;
4043 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4044 const char *next, struct obd_export **exp)
4046 struct obd_connect_data *data = NULL;
4047 struct obd_device *obd;
4051 OBD_ALLOC_PTR(data);
4053 GOTO(out, rc = -ENOMEM);
4055 obd = class_name2obd(next);
4057 CERROR("%s: can't locate next device: %s\n",
4058 mdt_obd_name(m), next);
4059 GOTO(out, rc = -ENOTCONN);
4062 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4063 data->ocd_version = LUSTRE_VERSION_CODE;
4065 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4067 CERROR("%s: cannot connect to next dev %s (%d)\n",
4068 mdt_obd_name(m), next, rc);
4078 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4079 struct lustre_cfg *cfg)
4081 char *dev = lustre_cfg_string(cfg, 0);
4082 int rc, name_size, uuid_size;
4083 char *name, *uuid, *p;
4084 struct lustre_cfg_bufs *bufs;
4085 struct lustre_cfg *lcfg;
4086 struct obd_device *obd;
4087 struct lustre_profile *lprof;
4088 struct lu_site *site;
4091 /* in 1.8 we had the only device in the stack - MDS.
4092 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4093 * in 2.3 OSD is instantiated by obd_mount.c, so we need
4094 * to generate names and setup MDT, MDD. MDT will be using
4095 * generated name to connect to MDD. for MDD the next device
4096 * will be LOD with name taken from so called "profile" which
4097 * is generated by mount_option line
4099 * 1.8 MGS generates config. commands like this:
4100 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
4101 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
4102 * 2.0 MGS generates config. commands like this:
4103 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
4104 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4105 * 3:lustre-MDT0000-mdtlov 4:f
4107 * we generate MDD name from MDT one, just replacing T with D
4109 * after all the preparations, the logical equivalent will be
4110 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
4111 * 3:lustre-MDT0000-mdtlov 4:f
4112 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4113 * 3:lustre-MDD0000 4:f
4115 * notice we build the stack from down to top: MDD first, then MDT */
4117 name_size = MAX_OBD_NAME;
4118 uuid_size = MAX_OBD_NAME;
4120 OBD_ALLOC(name, name_size);
4121 OBD_ALLOC(uuid, uuid_size);
4122 if (name == NULL || uuid == NULL)
4123 GOTO(cleanup_mem, rc = -ENOMEM);
4125 OBD_ALLOC_PTR(bufs);
4127 GOTO(cleanup_mem, rc = -ENOMEM);
4130 p = strstr(name, "-MDT");
4132 GOTO(free_bufs, rc = -ENOMEM);
4135 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4137 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4138 if (lprof == NULL || lprof->lp_dt == NULL) {
4139 CERROR("can't find the profile: %s\n",
4140 lustre_cfg_string(cfg, 0));
4141 GOTO(free_bufs, rc = -EINVAL);
4144 lustre_cfg_bufs_reset(bufs, name);
4145 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4146 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4147 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4149 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4151 GOTO(put_profile, rc = -ENOMEM);
4153 rc = class_attach(lcfg);
4155 GOTO(lcfg_cleanup, rc);
4157 obd = class_name2obd(name);
4159 CERROR("Can not find obd %s (%s in config)\n",
4160 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4161 GOTO(lcfg_cleanup, rc = -EINVAL);
4164 lustre_cfg_free(lcfg);
4166 lustre_cfg_bufs_reset(bufs, name);
4167 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4168 lustre_cfg_bufs_set_string(bufs, 2, dev);
4169 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4171 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4173 GOTO(class_detach, rc = -ENOMEM);
4175 rc = class_setup(obd, lcfg);
4177 GOTO(class_detach, rc);
4179 /* connect to MDD we just setup */
4180 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4182 GOTO(class_detach, rc);
4184 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4186 LASSERT(mdt_lu_site(mdt) == NULL);
4187 mdt->mdt_lu_dev.ld_site = site;
4188 site->ls_top_dev = &mdt->mdt_lu_dev;
4189 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4191 /* now connect to bottom OSD */
4192 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4193 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4195 GOTO(class_detach, rc);
4197 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4199 rc = lu_env_refill((struct lu_env *)env);
4201 CERROR("Failure to refill session: '%d'\n", rc);
4203 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4208 class_detach(obd, lcfg);
4210 lustre_cfg_free(lcfg);
4212 class_put_profile(lprof);
4217 OBD_FREE(name, name_size);
4219 OBD_FREE(uuid, uuid_size);
4223 /* setup quota master target on MDT0 */
4224 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4225 struct lustre_cfg *cfg)
4227 struct obd_device *obd;
4228 char *dev = lustre_cfg_string(cfg, 0);
4229 char *qmtname, *uuid, *p;
4230 struct lustre_cfg_bufs *bufs;
4231 struct lustre_cfg *lcfg;
4232 struct lustre_profile *lprof;
4233 struct obd_connect_data *data;
4237 LASSERT(mdt->mdt_qmt_exp == NULL);
4238 LASSERT(mdt->mdt_qmt_dev == NULL);
4240 /* quota master is on MDT0 only for now */
4241 if (mdt->mdt_seq_site.ss_node_id != 0)
4244 /* MGS generates config commands which look as follows:
4245 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4246 * 3:lustre-MDT0000-mdtlov 4:f
4248 * We generate the QMT name from the MDT one, just replacing MD with QM
4249 * after all the preparations, the logical equivalent will be:
4250 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4251 * 3:lustre-MDT0000-osd 4:f */
4252 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4253 OBD_ALLOC(uuid, UUID_MAX);
4254 OBD_ALLOC_PTR(bufs);
4255 OBD_ALLOC_PTR(data);
4256 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4257 GOTO(cleanup_mem, rc = -ENOMEM);
4259 strcpy(qmtname, dev);
4260 p = strstr(qmtname, "-MDT");
4262 GOTO(cleanup_mem, rc = -ENOMEM);
4263 /* replace MD with QM */
4267 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4269 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4270 if (lprof == NULL || lprof->lp_dt == NULL) {
4271 CERROR("can't find profile for %s\n",
4272 lustre_cfg_string(cfg, 0));
4273 GOTO(cleanup_mem, rc = -EINVAL);
4276 lustre_cfg_bufs_reset(bufs, qmtname);
4277 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4278 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4279 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4281 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4283 GOTO(put_profile, rc = -ENOMEM);
4285 rc = class_attach(lcfg);
4287 GOTO(lcfg_cleanup, rc);
4289 obd = class_name2obd(qmtname);
4291 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4292 lustre_cfg_string(cfg, 0));
4293 GOTO(lcfg_cleanup, rc = -EINVAL);
4296 lustre_cfg_free(lcfg);
4298 lustre_cfg_bufs_reset(bufs, qmtname);
4299 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4300 lustre_cfg_bufs_set_string(bufs, 2, dev);
4302 /* for quota, the next device should be the OSD device */
4303 lustre_cfg_bufs_set_string(bufs, 3,
4304 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4306 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4308 GOTO(class_detach, rc = -ENOMEM);
4310 rc = class_setup(obd, lcfg);
4312 GOTO(class_detach, rc);
4314 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4316 /* configure local quota objects */
4317 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4321 GOTO(class_cleanup, rc);
4323 /* connect to quota master target */
4324 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4325 data->ocd_version = LUSTRE_VERSION_CODE;
4326 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4329 CERROR("cannot connect to quota master device %s (%d)\n",
4331 GOTO(class_cleanup, rc);
4337 class_manual_cleanup(obd);
4338 mdt->mdt_qmt_dev = NULL;
4342 class_detach(obd, lcfg);
4344 lustre_cfg_free(lcfg);
4346 class_put_profile(lprof);
4351 OBD_FREE(qmtname, MAX_OBD_NAME);
4353 OBD_FREE(uuid, UUID_MAX);
4359 /* Shutdown quota master target associated with mdt */
4360 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4364 if (mdt->mdt_qmt_exp == NULL)
4366 LASSERT(mdt->mdt_qmt_dev != NULL);
4368 /* the qmt automatically shuts down when the mdt disconnects */
4369 obd_disconnect(mdt->mdt_qmt_exp);
4370 mdt->mdt_qmt_exp = NULL;
4371 mdt->mdt_qmt_dev = NULL;
4375 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4376 * for now. This will be removed along with converting rest of MDT code
4377 * to use tgt_session_info */
4378 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4380 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4383 rc = mdt_getxattr(info);
4385 mdt_thread_info_fini(info);
4389 static struct tgt_handler mdt_tgt_handlers[] = {
4390 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4391 0, MDS_CONNECT, mdt_tgt_connect,
4392 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4393 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4394 0, MDS_DISCONNECT, tgt_disconnect,
4395 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4396 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4397 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4398 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4399 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4400 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GET_ROOT, mdt_get_root),
4401 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4402 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4404 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4405 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4406 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4407 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4408 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4409 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4410 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4411 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4413 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4414 mdt_hsm_ct_register),
4415 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4416 mdt_hsm_ct_unregister),
4417 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4419 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4421 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4422 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4424 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4429 static struct tgt_handler mdt_sec_ctx_ops[] = {
4430 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4431 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4432 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4435 static struct tgt_handler mdt_quota_ops[] = {
4436 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4439 static struct tgt_opc_slice mdt_common_slice[] = {
4441 .tos_opc_start = MDS_FIRST_OPC,
4442 .tos_opc_end = MDS_LAST_OPC,
4443 .tos_hs = mdt_tgt_handlers
4446 .tos_opc_start = OBD_FIRST_OPC,
4447 .tos_opc_end = OBD_LAST_OPC,
4448 .tos_hs = tgt_obd_handlers
4451 .tos_opc_start = LDLM_FIRST_OPC,
4452 .tos_opc_end = LDLM_LAST_OPC,
4453 .tos_hs = tgt_dlm_handlers
4456 .tos_opc_start = SEC_FIRST_OPC,
4457 .tos_opc_end = SEC_LAST_OPC,
4458 .tos_hs = mdt_sec_ctx_ops
4461 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4462 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4463 .tos_hs = tgt_out_handlers
4466 .tos_opc_start = FLD_FIRST_OPC,
4467 .tos_opc_end = FLD_LAST_OPC,
4468 .tos_hs = fld_handlers
4471 .tos_opc_start = SEQ_FIRST_OPC,
4472 .tos_opc_end = SEQ_LAST_OPC,
4473 .tos_hs = seq_handlers
4476 .tos_opc_start = QUOTA_DQACQ,
4477 .tos_opc_end = QUOTA_LAST_OPC,
4478 .tos_hs = mdt_quota_ops
4481 .tos_opc_start = LLOG_FIRST_OPC,
4482 .tos_opc_end = LLOG_LAST_OPC,
4483 .tos_hs = tgt_llog_handlers
4486 .tos_opc_start = LFSCK_FIRST_OPC,
4487 .tos_opc_end = LFSCK_LAST_OPC,
4488 .tos_hs = tgt_lfsck_handlers
4496 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4498 struct md_device *next = m->mdt_child;
4499 struct lu_device *d = &m->mdt_lu_dev;
4500 struct obd_device *obd = mdt2obd_dev(m);
4501 struct lfsck_stop stop;
4504 stop.ls_status = LS_PAUSED;
4506 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4508 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4509 target_recovery_fini(obd);
4510 ping_evictor_stop();
4512 if (m->mdt_opts.mo_coordinator)
4513 mdt_hsm_cdt_stop(m);
4515 mdt_hsm_cdt_fini(m);
4517 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4518 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4520 if (m->mdt_namespace != NULL)
4521 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4522 d->ld_obd->obd_force);
4524 obd_exports_barrier(obd);
4525 obd_zombie_barrier();
4529 tgt_fini(env, &m->mdt_lut);
4530 mdt_fs_cleanup(env, m);
4531 upcall_cache_cleanup(m->mdt_identity_cache);
4532 m->mdt_identity_cache = NULL;
4534 if (m->mdt_namespace != NULL) {
4535 ldlm_namespace_free_post(m->mdt_namespace);
4536 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4539 mdt_quota_fini(env, m);
4541 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4543 mdt_seq_fini(env, m);
4544 mdt_fld_fini(env, m);
4549 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4551 LASSERT(atomic_read(&d->ld_ref) == 0);
4553 server_put_mount(mdt_obd_name(m), true);
4558 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4560 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4561 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4563 struct mdt_thread_info *info;
4564 struct obd_device *obd;
4565 const char *dev = lustre_cfg_string(cfg, 0);
4566 const char *num = lustre_cfg_string(cfg, 2);
4567 struct lustre_mount_info *lmi = NULL;
4568 struct lustre_sb_info *lsi;
4570 struct seq_server_site *ss_site;
4571 const char *identity_upcall = "NONE";
4572 struct md_device *next;
4578 lu_device_init(&m->mdt_lu_dev, ldt);
4580 * Environment (env) might be missing mdt_thread_key values at that
4581 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4584 * Usually device allocation path doesn't use module key values, but
4585 * mdt has to do a lot of work here, so allocate key value.
4587 rc = lu_env_refill((struct lu_env *)env);
4591 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4592 LASSERT(info != NULL);
4594 obd = class_name2obd(dev);
4595 LASSERT(obd != NULL);
4597 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4598 m->mdt_opts.mo_evict_tgt_nids = 1;
4599 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4601 /* default is coordinator off, it is started through conf_param
4603 m->mdt_opts.mo_coordinator = 0;
4605 lmi = server_get_mount(dev);
4607 CERROR("Cannot get mount info for %s!\n", dev);
4610 lsi = s2lsi(lmi->lmi_sb);
4611 /* CMD is supported only in IAM mode */
4613 node_id = simple_strtol(num, NULL, 10);
4614 obd->u.obt.obt_magic = OBT_MAGIC;
4615 if (lsi->lsi_lmd != NULL &&
4616 lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4617 m->mdt_skip_lfsck = 1;
4620 m->mdt_squash.rsi_uid = 0;
4621 m->mdt_squash.rsi_gid = 0;
4622 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4623 init_rwsem(&m->mdt_squash.rsi_sem);
4624 spin_lock_init(&m->mdt_osfs_lock);
4625 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4626 m->mdt_enable_remote_dir = 0;
4627 m->mdt_enable_remote_dir_gid = 0;
4629 atomic_set(&m->mdt_mds_mds_conns, 0);
4630 atomic_set(&m->mdt_async_commit_count, 0);
4632 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4633 m->mdt_lu_dev.ld_obd = obd;
4634 /* Set this lu_device to obd for error handling purposes. */
4635 obd->obd_lu_dev = &m->mdt_lu_dev;
4637 /* init the stack */
4638 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4640 CERROR("%s: Can't init device stack, rc %d\n",
4641 mdt_obd_name(m), rc);
4646 ss_site = mdt_seq_site(m);
4647 s->ld_seq_site = ss_site;
4650 /* set server index */
4651 ss_site->ss_node_id = node_id;
4653 /* failover is the default
4654 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4655 * assumed whose ss_node_id == 0 XXX
4657 obd->obd_replayable = 1;
4658 /* No connection accepted until configurations will finish */
4659 obd->obd_no_conn = 1;
4661 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4662 char *str = lustre_cfg_string(cfg, 4);
4663 if (strchr(str, 'n')) {
4664 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4665 obd->obd_replayable = 0;
4669 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4671 GOTO(err_fini_stack, rc);
4673 rc = mdt_seq_init(env, m);
4675 GOTO(err_fini_fld, rc);
4677 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4678 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4679 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4680 LDLM_NAMESPACE_SERVER,
4681 LDLM_NAMESPACE_GREEDY,
4683 if (m->mdt_namespace == NULL)
4684 GOTO(err_fini_seq, rc = -ENOMEM);
4686 m->mdt_namespace->ns_lvbp = m;
4687 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4689 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4690 /* set obd_namespace for compatibility with old code */
4691 obd->obd_namespace = m->mdt_namespace;
4693 rc = mdt_hsm_cdt_init(m);
4695 CERROR("%s: error initializing coordinator, rc %d\n",
4696 mdt_obd_name(m), rc);
4697 GOTO(err_free_ns, rc);
4700 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4701 OBD_FAIL_MDS_ALL_REQUEST_NET,
4702 OBD_FAIL_MDS_ALL_REPLY_NET);
4704 GOTO(err_free_hsm, rc);
4706 rc = mdt_fs_setup(env, m, obd, lsi);
4710 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4712 next = m->mdt_child;
4713 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4716 GOTO(err_fs_cleanup, rc);
4718 if (mntopts & MNTOPT_USERXATTR)
4719 m->mdt_opts.mo_user_xattr = 1;
4721 m->mdt_opts.mo_user_xattr = 0;
4723 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4725 GOTO(err_fs_cleanup, rc);
4727 if (mntopts & MNTOPT_ACL)
4728 m->mdt_opts.mo_acl = 1;
4730 m->mdt_opts.mo_acl = 0;
4732 /* XXX: to support suppgid for ACL, we enable identity_upcall
4733 * by default, otherwise, maybe got unexpected -EACCESS. */
4734 if (m->mdt_opts.mo_acl)
4735 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4737 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4739 &mdt_identity_upcall_cache_ops);
4740 if (IS_ERR(m->mdt_identity_cache)) {
4741 rc = PTR_ERR(m->mdt_identity_cache);
4742 m->mdt_identity_cache = NULL;
4743 GOTO(err_fs_cleanup, rc);
4746 rc = mdt_procfs_init(m, dev);
4748 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4749 GOTO(err_recovery, rc);
4752 rc = mdt_quota_init(env, m, cfg);
4754 GOTO(err_procfs, rc);
4756 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4757 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4758 "mdt_ldlm_client", m->mdt_ldlm_client);
4760 ping_evictor_start();
4762 /* recovery will be started upon mdt_prepare()
4763 * when the whole stack is complete and ready
4764 * to serve the requests */
4766 /* Reduce the initial timeout on an MDS because it doesn't need such
4767 * a long timeout as an OST does. Adaptive timeouts will adjust this
4768 * value appropriately. */
4769 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4770 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4776 target_recovery_fini(obd);
4777 upcall_cache_cleanup(m->mdt_identity_cache);
4778 m->mdt_identity_cache = NULL;
4780 mdt_fs_cleanup(env, m);
4782 tgt_fini(env, &m->mdt_lut);
4784 mdt_hsm_cdt_fini(m);
4786 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4787 obd->obd_namespace = m->mdt_namespace = NULL;
4789 mdt_seq_fini(env, m);
4791 mdt_fld_fini(env, m);
4793 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4796 server_put_mount(dev, true);
4800 /* For interoperability, the left element is old parameter, the right one
4801 * is the new version of the parameter, if some parameter is deprecated,
4802 * the new version should be set as NULL. */
4803 static struct cfg_interop_param mdt_interop_param[] = {
4804 { "mdt.group_upcall", NULL },
4805 { "mdt.quota_type", NULL },
4806 { "mdd.quota_type", NULL },
4807 { "mdt.som", NULL },
4808 { "mdt.rootsquash", "mdt.root_squash" },
4809 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4813 /* used by MGS to process specific configurations */
4814 static int mdt_process_config(const struct lu_env *env,
4815 struct lu_device *d, struct lustre_cfg *cfg)
4817 struct mdt_device *m = mdt_dev(d);
4818 struct md_device *md_next = m->mdt_child;
4819 struct lu_device *next = md2lu_dev(md_next);
4823 switch (cfg->lcfg_command) {
4825 struct obd_device *obd = d->ld_obd;
4827 /* For interoperability */
4828 struct cfg_interop_param *ptr = NULL;
4829 struct lustre_cfg *old_cfg = NULL;
4832 param = lustre_cfg_string(cfg, 1);
4833 if (param == NULL) {
4834 CERROR("param is empty\n");
4839 ptr = class_find_old_param(param, mdt_interop_param);
4841 if (ptr->new_param == NULL) {
4843 CWARN("For interoperability, skip this %s."
4844 " It is obsolete.\n", ptr->old_param);
4848 CWARN("Found old param %s, changed it to %s.\n",
4849 ptr->old_param, ptr->new_param);
4852 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4859 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4861 if (rc > 0 || rc == -ENOSYS) {
4862 /* is it an HSM var ? */
4863 rc = class_process_proc_param(PARAM_HSM,
4864 hsm_cdt_get_proc_vars(),
4866 if (rc > 0 || rc == -ENOSYS)
4867 /* we don't understand; pass it on */
4868 rc = next->ld_ops->ldo_process_config(env, next,
4872 if (old_cfg != NULL)
4873 lustre_cfg_free(cfg);
4878 /* others are passed further */
4879 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4885 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4886 const struct lu_object_header *hdr,
4887 struct lu_device *d)
4889 struct mdt_object *mo;
4893 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4895 struct lu_object *o;
4896 struct lu_object_header *h;
4899 h = &mo->mot_header;
4900 lu_object_header_init(h);
4901 lu_object_init(o, h, d);
4902 lu_object_add_top(h, o);
4903 o->lo_ops = &mdt_obj_ops;
4904 spin_lock_init(&mo->mot_write_lock);
4905 mutex_init(&mo->mot_lov_mutex);
4906 init_rwsem(&mo->mot_open_sem);
4912 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4913 const struct lu_object_conf *unused)
4915 struct mdt_device *d = mdt_dev(o->lo_dev);
4916 struct lu_device *under;
4917 struct lu_object *below;
4921 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4922 PFID(lu_object_fid(o)));
4924 under = &d->mdt_child->md_lu_dev;
4925 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4926 if (below != NULL) {
4927 lu_object_add(o, below);
4934 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4936 struct mdt_object *mo = mdt_obj(o);
4937 struct lu_object_header *h;
4941 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4942 PFID(lu_object_fid(o)));
4944 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4945 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4948 lu_object_header_fini(h);
4949 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4954 static int mdt_object_print(const struct lu_env *env, void *cookie,
4955 lu_printer_t p, const struct lu_object *o)
4957 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4959 return (*p)(env, cookie,
4960 LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4961 mdto, mdto->mot_flags, mdto->mot_write_count);
4964 static int mdt_prepare(const struct lu_env *env,
4965 struct lu_device *pdev,
4966 struct lu_device *cdev)
4968 struct mdt_device *mdt = mdt_dev(cdev);
4969 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4970 struct obd_device *obd = cdev->ld_obd;
4977 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4981 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4985 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4989 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4990 /* The LFSCK instance is registered just now, so it must be there when
4991 * register the namespace to such instance. */
4992 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4994 if (mdt->mdt_seq_site.ss_node_id == 0) {
4995 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4996 &mdt->mdt_md_root_fid);
5001 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5003 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5004 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5005 LASSERT(obd->obd_no_conn);
5006 spin_lock(&obd->obd_dev_lock);
5007 obd->obd_no_conn = 0;
5008 spin_unlock(&obd->obd_dev_lock);
5010 if (obd->obd_recovering == 0)
5011 mdt_postrecov(env, mdt);
5016 const struct lu_device_operations mdt_lu_ops = {
5017 .ldo_object_alloc = mdt_object_alloc,
5018 .ldo_process_config = mdt_process_config,
5019 .ldo_prepare = mdt_prepare,
5022 static const struct lu_object_operations mdt_obj_ops = {
5023 .loo_object_init = mdt_object_init,
5024 .loo_object_free = mdt_object_free,
5025 .loo_object_print = mdt_object_print
5028 static int mdt_obd_set_info_async(const struct lu_env *env,
5029 struct obd_export *exp,
5030 __u32 keylen, void *key,
5031 __u32 vallen, void *val,
5032 struct ptlrpc_request_set *set)
5038 if (KEY_IS(KEY_SPTLRPC_CONF)) {
5039 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
5047 * Match client and server connection feature flags.
5049 * Compute the compatibility flags for a connection request based on
5050 * features mutually supported by client and server.
5052 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5053 * must not be updated here, otherwise a partially initialized value may
5054 * be exposed. After the connection request is successfully processed,
5055 * the top-level MDT connect request handler atomically updates the export
5056 * connect flags from the obd_connect_data::ocd_connect_flags field of the
5057 * reply. \see mdt_connect().
5059 * \param exp the obd_export associated with this client/target pair
5060 * \param mdt the target device for the connection
5061 * \param data stores data for this connect request
5064 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5065 * \retval -EBADE client and server feature requirements are incompatible
5067 static int mdt_connect_internal(struct obd_export *exp,
5068 struct mdt_device *mdt,
5069 struct obd_connect_data *data)
5071 LASSERT(data != NULL);
5073 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5074 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5076 if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5077 !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5078 CWARN("%s: client %s does not support ibits lock, either "
5079 "very old or an invalid client: flags "LPX64"\n",
5080 mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5081 data->ocd_connect_flags);
5085 if (!mdt->mdt_opts.mo_acl)
5086 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5088 if (!mdt->mdt_opts.mo_user_xattr)
5089 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5091 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5092 data->ocd_brw_size = min(data->ocd_brw_size,
5093 (__u32)MD_MAX_BRW_SIZE);
5094 if (data->ocd_brw_size == 0) {
5095 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5096 " ocd_version: %x ocd_grant: %d "
5097 "ocd_index: %u ocd_brw_size is "
5098 "unexpectedly zero, network data "
5099 "corruption? Refusing connection of this"
5102 exp->exp_client_uuid.uuid,
5103 exp, data->ocd_connect_flags, data->ocd_version,
5104 data->ocd_grant, data->ocd_index);
5109 /* NB: Disregard the rule against updating
5110 * exp_connect_data.ocd_connect_flags in this case, since
5111 * tgt_client_new() needs to know if this is a lightweight
5112 * connection, and it is safe to expose this flag before
5113 * connection processing completes. */
5114 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5115 spin_lock(&exp->exp_lock);
5116 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5117 spin_unlock(&exp->exp_lock);
5120 data->ocd_version = LUSTRE_VERSION_CODE;
5122 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5123 CWARN("%s: MDS requires FID support, but client not\n",
5128 if (OCD_HAS_FLAG(data, PINGLESS)) {
5129 if (ptlrpc_pinger_suppress_pings()) {
5130 spin_lock(&exp->exp_obd->obd_dev_lock);
5131 list_del_init(&exp->exp_obd_chain_timed);
5132 spin_unlock(&exp->exp_obd->obd_dev_lock);
5134 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5138 data->ocd_max_easize = mdt->mdt_max_ea_size;
5140 /* NB: Disregard the rule against updating
5141 * exp_connect_data.ocd_connect_flags in this case, since
5142 * tgt_client_new() needs to know if this is client supports
5143 * multiple modify RPCs, and it is safe to expose this flag before
5144 * connection processing completes. */
5145 if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5146 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5147 spin_lock(&exp->exp_lock);
5148 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5149 spin_unlock(&exp->exp_lock);
5155 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5156 struct mdt_thread_info *info,
5157 struct mdt_file_data *mfd)
5159 struct lu_context ses;
5163 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5168 lu_context_enter(&ses);
5170 mdt_ucred(info)->uc_valid = UCRED_OLD;
5171 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5173 lu_context_exit(&ses);
5174 lu_context_fini(&ses);
5180 static int mdt_export_cleanup(struct obd_export *exp)
5182 struct list_head closing_list;
5183 struct mdt_export_data *med = &exp->exp_mdt_data;
5184 struct obd_device *obd = exp->exp_obd;
5185 struct mdt_device *mdt;
5186 struct mdt_thread_info *info;
5188 struct mdt_file_data *mfd, *n;
5192 INIT_LIST_HEAD(&closing_list);
5193 spin_lock(&med->med_open_lock);
5194 while (!list_empty(&med->med_open_head)) {
5195 struct list_head *tmp = med->med_open_head.next;
5196 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5198 /* Remove mfd handle so it can't be found again.
5199 * We are consuming the mfd_list reference here. */
5200 class_handle_unhash(&mfd->mfd_handle);
5201 list_move_tail(&mfd->mfd_list, &closing_list);
5203 spin_unlock(&med->med_open_lock);
5204 mdt = mdt_dev(obd->obd_lu_dev);
5205 LASSERT(mdt != NULL);
5207 rc = lu_env_init(&env, LCT_MD_THREAD);
5211 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5212 LASSERT(info != NULL);
5213 memset(info, 0, sizeof *info);
5214 info->mti_env = &env;
5215 info->mti_mdt = mdt;
5216 info->mti_exp = exp;
5218 if (!list_empty(&closing_list)) {
5219 struct md_attr *ma = &info->mti_attr;
5221 /* Close any open files (which may also cause orphan
5223 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5224 list_del_init(&mfd->mfd_list);
5225 ma->ma_need = ma->ma_valid = 0;
5227 /* This file is being closed due to an eviction, it
5228 * could have been modified and now dirty regarding to
5229 * HSM archive, check this!
5230 * The logic here is to mark a file dirty if there's a
5231 * chance it was dirtied before the client was evicted,
5232 * so that we don't have to wait for a release attempt
5233 * before finding out the file was actually dirty and
5234 * fail the release. Aggressively marking it dirty here
5235 * will cause the policy engine to attempt to
5236 * re-archive it; when rearchiving, we can compare the
5237 * current version to the HSM data_version and make the
5238 * archive request into a noop if it's not actually
5241 if (mfd->mfd_mode & FMODE_WRITE)
5242 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5244 /* Don't unlink orphan on failover umount, LU-184 */
5245 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5246 ma->ma_valid = MA_FLAGS;
5247 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5249 mdt_mfd_close(info, mfd);
5252 info->mti_mdt = NULL;
5253 /* cleanup client slot early */
5254 /* Do not erase record for recoverable client. */
5255 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5256 tgt_client_del(&env, exp);
5262 static inline void mdt_enable_slc(struct mdt_device *mdt)
5264 if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5265 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5268 static inline void mdt_disable_slc(struct mdt_device *mdt)
5270 if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5271 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5274 static int mdt_obd_disconnect(struct obd_export *exp)
5280 class_export_get(exp);
5282 if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5283 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5284 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5286 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5287 mdt_disable_slc(mdt);
5290 rc = server_disconnect_export(exp);
5292 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5294 rc = mdt_export_cleanup(exp);
5295 nodemap_del_member(exp);
5296 class_export_put(exp);
5300 /* mds_connect copy */
5301 static int mdt_obd_connect(const struct lu_env *env,
5302 struct obd_export **exp, struct obd_device *obd,
5303 struct obd_uuid *cluuid,
5304 struct obd_connect_data *data,
5307 struct obd_export *lexp;
5308 struct lustre_handle conn = { 0 };
5309 struct mdt_device *mdt;
5311 lnet_nid_t *client_nid = localdata;
5314 LASSERT(env != NULL);
5315 if (!exp || !obd || !cluuid)
5318 mdt = mdt_dev(obd->obd_lu_dev);
5320 if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5321 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5322 atomic_inc(&mdt->mdt_mds_mds_conns);
5323 mdt_enable_slc(mdt);
5327 * first, check whether the stack is ready to handle requests
5328 * XXX: probably not very appropriate method is used now
5329 * at some point we should find a better one
5331 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5332 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5333 !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5334 rc = obd_get_info(env, mdt->mdt_child_exp,
5335 sizeof(KEY_OSP_CONNECTED),
5336 KEY_OSP_CONNECTED, NULL, NULL);
5339 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5342 rc = class_connect(&conn, obd, cluuid);
5346 lexp = class_conn2export(&conn);
5347 LASSERT(lexp != NULL);
5349 rc = nodemap_add_member(*client_nid, lexp);
5350 if (rc != 0 && rc != -EEXIST)
5353 rc = mdt_connect_internal(lexp, mdt, data);
5355 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5358 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5359 rc = tgt_client_new(env, lexp);
5361 mdt_export_stats_init(obd, lexp, localdata);
5365 class_disconnect(lexp);
5366 nodemap_del_member(lexp);
5370 /* Because we do not want this export to be evicted by pinger,
5371 * let's not add this export to the timed chain list. */
5372 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5373 spin_lock(&lexp->exp_obd->obd_dev_lock);
5374 list_del_init(&lexp->exp_obd_chain_timed);
5375 spin_unlock(&lexp->exp_obd->obd_dev_lock);
5382 static int mdt_obd_reconnect(const struct lu_env *env,
5383 struct obd_export *exp, struct obd_device *obd,
5384 struct obd_uuid *cluuid,
5385 struct obd_connect_data *data,
5388 lnet_nid_t *client_nid = localdata;
5392 if (exp == NULL || obd == NULL || cluuid == NULL)
5395 rc = nodemap_add_member(*client_nid, exp);
5396 if (rc != 0 && rc != -EEXIST)
5399 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5401 mdt_export_stats_init(obd, exp, localdata);
5403 nodemap_del_member(exp);
5408 /* FIXME: Can we avoid using these two interfaces? */
5409 static int mdt_init_export(struct obd_export *exp)
5411 struct mdt_export_data *med = &exp->exp_mdt_data;
5415 INIT_LIST_HEAD(&med->med_open_head);
5416 spin_lock_init(&med->med_open_lock);
5417 mutex_init(&med->med_idmap_mutex);
5418 med->med_idmap = NULL;
5419 spin_lock(&exp->exp_lock);
5420 exp->exp_connecting = 1;
5421 spin_unlock(&exp->exp_lock);
5423 /* self-export doesn't need client data and ldlm initialization */
5424 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5425 &exp->exp_client_uuid)))
5428 rc = tgt_client_alloc(exp);
5432 rc = ldlm_init_export(exp);
5439 tgt_client_free(exp);
5441 CERROR("%s: Failed to initialize export: rc = %d\n",
5442 exp->exp_obd->obd_name, rc);
5446 static int mdt_destroy_export(struct obd_export *exp)
5450 if (exp_connect_rmtclient(exp))
5451 mdt_cleanup_idmap(&exp->exp_mdt_data);
5453 target_destroy_export(exp);
5454 /* destroy can be called from failed obd_setup, so
5455 * checking uuid is safer than obd_self_export */
5456 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5457 &exp->exp_client_uuid)))
5460 ldlm_destroy_export(exp);
5461 tgt_client_free(exp);
5463 LASSERT(list_empty(&exp->exp_outstanding_replies));
5464 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5469 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5470 struct linkea_data *ldata)
5474 LASSERT(ldata->ld_buf->lb_buf != NULL);
5476 if (!mdt_object_exists(mdt_obj))
5479 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5480 ldata->ld_buf, XATTR_NAME_LINK);
5481 if (rc == -ERANGE) {
5482 /* Buf was too small, figure out what we need. */
5483 lu_buf_free(ldata->ld_buf);
5484 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5485 ldata->ld_buf, XATTR_NAME_LINK);
5488 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5489 if (ldata->ld_buf->lb_buf == NULL)
5491 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5492 ldata->ld_buf, XATTR_NAME_LINK);
5497 return linkea_init(ldata);
5501 * Given an MDT object, try to look up the full path to the object.
5502 * Part of the MDT layer implementation of lfs fid2path.
5504 * \param[in] info Per-thread common data shared by MDT level handlers.
5505 * \param[in] obj Object to do path lookup of
5506 * \param[in,out] fp User-provided struct to store path information
5507 * \param[in] root_fid Root FID of current path should reach
5509 * \retval 0 Lookup successful, path information stored in fp
5510 * \retval -EAGAIN Lookup failed, usually because object is being moved
5511 * \retval negative errno if there was a problem
5513 static int mdt_path_current(struct mdt_thread_info *info,
5514 struct mdt_object *obj,
5515 struct getinfo_fid2path *fp,
5516 struct lu_fid *root_fid)
5518 struct mdt_device *mdt = info->mti_mdt;
5519 struct mdt_object *mdt_obj;
5520 struct link_ea_header *leh;
5521 struct link_ea_entry *lee;
5522 struct lu_name *tmpname = &info->mti_name;
5523 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5524 struct lu_buf *buf = &info->mti_big_buf;
5527 struct linkea_data ldata = { NULL };
5532 /* temp buffer for path element, the buffer will be finally freed
5533 * in mdt_thread_info_fini */
5534 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5535 if (buf->lb_buf == NULL)
5539 ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5542 *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5544 while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5545 struct lu_buf lmv_buf;
5547 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5548 lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5549 GOTO(out, rc = -ENOENT);
5551 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5552 if (IS_ERR(mdt_obj))
5553 GOTO(out, rc = PTR_ERR(mdt_obj));
5555 if (!mdt_object_exists(mdt_obj)) {
5556 mdt_object_put(info->mti_env, mdt_obj);
5557 GOTO(out, rc = -ENOENT);
5560 if (mdt_object_remote(mdt_obj)) {
5561 mdt_object_put(info->mti_env, mdt_obj);
5562 GOTO(remote_out, rc = -EREMOTE);
5565 rc = mdt_links_read(info, mdt_obj, &ldata);
5567 mdt_object_put(info->mti_env, mdt_obj);
5572 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5573 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5574 /* If set, use link #linkno for path lookup, otherwise use
5575 link #0. Only do this for the final path element. */
5576 if (first && fp->gf_linkno < leh->leh_reccount) {
5578 for (count = 0; count < fp->gf_linkno; count++) {
5579 lee = (struct link_ea_entry *)
5580 ((char *)lee + reclen);
5581 linkea_entry_unpack(lee, &reclen, tmpname,
5584 if (fp->gf_linkno < leh->leh_reccount - 1)
5585 /* indicate to user there are more links */
5589 lmv_buf.lb_buf = info->mti_xattr_buf;
5590 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5591 /* Check if it is slave stripes */
5592 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5593 &lmv_buf, XATTR_NAME_LMV);
5594 mdt_object_put(info->mti_env, mdt_obj);
5596 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5598 /* For slave stripes, get its master */
5599 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5600 fp->gf_fid = *tmpfid;
5603 } else if (rc < 0 && rc != -ENODATA) {
5609 /* Pack the name in the end of the buffer */
5610 ptr -= tmpname->ln_namelen;
5611 if (ptr - 1 <= fp->gf_u.gf_path)
5612 GOTO(out, rc = -EOVERFLOW);
5613 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5616 /* keep the last resolved fid to the client, so the
5617 * client will build the left path on another MDT for
5619 fp->gf_fid = *tmpfid;
5625 ptr++; /* skip leading / */
5626 memmove(fp->gf_u.gf_path, ptr,
5627 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5634 * Given an MDT object, use mdt_path_current to get the path.
5635 * Essentially a wrapper to retry mdt_path_current a set number of times
5636 * if -EAGAIN is returned (usually because an object is being moved).
5638 * Part of the MDT layer implementation of lfs fid2path.
5640 * \param[in] info Per-thread common data shared by mdt level handlers.
5641 * \param[in] obj Object to do path lookup of
5642 * \param[in,out] fp User-provided struct for arguments and to store path
5645 * \retval 0 Lookup successful, path information stored in fp
5646 * \retval negative errno if there was a problem
5648 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5649 struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5651 struct mdt_device *mdt = info->mti_mdt;
5656 if (fp->gf_pathlen < 3)
5659 if (root_fid == NULL)
5660 root_fid = &mdt->mdt_md_root_fid;
5662 if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5663 fp->gf_u.gf_path[0] = '\0';
5667 /* Retry multiple times in case file is being moved */
5668 while (tries-- && rc == -EAGAIN)
5669 rc = mdt_path_current(info, obj, fp, root_fid);
5675 * Get the full path of the provided FID, as of changelog record recno.
5677 * This checks sanity and looks up object for user provided FID
5678 * before calling the actual path lookup code.
5680 * Part of the MDT layer implementation of lfs fid2path.
5682 * \param[in] info Per-thread common data shared by mdt level handlers.
5683 * \param[in,out] fp User-provided struct for arguments and to store path
5686 * \retval 0 Lookup successful, path information and recno stored in fp
5687 * \retval -ENOENT, object does not exist
5688 * \retval negative errno if there was a problem
5690 static int mdt_fid2path(struct mdt_thread_info *info,
5691 struct lu_fid *root_fid,
5692 struct getinfo_fid2path *fp)
5694 struct mdt_device *mdt = info->mti_mdt;
5695 struct mdt_object *obj;
5699 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5700 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5702 if (!fid_is_sane(&fp->gf_fid))
5705 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5706 CWARN("%s: "DFID" is invalid, sequence should be "
5707 ">= "LPX64"\n", mdt_obd_name(mdt),
5708 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5712 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5715 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5716 PFID(&fp->gf_fid), rc);
5720 if (mdt_object_remote(obj))
5722 else if (!mdt_object_exists(obj))
5728 mdt_object_put(info->mti_env, obj);
5729 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5730 PFID(&fp->gf_fid), rc);
5734 rc = mdt_path(info, obj, fp, root_fid);
5736 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5737 PFID(&fp->gf_fid), fp->gf_u.gf_path,
5738 fp->gf_recno, fp->gf_linkno);
5740 mdt_object_put(info->mti_env, obj);
5745 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5746 void *val, int vallen)
5748 struct getinfo_fid2path *fpout, *fpin;
5749 struct lu_fid *root_fid = NULL;
5752 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5755 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5756 lustre_swab_fid2path(fpin);
5758 memcpy(fpout, fpin, sizeof(*fpin));
5759 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5762 if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5763 sizeof(struct lu_fid)) {
5764 /* client sent its root FID, which is normally fileset FID */
5765 root_fid = fpin->gf_u.gf_root_fid;
5766 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5767 lustre_swab_lu_fid(root_fid);
5769 if (root_fid != NULL && !fid_is_sane(root_fid))
5773 rc = mdt_fid2path(info, root_fid, fpout);
5777 int mdt_get_info(struct tgt_session_info *tsi)
5787 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5789 CDEBUG(D_IOCTL, "No GETINFO key\n");
5790 RETURN(err_serious(-EFAULT));
5792 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5795 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5796 if (vallen == NULL) {
5797 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5798 tgt_name(tsi->tsi_tgt));
5799 RETURN(err_serious(-EFAULT));
5802 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5804 rc = req_capsule_server_pack(tsi->tsi_pill);
5806 RETURN(err_serious(rc));
5808 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5809 if (valout == NULL) {
5810 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5811 tgt_name(tsi->tsi_tgt));
5812 RETURN(err_serious(-EFAULT));
5815 if (KEY_IS(KEY_FID2PATH)) {
5816 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5818 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5819 mdt_thread_info_fini(info);
5826 /* Pass the ioc down */
5827 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5828 unsigned int cmd, int len, void *data)
5830 struct lu_context ioctl_session;
5831 struct md_device *next = mdt->mdt_child;
5835 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5838 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5839 lu_context_enter(&ioctl_session);
5840 env->le_ses = &ioctl_session;
5842 LASSERT(next->md_ops->mdo_iocontrol);
5843 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5845 lu_context_exit(&ioctl_session);
5846 lu_context_fini(&ioctl_session);
5850 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5852 struct obd_ioctl_data *data = karg;
5855 struct mdt_object *obj;
5856 struct mdt_lock_handle *lh;
5860 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5861 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5864 fid = (struct lu_fid *)data->ioc_inlbuf1;
5866 if (!fid_is_sane(fid))
5869 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5871 lh = &mti->mti_lh[MDT_LH_PARENT];
5872 mdt_lock_reg_init(lh, LCK_CR);
5874 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5876 RETURN(PTR_ERR(obj));
5878 if (mdt_object_remote(obj)) {
5881 * before calling version get the correct MDS should be
5882 * fid, this is error to find remote object here
5884 CERROR("nonlocal object "DFID"\n", PFID(fid));
5885 } else if (!mdt_object_exists(obj)) {
5886 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5889 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5890 *(__u64 *)data->ioc_inlbuf2 = version;
5893 mdt_object_unlock_put(mti, obj, lh, 1);
5897 /* ioctls on obd dev */
5898 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5899 void *karg, void __user *uarg)
5902 struct obd_device *obd = exp->exp_obd;
5903 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5904 struct dt_device *dt = mdt->mdt_bottom;
5908 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5909 rc = lu_env_init(&env, LCT_MD_THREAD);
5915 rc = mdt_device_sync(&env, mdt);
5917 case OBD_IOC_SET_READONLY:
5918 rc = dt->dd_ops->dt_ro(&env, dt);
5920 case OBD_IOC_ABORT_RECOVERY:
5921 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5922 obd->obd_abort_recovery = 1;
5923 target_stop_recovery_thread(obd);
5926 case OBD_IOC_CHANGELOG_REG:
5927 case OBD_IOC_CHANGELOG_DEREG:
5928 case OBD_IOC_CHANGELOG_CLEAR:
5929 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5931 case OBD_IOC_START_LFSCK: {
5932 struct md_device *next = mdt->mdt_child;
5933 struct obd_ioctl_data *data = karg;
5934 struct lfsck_start_param lsp;
5936 if (unlikely(data == NULL)) {
5941 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5942 lsp.lsp_index_valid = 0;
5943 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5946 case OBD_IOC_STOP_LFSCK: {
5947 struct md_device *next = mdt->mdt_child;
5948 struct obd_ioctl_data *data = karg;
5949 struct lfsck_stop stop;
5951 stop.ls_status = LS_STOPPED;
5952 /* Old lfsck utils may pass NULL @stop. */
5953 if (data->ioc_inlbuf1 == NULL)
5957 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5959 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5962 case OBD_IOC_QUERY_LFSCK: {
5963 struct md_device *next = mdt->mdt_child;
5964 struct obd_ioctl_data *data = karg;
5966 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5970 case OBD_IOC_GET_OBJ_VERSION: {
5971 struct mdt_thread_info *mti;
5972 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5973 memset(mti, 0, sizeof *mti);
5974 mti->mti_env = &env;
5978 rc = mdt_ioc_version_get(mti, karg);
5981 case OBD_IOC_CATLOGLIST: {
5982 struct mdt_thread_info *mti;
5984 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5985 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5986 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5987 &mti->mti_tmp_fid1);
5992 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5993 mdt_obd_name(mdt), cmd, rc);
6000 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6002 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6006 if (!mdt->mdt_skip_lfsck) {
6007 struct lfsck_start_param lsp;
6009 lsp.lsp_start = NULL;
6010 lsp.lsp_index_valid = 0;
6011 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6012 OBD_IOC_START_LFSCK,
6014 if (rc != 0 && rc != -EALREADY)
6015 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6016 mdt_obd_name(mdt), rc);
6019 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6023 static int mdt_obd_postrecov(struct obd_device *obd)
6028 rc = lu_env_init(&env, LCT_MD_THREAD);
6031 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6036 static struct obd_ops mdt_obd_device_ops = {
6037 .o_owner = THIS_MODULE,
6038 .o_set_info_async = mdt_obd_set_info_async,
6039 .o_connect = mdt_obd_connect,
6040 .o_reconnect = mdt_obd_reconnect,
6041 .o_disconnect = mdt_obd_disconnect,
6042 .o_init_export = mdt_init_export,
6043 .o_destroy_export = mdt_destroy_export,
6044 .o_iocontrol = mdt_iocontrol,
6045 .o_postrecov = mdt_obd_postrecov,
6048 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6049 struct lu_device *d)
6051 struct mdt_device *m = mdt_dev(d);
6058 static struct lu_device *mdt_device_free(const struct lu_env *env,
6059 struct lu_device *d)
6061 struct mdt_device *m = mdt_dev(d);
6064 lu_device_fini(&m->mdt_lu_dev);
6070 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6071 struct lu_device_type *t,
6072 struct lustre_cfg *cfg)
6074 struct lu_device *l;
6075 struct mdt_device *m;
6082 rc = mdt_init0(env, m, t, cfg);
6084 mdt_device_free(env, l);
6089 l = ERR_PTR(-ENOMEM);
6093 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6094 LU_KEY_INIT(mdt, struct mdt_thread_info);
6096 static void mdt_key_fini(const struct lu_context *ctx,
6097 struct lu_context_key *key, void* data)
6099 struct mdt_thread_info *info = data;
6101 if (info->mti_big_lmm) {
6102 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6103 info->mti_big_lmm = NULL;
6104 info->mti_big_lmmsize = 0;
6109 /* context key: mdt_thread_key */
6110 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6112 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6114 return lu_ucred(info->mti_env);
6117 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6119 return lu_ucred_check(info->mti_env);
6123 * Enable/disable COS (Commit On Sharing).
6125 * Set/Clear the COS flag in mdt options.
6127 * \param mdt mdt device
6128 * \param val 0 disables COS, other values enable COS
6130 void mdt_enable_cos(struct mdt_device *mdt, int val)
6135 mdt->mdt_opts.mo_cos = !!val;
6136 rc = lu_env_init(&env, LCT_LOCAL);
6137 if (unlikely(rc != 0)) {
6138 CWARN("%s: lu_env initialization failed, cannot "
6139 "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6142 mdt_device_sync(&env, mdt);
6147 * Check COS (Commit On Sharing) status.
6149 * Return COS flag status.
6151 * \param mdt mdt device
6153 int mdt_cos_is_enabled(struct mdt_device *mdt)
6155 return mdt->mdt_opts.mo_cos != 0;
6158 static struct lu_device_type_operations mdt_device_type_ops = {
6159 .ldto_device_alloc = mdt_device_alloc,
6160 .ldto_device_free = mdt_device_free,
6161 .ldto_device_fini = mdt_device_fini
6164 static struct lu_device_type mdt_device_type = {
6165 .ldt_tags = LU_DEVICE_MD,
6166 .ldt_name = LUSTRE_MDT_NAME,
6167 .ldt_ops = &mdt_device_type_ops,
6168 .ldt_ctx_tags = LCT_MD_THREAD
6171 static int __init mdt_init(void)
6175 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6176 FID_NOBRACE_LEN + 1);
6177 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6179 rc = lu_kmem_init(mdt_caches);
6183 rc = mds_mod_init();
6187 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6188 LUSTRE_MDT_NAME, &mdt_device_type);
6193 lu_kmem_fini(mdt_caches);
6200 static void __exit mdt_exit(void)
6202 class_unregister_type(LUSTRE_MDT_NAME);
6204 lu_kmem_fini(mdt_caches);
6207 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6208 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6209 MODULE_VERSION(LUSTRE_VERSION_STRING);
6210 MODULE_LICENSE("GPL");
6212 module_init(mdt_init);
6213 module_exit(mdt_exit);