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 CFS_MODULE_PARM(max_mod_rpcs_per_client, "i", uint, 0644,
75 "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) {
315 strncpy(name, s1, lname->ln_namelen);
316 name[lname->ln_namelen] = '\0';
318 parent = mdt_object_find(info->mti_env, mdt, fid);
319 if (IS_ERR(parent)) {
320 rc = PTR_ERR(parent);
323 /* Only got the fid of this obj by name */
325 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
326 fid, &info->mti_spec);
327 mdt_object_put(info->mti_env, parent);
330 parent = mdt_object_find(info->mti_env, mdt, fid);
332 rc = PTR_ERR(parent);
334 mode = lu_object_attr(&parent->mot_obj);
335 mdt_object_put(info->mti_env, parent);
341 OBD_FREE(name, NAME_MAX + 1);
346 static int mdt_get_root(struct tgt_session_info *tsi)
348 struct mdt_thread_info *info = tsi2mdt_info(tsi);
349 struct mdt_device *mdt = info->mti_mdt;
350 struct mdt_body *repbody;
356 rc = mdt_check_ucred(info);
358 GOTO(out, rc = err_serious(rc));
360 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
361 GOTO(out, rc = err_serious(-ENOMEM));
363 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
364 if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
365 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
367 GOTO(out, rc = err_serious(-EFAULT));
369 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
371 GOTO(out, rc = err_serious(rc));
373 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
375 repbody->mbo_valid |= OBD_MD_FLID;
379 mdt_thread_info_fini(info);
383 static int mdt_statfs(struct tgt_session_info *tsi)
385 struct ptlrpc_request *req = tgt_ses_req(tsi);
386 struct mdt_thread_info *info = tsi2mdt_info(tsi);
387 struct md_device *next = info->mti_mdt->mdt_child;
388 struct ptlrpc_service_part *svcpt;
389 struct obd_statfs *osfs;
394 svcpt = req->rq_rqbd->rqbd_svcpt;
396 /* This will trigger a watchdog timeout */
397 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
398 (MDT_SERVICE_WATCHDOG_FACTOR *
399 at_get(&svcpt->scp_at_estimate)) + 1);
401 rc = mdt_check_ucred(info);
403 GOTO(out, rc = err_serious(rc));
405 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
406 GOTO(out, rc = err_serious(-ENOMEM));
408 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
410 GOTO(out, rc = -EPROTO);
412 /** statfs information are cached in the mdt_device */
413 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
414 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
415 /** statfs data is too old, get up-to-date one */
416 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
419 spin_lock(&info->mti_mdt->mdt_osfs_lock);
420 info->mti_mdt->mdt_osfs = *osfs;
421 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
422 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
424 /** use cached statfs data */
425 spin_lock(&info->mti_mdt->mdt_osfs_lock);
426 *osfs = info->mti_mdt->mdt_osfs;
427 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
431 mdt_counter_incr(req, LPROC_MDT_STATFS);
433 mdt_thread_info_fini(info);
437 #ifdef CONFIG_FS_POSIX_ACL
439 * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
441 * \param info thread info object
442 * \param repbody reply to pack ACLs into
443 * \param o mdt object of file to examine
444 * \param nodemap nodemap of client to reply to
446 * \retval -errno error getting or parsing ACL from disk
448 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
449 struct mdt_object *o, struct lu_nodemap *nodemap)
451 const struct lu_env *env = info->mti_env;
452 struct md_object *next = mdt_object_child(o);
453 struct lu_buf *buf = &info->mti_buf;
456 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
457 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
459 if (buf->lb_len == 0)
462 rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
464 if (rc == -ENODATA) {
465 repbody->mbo_aclsize = 0;
466 repbody->mbo_valid |= OBD_MD_FLACL;
468 } else if (rc == -EOPNOTSUPP) {
471 CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
472 mdt_obd_name(info->mti_mdt),
473 PFID(mdt_object_fid(o)), rc);
476 rc = nodemap_map_acl(nodemap, buf->lb_buf,
477 rc, NODEMAP_FS_TO_CLIENT);
478 /* if all ACLs mapped out, rc is still >= 0 */
480 CERROR("%s: nodemap_map_acl unable to parse "DFID
481 " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
482 PFID(mdt_object_fid(o)), rc);
484 repbody->mbo_aclsize = rc;
485 repbody->mbo_valid |= OBD_MD_FLACL;
493 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
494 const struct lu_attr *attr, const struct lu_fid *fid)
496 struct md_attr *ma = &info->mti_attr;
497 struct obd_export *exp = info->mti_exp;
498 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
500 LASSERT(ma->ma_valid & MA_INODE);
502 if (attr->la_valid & LA_ATIME) {
503 b->mbo_atime = attr->la_atime;
504 b->mbo_valid |= OBD_MD_FLATIME;
506 if (attr->la_valid & LA_MTIME) {
507 b->mbo_mtime = attr->la_mtime;
508 b->mbo_valid |= OBD_MD_FLMTIME;
510 if (attr->la_valid & LA_CTIME) {
511 b->mbo_ctime = attr->la_ctime;
512 b->mbo_valid |= OBD_MD_FLCTIME;
514 if (attr->la_valid & LA_FLAGS) {
515 b->mbo_flags = attr->la_flags;
516 b->mbo_valid |= OBD_MD_FLFLAGS;
518 if (attr->la_valid & LA_NLINK) {
519 b->mbo_nlink = attr->la_nlink;
520 b->mbo_valid |= OBD_MD_FLNLINK;
522 if (attr->la_valid & LA_UID) {
523 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
524 NODEMAP_FS_TO_CLIENT,
526 b->mbo_valid |= OBD_MD_FLUID;
528 if (attr->la_valid & LA_GID) {
529 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
530 NODEMAP_FS_TO_CLIENT,
532 b->mbo_valid |= OBD_MD_FLGID;
534 b->mbo_mode = attr->la_mode;
535 if (attr->la_valid & LA_MODE)
536 b->mbo_valid |= OBD_MD_FLMODE;
537 if (attr->la_valid & LA_TYPE)
538 b->mbo_valid |= OBD_MD_FLTYPE;
542 b->mbo_valid |= OBD_MD_FLID;
543 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
544 PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
548 mdt_body_reverse_idmap(info, b);
550 if (!(attr->la_valid & LA_TYPE))
553 b->mbo_rdev = attr->la_rdev;
554 b->mbo_size = attr->la_size;
555 b->mbo_blocks = attr->la_blocks;
557 if (!S_ISREG(attr->la_mode)) {
558 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
559 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
560 /* means no objects are allocated on osts. */
561 LASSERT(!(ma->ma_valid & MA_LOV));
562 /* just ignore blocks occupied by extend attributes on MDS */
564 /* if no object is allocated on osts, the size on mds is valid.
566 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
567 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
568 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
569 /* A released file stores its size on MDS. */
570 /* But return 1 block for released file, unless tools like tar
571 * will consider it fully sparse. (LU-3864)
573 if (unlikely(b->mbo_size == 0))
577 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
580 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
581 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
582 PFID(fid), (unsigned long long)b->mbo_size);
585 static inline int mdt_body_has_lov(const struct lu_attr *la,
586 const struct mdt_body *body)
588 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
589 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
592 void mdt_client_compatibility(struct mdt_thread_info *info)
594 struct mdt_body *body;
595 struct ptlrpc_request *req = mdt_info_req(info);
596 struct obd_export *exp = req->rq_export;
597 struct md_attr *ma = &info->mti_attr;
598 struct lu_attr *la = &ma->ma_attr;
601 if (exp_connect_layout(exp))
602 /* the client can deal with 16-bit lmm_stripe_count */
605 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
607 if (!mdt_body_has_lov(la, body))
610 /* now we have a reply with a lov for a client not compatible with the
611 * layout lock so we have to clean the layout generation number */
612 if (S_ISREG(la->la_mode))
613 ma->ma_lmm->lmm_layout_gen = 0;
617 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
618 struct mdt_object *o)
620 const struct lu_env *env = info->mti_env;
623 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
632 /* Is it a directory? Let's check for the LMV as well */
633 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
634 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
638 rc2 = mo_xattr_get(env, mdt_object_child(o),
640 XATTR_NAME_DEFAULT_LMV);
642 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
650 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
653 const struct lu_env *env = info->mti_env;
657 LASSERT(info->mti_big_lmm_used == 0);
658 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
662 /* big_lmm may need to be grown */
663 if (info->mti_big_lmmsize < rc) {
664 int size = size_roundup_power2(rc);
666 if (info->mti_big_lmmsize > 0) {
667 /* free old buffer */
668 LASSERT(info->mti_big_lmm);
669 OBD_FREE_LARGE(info->mti_big_lmm,
670 info->mti_big_lmmsize);
671 info->mti_big_lmm = NULL;
672 info->mti_big_lmmsize = 0;
675 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
676 if (info->mti_big_lmm == NULL)
678 info->mti_big_lmmsize = size;
680 LASSERT(info->mti_big_lmmsize >= rc);
682 info->mti_buf.lb_buf = info->mti_big_lmm;
683 info->mti_buf.lb_len = info->mti_big_lmmsize;
684 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
689 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
690 struct md_attr *ma, const char *name)
692 struct md_object *next = mdt_object_child(o);
693 struct lu_buf *buf = &info->mti_buf;
696 if (strcmp(name, XATTR_NAME_LOV) == 0) {
697 buf->lb_buf = ma->ma_lmm;
698 buf->lb_len = ma->ma_lmm_size;
699 LASSERT(!(ma->ma_valid & MA_LOV));
700 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
701 buf->lb_buf = ma->ma_lmv;
702 buf->lb_len = ma->ma_lmv_size;
703 LASSERT(!(ma->ma_valid & MA_LMV));
704 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
705 buf->lb_buf = ma->ma_lmv;
706 buf->lb_len = ma->ma_lmv_size;
707 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
712 rc = mo_xattr_get(info->mti_env, next, buf, name);
716 if (strcmp(name, XATTR_NAME_LOV) == 0) {
717 if (info->mti_big_lmm_used)
718 ma->ma_lmm = info->mti_big_lmm;
720 /* NOT return LOV EA with hole to old client. */
721 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
722 LOV_PATTERN_F_HOLE) &&
723 !(exp_connect_flags(info->mti_exp) &
724 OBD_CONNECT_LFSCK)) {
727 ma->ma_lmm_size = rc;
728 ma->ma_valid |= MA_LOV;
730 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
731 if (info->mti_big_lmm_used)
732 ma->ma_lmv = info->mti_big_lmm;
734 ma->ma_lmv_size = rc;
735 ma->ma_valid |= MA_LMV;
736 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
737 ma->ma_lmv_size = rc;
738 ma->ma_valid |= MA_LMV_DEF;
741 /* Update mdt_max_mdsize so all clients will be aware that */
742 if (info->mti_mdt->mdt_max_mdsize < rc)
743 info->mti_mdt->mdt_max_mdsize = rc;
746 } else if (rc == -ENODATA) {
749 } else if (rc == -ERANGE) {
750 /* Default LMV has fixed size, so it must be able to fit
751 * in the original buffer */
752 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
754 rc = mdt_big_xattr_get(info, o, name);
756 info->mti_big_lmm_used = 1;
764 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
765 struct mdt_object *o, struct lu_fid *pfid)
767 struct lu_buf *buf = &info->mti_buf;
768 struct link_ea_header *leh;
769 struct link_ea_entry *lee;
773 buf->lb_buf = info->mti_big_lmm;
774 buf->lb_len = info->mti_big_lmmsize;
775 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
776 buf, XATTR_NAME_LINK);
777 /* ignore errors, MA_PFID won't be set and it is
778 * up to the caller to treat this as an error */
779 if (rc == -ERANGE || buf->lb_len == 0) {
780 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
781 buf->lb_buf = info->mti_big_lmm;
782 buf->lb_len = info->mti_big_lmmsize;
787 if (rc < sizeof(*leh)) {
788 CERROR("short LinkEA on "DFID": rc = %d\n",
789 PFID(mdt_object_fid(o)), rc);
793 leh = (struct link_ea_header *) buf->lb_buf;
794 lee = (struct link_ea_entry *)(leh + 1);
795 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
796 leh->leh_magic = LINK_EA_MAGIC;
797 leh->leh_reccount = __swab32(leh->leh_reccount);
798 leh->leh_len = __swab64(leh->leh_len);
800 if (leh->leh_magic != LINK_EA_MAGIC)
802 if (leh->leh_reccount == 0)
805 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
806 fid_be_to_cpu(pfid, pfid);
811 int mdt_attr_get_complex(struct mdt_thread_info *info,
812 struct mdt_object *o, struct md_attr *ma)
814 const struct lu_env *env = info->mti_env;
815 struct md_object *next = mdt_object_child(o);
816 struct lu_buf *buf = &info->mti_buf;
817 int need = ma->ma_need;
824 if (mdt_object_exists(o) == 0)
825 GOTO(out, rc = -ENOENT);
826 mode = lu_object_attr(&next->mo_lu);
828 if (need & MA_INODE) {
829 ma->ma_need = MA_INODE;
830 rc = mo_attr_get(env, next, ma);
833 ma->ma_valid |= MA_INODE;
836 if (need & MA_PFID) {
837 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
839 ma->ma_valid |= MA_PFID;
840 /* ignore this error, parent fid is not mandatory */
844 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
845 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
850 if (need & MA_LMV && S_ISDIR(mode)) {
851 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
856 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
857 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
862 if (need & MA_HSM && S_ISREG(mode)) {
863 buf->lb_buf = info->mti_xattr_buf;
864 buf->lb_len = sizeof(info->mti_xattr_buf);
865 CLASSERT(sizeof(struct hsm_attrs) <=
866 sizeof(info->mti_xattr_buf));
867 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
868 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
870 ma->ma_valid |= MA_HSM;
871 else if (rc2 < 0 && rc2 != -ENODATA)
875 #ifdef CONFIG_FS_POSIX_ACL
876 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
877 buf->lb_buf = ma->ma_acl;
878 buf->lb_len = ma->ma_acl_size;
879 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
881 ma->ma_acl_size = rc2;
882 ma->ma_valid |= MA_ACL_DEF;
883 } else if (rc2 == -ENODATA) {
892 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
893 rc, ma->ma_valid, ma->ma_lmm);
897 static int mdt_getattr_internal(struct mdt_thread_info *info,
898 struct mdt_object *o, int ma_need)
900 struct md_object *next = mdt_object_child(o);
901 const struct mdt_body *reqbody = info->mti_body;
902 struct ptlrpc_request *req = mdt_info_req(info);
903 struct md_attr *ma = &info->mti_attr;
904 struct lu_attr *la = &ma->ma_attr;
905 struct req_capsule *pill = info->mti_pill;
906 const struct lu_env *env = info->mti_env;
907 struct mdt_body *repbody;
908 struct lu_buf *buffer = &info->mti_buf;
909 struct obd_export *exp = info->mti_exp;
910 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
915 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
916 RETURN(err_serious(-ENOMEM));
918 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
922 if (mdt_object_remote(o)) {
923 /* This object is located on remote node.*/
924 /* Return -ENOTSUPP for old client */
925 if (!mdt_is_dne_client(req->rq_export))
926 GOTO(out, rc = -ENOTSUPP);
928 repbody->mbo_fid1 = *mdt_object_fid(o);
929 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
933 if (reqbody->mbo_eadatasize > 0) {
934 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
935 if (buffer->lb_buf == NULL)
936 GOTO(out, rc = -EPROTO);
937 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
940 buffer->lb_buf = NULL;
942 ma_need &= ~(MA_LOV | MA_LMV);
943 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
944 mdt_obd_name(info->mti_mdt),
945 req->rq_export->exp_client_uuid.uuid);
948 /* If it is dir object and client require MEA, then we got MEA */
949 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
950 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
951 /* Assumption: MDT_MD size is enough for lmv size. */
952 ma->ma_lmv = buffer->lb_buf;
953 ma->ma_lmv_size = buffer->lb_len;
954 ma->ma_need = MA_INODE;
955 if (ma->ma_lmv_size > 0) {
956 if (reqbody->mbo_valid & OBD_MD_MEA)
957 ma->ma_need |= MA_LMV;
958 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
959 ma->ma_need |= MA_LMV_DEF;
962 ma->ma_lmm = buffer->lb_buf;
963 ma->ma_lmm_size = buffer->lb_len;
964 ma->ma_need = MA_INODE | MA_HSM;
965 if (ma->ma_lmm_size > 0)
966 ma->ma_need |= MA_LOV;
969 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
970 reqbody->mbo_valid & OBD_MD_FLDIREA &&
971 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
972 /* get default stripe info for this dir. */
973 ma->ma_need |= MA_LOV_DEF;
975 ma->ma_need |= ma_need;
977 rc = mdt_attr_get_complex(info, o, ma);
979 CERROR("%s: getattr error for "DFID": rc = %d\n",
980 mdt_obd_name(info->mti_mdt),
981 PFID(mdt_object_fid(o)), rc);
985 /* if file is released, check if a restore is running */
986 if (ma->ma_valid & MA_HSM) {
987 repbody->mbo_valid |= OBD_MD_TSTATE;
988 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
989 mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
990 repbody->mbo_t_state = MS_RESTORE;
993 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
995 /* the Lustre protocol supposes to return default striping
996 * on the user-visible root if explicitly requested */
997 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
998 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
999 struct lu_fid rootfid;
1000 struct mdt_object *root;
1001 struct mdt_device *mdt = info->mti_mdt;
1003 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1006 root = mdt_object_find(env, mdt, &rootfid);
1008 RETURN(PTR_ERR(root));
1009 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1010 mdt_object_put(info->mti_env, root);
1012 CERROR("%s: getattr error for "DFID": rc = %d\n",
1013 mdt_obd_name(info->mti_mdt),
1014 PFID(mdt_object_fid(o)), rc);
1019 if (likely(ma->ma_valid & MA_INODE))
1020 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1024 if (mdt_body_has_lov(la, reqbody)) {
1025 if (ma->ma_valid & MA_LOV) {
1026 LASSERT(ma->ma_lmm_size);
1027 repbody->mbo_eadatasize = ma->ma_lmm_size;
1028 if (S_ISDIR(la->la_mode))
1029 repbody->mbo_valid |= OBD_MD_FLDIREA;
1031 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1032 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1034 if (ma->ma_valid & MA_LMV) {
1035 /* Return -ENOTSUPP for old client */
1036 if (!mdt_is_striped_client(req->rq_export))
1039 LASSERT(S_ISDIR(la->la_mode));
1040 mdt_dump_lmv(D_INFO, ma->ma_lmv);
1041 repbody->mbo_eadatasize = ma->ma_lmv_size;
1042 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1044 if (ma->ma_valid & MA_LMV_DEF) {
1045 /* Return -ENOTSUPP for old client */
1046 if (!mdt_is_striped_client(req->rq_export))
1048 LASSERT(S_ISDIR(la->la_mode));
1049 repbody->mbo_eadatasize = ma->ma_lmv_size;
1050 repbody->mbo_valid |= (OBD_MD_FLDIREA |
1051 OBD_MD_DEFAULT_MEA);
1053 } else if (S_ISLNK(la->la_mode) &&
1054 reqbody->mbo_valid & OBD_MD_LINKNAME) {
1055 buffer->lb_buf = ma->ma_lmm;
1056 /* eadatasize from client includes NULL-terminator, so
1057 * there is no need to read it */
1058 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1059 rc = mo_readlink(env, next, buffer);
1060 if (unlikely(rc <= 0)) {
1061 CERROR("%s: readlink failed for "DFID": rc = %d\n",
1062 mdt_obd_name(info->mti_mdt),
1063 PFID(mdt_object_fid(o)), rc);
1066 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1068 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1070 repbody->mbo_valid |= OBD_MD_LINKNAME;
1071 /* we need to report back size with NULL-terminator
1072 * because client expects that */
1073 repbody->mbo_eadatasize = rc + 1;
1074 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1075 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1076 "on "DFID ", expected %d\n",
1077 mdt_obd_name(info->mti_mdt),
1078 rc, PFID(mdt_object_fid(o)),
1079 reqbody->mbo_eadatasize - 1);
1080 /* NULL terminate */
1081 ((char *)ma->ma_lmm)[rc] = 0;
1083 /* If the total CDEBUG() size is larger than a page, it
1084 * will print a warning to the console, avoid this by
1085 * printing just the last part of the symlink. */
1086 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1087 print_limit < rc ? "..." : "", print_limit,
1088 (char *)ma->ma_lmm + rc - print_limit, rc);
1093 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1094 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1095 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1096 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1097 repbody->mbo_max_mdsize);
1100 if (exp_connect_rmtclient(info->mti_exp) &&
1101 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1102 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1104 /* mdt_getattr_lock only */
1105 rc = mdt_pack_remote_perm(info, o, buf);
1107 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1108 repbody->mbo_aclsize = 0;
1111 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1112 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1115 #ifdef CONFIG_FS_POSIX_ACL
1116 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1117 (reqbody->mbo_valid & OBD_MD_FLACL))
1118 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1123 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1128 static int mdt_getattr(struct tgt_session_info *tsi)
1130 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1131 struct mdt_object *obj = info->mti_object;
1132 struct req_capsule *pill = info->mti_pill;
1133 struct mdt_body *reqbody;
1134 struct mdt_body *repbody;
1138 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1140 LASSERT(obj != NULL);
1141 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1143 /* Unlike intent case where we need to pre-fill out buffers early on
1144 * in intent policy for ldlm reasons, here we can have a much better
1145 * guess at EA size by just reading it from disk.
1146 * Exceptions are readdir and (missing) directory striping */
1148 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1149 /* No easy way to know how long is the symlink, but it cannot
1150 * be more than PATH_MAX, so we allocate +1 */
1153 /* A special case for fs ROOT: getattr there might fetch
1154 * default EA for entire fs, not just for this dir!
1156 } else if (lu_fid_eq(mdt_object_fid(obj),
1157 &info->mti_mdt->mdt_md_root_fid) &&
1158 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1159 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1161 /* Should the default strping be bigger, mdt_fix_reply
1162 * will reallocate */
1163 rc = DEF_REP_MD_SIZE;
1165 /* Read the actual EA size from disk */
1166 rc = mdt_attr_get_eabuf_size(info, obj);
1170 GOTO(out, rc = err_serious(rc));
1172 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1174 rc = req_capsule_server_pack(pill);
1175 if (unlikely(rc != 0))
1176 GOTO(out, rc = err_serious(rc));
1178 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1179 LASSERT(repbody != NULL);
1180 repbody->mbo_eadatasize = 0;
1181 repbody->mbo_aclsize = 0;
1183 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1184 rc = mdt_init_ucred(info, reqbody);
1186 rc = mdt_check_ucred(info);
1188 GOTO(out_shrink, rc);
1190 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1192 rc = mdt_getattr_internal(info, obj, 0);
1193 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1194 mdt_exit_ucred(info);
1197 mdt_client_compatibility(info);
1198 rc2 = mdt_fix_reply(info);
1202 mdt_thread_info_fini(info);
1207 * Exchange MOF_LOV_CREATED flags between two objects after a
1208 * layout swap. No assumption is made on whether o1 or o2 have
1209 * created objects or not.
1211 * \param[in,out] o1 First swap layout object
1212 * \param[in,out] o2 Second swap layout object
1214 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1218 mutex_lock(&o1->mot_lov_mutex);
1219 mutex_lock(&o2->mot_lov_mutex);
1221 o1_flags = o1->mot_flags;
1222 o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1223 (o2->mot_flags & MOF_LOV_CREATED);
1225 o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1226 (o1_flags & MOF_LOV_CREATED);
1228 mutex_unlock(&o2->mot_lov_mutex);
1229 mutex_unlock(&o1->mot_lov_mutex);
1232 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1234 struct mdt_thread_info *info;
1235 struct ptlrpc_request *req = tgt_ses_req(tsi);
1236 struct obd_export *exp = req->rq_export;
1237 struct mdt_object *o1, *o2, *o;
1238 struct mdt_lock_handle *lh1, *lh2;
1239 struct mdc_swap_layouts *msl;
1243 /* client does not support layout lock, so layout swaping
1245 * FIXME: there is a problem for old clients which don't support
1246 * layout lock yet. If those clients have already opened the file
1247 * they won't be notified at all so that old layout may still be
1248 * used to do IO. This can be fixed after file release is landed by
1249 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1250 if (!exp_connect_layout(exp))
1251 RETURN(-EOPNOTSUPP);
1253 info = tsi2mdt_info(tsi);
1255 if (info->mti_dlm_req != NULL)
1256 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1258 o1 = info->mti_object;
1259 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1260 &info->mti_body->mbo_fid2);
1262 GOTO(out, rc = PTR_ERR(o));
1264 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1265 GOTO(put, rc = -ENOENT);
1267 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1268 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1274 /* permission check. Make sure the calling process having permission
1275 * to write both files. */
1276 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1281 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1286 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1288 GOTO(put, rc = -EPROTO);
1290 lh1 = &info->mti_lh[MDT_LH_NEW];
1291 mdt_lock_reg_init(lh1, LCK_EX);
1292 lh2 = &info->mti_lh[MDT_LH_OLD];
1293 mdt_lock_reg_init(lh2, LCK_EX);
1295 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1296 MDS_INODELOCK_XATTR);
1300 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1301 MDS_INODELOCK_XATTR);
1305 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1306 mdt_object_child(o2), msl->msl_flags);
1310 mdt_swap_lov_flag(o1, o2);
1313 mdt_object_unlock(info, o2, lh2, rc);
1315 mdt_object_unlock(info, o1, lh1, rc);
1317 mdt_object_put(info->mti_env, o);
1319 mdt_thread_info_fini(info);
1323 static int mdt_raw_lookup(struct mdt_thread_info *info,
1324 struct mdt_object *parent,
1325 const struct lu_name *lname,
1326 struct ldlm_reply *ldlm_rep)
1328 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1332 LASSERT(!info->mti_cross_ref);
1334 /* Only got the fid of this obj by name */
1335 fid_zero(child_fid);
1336 rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1337 lname, child_fid, &info->mti_spec);
1339 struct mdt_body *repbody;
1341 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1342 repbody->mbo_fid1 = *child_fid;
1343 repbody->mbo_valid = OBD_MD_FLID;
1344 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1345 } else if (rc == -ENOENT) {
1346 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1353 * UPDATE lock should be taken against parent, and be released before exit;
1354 * child_bits lock should be taken against child, and be returned back:
1355 * (1)normal request should release the child lock;
1356 * (2)intent request will grant the lock to client.
1358 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1359 struct mdt_lock_handle *lhc,
1361 struct ldlm_reply *ldlm_rep)
1363 struct ptlrpc_request *req = mdt_info_req(info);
1364 struct mdt_body *reqbody = NULL;
1365 struct mdt_object *parent = info->mti_object;
1366 struct mdt_object *child;
1367 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1368 struct lu_name *lname = NULL;
1369 struct mdt_lock_handle *lhp = NULL;
1370 struct ldlm_lock *lock;
1377 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1378 LASSERT(ergo(is_resent,
1379 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1384 if (info->mti_cross_ref) {
1385 /* Only getattr on the child. Parent is on another node. */
1386 mdt_set_disposition(info, ldlm_rep,
1387 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1389 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1391 PFID(mdt_object_fid(child)), ldlm_rep);
1393 rc = mdt_check_resent_lock(info, child, lhc);
1396 } else if (rc > 0) {
1397 mdt_lock_handle_init(lhc);
1398 mdt_lock_reg_init(lhc, LCK_PR);
1401 * Object's name is on another MDS, no lookup or layout
1402 * lock is needed here but update lock is.
1404 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1405 MDS_INODELOCK_LAYOUT);
1406 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1408 rc = mdt_object_lock(info, child, lhc, child_bits);
1413 /* Finally, we can get attr for child. */
1414 if (!mdt_object_exists(child)) {
1415 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1417 "remote object doesn't exist.\n");
1418 mdt_object_unlock(info, child, lhc, 1);
1422 rc = mdt_getattr_internal(info, child, 0);
1423 if (unlikely(rc != 0))
1424 mdt_object_unlock(info, child, lhc, 1);
1429 lname = &info->mti_name;
1430 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1432 if (lu_name_is_valid(lname)) {
1433 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1434 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1435 PNAME(lname), ldlm_rep);
1437 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1438 if (unlikely(reqbody == NULL))
1439 RETURN(err_serious(-EPROTO));
1441 *child_fid = reqbody->mbo_fid2;
1443 if (unlikely(!fid_is_sane(child_fid)))
1444 RETURN(err_serious(-EINVAL));
1446 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1448 PFID(mdt_object_fid(parent)),
1449 PFID(&reqbody->mbo_fid2), ldlm_rep);
1452 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1454 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1455 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1457 "Parent doesn't exist!\n");
1461 if (mdt_object_remote(parent)) {
1462 CERROR("%s: parent "DFID" is on remote target\n",
1463 mdt_obd_name(info->mti_mdt),
1464 PFID(mdt_object_fid(parent)));
1468 if (lu_name_is_valid(lname)) {
1469 /* Always allow to lookup ".." */
1470 if (unlikely(lname->ln_namelen == 2 &&
1471 lname->ln_name[0] == '.' &&
1472 lname->ln_name[1] == '.'))
1473 info->mti_spec.sp_permitted = 1;
1475 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1476 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1481 /* step 1: lock parent only if parent is a directory */
1482 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1483 lhp = &info->mti_lh[MDT_LH_PARENT];
1484 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1485 rc = mdt_object_lock(info, parent, lhp,
1486 MDS_INODELOCK_UPDATE);
1487 if (unlikely(rc != 0))
1491 /* step 2: lookup child's fid by name */
1492 fid_zero(child_fid);
1493 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1494 child_fid, &info->mti_spec);
1496 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1499 GOTO(out_parent, rc);
1502 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1505 *step 3: find the child object by fid & lock it.
1506 * regardless if it is local or remote.
1508 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1509 * set parent dir fid the same as child fid in getattr by fid case
1510 * we should not lu_object_find() the object again, could lead
1511 * to hung if there is a concurrent unlink destroyed the object.
1513 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1514 mdt_object_get(info->mti_env, parent);
1517 child = mdt_object_find(info->mti_env, info->mti_mdt,
1521 if (unlikely(IS_ERR(child)))
1522 GOTO(out_parent, rc = PTR_ERR(child));
1524 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1525 if (!mdt_object_exists(child)) {
1526 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1528 "Object doesn't exist!\n");
1529 GOTO(out_child, rc = -ENOENT);
1532 rc = mdt_check_resent_lock(info, child, lhc);
1534 GOTO(out_child, rc);
1535 } else if (rc > 0) {
1536 mdt_lock_handle_init(lhc);
1537 mdt_lock_reg_init(lhc, LCK_PR);
1540 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1541 mdt_object_exists(child) && !mdt_object_remote(child)) {
1542 struct md_attr *ma = &info->mti_attr;
1545 ma->ma_need = MA_INODE;
1546 rc = mdt_attr_get_complex(info, child, ma);
1547 if (unlikely(rc != 0))
1548 GOTO(out_child, rc);
1550 /* If the file has not been changed for some time, we
1551 * return not only a LOOKUP lock, but also an UPDATE
1552 * lock and this might save us RPC on later STAT. For
1553 * directories, it also let negative dentry cache start
1554 * working for this dir. */
1555 if (ma->ma_valid & MA_INODE &&
1556 ma->ma_attr.la_valid & LA_CTIME &&
1557 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1558 ma->ma_attr.la_ctime < cfs_time_current_sec())
1559 child_bits |= MDS_INODELOCK_UPDATE;
1562 /* layout lock must be granted in a best-effort way
1563 * for IT operations */
1564 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1565 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1566 exp_connect_layout(info->mti_exp) &&
1567 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1568 !mdt_object_remote(child) && ldlm_rep != NULL) {
1569 /* try to grant layout lock for regular file. */
1575 child_bits |= MDS_INODELOCK_LAYOUT;
1576 /* try layout lock, it may fail to be granted due to
1577 * contention at LOOKUP or UPDATE */
1578 if (!mdt_object_lock_try(info, child, lhc,
1580 child_bits &= ~MDS_INODELOCK_LAYOUT;
1581 LASSERT(child_bits != 0);
1582 rc = mdt_object_lock(info, child, lhc,
1588 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1589 * client will enqueue the lock to the remote MDT */
1590 if (mdt_object_remote(child))
1591 child_bits &= ~MDS_INODELOCK_UPDATE;
1592 rc = mdt_object_lock(info, child, lhc, child_bits);
1594 if (unlikely(rc != 0))
1595 GOTO(out_child, rc);
1598 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1600 /* finally, we can get attr for child. */
1601 rc = mdt_getattr_internal(info, child, ma_need);
1602 if (unlikely(rc != 0)) {
1603 mdt_object_unlock(info, child, lhc, 1);
1605 /* Debugging code. */
1606 LDLM_DEBUG(lock, "Returning lock to client");
1607 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1608 &lock->l_resource->lr_name),
1609 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1610 PLDLMRES(lock->l_resource),
1611 PFID(mdt_object_fid(child)));
1614 LDLM_LOCK_PUT(lock);
1618 mdt_object_put(info->mti_env, child);
1621 mdt_object_unlock(info, parent, lhp, 1);
1625 /* normal handler: should release the child lock */
1626 static int mdt_getattr_name(struct tgt_session_info *tsi)
1628 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1629 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1630 struct mdt_body *reqbody;
1631 struct mdt_body *repbody;
1635 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1636 LASSERT(reqbody != NULL);
1637 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1638 LASSERT(repbody != NULL);
1640 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1641 repbody->mbo_eadatasize = 0;
1642 repbody->mbo_aclsize = 0;
1644 rc = mdt_init_ucred_intent_getattr(info, reqbody);
1646 GOTO(out_shrink, rc);
1648 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1649 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1650 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1651 lhc->mlh_reg_lh.cookie = 0;
1653 mdt_exit_ucred(info);
1656 mdt_client_compatibility(info);
1657 rc2 = mdt_fix_reply(info);
1660 mdt_thread_info_fini(info);
1664 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1665 void *karg, void __user *uarg);
1667 static int mdt_set_info(struct tgt_session_info *tsi)
1669 struct ptlrpc_request *req = tgt_ses_req(tsi);
1672 int keylen, vallen, rc = 0;
1676 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1678 DEBUG_REQ(D_HA, req, "no set_info key");
1679 RETURN(err_serious(-EFAULT));
1682 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1685 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1687 DEBUG_REQ(D_HA, req, "no set_info val");
1688 RETURN(err_serious(-EFAULT));
1691 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1694 /* Swab any part of val you need to here */
1695 if (KEY_IS(KEY_READ_ONLY)) {
1696 spin_lock(&req->rq_export->exp_lock);
1698 *exp_connect_flags_ptr(req->rq_export) |=
1701 *exp_connect_flags_ptr(req->rq_export) &=
1702 ~OBD_CONNECT_RDONLY;
1703 spin_unlock(&req->rq_export->exp_lock);
1704 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1705 struct changelog_setinfo *cs = val;
1707 if (vallen != sizeof(*cs)) {
1708 CERROR("%s: bad changelog_clear setinfo size %d\n",
1709 tgt_name(tsi->tsi_tgt), vallen);
1712 if (ptlrpc_req_need_swab(req)) {
1713 __swab64s(&cs->cs_recno);
1714 __swab32s(&cs->cs_id);
1717 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1719 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1721 obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1728 static int mdt_readpage(struct tgt_session_info *tsi)
1730 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1731 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1732 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1733 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1734 struct mdt_body *repbody;
1740 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1741 RETURN(err_serious(-ENOMEM));
1743 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1744 if (repbody == NULL || reqbody == NULL)
1745 RETURN(err_serious(-EFAULT));
1748 * prepare @rdpg before calling lower layers and transfer itself. Here
1749 * reqbody->size contains offset of where to start to read and
1750 * reqbody->nlink contains number bytes to read.
1752 rdpg->rp_hash = reqbody->mbo_size;
1753 if (rdpg->rp_hash != reqbody->mbo_size) {
1754 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1755 rdpg->rp_hash, reqbody->mbo_size);
1759 rdpg->rp_attrs = reqbody->mbo_mode;
1760 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1761 rdpg->rp_attrs |= LUDA_64BITHASH;
1762 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1763 exp_max_brw_size(tsi->tsi_exp));
1764 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1766 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1767 if (rdpg->rp_pages == NULL)
1770 for (i = 0; i < rdpg->rp_npages; ++i) {
1771 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1772 if (rdpg->rp_pages[i] == NULL)
1773 GOTO(free_rdpg, rc = -ENOMEM);
1776 /* call lower layers to fill allocated pages with directory data */
1777 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1779 GOTO(free_rdpg, rc);
1781 /* send pages to client */
1782 rc = tgt_sendpage(tsi, rdpg, rc);
1787 for (i = 0; i < rdpg->rp_npages; i++)
1788 if (rdpg->rp_pages[i] != NULL)
1789 __free_page(rdpg->rp_pages[i]);
1790 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1792 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1798 static int mdt_reint_internal(struct mdt_thread_info *info,
1799 struct mdt_lock_handle *lhc,
1802 struct req_capsule *pill = info->mti_pill;
1803 struct mdt_body *repbody;
1808 rc = mdt_reint_unpack(info, op);
1810 CERROR("Can't unpack reint, rc %d\n", rc);
1811 RETURN(err_serious(rc));
1814 /* for replay (no_create) lmm is not needed, client has it already */
1815 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1816 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1819 /* llog cookies are always 0, the field is kept for compatibility */
1820 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1821 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1823 rc = req_capsule_server_pack(pill);
1825 CERROR("Can't pack response, rc %d\n", rc);
1826 RETURN(err_serious(rc));
1829 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1830 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1832 repbody->mbo_eadatasize = 0;
1833 repbody->mbo_aclsize = 0;
1836 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1838 /* for replay no cookkie / lmm need, because client have this already */
1839 if (info->mti_spec.no_create)
1840 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1841 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1843 rc = mdt_init_ucred_reint(info);
1845 GOTO(out_shrink, rc);
1847 rc = mdt_fix_attr_ucred(info, op);
1849 GOTO(out_ucred, rc = err_serious(rc));
1851 rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1853 GOTO(out_ucred, rc);
1854 } else if (rc == 1) {
1855 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1856 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1857 GOTO(out_ucred, rc);
1859 rc = mdt_reint_rec(info, lhc);
1862 mdt_exit_ucred(info);
1864 mdt_client_compatibility(info);
1865 rc2 = mdt_fix_reply(info);
1871 static long mdt_reint_opcode(struct ptlrpc_request *req,
1872 const struct req_format **fmt)
1874 struct mdt_device *mdt;
1875 struct mdt_rec_reint *rec;
1878 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1880 opc = rec->rr_opcode;
1881 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1882 if (opc < REINT_MAX && fmt[opc] != NULL)
1883 req_capsule_extend(&req->rq_pill, fmt[opc]);
1885 mdt = mdt_exp2dev(req->rq_export);
1886 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1887 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1888 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1889 opc = err_serious(-EFAULT);
1892 opc = err_serious(-EFAULT);
1897 static int mdt_reint(struct tgt_session_info *tsi)
1901 static const struct req_format *reint_fmts[REINT_MAX] = {
1902 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1903 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1904 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1905 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1906 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1907 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1908 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1909 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1910 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1915 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1917 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1919 * No lock possible here from client to pass it to reint code
1922 rc = mdt_reint_internal(info, NULL, opc);
1923 mdt_thread_info_fini(info);
1928 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1932 /* this should sync the whole device */
1933 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1935 struct dt_device *dt = mdt->mdt_bottom;
1939 rc = dt->dd_ops->dt_sync(env, dt);
1943 /* this should sync this object */
1944 static int mdt_object_sync(struct mdt_thread_info *info)
1946 struct md_object *next;
1950 if (!mdt_object_exists(info->mti_object)) {
1951 CWARN("%s: non existing object "DFID": rc = %d\n",
1952 mdt_obd_name(info->mti_mdt),
1953 PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1956 next = mdt_object_child(info->mti_object);
1957 rc = mo_object_sync(info->mti_env, next);
1962 static int mdt_sync(struct tgt_session_info *tsi)
1964 struct ptlrpc_request *req = tgt_ses_req(tsi);
1965 struct req_capsule *pill = tsi->tsi_pill;
1966 struct mdt_body *body;
1971 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1972 RETURN(err_serious(-ENOMEM));
1974 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1975 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1977 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1979 /* sync an object */
1980 rc = mdt_object_sync(info);
1982 const struct lu_fid *fid;
1983 struct lu_attr *la = &info->mti_attr.ma_attr;
1985 info->mti_attr.ma_need = MA_INODE;
1986 info->mti_attr.ma_valid = 0;
1987 rc = mdt_attr_get_complex(info, info->mti_object,
1990 body = req_capsule_server_get(pill,
1992 fid = mdt_object_fid(info->mti_object);
1993 mdt_pack_attr2body(info, body, la, fid);
1996 mdt_thread_info_fini(info);
1999 mdt_counter_incr(req, LPROC_MDT_SYNC);
2005 * Handle quota control requests to consult current usage/limit, but also
2006 * to configure quota enforcement
2008 static int mdt_quotactl(struct tgt_session_info *tsi)
2010 struct obd_export *exp = tsi->tsi_exp;
2011 struct req_capsule *pill = tsi->tsi_pill;
2012 struct obd_quotactl *oqctl, *repoqc;
2014 struct mdt_device *mdt = mdt_exp2dev(exp);
2015 struct lu_device *qmt = mdt->mdt_qmt_dev;
2016 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
2019 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2021 RETURN(err_serious(-EPROTO));
2023 rc = req_capsule_server_pack(pill);
2025 RETURN(err_serious(rc));
2027 switch (oqctl->qc_cmd) {
2028 /* master quotactl */
2031 if (!nodemap_can_setquota(nodemap))
2036 RETURN(-EOPNOTSUPP);
2037 /* slave quotactl */
2042 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2046 /* map uid/gid for remote client */
2048 if (exp_connect_rmtclient(exp)) {
2049 struct lustre_idmap_table *idmap;
2051 idmap = exp->exp_mdt_data.med_idmap;
2053 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2054 oqctl->qc_cmd != Q_GETINFO))
2057 if (oqctl->qc_type == USRQUOTA)
2058 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2060 else if (oqctl->qc_type == GRPQUOTA)
2061 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2066 if (id == CFS_IDMAP_NOTFOUND) {
2067 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2072 if (oqctl->qc_type == USRQUOTA)
2073 id = nodemap_map_id(nodemap, NODEMAP_UID,
2074 NODEMAP_CLIENT_TO_FS, id);
2075 else if (oqctl->qc_type == GRPQUOTA)
2076 id = nodemap_map_id(nodemap, NODEMAP_UID,
2077 NODEMAP_CLIENT_TO_FS, id);
2079 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2081 RETURN(err_serious(-EFAULT));
2083 if (oqctl->qc_id != id)
2084 swap(oqctl->qc_id, id);
2086 switch (oqctl->qc_cmd) {
2092 /* forward quotactl request to QMT */
2093 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2098 /* slave quotactl */
2099 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2104 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2108 if (oqctl->qc_id != id)
2109 swap(oqctl->qc_id, id);
2115 /** clone llog ctxt from child (mdd)
2116 * This allows remote llog (replicator) access.
2117 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2118 * context was originally set up, or we can handle them directly.
2119 * I choose the latter, but that means I need any llog
2120 * contexts set up by child to be accessable by the mdt. So we clone the
2121 * context into our context list here.
2123 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2126 struct md_device *next = mdt->mdt_child;
2127 struct llog_ctxt *ctxt;
2130 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2133 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2134 if (rc || ctxt == NULL) {
2138 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2140 CERROR("Can't set mdt ctxt %d\n", rc);
2145 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2146 struct mdt_device *mdt, int idx)
2148 struct llog_ctxt *ctxt;
2150 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2153 /* Put once for the get we just did, and once for the clone */
2154 llog_ctxt_put(ctxt);
2155 llog_ctxt_put(ctxt);
2160 * sec context handlers
2162 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2166 rc = mdt_handle_idmap(tsi);
2168 struct ptlrpc_request *req = tgt_ses_req(tsi);
2171 opc = lustre_msg_get_opc(req->rq_reqmsg);
2172 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2173 sptlrpc_svc_ctx_invalidate(req);
2176 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2182 * quota request handlers
2184 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2186 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2187 struct lu_device *qmt = mdt->mdt_qmt_dev;
2192 RETURN(err_serious(-EOPNOTSUPP));
2194 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2198 struct mdt_object *mdt_object_new(const struct lu_env *env,
2199 struct mdt_device *d,
2200 const struct lu_fid *f)
2202 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2203 struct lu_object *o;
2204 struct mdt_object *m;
2207 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2208 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2209 if (unlikely(IS_ERR(o)))
2210 m = (struct mdt_object *)o;
2216 struct mdt_object *mdt_object_find(const struct lu_env *env,
2217 struct mdt_device *d,
2218 const struct lu_fid *f)
2220 struct lu_object *o;
2221 struct mdt_object *m;
2224 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2225 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2226 if (unlikely(IS_ERR(o)))
2227 m = (struct mdt_object *)o;
2235 * Asyncronous commit for mdt device.
2237 * Pass asynchonous commit call down the MDS stack.
2239 * \param env environment
2240 * \param mdt the mdt device
2242 static void mdt_device_commit_async(const struct lu_env *env,
2243 struct mdt_device *mdt)
2245 struct dt_device *dt = mdt->mdt_bottom;
2249 rc = dt->dd_ops->dt_commit_async(env, dt);
2250 if (unlikely(rc != 0))
2251 CWARN("%s: async commit start failed: rc = %d\n",
2252 mdt_obd_name(mdt), rc);
2253 atomic_inc(&mdt->mdt_async_commit_count);
2258 * Mark the lock as "synchonous".
2260 * Mark the lock to deffer transaction commit to the unlock time.
2262 * \param lock the lock to mark as "synchonous"
2264 * \see mdt_is_lock_sync
2265 * \see mdt_save_lock
2267 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2269 lock->l_ast_data = (void*)1;
2273 * Check whehter the lock "synchonous" or not.
2275 * \param lock the lock to check
2276 * \retval 1 the lock is "synchonous"
2277 * \retval 0 the lock isn't "synchronous"
2279 * \see mdt_set_lock_sync
2280 * \see mdt_save_lock
2282 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2284 return lock->l_ast_data != NULL;
2288 * Blocking AST for mdt locks.
2290 * Starts transaction commit if in case of COS lock conflict or
2291 * deffers such a commit to the mdt_save_lock.
2293 * \param lock the lock which blocks a request or cancelling lock
2294 * \param desc unused
2295 * \param data unused
2296 * \param flag indicates whether this cancelling or blocking callback
2298 * \see ldlm_blocking_ast_nocheck
2300 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2301 void *data, int flag)
2303 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2304 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2308 if (flag == LDLM_CB_CANCELING)
2311 lock_res_and_lock(lock);
2312 if (lock->l_blocking_ast != mdt_blocking_ast) {
2313 unlock_res_and_lock(lock);
2316 if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2317 lock->l_blocking_lock != NULL) {
2318 if (mdt_cos_is_enabled(mdt) &&
2319 lock->l_client_cookie !=
2320 lock->l_blocking_lock->l_client_cookie)
2321 mdt_set_lock_sync(lock);
2322 else if (mdt_slc_is_enabled(mdt) &&
2323 ldlm_is_cos_incompat(lock->l_blocking_lock))
2324 mdt_set_lock_sync(lock);
2326 rc = ldlm_blocking_ast_nocheck(lock);
2328 /* There is no lock conflict if l_blocking_lock == NULL,
2329 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2330 * when the last reference to a local lock was released */
2331 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2334 rc = lu_env_init(&env, LCT_LOCAL);
2335 if (unlikely(rc != 0))
2336 CWARN("%s: lu_env initialization failed, cannot "
2337 "start asynchronous commit: rc = %d\n",
2340 mdt_device_commit_async(&env, mdt);
2347 * Blocking AST for cross-MDT lock
2349 * Discard lock from uncommitted_slc_locks and cancel it.
2351 * \param lock the lock which blocks a request or cancelling lock
2352 * \param desc unused
2353 * \param data unused
2354 * \param flag indicates whether this cancelling or blocking callback
2355 * \retval 0 on success
2356 * \retval negative number on error
2358 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2359 void *data, int flag)
2361 struct lustre_handle lockh;
2366 case LDLM_CB_BLOCKING:
2367 ldlm_lock2handle(lock, &lockh);
2368 rc = ldlm_cli_cancel(&lockh,
2369 ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2371 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2375 case LDLM_CB_CANCELING:
2376 LDLM_DEBUG(lock, "Revoke remote lock\n");
2377 /* discard slc lock here so that it can be cleaned anytime,
2378 * especially for cleanup_resource() */
2379 tgt_discard_slc_lock(lock);
2388 int mdt_check_resent_lock(struct mdt_thread_info *info,
2389 struct mdt_object *mo,
2390 struct mdt_lock_handle *lhc)
2392 /* the lock might already be gotten in ldlm_handle_enqueue() */
2393 if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2394 struct ptlrpc_request *req = mdt_info_req(info);
2395 struct ldlm_lock *lock;
2397 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2398 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2400 /* Lock is pinned by ldlm_handle_enqueue0() as it is
2401 * a resend case, however, it could be already destroyed
2402 * due to client eviction or a raced cancel RPC. */
2403 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2404 lhc->mlh_reg_lh.cookie);
2408 if (!fid_res_name_eq(mdt_object_fid(mo),
2409 &lock->l_resource->lr_name)) {
2410 CWARN("%s: Although resent, but still not "
2411 "get child lock:"DFID"\n",
2412 info->mti_exp->exp_obd->obd_name,
2413 PFID(mdt_object_fid(mo)));
2414 LDLM_LOCK_PUT(lock);
2417 LDLM_LOCK_PUT(lock);
2423 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2424 const struct lu_fid *fid, struct lustre_handle *lh,
2425 enum ldlm_mode mode, __u64 ibits, bool nonblock)
2427 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2428 union ldlm_policy_data *policy = &mti->mti_policy;
2429 struct ldlm_res_id *res_id = &mti->mti_res_id;
2433 LASSERT(mdt_object_remote(o));
2435 fid_build_reg_res_name(fid, res_id);
2437 memset(einfo, 0, sizeof(*einfo));
2438 einfo->ei_type = LDLM_IBITS;
2439 einfo->ei_mode = mode;
2440 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2441 einfo->ei_cb_cp = ldlm_completion_ast;
2442 einfo->ei_enq_slave = 0;
2443 einfo->ei_res_id = res_id;
2445 einfo->ei_nonblock = 1;
2447 memset(policy, 0, sizeof(*policy));
2448 policy->l_inodebits.bits = ibits;
2450 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2455 static int mdt_object_local_lock(struct mdt_thread_info *info,
2456 struct mdt_object *o,
2457 struct mdt_lock_handle *lh, __u64 ibits,
2458 bool nonblock, bool cos_incompat)
2460 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2461 union ldlm_policy_data *policy = &info->mti_policy;
2462 struct ldlm_res_id *res_id = &info->mti_res_id;
2467 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2468 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2469 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2470 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2473 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2474 lh->mlh_reg_mode == LCK_EX);
2475 dlmflags |= LDLM_FL_COS_INCOMPAT;
2476 } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2477 dlmflags |= LDLM_FL_COS_ENABLED;
2480 /* Only enqueue LOOKUP lock for remote object */
2481 if (mdt_object_remote(o))
2482 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2484 if (lh->mlh_type == MDT_PDO_LOCK) {
2485 /* check for exists after object is locked */
2486 if (mdt_object_exists(o) == 0) {
2487 /* Non-existent object shouldn't have PDO lock */
2490 /* Non-dir object shouldn't have PDO lock */
2491 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2496 memset(policy, 0, sizeof(*policy));
2497 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2499 dlmflags |= LDLM_FL_ATOMIC_CB;
2501 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2504 * Take PDO lock on whole directory and build correct @res_id for lock
2505 * on part of directory.
2507 if (lh->mlh_pdo_hash != 0) {
2508 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2509 mdt_lock_pdo_mode(info, o, lh);
2510 if (lh->mlh_pdo_mode != LCK_NL) {
2512 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2513 * is never going to be sent to client and we do not
2514 * want it slowed down due to possible cancels.
2516 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2517 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2518 policy, res_id, dlmflags,
2519 info->mti_exp == NULL ? NULL :
2520 &info->mti_exp->exp_handle.h_cookie);
2521 if (unlikely(rc != 0))
2522 GOTO(out_unlock, rc);
2526 * Finish res_id initializing by name hash marking part of
2527 * directory which is taking modification.
2529 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2532 policy->l_inodebits.bits = ibits;
2535 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2536 * going to be sent to client. If it is - mdt_intent_policy() path will
2537 * fix it up and turn FL_LOCAL flag off.
2539 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2540 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2541 info->mti_exp == NULL ? NULL :
2542 &info->mti_exp->exp_handle.h_cookie);
2545 mdt_object_unlock(info, o, lh, 1);
2546 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2547 lh->mlh_pdo_hash != 0 &&
2548 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2549 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2555 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2556 struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2559 struct mdt_lock_handle *local_lh = NULL;
2563 if (!mdt_object_remote(o)) {
2564 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2569 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2570 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2571 MDS_INODELOCK_XATTR);
2573 /* Only enqueue LOOKUP lock for remote object */
2574 if (ibits & MDS_INODELOCK_LOOKUP) {
2575 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2576 nonblock, cos_incompat);
2583 if (ibits & MDS_INODELOCK_UPDATE) {
2584 /* Sigh, PDO needs to enqueue 2 locks right now, but
2585 * enqueue RPC can only request 1 lock, to avoid extra
2586 * RPC, so it will instead enqueue EX lock for remote
2587 * object anyway XXX*/
2588 if (lh->mlh_type == MDT_PDO_LOCK &&
2589 lh->mlh_pdo_hash != 0) {
2590 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2591 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2592 PFID(mdt_object_fid(o)));
2593 lh->mlh_pdo_hash = 0;
2594 lh->mlh_rreg_mode = LCK_EX;
2595 lh->mlh_type = MDT_REG_LOCK;
2597 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2600 MDS_INODELOCK_UPDATE, nonblock);
2601 if (rc != ELDLM_OK) {
2602 if (local_lh != NULL)
2603 mdt_object_unlock(info, o, local_lh, rc);
2611 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2612 struct mdt_lock_handle *lh, __u64 ibits)
2614 return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2617 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2618 struct mdt_lock_handle *lh, __u64 ibits,
2621 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2622 return mdt_object_lock_internal(info, o, lh, ibits, false,
2626 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2627 struct mdt_lock_handle *lh, __u64 ibits)
2629 struct mdt_lock_handle tmp = *lh;
2632 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2639 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2640 struct mdt_object *o, struct mdt_lock_handle *lh,
2641 __u64 ibits, bool cos_incompat)
2643 struct mdt_lock_handle tmp = *lh;
2646 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2647 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2655 * Save a lock within request object.
2657 * Keep the lock referenced until whether client ACK or transaction
2658 * commit happens or release the lock immediately depending on input
2659 * parameters. If COS is ON, a write lock is converted to COS lock
2662 * \param info thead info object
2663 * \param h lock handle
2664 * \param mode lock mode
2665 * \param decref force immediate lock releasing
2667 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2668 enum ldlm_mode mode, int decref)
2672 if (lustre_handle_is_used(h)) {
2673 if (decref || !info->mti_has_trans ||
2674 !(mode & (LCK_PW | LCK_EX))) {
2675 mdt_fid_unlock(h, mode);
2677 struct mdt_device *mdt = info->mti_mdt;
2678 struct ldlm_lock *lock = ldlm_handle2lock(h);
2679 struct ptlrpc_request *req = mdt_info_req(info);
2682 cos = (mdt_cos_is_enabled(mdt) ||
2683 mdt_slc_is_enabled(mdt));
2685 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2688 /* there is no request if mdt_object_unlock() is called
2689 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2690 if (likely(req != NULL)) {
2691 CDEBUG(D_HA, "request = %p reply state = %p"
2692 " transno = "LPD64"\n", req,
2693 req->rq_reply_state, req->rq_transno);
2695 ldlm_lock_downgrade(lock, LCK_COS);
2698 ptlrpc_save_lock(req, h, mode, cos);
2700 mdt_fid_unlock(h, mode);
2702 if (mdt_is_lock_sync(lock)) {
2703 CDEBUG(D_HA, "found sync-lock,"
2704 " async commit started\n");
2705 mdt_device_commit_async(info->mti_env,
2708 LDLM_LOCK_PUT(lock);
2717 * Save cross-MDT lock in uncommitted_slc_locks
2719 * Keep the lock referenced until transaction commit happens or release the lock
2720 * immediately depending on input parameters.
2722 * \param info thead info object
2723 * \param h lock handle
2724 * \param mode lock mode
2725 * \param decref force immediate lock releasing
2727 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2728 struct lustre_handle *h, enum ldlm_mode mode,
2733 if (lustre_handle_is_used(h)) {
2734 if (decref || !info->mti_has_trans ||
2735 !(mode & (LCK_PW | LCK_EX))) {
2736 ldlm_lock_decref_and_cancel(h, mode);
2738 struct ldlm_lock *lock = ldlm_handle2lock(h);
2739 struct ptlrpc_request *req = mdt_info_req(info);
2741 LASSERT(req != NULL);
2742 tgt_save_slc_lock(lock, req->rq_transno);
2743 ldlm_lock_decref(h, mode);
2752 * Unlock mdt object.
2754 * Immeditely release the regular lock and the PDO lock or save the
2755 * lock in request and keep them referenced until client ACK or
2756 * transaction commit.
2758 * \param info thread info object
2759 * \param o mdt object
2760 * \param lh mdt lock handle referencing regular and PDO locks
2761 * \param decref force immediate lock releasing
2763 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2764 struct mdt_lock_handle *lh, int decref)
2768 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2769 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2770 mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
2775 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2776 const struct lu_fid *f,
2777 struct mdt_lock_handle *lh,
2780 struct mdt_object *o;
2782 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2786 rc = mdt_object_lock(info, o, lh, ibits);
2788 mdt_object_put(info->mti_env, o);
2795 void mdt_object_unlock_put(struct mdt_thread_info * info,
2796 struct mdt_object * o,
2797 struct mdt_lock_handle *lh,
2800 mdt_object_unlock(info, o, lh, decref);
2801 mdt_object_put(info->mti_env, o);
2805 * Generic code handling requests that have struct mdt_body passed in:
2807 * - extract mdt_body from request and save it in @info, if present;
2809 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2812 * - if HABEO_CORPUS flag is set for this request type check whether object
2813 * actually exists on storage (lu_object_exists()).
2816 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2818 const struct mdt_body *body;
2819 struct mdt_object *obj;
2820 const struct lu_env *env;
2821 struct req_capsule *pill;
2825 env = info->mti_env;
2826 pill = info->mti_pill;
2828 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2832 if (!(body->mbo_valid & OBD_MD_FLID))
2835 if (!fid_is_sane(&body->mbo_fid1)) {
2836 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2840 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2842 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2843 mdt_object_put(env, obj);
2846 info->mti_object = obj;
2855 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2857 struct req_capsule *pill = info->mti_pill;
2861 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2862 rc = mdt_body_unpack(info, flags);
2866 if (rc == 0 && (flags & HABEO_REFERO)) {
2868 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2869 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2871 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2872 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2875 rc = req_capsule_server_pack(pill);
2880 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2882 lh->mlh_type = MDT_NUL_LOCK;
2883 lh->mlh_reg_lh.cookie = 0ull;
2884 lh->mlh_reg_mode = LCK_MINMODE;
2885 lh->mlh_pdo_lh.cookie = 0ull;
2886 lh->mlh_pdo_mode = LCK_MINMODE;
2887 lh->mlh_rreg_lh.cookie = 0ull;
2888 lh->mlh_rreg_mode = LCK_MINMODE;
2891 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2893 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2894 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2898 * Initialize fields of struct mdt_thread_info. Other fields are left in
2899 * uninitialized state, because it's too expensive to zero out whole
2900 * mdt_thread_info (> 1K) on each request arrival.
2902 void mdt_thread_info_init(struct ptlrpc_request *req,
2903 struct mdt_thread_info *info)
2907 info->mti_pill = &req->rq_pill;
2910 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2911 mdt_lock_handle_init(&info->mti_lh[i]);
2913 /* mdt device: it can be NULL while CONNECT */
2914 if (req->rq_export) {
2915 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2916 info->mti_exp = req->rq_export;
2918 info->mti_mdt = NULL;
2919 info->mti_env = req->rq_svc_thread->t_env;
2920 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2922 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2923 info->mti_big_buf = LU_BUF_NULL;
2924 info->mti_body = NULL;
2925 info->mti_object = NULL;
2926 info->mti_dlm_req = NULL;
2927 info->mti_has_trans = 0;
2928 info->mti_cross_ref = 0;
2929 info->mti_opdata = 0;
2930 info->mti_big_lmm_used = 0;
2932 info->mti_spec.no_create = 0;
2933 info->mti_spec.sp_rm_entry = 0;
2934 info->mti_spec.sp_permitted = 0;
2935 info->mti_spec.sp_migrate_close = 0;
2937 info->mti_spec.u.sp_ea.eadata = NULL;
2938 info->mti_spec.u.sp_ea.eadatalen = 0;
2941 void mdt_thread_info_fini(struct mdt_thread_info *info)
2945 if (info->mti_object != NULL) {
2946 mdt_object_put(info->mti_env, info->mti_object);
2947 info->mti_object = NULL;
2950 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2951 mdt_lock_handle_fini(&info->mti_lh[i]);
2952 info->mti_env = NULL;
2953 info->mti_pill = NULL;
2954 info->mti_exp = NULL;
2956 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2957 lu_buf_free(&info->mti_big_buf);
2960 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2962 struct mdt_thread_info *mti;
2964 mti = mdt_th_info(tsi->tsi_env);
2965 LASSERT(mti != NULL);
2967 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2968 if (tsi->tsi_corpus != NULL) {
2969 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2970 lu_object_get(tsi->tsi_corpus);
2972 mti->mti_body = tsi->tsi_mdt_body;
2973 mti->mti_dlm_req = tsi->tsi_dlm_req;
2978 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2980 struct ptlrpc_request *req = tgt_ses_req(tsi);
2985 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2987 tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2988 set_current_state(TASK_UNINTERRUPTIBLE);
2989 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2992 rc = tgt_connect(tsi);
2996 rc = mdt_init_idmap(tsi);
3001 obd_disconnect(class_export_get(req->rq_export));
3020 static int mdt_intent_getattr(enum mdt_it_code opcode,
3021 struct mdt_thread_info *info,
3022 struct ldlm_lock **,
3025 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3026 struct mdt_thread_info *info,
3027 struct ldlm_lock **lockp,
3030 static int mdt_intent_layout(enum mdt_it_code opcode,
3031 struct mdt_thread_info *info,
3032 struct ldlm_lock **,
3034 static int mdt_intent_reint(enum mdt_it_code opcode,
3035 struct mdt_thread_info *info,
3036 struct ldlm_lock **,
3039 static struct mdt_it_flavor {
3040 const struct req_format *it_fmt;
3042 int (*it_act)(enum mdt_it_code ,
3043 struct mdt_thread_info *,
3044 struct ldlm_lock **,
3047 } mdt_it_flavor[] = {
3049 .it_fmt = &RQF_LDLM_INTENT,
3050 /*.it_flags = HABEO_REFERO,*/
3052 .it_act = mdt_intent_reint,
3053 .it_reint = REINT_OPEN
3056 .it_fmt = &RQF_LDLM_INTENT,
3058 * OCREAT is not a MUTABOR request as if the file
3060 * We do the extra check of OBD_CONNECT_RDONLY in
3061 * mdt_reint_open() when we really need to create
3065 .it_act = mdt_intent_reint,
3066 .it_reint = REINT_OPEN
3069 .it_fmt = &RQF_LDLM_INTENT,
3070 .it_flags = MUTABOR,
3071 .it_act = mdt_intent_reint,
3072 .it_reint = REINT_CREATE
3074 [MDT_IT_GETATTR] = {
3075 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3076 .it_flags = HABEO_REFERO,
3077 .it_act = mdt_intent_getattr
3079 [MDT_IT_READDIR] = {
3085 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3086 .it_flags = HABEO_REFERO,
3087 .it_act = mdt_intent_getattr
3090 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3091 .it_flags = MUTABOR,
3093 .it_reint = REINT_UNLINK
3097 .it_flags = MUTABOR,
3100 [MDT_IT_GETXATTR] = {
3101 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
3102 .it_flags = HABEO_CORPUS,
3103 .it_act = mdt_intent_getxattr
3106 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3108 .it_act = mdt_intent_layout
3113 mdt_intent_lock_replace(struct mdt_thread_info *info,
3114 struct ldlm_lock **lockp,
3115 struct mdt_lock_handle *lh,
3118 struct ptlrpc_request *req = mdt_info_req(info);
3119 struct ldlm_lock *lock = *lockp;
3120 struct ldlm_lock *new_lock;
3122 /* If possible resent found a lock, @lh is set to its handle */
3123 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3125 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3126 lh->mlh_reg_lh.cookie = 0;
3130 LASSERTF(new_lock != NULL,
3131 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3134 * If we've already given this lock to a client once, then we should
3135 * have no readers or writers. Otherwise, we should have one reader
3136 * _or_ writer ref (which will be zeroed below) before returning the
3139 if (new_lock->l_export == req->rq_export) {
3140 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3142 LASSERT(new_lock->l_export == NULL);
3143 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3148 if (new_lock->l_export == req->rq_export) {
3150 * Already gave this to the client, which means that we
3151 * reconstructed a reply.
3153 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3156 LDLM_LOCK_RELEASE(new_lock);
3157 lh->mlh_reg_lh.cookie = 0;
3158 RETURN(ELDLM_LOCK_REPLACED);
3162 * Fixup the lock to be given to the client.
3164 lock_res_and_lock(new_lock);
3165 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3166 * possible blocking AST. */
3167 while (new_lock->l_readers > 0) {
3168 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3169 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3170 new_lock->l_readers--;
3172 while (new_lock->l_writers > 0) {
3173 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3174 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3175 new_lock->l_writers--;
3178 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3179 new_lock->l_blocking_ast = lock->l_blocking_ast;
3180 new_lock->l_completion_ast = lock->l_completion_ast;
3181 new_lock->l_remote_handle = lock->l_remote_handle;
3182 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3184 unlock_res_and_lock(new_lock);
3186 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3187 &new_lock->l_remote_handle,
3188 &new_lock->l_exp_hash);
3190 LDLM_LOCK_RELEASE(new_lock);
3191 lh->mlh_reg_lh.cookie = 0;
3193 RETURN(ELDLM_LOCK_REPLACED);
3196 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3197 struct ldlm_lock *new_lock,
3198 struct mdt_lock_handle *lh,
3201 struct ptlrpc_request *req = mdt_info_req(info);
3202 struct ldlm_request *dlmreq;
3204 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3207 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3209 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3210 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3212 if (flags & LDLM_FL_RESENT) {
3213 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3214 lh->mlh_reg_mode = new_lock->l_granted_mode;
3216 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3217 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3218 lh->mlh_reg_lh.cookie);
3223 * If the xid matches, then we know this is a resent request, and allow
3224 * it. (It's probably an OPEN, for which we don't send a lock.
3226 if (req_can_reconstruct(req, NULL))
3230 * This remote handle isn't enqueued, so we never received or processed
3231 * this request. Clear MSG_RESENT, because it can be handled like any
3232 * normal request now.
3234 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3236 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3237 dlmreq->lock_handle[0].cookie);
3240 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3241 struct mdt_thread_info *info,
3242 struct ldlm_lock **lockp,
3245 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3246 struct ldlm_reply *ldlm_rep = NULL;
3250 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3251 * (for the resend case) or a new lock. Below we will use it to
3252 * replace the original lock.
3254 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3255 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3256 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3257 rc = mdt_object_lock(info, info->mti_object, lhc,
3258 MDS_INODELOCK_XATTR);
3263 grc = mdt_getxattr(info);
3265 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3267 if (mdt_info_req(info)->rq_repmsg != NULL)
3268 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3269 if (ldlm_rep == NULL)
3270 RETURN(err_serious(-EFAULT));
3272 ldlm_rep->lock_policy_res2 = grc;
3277 static int mdt_intent_getattr(enum mdt_it_code opcode,
3278 struct mdt_thread_info *info,
3279 struct ldlm_lock **lockp,
3282 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3284 struct ldlm_reply *ldlm_rep;
3285 struct mdt_body *reqbody;
3286 struct mdt_body *repbody;
3290 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3293 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3296 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3297 repbody->mbo_eadatasize = 0;
3298 repbody->mbo_aclsize = 0;
3302 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3304 case MDT_IT_GETATTR:
3305 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3309 CERROR("Unsupported intent (%d)\n", opcode);
3310 GOTO(out_shrink, rc = -EINVAL);
3313 rc = mdt_init_ucred_intent_getattr(info, reqbody);
3315 GOTO(out_shrink, rc);
3317 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3318 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3320 /* Get lock from request for possible resent case. */
3321 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3323 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3324 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3326 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3327 ldlm_rep->lock_policy_res2 = 0;
3328 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3329 ldlm_rep->lock_policy_res2) {
3330 lhc->mlh_reg_lh.cookie = 0ull;
3331 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3334 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3337 mdt_exit_ucred(info);
3339 mdt_client_compatibility(info);
3340 rc2 = mdt_fix_reply(info);
3346 static int mdt_intent_layout(enum mdt_it_code opcode,
3347 struct mdt_thread_info *info,
3348 struct ldlm_lock **lockp,
3351 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3352 struct layout_intent *layout;
3354 struct mdt_object *obj = NULL;
3355 int layout_size = 0;
3359 if (opcode != MDT_IT_LAYOUT) {
3360 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3365 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3369 if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3370 CERROR("%s: Unsupported layout intent opc %d\n",
3371 mdt_obd_name(info->mti_mdt), layout->li_opc);
3375 fid = &info->mti_tmp_fid2;
3376 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3378 /* Get lock from request for possible resent case. */
3379 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3381 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3383 GOTO(out, rc = PTR_ERR(obj));
3385 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3386 layout_size = mdt_attr_get_eabuf_size(info, obj);
3387 if (layout_size < 0)
3388 GOTO(out_obj, rc = layout_size);
3390 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3391 info->mti_mdt->mdt_max_mdsize = layout_size;
3394 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3395 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3397 rc = req_capsule_server_pack(info->mti_pill);
3401 mdt_object_put(info->mti_env, obj);
3403 if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3404 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3407 lhc->mlh_reg_lh.cookie = 0;
3412 static int mdt_intent_reint(enum mdt_it_code opcode,
3413 struct mdt_thread_info *info,
3414 struct ldlm_lock **lockp,
3417 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3418 struct ldlm_reply *rep = NULL;
3422 static const struct req_format *intent_fmts[REINT_MAX] = {
3423 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3424 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3429 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3433 if (mdt_it_flavor[opcode].it_reint != opc) {
3434 CERROR("Reint code %ld doesn't match intent: %d\n",
3436 RETURN(err_serious(-EPROTO));
3439 /* Get lock from request for possible resent case. */
3440 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3442 rc = mdt_reint_internal(info, lhc, opc);
3444 /* Check whether the reply has been packed successfully. */
3445 if (mdt_info_req(info)->rq_repmsg != NULL)
3446 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3448 RETURN(err_serious(-EFAULT));
3450 /* MDC expects this in any case */
3452 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3454 /* the open lock or the lock for cross-ref object should be
3455 * returned to the client */
3456 if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3457 (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3458 rep->lock_policy_res2 = 0;
3459 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3463 rep->lock_policy_res2 = clear_serious(rc);
3465 if (rep->lock_policy_res2 == -ENOENT &&
3466 mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3467 !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3468 rep->lock_policy_res2 = 0;
3470 lhc->mlh_reg_lh.cookie = 0ull;
3471 if (rc == -ENOTCONN || rc == -ENODEV ||
3472 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3474 * If it is the disconnect error (ENODEV & ENOCONN), the error
3475 * will be returned by rq_status, and client at ptlrpc layer
3476 * will detect this, then disconnect, reconnect the import
3477 * immediately, instead of impacting the following the rpc.
3482 * For other cases, the error will be returned by intent, and client
3483 * will retrieve the result from intent.
3485 RETURN(ELDLM_LOCK_ABORTED);
3488 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3496 case IT_OPEN|IT_CREAT:
3503 rc = MDT_IT_READDIR;
3506 rc = MDT_IT_GETATTR;
3518 rc = MDT_IT_GETXATTR;
3523 case IT_QUOTA_DQACQ:
3528 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3535 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3536 struct mdt_thread_info *info,
3537 struct ldlm_lock **lockp, __u64 flags)
3539 struct req_capsule *pill = info->mti_pill;
3540 struct ptlrpc_request *req = mdt_info_req(info);
3541 struct mdt_it_flavor *flv;
3546 opc = mdt_intent_code(itopc);
3550 if (opc == MDT_IT_QUOTA) {
3551 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3554 RETURN(-EOPNOTSUPP);
3556 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3557 /* pass the request to quota master */
3558 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3559 mdt_info_req(info), lockp,
3564 flv = &mdt_it_flavor[opc];
3565 if (flv->it_fmt != NULL)
3566 req_capsule_extend(pill, flv->it_fmt);
3568 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3572 if (flv->it_flags & MUTABOR &&
3573 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3576 if (flv->it_act != NULL) {
3577 struct ldlm_reply *rep;
3579 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3581 /* execute policy */
3582 rc = flv->it_act(opc, info, lockp, flags);
3584 /* Check whether the reply has been packed successfully. */
3585 if (req->rq_repmsg != NULL) {
3586 rep = req_capsule_server_get(info->mti_pill,
3588 rep->lock_policy_res2 =
3589 ptlrpc_status_hton(rep->lock_policy_res2);
3596 static int mdt_intent_policy(struct ldlm_namespace *ns,
3597 struct ldlm_lock **lockp, void *req_cookie,
3598 enum ldlm_mode mode, __u64 flags, void *data)
3600 struct tgt_session_info *tsi;
3601 struct mdt_thread_info *info;
3602 struct ptlrpc_request *req = req_cookie;
3603 struct ldlm_intent *it;
3604 struct req_capsule *pill;
3609 LASSERT(req != NULL);
3611 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3613 info = tsi2mdt_info(tsi);
3614 LASSERT(info != NULL);
3615 pill = info->mti_pill;
3616 LASSERT(pill->rc_req == req);
3618 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3619 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3620 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3622 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3626 /* Lock without inodebits makes no sense and will oops
3627 * later in ldlm. Let's check it now to see if we have
3628 * ibits corrupted somewhere in mdt_intent_opc().
3629 * The case for client miss to set ibits has been
3630 * processed by others. */
3631 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3632 lr_type == LDLM_IBITS,
3633 info->mti_dlm_req->lock_desc.\
3634 l_policy_data.l_inodebits.bits != 0));
3636 rc = err_serious(-EFAULT);
3638 /* No intent was provided */
3639 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3640 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3641 rc = req_capsule_server_pack(pill);
3643 rc = err_serious(rc);
3645 mdt_thread_info_fini(info);
3649 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3651 struct seq_server_site *ss = mdt_seq_site(mdt);
3653 if (ss->ss_node_id == 0)
3656 if (ss->ss_client_seq != NULL) {
3657 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3658 ss->ss_client_seq->lcs_exp = NULL;
3661 if (ss->ss_server_fld != NULL) {
3662 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3663 ss->ss_server_fld->lsf_control_exp = NULL;
3667 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3669 struct seq_server_site *ss = mdt_seq_site(mdt);
3674 if (ss->ss_server_seq != NULL)
3675 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3678 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3680 mdt_seq_fini_cli(mdt);
3681 mdt_deregister_seq_exp(mdt);
3683 return seq_site_fini(env, mdt_seq_site(mdt));
3687 * It will retrieve its FLDB entries from MDT0, and it only happens
3688 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3689 * and it needs to refresh FLDB from the MDT0.
3691 static int mdt_register_lwp_callback(void *data)
3694 struct mdt_device *mdt = data;
3695 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3699 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3701 rc = lu_env_init(&env, LCT_MD_THREAD);
3703 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3707 /* Allocate new sequence now to avoid creating local transaction
3708 * in the normal transaction process */
3709 rc = seq_server_check_and_alloc_super(&env,
3710 mdt_seq_site(mdt)->ss_server_seq);
3715 rc = fld_update_from_controller(&env, fld);
3717 CERROR("%s: cannot update controller: rc = %d\n",
3718 mdt_obd_name(mdt), rc);
3727 static int mdt_register_seq_exp(struct mdt_device *mdt)
3729 struct seq_server_site *ss = mdt_seq_site(mdt);
3730 char *lwp_name = NULL;
3733 if (ss->ss_node_id == 0)
3736 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3737 if (lwp_name == NULL)
3738 GOTO(out_free, rc = -ENOMEM);
3740 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3744 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3749 rc = lustre_register_lwp_item(lwp_name,
3750 &ss->ss_server_fld->lsf_control_exp,
3751 mdt_register_lwp_callback, mdt);
3753 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3754 ss->ss_client_seq->lcs_exp = NULL;
3758 if (lwp_name != NULL)
3759 OBD_FREE(lwp_name, MAX_OBD_NAME);
3765 * Init client sequence manager which is used by local MDS to talk to sequence
3766 * controller on remote node.
3768 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3770 struct seq_server_site *ss = mdt_seq_site(mdt);
3775 /* check if this is adding the first MDC and controller is not yet
3777 OBD_ALLOC_PTR(ss->ss_client_seq);
3778 if (ss->ss_client_seq == NULL)
3781 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3782 if (prefix == NULL) {
3783 OBD_FREE_PTR(ss->ss_client_seq);
3784 ss->ss_client_seq = NULL;
3788 /* Note: seq_client_fini will be called in seq_site_fini */
3789 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3790 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3791 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3793 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3795 OBD_FREE_PTR(ss->ss_client_seq);
3796 ss->ss_client_seq = NULL;
3800 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3805 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3807 struct seq_server_site *ss;
3811 ss = mdt_seq_site(mdt);
3812 /* init sequence controller server(MDT0) */
3813 if (ss->ss_node_id == 0) {
3814 OBD_ALLOC_PTR(ss->ss_control_seq);
3815 if (ss->ss_control_seq == NULL)
3818 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3819 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3822 GOTO(out_seq_fini, rc);
3825 /* Init normal sequence server */
3826 OBD_ALLOC_PTR(ss->ss_server_seq);
3827 if (ss->ss_server_seq == NULL)
3828 GOTO(out_seq_fini, rc = -ENOMEM);
3830 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3831 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3833 GOTO(out_seq_fini, rc);
3835 /* init seq client for seq server to talk to seq controller(MDT0) */
3836 rc = mdt_seq_init_cli(env, mdt);
3838 GOTO(out_seq_fini, rc);
3840 if (ss->ss_node_id != 0)
3841 /* register controller export through lwp */
3842 rc = mdt_register_seq_exp(mdt);
3847 mdt_seq_fini(env, mdt);
3855 static int mdt_fld_fini(const struct lu_env *env,
3856 struct mdt_device *m)
3858 struct seq_server_site *ss = mdt_seq_site(m);
3861 if (ss && ss->ss_server_fld) {
3862 fld_server_fini(env, ss->ss_server_fld);
3863 OBD_FREE_PTR(ss->ss_server_fld);
3864 ss->ss_server_fld = NULL;
3870 static int mdt_fld_init(const struct lu_env *env,
3872 struct mdt_device *m)
3874 struct seq_server_site *ss;
3878 ss = mdt_seq_site(m);
3880 OBD_ALLOC_PTR(ss->ss_server_fld);
3881 if (ss->ss_server_fld == NULL)
3882 RETURN(rc = -ENOMEM);
3884 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3887 OBD_FREE_PTR(ss->ss_server_fld);
3888 ss->ss_server_fld = NULL;
3895 static void mdt_stack_pre_fini(const struct lu_env *env,
3896 struct mdt_device *m, struct lu_device *top)
3898 struct lustre_cfg_bufs *bufs;
3899 struct lustre_cfg *lcfg;
3900 struct mdt_thread_info *info;
3905 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3906 LASSERT(info != NULL);
3908 bufs = &info->mti_u.bufs;
3910 LASSERT(m->mdt_child_exp);
3911 LASSERT(m->mdt_child_exp->exp_obd);
3913 /* process cleanup, pass mdt obd name to get obd umount flags */
3914 /* XXX: this is needed because all layers are referenced by
3915 * objects (some of them are pinned by osd, for example *
3916 * the proper solution should be a model where object used
3917 * by osd only doesn't have mdt/mdd slices -bzzz */
3918 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3919 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3920 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3924 top->ld_ops->ldo_process_config(env, top, lcfg);
3925 lustre_cfg_free(lcfg);
3929 static void mdt_stack_fini(const struct lu_env *env,
3930 struct mdt_device *m, struct lu_device *top)
3932 struct obd_device *obd = mdt2obd_dev(m);
3933 struct lustre_cfg_bufs *bufs;
3934 struct lustre_cfg *lcfg;
3935 struct mdt_thread_info *info;
3939 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3940 LASSERT(info != NULL);
3942 lu_dev_del_linkage(top->ld_site, top);
3944 lu_site_purge(env, top->ld_site, -1);
3946 bufs = &info->mti_u.bufs;
3947 /* process cleanup, pass mdt obd name to get obd umount flags */
3948 /* another purpose is to let all layers to release their objects */
3949 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3954 lustre_cfg_bufs_set_string(bufs, 1, flags);
3955 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3960 top->ld_ops->ldo_process_config(env, top, lcfg);
3961 lustre_cfg_free(lcfg);
3963 lu_site_purge(env, top->ld_site, -1);
3965 m->mdt_child = NULL;
3966 m->mdt_bottom = NULL;
3968 obd_disconnect(m->mdt_child_exp);
3969 m->mdt_child_exp = NULL;
3971 obd_disconnect(m->mdt_bottom_exp);
3972 m->mdt_child_exp = NULL;
3975 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3976 const char *next, struct obd_export **exp)
3978 struct obd_connect_data *data = NULL;
3979 struct obd_device *obd;
3983 OBD_ALLOC_PTR(data);
3985 GOTO(out, rc = -ENOMEM);
3987 obd = class_name2obd(next);
3989 CERROR("%s: can't locate next device: %s\n",
3990 mdt_obd_name(m), next);
3991 GOTO(out, rc = -ENOTCONN);
3994 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3995 data->ocd_version = LUSTRE_VERSION_CODE;
3997 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3999 CERROR("%s: cannot connect to next dev %s (%d)\n",
4000 mdt_obd_name(m), next, rc);
4010 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4011 struct lustre_cfg *cfg)
4013 char *dev = lustre_cfg_string(cfg, 0);
4014 int rc, name_size, uuid_size;
4015 char *name, *uuid, *p;
4016 struct lustre_cfg_bufs *bufs;
4017 struct lustre_cfg *lcfg;
4018 struct obd_device *obd;
4019 struct lustre_profile *lprof;
4020 struct lu_site *site;
4023 /* in 1.8 we had the only device in the stack - MDS.
4024 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4025 * in 2.3 OSD is instantiated by obd_mount.c, so we need
4026 * to generate names and setup MDT, MDD. MDT will be using
4027 * generated name to connect to MDD. for MDD the next device
4028 * will be LOD with name taken from so called "profile" which
4029 * is generated by mount_option line
4031 * 1.8 MGS generates config. commands like this:
4032 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
4033 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
4034 * 2.0 MGS generates config. commands like this:
4035 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
4036 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4037 * 3:lustre-MDT0000-mdtlov 4:f
4039 * we generate MDD name from MDT one, just replacing T with D
4041 * after all the preparations, the logical equivalent will be
4042 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
4043 * 3:lustre-MDT0000-mdtlov 4:f
4044 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4045 * 3:lustre-MDD0000 4:f
4047 * notice we build the stack from down to top: MDD first, then MDT */
4049 name_size = MAX_OBD_NAME;
4050 uuid_size = MAX_OBD_NAME;
4052 OBD_ALLOC(name, name_size);
4053 OBD_ALLOC(uuid, uuid_size);
4054 if (name == NULL || uuid == NULL)
4055 GOTO(cleanup_mem, rc = -ENOMEM);
4057 OBD_ALLOC_PTR(bufs);
4059 GOTO(cleanup_mem, rc = -ENOMEM);
4062 p = strstr(name, "-MDT");
4064 GOTO(free_bufs, rc = -ENOMEM);
4067 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4069 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4070 if (lprof == NULL || lprof->lp_dt == NULL) {
4071 CERROR("can't find the profile: %s\n",
4072 lustre_cfg_string(cfg, 0));
4073 GOTO(free_bufs, rc = -EINVAL);
4076 lustre_cfg_bufs_reset(bufs, name);
4077 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4078 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4079 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4081 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4083 GOTO(put_profile, rc = -ENOMEM);
4085 rc = class_attach(lcfg);
4087 GOTO(lcfg_cleanup, rc);
4089 obd = class_name2obd(name);
4091 CERROR("Can not find obd %s (%s in config)\n",
4092 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4093 GOTO(lcfg_cleanup, rc = -EINVAL);
4096 lustre_cfg_free(lcfg);
4098 lustre_cfg_bufs_reset(bufs, name);
4099 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4100 lustre_cfg_bufs_set_string(bufs, 2, dev);
4101 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4103 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4105 GOTO(class_detach, rc = -ENOMEM);
4107 rc = class_setup(obd, lcfg);
4109 GOTO(class_detach, rc);
4111 /* connect to MDD we just setup */
4112 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4114 GOTO(class_detach, rc);
4116 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4118 LASSERT(mdt_lu_site(mdt) == NULL);
4119 mdt->mdt_lu_dev.ld_site = site;
4120 site->ls_top_dev = &mdt->mdt_lu_dev;
4121 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4123 /* now connect to bottom OSD */
4124 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4125 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4127 GOTO(class_detach, rc);
4129 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4131 rc = lu_env_refill((struct lu_env *)env);
4133 CERROR("Failure to refill session: '%d'\n", rc);
4135 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4140 class_detach(obd, lcfg);
4142 lustre_cfg_free(lcfg);
4144 class_put_profile(lprof);
4149 OBD_FREE(name, name_size);
4151 OBD_FREE(uuid, uuid_size);
4155 /* setup quota master target on MDT0 */
4156 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4157 struct lustre_cfg *cfg)
4159 struct obd_device *obd;
4160 char *dev = lustre_cfg_string(cfg, 0);
4161 char *qmtname, *uuid, *p;
4162 struct lustre_cfg_bufs *bufs;
4163 struct lustre_cfg *lcfg;
4164 struct lustre_profile *lprof;
4165 struct obd_connect_data *data;
4169 LASSERT(mdt->mdt_qmt_exp == NULL);
4170 LASSERT(mdt->mdt_qmt_dev == NULL);
4172 /* quota master is on MDT0 only for now */
4173 if (mdt->mdt_seq_site.ss_node_id != 0)
4176 /* MGS generates config commands which look as follows:
4177 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4178 * 3:lustre-MDT0000-mdtlov 4:f
4180 * We generate the QMT name from the MDT one, just replacing MD with QM
4181 * after all the preparations, the logical equivalent will be:
4182 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4183 * 3:lustre-MDT0000-osd 4:f */
4184 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4185 OBD_ALLOC(uuid, UUID_MAX);
4186 OBD_ALLOC_PTR(bufs);
4187 OBD_ALLOC_PTR(data);
4188 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4189 GOTO(cleanup_mem, rc = -ENOMEM);
4191 strcpy(qmtname, dev);
4192 p = strstr(qmtname, "-MDT");
4194 GOTO(cleanup_mem, rc = -ENOMEM);
4195 /* replace MD with QM */
4199 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4201 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4202 if (lprof == NULL || lprof->lp_dt == NULL) {
4203 CERROR("can't find profile for %s\n",
4204 lustre_cfg_string(cfg, 0));
4205 GOTO(cleanup_mem, rc = -EINVAL);
4208 lustre_cfg_bufs_reset(bufs, qmtname);
4209 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4210 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4211 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4213 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4215 GOTO(put_profile, rc = -ENOMEM);
4217 rc = class_attach(lcfg);
4219 GOTO(lcfg_cleanup, rc);
4221 obd = class_name2obd(qmtname);
4223 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4224 lustre_cfg_string(cfg, 0));
4225 GOTO(lcfg_cleanup, rc = -EINVAL);
4228 lustre_cfg_free(lcfg);
4230 lustre_cfg_bufs_reset(bufs, qmtname);
4231 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4232 lustre_cfg_bufs_set_string(bufs, 2, dev);
4234 /* for quota, the next device should be the OSD device */
4235 lustre_cfg_bufs_set_string(bufs, 3,
4236 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4238 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4240 GOTO(class_detach, rc = -ENOMEM);
4242 rc = class_setup(obd, lcfg);
4244 GOTO(class_detach, rc);
4246 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4248 /* configure local quota objects */
4249 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4253 GOTO(class_cleanup, rc);
4255 /* connect to quota master target */
4256 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4257 data->ocd_version = LUSTRE_VERSION_CODE;
4258 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4261 CERROR("cannot connect to quota master device %s (%d)\n",
4263 GOTO(class_cleanup, rc);
4269 class_manual_cleanup(obd);
4270 mdt->mdt_qmt_dev = NULL;
4274 class_detach(obd, lcfg);
4276 lustre_cfg_free(lcfg);
4278 class_put_profile(lprof);
4283 OBD_FREE(qmtname, MAX_OBD_NAME);
4285 OBD_FREE(uuid, UUID_MAX);
4291 /* Shutdown quota master target associated with mdt */
4292 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4296 if (mdt->mdt_qmt_exp == NULL)
4298 LASSERT(mdt->mdt_qmt_dev != NULL);
4300 /* the qmt automatically shuts down when the mdt disconnects */
4301 obd_disconnect(mdt->mdt_qmt_exp);
4302 mdt->mdt_qmt_exp = NULL;
4303 mdt->mdt_qmt_dev = NULL;
4307 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4308 * for now. This will be removed along with converting rest of MDT code
4309 * to use tgt_session_info */
4310 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4312 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4315 rc = mdt_getxattr(info);
4317 mdt_thread_info_fini(info);
4321 static struct tgt_handler mdt_tgt_handlers[] = {
4322 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4323 0, MDS_CONNECT, mdt_tgt_connect,
4324 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4325 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4326 0, MDS_DISCONNECT, tgt_disconnect,
4327 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4328 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4329 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4330 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4331 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4332 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GET_ROOT, mdt_get_root),
4333 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4334 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4336 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4337 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4338 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4339 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4340 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4341 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4342 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4343 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4345 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4346 mdt_hsm_ct_register),
4347 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4348 mdt_hsm_ct_unregister),
4349 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4351 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4353 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4354 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4356 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4361 static struct tgt_handler mdt_sec_ctx_ops[] = {
4362 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4363 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4364 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4367 static struct tgt_handler mdt_quota_ops[] = {
4368 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4371 static struct tgt_opc_slice mdt_common_slice[] = {
4373 .tos_opc_start = MDS_FIRST_OPC,
4374 .tos_opc_end = MDS_LAST_OPC,
4375 .tos_hs = mdt_tgt_handlers
4378 .tos_opc_start = OBD_FIRST_OPC,
4379 .tos_opc_end = OBD_LAST_OPC,
4380 .tos_hs = tgt_obd_handlers
4383 .tos_opc_start = LDLM_FIRST_OPC,
4384 .tos_opc_end = LDLM_LAST_OPC,
4385 .tos_hs = tgt_dlm_handlers
4388 .tos_opc_start = SEC_FIRST_OPC,
4389 .tos_opc_end = SEC_LAST_OPC,
4390 .tos_hs = mdt_sec_ctx_ops
4393 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4394 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4395 .tos_hs = tgt_out_handlers
4398 .tos_opc_start = FLD_FIRST_OPC,
4399 .tos_opc_end = FLD_LAST_OPC,
4400 .tos_hs = fld_handlers
4403 .tos_opc_start = SEQ_FIRST_OPC,
4404 .tos_opc_end = SEQ_LAST_OPC,
4405 .tos_hs = seq_handlers
4408 .tos_opc_start = QUOTA_DQACQ,
4409 .tos_opc_end = QUOTA_LAST_OPC,
4410 .tos_hs = mdt_quota_ops
4413 .tos_opc_start = LLOG_FIRST_OPC,
4414 .tos_opc_end = LLOG_LAST_OPC,
4415 .tos_hs = tgt_llog_handlers
4418 .tos_opc_start = LFSCK_FIRST_OPC,
4419 .tos_opc_end = LFSCK_LAST_OPC,
4420 .tos_hs = tgt_lfsck_handlers
4428 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4430 struct md_device *next = m->mdt_child;
4431 struct lu_device *d = &m->mdt_lu_dev;
4432 struct obd_device *obd = mdt2obd_dev(m);
4433 struct lfsck_stop stop;
4436 stop.ls_status = LS_PAUSED;
4438 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4440 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4441 target_recovery_fini(obd);
4442 ping_evictor_stop();
4444 if (m->mdt_opts.mo_coordinator)
4445 mdt_hsm_cdt_stop(m);
4447 mdt_hsm_cdt_fini(m);
4449 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4450 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4452 if (m->mdt_namespace != NULL)
4453 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4454 d->ld_obd->obd_force);
4456 obd_exports_barrier(obd);
4457 obd_zombie_barrier();
4461 tgt_fini(env, &m->mdt_lut);
4462 mdt_fs_cleanup(env, m);
4463 upcall_cache_cleanup(m->mdt_identity_cache);
4464 m->mdt_identity_cache = NULL;
4466 if (m->mdt_namespace != NULL) {
4467 ldlm_namespace_free_post(m->mdt_namespace);
4468 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4471 mdt_quota_fini(env, m);
4473 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4475 mdt_seq_fini(env, m);
4476 mdt_fld_fini(env, m);
4481 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4483 LASSERT(atomic_read(&d->ld_ref) == 0);
4485 server_put_mount(mdt_obd_name(m), true);
4490 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4492 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4493 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4495 struct mdt_thread_info *info;
4496 struct obd_device *obd;
4497 const char *dev = lustre_cfg_string(cfg, 0);
4498 const char *num = lustre_cfg_string(cfg, 2);
4499 struct lustre_mount_info *lmi = NULL;
4500 struct lustre_sb_info *lsi;
4502 struct seq_server_site *ss_site;
4503 const char *identity_upcall = "NONE";
4504 struct md_device *next;
4510 lu_device_init(&m->mdt_lu_dev, ldt);
4512 * Environment (env) might be missing mdt_thread_key values at that
4513 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4516 * Usually device allocation path doesn't use module key values, but
4517 * mdt has to do a lot of work here, so allocate key value.
4519 rc = lu_env_refill((struct lu_env *)env);
4523 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4524 LASSERT(info != NULL);
4526 obd = class_name2obd(dev);
4527 LASSERT(obd != NULL);
4529 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4530 m->mdt_opts.mo_evict_tgt_nids = 1;
4531 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4533 /* default is coordinator off, it is started through conf_param
4535 m->mdt_opts.mo_coordinator = 0;
4537 lmi = server_get_mount(dev);
4539 CERROR("Cannot get mount info for %s!\n", dev);
4542 lsi = s2lsi(lmi->lmi_sb);
4543 /* CMD is supported only in IAM mode */
4545 node_id = simple_strtol(num, NULL, 10);
4546 obd->u.obt.obt_magic = OBT_MAGIC;
4547 if (lsi->lsi_lmd != NULL &&
4548 lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4549 m->mdt_skip_lfsck = 1;
4552 m->mdt_squash.rsi_uid = 0;
4553 m->mdt_squash.rsi_gid = 0;
4554 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4555 init_rwsem(&m->mdt_squash.rsi_sem);
4556 spin_lock_init(&m->mdt_osfs_lock);
4557 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4558 m->mdt_enable_remote_dir = 0;
4559 m->mdt_enable_remote_dir_gid = 0;
4561 atomic_set(&m->mdt_mds_mds_conns, 0);
4562 atomic_set(&m->mdt_async_commit_count, 0);
4564 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4565 m->mdt_lu_dev.ld_obd = obd;
4566 /* Set this lu_device to obd for error handling purposes. */
4567 obd->obd_lu_dev = &m->mdt_lu_dev;
4569 /* init the stack */
4570 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4572 CERROR("%s: Can't init device stack, rc %d\n",
4573 mdt_obd_name(m), rc);
4578 ss_site = mdt_seq_site(m);
4579 s->ld_seq_site = ss_site;
4582 /* set server index */
4583 ss_site->ss_node_id = node_id;
4585 /* failover is the default
4586 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4587 * assumed whose ss_node_id == 0 XXX
4589 obd->obd_replayable = 1;
4590 /* No connection accepted until configurations will finish */
4591 obd->obd_no_conn = 1;
4593 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4594 char *str = lustre_cfg_string(cfg, 4);
4595 if (strchr(str, 'n')) {
4596 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4597 obd->obd_replayable = 0;
4601 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4603 GOTO(err_fini_stack, rc);
4605 rc = mdt_seq_init(env, m);
4607 GOTO(err_fini_fld, rc);
4609 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4610 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4611 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4612 LDLM_NAMESPACE_SERVER,
4613 LDLM_NAMESPACE_GREEDY,
4615 if (m->mdt_namespace == NULL)
4616 GOTO(err_fini_seq, rc = -ENOMEM);
4618 m->mdt_namespace->ns_lvbp = m;
4619 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4621 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4622 /* set obd_namespace for compatibility with old code */
4623 obd->obd_namespace = m->mdt_namespace;
4625 rc = mdt_hsm_cdt_init(m);
4627 CERROR("%s: error initializing coordinator, rc %d\n",
4628 mdt_obd_name(m), rc);
4629 GOTO(err_free_ns, rc);
4632 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4633 OBD_FAIL_MDS_ALL_REQUEST_NET,
4634 OBD_FAIL_MDS_ALL_REPLY_NET);
4636 GOTO(err_free_hsm, rc);
4638 rc = mdt_fs_setup(env, m, obd, lsi);
4642 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4644 next = m->mdt_child;
4645 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4648 GOTO(err_fs_cleanup, rc);
4650 if (mntopts & MNTOPT_USERXATTR)
4651 m->mdt_opts.mo_user_xattr = 1;
4653 m->mdt_opts.mo_user_xattr = 0;
4655 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4657 GOTO(err_fs_cleanup, rc);
4659 if (mntopts & MNTOPT_ACL)
4660 m->mdt_opts.mo_acl = 1;
4662 m->mdt_opts.mo_acl = 0;
4664 /* XXX: to support suppgid for ACL, we enable identity_upcall
4665 * by default, otherwise, maybe got unexpected -EACCESS. */
4666 if (m->mdt_opts.mo_acl)
4667 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4669 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4671 &mdt_identity_upcall_cache_ops);
4672 if (IS_ERR(m->mdt_identity_cache)) {
4673 rc = PTR_ERR(m->mdt_identity_cache);
4674 m->mdt_identity_cache = NULL;
4675 GOTO(err_fs_cleanup, rc);
4678 rc = mdt_procfs_init(m, dev);
4680 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4681 GOTO(err_recovery, rc);
4684 rc = mdt_quota_init(env, m, cfg);
4686 GOTO(err_procfs, rc);
4688 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4689 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4690 "mdt_ldlm_client", m->mdt_ldlm_client);
4692 ping_evictor_start();
4694 /* recovery will be started upon mdt_prepare()
4695 * when the whole stack is complete and ready
4696 * to serve the requests */
4698 /* Reduce the initial timeout on an MDS because it doesn't need such
4699 * a long timeout as an OST does. Adaptive timeouts will adjust this
4700 * value appropriately. */
4701 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4702 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4708 target_recovery_fini(obd);
4709 upcall_cache_cleanup(m->mdt_identity_cache);
4710 m->mdt_identity_cache = NULL;
4712 mdt_fs_cleanup(env, m);
4714 tgt_fini(env, &m->mdt_lut);
4716 mdt_hsm_cdt_fini(m);
4718 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4719 obd->obd_namespace = m->mdt_namespace = NULL;
4721 mdt_seq_fini(env, m);
4723 mdt_fld_fini(env, m);
4725 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4728 server_put_mount(dev, true);
4732 /* For interoperability, the left element is old parameter, the right one
4733 * is the new version of the parameter, if some parameter is deprecated,
4734 * the new version should be set as NULL. */
4735 static struct cfg_interop_param mdt_interop_param[] = {
4736 { "mdt.group_upcall", NULL },
4737 { "mdt.quota_type", NULL },
4738 { "mdd.quota_type", NULL },
4739 { "mdt.rootsquash", "mdt.root_squash" },
4740 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4744 /* used by MGS to process specific configurations */
4745 static int mdt_process_config(const struct lu_env *env,
4746 struct lu_device *d, struct lustre_cfg *cfg)
4748 struct mdt_device *m = mdt_dev(d);
4749 struct md_device *md_next = m->mdt_child;
4750 struct lu_device *next = md2lu_dev(md_next);
4754 switch (cfg->lcfg_command) {
4756 struct obd_device *obd = d->ld_obd;
4758 /* For interoperability */
4759 struct cfg_interop_param *ptr = NULL;
4760 struct lustre_cfg *old_cfg = NULL;
4763 param = lustre_cfg_string(cfg, 1);
4764 if (param == NULL) {
4765 CERROR("param is empty\n");
4770 ptr = class_find_old_param(param, mdt_interop_param);
4772 if (ptr->new_param == NULL) {
4774 CWARN("For interoperability, skip this %s."
4775 " It is obsolete.\n", ptr->old_param);
4779 CWARN("Found old param %s, changed it to %s.\n",
4780 ptr->old_param, ptr->new_param);
4783 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4790 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4792 if (rc > 0 || rc == -ENOSYS) {
4793 /* is it an HSM var ? */
4794 rc = class_process_proc_param(PARAM_HSM,
4795 hsm_cdt_get_proc_vars(),
4797 if (rc > 0 || rc == -ENOSYS)
4798 /* we don't understand; pass it on */
4799 rc = next->ld_ops->ldo_process_config(env, next,
4803 if (old_cfg != NULL)
4804 lustre_cfg_free(cfg);
4809 /* others are passed further */
4810 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4816 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4817 const struct lu_object_header *hdr,
4818 struct lu_device *d)
4820 struct mdt_object *mo;
4824 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4826 struct lu_object *o;
4827 struct lu_object_header *h;
4830 h = &mo->mot_header;
4831 lu_object_header_init(h);
4832 lu_object_init(o, h, d);
4833 lu_object_add_top(h, o);
4834 o->lo_ops = &mdt_obj_ops;
4835 spin_lock_init(&mo->mot_write_lock);
4836 mutex_init(&mo->mot_lov_mutex);
4837 init_rwsem(&mo->mot_open_sem);
4843 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4844 const struct lu_object_conf *unused)
4846 struct mdt_device *d = mdt_dev(o->lo_dev);
4847 struct lu_device *under;
4848 struct lu_object *below;
4852 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4853 PFID(lu_object_fid(o)));
4855 under = &d->mdt_child->md_lu_dev;
4856 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4857 if (below != NULL) {
4858 lu_object_add(o, below);
4865 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4867 struct mdt_object *mo = mdt_obj(o);
4868 struct lu_object_header *h;
4872 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4873 PFID(lu_object_fid(o)));
4875 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4876 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4879 lu_object_header_fini(h);
4880 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4885 static int mdt_object_print(const struct lu_env *env, void *cookie,
4886 lu_printer_t p, const struct lu_object *o)
4888 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4890 return (*p)(env, cookie,
4891 LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4892 mdto, mdto->mot_flags, mdto->mot_write_count);
4895 static int mdt_prepare(const struct lu_env *env,
4896 struct lu_device *pdev,
4897 struct lu_device *cdev)
4899 struct mdt_device *mdt = mdt_dev(cdev);
4900 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4901 struct obd_device *obd = cdev->ld_obd;
4908 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4912 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4916 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4920 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4921 /* The LFSCK instance is registered just now, so it must be there when
4922 * register the namespace to such instance. */
4923 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4925 if (mdt->mdt_seq_site.ss_node_id == 0) {
4926 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4927 &mdt->mdt_md_root_fid);
4932 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4934 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4935 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4936 LASSERT(obd->obd_no_conn);
4937 spin_lock(&obd->obd_dev_lock);
4938 obd->obd_no_conn = 0;
4939 spin_unlock(&obd->obd_dev_lock);
4941 if (obd->obd_recovering == 0)
4942 mdt_postrecov(env, mdt);
4947 const struct lu_device_operations mdt_lu_ops = {
4948 .ldo_object_alloc = mdt_object_alloc,
4949 .ldo_process_config = mdt_process_config,
4950 .ldo_prepare = mdt_prepare,
4953 static const struct lu_object_operations mdt_obj_ops = {
4954 .loo_object_init = mdt_object_init,
4955 .loo_object_free = mdt_object_free,
4956 .loo_object_print = mdt_object_print
4959 static int mdt_obd_set_info_async(const struct lu_env *env,
4960 struct obd_export *exp,
4961 __u32 keylen, void *key,
4962 __u32 vallen, void *val,
4963 struct ptlrpc_request_set *set)
4969 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4970 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4978 * Match client and server connection feature flags.
4980 * Compute the compatibility flags for a connection request based on
4981 * features mutually supported by client and server.
4983 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4984 * must not be updated here, otherwise a partially initialized value may
4985 * be exposed. After the connection request is successfully processed,
4986 * the top-level MDT connect request handler atomically updates the export
4987 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4988 * reply. \see mdt_connect().
4990 * \param exp the obd_export associated with this client/target pair
4991 * \param mdt the target device for the connection
4992 * \param data stores data for this connect request
4995 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4996 * \retval -EBADE client and server feature requirements are incompatible
4998 static int mdt_connect_internal(struct obd_export *exp,
4999 struct mdt_device *mdt,
5000 struct obd_connect_data *data)
5002 LASSERT(data != NULL);
5004 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5005 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5007 if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5008 !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5009 CWARN("%s: client %s does not support ibits lock, either "
5010 "very old or an invalid client: flags "LPX64"\n",
5011 mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5012 data->ocd_connect_flags);
5016 if (!mdt->mdt_opts.mo_acl)
5017 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5019 if (!mdt->mdt_opts.mo_user_xattr)
5020 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5022 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5023 data->ocd_brw_size = min(data->ocd_brw_size,
5024 (__u32)MD_MAX_BRW_SIZE);
5025 if (data->ocd_brw_size == 0) {
5026 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5027 " ocd_version: %x ocd_grant: %d "
5028 "ocd_index: %u ocd_brw_size is "
5029 "unexpectedly zero, network data "
5030 "corruption? Refusing connection of this"
5033 exp->exp_client_uuid.uuid,
5034 exp, data->ocd_connect_flags, data->ocd_version,
5035 data->ocd_grant, data->ocd_index);
5040 /* NB: Disregard the rule against updating
5041 * exp_connect_data.ocd_connect_flags in this case, since
5042 * tgt_client_new() needs to know if this is a lightweight
5043 * connection, and it is safe to expose this flag before
5044 * connection processing completes. */
5045 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5046 spin_lock(&exp->exp_lock);
5047 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5048 spin_unlock(&exp->exp_lock);
5051 data->ocd_version = LUSTRE_VERSION_CODE;
5053 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5054 CWARN("%s: MDS requires FID support, but client not\n",
5059 if (OCD_HAS_FLAG(data, PINGLESS)) {
5060 if (ptlrpc_pinger_suppress_pings()) {
5061 spin_lock(&exp->exp_obd->obd_dev_lock);
5062 list_del_init(&exp->exp_obd_chain_timed);
5063 spin_unlock(&exp->exp_obd->obd_dev_lock);
5065 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5069 data->ocd_max_easize = mdt->mdt_max_ea_size;
5071 /* NB: Disregard the rule against updating
5072 * exp_connect_data.ocd_connect_flags in this case, since
5073 * tgt_client_new() needs to know if this is client supports
5074 * multiple modify RPCs, and it is safe to expose this flag before
5075 * connection processing completes. */
5076 if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5077 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5078 spin_lock(&exp->exp_lock);
5079 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5080 spin_unlock(&exp->exp_lock);
5086 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5087 struct mdt_thread_info *info,
5088 struct mdt_file_data *mfd)
5090 struct lu_context ses;
5094 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5099 lu_context_enter(&ses);
5101 mdt_ucred(info)->uc_valid = UCRED_OLD;
5102 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5104 lu_context_exit(&ses);
5105 lu_context_fini(&ses);
5111 static int mdt_export_cleanup(struct obd_export *exp)
5113 struct list_head closing_list;
5114 struct mdt_export_data *med = &exp->exp_mdt_data;
5115 struct obd_device *obd = exp->exp_obd;
5116 struct mdt_device *mdt;
5117 struct mdt_thread_info *info;
5119 struct mdt_file_data *mfd, *n;
5123 INIT_LIST_HEAD(&closing_list);
5124 spin_lock(&med->med_open_lock);
5125 while (!list_empty(&med->med_open_head)) {
5126 struct list_head *tmp = med->med_open_head.next;
5127 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5129 /* Remove mfd handle so it can't be found again.
5130 * We are consuming the mfd_list reference here. */
5131 class_handle_unhash(&mfd->mfd_handle);
5132 list_move_tail(&mfd->mfd_list, &closing_list);
5134 spin_unlock(&med->med_open_lock);
5135 mdt = mdt_dev(obd->obd_lu_dev);
5136 LASSERT(mdt != NULL);
5138 rc = lu_env_init(&env, LCT_MD_THREAD);
5142 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5143 LASSERT(info != NULL);
5144 memset(info, 0, sizeof *info);
5145 info->mti_env = &env;
5146 info->mti_mdt = mdt;
5147 info->mti_exp = exp;
5149 if (!list_empty(&closing_list)) {
5150 struct md_attr *ma = &info->mti_attr;
5152 /* Close any open files (which may also cause orphan
5154 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5155 list_del_init(&mfd->mfd_list);
5156 ma->ma_need = ma->ma_valid = 0;
5158 /* This file is being closed due to an eviction, it
5159 * could have been modified and now dirty regarding to
5160 * HSM archive, check this!
5161 * The logic here is to mark a file dirty if there's a
5162 * chance it was dirtied before the client was evicted,
5163 * so that we don't have to wait for a release attempt
5164 * before finding out the file was actually dirty and
5165 * fail the release. Aggressively marking it dirty here
5166 * will cause the policy engine to attempt to
5167 * re-archive it; when rearchiving, we can compare the
5168 * current version to the HSM data_version and make the
5169 * archive request into a noop if it's not actually
5172 if (mfd->mfd_mode & FMODE_WRITE)
5173 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5175 /* Don't unlink orphan on failover umount, LU-184 */
5176 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5177 ma->ma_valid = MA_FLAGS;
5178 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5180 mdt_mfd_close(info, mfd);
5183 info->mti_mdt = NULL;
5184 /* cleanup client slot early */
5185 /* Do not erase record for recoverable client. */
5186 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5187 tgt_client_del(&env, exp);
5193 static inline void mdt_enable_slc(struct mdt_device *mdt)
5195 if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5196 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5199 static inline void mdt_disable_slc(struct mdt_device *mdt)
5201 if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5202 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5205 static int mdt_obd_disconnect(struct obd_export *exp)
5211 class_export_get(exp);
5213 if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5214 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5215 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5217 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5218 mdt_disable_slc(mdt);
5221 rc = server_disconnect_export(exp);
5223 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5225 rc = mdt_export_cleanup(exp);
5226 nodemap_del_member(exp);
5227 class_export_put(exp);
5231 /* mds_connect copy */
5232 static int mdt_obd_connect(const struct lu_env *env,
5233 struct obd_export **exp, struct obd_device *obd,
5234 struct obd_uuid *cluuid,
5235 struct obd_connect_data *data,
5238 struct obd_export *lexp;
5239 struct lustre_handle conn = { 0 };
5240 struct mdt_device *mdt;
5242 lnet_nid_t *client_nid = localdata;
5245 LASSERT(env != NULL);
5246 if (!exp || !obd || !cluuid)
5249 mdt = mdt_dev(obd->obd_lu_dev);
5251 if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5252 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5253 atomic_inc(&mdt->mdt_mds_mds_conns);
5254 mdt_enable_slc(mdt);
5258 * first, check whether the stack is ready to handle requests
5259 * XXX: probably not very appropriate method is used now
5260 * at some point we should find a better one
5262 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5263 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5264 !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5265 rc = obd_get_info(env, mdt->mdt_child_exp,
5266 sizeof(KEY_OSP_CONNECTED),
5267 KEY_OSP_CONNECTED, NULL, NULL);
5270 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5273 rc = class_connect(&conn, obd, cluuid);
5277 lexp = class_conn2export(&conn);
5278 LASSERT(lexp != NULL);
5280 rc = nodemap_add_member(*client_nid, lexp);
5281 if (rc != 0 && rc != -EEXIST)
5284 rc = mdt_connect_internal(lexp, mdt, data);
5286 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5289 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5290 rc = tgt_client_new(env, lexp);
5292 mdt_export_stats_init(obd, lexp, localdata);
5296 class_disconnect(lexp);
5297 nodemap_del_member(lexp);
5301 /* Because we do not want this export to be evicted by pinger,
5302 * let's not add this export to the timed chain list. */
5303 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5304 spin_lock(&lexp->exp_obd->obd_dev_lock);
5305 list_del_init(&lexp->exp_obd_chain_timed);
5306 spin_unlock(&lexp->exp_obd->obd_dev_lock);
5313 static int mdt_obd_reconnect(const struct lu_env *env,
5314 struct obd_export *exp, struct obd_device *obd,
5315 struct obd_uuid *cluuid,
5316 struct obd_connect_data *data,
5319 lnet_nid_t *client_nid = localdata;
5323 if (exp == NULL || obd == NULL || cluuid == NULL)
5326 rc = nodemap_add_member(*client_nid, exp);
5327 if (rc != 0 && rc != -EEXIST)
5330 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5332 mdt_export_stats_init(obd, exp, localdata);
5334 nodemap_del_member(exp);
5339 /* FIXME: Can we avoid using these two interfaces? */
5340 static int mdt_init_export(struct obd_export *exp)
5342 struct mdt_export_data *med = &exp->exp_mdt_data;
5346 INIT_LIST_HEAD(&med->med_open_head);
5347 spin_lock_init(&med->med_open_lock);
5348 mutex_init(&med->med_idmap_mutex);
5349 med->med_idmap = NULL;
5350 spin_lock(&exp->exp_lock);
5351 exp->exp_connecting = 1;
5352 spin_unlock(&exp->exp_lock);
5354 /* self-export doesn't need client data and ldlm initialization */
5355 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5356 &exp->exp_client_uuid)))
5359 rc = tgt_client_alloc(exp);
5363 rc = ldlm_init_export(exp);
5370 tgt_client_free(exp);
5372 CERROR("%s: Failed to initialize export: rc = %d\n",
5373 exp->exp_obd->obd_name, rc);
5377 static int mdt_destroy_export(struct obd_export *exp)
5381 if (exp_connect_rmtclient(exp))
5382 mdt_cleanup_idmap(&exp->exp_mdt_data);
5384 target_destroy_export(exp);
5385 /* destroy can be called from failed obd_setup, so
5386 * checking uuid is safer than obd_self_export */
5387 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5388 &exp->exp_client_uuid)))
5391 ldlm_destroy_export(exp);
5392 tgt_client_free(exp);
5394 LASSERT(list_empty(&exp->exp_outstanding_replies));
5395 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5400 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5401 struct linkea_data *ldata)
5405 LASSERT(ldata->ld_buf->lb_buf != NULL);
5407 if (!mdt_object_exists(mdt_obj))
5410 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5411 ldata->ld_buf, XATTR_NAME_LINK);
5412 if (rc == -ERANGE) {
5413 /* Buf was too small, figure out what we need. */
5414 lu_buf_free(ldata->ld_buf);
5415 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5416 ldata->ld_buf, XATTR_NAME_LINK);
5419 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5420 if (ldata->ld_buf->lb_buf == NULL)
5422 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5423 ldata->ld_buf, XATTR_NAME_LINK);
5428 return linkea_init(ldata);
5432 * Given an MDT object, try to look up the full path to the object.
5433 * Part of the MDT layer implementation of lfs fid2path.
5435 * \param[in] info Per-thread common data shared by MDT level handlers.
5436 * \param[in] obj Object to do path lookup of
5437 * \param[in,out] fp User-provided struct to store path information
5438 * \param[in] root_fid Root FID of current path should reach
5440 * \retval 0 Lookup successful, path information stored in fp
5441 * \retval -EAGAIN Lookup failed, usually because object is being moved
5442 * \retval negative errno if there was a problem
5444 static int mdt_path_current(struct mdt_thread_info *info,
5445 struct mdt_object *obj,
5446 struct getinfo_fid2path *fp,
5447 struct lu_fid *root_fid)
5449 struct mdt_device *mdt = info->mti_mdt;
5450 struct mdt_object *mdt_obj;
5451 struct link_ea_header *leh;
5452 struct link_ea_entry *lee;
5453 struct lu_name *tmpname = &info->mti_name;
5454 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5455 struct lu_buf *buf = &info->mti_big_buf;
5458 struct linkea_data ldata = { NULL };
5463 /* temp buffer for path element, the buffer will be finally freed
5464 * in mdt_thread_info_fini */
5465 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5466 if (buf->lb_buf == NULL)
5470 ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5473 *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5475 while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5476 struct lu_buf lmv_buf;
5478 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5479 lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5480 GOTO(out, rc = -ENOENT);
5482 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5483 if (IS_ERR(mdt_obj))
5484 GOTO(out, rc = PTR_ERR(mdt_obj));
5486 if (!mdt_object_exists(mdt_obj)) {
5487 mdt_object_put(info->mti_env, mdt_obj);
5488 GOTO(out, rc = -ENOENT);
5491 if (mdt_object_remote(mdt_obj)) {
5492 mdt_object_put(info->mti_env, mdt_obj);
5493 GOTO(remote_out, rc = -EREMOTE);
5496 rc = mdt_links_read(info, mdt_obj, &ldata);
5498 mdt_object_put(info->mti_env, mdt_obj);
5503 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5504 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5505 /* If set, use link #linkno for path lookup, otherwise use
5506 link #0. Only do this for the final path element. */
5507 if (first && fp->gf_linkno < leh->leh_reccount) {
5509 for (count = 0; count < fp->gf_linkno; count++) {
5510 lee = (struct link_ea_entry *)
5511 ((char *)lee + reclen);
5512 linkea_entry_unpack(lee, &reclen, tmpname,
5515 if (fp->gf_linkno < leh->leh_reccount - 1)
5516 /* indicate to user there are more links */
5520 lmv_buf.lb_buf = info->mti_xattr_buf;
5521 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5522 /* Check if it is slave stripes */
5523 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5524 &lmv_buf, XATTR_NAME_LMV);
5525 mdt_object_put(info->mti_env, mdt_obj);
5527 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5529 /* For slave stripes, get its master */
5530 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5531 fp->gf_fid = *tmpfid;
5534 } else if (rc < 0 && rc != -ENODATA) {
5540 /* Pack the name in the end of the buffer */
5541 ptr -= tmpname->ln_namelen;
5542 if (ptr - 1 <= fp->gf_u.gf_path)
5543 GOTO(out, rc = -EOVERFLOW);
5544 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5547 /* keep the last resolved fid to the client, so the
5548 * client will build the left path on another MDT for
5550 fp->gf_fid = *tmpfid;
5556 ptr++; /* skip leading / */
5557 memmove(fp->gf_u.gf_path, ptr,
5558 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5565 * Given an MDT object, use mdt_path_current to get the path.
5566 * Essentially a wrapper to retry mdt_path_current a set number of times
5567 * if -EAGAIN is returned (usually because an object is being moved).
5569 * Part of the MDT layer implementation of lfs fid2path.
5571 * \param[in] info Per-thread common data shared by mdt level handlers.
5572 * \param[in] obj Object to do path lookup of
5573 * \param[in,out] fp User-provided struct for arguments and to store path
5576 * \retval 0 Lookup successful, path information stored in fp
5577 * \retval negative errno if there was a problem
5579 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5580 struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5582 struct mdt_device *mdt = info->mti_mdt;
5587 if (fp->gf_pathlen < 3)
5590 if (root_fid == NULL)
5591 root_fid = &mdt->mdt_md_root_fid;
5593 if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5594 fp->gf_u.gf_path[0] = '\0';
5598 /* Retry multiple times in case file is being moved */
5599 while (tries-- && rc == -EAGAIN)
5600 rc = mdt_path_current(info, obj, fp, root_fid);
5606 * Get the full path of the provided FID, as of changelog record recno.
5608 * This checks sanity and looks up object for user provided FID
5609 * before calling the actual path lookup code.
5611 * Part of the MDT layer implementation of lfs fid2path.
5613 * \param[in] info Per-thread common data shared by mdt level handlers.
5614 * \param[in,out] fp User-provided struct for arguments and to store path
5617 * \retval 0 Lookup successful, path information and recno stored in fp
5618 * \retval -ENOENT, object does not exist
5619 * \retval negative errno if there was a problem
5621 static int mdt_fid2path(struct mdt_thread_info *info,
5622 struct lu_fid *root_fid,
5623 struct getinfo_fid2path *fp)
5625 struct mdt_device *mdt = info->mti_mdt;
5626 struct mdt_object *obj;
5630 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5631 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5633 if (!fid_is_sane(&fp->gf_fid))
5636 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5637 CWARN("%s: "DFID" is invalid, sequence should be "
5638 ">= "LPX64"\n", mdt_obd_name(mdt),
5639 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5643 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5646 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5647 PFID(&fp->gf_fid), rc);
5651 if (mdt_object_remote(obj))
5653 else if (!mdt_object_exists(obj))
5659 mdt_object_put(info->mti_env, obj);
5660 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5661 PFID(&fp->gf_fid), rc);
5665 rc = mdt_path(info, obj, fp, root_fid);
5667 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5668 PFID(&fp->gf_fid), fp->gf_u.gf_path,
5669 fp->gf_recno, fp->gf_linkno);
5671 mdt_object_put(info->mti_env, obj);
5676 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5677 void *val, int vallen)
5679 struct getinfo_fid2path *fpout, *fpin;
5680 struct lu_fid *root_fid = NULL;
5683 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5686 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5687 lustre_swab_fid2path(fpin);
5689 memcpy(fpout, fpin, sizeof(*fpin));
5690 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5693 if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5694 sizeof(struct lu_fid)) {
5695 /* client sent its root FID, which is normally fileset FID */
5696 root_fid = fpin->gf_u.gf_root_fid;
5697 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5698 lustre_swab_lu_fid(root_fid);
5700 if (root_fid != NULL && !fid_is_sane(root_fid))
5704 rc = mdt_fid2path(info, root_fid, fpout);
5708 int mdt_get_info(struct tgt_session_info *tsi)
5718 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5720 CDEBUG(D_IOCTL, "No GETINFO key\n");
5721 RETURN(err_serious(-EFAULT));
5723 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5726 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5727 if (vallen == NULL) {
5728 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5729 tgt_name(tsi->tsi_tgt));
5730 RETURN(err_serious(-EFAULT));
5733 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5735 rc = req_capsule_server_pack(tsi->tsi_pill);
5737 RETURN(err_serious(rc));
5739 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5740 if (valout == NULL) {
5741 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5742 tgt_name(tsi->tsi_tgt));
5743 RETURN(err_serious(-EFAULT));
5746 if (KEY_IS(KEY_FID2PATH)) {
5747 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5749 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5750 mdt_thread_info_fini(info);
5757 /* Pass the ioc down */
5758 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5759 unsigned int cmd, int len, void *data)
5761 struct lu_context ioctl_session;
5762 struct md_device *next = mdt->mdt_child;
5766 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5769 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5770 lu_context_enter(&ioctl_session);
5771 env->le_ses = &ioctl_session;
5773 LASSERT(next->md_ops->mdo_iocontrol);
5774 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5776 lu_context_exit(&ioctl_session);
5777 lu_context_fini(&ioctl_session);
5781 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5783 struct obd_ioctl_data *data = karg;
5786 struct mdt_object *obj;
5787 struct mdt_lock_handle *lh;
5791 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5792 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5795 fid = (struct lu_fid *)data->ioc_inlbuf1;
5797 if (!fid_is_sane(fid))
5800 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5802 lh = &mti->mti_lh[MDT_LH_PARENT];
5803 mdt_lock_reg_init(lh, LCK_CR);
5805 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5807 RETURN(PTR_ERR(obj));
5809 if (mdt_object_remote(obj)) {
5812 * before calling version get the correct MDS should be
5813 * fid, this is error to find remote object here
5815 CERROR("nonlocal object "DFID"\n", PFID(fid));
5816 } else if (!mdt_object_exists(obj)) {
5817 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5820 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5821 *(__u64 *)data->ioc_inlbuf2 = version;
5824 mdt_object_unlock_put(mti, obj, lh, 1);
5828 /* ioctls on obd dev */
5829 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5830 void *karg, void __user *uarg)
5833 struct obd_device *obd = exp->exp_obd;
5834 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5835 struct dt_device *dt = mdt->mdt_bottom;
5839 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5840 rc = lu_env_init(&env, LCT_MD_THREAD);
5846 rc = mdt_device_sync(&env, mdt);
5848 case OBD_IOC_SET_READONLY:
5849 rc = dt->dd_ops->dt_ro(&env, dt);
5851 case OBD_IOC_ABORT_RECOVERY:
5852 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5853 obd->obd_abort_recovery = 1;
5854 target_stop_recovery_thread(obd);
5857 case OBD_IOC_CHANGELOG_REG:
5858 case OBD_IOC_CHANGELOG_DEREG:
5859 case OBD_IOC_CHANGELOG_CLEAR:
5860 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5862 case OBD_IOC_START_LFSCK: {
5863 struct md_device *next = mdt->mdt_child;
5864 struct obd_ioctl_data *data = karg;
5865 struct lfsck_start_param lsp;
5867 if (unlikely(data == NULL)) {
5872 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5873 lsp.lsp_index_valid = 0;
5874 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5877 case OBD_IOC_STOP_LFSCK: {
5878 struct md_device *next = mdt->mdt_child;
5879 struct obd_ioctl_data *data = karg;
5880 struct lfsck_stop stop;
5882 stop.ls_status = LS_STOPPED;
5883 /* Old lfsck utils may pass NULL @stop. */
5884 if (data->ioc_inlbuf1 == NULL)
5888 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5890 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5893 case OBD_IOC_QUERY_LFSCK: {
5894 struct md_device *next = mdt->mdt_child;
5895 struct obd_ioctl_data *data = karg;
5897 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5901 case OBD_IOC_GET_OBJ_VERSION: {
5902 struct mdt_thread_info *mti;
5903 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5904 memset(mti, 0, sizeof *mti);
5905 mti->mti_env = &env;
5909 rc = mdt_ioc_version_get(mti, karg);
5912 case OBD_IOC_CATLOGLIST: {
5913 struct mdt_thread_info *mti;
5915 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5916 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5917 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5918 &mti->mti_tmp_fid1);
5923 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5924 mdt_obd_name(mdt), cmd, rc);
5931 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5933 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5937 if (!mdt->mdt_skip_lfsck) {
5938 struct lfsck_start_param lsp;
5940 lsp.lsp_start = NULL;
5941 lsp.lsp_index_valid = 0;
5942 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5943 OBD_IOC_START_LFSCK,
5945 if (rc != 0 && rc != -EALREADY)
5946 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5947 mdt_obd_name(mdt), rc);
5950 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5954 static int mdt_obd_postrecov(struct obd_device *obd)
5959 rc = lu_env_init(&env, LCT_MD_THREAD);
5962 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5967 static struct obd_ops mdt_obd_device_ops = {
5968 .o_owner = THIS_MODULE,
5969 .o_set_info_async = mdt_obd_set_info_async,
5970 .o_connect = mdt_obd_connect,
5971 .o_reconnect = mdt_obd_reconnect,
5972 .o_disconnect = mdt_obd_disconnect,
5973 .o_init_export = mdt_init_export,
5974 .o_destroy_export = mdt_destroy_export,
5975 .o_iocontrol = mdt_iocontrol,
5976 .o_postrecov = mdt_obd_postrecov,
5979 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5980 struct lu_device *d)
5982 struct mdt_device *m = mdt_dev(d);
5989 static struct lu_device *mdt_device_free(const struct lu_env *env,
5990 struct lu_device *d)
5992 struct mdt_device *m = mdt_dev(d);
5995 lu_device_fini(&m->mdt_lu_dev);
6001 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6002 struct lu_device_type *t,
6003 struct lustre_cfg *cfg)
6005 struct lu_device *l;
6006 struct mdt_device *m;
6013 rc = mdt_init0(env, m, t, cfg);
6015 mdt_device_free(env, l);
6020 l = ERR_PTR(-ENOMEM);
6024 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6025 LU_KEY_INIT(mdt, struct mdt_thread_info);
6027 static void mdt_key_fini(const struct lu_context *ctx,
6028 struct lu_context_key *key, void* data)
6030 struct mdt_thread_info *info = data;
6032 if (info->mti_big_lmm) {
6033 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6034 info->mti_big_lmm = NULL;
6035 info->mti_big_lmmsize = 0;
6040 /* context key: mdt_thread_key */
6041 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6043 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6045 return lu_ucred(info->mti_env);
6048 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6050 return lu_ucred_check(info->mti_env);
6054 * Enable/disable COS (Commit On Sharing).
6056 * Set/Clear the COS flag in mdt options.
6058 * \param mdt mdt device
6059 * \param val 0 disables COS, other values enable COS
6061 void mdt_enable_cos(struct mdt_device *mdt, int val)
6066 mdt->mdt_opts.mo_cos = !!val;
6067 rc = lu_env_init(&env, LCT_LOCAL);
6068 if (unlikely(rc != 0)) {
6069 CWARN("%s: lu_env initialization failed, cannot "
6070 "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6073 mdt_device_sync(&env, mdt);
6078 * Check COS (Commit On Sharing) status.
6080 * Return COS flag status.
6082 * \param mdt mdt device
6084 int mdt_cos_is_enabled(struct mdt_device *mdt)
6086 return mdt->mdt_opts.mo_cos != 0;
6089 static struct lu_device_type_operations mdt_device_type_ops = {
6090 .ldto_device_alloc = mdt_device_alloc,
6091 .ldto_device_free = mdt_device_free,
6092 .ldto_device_fini = mdt_device_fini
6095 static struct lu_device_type mdt_device_type = {
6096 .ldt_tags = LU_DEVICE_MD,
6097 .ldt_name = LUSTRE_MDT_NAME,
6098 .ldt_ops = &mdt_device_type_ops,
6099 .ldt_ctx_tags = LCT_MD_THREAD
6102 static int __init mdt_init(void)
6106 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6107 FID_NOBRACE_LEN + 1);
6108 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6110 rc = lu_kmem_init(mdt_caches);
6114 rc = mds_mod_init();
6118 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6119 LUSTRE_MDT_NAME, &mdt_device_type);
6124 lu_kmem_fini(mdt_caches);
6131 static void __exit mdt_exit(void)
6133 class_unregister_type(LUSTRE_MDT_NAME);
6135 lu_kmem_fini(mdt_caches);
6138 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6139 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6140 MODULE_VERSION(LUSTRE_VERSION_STRING);
6141 MODULE_LICENSE("GPL");
6143 module_init(mdt_init);
6144 module_exit(mdt_exit);