4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 #include <dt_object.h>
54 #include <lustre_acl.h>
55 #include <lustre_export.h>
56 #include <lustre_ioctl.h>
57 #include <lustre_lfsck.h>
58 #include <lustre_log.h>
59 #include <lustre_net.h>
60 #include <lustre_nodemap.h>
61 #include <lustre_mds.h>
62 #include <lustre_param.h>
63 #include <lustre_quota.h>
64 #include <lustre_swab.h>
66 #include <obd_support.h>
68 #include <llog_swab.h>
70 #include "mdt_internal.h"
73 static unsigned int max_mod_rpcs_per_client = 8;
74 module_param(max_mod_rpcs_per_client, uint, 0644);
75 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78 [LCK_MINMODE] = MDL_MINMODE,
85 [LCK_GROUP] = MDL_GROUP
88 enum ldlm_mode mdt_dlm_lock_modes[] = {
89 [MDL_MINMODE] = LCK_MINMODE,
96 [MDL_GROUP] = LCK_GROUP
99 static struct mdt_device *mdt_dev(struct lu_device *d);
100 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
102 static const struct lu_object_operations mdt_obj_ops;
104 /* Slab for MDT object allocation */
105 static struct kmem_cache *mdt_object_kmem;
107 /* For HSM restore handles */
108 struct kmem_cache *mdt_hsm_cdt_kmem;
110 /* For HSM request handles */
111 struct kmem_cache *mdt_hsm_car_kmem;
113 static struct lu_kmem_descr mdt_caches[] = {
115 .ckd_cache = &mdt_object_kmem,
116 .ckd_name = "mdt_obj",
117 .ckd_size = sizeof(struct mdt_object)
120 .ckd_cache = &mdt_hsm_cdt_kmem,
121 .ckd_name = "mdt_cdt_restore_handle",
122 .ckd_size = sizeof(struct cdt_restore_handle)
125 .ckd_cache = &mdt_hsm_car_kmem,
126 .ckd_name = "mdt_cdt_agent_req",
127 .ckd_size = sizeof(struct cdt_agent_req)
134 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
138 return rep->lock_policy_res1 & op_flag;
141 void mdt_clear_disposition(struct mdt_thread_info *info,
142 struct ldlm_reply *rep, __u64 op_flag)
145 info->mti_opdata &= ~op_flag;
146 tgt_opdata_clear(info->mti_env, op_flag);
149 rep->lock_policy_res1 &= ~op_flag;
152 void mdt_set_disposition(struct mdt_thread_info *info,
153 struct ldlm_reply *rep, __u64 op_flag)
156 info->mti_opdata |= op_flag;
157 tgt_opdata_set(info->mti_env, op_flag);
160 rep->lock_policy_res1 |= op_flag;
163 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
165 lh->mlh_pdo_hash = 0;
166 lh->mlh_reg_mode = lm;
167 lh->mlh_rreg_mode = lm;
168 lh->mlh_type = MDT_REG_LOCK;
171 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
172 const struct lu_name *lname)
174 lh->mlh_reg_mode = lock_mode;
175 lh->mlh_pdo_mode = LCK_MINMODE;
176 lh->mlh_rreg_mode = lock_mode;
177 lh->mlh_type = MDT_PDO_LOCK;
179 if (lu_name_is_valid(lname)) {
180 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
182 /* XXX Workaround for LU-2856
184 * Zero is a valid return value of full_name_hash, but
185 * several users of mlh_pdo_hash assume a non-zero
186 * hash value. We therefore map zero onto an
187 * arbitrary, but consistent value (1) to avoid
188 * problems further down the road. */
189 if (unlikely(lh->mlh_pdo_hash == 0))
190 lh->mlh_pdo_hash = 1;
192 lh->mlh_pdo_hash = 0;
196 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
197 struct mdt_lock_handle *lh)
203 * Any dir access needs couple of locks:
205 * 1) on part of dir we gonna take lookup/modify;
207 * 2) on whole dir to protect it from concurrent splitting and/or to
208 * flush client's cache for readdir().
210 * so, for a given mode and object this routine decides what lock mode
211 * to use for lock #2:
213 * 1) if caller's gonna lookup in dir then we need to protect dir from
214 * being splitted only - LCK_CR
216 * 2) if caller's gonna modify dir then we need to protect dir from
217 * being splitted and to flush cache - LCK_CW
219 * 3) if caller's gonna modify dir and that dir seems ready for
220 * splitting then we need to protect it from any type of access
221 * (lookup/modify/split) - LCK_EX --bzzz
224 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
225 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
228 * Ask underlaying level its opinion about preferable PDO lock mode
229 * having access type passed as regular lock mode:
231 * - MDL_MINMODE means that lower layer does not want to specify lock
234 * - MDL_NL means that no PDO lock should be taken. This is used in some
235 * cases. Say, for non-splittable directories no need to use PDO locks
238 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
239 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
241 if (mode != MDL_MINMODE) {
242 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
245 * Lower layer does not want to specify locking mode. We do it
246 * our selves. No special protection is needed, just flush
247 * client's cache on modification and allow concurrent
250 switch (lh->mlh_reg_mode) {
252 lh->mlh_pdo_mode = LCK_EX;
255 lh->mlh_pdo_mode = LCK_CR;
258 lh->mlh_pdo_mode = LCK_CW;
261 CERROR("Not expected lock type (0x%x)\n",
262 (int)lh->mlh_reg_mode);
267 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
271 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
274 struct mdt_device *mdt = info->mti_mdt;
275 struct lu_name *lname = &info->mti_name;
277 struct mdt_object *parent;
281 LASSERT(!info->mti_cross_ref);
283 OBD_ALLOC(name, NAME_MAX + 1);
286 lname->ln_name = name;
289 * We may want to allow this to mount a completely separate
290 * fileset from the MDT in the future, but keeping it to
291 * ROOT/ only for now avoid potential security issues.
293 *fid = mdt->mdt_md_root_fid;
295 while (rc == 0 && fileset != NULL && *fileset != '\0') {
296 const char *s1 = fileset;
302 while (*s2 != '/' && *s2 != '\0')
310 lname->ln_namelen = s2 - s1;
311 if (lname->ln_namelen > NAME_MAX) {
316 /* reject .. as a path component */
317 if (lname->ln_namelen == 2 &&
318 strncmp(s1, "..", 2) == 0) {
323 strncpy(name, s1, lname->ln_namelen);
324 name[lname->ln_namelen] = '\0';
326 parent = mdt_object_find(info->mti_env, mdt, fid);
327 if (IS_ERR(parent)) {
328 rc = PTR_ERR(parent);
331 /* Only got the fid of this obj by name */
333 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
334 fid, &info->mti_spec);
335 mdt_object_put(info->mti_env, parent);
338 parent = mdt_object_find(info->mti_env, mdt, fid);
340 rc = PTR_ERR(parent);
342 mode = lu_object_attr(&parent->mot_obj);
343 mdt_object_put(info->mti_env, parent);
349 OBD_FREE(name, NAME_MAX + 1);
354 static int mdt_get_root(struct tgt_session_info *tsi)
356 struct mdt_thread_info *info = tsi2mdt_info(tsi);
357 struct mdt_device *mdt = info->mti_mdt;
358 struct mdt_body *repbody;
359 char *fileset = NULL, *buffer = NULL;
361 struct obd_export *exp = info->mti_exp;
362 char *nodemap_fileset;
366 rc = mdt_check_ucred(info);
368 GOTO(out, rc = err_serious(rc));
370 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
371 GOTO(out, rc = err_serious(-ENOMEM));
373 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
374 if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
375 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
377 GOTO(out, rc = err_serious(-EFAULT));
380 nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
381 if (nodemap_fileset && nodemap_fileset[0]) {
383 /* consider fileset from client as a sub-fileset
384 * of the nodemap one */
385 OBD_ALLOC(buffer, PATH_MAX + 1);
387 GOTO(out, rc = err_serious(-ENOMEM));
388 if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
389 nodemap_fileset, fileset) >= PATH_MAX + 1)
390 GOTO(out, rc = err_serious(-EINVAL));
393 /* enforce fileset as specified in the nodemap */
394 fileset = nodemap_fileset;
399 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
401 GOTO(out, rc = err_serious(rc));
403 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
405 repbody->mbo_valid |= OBD_MD_FLID;
409 mdt_thread_info_fini(info);
411 OBD_FREE(buffer, PATH_MAX+1);
415 static int mdt_statfs(struct tgt_session_info *tsi)
417 struct ptlrpc_request *req = tgt_ses_req(tsi);
418 struct mdt_thread_info *info = tsi2mdt_info(tsi);
419 struct md_device *next = info->mti_mdt->mdt_child;
420 struct ptlrpc_service_part *svcpt;
421 struct obd_statfs *osfs;
426 svcpt = req->rq_rqbd->rqbd_svcpt;
428 /* This will trigger a watchdog timeout */
429 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
430 (MDT_SERVICE_WATCHDOG_FACTOR *
431 at_get(&svcpt->scp_at_estimate)) + 1);
433 rc = mdt_check_ucred(info);
435 GOTO(out, rc = err_serious(rc));
437 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
438 GOTO(out, rc = err_serious(-ENOMEM));
440 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
442 GOTO(out, rc = -EPROTO);
444 /** statfs information are cached in the mdt_device */
445 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
446 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
447 /** statfs data is too old, get up-to-date one */
448 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
451 spin_lock(&info->mti_mdt->mdt_lock);
452 info->mti_mdt->mdt_osfs = *osfs;
453 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
454 spin_unlock(&info->mti_mdt->mdt_lock);
456 /** use cached statfs data */
457 spin_lock(&info->mti_mdt->mdt_lock);
458 *osfs = info->mti_mdt->mdt_osfs;
459 spin_unlock(&info->mti_mdt->mdt_lock);
463 mdt_counter_incr(req, LPROC_MDT_STATFS);
465 mdt_thread_info_fini(info);
469 #ifdef CONFIG_FS_POSIX_ACL
471 * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
473 * \param info thread info object
474 * \param repbody reply to pack ACLs into
475 * \param o mdt object of file to examine
476 * \param nodemap nodemap of client to reply to
478 * \retval -errno error getting or parsing ACL from disk
480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
481 struct mdt_object *o, struct lu_nodemap *nodemap)
483 const struct lu_env *env = info->mti_env;
484 struct md_object *next = mdt_object_child(o);
485 struct lu_buf *buf = &info->mti_buf;
488 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
489 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
491 if (buf->lb_len == 0)
494 rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
496 if (rc == -ENODATA) {
497 repbody->mbo_aclsize = 0;
498 repbody->mbo_valid |= OBD_MD_FLACL;
500 } else if (rc == -EOPNOTSUPP) {
503 CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
504 mdt_obd_name(info->mti_mdt),
505 PFID(mdt_object_fid(o)), rc);
508 rc = nodemap_map_acl(nodemap, buf->lb_buf,
509 rc, NODEMAP_FS_TO_CLIENT);
510 /* if all ACLs mapped out, rc is still >= 0 */
512 CERROR("%s: nodemap_map_acl unable to parse "DFID
513 " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
514 PFID(mdt_object_fid(o)), rc);
516 repbody->mbo_aclsize = rc;
517 repbody->mbo_valid |= OBD_MD_FLACL;
525 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
526 const struct lu_attr *attr, const struct lu_fid *fid)
528 struct md_attr *ma = &info->mti_attr;
529 struct obd_export *exp = info->mti_exp;
530 struct lu_nodemap *nodemap = NULL;
532 LASSERT(ma->ma_valid & MA_INODE);
534 if (attr->la_valid & LA_ATIME) {
535 b->mbo_atime = attr->la_atime;
536 b->mbo_valid |= OBD_MD_FLATIME;
538 if (attr->la_valid & LA_MTIME) {
539 b->mbo_mtime = attr->la_mtime;
540 b->mbo_valid |= OBD_MD_FLMTIME;
542 if (attr->la_valid & LA_CTIME) {
543 b->mbo_ctime = attr->la_ctime;
544 b->mbo_valid |= OBD_MD_FLCTIME;
546 if (attr->la_valid & LA_FLAGS) {
547 b->mbo_flags = attr->la_flags;
548 b->mbo_valid |= OBD_MD_FLFLAGS;
550 if (attr->la_valid & LA_NLINK) {
551 b->mbo_nlink = attr->la_nlink;
552 b->mbo_valid |= OBD_MD_FLNLINK;
554 if (attr->la_valid & (LA_UID|LA_GID)) {
555 nodemap = nodemap_get_from_exp(exp);
559 if (attr->la_valid & LA_UID) {
560 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
561 NODEMAP_FS_TO_CLIENT,
563 b->mbo_valid |= OBD_MD_FLUID;
565 if (attr->la_valid & LA_GID) {
566 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
567 NODEMAP_FS_TO_CLIENT,
569 b->mbo_valid |= OBD_MD_FLGID;
572 b->mbo_mode = attr->la_mode;
573 if (attr->la_valid & LA_MODE)
574 b->mbo_valid |= OBD_MD_FLMODE;
575 if (attr->la_valid & LA_TYPE)
576 b->mbo_valid |= OBD_MD_FLTYPE;
580 b->mbo_valid |= OBD_MD_FLID;
581 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
582 PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
585 if (!(attr->la_valid & LA_TYPE))
588 b->mbo_rdev = attr->la_rdev;
589 b->mbo_size = attr->la_size;
590 b->mbo_blocks = attr->la_blocks;
592 if (!S_ISREG(attr->la_mode)) {
593 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
594 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
595 /* means no objects are allocated on osts. */
596 LASSERT(!(ma->ma_valid & MA_LOV));
597 /* just ignore blocks occupied by extend attributes on MDS */
599 /* if no object is allocated on osts, the size on mds is valid.
601 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
602 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
603 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
604 /* A released file stores its size on MDS. */
605 /* But return 1 block for released file, unless tools like tar
606 * will consider it fully sparse. (LU-3864)
608 if (unlikely(b->mbo_size == 0))
612 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
615 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
616 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
617 PFID(fid), (unsigned long long)b->mbo_size);
620 if (!IS_ERR_OR_NULL(nodemap))
621 nodemap_putref(nodemap);
624 static inline int mdt_body_has_lov(const struct lu_attr *la,
625 const struct mdt_body *body)
627 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
628 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
631 void mdt_client_compatibility(struct mdt_thread_info *info)
633 struct mdt_body *body;
634 struct ptlrpc_request *req = mdt_info_req(info);
635 struct obd_export *exp = req->rq_export;
636 struct md_attr *ma = &info->mti_attr;
637 struct lu_attr *la = &ma->ma_attr;
640 if (exp_connect_layout(exp))
641 /* the client can deal with 16-bit lmm_stripe_count */
644 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
646 if (!mdt_body_has_lov(la, body))
649 /* now we have a reply with a lov for a client not compatible with the
650 * layout lock so we have to clean the layout generation number */
651 if (S_ISREG(la->la_mode))
652 ma->ma_lmm->lmm_layout_gen = 0;
656 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
657 struct mdt_object *o)
659 const struct lu_env *env = info->mti_env;
662 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
671 /* Is it a directory? Let's check for the LMV as well */
672 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
673 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
677 rc2 = mo_xattr_get(env, mdt_object_child(o),
679 XATTR_NAME_DEFAULT_LMV);
681 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
689 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
692 const struct lu_env *env = info->mti_env;
696 LASSERT(info->mti_big_lmm_used == 0);
697 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
701 /* big_lmm may need to be grown */
702 if (info->mti_big_lmmsize < rc) {
703 int size = size_roundup_power2(rc);
705 if (info->mti_big_lmmsize > 0) {
706 /* free old buffer */
707 LASSERT(info->mti_big_lmm);
708 OBD_FREE_LARGE(info->mti_big_lmm,
709 info->mti_big_lmmsize);
710 info->mti_big_lmm = NULL;
711 info->mti_big_lmmsize = 0;
714 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
715 if (info->mti_big_lmm == NULL)
717 info->mti_big_lmmsize = size;
719 LASSERT(info->mti_big_lmmsize >= rc);
721 info->mti_buf.lb_buf = info->mti_big_lmm;
722 info->mti_buf.lb_len = info->mti_big_lmmsize;
723 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
728 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
729 struct md_attr *ma, const char *name)
731 struct md_object *next = mdt_object_child(o);
732 struct lu_buf *buf = &info->mti_buf;
735 if (strcmp(name, XATTR_NAME_LOV) == 0) {
736 buf->lb_buf = ma->ma_lmm;
737 buf->lb_len = ma->ma_lmm_size;
738 LASSERT(!(ma->ma_valid & MA_LOV));
739 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
740 buf->lb_buf = ma->ma_lmv;
741 buf->lb_len = ma->ma_lmv_size;
742 LASSERT(!(ma->ma_valid & MA_LMV));
743 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
744 buf->lb_buf = ma->ma_lmv;
745 buf->lb_len = ma->ma_lmv_size;
746 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
751 rc = mo_xattr_get(info->mti_env, next, buf, name);
755 if (strcmp(name, XATTR_NAME_LOV) == 0) {
756 if (info->mti_big_lmm_used)
757 ma->ma_lmm = info->mti_big_lmm;
759 /* NOT return LOV EA with hole to old client. */
760 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
761 LOV_PATTERN_F_HOLE) &&
762 !(exp_connect_flags(info->mti_exp) &
763 OBD_CONNECT_LFSCK)) {
766 ma->ma_lmm_size = rc;
767 ma->ma_valid |= MA_LOV;
769 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
770 if (info->mti_big_lmm_used)
771 ma->ma_lmv = info->mti_big_lmm;
773 ma->ma_lmv_size = rc;
774 ma->ma_valid |= MA_LMV;
775 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
776 ma->ma_lmv_size = rc;
777 ma->ma_valid |= MA_LMV_DEF;
780 /* Update mdt_max_mdsize so all clients will be aware that */
781 if (info->mti_mdt->mdt_max_mdsize < rc)
782 info->mti_mdt->mdt_max_mdsize = rc;
785 } else if (rc == -ENODATA) {
788 } else if (rc == -ERANGE) {
789 /* Default LMV has fixed size, so it must be able to fit
790 * in the original buffer */
791 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
793 rc = mdt_big_xattr_get(info, o, name);
795 info->mti_big_lmm_used = 1;
803 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
804 struct mdt_object *o, struct lu_fid *pfid)
806 struct lu_buf *buf = &info->mti_buf;
807 struct link_ea_header *leh;
808 struct link_ea_entry *lee;
812 buf->lb_buf = info->mti_big_lmm;
813 buf->lb_len = info->mti_big_lmmsize;
814 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
815 buf, XATTR_NAME_LINK);
816 /* ignore errors, MA_PFID won't be set and it is
817 * up to the caller to treat this as an error */
818 if (rc == -ERANGE || buf->lb_len == 0) {
819 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
820 buf->lb_buf = info->mti_big_lmm;
821 buf->lb_len = info->mti_big_lmmsize;
826 if (rc < sizeof(*leh)) {
827 CERROR("short LinkEA on "DFID": rc = %d\n",
828 PFID(mdt_object_fid(o)), rc);
832 leh = (struct link_ea_header *) buf->lb_buf;
833 lee = (struct link_ea_entry *)(leh + 1);
834 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
835 leh->leh_magic = LINK_EA_MAGIC;
836 leh->leh_reccount = __swab32(leh->leh_reccount);
837 leh->leh_len = __swab64(leh->leh_len);
839 if (leh->leh_magic != LINK_EA_MAGIC)
841 if (leh->leh_reccount == 0)
844 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
845 fid_be_to_cpu(pfid, pfid);
850 int mdt_attr_get_complex(struct mdt_thread_info *info,
851 struct mdt_object *o, struct md_attr *ma)
853 const struct lu_env *env = info->mti_env;
854 struct md_object *next = mdt_object_child(o);
855 struct lu_buf *buf = &info->mti_buf;
856 int need = ma->ma_need;
863 if (mdt_object_exists(o) == 0)
864 GOTO(out, rc = -ENOENT);
865 mode = lu_object_attr(&next->mo_lu);
867 if (need & MA_INODE) {
868 ma->ma_need = MA_INODE;
869 rc = mo_attr_get(env, next, ma);
872 ma->ma_valid |= MA_INODE;
875 if (need & MA_PFID) {
876 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
878 ma->ma_valid |= MA_PFID;
879 /* ignore this error, parent fid is not mandatory */
883 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
884 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
889 if (need & MA_LMV && S_ISDIR(mode)) {
890 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
895 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
896 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
901 if (need & MA_HSM && S_ISREG(mode)) {
902 buf->lb_buf = info->mti_xattr_buf;
903 buf->lb_len = sizeof(info->mti_xattr_buf);
904 CLASSERT(sizeof(struct hsm_attrs) <=
905 sizeof(info->mti_xattr_buf));
906 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
907 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
909 ma->ma_valid |= MA_HSM;
910 else if (rc2 < 0 && rc2 != -ENODATA)
914 #ifdef CONFIG_FS_POSIX_ACL
915 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
916 buf->lb_buf = ma->ma_acl;
917 buf->lb_len = ma->ma_acl_size;
918 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
920 ma->ma_acl_size = rc2;
921 ma->ma_valid |= MA_ACL_DEF;
922 } else if (rc2 == -ENODATA) {
931 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
932 rc, ma->ma_valid, ma->ma_lmm);
936 static int mdt_getattr_internal(struct mdt_thread_info *info,
937 struct mdt_object *o, int ma_need)
939 struct md_object *next = mdt_object_child(o);
940 const struct mdt_body *reqbody = info->mti_body;
941 struct ptlrpc_request *req = mdt_info_req(info);
942 struct md_attr *ma = &info->mti_attr;
943 struct lu_attr *la = &ma->ma_attr;
944 struct req_capsule *pill = info->mti_pill;
945 const struct lu_env *env = info->mti_env;
946 struct mdt_body *repbody;
947 struct lu_buf *buffer = &info->mti_buf;
948 struct obd_export *exp = info->mti_exp;
953 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
954 RETURN(err_serious(-ENOMEM));
956 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
960 if (mdt_object_remote(o)) {
961 /* This object is located on remote node.*/
962 /* Return -ENOTSUPP for old client */
963 if (!mdt_is_dne_client(req->rq_export))
964 GOTO(out, rc = -ENOTSUPP);
966 repbody->mbo_fid1 = *mdt_object_fid(o);
967 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
971 if (reqbody->mbo_eadatasize > 0) {
972 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
973 if (buffer->lb_buf == NULL)
974 GOTO(out, rc = -EPROTO);
975 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
978 buffer->lb_buf = NULL;
980 ma_need &= ~(MA_LOV | MA_LMV);
981 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
982 mdt_obd_name(info->mti_mdt),
983 req->rq_export->exp_client_uuid.uuid);
986 /* If it is dir object and client require MEA, then we got MEA */
987 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
988 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
989 /* Assumption: MDT_MD size is enough for lmv size. */
990 ma->ma_lmv = buffer->lb_buf;
991 ma->ma_lmv_size = buffer->lb_len;
992 ma->ma_need = MA_INODE;
993 if (ma->ma_lmv_size > 0) {
994 if (reqbody->mbo_valid & OBD_MD_MEA)
995 ma->ma_need |= MA_LMV;
996 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
997 ma->ma_need |= MA_LMV_DEF;
1000 ma->ma_lmm = buffer->lb_buf;
1001 ma->ma_lmm_size = buffer->lb_len;
1002 ma->ma_need = MA_INODE | MA_HSM;
1003 if (ma->ma_lmm_size > 0)
1004 ma->ma_need |= MA_LOV;
1007 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1008 reqbody->mbo_valid & OBD_MD_FLDIREA &&
1009 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1010 /* get default stripe info for this dir. */
1011 ma->ma_need |= MA_LOV_DEF;
1013 ma->ma_need |= ma_need;
1015 rc = mdt_attr_get_complex(info, o, ma);
1017 CERROR("%s: getattr error for "DFID": rc = %d\n",
1018 mdt_obd_name(info->mti_mdt),
1019 PFID(mdt_object_fid(o)), rc);
1023 /* if file is released, check if a restore is running */
1024 if (ma->ma_valid & MA_HSM) {
1025 repbody->mbo_valid |= OBD_MD_TSTATE;
1026 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1027 mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1028 repbody->mbo_t_state = MS_RESTORE;
1031 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
1033 /* the Lustre protocol supposes to return default striping
1034 * on the user-visible root if explicitly requested */
1035 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
1036 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
1037 struct lu_fid rootfid;
1038 struct mdt_object *root;
1039 struct mdt_device *mdt = info->mti_mdt;
1041 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
1044 root = mdt_object_find(env, mdt, &rootfid);
1046 RETURN(PTR_ERR(root));
1047 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
1048 mdt_object_put(info->mti_env, root);
1050 CERROR("%s: getattr error for "DFID": rc = %d\n",
1051 mdt_obd_name(info->mti_mdt),
1052 PFID(mdt_object_fid(o)), rc);
1057 if (likely(ma->ma_valid & MA_INODE))
1058 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1062 if (mdt_body_has_lov(la, reqbody)) {
1063 if (ma->ma_valid & MA_LOV) {
1064 LASSERT(ma->ma_lmm_size);
1065 repbody->mbo_eadatasize = ma->ma_lmm_size;
1066 if (S_ISDIR(la->la_mode))
1067 repbody->mbo_valid |= OBD_MD_FLDIREA;
1069 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1070 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1072 if (ma->ma_valid & MA_LMV) {
1073 /* Return -ENOTSUPP for old client */
1074 if (!mdt_is_striped_client(req->rq_export))
1077 LASSERT(S_ISDIR(la->la_mode));
1078 mdt_dump_lmv(D_INFO, ma->ma_lmv);
1079 repbody->mbo_eadatasize = ma->ma_lmv_size;
1080 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1082 if (ma->ma_valid & MA_LMV_DEF) {
1083 /* Return -ENOTSUPP for old client */
1084 if (!mdt_is_striped_client(req->rq_export))
1086 LASSERT(S_ISDIR(la->la_mode));
1087 repbody->mbo_eadatasize = ma->ma_lmv_size;
1088 repbody->mbo_valid |= (OBD_MD_FLDIREA |
1089 OBD_MD_DEFAULT_MEA);
1091 } else if (S_ISLNK(la->la_mode) &&
1092 reqbody->mbo_valid & OBD_MD_LINKNAME) {
1093 buffer->lb_buf = ma->ma_lmm;
1094 /* eadatasize from client includes NULL-terminator, so
1095 * there is no need to read it */
1096 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1097 rc = mo_readlink(env, next, buffer);
1098 if (unlikely(rc <= 0)) {
1099 CERROR("%s: readlink failed for "DFID": rc = %d\n",
1100 mdt_obd_name(info->mti_mdt),
1101 PFID(mdt_object_fid(o)), rc);
1104 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1106 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1108 repbody->mbo_valid |= OBD_MD_LINKNAME;
1109 /* we need to report back size with NULL-terminator
1110 * because client expects that */
1111 repbody->mbo_eadatasize = rc + 1;
1112 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1113 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1114 "on "DFID ", expected %d\n",
1115 mdt_obd_name(info->mti_mdt),
1116 rc, PFID(mdt_object_fid(o)),
1117 reqbody->mbo_eadatasize - 1);
1118 /* NULL terminate */
1119 ((char *)ma->ma_lmm)[rc] = 0;
1121 /* If the total CDEBUG() size is larger than a page, it
1122 * will print a warning to the console, avoid this by
1123 * printing just the last part of the symlink. */
1124 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1125 print_limit < rc ? "..." : "", print_limit,
1126 (char *)ma->ma_lmm + rc - print_limit, rc);
1131 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1132 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1133 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1134 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1135 repbody->mbo_max_mdsize);
1138 #ifdef CONFIG_FS_POSIX_ACL
1139 if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1140 (reqbody->mbo_valid & OBD_MD_FLACL)) {
1141 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1142 if (IS_ERR(nodemap))
1143 RETURN(PTR_ERR(nodemap));
1145 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1146 nodemap_putref(nodemap);
1152 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1157 static int mdt_getattr(struct tgt_session_info *tsi)
1159 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1160 struct mdt_object *obj = info->mti_object;
1161 struct req_capsule *pill = info->mti_pill;
1162 struct mdt_body *reqbody;
1163 struct mdt_body *repbody;
1167 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1169 LASSERT(obj != NULL);
1170 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1172 /* Unlike intent case where we need to pre-fill out buffers early on
1173 * in intent policy for ldlm reasons, here we can have a much better
1174 * guess at EA size by just reading it from disk.
1175 * Exceptions are readdir and (missing) directory striping */
1177 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1178 /* No easy way to know how long is the symlink, but it cannot
1179 * be more than PATH_MAX, so we allocate +1 */
1182 /* A special case for fs ROOT: getattr there might fetch
1183 * default EA for entire fs, not just for this dir!
1185 } else if (lu_fid_eq(mdt_object_fid(obj),
1186 &info->mti_mdt->mdt_md_root_fid) &&
1187 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1188 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1190 /* Should the default strping be bigger, mdt_fix_reply
1191 * will reallocate */
1192 rc = DEF_REP_MD_SIZE;
1194 /* Read the actual EA size from disk */
1195 rc = mdt_attr_get_eabuf_size(info, obj);
1199 GOTO(out, rc = err_serious(rc));
1201 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1203 rc = req_capsule_server_pack(pill);
1204 if (unlikely(rc != 0))
1205 GOTO(out, rc = err_serious(rc));
1207 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1208 LASSERT(repbody != NULL);
1209 repbody->mbo_eadatasize = 0;
1210 repbody->mbo_aclsize = 0;
1212 rc = mdt_check_ucred(info);
1214 GOTO(out_shrink, rc);
1216 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1218 rc = mdt_getattr_internal(info, obj, 0);
1221 mdt_client_compatibility(info);
1222 rc2 = mdt_fix_reply(info);
1226 mdt_thread_info_fini(info);
1231 * Exchange MOF_LOV_CREATED flags between two objects after a
1232 * layout swap. No assumption is made on whether o1 or o2 have
1233 * created objects or not.
1235 * \param[in,out] o1 First swap layout object
1236 * \param[in,out] o2 Second swap layout object
1238 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1240 unsigned int o1_lov_created = o1->mot_lov_created;
1242 mutex_lock(&o1->mot_lov_mutex);
1243 mutex_lock(&o2->mot_lov_mutex);
1245 o1->mot_lov_created = o2->mot_lov_created;
1246 o2->mot_lov_created = o1_lov_created;
1248 mutex_unlock(&o2->mot_lov_mutex);
1249 mutex_unlock(&o1->mot_lov_mutex);
1252 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1254 struct mdt_thread_info *info;
1255 struct ptlrpc_request *req = tgt_ses_req(tsi);
1256 struct obd_export *exp = req->rq_export;
1257 struct mdt_object *o1, *o2, *o;
1258 struct mdt_lock_handle *lh1, *lh2;
1259 struct mdc_swap_layouts *msl;
1263 /* client does not support layout lock, so layout swaping
1265 * FIXME: there is a problem for old clients which don't support
1266 * layout lock yet. If those clients have already opened the file
1267 * they won't be notified at all so that old layout may still be
1268 * used to do IO. This can be fixed after file release is landed by
1269 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1270 if (!exp_connect_layout(exp))
1271 RETURN(-EOPNOTSUPP);
1273 info = tsi2mdt_info(tsi);
1275 if (info->mti_dlm_req != NULL)
1276 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1278 o1 = info->mti_object;
1279 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1280 &info->mti_body->mbo_fid2);
1282 GOTO(out, rc = PTR_ERR(o));
1284 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1285 GOTO(put, rc = -ENOENT);
1287 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1288 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1294 /* permission check. Make sure the calling process having permission
1295 * to write both files. */
1296 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1301 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1306 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1308 GOTO(put, rc = -EPROTO);
1310 lh1 = &info->mti_lh[MDT_LH_NEW];
1311 mdt_lock_reg_init(lh1, LCK_EX);
1312 lh2 = &info->mti_lh[MDT_LH_OLD];
1313 mdt_lock_reg_init(lh2, LCK_EX);
1315 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1316 MDS_INODELOCK_XATTR);
1320 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1321 MDS_INODELOCK_XATTR);
1325 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1326 mdt_object_child(o2), msl->msl_flags);
1330 mdt_swap_lov_flag(o1, o2);
1333 mdt_object_unlock(info, o2, lh2, rc);
1335 mdt_object_unlock(info, o1, lh1, rc);
1337 mdt_object_put(info->mti_env, o);
1339 mdt_thread_info_fini(info);
1343 static int mdt_raw_lookup(struct mdt_thread_info *info,
1344 struct mdt_object *parent,
1345 const struct lu_name *lname,
1346 struct ldlm_reply *ldlm_rep)
1348 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1352 LASSERT(!info->mti_cross_ref);
1354 /* Only got the fid of this obj by name */
1355 fid_zero(child_fid);
1356 rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1357 lname, child_fid, &info->mti_spec);
1359 struct mdt_body *repbody;
1361 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1362 repbody->mbo_fid1 = *child_fid;
1363 repbody->mbo_valid = OBD_MD_FLID;
1364 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1365 } else if (rc == -ENOENT) {
1366 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1373 * UPDATE lock should be taken against parent, and be released before exit;
1374 * child_bits lock should be taken against child, and be returned back:
1375 * (1)normal request should release the child lock;
1376 * (2)intent request will grant the lock to client.
1378 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1379 struct mdt_lock_handle *lhc,
1381 struct ldlm_reply *ldlm_rep)
1383 struct ptlrpc_request *req = mdt_info_req(info);
1384 struct mdt_body *reqbody = NULL;
1385 struct mdt_object *parent = info->mti_object;
1386 struct mdt_object *child;
1387 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1388 struct lu_name *lname = NULL;
1389 struct mdt_lock_handle *lhp = NULL;
1390 struct ldlm_lock *lock;
1397 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1398 LASSERT(ergo(is_resent,
1399 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1404 if (info->mti_cross_ref) {
1405 /* Only getattr on the child. Parent is on another node. */
1406 mdt_set_disposition(info, ldlm_rep,
1407 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1409 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1411 PFID(mdt_object_fid(child)), ldlm_rep);
1413 rc = mdt_check_resent_lock(info, child, lhc);
1416 } else if (rc > 0) {
1417 mdt_lock_handle_init(lhc);
1418 mdt_lock_reg_init(lhc, LCK_PR);
1421 * Object's name is on another MDS, no lookup or layout
1422 * lock is needed here but update lock is.
1424 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1425 MDS_INODELOCK_LAYOUT);
1426 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1428 rc = mdt_object_lock(info, child, lhc, child_bits);
1433 /* Finally, we can get attr for child. */
1434 if (!mdt_object_exists(child)) {
1435 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1437 "remote object doesn't exist.");
1438 mdt_object_unlock(info, child, lhc, 1);
1442 rc = mdt_getattr_internal(info, child, 0);
1443 if (unlikely(rc != 0))
1444 mdt_object_unlock(info, child, lhc, 1);
1449 lname = &info->mti_name;
1450 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1452 if (lu_name_is_valid(lname)) {
1453 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1454 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1455 PNAME(lname), ldlm_rep);
1457 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1458 if (unlikely(reqbody == NULL))
1459 RETURN(err_serious(-EPROTO));
1461 *child_fid = reqbody->mbo_fid2;
1463 if (unlikely(!fid_is_sane(child_fid)))
1464 RETURN(err_serious(-EINVAL));
1466 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1468 PFID(mdt_object_fid(parent)),
1469 PFID(&reqbody->mbo_fid2), ldlm_rep);
1472 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1474 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1475 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1477 "Parent doesn't exist!");
1481 if (mdt_object_remote(parent)) {
1482 CERROR("%s: parent "DFID" is on remote target\n",
1483 mdt_obd_name(info->mti_mdt),
1484 PFID(mdt_object_fid(parent)));
1488 if (lu_name_is_valid(lname)) {
1489 /* Always allow to lookup ".." */
1490 if (unlikely(lname->ln_namelen == 2 &&
1491 lname->ln_name[0] == '.' &&
1492 lname->ln_name[1] == '.'))
1493 info->mti_spec.sp_permitted = 1;
1495 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1496 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1501 /* step 1: lock parent only if parent is a directory */
1502 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1503 lhp = &info->mti_lh[MDT_LH_PARENT];
1504 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1505 rc = mdt_object_lock(info, parent, lhp,
1506 MDS_INODELOCK_UPDATE);
1507 if (unlikely(rc != 0))
1511 /* step 2: lookup child's fid by name */
1512 fid_zero(child_fid);
1513 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1514 child_fid, &info->mti_spec);
1516 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1519 GOTO(out_parent, rc);
1522 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1525 *step 3: find the child object by fid & lock it.
1526 * regardless if it is local or remote.
1528 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1529 * set parent dir fid the same as child fid in getattr by fid case
1530 * we should not lu_object_find() the object again, could lead
1531 * to hung if there is a concurrent unlink destroyed the object.
1533 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1534 mdt_object_get(info->mti_env, parent);
1537 child = mdt_object_find(info->mti_env, info->mti_mdt,
1541 if (unlikely(IS_ERR(child)))
1542 GOTO(out_parent, rc = PTR_ERR(child));
1544 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1545 if (!mdt_object_exists(child)) {
1546 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1548 "Object doesn't exist!");
1549 GOTO(out_child, rc = -ENOENT);
1552 rc = mdt_check_resent_lock(info, child, lhc);
1554 GOTO(out_child, rc);
1555 } else if (rc > 0) {
1556 mdt_lock_handle_init(lhc);
1557 mdt_lock_reg_init(lhc, LCK_PR);
1560 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1561 mdt_object_exists(child) && !mdt_object_remote(child)) {
1562 struct md_attr *ma = &info->mti_attr;
1565 ma->ma_need = MA_INODE;
1566 rc = mdt_attr_get_complex(info, child, ma);
1567 if (unlikely(rc != 0))
1568 GOTO(out_child, rc);
1570 /* If the file has not been changed for some time, we
1571 * return not only a LOOKUP lock, but also an UPDATE
1572 * lock and this might save us RPC on later STAT. For
1573 * directories, it also let negative dentry cache start
1574 * working for this dir. */
1575 if (ma->ma_valid & MA_INODE &&
1576 ma->ma_attr.la_valid & LA_CTIME &&
1577 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1578 ma->ma_attr.la_ctime < cfs_time_current_sec())
1579 child_bits |= MDS_INODELOCK_UPDATE;
1582 /* layout lock must be granted in a best-effort way
1583 * for IT operations */
1584 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1585 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1586 exp_connect_layout(info->mti_exp) &&
1587 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1588 !mdt_object_remote(child) && ldlm_rep != NULL) {
1589 /* try to grant layout lock for regular file. */
1595 child_bits |= MDS_INODELOCK_LAYOUT;
1596 /* try layout lock, it may fail to be granted due to
1597 * contention at LOOKUP or UPDATE */
1598 if (!mdt_object_lock_try(info, child, lhc,
1600 child_bits &= ~MDS_INODELOCK_LAYOUT;
1601 LASSERT(child_bits != 0);
1602 rc = mdt_object_lock(info, child, lhc,
1608 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1609 * client will enqueue the lock to the remote MDT */
1610 if (mdt_object_remote(child))
1611 child_bits &= ~MDS_INODELOCK_UPDATE;
1612 rc = mdt_object_lock(info, child, lhc, child_bits);
1614 if (unlikely(rc != 0))
1615 GOTO(out_child, rc);
1618 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1620 /* finally, we can get attr for child. */
1621 rc = mdt_getattr_internal(info, child, ma_need);
1622 if (unlikely(rc != 0)) {
1623 mdt_object_unlock(info, child, lhc, 1);
1625 /* Debugging code. */
1626 LDLM_DEBUG(lock, "Returning lock to client");
1627 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1628 &lock->l_resource->lr_name),
1629 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1630 PLDLMRES(lock->l_resource),
1631 PFID(mdt_object_fid(child)));
1634 LDLM_LOCK_PUT(lock);
1638 mdt_object_put(info->mti_env, child);
1641 mdt_object_unlock(info, parent, lhp, 1);
1645 /* normal handler: should release the child lock */
1646 static int mdt_getattr_name(struct tgt_session_info *tsi)
1648 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1649 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1650 struct mdt_body *reqbody;
1651 struct mdt_body *repbody;
1655 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1656 LASSERT(reqbody != NULL);
1657 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1658 LASSERT(repbody != NULL);
1660 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1661 repbody->mbo_eadatasize = 0;
1662 repbody->mbo_aclsize = 0;
1664 rc = mdt_init_ucred_intent_getattr(info, reqbody);
1666 GOTO(out_shrink, rc);
1668 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1669 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1670 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1671 lhc->mlh_reg_lh.cookie = 0;
1673 mdt_exit_ucred(info);
1676 mdt_client_compatibility(info);
1677 rc2 = mdt_fix_reply(info);
1680 mdt_thread_info_fini(info);
1684 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1685 void *karg, void __user *uarg);
1687 static int mdt_set_info(struct tgt_session_info *tsi)
1689 struct ptlrpc_request *req = tgt_ses_req(tsi);
1692 int keylen, vallen, rc = 0;
1696 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1698 DEBUG_REQ(D_HA, req, "no set_info key");
1699 RETURN(err_serious(-EFAULT));
1702 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1705 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1707 DEBUG_REQ(D_HA, req, "no set_info val");
1708 RETURN(err_serious(-EFAULT));
1711 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1714 /* Swab any part of val you need to here */
1715 if (KEY_IS(KEY_READ_ONLY)) {
1716 spin_lock(&req->rq_export->exp_lock);
1718 *exp_connect_flags_ptr(req->rq_export) |=
1721 *exp_connect_flags_ptr(req->rq_export) &=
1722 ~OBD_CONNECT_RDONLY;
1723 spin_unlock(&req->rq_export->exp_lock);
1724 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1725 struct changelog_setinfo *cs = val;
1727 if (vallen != sizeof(*cs)) {
1728 CERROR("%s: bad changelog_clear setinfo size %d\n",
1729 tgt_name(tsi->tsi_tgt), vallen);
1732 if (ptlrpc_req_need_swab(req)) {
1733 __swab64s(&cs->cs_recno);
1734 __swab32s(&cs->cs_id);
1737 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1739 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1741 obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1748 static int mdt_readpage(struct tgt_session_info *tsi)
1750 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1751 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1752 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1753 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1754 struct mdt_body *repbody;
1760 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1761 RETURN(err_serious(-ENOMEM));
1763 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1764 if (repbody == NULL || reqbody == NULL)
1765 RETURN(err_serious(-EFAULT));
1768 * prepare @rdpg before calling lower layers and transfer itself. Here
1769 * reqbody->size contains offset of where to start to read and
1770 * reqbody->nlink contains number bytes to read.
1772 rdpg->rp_hash = reqbody->mbo_size;
1773 if (rdpg->rp_hash != reqbody->mbo_size) {
1774 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1775 rdpg->rp_hash, reqbody->mbo_size);
1779 rdpg->rp_attrs = reqbody->mbo_mode;
1780 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1781 rdpg->rp_attrs |= LUDA_64BITHASH;
1782 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1783 exp_max_brw_size(tsi->tsi_exp));
1784 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1786 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1787 if (rdpg->rp_pages == NULL)
1790 for (i = 0; i < rdpg->rp_npages; ++i) {
1791 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1792 if (rdpg->rp_pages[i] == NULL)
1793 GOTO(free_rdpg, rc = -ENOMEM);
1796 /* call lower layers to fill allocated pages with directory data */
1797 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1799 GOTO(free_rdpg, rc);
1801 /* send pages to client */
1802 rc = tgt_sendpage(tsi, rdpg, rc);
1807 for (i = 0; i < rdpg->rp_npages; i++)
1808 if (rdpg->rp_pages[i] != NULL)
1809 __free_page(rdpg->rp_pages[i]);
1810 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1812 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1818 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1820 struct lu_ucred *uc = mdt_ucred_check(info);
1821 struct lu_attr *attr = &info->mti_attr.ma_attr;
1826 if (op != REINT_SETATTR) {
1827 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1828 attr->la_uid = uc->uc_fsuid;
1829 /* for S_ISGID, inherit gid from his parent, such work will be
1830 * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1831 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1832 attr->la_gid = uc->uc_fsgid;
1838 static int mdt_reint_internal(struct mdt_thread_info *info,
1839 struct mdt_lock_handle *lhc,
1842 struct req_capsule *pill = info->mti_pill;
1843 struct mdt_body *repbody;
1848 rc = mdt_reint_unpack(info, op);
1850 CERROR("Can't unpack reint, rc %d\n", rc);
1851 RETURN(err_serious(rc));
1854 /* for replay (no_create) lmm is not needed, client has it already */
1855 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1856 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1859 /* llog cookies are always 0, the field is kept for compatibility */
1860 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1861 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1863 rc = req_capsule_server_pack(pill);
1865 CERROR("Can't pack response, rc %d\n", rc);
1866 RETURN(err_serious(rc));
1869 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1870 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1872 repbody->mbo_eadatasize = 0;
1873 repbody->mbo_aclsize = 0;
1876 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1878 /* for replay no cookkie / lmm need, because client have this already */
1879 if (info->mti_spec.no_create)
1880 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1881 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1883 rc = mdt_init_ucred_reint(info);
1885 GOTO(out_shrink, rc);
1887 rc = mdt_fix_attr_ucred(info, op);
1889 GOTO(out_ucred, rc = err_serious(rc));
1891 rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1893 GOTO(out_ucred, rc);
1894 } else if (rc == 1) {
1895 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1896 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1897 GOTO(out_ucred, rc);
1899 rc = mdt_reint_rec(info, lhc);
1902 mdt_exit_ucred(info);
1904 mdt_client_compatibility(info);
1905 rc2 = mdt_fix_reply(info);
1911 static long mdt_reint_opcode(struct ptlrpc_request *req,
1912 const struct req_format **fmt)
1914 struct mdt_device *mdt;
1915 struct mdt_rec_reint *rec;
1918 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1920 opc = rec->rr_opcode;
1921 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1922 if (opc < REINT_MAX && fmt[opc] != NULL)
1923 req_capsule_extend(&req->rq_pill, fmt[opc]);
1925 mdt = mdt_exp2dev(req->rq_export);
1926 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1927 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1928 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1929 opc = err_serious(-EFAULT);
1932 opc = err_serious(-EFAULT);
1937 static int mdt_reint(struct tgt_session_info *tsi)
1941 static const struct req_format *reint_fmts[REINT_MAX] = {
1942 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1943 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1944 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1945 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1946 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1947 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1948 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1949 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1950 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1955 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1957 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1959 * No lock possible here from client to pass it to reint code
1962 rc = mdt_reint_internal(info, NULL, opc);
1963 mdt_thread_info_fini(info);
1968 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1972 /* this should sync the whole device */
1973 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1975 struct dt_device *dt = mdt->mdt_bottom;
1979 rc = dt->dd_ops->dt_sync(env, dt);
1983 /* this should sync this object */
1984 static int mdt_object_sync(struct mdt_thread_info *info)
1986 struct md_object *next;
1990 if (!mdt_object_exists(info->mti_object)) {
1991 CWARN("%s: non existing object "DFID": rc = %d\n",
1992 mdt_obd_name(info->mti_mdt),
1993 PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1996 next = mdt_object_child(info->mti_object);
1997 rc = mo_object_sync(info->mti_env, next);
2002 static int mdt_sync(struct tgt_session_info *tsi)
2004 struct ptlrpc_request *req = tgt_ses_req(tsi);
2005 struct req_capsule *pill = tsi->tsi_pill;
2006 struct mdt_body *body;
2011 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2012 RETURN(err_serious(-ENOMEM));
2014 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2015 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2017 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2019 /* sync an object */
2020 rc = mdt_object_sync(info);
2022 const struct lu_fid *fid;
2023 struct lu_attr *la = &info->mti_attr.ma_attr;
2025 info->mti_attr.ma_need = MA_INODE;
2026 info->mti_attr.ma_valid = 0;
2027 rc = mdt_attr_get_complex(info, info->mti_object,
2030 body = req_capsule_server_get(pill,
2032 fid = mdt_object_fid(info->mti_object);
2033 mdt_pack_attr2body(info, body, la, fid);
2036 mdt_thread_info_fini(info);
2039 mdt_counter_incr(req, LPROC_MDT_SYNC);
2045 * Handle quota control requests to consult current usage/limit, but also
2046 * to configure quota enforcement
2048 static int mdt_quotactl(struct tgt_session_info *tsi)
2050 struct obd_export *exp = tsi->tsi_exp;
2051 struct req_capsule *pill = tsi->tsi_pill;
2052 struct obd_quotactl *oqctl, *repoqc;
2054 struct mdt_device *mdt = mdt_exp2dev(exp);
2055 struct lu_device *qmt = mdt->mdt_qmt_dev;
2056 struct lu_nodemap *nodemap;
2059 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2061 RETURN(err_serious(-EPROTO));
2063 rc = req_capsule_server_pack(pill);
2065 RETURN(err_serious(rc));
2067 nodemap = nodemap_get_from_exp(exp);
2068 if (IS_ERR(nodemap))
2069 RETURN(PTR_ERR(nodemap));
2071 switch (oqctl->qc_cmd) {
2072 /* master quotactl */
2075 if (!nodemap_can_setquota(nodemap))
2076 GOTO(out_nodemap, rc = -EPERM);
2080 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2081 /* slave quotactl */
2086 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2087 GOTO(out_nodemap, rc = -EFAULT);
2091 if (oqctl->qc_type == USRQUOTA)
2092 id = nodemap_map_id(nodemap, NODEMAP_UID,
2093 NODEMAP_CLIENT_TO_FS, id);
2094 else if (oqctl->qc_type == GRPQUOTA)
2095 id = nodemap_map_id(nodemap, NODEMAP_UID,
2096 NODEMAP_CLIENT_TO_FS, id);
2098 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2100 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2102 if (oqctl->qc_id != id)
2103 swap(oqctl->qc_id, id);
2105 switch (oqctl->qc_cmd) {
2111 /* forward quotactl request to QMT */
2112 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2117 /* slave quotactl */
2118 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2123 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2124 GOTO(out_nodemap, rc = -EFAULT);
2127 if (oqctl->qc_id != id)
2128 swap(oqctl->qc_id, id);
2135 nodemap_putref(nodemap);
2140 /** clone llog ctxt from child (mdd)
2141 * This allows remote llog (replicator) access.
2142 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2143 * context was originally set up, or we can handle them directly.
2144 * I choose the latter, but that means I need any llog
2145 * contexts set up by child to be accessable by the mdt. So we clone the
2146 * context into our context list here.
2148 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2151 struct md_device *next = mdt->mdt_child;
2152 struct llog_ctxt *ctxt;
2155 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2158 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2159 if (rc || ctxt == NULL) {
2163 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2165 CERROR("Can't set mdt ctxt %d\n", rc);
2170 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2171 struct mdt_device *mdt, int idx)
2173 struct llog_ctxt *ctxt;
2175 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2178 /* Put once for the get we just did, and once for the clone */
2179 llog_ctxt_put(ctxt);
2180 llog_ctxt_put(ctxt);
2185 * sec context handlers
2187 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2189 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2195 * quota request handlers
2197 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2199 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2200 struct lu_device *qmt = mdt->mdt_qmt_dev;
2205 RETURN(err_serious(-EOPNOTSUPP));
2207 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2211 struct mdt_object *mdt_object_new(const struct lu_env *env,
2212 struct mdt_device *d,
2213 const struct lu_fid *f)
2215 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2216 struct lu_object *o;
2217 struct mdt_object *m;
2220 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2221 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2222 if (unlikely(IS_ERR(o)))
2223 m = (struct mdt_object *)o;
2229 struct mdt_object *mdt_object_find(const struct lu_env *env,
2230 struct mdt_device *d,
2231 const struct lu_fid *f)
2233 struct lu_object *o;
2234 struct mdt_object *m;
2237 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2238 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2239 if (unlikely(IS_ERR(o)))
2240 m = (struct mdt_object *)o;
2248 * Asyncronous commit for mdt device.
2250 * Pass asynchonous commit call down the MDS stack.
2252 * \param env environment
2253 * \param mdt the mdt device
2255 static void mdt_device_commit_async(const struct lu_env *env,
2256 struct mdt_device *mdt)
2258 struct dt_device *dt = mdt->mdt_bottom;
2262 rc = dt->dd_ops->dt_commit_async(env, dt);
2263 if (unlikely(rc != 0))
2264 CWARN("%s: async commit start failed: rc = %d\n",
2265 mdt_obd_name(mdt), rc);
2266 atomic_inc(&mdt->mdt_async_commit_count);
2271 * Mark the lock as "synchonous".
2273 * Mark the lock to deffer transaction commit to the unlock time.
2275 * \param lock the lock to mark as "synchonous"
2277 * \see mdt_is_lock_sync
2278 * \see mdt_save_lock
2280 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2282 lock->l_ast_data = (void*)1;
2286 * Check whehter the lock "synchonous" or not.
2288 * \param lock the lock to check
2289 * \retval 1 the lock is "synchonous"
2290 * \retval 0 the lock isn't "synchronous"
2292 * \see mdt_set_lock_sync
2293 * \see mdt_save_lock
2295 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2297 return lock->l_ast_data != NULL;
2301 * Blocking AST for mdt locks.
2303 * Starts transaction commit if in case of COS lock conflict or
2304 * deffers such a commit to the mdt_save_lock.
2306 * \param lock the lock which blocks a request or cancelling lock
2307 * \param desc unused
2308 * \param data unused
2309 * \param flag indicates whether this cancelling or blocking callback
2311 * \see ldlm_blocking_ast_nocheck
2313 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2314 void *data, int flag)
2316 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2317 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2321 if (flag == LDLM_CB_CANCELING)
2324 lock_res_and_lock(lock);
2325 if (lock->l_blocking_ast != mdt_blocking_ast) {
2326 unlock_res_and_lock(lock);
2329 if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2330 lock->l_blocking_lock != NULL) {
2331 if (mdt_cos_is_enabled(mdt) &&
2332 lock->l_client_cookie !=
2333 lock->l_blocking_lock->l_client_cookie)
2334 mdt_set_lock_sync(lock);
2335 else if (mdt_slc_is_enabled(mdt) &&
2336 ldlm_is_cos_incompat(lock->l_blocking_lock))
2337 mdt_set_lock_sync(lock);
2339 rc = ldlm_blocking_ast_nocheck(lock);
2341 /* There is no lock conflict if l_blocking_lock == NULL,
2342 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2343 * when the last reference to a local lock was released */
2344 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2347 rc = lu_env_init(&env, LCT_LOCAL);
2348 if (unlikely(rc != 0))
2349 CWARN("%s: lu_env initialization failed, cannot "
2350 "start asynchronous commit: rc = %d\n",
2353 mdt_device_commit_async(&env, mdt);
2360 * Blocking AST for cross-MDT lock
2362 * Discard lock from uncommitted_slc_locks and cancel it.
2364 * \param lock the lock which blocks a request or cancelling lock
2365 * \param desc unused
2366 * \param data unused
2367 * \param flag indicates whether this cancelling or blocking callback
2368 * \retval 0 on success
2369 * \retval negative number on error
2371 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2372 void *data, int flag)
2378 case LDLM_CB_BLOCKING: {
2379 struct lustre_handle lockh;
2381 ldlm_lock2handle(lock, &lockh);
2382 rc = ldlm_cli_cancel(&lockh,
2383 ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2385 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2390 case LDLM_CB_CANCELING: {
2391 LDLM_DEBUG(lock, "Revoke remote lock\n");
2393 /* discard slc lock here so that it can be cleaned anytime,
2394 * especially for cleanup_resource() */
2395 tgt_discard_slc_lock(lock);
2397 /* once we cache lock, l_ast_data is set to mdt_object */
2398 if (lock->l_ast_data != NULL) {
2399 struct mdt_object *mo = lock->l_ast_data;
2402 rc = lu_env_init(&env, LCT_MD_THREAD);
2403 if (unlikely(rc != 0)) {
2404 struct obd_device *obd;
2406 obd = ldlm_lock_to_ns(lock)->ns_obd;
2407 CWARN("%s: lu_env initialization failed, object"
2408 "%p "DFID" is leaked!\n",
2410 PFID(mdt_object_fid(mo)));
2414 if (lock->l_policy_data.l_inodebits.bits &
2415 (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2416 rc = mo_invalidate(&env, mdt_object_child(mo));
2417 mo->mot_cache_attr = 0;
2419 mdt_object_put(&env, mo);
2431 int mdt_check_resent_lock(struct mdt_thread_info *info,
2432 struct mdt_object *mo,
2433 struct mdt_lock_handle *lhc)
2435 /* the lock might already be gotten in ldlm_handle_enqueue() */
2436 if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2437 struct ptlrpc_request *req = mdt_info_req(info);
2438 struct ldlm_lock *lock;
2440 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2441 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2443 /* Lock is pinned by ldlm_handle_enqueue0() as it is
2444 * a resend case, however, it could be already destroyed
2445 * due to client eviction or a raced cancel RPC. */
2446 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64,
2447 lhc->mlh_reg_lh.cookie);
2451 if (!fid_res_name_eq(mdt_object_fid(mo),
2452 &lock->l_resource->lr_name)) {
2453 CWARN("%s: Although resent, but still not "
2454 "get child lock:"DFID"\n",
2455 info->mti_exp->exp_obd->obd_name,
2456 PFID(mdt_object_fid(mo)));
2457 LDLM_LOCK_PUT(lock);
2460 LDLM_LOCK_PUT(lock);
2466 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2467 const struct lu_fid *fid, struct lustre_handle *lh,
2468 enum ldlm_mode mode, __u64 ibits, bool nonblock,
2471 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2472 union ldlm_policy_data *policy = &mti->mti_policy;
2473 struct ldlm_res_id *res_id = &mti->mti_res_id;
2477 LASSERT(mdt_object_remote(o));
2479 fid_build_reg_res_name(fid, res_id);
2481 memset(einfo, 0, sizeof(*einfo));
2482 einfo->ei_type = LDLM_IBITS;
2483 einfo->ei_mode = mode;
2484 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2485 einfo->ei_cb_cp = ldlm_completion_ast;
2486 einfo->ei_enq_slave = 0;
2487 einfo->ei_res_id = res_id;
2489 einfo->ei_nonblock = 1;
2492 * if we cache lock, couple lock with mdt_object, so that object
2493 * can be easily found in lock ASTs.
2495 mdt_object_get(mti->mti_env, o);
2496 einfo->ei_cbdata = o;
2499 memset(policy, 0, sizeof(*policy));
2500 policy->l_inodebits.bits = ibits;
2502 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2504 if (rc < 0 && cache) {
2505 mdt_object_put(mti->mti_env, o);
2506 einfo->ei_cbdata = NULL;
2511 static int mdt_object_local_lock(struct mdt_thread_info *info,
2512 struct mdt_object *o,
2513 struct mdt_lock_handle *lh, __u64 ibits,
2514 bool nonblock, bool cos_incompat)
2516 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2517 union ldlm_policy_data *policy = &info->mti_policy;
2518 struct ldlm_res_id *res_id = &info->mti_res_id;
2523 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2524 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2525 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2526 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2529 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2530 lh->mlh_reg_mode == LCK_EX);
2531 dlmflags |= LDLM_FL_COS_INCOMPAT;
2532 } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2533 dlmflags |= LDLM_FL_COS_ENABLED;
2536 /* Only enqueue LOOKUP lock for remote object */
2537 if (mdt_object_remote(o))
2538 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2540 if (lh->mlh_type == MDT_PDO_LOCK) {
2541 /* check for exists after object is locked */
2542 if (mdt_object_exists(o) == 0) {
2543 /* Non-existent object shouldn't have PDO lock */
2546 /* Non-dir object shouldn't have PDO lock */
2547 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2552 memset(policy, 0, sizeof(*policy));
2553 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2555 dlmflags |= LDLM_FL_ATOMIC_CB;
2557 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2560 * Take PDO lock on whole directory and build correct @res_id for lock
2561 * on part of directory.
2563 if (lh->mlh_pdo_hash != 0) {
2564 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2565 mdt_lock_pdo_mode(info, o, lh);
2566 if (lh->mlh_pdo_mode != LCK_NL) {
2568 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2569 * is never going to be sent to client and we do not
2570 * want it slowed down due to possible cancels.
2572 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2573 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2574 policy, res_id, dlmflags,
2575 info->mti_exp == NULL ? NULL :
2576 &info->mti_exp->exp_handle.h_cookie);
2577 if (unlikely(rc != 0))
2578 GOTO(out_unlock, rc);
2582 * Finish res_id initializing by name hash marking part of
2583 * directory which is taking modification.
2585 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2588 policy->l_inodebits.bits = ibits;
2591 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2592 * going to be sent to client. If it is - mdt_intent_policy() path will
2593 * fix it up and turn FL_LOCAL flag off.
2595 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2596 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2597 info->mti_exp == NULL ? NULL :
2598 &info->mti_exp->exp_handle.h_cookie);
2601 mdt_object_unlock(info, o, lh, 1);
2602 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2603 lh->mlh_pdo_hash != 0 &&
2604 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2605 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2611 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2612 struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2615 struct mdt_lock_handle *local_lh = NULL;
2619 if (!mdt_object_remote(o)) {
2620 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2625 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2626 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2627 MDS_INODELOCK_XATTR);
2629 /* Only enqueue LOOKUP lock for remote object */
2630 if (ibits & MDS_INODELOCK_LOOKUP) {
2631 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2632 nonblock, cos_incompat);
2639 if (ibits & MDS_INODELOCK_UPDATE) {
2640 /* Sigh, PDO needs to enqueue 2 locks right now, but
2641 * enqueue RPC can only request 1 lock, to avoid extra
2642 * RPC, so it will instead enqueue EX lock for remote
2643 * object anyway XXX*/
2644 if (lh->mlh_type == MDT_PDO_LOCK &&
2645 lh->mlh_pdo_hash != 0) {
2646 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2647 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2648 PFID(mdt_object_fid(o)));
2649 lh->mlh_pdo_hash = 0;
2650 lh->mlh_rreg_mode = LCK_EX;
2651 lh->mlh_type = MDT_REG_LOCK;
2653 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2656 MDS_INODELOCK_UPDATE, nonblock,
2658 if (rc != ELDLM_OK) {
2659 if (local_lh != NULL)
2660 mdt_object_unlock(info, o, local_lh, rc);
2668 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2669 struct mdt_lock_handle *lh, __u64 ibits)
2671 return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2674 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2675 struct mdt_lock_handle *lh, __u64 ibits,
2678 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2679 return mdt_object_lock_internal(info, o, lh, ibits, false,
2683 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2684 struct mdt_lock_handle *lh, __u64 ibits)
2686 struct mdt_lock_handle tmp = *lh;
2689 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2696 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2697 struct mdt_object *o, struct mdt_lock_handle *lh,
2698 __u64 ibits, bool cos_incompat)
2700 struct mdt_lock_handle tmp = *lh;
2703 LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2704 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2712 * Save a lock within request object.
2714 * Keep the lock referenced until whether client ACK or transaction
2715 * commit happens or release the lock immediately depending on input
2716 * parameters. If COS is ON, a write lock is converted to COS lock
2719 * \param info thead info object
2720 * \param h lock handle
2721 * \param mode lock mode
2722 * \param decref force immediate lock releasing
2724 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2725 enum ldlm_mode mode, int decref)
2729 if (lustre_handle_is_used(h)) {
2730 if (decref || !info->mti_has_trans ||
2731 !(mode & (LCK_PW | LCK_EX))) {
2732 mdt_fid_unlock(h, mode);
2734 struct mdt_device *mdt = info->mti_mdt;
2735 struct ldlm_lock *lock = ldlm_handle2lock(h);
2736 struct ptlrpc_request *req = mdt_info_req(info);
2739 cos = (mdt_cos_is_enabled(mdt) ||
2740 mdt_slc_is_enabled(mdt));
2742 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2745 /* there is no request if mdt_object_unlock() is called
2746 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2747 if (likely(req != NULL)) {
2748 CDEBUG(D_HA, "request = %p reply state = %p"
2749 " transno = "LPD64"\n", req,
2750 req->rq_reply_state, req->rq_transno);
2752 ldlm_lock_downgrade(lock, LCK_COS);
2755 ptlrpc_save_lock(req, h, mode, cos);
2757 mdt_fid_unlock(h, mode);
2759 if (mdt_is_lock_sync(lock)) {
2760 CDEBUG(D_HA, "found sync-lock,"
2761 " async commit started\n");
2762 mdt_device_commit_async(info->mti_env,
2765 LDLM_LOCK_PUT(lock);
2774 * Save cross-MDT lock in uncommitted_slc_locks
2776 * Keep the lock referenced until transaction commit happens or release the lock
2777 * immediately depending on input parameters.
2779 * \param info thead info object
2780 * \param h lock handle
2781 * \param mode lock mode
2782 * \param decref force immediate lock releasing
2784 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2785 struct mdt_object *o, struct lustre_handle *h,
2786 enum ldlm_mode mode, int decref)
2790 if (lustre_handle_is_used(h)) {
2791 struct ldlm_lock *lock = ldlm_handle2lock(h);
2794 (lock->l_policy_data.l_inodebits.bits &
2795 (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2796 mo_invalidate(info->mti_env, mdt_object_child(o));
2798 if (decref || !info->mti_has_trans ||
2799 !(mode & (LCK_PW | LCK_EX))) {
2800 ldlm_lock_decref_and_cancel(h, mode);
2801 LDLM_LOCK_PUT(lock);
2803 struct ptlrpc_request *req = mdt_info_req(info);
2805 LASSERT(req != NULL);
2806 tgt_save_slc_lock(lock, req->rq_transno);
2807 ldlm_lock_decref(h, mode);
2816 * Unlock mdt object.
2818 * Immeditely release the regular lock and the PDO lock or save the
2819 * lock in request and keep them referenced until client ACK or
2820 * transaction commit.
2822 * \param info thread info object
2823 * \param o mdt object
2824 * \param lh mdt lock handle referencing regular and PDO locks
2825 * \param decref force immediate lock releasing
2827 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2828 struct mdt_lock_handle *lh, int decref)
2832 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2833 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2834 mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2840 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2841 const struct lu_fid *f,
2842 struct mdt_lock_handle *lh,
2845 struct mdt_object *o;
2847 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2851 rc = mdt_object_lock(info, o, lh, ibits);
2853 mdt_object_put(info->mti_env, o);
2860 void mdt_object_unlock_put(struct mdt_thread_info * info,
2861 struct mdt_object * o,
2862 struct mdt_lock_handle *lh,
2865 mdt_object_unlock(info, o, lh, decref);
2866 mdt_object_put(info->mti_env, o);
2870 * Generic code handling requests that have struct mdt_body passed in:
2872 * - extract mdt_body from request and save it in @info, if present;
2874 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2877 * - if HABEO_CORPUS flag is set for this request type check whether object
2878 * actually exists on storage (lu_object_exists()).
2881 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2883 const struct mdt_body *body;
2884 struct mdt_object *obj;
2885 const struct lu_env *env;
2886 struct req_capsule *pill;
2890 env = info->mti_env;
2891 pill = info->mti_pill;
2893 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2897 if (!(body->mbo_valid & OBD_MD_FLID))
2900 if (!fid_is_sane(&body->mbo_fid1)) {
2901 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2905 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2907 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2908 mdt_object_put(env, obj);
2911 info->mti_object = obj;
2920 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2922 struct req_capsule *pill = info->mti_pill;
2926 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2927 rc = mdt_body_unpack(info, flags);
2931 if (rc == 0 && (flags & HABEO_REFERO)) {
2933 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2934 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2936 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2937 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2940 rc = req_capsule_server_pack(pill);
2945 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2947 lh->mlh_type = MDT_NUL_LOCK;
2948 lh->mlh_reg_lh.cookie = 0ull;
2949 lh->mlh_reg_mode = LCK_MINMODE;
2950 lh->mlh_pdo_lh.cookie = 0ull;
2951 lh->mlh_pdo_mode = LCK_MINMODE;
2952 lh->mlh_rreg_lh.cookie = 0ull;
2953 lh->mlh_rreg_mode = LCK_MINMODE;
2956 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2958 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2959 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2963 * Initialize fields of struct mdt_thread_info. Other fields are left in
2964 * uninitialized state, because it's too expensive to zero out whole
2965 * mdt_thread_info (> 1K) on each request arrival.
2967 void mdt_thread_info_init(struct ptlrpc_request *req,
2968 struct mdt_thread_info *info)
2972 info->mti_pill = &req->rq_pill;
2975 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2976 mdt_lock_handle_init(&info->mti_lh[i]);
2978 /* mdt device: it can be NULL while CONNECT */
2979 if (req->rq_export) {
2980 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2981 info->mti_exp = req->rq_export;
2983 info->mti_mdt = NULL;
2984 info->mti_env = req->rq_svc_thread->t_env;
2985 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2987 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2988 info->mti_big_buf = LU_BUF_NULL;
2989 info->mti_body = NULL;
2990 info->mti_object = NULL;
2991 info->mti_dlm_req = NULL;
2992 info->mti_has_trans = 0;
2993 info->mti_cross_ref = 0;
2994 info->mti_opdata = 0;
2995 info->mti_big_lmm_used = 0;
2997 info->mti_spec.no_create = 0;
2998 info->mti_spec.sp_rm_entry = 0;
2999 info->mti_spec.sp_permitted = 0;
3000 info->mti_spec.sp_migrate_close = 0;
3002 info->mti_spec.u.sp_ea.eadata = NULL;
3003 info->mti_spec.u.sp_ea.eadatalen = 0;
3006 void mdt_thread_info_fini(struct mdt_thread_info *info)
3010 if (info->mti_object != NULL) {
3011 mdt_object_put(info->mti_env, info->mti_object);
3012 info->mti_object = NULL;
3015 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3016 mdt_lock_handle_fini(&info->mti_lh[i]);
3017 info->mti_env = NULL;
3018 info->mti_pill = NULL;
3019 info->mti_exp = NULL;
3021 if (unlikely(info->mti_big_buf.lb_buf != NULL))
3022 lu_buf_free(&info->mti_big_buf);
3025 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3027 struct mdt_thread_info *mti;
3029 mti = mdt_th_info(tsi->tsi_env);
3030 LASSERT(mti != NULL);
3032 mdt_thread_info_init(tgt_ses_req(tsi), mti);
3033 if (tsi->tsi_corpus != NULL) {
3034 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3035 lu_object_get(tsi->tsi_corpus);
3037 mti->mti_body = tsi->tsi_mdt_body;
3038 mti->mti_dlm_req = tsi->tsi_dlm_req;
3043 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3045 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3047 tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3048 set_current_state(TASK_UNINTERRUPTIBLE);
3049 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3052 return tgt_connect(tsi);
3070 static int mdt_intent_getattr(enum mdt_it_code opcode,
3071 struct mdt_thread_info *info,
3072 struct ldlm_lock **,
3075 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3076 struct mdt_thread_info *info,
3077 struct ldlm_lock **lockp,
3080 static int mdt_intent_layout(enum mdt_it_code opcode,
3081 struct mdt_thread_info *info,
3082 struct ldlm_lock **,
3084 static int mdt_intent_reint(enum mdt_it_code opcode,
3085 struct mdt_thread_info *info,
3086 struct ldlm_lock **,
3089 static struct mdt_it_flavor {
3090 const struct req_format *it_fmt;
3092 int (*it_act)(enum mdt_it_code ,
3093 struct mdt_thread_info *,
3094 struct ldlm_lock **,
3097 } mdt_it_flavor[] = {
3099 .it_fmt = &RQF_LDLM_INTENT,
3100 /*.it_flags = HABEO_REFERO,*/
3102 .it_act = mdt_intent_reint,
3103 .it_reint = REINT_OPEN
3106 .it_fmt = &RQF_LDLM_INTENT,
3108 * OCREAT is not a MUTABOR request as if the file
3110 * We do the extra check of OBD_CONNECT_RDONLY in
3111 * mdt_reint_open() when we really need to create
3115 .it_act = mdt_intent_reint,
3116 .it_reint = REINT_OPEN
3119 .it_fmt = &RQF_LDLM_INTENT,
3120 .it_flags = MUTABOR,
3121 .it_act = mdt_intent_reint,
3122 .it_reint = REINT_CREATE
3124 [MDT_IT_GETATTR] = {
3125 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3126 .it_flags = HABEO_REFERO,
3127 .it_act = mdt_intent_getattr
3129 [MDT_IT_READDIR] = {
3135 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3136 .it_flags = HABEO_REFERO,
3137 .it_act = mdt_intent_getattr
3140 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3141 .it_flags = MUTABOR,
3143 .it_reint = REINT_UNLINK
3147 .it_flags = MUTABOR,
3150 [MDT_IT_GETXATTR] = {
3151 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
3152 .it_flags = HABEO_CORPUS,
3153 .it_act = mdt_intent_getxattr
3156 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3158 .it_act = mdt_intent_layout
3163 mdt_intent_lock_replace(struct mdt_thread_info *info,
3164 struct ldlm_lock **lockp,
3165 struct mdt_lock_handle *lh,
3166 __u64 flags, int result)
3168 struct ptlrpc_request *req = mdt_info_req(info);
3169 struct ldlm_lock *lock = *lockp;
3170 struct ldlm_lock *new_lock;
3172 /* If possible resent found a lock, @lh is set to its handle */
3173 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3175 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3176 lh->mlh_reg_lh.cookie = 0;
3180 if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3181 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3182 * a resend case, however, it could be already destroyed
3183 * due to client eviction or a raced cancel RPC. */
3184 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
3185 lh->mlh_reg_lh.cookie);
3186 lh->mlh_reg_lh.cookie = 0;
3190 LASSERTF(new_lock != NULL,
3191 "lockh "LPX64" flags "LPX64" rc %d\n",
3192 lh->mlh_reg_lh.cookie, flags, result);
3195 * If we've already given this lock to a client once, then we should
3196 * have no readers or writers. Otherwise, we should have one reader
3197 * _or_ writer ref (which will be zeroed below) before returning the
3200 if (new_lock->l_export == req->rq_export) {
3201 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3203 LASSERT(new_lock->l_export == NULL);
3204 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3209 if (new_lock->l_export == req->rq_export) {
3211 * Already gave this to the client, which means that we
3212 * reconstructed a reply.
3214 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3217 LDLM_LOCK_RELEASE(new_lock);
3218 lh->mlh_reg_lh.cookie = 0;
3219 RETURN(ELDLM_LOCK_REPLACED);
3223 * Fixup the lock to be given to the client.
3225 lock_res_and_lock(new_lock);
3226 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3227 * possible blocking AST. */
3228 while (new_lock->l_readers > 0) {
3229 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3230 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3231 new_lock->l_readers--;
3233 while (new_lock->l_writers > 0) {
3234 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3235 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3236 new_lock->l_writers--;
3239 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3240 new_lock->l_blocking_ast = lock->l_blocking_ast;
3241 new_lock->l_completion_ast = lock->l_completion_ast;
3242 new_lock->l_remote_handle = lock->l_remote_handle;
3243 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3245 unlock_res_and_lock(new_lock);
3247 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3248 &new_lock->l_remote_handle,
3249 &new_lock->l_exp_hash);
3251 LDLM_LOCK_RELEASE(new_lock);
3252 lh->mlh_reg_lh.cookie = 0;
3254 RETURN(ELDLM_LOCK_REPLACED);
3257 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3258 struct ldlm_lock *new_lock,
3259 struct mdt_lock_handle *lh,
3262 struct ptlrpc_request *req = mdt_info_req(info);
3263 struct ldlm_request *dlmreq;
3265 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3268 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3270 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3271 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3273 if (flags & LDLM_FL_RESENT) {
3274 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3275 lh->mlh_reg_mode = new_lock->l_granted_mode;
3277 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3278 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3279 lh->mlh_reg_lh.cookie);
3284 * If the xid matches, then we know this is a resent request, and allow
3285 * it. (It's probably an OPEN, for which we don't send a lock.
3287 if (req_can_reconstruct(req, NULL))
3291 * This remote handle isn't enqueued, so we never received or processed
3292 * this request. Clear MSG_RESENT, because it can be handled like any
3293 * normal request now.
3295 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3297 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3298 dlmreq->lock_handle[0].cookie);
3301 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3302 struct mdt_thread_info *info,
3303 struct ldlm_lock **lockp,
3306 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3307 struct ldlm_reply *ldlm_rep = NULL;
3311 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3312 * (for the resend case) or a new lock. Below we will use it to
3313 * replace the original lock.
3315 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3316 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3317 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3318 rc = mdt_object_lock(info, info->mti_object, lhc,
3319 MDS_INODELOCK_XATTR);
3324 grc = mdt_getxattr(info);
3326 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, 0);
3328 if (mdt_info_req(info)->rq_repmsg != NULL)
3329 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3330 if (ldlm_rep == NULL)
3331 RETURN(err_serious(-EFAULT));
3333 ldlm_rep->lock_policy_res2 = grc;
3338 static int mdt_intent_getattr(enum mdt_it_code opcode,
3339 struct mdt_thread_info *info,
3340 struct ldlm_lock **lockp,
3343 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3345 struct ldlm_reply *ldlm_rep;
3346 struct mdt_body *reqbody;
3347 struct mdt_body *repbody;
3351 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3354 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3357 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3358 repbody->mbo_eadatasize = 0;
3359 repbody->mbo_aclsize = 0;
3363 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3365 case MDT_IT_GETATTR:
3366 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3370 CERROR("Unsupported intent (%d)\n", opcode);
3371 GOTO(out_shrink, rc = -EINVAL);
3374 rc = mdt_init_ucred_intent_getattr(info, reqbody);
3376 GOTO(out_shrink, rc);
3378 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3379 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3381 /* Get lock from request for possible resent case. */
3382 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3384 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3385 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3387 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3388 ldlm_rep->lock_policy_res2 = 0;
3389 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3390 ldlm_rep->lock_policy_res2) {
3391 lhc->mlh_reg_lh.cookie = 0ull;
3392 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3395 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3398 mdt_exit_ucred(info);
3400 mdt_client_compatibility(info);
3401 rc2 = mdt_fix_reply(info);
3407 static int mdt_intent_layout(enum mdt_it_code opcode,
3408 struct mdt_thread_info *info,
3409 struct ldlm_lock **lockp,
3412 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3413 struct layout_intent *layout;
3415 struct mdt_object *obj = NULL;
3416 int layout_size = 0;
3420 if (opcode != MDT_IT_LAYOUT) {
3421 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3426 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3430 if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3431 CERROR("%s: Unsupported layout intent opc %d\n",
3432 mdt_obd_name(info->mti_mdt), layout->li_opc);
3436 fid = &info->mti_tmp_fid2;
3437 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3439 /* Get lock from request for possible resent case. */
3440 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3442 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3444 GOTO(out, rc = PTR_ERR(obj));
3446 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3447 layout_size = mdt_attr_get_eabuf_size(info, obj);
3448 if (layout_size < 0)
3449 GOTO(out_obj, rc = layout_size);
3451 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3452 info->mti_mdt->mdt_max_mdsize = layout_size;
3455 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3456 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3458 rc = req_capsule_server_pack(info->mti_pill);
3462 mdt_object_put(info->mti_env, obj);
3464 if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3465 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3468 lhc->mlh_reg_lh.cookie = 0;
3473 static int mdt_intent_reint(enum mdt_it_code opcode,
3474 struct mdt_thread_info *info,
3475 struct ldlm_lock **lockp,
3478 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3479 struct ldlm_reply *rep = NULL;
3483 static const struct req_format *intent_fmts[REINT_MAX] = {
3484 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3485 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3490 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3494 if (mdt_it_flavor[opcode].it_reint != opc) {
3495 CERROR("Reint code %ld doesn't match intent: %d\n",
3497 RETURN(err_serious(-EPROTO));
3500 /* Get lock from request for possible resent case. */
3501 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3503 rc = mdt_reint_internal(info, lhc, opc);
3505 /* Check whether the reply has been packed successfully. */
3506 if (mdt_info_req(info)->rq_repmsg != NULL)
3507 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3509 RETURN(err_serious(-EFAULT));
3511 /* MDC expects this in any case */
3513 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3515 /* the open lock or the lock for cross-ref object should be
3516 * returned to the client */
3517 if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3518 (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3519 rep->lock_policy_res2 = 0;
3520 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3524 rep->lock_policy_res2 = clear_serious(rc);
3526 if (rep->lock_policy_res2 == -ENOENT &&
3527 mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3528 !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3529 rep->lock_policy_res2 = 0;
3531 lhc->mlh_reg_lh.cookie = 0ull;
3532 if (rc == -ENOTCONN || rc == -ENODEV ||
3533 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3535 * If it is the disconnect error (ENODEV & ENOCONN), the error
3536 * will be returned by rq_status, and client at ptlrpc layer
3537 * will detect this, then disconnect, reconnect the import
3538 * immediately, instead of impacting the following the rpc.
3543 * For other cases, the error will be returned by intent, and client
3544 * will retrieve the result from intent.
3546 RETURN(ELDLM_LOCK_ABORTED);
3549 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3557 case IT_OPEN|IT_CREAT:
3564 rc = MDT_IT_READDIR;
3567 rc = MDT_IT_GETATTR;
3579 rc = MDT_IT_GETXATTR;
3584 case IT_QUOTA_DQACQ:
3589 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3596 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3597 struct mdt_thread_info *info,
3598 struct ldlm_lock **lockp, __u64 flags)
3600 struct req_capsule *pill = info->mti_pill;
3601 struct ptlrpc_request *req = mdt_info_req(info);
3602 struct mdt_it_flavor *flv;
3607 opc = mdt_intent_code(itopc);
3611 if (opc == MDT_IT_QUOTA) {
3612 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3615 RETURN(-EOPNOTSUPP);
3617 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3618 /* pass the request to quota master */
3619 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3620 mdt_info_req(info), lockp,
3625 flv = &mdt_it_flavor[opc];
3626 if (flv->it_fmt != NULL)
3627 req_capsule_extend(pill, flv->it_fmt);
3629 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3633 if (flv->it_flags & MUTABOR &&
3634 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3637 if (flv->it_act != NULL) {
3638 struct ldlm_reply *rep;
3640 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3642 /* execute policy */
3643 rc = flv->it_act(opc, info, lockp, flags);
3645 /* Check whether the reply has been packed successfully. */
3646 if (req->rq_repmsg != NULL) {
3647 rep = req_capsule_server_get(info->mti_pill,
3649 rep->lock_policy_res2 =
3650 ptlrpc_status_hton(rep->lock_policy_res2);
3657 static int mdt_intent_policy(struct ldlm_namespace *ns,
3658 struct ldlm_lock **lockp, void *req_cookie,
3659 enum ldlm_mode mode, __u64 flags, void *data)
3661 struct tgt_session_info *tsi;
3662 struct mdt_thread_info *info;
3663 struct ptlrpc_request *req = req_cookie;
3664 struct ldlm_intent *it;
3665 struct req_capsule *pill;
3670 LASSERT(req != NULL);
3672 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3674 info = tsi2mdt_info(tsi);
3675 LASSERT(info != NULL);
3676 pill = info->mti_pill;
3677 LASSERT(pill->rc_req == req);
3679 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3680 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3681 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3683 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3687 /* Lock without inodebits makes no sense and will oops
3688 * later in ldlm. Let's check it now to see if we have
3689 * ibits corrupted somewhere in mdt_intent_opc().
3690 * The case for client miss to set ibits has been
3691 * processed by others. */
3692 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3693 lr_type == LDLM_IBITS,
3694 info->mti_dlm_req->lock_desc.\
3695 l_policy_data.l_inodebits.bits != 0));
3697 rc = err_serious(-EFAULT);
3699 /* No intent was provided */
3700 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3701 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3702 rc = req_capsule_server_pack(pill);
3704 rc = err_serious(rc);
3706 mdt_thread_info_fini(info);
3710 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3712 struct seq_server_site *ss = mdt_seq_site(mdt);
3714 if (ss->ss_node_id == 0)
3717 if (ss->ss_client_seq != NULL) {
3718 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3719 ss->ss_client_seq->lcs_exp = NULL;
3722 if (ss->ss_server_fld != NULL) {
3723 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3724 ss->ss_server_fld->lsf_control_exp = NULL;
3728 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3730 struct seq_server_site *ss = mdt_seq_site(mdt);
3735 if (ss->ss_server_seq != NULL)
3736 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3739 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3741 mdt_seq_fini_cli(mdt);
3742 mdt_deregister_seq_exp(mdt);
3744 return seq_site_fini(env, mdt_seq_site(mdt));
3748 * It will retrieve its FLDB entries from MDT0, and it only happens
3749 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3750 * and it needs to refresh FLDB from the MDT0.
3752 static int mdt_register_lwp_callback(void *data)
3755 struct mdt_device *mdt = data;
3756 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3760 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3762 rc = lu_env_init(&env, LCT_MD_THREAD);
3764 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3768 /* Allocate new sequence now to avoid creating local transaction
3769 * in the normal transaction process */
3770 rc = seq_server_check_and_alloc_super(&env,
3771 mdt_seq_site(mdt)->ss_server_seq);
3776 rc = fld_update_from_controller(&env, fld);
3778 CERROR("%s: cannot update controller: rc = %d\n",
3779 mdt_obd_name(mdt), rc);
3788 static int mdt_register_seq_exp(struct mdt_device *mdt)
3790 struct seq_server_site *ss = mdt_seq_site(mdt);
3791 char *lwp_name = NULL;
3794 if (ss->ss_node_id == 0)
3797 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3798 if (lwp_name == NULL)
3799 GOTO(out_free, rc = -ENOMEM);
3801 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3805 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3810 rc = lustre_register_lwp_item(lwp_name,
3811 &ss->ss_server_fld->lsf_control_exp,
3812 mdt_register_lwp_callback, mdt);
3814 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3815 ss->ss_client_seq->lcs_exp = NULL;
3819 if (lwp_name != NULL)
3820 OBD_FREE(lwp_name, MAX_OBD_NAME);
3826 * Init client sequence manager which is used by local MDS to talk to sequence
3827 * controller on remote node.
3829 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3831 struct seq_server_site *ss = mdt_seq_site(mdt);
3836 /* check if this is adding the first MDC and controller is not yet
3838 OBD_ALLOC_PTR(ss->ss_client_seq);
3839 if (ss->ss_client_seq == NULL)
3842 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3843 if (prefix == NULL) {
3844 OBD_FREE_PTR(ss->ss_client_seq);
3845 ss->ss_client_seq = NULL;
3849 /* Note: seq_client_fini will be called in seq_site_fini */
3850 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3851 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3852 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3854 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3856 OBD_FREE_PTR(ss->ss_client_seq);
3857 ss->ss_client_seq = NULL;
3861 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3866 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3868 struct seq_server_site *ss;
3872 ss = mdt_seq_site(mdt);
3873 /* init sequence controller server(MDT0) */
3874 if (ss->ss_node_id == 0) {
3875 OBD_ALLOC_PTR(ss->ss_control_seq);
3876 if (ss->ss_control_seq == NULL)
3879 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3880 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3883 GOTO(out_seq_fini, rc);
3886 /* Init normal sequence server */
3887 OBD_ALLOC_PTR(ss->ss_server_seq);
3888 if (ss->ss_server_seq == NULL)
3889 GOTO(out_seq_fini, rc = -ENOMEM);
3891 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3892 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3894 GOTO(out_seq_fini, rc);
3896 /* init seq client for seq server to talk to seq controller(MDT0) */
3897 rc = mdt_seq_init_cli(env, mdt);
3899 GOTO(out_seq_fini, rc);
3901 if (ss->ss_node_id != 0)
3902 /* register controller export through lwp */
3903 rc = mdt_register_seq_exp(mdt);
3908 mdt_seq_fini(env, mdt);
3916 static int mdt_fld_fini(const struct lu_env *env,
3917 struct mdt_device *m)
3919 struct seq_server_site *ss = mdt_seq_site(m);
3922 if (ss && ss->ss_server_fld) {
3923 fld_server_fini(env, ss->ss_server_fld);
3924 OBD_FREE_PTR(ss->ss_server_fld);
3925 ss->ss_server_fld = NULL;
3931 static int mdt_fld_init(const struct lu_env *env,
3933 struct mdt_device *m)
3935 struct seq_server_site *ss;
3939 ss = mdt_seq_site(m);
3941 OBD_ALLOC_PTR(ss->ss_server_fld);
3942 if (ss->ss_server_fld == NULL)
3943 RETURN(rc = -ENOMEM);
3945 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3948 OBD_FREE_PTR(ss->ss_server_fld);
3949 ss->ss_server_fld = NULL;
3956 static void mdt_stack_pre_fini(const struct lu_env *env,
3957 struct mdt_device *m, struct lu_device *top)
3959 struct lustre_cfg_bufs *bufs;
3960 struct lustre_cfg *lcfg;
3961 struct mdt_thread_info *info;
3966 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3967 LASSERT(info != NULL);
3969 bufs = &info->mti_u.bufs;
3971 LASSERT(m->mdt_child_exp);
3972 LASSERT(m->mdt_child_exp->exp_obd);
3974 /* process cleanup, pass mdt obd name to get obd umount flags */
3975 /* XXX: this is needed because all layers are referenced by
3976 * objects (some of them are pinned by osd, for example *
3977 * the proper solution should be a model where object used
3978 * by osd only doesn't have mdt/mdd slices -bzzz */
3979 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3980 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3981 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3985 top->ld_ops->ldo_process_config(env, top, lcfg);
3986 lustre_cfg_free(lcfg);
3990 static void mdt_stack_fini(const struct lu_env *env,
3991 struct mdt_device *m, struct lu_device *top)
3993 struct obd_device *obd = mdt2obd_dev(m);
3994 struct lustre_cfg_bufs *bufs;
3995 struct lustre_cfg *lcfg;
3996 struct mdt_thread_info *info;
4000 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4001 LASSERT(info != NULL);
4003 lu_dev_del_linkage(top->ld_site, top);
4005 lu_site_purge(env, top->ld_site, -1);
4007 bufs = &info->mti_u.bufs;
4008 /* process cleanup, pass mdt obd name to get obd umount flags */
4009 /* another purpose is to let all layers to release their objects */
4010 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4015 lustre_cfg_bufs_set_string(bufs, 1, flags);
4016 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4021 top->ld_ops->ldo_process_config(env, top, lcfg);
4022 lustre_cfg_free(lcfg);
4024 lu_site_purge(env, top->ld_site, -1);
4026 m->mdt_child = NULL;
4027 m->mdt_bottom = NULL;
4029 obd_disconnect(m->mdt_child_exp);
4030 m->mdt_child_exp = NULL;
4032 obd_disconnect(m->mdt_bottom_exp);
4033 m->mdt_child_exp = NULL;
4036 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4037 const char *next, struct obd_export **exp)
4039 struct obd_connect_data *data = NULL;
4040 struct obd_device *obd;
4044 OBD_ALLOC_PTR(data);
4046 GOTO(out, rc = -ENOMEM);
4048 obd = class_name2obd(next);
4050 CERROR("%s: can't locate next device: %s\n",
4051 mdt_obd_name(m), next);
4052 GOTO(out, rc = -ENOTCONN);
4055 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4056 data->ocd_version = LUSTRE_VERSION_CODE;
4058 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4060 CERROR("%s: cannot connect to next dev %s (%d)\n",
4061 mdt_obd_name(m), next, rc);
4071 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4072 struct lustre_cfg *cfg)
4074 char *dev = lustre_cfg_string(cfg, 0);
4075 int rc, name_size, uuid_size;
4076 char *name, *uuid, *p;
4077 struct lustre_cfg_bufs *bufs;
4078 struct lustre_cfg *lcfg;
4079 struct obd_device *obd;
4080 struct lustre_profile *lprof;
4081 struct lu_site *site;
4084 /* in 1.8 we had the only device in the stack - MDS.
4085 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4086 * in 2.3 OSD is instantiated by obd_mount.c, so we need
4087 * to generate names and setup MDT, MDD. MDT will be using
4088 * generated name to connect to MDD. for MDD the next device
4089 * will be LOD with name taken from so called "profile" which
4090 * is generated by mount_option line
4092 * 1.8 MGS generates config. commands like this:
4093 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
4094 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
4095 * 2.0 MGS generates config. commands like this:
4096 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
4097 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4098 * 3:lustre-MDT0000-mdtlov 4:f
4100 * we generate MDD name from MDT one, just replacing T with D
4102 * after all the preparations, the logical equivalent will be
4103 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
4104 * 3:lustre-MDT0000-mdtlov 4:f
4105 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4106 * 3:lustre-MDD0000 4:f
4108 * notice we build the stack from down to top: MDD first, then MDT */
4110 name_size = MAX_OBD_NAME;
4111 uuid_size = MAX_OBD_NAME;
4113 OBD_ALLOC(name, name_size);
4114 OBD_ALLOC(uuid, uuid_size);
4115 if (name == NULL || uuid == NULL)
4116 GOTO(cleanup_mem, rc = -ENOMEM);
4118 OBD_ALLOC_PTR(bufs);
4120 GOTO(cleanup_mem, rc = -ENOMEM);
4123 p = strstr(name, "-MDT");
4125 GOTO(free_bufs, rc = -ENOMEM);
4128 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4130 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4131 if (lprof == NULL || lprof->lp_dt == NULL) {
4132 CERROR("can't find the profile: %s\n",
4133 lustre_cfg_string(cfg, 0));
4134 GOTO(free_bufs, rc = -EINVAL);
4137 lustre_cfg_bufs_reset(bufs, name);
4138 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4139 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4140 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4142 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4144 GOTO(put_profile, rc = -ENOMEM);
4146 rc = class_attach(lcfg);
4148 GOTO(lcfg_cleanup, rc);
4150 obd = class_name2obd(name);
4152 CERROR("Can not find obd %s (%s in config)\n",
4153 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4154 GOTO(lcfg_cleanup, rc = -EINVAL);
4157 lustre_cfg_free(lcfg);
4159 lustre_cfg_bufs_reset(bufs, name);
4160 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4161 lustre_cfg_bufs_set_string(bufs, 2, dev);
4162 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4164 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4166 GOTO(class_detach, rc = -ENOMEM);
4168 rc = class_setup(obd, lcfg);
4170 GOTO(class_detach, rc);
4172 /* connect to MDD we just setup */
4173 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4175 GOTO(class_detach, rc);
4177 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4179 LASSERT(mdt_lu_site(mdt) == NULL);
4180 mdt->mdt_lu_dev.ld_site = site;
4181 site->ls_top_dev = &mdt->mdt_lu_dev;
4182 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4184 /* now connect to bottom OSD */
4185 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4186 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4188 GOTO(class_detach, rc);
4190 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4192 rc = lu_env_refill((struct lu_env *)env);
4194 CERROR("Failure to refill session: '%d'\n", rc);
4196 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4201 class_detach(obd, lcfg);
4203 lustre_cfg_free(lcfg);
4205 class_put_profile(lprof);
4210 OBD_FREE(name, name_size);
4212 OBD_FREE(uuid, uuid_size);
4216 /* setup quota master target on MDT0 */
4217 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4218 struct lustre_cfg *cfg)
4220 struct obd_device *obd;
4221 char *dev = lustre_cfg_string(cfg, 0);
4222 char *qmtname, *uuid, *p;
4223 struct lustre_cfg_bufs *bufs;
4224 struct lustre_cfg *lcfg;
4225 struct lustre_profile *lprof;
4226 struct obd_connect_data *data;
4230 LASSERT(mdt->mdt_qmt_exp == NULL);
4231 LASSERT(mdt->mdt_qmt_dev == NULL);
4233 /* quota master is on MDT0 only for now */
4234 if (mdt->mdt_seq_site.ss_node_id != 0)
4237 /* MGS generates config commands which look as follows:
4238 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4239 * 3:lustre-MDT0000-mdtlov 4:f
4241 * We generate the QMT name from the MDT one, just replacing MD with QM
4242 * after all the preparations, the logical equivalent will be:
4243 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4244 * 3:lustre-MDT0000-osd 4:f */
4245 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4246 OBD_ALLOC(uuid, UUID_MAX);
4247 OBD_ALLOC_PTR(bufs);
4248 OBD_ALLOC_PTR(data);
4249 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4250 GOTO(cleanup_mem, rc = -ENOMEM);
4252 strcpy(qmtname, dev);
4253 p = strstr(qmtname, "-MDT");
4255 GOTO(cleanup_mem, rc = -ENOMEM);
4256 /* replace MD with QM */
4260 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4262 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4263 if (lprof == NULL || lprof->lp_dt == NULL) {
4264 CERROR("can't find profile for %s\n",
4265 lustre_cfg_string(cfg, 0));
4266 GOTO(cleanup_mem, rc = -EINVAL);
4269 lustre_cfg_bufs_reset(bufs, qmtname);
4270 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4271 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4272 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4274 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4276 GOTO(put_profile, rc = -ENOMEM);
4278 rc = class_attach(lcfg);
4280 GOTO(lcfg_cleanup, rc);
4282 obd = class_name2obd(qmtname);
4284 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4285 lustre_cfg_string(cfg, 0));
4286 GOTO(lcfg_cleanup, rc = -EINVAL);
4289 lustre_cfg_free(lcfg);
4291 lustre_cfg_bufs_reset(bufs, qmtname);
4292 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4293 lustre_cfg_bufs_set_string(bufs, 2, dev);
4295 /* for quota, the next device should be the OSD device */
4296 lustre_cfg_bufs_set_string(bufs, 3,
4297 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4299 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4301 GOTO(class_detach, rc = -ENOMEM);
4303 rc = class_setup(obd, lcfg);
4305 GOTO(class_detach, rc);
4307 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4309 /* configure local quota objects */
4310 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4314 GOTO(class_cleanup, rc);
4316 /* connect to quota master target */
4317 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4318 data->ocd_version = LUSTRE_VERSION_CODE;
4319 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4322 CERROR("cannot connect to quota master device %s (%d)\n",
4324 GOTO(class_cleanup, rc);
4330 class_manual_cleanup(obd);
4331 mdt->mdt_qmt_dev = NULL;
4335 class_detach(obd, lcfg);
4337 lustre_cfg_free(lcfg);
4339 class_put_profile(lprof);
4344 OBD_FREE(qmtname, MAX_OBD_NAME);
4346 OBD_FREE(uuid, UUID_MAX);
4352 /* Shutdown quota master target associated with mdt */
4353 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4357 if (mdt->mdt_qmt_exp == NULL)
4359 LASSERT(mdt->mdt_qmt_dev != NULL);
4361 /* the qmt automatically shuts down when the mdt disconnects */
4362 obd_disconnect(mdt->mdt_qmt_exp);
4363 mdt->mdt_qmt_exp = NULL;
4364 mdt->mdt_qmt_dev = NULL;
4368 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4369 * for now. This will be removed along with converting rest of MDT code
4370 * to use tgt_session_info */
4371 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4373 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4376 rc = mdt_getxattr(info);
4378 mdt_thread_info_fini(info);
4382 static struct tgt_handler mdt_tgt_handlers[] = {
4383 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4384 0, MDS_CONNECT, mdt_tgt_connect,
4385 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4386 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4387 0, MDS_DISCONNECT, tgt_disconnect,
4388 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4389 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4390 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4391 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4392 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4393 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GET_ROOT, mdt_get_root),
4394 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4395 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4397 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4398 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4399 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4400 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4401 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4402 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4403 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4404 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4406 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4407 mdt_hsm_ct_register),
4408 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4409 mdt_hsm_ct_unregister),
4410 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4412 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4414 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4415 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4417 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4422 static struct tgt_handler mdt_sec_ctx_ops[] = {
4423 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4424 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4425 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4428 static struct tgt_handler mdt_quota_ops[] = {
4429 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4432 static struct tgt_opc_slice mdt_common_slice[] = {
4434 .tos_opc_start = MDS_FIRST_OPC,
4435 .tos_opc_end = MDS_LAST_OPC,
4436 .tos_hs = mdt_tgt_handlers
4439 .tos_opc_start = OBD_FIRST_OPC,
4440 .tos_opc_end = OBD_LAST_OPC,
4441 .tos_hs = tgt_obd_handlers
4444 .tos_opc_start = LDLM_FIRST_OPC,
4445 .tos_opc_end = LDLM_LAST_OPC,
4446 .tos_hs = tgt_dlm_handlers
4449 .tos_opc_start = SEC_FIRST_OPC,
4450 .tos_opc_end = SEC_LAST_OPC,
4451 .tos_hs = mdt_sec_ctx_ops
4454 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4455 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4456 .tos_hs = tgt_out_handlers
4459 .tos_opc_start = FLD_FIRST_OPC,
4460 .tos_opc_end = FLD_LAST_OPC,
4461 .tos_hs = fld_handlers
4464 .tos_opc_start = SEQ_FIRST_OPC,
4465 .tos_opc_end = SEQ_LAST_OPC,
4466 .tos_hs = seq_handlers
4469 .tos_opc_start = QUOTA_DQACQ,
4470 .tos_opc_end = QUOTA_LAST_OPC,
4471 .tos_hs = mdt_quota_ops
4474 .tos_opc_start = LLOG_FIRST_OPC,
4475 .tos_opc_end = LLOG_LAST_OPC,
4476 .tos_hs = tgt_llog_handlers
4479 .tos_opc_start = LFSCK_FIRST_OPC,
4480 .tos_opc_end = LFSCK_LAST_OPC,
4481 .tos_hs = tgt_lfsck_handlers
4489 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4491 struct md_device *next = m->mdt_child;
4492 struct lu_device *d = &m->mdt_lu_dev;
4493 struct obd_device *obd = mdt2obd_dev(m);
4494 struct lfsck_stop stop;
4497 if (m->mdt_md_root != NULL) {
4498 mdt_object_put(env, m->mdt_md_root);
4499 m->mdt_md_root = NULL;
4502 stop.ls_status = LS_PAUSED;
4504 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4506 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4507 target_recovery_fini(obd);
4508 ping_evictor_stop();
4510 if (m->mdt_opts.mo_coordinator)
4511 mdt_hsm_cdt_stop(m);
4513 mdt_hsm_cdt_fini(m);
4515 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4516 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4518 if (m->mdt_namespace != NULL)
4519 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4520 d->ld_obd->obd_force);
4522 obd_exports_barrier(obd);
4523 obd_zombie_barrier();
4527 tgt_fini(env, &m->mdt_lut);
4528 mdt_fs_cleanup(env, m);
4529 upcall_cache_cleanup(m->mdt_identity_cache);
4530 m->mdt_identity_cache = NULL;
4532 if (m->mdt_namespace != NULL) {
4533 ldlm_namespace_free_post(m->mdt_namespace);
4534 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4537 mdt_quota_fini(env, m);
4539 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4541 mdt_seq_fini(env, m);
4542 mdt_fld_fini(env, m);
4547 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4549 LASSERT(atomic_read(&d->ld_ref) == 0);
4551 server_put_mount(mdt_obd_name(m), true);
4556 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4558 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4559 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4561 struct mdt_thread_info *info;
4562 struct obd_device *obd;
4563 const char *dev = lustre_cfg_string(cfg, 0);
4564 const char *num = lustre_cfg_string(cfg, 2);
4565 struct lustre_mount_info *lmi = NULL;
4566 struct lustre_sb_info *lsi;
4568 struct seq_server_site *ss_site;
4569 const char *identity_upcall = "NONE";
4570 struct md_device *next;
4576 lu_device_init(&m->mdt_lu_dev, ldt);
4578 * Environment (env) might be missing mdt_thread_key values at that
4579 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4582 * Usually device allocation path doesn't use module key values, but
4583 * mdt has to do a lot of work here, so allocate key value.
4585 rc = lu_env_refill((struct lu_env *)env);
4589 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4590 LASSERT(info != NULL);
4592 obd = class_name2obd(dev);
4593 LASSERT(obd != NULL);
4595 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4596 m->mdt_opts.mo_evict_tgt_nids = 1;
4597 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4599 /* default is coordinator off, it is started through conf_param
4601 m->mdt_opts.mo_coordinator = 0;
4603 lmi = server_get_mount(dev);
4605 CERROR("Cannot get mount info for %s!\n", dev);
4608 lsi = s2lsi(lmi->lmi_sb);
4609 /* CMD is supported only in IAM mode */
4611 node_id = simple_strtol(num, NULL, 10);
4612 obd->u.obt.obt_magic = OBT_MAGIC;
4613 if (lsi->lsi_lmd != NULL &&
4614 lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4615 m->mdt_skip_lfsck = 1;
4618 m->mdt_squash.rsi_uid = 0;
4619 m->mdt_squash.rsi_gid = 0;
4620 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4621 init_rwsem(&m->mdt_squash.rsi_sem);
4622 spin_lock_init(&m->mdt_lock);
4623 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4624 m->mdt_enable_remote_dir = 0;
4625 m->mdt_enable_remote_dir_gid = 0;
4627 atomic_set(&m->mdt_mds_mds_conns, 0);
4628 atomic_set(&m->mdt_async_commit_count, 0);
4630 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4631 m->mdt_lu_dev.ld_obd = obd;
4632 /* Set this lu_device to obd for error handling purposes. */
4633 obd->obd_lu_dev = &m->mdt_lu_dev;
4635 /* init the stack */
4636 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4638 CERROR("%s: Can't init device stack, rc %d\n",
4639 mdt_obd_name(m), rc);
4644 ss_site = mdt_seq_site(m);
4645 s->ld_seq_site = ss_site;
4648 /* set server index */
4649 ss_site->ss_node_id = node_id;
4651 /* failover is the default
4652 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4653 * assumed whose ss_node_id == 0 XXX
4655 obd->obd_replayable = 1;
4656 /* No connection accepted until configurations will finish */
4657 obd->obd_no_conn = 1;
4659 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4660 char *str = lustre_cfg_string(cfg, 4);
4661 if (strchr(str, 'n')) {
4662 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4663 obd->obd_replayable = 0;
4667 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4669 GOTO(err_fini_stack, rc);
4671 rc = mdt_seq_init(env, m);
4673 GOTO(err_fini_fld, rc);
4675 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4676 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4677 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4678 LDLM_NAMESPACE_SERVER,
4679 LDLM_NAMESPACE_GREEDY,
4681 if (m->mdt_namespace == NULL)
4682 GOTO(err_fini_seq, rc = -ENOMEM);
4684 m->mdt_namespace->ns_lvbp = m;
4685 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4687 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4688 /* set obd_namespace for compatibility with old code */
4689 obd->obd_namespace = m->mdt_namespace;
4691 rc = mdt_hsm_cdt_init(m);
4693 CERROR("%s: error initializing coordinator, rc %d\n",
4694 mdt_obd_name(m), rc);
4695 GOTO(err_free_ns, rc);
4698 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4699 OBD_FAIL_MDS_ALL_REQUEST_NET,
4700 OBD_FAIL_MDS_ALL_REPLY_NET);
4702 GOTO(err_free_hsm, rc);
4704 rc = mdt_fs_setup(env, m, obd, lsi);
4708 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4710 next = m->mdt_child;
4711 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4714 GOTO(err_fs_cleanup, rc);
4716 if (mntopts & MNTOPT_USERXATTR)
4717 m->mdt_opts.mo_user_xattr = 1;
4719 m->mdt_opts.mo_user_xattr = 0;
4721 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4723 GOTO(err_fs_cleanup, rc);
4725 if (mntopts & MNTOPT_ACL)
4726 m->mdt_opts.mo_acl = 1;
4728 m->mdt_opts.mo_acl = 0;
4730 /* XXX: to support suppgid for ACL, we enable identity_upcall
4731 * by default, otherwise, maybe got unexpected -EACCESS. */
4732 if (m->mdt_opts.mo_acl)
4733 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4735 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4737 &mdt_identity_upcall_cache_ops);
4738 if (IS_ERR(m->mdt_identity_cache)) {
4739 rc = PTR_ERR(m->mdt_identity_cache);
4740 m->mdt_identity_cache = NULL;
4741 GOTO(err_fs_cleanup, rc);
4744 rc = mdt_procfs_init(m, dev);
4746 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4747 GOTO(err_recovery, rc);
4750 rc = mdt_quota_init(env, m, cfg);
4752 GOTO(err_procfs, rc);
4754 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4755 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4756 "mdt_ldlm_client", m->mdt_ldlm_client);
4758 ping_evictor_start();
4760 /* recovery will be started upon mdt_prepare()
4761 * when the whole stack is complete and ready
4762 * to serve the requests */
4764 /* Reduce the initial timeout on an MDS because it doesn't need such
4765 * a long timeout as an OST does. Adaptive timeouts will adjust this
4766 * value appropriately. */
4767 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4768 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4774 target_recovery_fini(obd);
4775 upcall_cache_cleanup(m->mdt_identity_cache);
4776 m->mdt_identity_cache = NULL;
4778 mdt_fs_cleanup(env, m);
4780 tgt_fini(env, &m->mdt_lut);
4782 mdt_hsm_cdt_fini(m);
4784 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4785 obd->obd_namespace = m->mdt_namespace = NULL;
4787 mdt_seq_fini(env, m);
4789 mdt_fld_fini(env, m);
4791 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4794 server_put_mount(dev, true);
4798 /* For interoperability, the left element is old parameter, the right one
4799 * is the new version of the parameter, if some parameter is deprecated,
4800 * the new version should be set as NULL. */
4801 static struct cfg_interop_param mdt_interop_param[] = {
4802 { "mdt.group_upcall", NULL },
4803 { "mdt.quota_type", NULL },
4804 { "mdd.quota_type", NULL },
4805 { "mdt.som", NULL },
4806 { "mdt.rootsquash", "mdt.root_squash" },
4807 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4811 /* used by MGS to process specific configurations */
4812 static int mdt_process_config(const struct lu_env *env,
4813 struct lu_device *d, struct lustre_cfg *cfg)
4815 struct mdt_device *m = mdt_dev(d);
4816 struct md_device *md_next = m->mdt_child;
4817 struct lu_device *next = md2lu_dev(md_next);
4821 switch (cfg->lcfg_command) {
4823 struct obd_device *obd = d->ld_obd;
4825 /* For interoperability */
4826 struct cfg_interop_param *ptr = NULL;
4827 struct lustre_cfg *old_cfg = NULL;
4830 param = lustre_cfg_string(cfg, 1);
4831 if (param == NULL) {
4832 CERROR("param is empty\n");
4837 ptr = class_find_old_param(param, mdt_interop_param);
4839 if (ptr->new_param == NULL) {
4841 CWARN("For interoperability, skip this %s."
4842 " It is obsolete.\n", ptr->old_param);
4846 CWARN("Found old param %s, changed it to %s.\n",
4847 ptr->old_param, ptr->new_param);
4850 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4857 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4859 if (rc > 0 || rc == -ENOSYS) {
4860 /* is it an HSM var ? */
4861 rc = class_process_proc_param(PARAM_HSM,
4862 hsm_cdt_get_proc_vars(),
4864 if (rc > 0 || rc == -ENOSYS)
4865 /* we don't understand; pass it on */
4866 rc = next->ld_ops->ldo_process_config(env, next,
4870 if (old_cfg != NULL)
4871 lustre_cfg_free(cfg);
4876 /* others are passed further */
4877 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4883 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4884 const struct lu_object_header *hdr,
4885 struct lu_device *d)
4887 struct mdt_object *mo;
4891 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4893 struct lu_object *o;
4894 struct lu_object_header *h;
4897 h = &mo->mot_header;
4898 lu_object_header_init(h);
4899 lu_object_init(o, h, d);
4900 lu_object_add_top(h, o);
4901 o->lo_ops = &mdt_obj_ops;
4902 spin_lock_init(&mo->mot_write_lock);
4903 mutex_init(&mo->mot_lov_mutex);
4904 init_rwsem(&mo->mot_open_sem);
4910 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4911 const struct lu_object_conf *unused)
4913 struct mdt_device *d = mdt_dev(o->lo_dev);
4914 struct lu_device *under;
4915 struct lu_object *below;
4919 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4920 PFID(lu_object_fid(o)));
4922 under = &d->mdt_child->md_lu_dev;
4923 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4924 if (below != NULL) {
4925 lu_object_add(o, below);
4932 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4934 struct mdt_object *mo = mdt_obj(o);
4935 struct lu_object_header *h;
4939 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4940 PFID(lu_object_fid(o)));
4942 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4943 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4946 lu_object_header_fini(h);
4947 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4952 static int mdt_object_print(const struct lu_env *env, void *cookie,
4953 lu_printer_t p, const struct lu_object *o)
4955 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4957 return (*p)(env, cookie,
4958 LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
4959 mdto, mdto->mot_lov_created ? "lov_created" : "",
4960 mdto->mot_cache_attr ? "cache_attr" : "",
4961 mdto->mot_write_count);
4964 static int mdt_prepare(const struct lu_env *env,
4965 struct lu_device *pdev,
4966 struct lu_device *cdev)
4968 struct mdt_device *mdt = mdt_dev(cdev);
4969 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4970 struct obd_device *obd = cdev->ld_obd;
4977 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4981 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4985 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4989 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4990 /* The LFSCK instance is registered just now, so it must be there when
4991 * register the namespace to such instance. */
4992 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4994 if (mdt->mdt_seq_site.ss_node_id == 0) {
4995 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4996 &mdt->mdt_md_root_fid);
5001 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5003 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5004 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5005 LASSERT(obd->obd_no_conn);
5006 spin_lock(&obd->obd_dev_lock);
5007 obd->obd_no_conn = 0;
5008 spin_unlock(&obd->obd_dev_lock);
5010 if (obd->obd_recovering == 0)
5011 mdt_postrecov(env, mdt);
5016 const struct lu_device_operations mdt_lu_ops = {
5017 .ldo_object_alloc = mdt_object_alloc,
5018 .ldo_process_config = mdt_process_config,
5019 .ldo_prepare = mdt_prepare,
5022 static const struct lu_object_operations mdt_obj_ops = {
5023 .loo_object_init = mdt_object_init,
5024 .loo_object_free = mdt_object_free,
5025 .loo_object_print = mdt_object_print
5028 static int mdt_obd_set_info_async(const struct lu_env *env,
5029 struct obd_export *exp,
5030 __u32 keylen, void *key,
5031 __u32 vallen, void *val,
5032 struct ptlrpc_request_set *set)
5038 if (KEY_IS(KEY_SPTLRPC_CONF)) {
5039 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
5047 * Match client and server connection feature flags.
5049 * Compute the compatibility flags for a connection request based on
5050 * features mutually supported by client and server.
5052 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5053 * must not be updated here, otherwise a partially initialized value may
5054 * be exposed. After the connection request is successfully processed,
5055 * the top-level MDT connect request handler atomically updates the export
5056 * connect flags from the obd_connect_data::ocd_connect_flags field of the
5057 * reply. \see mdt_connect().
5059 * \param exp the obd_export associated with this client/target pair
5060 * \param mdt the target device for the connection
5061 * \param data stores data for this connect request
5064 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5065 * \retval -EBADE client and server feature requirements are incompatible
5067 static int mdt_connect_internal(struct obd_export *exp,
5068 struct mdt_device *mdt,
5069 struct obd_connect_data *data)
5071 LASSERT(data != NULL);
5073 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5074 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5075 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5077 if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5078 !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5079 CWARN("%s: client %s does not support ibits lock, either "
5080 "very old or an invalid client: flags "LPX64"\n",
5081 mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5082 data->ocd_connect_flags);
5086 if (!mdt->mdt_opts.mo_acl)
5087 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5089 if (!mdt->mdt_opts.mo_user_xattr)
5090 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5092 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5093 data->ocd_brw_size = min(data->ocd_brw_size,
5094 (__u32)MD_MAX_BRW_SIZE);
5095 if (data->ocd_brw_size == 0) {
5096 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5097 " ocd_version: %x ocd_grant: %d "
5098 "ocd_index: %u ocd_brw_size is "
5099 "unexpectedly zero, network data "
5100 "corruption? Refusing connection of this"
5103 exp->exp_client_uuid.uuid,
5104 exp, data->ocd_connect_flags, data->ocd_version,
5105 data->ocd_grant, data->ocd_index);
5110 /* NB: Disregard the rule against updating
5111 * exp_connect_data.ocd_connect_flags in this case, since
5112 * tgt_client_new() needs to know if this is a lightweight
5113 * connection, and it is safe to expose this flag before
5114 * connection processing completes. */
5115 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5116 spin_lock(&exp->exp_lock);
5117 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5118 spin_unlock(&exp->exp_lock);
5121 data->ocd_version = LUSTRE_VERSION_CODE;
5123 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5124 CWARN("%s: MDS requires FID support, but client not\n",
5129 if (OCD_HAS_FLAG(data, PINGLESS)) {
5130 if (ptlrpc_pinger_suppress_pings()) {
5131 spin_lock(&exp->exp_obd->obd_dev_lock);
5132 list_del_init(&exp->exp_obd_chain_timed);
5133 spin_unlock(&exp->exp_obd->obd_dev_lock);
5135 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5139 data->ocd_max_easize = mdt->mdt_max_ea_size;
5141 /* NB: Disregard the rule against updating
5142 * exp_connect_data.ocd_connect_flags in this case, since
5143 * tgt_client_new() needs to know if this is client supports
5144 * multiple modify RPCs, and it is safe to expose this flag before
5145 * connection processing completes. */
5146 if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5147 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5148 spin_lock(&exp->exp_lock);
5149 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5150 spin_unlock(&exp->exp_lock);
5156 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5157 struct mdt_thread_info *info,
5158 struct mdt_file_data *mfd)
5160 struct lu_context ses;
5164 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5169 lu_context_enter(&ses);
5171 mdt_ucred(info)->uc_valid = UCRED_OLD;
5172 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5174 lu_context_exit(&ses);
5175 lu_context_fini(&ses);
5181 static int mdt_export_cleanup(struct obd_export *exp)
5183 struct list_head closing_list;
5184 struct mdt_export_data *med = &exp->exp_mdt_data;
5185 struct obd_device *obd = exp->exp_obd;
5186 struct mdt_device *mdt;
5187 struct mdt_thread_info *info;
5189 struct mdt_file_data *mfd, *n;
5193 INIT_LIST_HEAD(&closing_list);
5194 spin_lock(&med->med_open_lock);
5195 while (!list_empty(&med->med_open_head)) {
5196 struct list_head *tmp = med->med_open_head.next;
5197 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5199 /* Remove mfd handle so it can't be found again.
5200 * We are consuming the mfd_list reference here. */
5201 class_handle_unhash(&mfd->mfd_handle);
5202 list_move_tail(&mfd->mfd_list, &closing_list);
5204 spin_unlock(&med->med_open_lock);
5205 mdt = mdt_dev(obd->obd_lu_dev);
5206 LASSERT(mdt != NULL);
5208 rc = lu_env_init(&env, LCT_MD_THREAD);
5212 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5213 LASSERT(info != NULL);
5214 memset(info, 0, sizeof *info);
5215 info->mti_env = &env;
5216 info->mti_mdt = mdt;
5217 info->mti_exp = exp;
5219 if (!list_empty(&closing_list)) {
5220 struct md_attr *ma = &info->mti_attr;
5222 /* Close any open files (which may also cause orphan
5224 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5225 list_del_init(&mfd->mfd_list);
5226 ma->ma_need = ma->ma_valid = 0;
5228 /* This file is being closed due to an eviction, it
5229 * could have been modified and now dirty regarding to
5230 * HSM archive, check this!
5231 * The logic here is to mark a file dirty if there's a
5232 * chance it was dirtied before the client was evicted,
5233 * so that we don't have to wait for a release attempt
5234 * before finding out the file was actually dirty and
5235 * fail the release. Aggressively marking it dirty here
5236 * will cause the policy engine to attempt to
5237 * re-archive it; when rearchiving, we can compare the
5238 * current version to the HSM data_version and make the
5239 * archive request into a noop if it's not actually
5242 if (mfd->mfd_mode & FMODE_WRITE)
5243 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5245 /* Don't unlink orphan on failover umount, LU-184 */
5246 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5247 ma->ma_valid = MA_FLAGS;
5248 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5250 mdt_mfd_close(info, mfd);
5253 info->mti_mdt = NULL;
5254 /* cleanup client slot early */
5255 /* Do not erase record for recoverable client. */
5256 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5257 tgt_client_del(&env, exp);
5263 static inline void mdt_enable_slc(struct mdt_device *mdt)
5265 if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5266 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5269 static inline void mdt_disable_slc(struct mdt_device *mdt)
5271 if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5272 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5275 static int mdt_obd_disconnect(struct obd_export *exp)
5281 class_export_get(exp);
5283 if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5284 !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5285 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5287 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5288 mdt_disable_slc(mdt);
5291 rc = server_disconnect_export(exp);
5293 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5295 rc = mdt_export_cleanup(exp);
5296 nodemap_del_member(exp);
5297 class_export_put(exp);
5301 /* mds_connect copy */
5302 static int mdt_obd_connect(const struct lu_env *env,
5303 struct obd_export **exp, struct obd_device *obd,
5304 struct obd_uuid *cluuid,
5305 struct obd_connect_data *data,
5308 struct obd_export *lexp;
5309 struct lustre_handle conn = { 0 };
5310 struct mdt_device *mdt;
5312 lnet_nid_t *client_nid = localdata;
5315 LASSERT(env != NULL);
5316 LASSERT(data != NULL);
5318 if (!exp || !obd || !cluuid)
5321 mdt = mdt_dev(obd->obd_lu_dev);
5323 if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5324 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5325 atomic_inc(&mdt->mdt_mds_mds_conns);
5326 mdt_enable_slc(mdt);
5330 * first, check whether the stack is ready to handle requests
5331 * XXX: probably not very appropriate method is used now
5332 * at some point we should find a better one
5334 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5335 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5336 !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5337 rc = obd_get_info(env, mdt->mdt_child_exp,
5338 sizeof(KEY_OSP_CONNECTED),
5339 KEY_OSP_CONNECTED, NULL, NULL);
5342 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5345 rc = class_connect(&conn, obd, cluuid);
5349 lexp = class_conn2export(&conn);
5350 LASSERT(lexp != NULL);
5352 rc = nodemap_add_member(*client_nid, lexp);
5353 if (rc != 0 && rc != -EEXIST)
5356 rc = mdt_connect_internal(lexp, mdt, data);
5358 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5361 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5362 rc = tgt_client_new(env, lexp);
5364 mdt_export_stats_init(obd, lexp, localdata);
5368 class_disconnect(lexp);
5369 nodemap_del_member(lexp);
5373 /* Because we do not want this export to be evicted by pinger,
5374 * let's not add this export to the timed chain list. */
5375 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5376 spin_lock(&lexp->exp_obd->obd_dev_lock);
5377 list_del_init(&lexp->exp_obd_chain_timed);
5378 spin_unlock(&lexp->exp_obd->obd_dev_lock);
5385 static int mdt_obd_reconnect(const struct lu_env *env,
5386 struct obd_export *exp, struct obd_device *obd,
5387 struct obd_uuid *cluuid,
5388 struct obd_connect_data *data,
5391 lnet_nid_t *client_nid = localdata;
5395 if (exp == NULL || obd == NULL || cluuid == NULL)
5398 rc = nodemap_add_member(*client_nid, exp);
5399 if (rc != 0 && rc != -EEXIST)
5402 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5404 mdt_export_stats_init(obd, exp, localdata);
5406 nodemap_del_member(exp);
5411 /* FIXME: Can we avoid using these two interfaces? */
5412 static int mdt_init_export(struct obd_export *exp)
5414 struct mdt_export_data *med = &exp->exp_mdt_data;
5418 INIT_LIST_HEAD(&med->med_open_head);
5419 spin_lock_init(&med->med_open_lock);
5420 spin_lock(&exp->exp_lock);
5421 exp->exp_connecting = 1;
5422 spin_unlock(&exp->exp_lock);
5424 /* self-export doesn't need client data and ldlm initialization */
5425 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5426 &exp->exp_client_uuid)))
5429 rc = tgt_client_alloc(exp);
5433 rc = ldlm_init_export(exp);
5440 tgt_client_free(exp);
5442 CERROR("%s: Failed to initialize export: rc = %d\n",
5443 exp->exp_obd->obd_name, rc);
5447 static int mdt_destroy_export(struct obd_export *exp)
5451 target_destroy_export(exp);
5452 /* destroy can be called from failed obd_setup, so
5453 * checking uuid is safer than obd_self_export */
5454 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5455 &exp->exp_client_uuid)))
5458 ldlm_destroy_export(exp);
5459 tgt_client_free(exp);
5461 LASSERT(list_empty(&exp->exp_outstanding_replies));
5462 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5467 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5468 struct linkea_data *ldata)
5472 LASSERT(ldata->ld_buf->lb_buf != NULL);
5474 if (!mdt_object_exists(mdt_obj))
5477 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5478 ldata->ld_buf, XATTR_NAME_LINK);
5479 if (rc == -ERANGE) {
5480 /* Buf was too small, figure out what we need. */
5481 lu_buf_free(ldata->ld_buf);
5482 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5483 ldata->ld_buf, XATTR_NAME_LINK);
5486 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5487 if (ldata->ld_buf->lb_buf == NULL)
5489 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5490 ldata->ld_buf, XATTR_NAME_LINK);
5495 return linkea_init(ldata);
5499 * Given an MDT object, try to look up the full path to the object.
5500 * Part of the MDT layer implementation of lfs fid2path.
5502 * \param[in] info Per-thread common data shared by MDT level handlers.
5503 * \param[in] obj Object to do path lookup of
5504 * \param[in,out] fp User-provided struct to store path information
5505 * \param[in] root_fid Root FID of current path should reach
5507 * \retval 0 Lookup successful, path information stored in fp
5508 * \retval -EAGAIN Lookup failed, usually because object is being moved
5509 * \retval negative errno if there was a problem
5511 static int mdt_path_current(struct mdt_thread_info *info,
5512 struct mdt_object *obj,
5513 struct getinfo_fid2path *fp,
5514 struct lu_fid *root_fid)
5516 struct mdt_device *mdt = info->mti_mdt;
5517 struct mdt_object *mdt_obj;
5518 struct link_ea_header *leh;
5519 struct link_ea_entry *lee;
5520 struct lu_name *tmpname = &info->mti_name;
5521 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5522 struct lu_buf *buf = &info->mti_big_buf;
5525 struct linkea_data ldata = { NULL };
5530 /* temp buffer for path element, the buffer will be finally freed
5531 * in mdt_thread_info_fini */
5532 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5533 if (buf->lb_buf == NULL)
5537 ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5540 *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5542 while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5543 struct lu_buf lmv_buf;
5545 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5546 lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5547 GOTO(out, rc = -ENOENT);
5549 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5550 if (IS_ERR(mdt_obj))
5551 GOTO(out, rc = PTR_ERR(mdt_obj));
5553 if (!mdt_object_exists(mdt_obj)) {
5554 mdt_object_put(info->mti_env, mdt_obj);
5555 GOTO(out, rc = -ENOENT);
5558 if (mdt_object_remote(mdt_obj)) {
5559 mdt_object_put(info->mti_env, mdt_obj);
5560 GOTO(remote_out, rc = -EREMOTE);
5563 rc = mdt_links_read(info, mdt_obj, &ldata);
5565 mdt_object_put(info->mti_env, mdt_obj);
5570 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5571 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5572 /* If set, use link #linkno for path lookup, otherwise use
5573 link #0. Only do this for the final path element. */
5574 if (first && fp->gf_linkno < leh->leh_reccount) {
5576 for (count = 0; count < fp->gf_linkno; count++) {
5577 lee = (struct link_ea_entry *)
5578 ((char *)lee + reclen);
5579 linkea_entry_unpack(lee, &reclen, tmpname,
5582 if (fp->gf_linkno < leh->leh_reccount - 1)
5583 /* indicate to user there are more links */
5587 lmv_buf.lb_buf = info->mti_xattr_buf;
5588 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5589 /* Check if it is slave stripes */
5590 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5591 &lmv_buf, XATTR_NAME_LMV);
5592 mdt_object_put(info->mti_env, mdt_obj);
5594 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5596 /* For slave stripes, get its master */
5597 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5598 fp->gf_fid = *tmpfid;
5601 } else if (rc < 0 && rc != -ENODATA) {
5607 /* Pack the name in the end of the buffer */
5608 ptr -= tmpname->ln_namelen;
5609 if (ptr - 1 <= fp->gf_u.gf_path)
5610 GOTO(out, rc = -EOVERFLOW);
5611 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5614 /* keep the last resolved fid to the client, so the
5615 * client will build the left path on another MDT for
5617 fp->gf_fid = *tmpfid;
5623 ptr++; /* skip leading / */
5624 memmove(fp->gf_u.gf_path, ptr,
5625 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5632 * Given an MDT object, use mdt_path_current to get the path.
5633 * Essentially a wrapper to retry mdt_path_current a set number of times
5634 * if -EAGAIN is returned (usually because an object is being moved).
5636 * Part of the MDT layer implementation of lfs fid2path.
5638 * \param[in] info Per-thread common data shared by mdt level handlers.
5639 * \param[in] obj Object to do path lookup of
5640 * \param[in,out] fp User-provided struct for arguments and to store path
5643 * \retval 0 Lookup successful, path information stored in fp
5644 * \retval negative errno if there was a problem
5646 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5647 struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5649 struct mdt_device *mdt = info->mti_mdt;
5654 if (fp->gf_pathlen < 3)
5657 if (root_fid == NULL)
5658 root_fid = &mdt->mdt_md_root_fid;
5660 if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5661 fp->gf_u.gf_path[0] = '\0';
5665 /* Retry multiple times in case file is being moved */
5666 while (tries-- && rc == -EAGAIN)
5667 rc = mdt_path_current(info, obj, fp, root_fid);
5673 * Get the full path of the provided FID, as of changelog record recno.
5675 * This checks sanity and looks up object for user provided FID
5676 * before calling the actual path lookup code.
5678 * Part of the MDT layer implementation of lfs fid2path.
5680 * \param[in] info Per-thread common data shared by mdt level handlers.
5681 * \param[in,out] fp User-provided struct for arguments and to store path
5684 * \retval 0 Lookup successful, path information and recno stored in fp
5685 * \retval -ENOENT, object does not exist
5686 * \retval negative errno if there was a problem
5688 static int mdt_fid2path(struct mdt_thread_info *info,
5689 struct lu_fid *root_fid,
5690 struct getinfo_fid2path *fp)
5692 struct mdt_device *mdt = info->mti_mdt;
5693 struct mdt_object *obj;
5697 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5698 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5700 if (!fid_is_sane(&fp->gf_fid))
5703 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5704 CWARN("%s: "DFID" is invalid, sequence should be "
5705 ">= "LPX64"\n", mdt_obd_name(mdt),
5706 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5710 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5713 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5714 PFID(&fp->gf_fid), rc);
5718 if (mdt_object_remote(obj))
5720 else if (!mdt_object_exists(obj))
5726 mdt_object_put(info->mti_env, obj);
5727 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5728 PFID(&fp->gf_fid), rc);
5732 rc = mdt_path(info, obj, fp, root_fid);
5734 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5735 PFID(&fp->gf_fid), fp->gf_u.gf_path,
5736 fp->gf_recno, fp->gf_linkno);
5738 mdt_object_put(info->mti_env, obj);
5743 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5744 void *val, int vallen)
5746 struct getinfo_fid2path *fpout, *fpin;
5747 struct lu_fid *root_fid = NULL;
5750 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5753 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5754 lustre_swab_fid2path(fpin);
5756 memcpy(fpout, fpin, sizeof(*fpin));
5757 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5760 if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5761 sizeof(struct lu_fid)) {
5762 /* client sent its root FID, which is normally fileset FID */
5763 root_fid = fpin->gf_u.gf_root_fid;
5764 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5765 lustre_swab_lu_fid(root_fid);
5767 if (root_fid != NULL && !fid_is_sane(root_fid))
5771 rc = mdt_fid2path(info, root_fid, fpout);
5775 int mdt_get_info(struct tgt_session_info *tsi)
5785 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5787 CDEBUG(D_IOCTL, "No GETINFO key\n");
5788 RETURN(err_serious(-EFAULT));
5790 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5793 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5794 if (vallen == NULL) {
5795 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5796 tgt_name(tsi->tsi_tgt));
5797 RETURN(err_serious(-EFAULT));
5800 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5802 rc = req_capsule_server_pack(tsi->tsi_pill);
5804 RETURN(err_serious(rc));
5806 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5807 if (valout == NULL) {
5808 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5809 tgt_name(tsi->tsi_tgt));
5810 RETURN(err_serious(-EFAULT));
5813 if (KEY_IS(KEY_FID2PATH)) {
5814 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5816 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5817 mdt_thread_info_fini(info);
5824 /* Pass the ioc down */
5825 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5826 unsigned int cmd, int len, void *data)
5828 struct lu_context ioctl_session;
5829 struct md_device *next = mdt->mdt_child;
5833 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5836 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5837 lu_context_enter(&ioctl_session);
5838 env->le_ses = &ioctl_session;
5840 LASSERT(next->md_ops->mdo_iocontrol);
5841 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5843 lu_context_exit(&ioctl_session);
5844 lu_context_fini(&ioctl_session);
5848 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5850 struct obd_ioctl_data *data = karg;
5853 struct mdt_object *obj;
5854 struct mdt_lock_handle *lh;
5858 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5859 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5862 fid = (struct lu_fid *)data->ioc_inlbuf1;
5864 if (!fid_is_sane(fid))
5867 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5869 lh = &mti->mti_lh[MDT_LH_PARENT];
5870 mdt_lock_reg_init(lh, LCK_CR);
5872 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5874 RETURN(PTR_ERR(obj));
5876 if (mdt_object_remote(obj)) {
5879 * before calling version get the correct MDS should be
5880 * fid, this is error to find remote object here
5882 CERROR("nonlocal object "DFID"\n", PFID(fid));
5883 } else if (!mdt_object_exists(obj)) {
5884 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5887 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5888 *(__u64 *)data->ioc_inlbuf2 = version;
5891 mdt_object_unlock_put(mti, obj, lh, 1);
5895 /* ioctls on obd dev */
5896 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5897 void *karg, void __user *uarg)
5900 struct obd_device *obd = exp->exp_obd;
5901 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5902 struct dt_device *dt = mdt->mdt_bottom;
5906 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5907 rc = lu_env_init(&env, LCT_MD_THREAD);
5913 rc = mdt_device_sync(&env, mdt);
5915 case OBD_IOC_SET_READONLY:
5916 rc = dt_sync(&env, dt);
5918 rc = dt_ro(&env, dt);
5920 case OBD_IOC_ABORT_RECOVERY:
5921 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5922 obd->obd_abort_recovery = 1;
5923 target_stop_recovery_thread(obd);
5926 case OBD_IOC_CHANGELOG_REG:
5927 case OBD_IOC_CHANGELOG_DEREG:
5928 case OBD_IOC_CHANGELOG_CLEAR:
5929 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5931 case OBD_IOC_START_LFSCK: {
5932 struct md_device *next = mdt->mdt_child;
5933 struct obd_ioctl_data *data = karg;
5934 struct lfsck_start_param lsp;
5936 if (unlikely(data == NULL)) {
5941 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5942 lsp.lsp_index_valid = 0;
5943 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5946 case OBD_IOC_STOP_LFSCK: {
5947 struct md_device *next = mdt->mdt_child;
5948 struct obd_ioctl_data *data = karg;
5949 struct lfsck_stop stop;
5951 stop.ls_status = LS_STOPPED;
5952 /* Old lfsck utils may pass NULL @stop. */
5953 if (data->ioc_inlbuf1 == NULL)
5957 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5959 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5962 case OBD_IOC_QUERY_LFSCK: {
5963 struct md_device *next = mdt->mdt_child;
5964 struct obd_ioctl_data *data = karg;
5966 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
5970 case OBD_IOC_GET_OBJ_VERSION: {
5971 struct mdt_thread_info *mti;
5972 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5973 memset(mti, 0, sizeof *mti);
5974 mti->mti_env = &env;
5978 rc = mdt_ioc_version_get(mti, karg);
5981 case OBD_IOC_CATLOGLIST: {
5982 struct mdt_thread_info *mti;
5984 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5985 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5986 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5987 &mti->mti_tmp_fid1);
5992 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5993 mdt_obd_name(mdt), cmd, rc);
6000 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6002 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6006 if (!mdt->mdt_skip_lfsck) {
6007 struct lfsck_start_param lsp;
6009 lsp.lsp_start = NULL;
6010 lsp.lsp_index_valid = 0;
6011 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6012 OBD_IOC_START_LFSCK,
6014 if (rc != 0 && rc != -EALREADY)
6015 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6016 mdt_obd_name(mdt), rc);
6019 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6023 static int mdt_obd_postrecov(struct obd_device *obd)
6028 rc = lu_env_init(&env, LCT_MD_THREAD);
6031 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6036 static struct obd_ops mdt_obd_device_ops = {
6037 .o_owner = THIS_MODULE,
6038 .o_set_info_async = mdt_obd_set_info_async,
6039 .o_connect = mdt_obd_connect,
6040 .o_reconnect = mdt_obd_reconnect,
6041 .o_disconnect = mdt_obd_disconnect,
6042 .o_init_export = mdt_init_export,
6043 .o_destroy_export = mdt_destroy_export,
6044 .o_iocontrol = mdt_iocontrol,
6045 .o_postrecov = mdt_obd_postrecov,
6048 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6049 struct lu_device *d)
6051 struct mdt_device *m = mdt_dev(d);
6058 static struct lu_device *mdt_device_free(const struct lu_env *env,
6059 struct lu_device *d)
6061 struct mdt_device *m = mdt_dev(d);
6064 lu_device_fini(&m->mdt_lu_dev);
6070 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6071 struct lu_device_type *t,
6072 struct lustre_cfg *cfg)
6074 struct lu_device *l;
6075 struct mdt_device *m;
6082 rc = mdt_init0(env, m, t, cfg);
6084 mdt_device_free(env, l);
6089 l = ERR_PTR(-ENOMEM);
6093 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6094 LU_KEY_INIT(mdt, struct mdt_thread_info);
6096 static void mdt_key_fini(const struct lu_context *ctx,
6097 struct lu_context_key *key, void* data)
6099 struct mdt_thread_info *info = data;
6101 if (info->mti_big_lmm) {
6102 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6103 info->mti_big_lmm = NULL;
6104 info->mti_big_lmmsize = 0;
6109 /* context key: mdt_thread_key */
6110 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6112 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6114 return lu_ucred(info->mti_env);
6117 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6119 return lu_ucred_check(info->mti_env);
6123 * Enable/disable COS (Commit On Sharing).
6125 * Set/Clear the COS flag in mdt options.
6127 * \param mdt mdt device
6128 * \param val 0 disables COS, other values enable COS
6130 void mdt_enable_cos(struct mdt_device *mdt, int val)
6135 mdt->mdt_opts.mo_cos = !!val;
6136 rc = lu_env_init(&env, LCT_LOCAL);
6137 if (unlikely(rc != 0)) {
6138 CWARN("%s: lu_env initialization failed, cannot "
6139 "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6142 mdt_device_sync(&env, mdt);
6147 * Check COS (Commit On Sharing) status.
6149 * Return COS flag status.
6151 * \param mdt mdt device
6153 int mdt_cos_is_enabled(struct mdt_device *mdt)
6155 return mdt->mdt_opts.mo_cos != 0;
6158 static struct lu_device_type_operations mdt_device_type_ops = {
6159 .ldto_device_alloc = mdt_device_alloc,
6160 .ldto_device_free = mdt_device_free,
6161 .ldto_device_fini = mdt_device_fini
6164 static struct lu_device_type mdt_device_type = {
6165 .ldt_tags = LU_DEVICE_MD,
6166 .ldt_name = LUSTRE_MDT_NAME,
6167 .ldt_ops = &mdt_device_type_ops,
6168 .ldt_ctx_tags = LCT_MD_THREAD
6171 static int __init mdt_init(void)
6175 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6176 FID_NOBRACE_LEN + 1);
6177 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6179 rc = lu_kmem_init(mdt_caches);
6183 rc = mds_mod_init();
6187 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6188 LUSTRE_MDT_NAME, &mdt_device_type);
6193 lu_kmem_fini(mdt_caches);
6200 static void __exit mdt_exit(void)
6202 class_unregister_type(LUSTRE_MDT_NAME);
6204 lu_kmem_fini(mdt_caches);
6207 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6208 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6209 MODULE_VERSION(LUSTRE_VERSION_STRING);
6210 MODULE_LICENSE("GPL");
6212 module_init(mdt_init);
6213 module_exit(mdt_exit);