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, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/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 * struct OBD_{ALLOC,FREE}*()
55 #include <obd_support.h>
56 #include <lustre_ioctl.h>
57 /* struct ptlrpc_request */
58 #include <lustre_net.h>
59 /* struct obd_export */
60 #include <lustre_export.h>
61 /* struct obd_device */
64 #include <dt_object.h>
65 #include <lustre_mds.h>
66 #include <lustre_log.h>
67 #include "mdt_internal.h"
68 #include <lustre_acl.h>
69 #include <lustre_param.h>
70 #include <lustre_quota.h>
71 #include <lustre_lfsck.h>
73 mdl_mode_t mdt_mdl_lock_modes[] = {
74 [LCK_MINMODE] = MDL_MINMODE,
81 [LCK_GROUP] = MDL_GROUP
84 ldlm_mode_t mdt_dlm_lock_modes[] = {
85 [MDL_MINMODE] = LCK_MINMODE,
92 [MDL_GROUP] = LCK_GROUP
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
98 static const struct lu_object_operations mdt_obj_ops;
100 /* Slab for MDT object allocation */
101 static struct kmem_cache *mdt_object_kmem;
103 /* For HSM restore handles */
104 struct kmem_cache *mdt_hsm_cdt_kmem;
106 /* For HSM request handles */
107 struct kmem_cache *mdt_hsm_car_kmem;
109 static struct lu_kmem_descr mdt_caches[] = {
111 .ckd_cache = &mdt_object_kmem,
112 .ckd_name = "mdt_obj",
113 .ckd_size = sizeof(struct mdt_object)
116 .ckd_cache = &mdt_hsm_cdt_kmem,
117 .ckd_name = "mdt_cdt_restore_handle",
118 .ckd_size = sizeof(struct cdt_restore_handle)
121 .ckd_cache = &mdt_hsm_car_kmem,
122 .ckd_name = "mdt_cdt_agent_req",
123 .ckd_size = sizeof(struct cdt_agent_req)
130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
134 return (rep->lock_policy_res1 & flag);
137 void mdt_clear_disposition(struct mdt_thread_info *info,
138 struct ldlm_reply *rep, int flag)
141 info->mti_opdata &= ~flag;
142 tgt_opdata_clear(info->mti_env, flag);
145 rep->lock_policy_res1 &= ~flag;
148 void mdt_set_disposition(struct mdt_thread_info *info,
149 struct ldlm_reply *rep, int flag)
152 info->mti_opdata |= flag;
153 tgt_opdata_set(info->mti_env, flag);
156 rep->lock_policy_res1 |= flag;
159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
161 lh->mlh_pdo_hash = 0;
162 lh->mlh_reg_mode = lm;
163 lh->mlh_rreg_mode = lm;
164 lh->mlh_type = MDT_REG_LOCK;
167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
168 const struct lu_name *lname)
170 lh->mlh_reg_mode = lock_mode;
171 lh->mlh_rreg_mode = lock_mode;
172 lh->mlh_type = MDT_PDO_LOCK;
174 if (lu_name_is_valid(lname)) {
175 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
177 /* XXX Workaround for LU-2856
179 * Zero is a valid return value of full_name_hash, but
180 * several users of mlh_pdo_hash assume a non-zero
181 * hash value. We therefore map zero onto an
182 * arbitrary, but consistent value (1) to avoid
183 * problems further down the road. */
184 if (unlikely(lh->mlh_pdo_hash == 0))
185 lh->mlh_pdo_hash = 1;
187 lh->mlh_pdo_hash = 0;
191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
192 struct mdt_lock_handle *lh)
198 * Any dir access needs couple of locks:
200 * 1) on part of dir we gonna take lookup/modify;
202 * 2) on whole dir to protect it from concurrent splitting and/or to
203 * flush client's cache for readdir().
205 * so, for a given mode and object this routine decides what lock mode
206 * to use for lock #2:
208 * 1) if caller's gonna lookup in dir then we need to protect dir from
209 * being splitted only - LCK_CR
211 * 2) if caller's gonna modify dir then we need to protect dir from
212 * being splitted and to flush cache - LCK_CW
214 * 3) if caller's gonna modify dir and that dir seems ready for
215 * splitting then we need to protect it from any type of access
216 * (lookup/modify/split) - LCK_EX --bzzz
219 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
220 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
223 * Ask underlaying level its opinion about preferable PDO lock mode
224 * having access type passed as regular lock mode:
226 * - MDL_MINMODE means that lower layer does not want to specify lock
229 * - MDL_NL means that no PDO lock should be taken. This is used in some
230 * cases. Say, for non-splittable directories no need to use PDO locks
233 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
234 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
236 if (mode != MDL_MINMODE) {
237 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
240 * Lower layer does not want to specify locking mode. We do it
241 * our selves. No special protection is needed, just flush
242 * client's cache on modification and allow concurrent
245 switch (lh->mlh_reg_mode) {
247 lh->mlh_pdo_mode = LCK_EX;
250 lh->mlh_pdo_mode = LCK_CR;
253 lh->mlh_pdo_mode = LCK_CW;
256 CERROR("Not expected lock type (0x%x)\n",
257 (int)lh->mlh_reg_mode);
262 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
266 static int mdt_getstatus(struct tgt_session_info *tsi)
268 struct mdt_thread_info *info = tsi2mdt_info(tsi);
269 struct mdt_device *mdt = info->mti_mdt;
270 struct mdt_body *repbody;
275 rc = mdt_check_ucred(info);
277 GOTO(out, rc = err_serious(rc));
279 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
280 GOTO(out, rc = err_serious(-ENOMEM));
282 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
283 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
284 repbody->mbo_valid |= OBD_MD_FLID;
286 if (tsi->tsi_tgt->lut_mds_capa &&
287 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
288 struct mdt_object *root;
289 struct lustre_capa *capa;
291 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
293 GOTO(out, rc = PTR_ERR(root));
295 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
297 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
298 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
300 mdt_object_put(info->mti_env, root);
302 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
306 mdt_thread_info_fini(info);
310 static int mdt_statfs(struct tgt_session_info *tsi)
312 struct ptlrpc_request *req = tgt_ses_req(tsi);
313 struct mdt_thread_info *info = tsi2mdt_info(tsi);
314 struct md_device *next = info->mti_mdt->mdt_child;
315 struct ptlrpc_service_part *svcpt;
316 struct obd_statfs *osfs;
321 svcpt = req->rq_rqbd->rqbd_svcpt;
323 /* This will trigger a watchdog timeout */
324 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
325 (MDT_SERVICE_WATCHDOG_FACTOR *
326 at_get(&svcpt->scp_at_estimate)) + 1);
328 rc = mdt_check_ucred(info);
330 GOTO(out, rc = err_serious(rc));
332 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
333 GOTO(out, rc = err_serious(-ENOMEM));
335 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
337 GOTO(out, rc = -EPROTO);
339 /** statfs information are cached in the mdt_device */
340 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
341 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
342 /** statfs data is too old, get up-to-date one */
343 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
346 spin_lock(&info->mti_mdt->mdt_osfs_lock);
347 info->mti_mdt->mdt_osfs = *osfs;
348 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
349 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
351 /** use cached statfs data */
352 spin_lock(&info->mti_mdt->mdt_osfs_lock);
353 *osfs = info->mti_mdt->mdt_osfs;
354 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
358 mdt_counter_incr(req, LPROC_MDT_STATFS);
360 mdt_thread_info_fini(info);
365 * Pack SOM attributes into the reply.
366 * Call under a DLM UPDATE lock.
368 static void mdt_pack_size2body(struct mdt_thread_info *info,
369 struct mdt_object *mo)
372 struct md_attr *ma = &info->mti_attr;
374 LASSERT(ma->ma_attr.la_valid & LA_MODE);
375 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
377 /* Check if Size-on-MDS is supported, if this is a regular file,
378 * if SOM is enabled on the object and if SOM cache exists and valid.
379 * Otherwise do not pack Size-on-MDS attributes to the reply. */
380 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
381 !S_ISREG(ma->ma_attr.la_mode) ||
382 !mdt_object_is_som_enabled(mo) ||
383 !(ma->ma_valid & MA_SOM))
386 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
387 b->mbo_size = ma->ma_som->msd_size;
388 b->mbo_blocks = ma->ma_som->msd_blocks;
391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
392 const struct lu_attr *attr, const struct lu_fid *fid)
394 struct md_attr *ma = &info->mti_attr;
396 LASSERT(ma->ma_valid & MA_INODE);
398 b->mbo_atime = attr->la_atime;
399 b->mbo_mtime = attr->la_mtime;
400 b->mbo_ctime = attr->la_ctime;
401 b->mbo_mode = attr->la_mode;
402 b->mbo_size = attr->la_size;
403 b->mbo_blocks = attr->la_blocks;
404 b->mbo_uid = attr->la_uid;
405 b->mbo_gid = attr->la_gid;
406 b->mbo_flags = attr->la_flags;
407 b->mbo_nlink = attr->la_nlink;
408 b->mbo_rdev = attr->la_rdev;
410 /* XXX should pack the reply body according to lu_valid */
411 b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
412 OBD_MD_FLGID | OBD_MD_FLTYPE |
413 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
414 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
416 if (!S_ISREG(attr->la_mode)) {
417 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
418 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
419 /* means no objects are allocated on osts. */
420 LASSERT(!(ma->ma_valid & MA_LOV));
421 /* just ignore blocks occupied by extend attributes on MDS */
423 /* if no object is allocated on osts, the size on mds is valid.
425 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
426 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
427 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
428 /* A released file stores its size on MDS. */
429 /* But return 1 block for released file, unless tools like tar
430 * will consider it fully sparse. (LU-3864)
432 if (unlikely(b->mbo_size == 0))
436 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
441 b->mbo_valid |= OBD_MD_FLID;
442 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
443 PFID(fid), b->mbo_nlink,
444 b->mbo_mode, b->mbo_size);
448 mdt_body_reverse_idmap(info, b);
450 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
451 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
452 PFID(fid), (unsigned long long)b->mbo_size);
455 static inline int mdt_body_has_lov(const struct lu_attr *la,
456 const struct mdt_body *body)
458 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
459 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
462 void mdt_client_compatibility(struct mdt_thread_info *info)
464 struct mdt_body *body;
465 struct ptlrpc_request *req = mdt_info_req(info);
466 struct obd_export *exp = req->rq_export;
467 struct md_attr *ma = &info->mti_attr;
468 struct lu_attr *la = &ma->ma_attr;
471 if (exp_connect_layout(exp))
472 /* the client can deal with 16-bit lmm_stripe_count */
475 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
477 if (!mdt_body_has_lov(la, body))
480 /* now we have a reply with a lov for a client not compatible with the
481 * layout lock so we have to clean the layout generation number */
482 if (S_ISREG(la->la_mode))
483 ma->ma_lmm->lmm_layout_gen = 0;
487 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
489 const struct lu_env *env = info->mti_env;
492 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
501 /* Is it a directory? Let's check for the LMV as well */
502 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
503 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
507 rc2 = mo_xattr_get(env, mdt_object_child(o),
509 XATTR_NAME_DEFAULT_LMV);
511 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
519 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
522 const struct lu_env *env = info->mti_env;
526 LASSERT(info->mti_big_lmm_used == 0);
527 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
531 /* big_lmm may need to be grown */
532 if (info->mti_big_lmmsize < rc) {
533 int size = size_roundup_power2(rc);
535 if (info->mti_big_lmmsize > 0) {
536 /* free old buffer */
537 LASSERT(info->mti_big_lmm);
538 OBD_FREE_LARGE(info->mti_big_lmm,
539 info->mti_big_lmmsize);
540 info->mti_big_lmm = NULL;
541 info->mti_big_lmmsize = 0;
544 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
545 if (info->mti_big_lmm == NULL)
547 info->mti_big_lmmsize = size;
549 LASSERT(info->mti_big_lmmsize >= rc);
551 info->mti_buf.lb_buf = info->mti_big_lmm;
552 info->mti_buf.lb_len = info->mti_big_lmmsize;
553 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
558 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
559 struct md_attr *ma, const char *name)
561 struct md_object *next = mdt_object_child(o);
562 struct lu_buf *buf = &info->mti_buf;
565 if (strcmp(name, XATTR_NAME_LOV) == 0) {
566 buf->lb_buf = ma->ma_lmm;
567 buf->lb_len = ma->ma_lmm_size;
568 LASSERT(!(ma->ma_valid & MA_LOV));
569 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
570 buf->lb_buf = ma->ma_lmv;
571 buf->lb_len = ma->ma_lmv_size;
572 LASSERT(!(ma->ma_valid & MA_LMV));
573 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
574 buf->lb_buf = ma->ma_lmv;
575 buf->lb_len = ma->ma_lmv_size;
576 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
581 rc = mo_xattr_get(info->mti_env, next, buf, name);
585 if (strcmp(name, XATTR_NAME_LOV) == 0) {
586 if (info->mti_big_lmm_used)
587 ma->ma_lmm = info->mti_big_lmm;
589 /* NOT return LOV EA with hole to old client. */
590 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
591 LOV_PATTERN_F_HOLE) &&
592 !(exp_connect_flags(info->mti_exp) &
593 OBD_CONNECT_LFSCK)) {
596 ma->ma_lmm_size = rc;
597 ma->ma_valid |= MA_LOV;
599 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
600 if (info->mti_big_lmm_used)
601 ma->ma_lmv = info->mti_big_lmm;
603 ma->ma_lmv_size = rc;
604 ma->ma_valid |= MA_LMV;
605 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
606 ma->ma_lmv_size = rc;
607 ma->ma_valid |= MA_LMV_DEF;
610 /* Update mdt_max_mdsize so all clients will be aware that */
611 if (info->mti_mdt->mdt_max_mdsize < rc)
612 info->mti_mdt->mdt_max_mdsize = rc;
615 } else if (rc == -ENODATA) {
618 } else if (rc == -ERANGE) {
619 /* Default LMV has fixed size, so it must be able to fit
620 * in the original buffer */
621 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
623 rc = mdt_big_xattr_get(info, o, name);
625 info->mti_big_lmm_used = 1;
633 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
634 struct mdt_object *o, struct lu_fid *pfid)
636 struct lu_buf *buf = &info->mti_buf;
637 struct link_ea_header *leh;
638 struct link_ea_entry *lee;
642 buf->lb_buf = info->mti_big_lmm;
643 buf->lb_len = info->mti_big_lmmsize;
644 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
645 buf, XATTR_NAME_LINK);
646 /* ignore errors, MA_PFID won't be set and it is
647 * up to the caller to treat this as an error */
648 if (rc == -ERANGE || buf->lb_len == 0) {
649 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
650 buf->lb_buf = info->mti_big_lmm;
651 buf->lb_len = info->mti_big_lmmsize;
656 if (rc < sizeof(*leh)) {
657 CERROR("short LinkEA on "DFID": rc = %d\n",
658 PFID(mdt_object_fid(o)), rc);
662 leh = (struct link_ea_header *) buf->lb_buf;
663 lee = (struct link_ea_entry *)(leh + 1);
664 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
665 leh->leh_magic = LINK_EA_MAGIC;
666 leh->leh_reccount = __swab32(leh->leh_reccount);
667 leh->leh_len = __swab64(leh->leh_len);
669 if (leh->leh_magic != LINK_EA_MAGIC)
671 if (leh->leh_reccount == 0)
674 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
675 fid_be_to_cpu(pfid, pfid);
680 int mdt_attr_get_complex(struct mdt_thread_info *info,
681 struct mdt_object *o, struct md_attr *ma)
683 const struct lu_env *env = info->mti_env;
684 struct md_object *next = mdt_object_child(o);
685 struct lu_buf *buf = &info->mti_buf;
686 u32 mode = lu_object_attr(&next->mo_lu);
687 int need = ma->ma_need;
693 if (need & MA_INODE) {
694 ma->ma_need = MA_INODE;
695 rc = mo_attr_get(env, next, ma);
698 ma->ma_valid |= MA_INODE;
701 if (need & MA_PFID) {
702 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
704 ma->ma_valid |= MA_PFID;
705 /* ignore this error, parent fid is not mandatory */
709 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
710 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
715 if (need & MA_LMV && S_ISDIR(mode)) {
716 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
721 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
722 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
727 if (need & MA_SOM && S_ISREG(mode)) {
728 buf->lb_buf = info->mti_xattr_buf;
729 buf->lb_len = sizeof(info->mti_xattr_buf);
730 CLASSERT(sizeof(struct som_attrs) <=
731 sizeof(info->mti_xattr_buf));
732 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
733 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
735 ma->ma_valid |= MA_SOM;
736 else if (rc2 < 0 && rc2 != -ENODATA)
740 if (need & MA_HSM && S_ISREG(mode)) {
741 buf->lb_buf = info->mti_xattr_buf;
742 buf->lb_len = sizeof(info->mti_xattr_buf);
743 CLASSERT(sizeof(struct hsm_attrs) <=
744 sizeof(info->mti_xattr_buf));
745 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
746 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
748 ma->ma_valid |= MA_HSM;
749 else if (rc2 < 0 && rc2 != -ENODATA)
753 #ifdef CONFIG_FS_POSIX_ACL
754 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
755 buf->lb_buf = ma->ma_acl;
756 buf->lb_len = ma->ma_acl_size;
757 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
759 ma->ma_acl_size = rc2;
760 ma->ma_valid |= MA_ACL_DEF;
761 } else if (rc2 == -ENODATA) {
770 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
771 rc, ma->ma_valid, ma->ma_lmm);
775 static int mdt_getattr_internal(struct mdt_thread_info *info,
776 struct mdt_object *o, int ma_need)
778 struct md_object *next = mdt_object_child(o);
779 const struct mdt_body *reqbody = info->mti_body;
780 struct ptlrpc_request *req = mdt_info_req(info);
781 struct md_attr *ma = &info->mti_attr;
782 struct lu_attr *la = &ma->ma_attr;
783 struct req_capsule *pill = info->mti_pill;
784 const struct lu_env *env = info->mti_env;
785 struct mdt_body *repbody;
786 struct lu_buf *buffer = &info->mti_buf;
791 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
792 RETURN(err_serious(-ENOMEM));
794 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
798 if (mdt_object_remote(o)) {
799 /* This object is located on remote node.*/
800 /* Return -ENOTSUPP for old client */
801 if (!mdt_is_dne_client(req->rq_export))
802 GOTO(out, rc = -ENOTSUPP);
804 repbody->mbo_fid1 = *mdt_object_fid(o);
805 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
809 if (reqbody->mbo_eadatasize > 0) {
810 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
811 if (buffer->lb_buf == NULL)
812 GOTO(out, rc = -EPROTO);
813 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
816 buffer->lb_buf = NULL;
818 ma_need &= ~(MA_LOV | MA_LMV);
819 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
820 mdt_obd_name(info->mti_mdt),
821 req->rq_export->exp_client_uuid.uuid);
824 /* If it is dir object and client require MEA, then we got MEA */
825 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
826 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
827 /* Assumption: MDT_MD size is enough for lmv size. */
828 ma->ma_lmv = buffer->lb_buf;
829 ma->ma_lmv_size = buffer->lb_len;
830 ma->ma_need = MA_INODE;
831 if (ma->ma_lmv_size > 0) {
832 if (reqbody->mbo_valid & OBD_MD_MEA)
833 ma->ma_need |= MA_LMV;
834 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
835 ma->ma_need |= MA_LMV_DEF;
838 ma->ma_lmm = buffer->lb_buf;
839 ma->ma_lmm_size = buffer->lb_len;
840 ma->ma_need = MA_INODE | MA_HSM;
841 if (ma->ma_lmm_size > 0)
842 ma->ma_need |= MA_LOV;
845 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
846 reqbody->mbo_valid & OBD_MD_FLDIREA &&
847 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
848 /* get default stripe info for this dir. */
849 ma->ma_need |= MA_LOV_DEF;
851 ma->ma_need |= ma_need;
852 if (ma->ma_need & MA_SOM)
853 ma->ma_som = &info->mti_u.som.data;
855 rc = mdt_attr_get_complex(info, o, ma);
857 CERROR("%s: getattr error for "DFID": rc = %d\n",
858 mdt_obd_name(info->mti_mdt),
859 PFID(mdt_object_fid(o)), rc);
863 /* if file is released, check if a restore is running */
864 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
865 mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
866 repbody->mbo_t_state = MS_RESTORE;
867 repbody->mbo_valid |= OBD_MD_TSTATE;
870 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
872 /* the Lustre protocol supposes to return default striping
873 * on the user-visible root if explicitly requested */
874 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
875 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
876 struct lu_fid rootfid;
877 struct mdt_object *root;
878 struct mdt_device *mdt = info->mti_mdt;
880 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
883 root = mdt_object_find(env, mdt, &rootfid);
885 RETURN(PTR_ERR(root));
886 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
887 mdt_object_put(info->mti_env, root);
889 CERROR("%s: getattr error for "DFID": rc = %d\n",
890 mdt_obd_name(info->mti_mdt),
891 PFID(mdt_object_fid(o)), rc);
896 if (likely(ma->ma_valid & MA_INODE))
897 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
901 if (mdt_body_has_lov(la, reqbody)) {
902 if (ma->ma_valid & MA_LOV) {
903 LASSERT(ma->ma_lmm_size);
904 repbody->mbo_eadatasize = ma->ma_lmm_size;
905 if (S_ISDIR(la->la_mode))
906 repbody->mbo_valid |= OBD_MD_FLDIREA;
908 repbody->mbo_valid |= OBD_MD_FLEASIZE;
909 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
911 if (ma->ma_valid & MA_LMV) {
912 /* Return -ENOTSUPP for old client */
913 if (!mdt_is_striped_client(req->rq_export))
916 LASSERT(S_ISDIR(la->la_mode));
917 mdt_dump_lmv(D_INFO, ma->ma_lmv);
918 repbody->mbo_eadatasize = ma->ma_lmv_size;
919 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
921 if (ma->ma_valid & MA_LMV_DEF) {
922 /* Return -ENOTSUPP for old client */
923 if (!mdt_is_striped_client(req->rq_export))
925 LASSERT(S_ISDIR(la->la_mode));
926 repbody->mbo_eadatasize = ma->ma_lmv_size;
927 repbody->mbo_valid |= (OBD_MD_FLDIREA |
930 } else if (S_ISLNK(la->la_mode) &&
931 reqbody->mbo_valid & OBD_MD_LINKNAME) {
932 buffer->lb_buf = ma->ma_lmm;
933 /* eadatasize from client includes NULL-terminator, so
934 * there is no need to read it */
935 buffer->lb_len = reqbody->mbo_eadatasize - 1;
936 rc = mo_readlink(env, next, buffer);
937 if (unlikely(rc <= 0)) {
938 CERROR("%s: readlink failed for "DFID": rc = %d\n",
939 mdt_obd_name(info->mti_mdt),
940 PFID(mdt_object_fid(o)), rc);
943 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
945 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
947 repbody->mbo_valid |= OBD_MD_LINKNAME;
948 /* we need to report back size with NULL-terminator
949 * because client expects that */
950 repbody->mbo_eadatasize = rc + 1;
951 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
952 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
953 "on "DFID ", expected %d\n",
954 mdt_obd_name(info->mti_mdt),
955 rc, PFID(mdt_object_fid(o)),
956 reqbody->mbo_eadatasize - 1);
958 ((char *)ma->ma_lmm)[rc] = 0;
960 /* If the total CDEBUG() size is larger than a page, it
961 * will print a warning to the console, avoid this by
962 * printing just the last part of the symlink. */
963 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
964 print_limit < rc ? "..." : "", print_limit,
965 (char *)ma->ma_lmm + rc - print_limit, rc);
970 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
971 repbody->mbo_max_cookiesize = 0;
972 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
973 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
974 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
975 "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
976 repbody->mbo_max_cookiesize);
979 if (exp_connect_rmtclient(info->mti_exp) &&
980 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
981 void *buf = req_capsule_server_get(pill, &RMF_ACL);
983 /* mdt_getattr_lock only */
984 rc = mdt_pack_remote_perm(info, o, buf);
986 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
987 repbody->mbo_aclsize = 0;
990 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
991 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
994 #ifdef CONFIG_FS_POSIX_ACL
995 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
996 (reqbody->mbo_valid & OBD_MD_FLACL)) {
997 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
998 buffer->lb_len = req_capsule_get_size(pill,
999 &RMF_ACL, RCL_SERVER);
1000 if (buffer->lb_len > 0) {
1001 rc = mo_xattr_get(env, next, buffer,
1002 XATTR_NAME_ACL_ACCESS);
1004 if (rc == -ENODATA) {
1005 repbody->mbo_aclsize = 0;
1006 repbody->mbo_valid |= OBD_MD_FLACL;
1008 } else if (rc == -EOPNOTSUPP) {
1011 CERROR("%s: unable to read "DFID
1013 mdt_obd_name(info->mti_mdt),
1014 PFID(mdt_object_fid(o)), rc);
1017 repbody->mbo_aclsize = rc;
1018 repbody->mbo_valid |= OBD_MD_FLACL;
1025 if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1026 info->mti_mdt->mdt_lut.lut_mds_capa &&
1027 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1028 struct lustre_capa *capa;
1030 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1032 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1033 rc = mo_capa_get(env, next, capa, 0);
1036 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1041 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1046 static int mdt_renew_capa(struct mdt_thread_info *info)
1048 struct mdt_object *obj = info->mti_object;
1049 struct mdt_body *body;
1050 struct lustre_capa *capa, *c;
1054 /* if object doesn't exist, or server has disabled capability,
1055 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1058 if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1059 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1062 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1063 LASSERT(body != NULL);
1065 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1068 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1072 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1074 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1078 static int mdt_getattr(struct tgt_session_info *tsi)
1080 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1081 struct mdt_object *obj = info->mti_object;
1082 struct req_capsule *pill = info->mti_pill;
1083 struct mdt_body *reqbody;
1084 struct mdt_body *repbody;
1088 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1091 if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1092 rc = req_capsule_server_pack(pill);
1094 RETURN(err_serious(rc));
1095 rc = mdt_renew_capa(info);
1096 GOTO(out_shrink, rc);
1099 LASSERT(obj != NULL);
1100 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1102 /* Unlike intent case where we need to pre-fill out buffers early on
1103 * in intent policy for ldlm reasons, here we can have a much better
1104 * guess at EA size by just reading it from disk.
1105 * Exceptions are readdir and (missing) directory striping */
1107 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1108 /* No easy way to know how long is the symlink, but it cannot
1109 * be more than PATH_MAX, so we allocate +1 */
1112 /* A special case for fs ROOT: getattr there might fetch
1113 * default EA for entire fs, not just for this dir!
1115 } else if (lu_fid_eq(mdt_object_fid(obj),
1116 &info->mti_mdt->mdt_md_root_fid) &&
1117 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1118 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1120 /* Should the default strping be bigger, mdt_fix_reply
1121 * will reallocate */
1122 rc = DEF_REP_MD_SIZE;
1124 /* Read the actual EA size from disk */
1125 rc = mdt_attr_get_eabuf_size(info, obj);
1129 GOTO(out_shrink, rc);
1131 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1133 rc = req_capsule_server_pack(pill);
1134 if (unlikely(rc != 0))
1135 GOTO(out, rc = err_serious(rc));
1137 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1138 LASSERT(repbody != NULL);
1139 repbody->mbo_eadatasize = 0;
1140 repbody->mbo_aclsize = 0;
1142 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1143 rc = mdt_init_ucred(info, reqbody);
1145 rc = mdt_check_ucred(info);
1147 GOTO(out_shrink, rc);
1149 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1152 * Don't check capability at all, because rename might getattr for
1153 * remote obj, and at that time no capability is available.
1155 mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1156 rc = mdt_getattr_internal(info, obj, 0);
1157 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1158 mdt_exit_ucred(info);
1161 mdt_client_compatibility(info);
1162 rc2 = mdt_fix_reply(info);
1166 mdt_thread_info_fini(info);
1170 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1172 struct mdt_thread_info *info;
1173 struct ptlrpc_request *req = tgt_ses_req(tsi);
1174 struct obd_export *exp = req->rq_export;
1175 struct mdt_object *o1, *o2, *o;
1176 struct mdt_lock_handle *lh1, *lh2;
1177 struct mdc_swap_layouts *msl;
1181 /* client does not support layout lock, so layout swaping
1183 * FIXME: there is a problem for old clients which don't support
1184 * layout lock yet. If those clients have already opened the file
1185 * they won't be notified at all so that old layout may still be
1186 * used to do IO. This can be fixed after file release is landed by
1187 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1188 if (!exp_connect_layout(exp))
1189 RETURN(-EOPNOTSUPP);
1191 info = tsi2mdt_info(tsi);
1193 if (info->mti_dlm_req != NULL)
1194 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1196 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1197 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1198 req_capsule_client_get(info->mti_pill,
1201 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1202 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1203 req_capsule_client_get(info->mti_pill,
1206 o1 = info->mti_object;
1207 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1208 &info->mti_body->mbo_fid2);
1210 GOTO(out, rc = PTR_ERR(o));
1212 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1213 GOTO(put, rc = -ENOENT);
1215 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1216 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1222 /* permission check. Make sure the calling process having permission
1223 * to write both files. */
1224 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1229 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1234 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1236 GOTO(put, rc = -EPROTO);
1238 lh1 = &info->mti_lh[MDT_LH_NEW];
1239 mdt_lock_reg_init(lh1, LCK_EX);
1240 lh2 = &info->mti_lh[MDT_LH_OLD];
1241 mdt_lock_reg_init(lh2, LCK_EX);
1243 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1244 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1248 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1249 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1253 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1254 mdt_object_child(o2), msl->msl_flags);
1257 mdt_object_unlock(info, o2, lh2, rc);
1259 mdt_object_unlock(info, o1, lh1, rc);
1261 mdt_object_put(info->mti_env, o);
1263 mdt_thread_info_fini(info);
1267 static int mdt_raw_lookup(struct mdt_thread_info *info,
1268 struct mdt_object *parent,
1269 const struct lu_name *lname,
1270 struct ldlm_reply *ldlm_rep)
1272 struct md_object *next = mdt_object_child(info->mti_object);
1273 const struct mdt_body *reqbody = info->mti_body;
1274 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1275 struct mdt_body *repbody;
1279 if (reqbody->mbo_valid != OBD_MD_FLID)
1282 LASSERT(!info->mti_cross_ref);
1284 /* Only got the fid of this obj by name */
1285 fid_zero(child_fid);
1286 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1289 /* XXX is raw_lookup possible as intent operation? */
1292 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1295 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1297 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1300 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1301 repbody->mbo_fid1 = *child_fid;
1302 repbody->mbo_valid = OBD_MD_FLID;
1308 * UPDATE lock should be taken against parent, and be release before exit;
1309 * child_bits lock should be taken against child, and be returned back:
1310 * (1)normal request should release the child lock;
1311 * (2)intent request will grant the lock to client.
1313 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1314 struct mdt_lock_handle *lhc,
1316 struct ldlm_reply *ldlm_rep)
1318 struct ptlrpc_request *req = mdt_info_req(info);
1319 struct mdt_body *reqbody = NULL;
1320 struct mdt_object *parent = info->mti_object;
1321 struct mdt_object *child;
1322 struct md_object *next = mdt_object_child(parent);
1323 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1324 struct lu_name *lname = NULL;
1325 struct mdt_lock_handle *lhp = NULL;
1326 struct ldlm_lock *lock;
1333 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1334 LASSERT(ergo(is_resent,
1335 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1337 LASSERT(parent != NULL);
1339 if (info->mti_cross_ref) {
1340 /* Only getattr on the child. Parent is on another node. */
1341 mdt_set_disposition(info, ldlm_rep,
1342 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1344 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1346 PFID(mdt_object_fid(child)), ldlm_rep);
1349 /* Do not take lock for resent case. */
1350 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1351 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1352 lhc->mlh_reg_lh.cookie);
1353 LASSERT(fid_res_name_eq(mdt_object_fid(child),
1354 &lock->l_resource->lr_name));
1355 LDLM_LOCK_PUT(lock);
1357 mdt_lock_handle_init(lhc);
1358 mdt_lock_reg_init(lhc, LCK_PR);
1361 * Object's name is on another MDS, no lookup or layout
1362 * lock is needed here but update lock is.
1364 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1365 MDS_INODELOCK_LAYOUT);
1366 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1368 rc = mdt_object_lock(info, child, lhc, child_bits,
1374 /* Finally, we can get attr for child. */
1375 if (!mdt_object_exists(child)) {
1376 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1378 "remote object doesn't exist.\n");
1379 mdt_object_unlock(info, child, lhc, 1);
1383 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1384 rc = mdt_getattr_internal(info, child, 0);
1385 if (unlikely(rc != 0))
1386 mdt_object_unlock(info, child, lhc, 1);
1391 lname = &info->mti_name;
1392 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1394 if (lu_name_is_valid(lname)) {
1395 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1396 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1397 PNAME(lname), ldlm_rep);
1399 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1400 if (unlikely(reqbody == NULL))
1401 RETURN(err_serious(-EPROTO));
1403 *child_fid = reqbody->mbo_fid2;
1405 if (unlikely(!fid_is_sane(child_fid)))
1406 RETURN(err_serious(-EINVAL));
1408 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1410 PFID(mdt_object_fid(parent)),
1411 PFID(&reqbody->mbo_fid2), ldlm_rep);
1414 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1416 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1417 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1419 "Parent doesn't exist!\n");
1423 if (mdt_object_remote(parent)) {
1424 CERROR("%s: parent "DFID" is on remote target\n",
1425 mdt_obd_name(info->mti_mdt),
1426 PFID(mdt_object_fid(parent)));
1430 if (lu_name_is_valid(lname)) {
1431 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1438 /* step 1: lock parent only if parent is a directory */
1439 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1440 lhp = &info->mti_lh[MDT_LH_PARENT];
1441 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1442 rc = mdt_object_lock(info, parent, lhp,
1443 MDS_INODELOCK_UPDATE,
1445 if (unlikely(rc != 0))
1449 /* step 2: lookup child's fid by name */
1450 fid_zero(child_fid);
1451 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1454 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1457 GOTO(out_parent, rc);
1460 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1463 *step 3: find the child object by fid & lock it.
1464 * regardless if it is local or remote.
1466 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1467 * set parent dir fid the same as child fid in getattr by fid case
1468 * we should not lu_object_find() the object again, could lead
1469 * to hung if there is a concurrent unlink destroyed the object.
1471 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1472 mdt_object_get(info->mti_env, parent);
1475 child = mdt_object_find(info->mti_env, info->mti_mdt,
1479 if (unlikely(IS_ERR(child)))
1480 GOTO(out_parent, rc = PTR_ERR(child));
1482 /* Do not take lock for resent case. */
1483 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1484 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1485 lhc->mlh_reg_lh.cookie);
1487 if (!fid_res_name_eq(mdt_object_fid(child),
1488 &lock->l_resource->lr_name)) {
1489 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1490 &lock->l_resource->lr_name),
1491 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1492 PLDLMRES(lock->l_resource),
1493 PFID(mdt_object_fid(parent)));
1494 CWARN("Although resent, but still not get child lock"
1495 "parent:"DFID" child:"DFID"\n",
1496 PFID(mdt_object_fid(parent)),
1497 PFID(mdt_object_fid(child)));
1498 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1499 LDLM_LOCK_PUT(lock);
1502 LDLM_LOCK_PUT(lock);
1505 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1506 mdt_lock_handle_init(lhc);
1507 mdt_lock_reg_init(lhc, LCK_PR);
1510 if (!mdt_object_exists(child)) {
1511 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1513 "Object doesn't exist!\n");
1514 GOTO(out_child, rc = -ENOENT);
1517 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1518 mdt_object_exists(child) && !mdt_object_remote(child)) {
1519 struct md_attr *ma = &info->mti_attr;
1522 ma->ma_need = MA_INODE;
1523 rc = mdt_attr_get_complex(info, child, ma);
1524 if (unlikely(rc != 0))
1525 GOTO(out_child, rc);
1527 /* If the file has not been changed for some time, we
1528 * return not only a LOOKUP lock, but also an UPDATE
1529 * lock and this might save us RPC on later STAT. For
1530 * directories, it also let negative dentry cache start
1531 * working for this dir. */
1532 if (ma->ma_valid & MA_INODE &&
1533 ma->ma_attr.la_valid & LA_CTIME &&
1534 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1535 ma->ma_attr.la_ctime < cfs_time_current_sec())
1536 child_bits |= MDS_INODELOCK_UPDATE;
1539 /* layout lock must be granted in a best-effort way
1540 * for IT operations */
1541 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1542 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1543 exp_connect_layout(info->mti_exp) &&
1544 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1545 !mdt_object_remote(child) && ldlm_rep != NULL) {
1546 /* try to grant layout lock for regular file. */
1552 child_bits |= MDS_INODELOCK_LAYOUT;
1553 /* try layout lock, it may fail to be granted due to
1554 * contention at LOOKUP or UPDATE */
1555 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1557 child_bits &= ~MDS_INODELOCK_LAYOUT;
1558 LASSERT(child_bits != 0);
1559 rc = mdt_object_lock(info, child, lhc,
1560 child_bits, MDT_CROSS_LOCK);
1565 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1566 * client will enqueue the lock to the remote MDT */
1567 if (mdt_object_remote(child))
1568 child_bits &= ~MDS_INODELOCK_UPDATE;
1569 rc = mdt_object_lock(info, child, lhc, child_bits,
1572 if (unlikely(rc != 0))
1573 GOTO(out_child, rc);
1576 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1577 /* Get MA_SOM attributes if update lock is given. */
1579 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1580 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1583 /* finally, we can get attr for child. */
1584 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1585 rc = mdt_getattr_internal(info, child, ma_need);
1586 if (unlikely(rc != 0)) {
1587 mdt_object_unlock(info, child, lhc, 1);
1589 /* Debugging code. */
1590 LDLM_DEBUG(lock, "Returning lock to client");
1591 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1592 &lock->l_resource->lr_name),
1593 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1594 PLDLMRES(lock->l_resource),
1595 PFID(mdt_object_fid(child)));
1596 if (mdt_object_exists(child) && !mdt_object_remote(child))
1597 mdt_pack_size2body(info, child);
1600 LDLM_LOCK_PUT(lock);
1604 mdt_object_put(info->mti_env, child);
1607 mdt_object_unlock(info, parent, lhp, 1);
1611 /* normal handler: should release the child lock */
1612 static int mdt_getattr_name(struct tgt_session_info *tsi)
1614 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1615 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1616 struct mdt_body *reqbody;
1617 struct mdt_body *repbody;
1621 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1622 LASSERT(reqbody != NULL);
1623 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1624 LASSERT(repbody != NULL);
1626 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1627 repbody->mbo_eadatasize = 0;
1628 repbody->mbo_aclsize = 0;
1630 rc = mdt_init_ucred(info, reqbody);
1632 GOTO(out_shrink, rc);
1634 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1635 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1636 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1637 lhc->mlh_reg_lh.cookie = 0;
1639 mdt_exit_ucred(info);
1642 mdt_client_compatibility(info);
1643 rc2 = mdt_fix_reply(info);
1646 mdt_thread_info_fini(info);
1650 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1651 void *karg, void *uarg);
1653 static int mdt_set_info(struct tgt_session_info *tsi)
1655 struct ptlrpc_request *req = tgt_ses_req(tsi);
1658 int keylen, vallen, rc = 0;
1662 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1664 DEBUG_REQ(D_HA, req, "no set_info key");
1665 RETURN(err_serious(-EFAULT));
1668 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1671 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1673 DEBUG_REQ(D_HA, req, "no set_info val");
1674 RETURN(err_serious(-EFAULT));
1677 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1680 /* Swab any part of val you need to here */
1681 if (KEY_IS(KEY_READ_ONLY)) {
1682 spin_lock(&req->rq_export->exp_lock);
1684 *exp_connect_flags_ptr(req->rq_export) |=
1687 *exp_connect_flags_ptr(req->rq_export) &=
1688 ~OBD_CONNECT_RDONLY;
1689 spin_unlock(&req->rq_export->exp_lock);
1690 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1691 struct changelog_setinfo *cs = val;
1693 if (vallen != sizeof(*cs)) {
1694 CERROR("%s: bad changelog_clear setinfo size %d\n",
1695 tgt_name(tsi->tsi_tgt), vallen);
1698 if (ptlrpc_req_need_swab(req)) {
1699 __swab64s(&cs->cs_recno);
1700 __swab32s(&cs->cs_id);
1703 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1711 static int mdt_readpage(struct tgt_session_info *tsi)
1713 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1714 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1715 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1716 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1717 struct mdt_body *repbody;
1723 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1724 RETURN(err_serious(-ENOMEM));
1726 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1727 if (repbody == NULL || reqbody == NULL)
1728 RETURN(err_serious(-EFAULT));
1731 * prepare @rdpg before calling lower layers and transfer itself. Here
1732 * reqbody->size contains offset of where to start to read and
1733 * reqbody->nlink contains number bytes to read.
1735 rdpg->rp_hash = reqbody->mbo_size;
1736 if (rdpg->rp_hash != reqbody->mbo_size) {
1737 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1738 rdpg->rp_hash, reqbody->mbo_size);
1742 rdpg->rp_attrs = reqbody->mbo_mode;
1743 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1744 rdpg->rp_attrs |= LUDA_64BITHASH;
1745 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1746 exp_max_brw_size(tsi->tsi_exp));
1747 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1749 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1750 if (rdpg->rp_pages == NULL)
1753 for (i = 0; i < rdpg->rp_npages; ++i) {
1754 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1755 if (rdpg->rp_pages[i] == NULL)
1756 GOTO(free_rdpg, rc = -ENOMEM);
1759 /* call lower layers to fill allocated pages with directory data */
1760 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1762 GOTO(free_rdpg, rc);
1764 /* send pages to client */
1765 rc = tgt_sendpage(tsi, rdpg, rc);
1770 for (i = 0; i < rdpg->rp_npages; i++)
1771 if (rdpg->rp_pages[i] != NULL)
1772 __free_page(rdpg->rp_pages[i]);
1773 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1775 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1781 static int mdt_reint_internal(struct mdt_thread_info *info,
1782 struct mdt_lock_handle *lhc,
1785 struct req_capsule *pill = info->mti_pill;
1786 struct mdt_body *repbody;
1791 rc = mdt_reint_unpack(info, op);
1793 CERROR("Can't unpack reint, rc %d\n", rc);
1794 RETURN(err_serious(rc));
1797 /* for replay (no_create) lmm is not needed, client has it already */
1798 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1799 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1802 /* llog cookies are always 0, the field is kept for compatibility */
1803 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1804 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1806 rc = req_capsule_server_pack(pill);
1808 CERROR("Can't pack response, rc %d\n", rc);
1809 RETURN(err_serious(rc));
1812 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1813 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1815 repbody->mbo_eadatasize = 0;
1816 repbody->mbo_aclsize = 0;
1819 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1821 /* for replay no cookkie / lmm need, because client have this already */
1822 if (info->mti_spec.no_create)
1823 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1824 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1826 rc = mdt_init_ucred_reint(info);
1828 GOTO(out_shrink, rc);
1830 rc = mdt_fix_attr_ucred(info, op);
1832 GOTO(out_ucred, rc = err_serious(rc));
1834 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1835 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1836 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1837 GOTO(out_ucred, rc);
1839 rc = mdt_reint_rec(info, lhc);
1842 mdt_exit_ucred(info);
1844 mdt_client_compatibility(info);
1845 rc2 = mdt_fix_reply(info);
1851 static long mdt_reint_opcode(struct ptlrpc_request *req,
1852 const struct req_format **fmt)
1854 struct mdt_device *mdt;
1855 struct mdt_rec_reint *rec;
1858 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1860 opc = rec->rr_opcode;
1861 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1862 if (opc < REINT_MAX && fmt[opc] != NULL)
1863 req_capsule_extend(&req->rq_pill, fmt[opc]);
1865 mdt = mdt_exp2dev(req->rq_export);
1866 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1867 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1868 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1869 opc = err_serious(-EFAULT);
1872 opc = err_serious(-EFAULT);
1877 static int mdt_reint(struct tgt_session_info *tsi)
1881 static const struct req_format *reint_fmts[REINT_MAX] = {
1882 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1883 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1884 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1885 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1886 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1887 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1888 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1889 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1890 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1895 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1897 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1899 * No lock possible here from client to pass it to reint code
1902 rc = mdt_reint_internal(info, NULL, opc);
1903 mdt_thread_info_fini(info);
1908 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1912 /* this should sync the whole device */
1913 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1915 struct dt_device *dt = mdt->mdt_bottom;
1919 rc = dt->dd_ops->dt_sync(env, dt);
1923 /* this should sync this object */
1924 static int mdt_object_sync(struct mdt_thread_info *info)
1926 struct md_object *next;
1930 if (!mdt_object_exists(info->mti_object)) {
1931 CWARN("Non existing object "DFID"!\n",
1932 PFID(mdt_object_fid(info->mti_object)));
1935 next = mdt_object_child(info->mti_object);
1936 rc = mo_object_sync(info->mti_env, next);
1941 static int mdt_sync(struct tgt_session_info *tsi)
1943 struct ptlrpc_request *req = tgt_ses_req(tsi);
1944 struct req_capsule *pill = tsi->tsi_pill;
1945 struct mdt_body *body;
1950 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1951 RETURN(err_serious(-ENOMEM));
1953 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1954 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1956 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1958 /* sync an object */
1959 rc = mdt_object_sync(info);
1961 const struct lu_fid *fid;
1962 struct lu_attr *la = &info->mti_attr.ma_attr;
1964 info->mti_attr.ma_need = MA_INODE;
1965 info->mti_attr.ma_valid = 0;
1966 rc = mdt_attr_get_complex(info, info->mti_object,
1969 body = req_capsule_server_get(pill,
1971 fid = mdt_object_fid(info->mti_object);
1972 mdt_pack_attr2body(info, body, la, fid);
1975 mdt_thread_info_fini(info);
1978 mdt_counter_incr(req, LPROC_MDT_SYNC);
1984 * Handle quota control requests to consult current usage/limit, but also
1985 * to configure quota enforcement
1987 static int mdt_quotactl(struct tgt_session_info *tsi)
1989 struct obd_export *exp = tsi->tsi_exp;
1990 struct req_capsule *pill = tsi->tsi_pill;
1991 struct obd_quotactl *oqctl, *repoqc;
1993 struct mdt_device *mdt = mdt_exp2dev(exp);
1994 struct lu_device *qmt = mdt->mdt_qmt_dev;
1997 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1999 RETURN(err_serious(-EPROTO));
2001 rc = req_capsule_server_pack(pill);
2003 RETURN(err_serious(rc));
2005 switch (oqctl->qc_cmd) {
2007 case LUSTRE_Q_INVALIDATE:
2008 case LUSTRE_Q_FINVALIDATE:
2012 /* deprecated, not used any more */
2013 RETURN(-EOPNOTSUPP);
2014 /* master quotactl */
2020 RETURN(-EOPNOTSUPP);
2021 /* slave quotactl */
2026 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2030 /* map uid/gid for remote client */
2032 if (exp_connect_rmtclient(exp)) {
2033 struct lustre_idmap_table *idmap;
2035 idmap = exp->exp_mdt_data.med_idmap;
2037 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2038 oqctl->qc_cmd != Q_GETINFO))
2041 if (oqctl->qc_type == USRQUOTA)
2042 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2044 else if (oqctl->qc_type == GRPQUOTA)
2045 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2050 if (id == CFS_IDMAP_NOTFOUND) {
2051 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2056 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2058 RETURN(err_serious(-EFAULT));
2060 if (oqctl->qc_id != id)
2061 swap(oqctl->qc_id, id);
2063 switch (oqctl->qc_cmd) {
2069 /* forward quotactl request to QMT */
2070 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2075 /* slave quotactl */
2076 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2081 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2085 if (oqctl->qc_id != id)
2086 swap(oqctl->qc_id, id);
2092 /** clone llog ctxt from child (mdd)
2093 * This allows remote llog (replicator) access.
2094 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2095 * context was originally set up, or we can handle them directly.
2096 * I choose the latter, but that means I need any llog
2097 * contexts set up by child to be accessable by the mdt. So we clone the
2098 * context into our context list here.
2100 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2103 struct md_device *next = mdt->mdt_child;
2104 struct llog_ctxt *ctxt;
2107 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2110 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2111 if (rc || ctxt == NULL) {
2115 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2117 CERROR("Can't set mdt ctxt %d\n", rc);
2122 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2123 struct mdt_device *mdt, int idx)
2125 struct llog_ctxt *ctxt;
2127 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2130 /* Put once for the get we just did, and once for the clone */
2131 llog_ctxt_put(ctxt);
2132 llog_ctxt_put(ctxt);
2137 * sec context handlers
2139 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2143 rc = mdt_handle_idmap(tsi);
2145 struct ptlrpc_request *req = tgt_ses_req(tsi);
2148 opc = lustre_msg_get_opc(req->rq_reqmsg);
2149 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2150 sptlrpc_svc_ctx_invalidate(req);
2153 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2159 * quota request handlers
2161 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2163 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2164 struct lu_device *qmt = mdt->mdt_qmt_dev;
2169 RETURN(err_serious(-EOPNOTSUPP));
2171 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2175 struct mdt_object *mdt_object_new(const struct lu_env *env,
2176 struct mdt_device *d,
2177 const struct lu_fid *f)
2179 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2180 struct lu_object *o;
2181 struct mdt_object *m;
2184 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2185 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2186 if (unlikely(IS_ERR(o)))
2187 m = (struct mdt_object *)o;
2193 struct mdt_object *mdt_object_find(const struct lu_env *env,
2194 struct mdt_device *d,
2195 const struct lu_fid *f)
2197 struct lu_object *o;
2198 struct mdt_object *m;
2201 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2202 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2203 if (unlikely(IS_ERR(o)))
2204 m = (struct mdt_object *)o;
2212 * Asyncronous commit for mdt device.
2214 * Pass asynchonous commit call down the MDS stack.
2216 * \param env environment
2217 * \param mdt the mdt device
2219 static void mdt_device_commit_async(const struct lu_env *env,
2220 struct mdt_device *mdt)
2222 struct dt_device *dt = mdt->mdt_bottom;
2225 rc = dt->dd_ops->dt_commit_async(env, dt);
2226 if (unlikely(rc != 0))
2227 CWARN("async commit start failed with rc = %d", rc);
2231 * Mark the lock as "synchonous".
2233 * Mark the lock to deffer transaction commit to the unlock time.
2235 * \param lock the lock to mark as "synchonous"
2237 * \see mdt_is_lock_sync
2238 * \see mdt_save_lock
2240 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2242 lock->l_ast_data = (void*)1;
2246 * Check whehter the lock "synchonous" or not.
2248 * \param lock the lock to check
2249 * \retval 1 the lock is "synchonous"
2250 * \retval 0 the lock isn't "synchronous"
2252 * \see mdt_set_lock_sync
2253 * \see mdt_save_lock
2255 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2257 return lock->l_ast_data != NULL;
2261 * Blocking AST for mdt locks.
2263 * Starts transaction commit if in case of COS lock conflict or
2264 * deffers such a commit to the mdt_save_lock.
2266 * \param lock the lock which blocks a request or cancelling lock
2267 * \param desc unused
2268 * \param data unused
2269 * \param flag indicates whether this cancelling or blocking callback
2271 * \see ldlm_blocking_ast_nocheck
2273 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2274 void *data, int flag)
2276 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2277 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2281 if (flag == LDLM_CB_CANCELING)
2283 lock_res_and_lock(lock);
2284 if (lock->l_blocking_ast != mdt_blocking_ast) {
2285 unlock_res_and_lock(lock);
2288 if (mdt_cos_is_enabled(mdt) &&
2289 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2290 lock->l_blocking_lock != NULL &&
2291 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2292 mdt_set_lock_sync(lock);
2294 rc = ldlm_blocking_ast_nocheck(lock);
2296 /* There is no lock conflict if l_blocking_lock == NULL,
2297 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2298 * when the last reference to a local lock was released */
2299 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2302 rc = lu_env_init(&env, LCT_LOCAL);
2303 if (unlikely(rc != 0))
2304 CWARN("lu_env initialization failed with rc = %d,"
2305 "cannot start asynchronous commit\n", rc);
2307 mdt_device_commit_async(&env, mdt);
2313 /* Used for cross-MDT lock */
2314 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2315 void *data, int flag)
2317 struct lustre_handle lockh;
2321 case LDLM_CB_BLOCKING:
2322 ldlm_lock2handle(lock, &lockh);
2323 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2325 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2329 case LDLM_CB_CANCELING:
2330 LDLM_DEBUG(lock, "Revoke remote lock\n");
2338 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2339 struct mdt_object *o, const struct lu_fid *fid,
2340 struct lustre_handle *lh, ldlm_mode_t mode,
2343 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2344 ldlm_policy_data_t *policy = &mti->mti_policy;
2345 struct ldlm_res_id *res_id = &mti->mti_res_id;
2349 LASSERT(mdt_object_remote(o));
2351 LASSERT(ibits == MDS_INODELOCK_UPDATE);
2353 fid_build_reg_res_name(fid, res_id);
2355 memset(einfo, 0, sizeof(*einfo));
2356 einfo->ei_type = LDLM_IBITS;
2357 einfo->ei_mode = mode;
2358 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2359 einfo->ei_cb_cp = ldlm_completion_ast;
2360 einfo->ei_enq_slave = 0;
2361 einfo->ei_res_id = res_id;
2363 memset(policy, 0, sizeof(*policy));
2364 policy->l_inodebits.bits = ibits;
2366 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2371 static int mdt_object_local_lock(struct mdt_thread_info *info,
2372 struct mdt_object *o,
2373 struct mdt_lock_handle *lh, __u64 ibits,
2374 bool nonblock, int locality)
2376 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2377 ldlm_policy_data_t *policy = &info->mti_policy;
2378 struct ldlm_res_id *res_id = &info->mti_res_id;
2383 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2384 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2385 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2386 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2388 /* Only enqueue LOOKUP lock for remote object */
2389 if (mdt_object_remote(o))
2390 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2392 if (lh->mlh_type == MDT_PDO_LOCK) {
2393 /* check for exists after object is locked */
2394 if (mdt_object_exists(o) == 0) {
2395 /* Non-existent object shouldn't have PDO lock */
2398 /* Non-dir object shouldn't have PDO lock */
2399 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2404 memset(policy, 0, sizeof(*policy));
2405 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2407 dlmflags = LDLM_FL_ATOMIC_CB;
2409 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2412 * Take PDO lock on whole directory and build correct @res_id for lock
2413 * on part of directory.
2415 if (lh->mlh_pdo_hash != 0) {
2416 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2417 mdt_lock_pdo_mode(info, o, lh);
2418 if (lh->mlh_pdo_mode != LCK_NL) {
2420 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2421 * is never going to be sent to client and we do not
2422 * want it slowed down due to possible cancels.
2424 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2425 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2426 policy, res_id, dlmflags,
2427 info->mti_exp == NULL ? NULL :
2428 &info->mti_exp->exp_handle.h_cookie);
2434 * Finish res_id initializing by name hash marking part of
2435 * directory which is taking modification.
2437 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2440 policy->l_inodebits.bits = ibits;
2443 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2444 * going to be sent to client. If it is - mdt_intent_policy() path will
2445 * fix it up and turn FL_LOCAL flag off.
2447 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2448 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2449 info->mti_exp == NULL ? NULL :
2450 &info->mti_exp->exp_handle.h_cookie);
2452 mdt_object_unlock(info, o, lh, 1);
2453 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2454 lh->mlh_pdo_hash != 0 &&
2455 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2456 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2463 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2464 struct mdt_lock_handle *lh, __u64 ibits,
2465 bool nonblock, int locality)
2470 if (!mdt_object_remote(o))
2471 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2474 if (locality == MDT_LOCAL_LOCK) {
2475 CERROR("%s: try to get local lock for remote object"
2476 DFID".\n", mdt_obd_name(info->mti_mdt),
2477 PFID(mdt_object_fid(o)));
2481 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2482 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2483 MDS_INODELOCK_XATTR);
2484 if (ibits & MDS_INODELOCK_UPDATE) {
2485 /* Sigh, PDO needs to enqueue 2 locks right now, but
2486 * enqueue RPC can only request 1 lock, to avoid extra
2487 * RPC, so it will instead enqueue EX lock for remote
2488 * object anyway XXX*/
2489 if (lh->mlh_type == MDT_PDO_LOCK &&
2490 lh->mlh_pdo_hash != 0) {
2491 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2492 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2493 PFID(mdt_object_fid(o)));
2494 lh->mlh_pdo_hash = 0;
2495 lh->mlh_rreg_mode = LCK_EX;
2496 lh->mlh_type = MDT_REG_LOCK;
2498 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2501 MDS_INODELOCK_UPDATE);
2506 /* Only enqueue LOOKUP lock for remote object */
2507 if (ibits & MDS_INODELOCK_LOOKUP) {
2508 rc = mdt_object_local_lock(info, o, lh,
2509 MDS_INODELOCK_LOOKUP,
2510 nonblock, locality);
2518 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2519 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2521 return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2524 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2525 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2527 struct mdt_lock_handle tmp = *lh;
2530 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2538 * Save a lock within request object.
2540 * Keep the lock referenced until whether client ACK or transaction
2541 * commit happens or release the lock immediately depending on input
2542 * parameters. If COS is ON, a write lock is converted to COS lock
2545 * \param info thead info object
2546 * \param h lock handle
2547 * \param mode lock mode
2548 * \param decref force immediate lock releasing
2551 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2552 ldlm_mode_t mode, int decref)
2556 if (lustre_handle_is_used(h)) {
2557 if (decref || !info->mti_has_trans ||
2558 !(mode & (LCK_PW | LCK_EX))){
2559 mdt_fid_unlock(h, mode);
2561 struct mdt_device *mdt = info->mti_mdt;
2562 struct ldlm_lock *lock = ldlm_handle2lock(h);
2563 struct ptlrpc_request *req = mdt_info_req(info);
2566 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2568 /* there is no request if mdt_object_unlock() is called
2569 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2570 if (likely(req != NULL)) {
2571 CDEBUG(D_HA, "request = %p reply state = %p"
2572 " transno = "LPD64"\n", req,
2573 req->rq_reply_state, req->rq_transno);
2574 if (mdt_cos_is_enabled(mdt)) {
2576 ldlm_lock_downgrade(lock, LCK_COS);
2579 ptlrpc_save_lock(req, h, mode, no_ack);
2581 ldlm_lock_decref(h, mode);
2583 if (mdt_is_lock_sync(lock)) {
2584 CDEBUG(D_HA, "found sync-lock,"
2585 " async commit started\n");
2586 mdt_device_commit_async(info->mti_env,
2589 LDLM_LOCK_PUT(lock);
2598 * Unlock mdt object.
2600 * Immeditely release the regular lock and the PDO lock or save the
2601 * lock in reqeuest and keep them referenced until client ACK or
2602 * transaction commit.
2604 * \param info thread info object
2605 * \param o mdt object
2606 * \param lh mdt lock handle referencing regular and PDO locks
2607 * \param decref force immediate lock releasing
2609 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2610 struct mdt_lock_handle *lh, int decref)
2614 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2615 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2617 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2618 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2623 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2624 const struct lu_fid *f,
2625 struct mdt_lock_handle *lh,
2628 struct mdt_object *o;
2630 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2634 rc = mdt_object_lock(info, o, lh, ibits,
2637 mdt_object_put(info->mti_env, o);
2644 void mdt_object_unlock_put(struct mdt_thread_info * info,
2645 struct mdt_object * o,
2646 struct mdt_lock_handle *lh,
2649 mdt_object_unlock(info, o, lh, decref);
2650 mdt_object_put(info->mti_env, o);
2654 * Generic code handling requests that have struct mdt_body passed in:
2656 * - extract mdt_body from request and save it in @info, if present;
2658 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2661 * - if HABEO_CORPUS flag is set for this request type check whether object
2662 * actually exists on storage (lu_object_exists()).
2665 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2667 const struct mdt_body *body;
2668 struct mdt_object *obj;
2669 const struct lu_env *env;
2670 struct req_capsule *pill;
2674 env = info->mti_env;
2675 pill = info->mti_pill;
2677 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2681 if (!(body->mbo_valid & OBD_MD_FLID))
2684 if (!fid_is_sane(&body->mbo_fid1)) {
2685 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2690 * Do not get size or any capa fields before we check that request
2691 * contains capa actually. There are some requests which do not, for
2692 * instance MDS_IS_SUBDIR.
2694 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2695 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2696 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2697 req_capsule_client_get(pill, &RMF_CAPA1));
2699 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2701 if ((flags & HABEO_CORPUS) &&
2702 !mdt_object_exists(obj)) {
2703 mdt_object_put(env, obj);
2704 /* for capability renew ENOENT will be handled in
2706 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2711 info->mti_object = obj;
2720 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2722 struct req_capsule *pill = info->mti_pill;
2726 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2727 rc = mdt_body_unpack(info, flags);
2731 if (rc == 0 && (flags & HABEO_REFERO)) {
2733 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2734 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2736 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2737 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2740 rc = req_capsule_server_pack(pill);
2745 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2747 struct md_device *next = m->mdt_child;
2749 return next->md_ops->mdo_init_capa_ctxt(env, next,
2750 m->mdt_lut.lut_mds_capa,
2751 m->mdt_capa_timeout,
2756 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2758 lh->mlh_type = MDT_NUL_LOCK;
2759 lh->mlh_reg_lh.cookie = 0ull;
2760 lh->mlh_reg_mode = LCK_MINMODE;
2761 lh->mlh_pdo_lh.cookie = 0ull;
2762 lh->mlh_pdo_mode = LCK_MINMODE;
2763 lh->mlh_rreg_lh.cookie = 0ull;
2764 lh->mlh_rreg_mode = LCK_MINMODE;
2767 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2769 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2770 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2774 * Initialize fields of struct mdt_thread_info. Other fields are left in
2775 * uninitialized state, because it's too expensive to zero out whole
2776 * mdt_thread_info (> 1K) on each request arrival.
2778 void mdt_thread_info_init(struct ptlrpc_request *req,
2779 struct mdt_thread_info *info)
2783 info->mti_pill = &req->rq_pill;
2786 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2787 mdt_lock_handle_init(&info->mti_lh[i]);
2789 /* mdt device: it can be NULL while CONNECT */
2790 if (req->rq_export) {
2791 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2792 info->mti_exp = req->rq_export;
2794 info->mti_mdt = NULL;
2795 info->mti_env = req->rq_svc_thread->t_env;
2796 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2798 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2799 info->mti_big_buf = LU_BUF_NULL;
2800 info->mti_body = NULL;
2801 info->mti_object = NULL;
2802 info->mti_dlm_req = NULL;
2803 info->mti_has_trans = 0;
2804 info->mti_cross_ref = 0;
2805 info->mti_opdata = 0;
2806 info->mti_big_lmm_used = 0;
2808 info->mti_spec.no_create = 0;
2809 info->mti_spec.sp_rm_entry = 0;
2811 info->mti_spec.u.sp_ea.eadata = NULL;
2812 info->mti_spec.u.sp_ea.eadatalen = 0;
2815 void mdt_thread_info_fini(struct mdt_thread_info *info)
2819 if (info->mti_object != NULL) {
2820 mdt_object_put(info->mti_env, info->mti_object);
2821 info->mti_object = NULL;
2824 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2825 mdt_lock_handle_fini(&info->mti_lh[i]);
2826 info->mti_env = NULL;
2827 info->mti_pill = NULL;
2828 info->mti_exp = NULL;
2830 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2831 lu_buf_free(&info->mti_big_buf);
2834 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2836 struct mdt_thread_info *mti;
2837 struct lustre_capa *lc;
2839 mti = mdt_th_info(tsi->tsi_env);
2840 LASSERT(mti != NULL);
2842 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2843 if (tsi->tsi_corpus != NULL) {
2844 struct req_capsule *pill = tsi->tsi_pill;
2846 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2847 lu_object_get(tsi->tsi_corpus);
2850 * XXX: must be part of tgt_mdt_body_unpack but moved here
2851 * due to mdt_set_capainfo().
2853 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2854 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2855 lc = req_capsule_client_get(pill, &RMF_CAPA1);
2856 mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2860 mti->mti_body = tsi->tsi_mdt_body;
2861 mti->mti_dlm_req = tsi->tsi_dlm_req;
2866 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2868 struct ptlrpc_request *req = tgt_ses_req(tsi);
2873 rc = tgt_connect(tsi);
2877 rc = mdt_init_idmap(tsi);
2882 obd_disconnect(class_export_get(req->rq_export));
2901 static int mdt_intent_getattr(enum mdt_it_code opcode,
2902 struct mdt_thread_info *info,
2903 struct ldlm_lock **,
2906 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2907 struct mdt_thread_info *info,
2908 struct ldlm_lock **lockp,
2911 static int mdt_intent_layout(enum mdt_it_code opcode,
2912 struct mdt_thread_info *info,
2913 struct ldlm_lock **,
2915 static int mdt_intent_reint(enum mdt_it_code opcode,
2916 struct mdt_thread_info *info,
2917 struct ldlm_lock **,
2920 static struct mdt_it_flavor {
2921 const struct req_format *it_fmt;
2923 int (*it_act)(enum mdt_it_code ,
2924 struct mdt_thread_info *,
2925 struct ldlm_lock **,
2928 } mdt_it_flavor[] = {
2930 .it_fmt = &RQF_LDLM_INTENT,
2931 /*.it_flags = HABEO_REFERO,*/
2933 .it_act = mdt_intent_reint,
2934 .it_reint = REINT_OPEN
2937 .it_fmt = &RQF_LDLM_INTENT,
2939 * OCREAT is not a MUTABOR request as if the file
2941 * We do the extra check of OBD_CONNECT_RDONLY in
2942 * mdt_reint_open() when we really need to create
2946 .it_act = mdt_intent_reint,
2947 .it_reint = REINT_OPEN
2950 .it_fmt = &RQF_LDLM_INTENT,
2951 .it_flags = MUTABOR,
2952 .it_act = mdt_intent_reint,
2953 .it_reint = REINT_CREATE
2955 [MDT_IT_GETATTR] = {
2956 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2957 .it_flags = HABEO_REFERO,
2958 .it_act = mdt_intent_getattr
2960 [MDT_IT_READDIR] = {
2966 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2967 .it_flags = HABEO_REFERO,
2968 .it_act = mdt_intent_getattr
2971 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
2972 .it_flags = MUTABOR,
2974 .it_reint = REINT_UNLINK
2978 .it_flags = MUTABOR,
2981 [MDT_IT_GETXATTR] = {
2982 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
2983 .it_flags = HABEO_CORPUS,
2984 .it_act = mdt_intent_getxattr
2987 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
2989 .it_act = mdt_intent_layout
2994 mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp,
2995 struct ldlm_lock *new_lock, struct mdt_lock_handle *lh,
2998 struct ptlrpc_request *req = mdt_info_req(info);
2999 struct ldlm_lock *lock = *lockp;
3002 * Get new lock only for cases when possible resent did not find any
3005 if (new_lock == NULL)
3006 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3008 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3009 lh->mlh_reg_lh.cookie = 0;
3013 LASSERTF(new_lock != NULL,
3014 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3017 * If we've already given this lock to a client once, then we should
3018 * have no readers or writers. Otherwise, we should have one reader
3019 * _or_ writer ref (which will be zeroed below) before returning the
3022 if (new_lock->l_export == req->rq_export) {
3023 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3025 LASSERT(new_lock->l_export == NULL);
3026 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3031 if (new_lock->l_export == req->rq_export) {
3033 * Already gave this to the client, which means that we
3034 * reconstructed a reply.
3036 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3038 lh->mlh_reg_lh.cookie = 0;
3039 RETURN(ELDLM_LOCK_REPLACED);
3043 * Fixup the lock to be given to the client.
3045 lock_res_and_lock(new_lock);
3046 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3047 * possible blocking AST. */
3048 while (new_lock->l_readers > 0) {
3049 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3050 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3051 new_lock->l_readers--;
3053 while (new_lock->l_writers > 0) {
3054 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3055 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3056 new_lock->l_writers--;
3059 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3060 new_lock->l_blocking_ast = lock->l_blocking_ast;
3061 new_lock->l_completion_ast = lock->l_completion_ast;
3062 new_lock->l_remote_handle = lock->l_remote_handle;
3063 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3065 unlock_res_and_lock(new_lock);
3067 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3068 &new_lock->l_remote_handle,
3069 &new_lock->l_exp_hash);
3071 LDLM_LOCK_RELEASE(new_lock);
3072 lh->mlh_reg_lh.cookie = 0;
3074 RETURN(ELDLM_LOCK_REPLACED);
3077 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3078 struct ldlm_lock *new_lock,
3079 struct ldlm_lock **old_lock,
3080 struct mdt_lock_handle *lh,
3081 enum mdt_it_code opcode)
3083 struct ptlrpc_request *req = mdt_info_req(info);
3084 struct obd_export *exp = req->rq_export;
3085 struct lustre_handle remote_hdl;
3086 struct ldlm_request *dlmreq;
3087 struct ldlm_lock *lock;
3089 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3092 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3093 remote_hdl = dlmreq->lock_handle[0];
3094 /* If the client does not require open lock, it does not need to
3095 * search lock in exp_lock_hash, since the server thread will
3096 * make sure the lock will be released, and the resend request
3097 * can always re-enqueue the lock */
3098 if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3099 info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3100 /* In the function below, .hs_keycmp resolves to
3101 * ldlm_export_lock_keycmp() */
3102 /* coverity[overrun-buffer-val] */
3103 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3105 lock_res_and_lock(lock);
3106 if (lock != new_lock) {
3107 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3108 lh->mlh_reg_mode = lock->l_granted_mode;
3110 LDLM_DEBUG(lock, "Restoring lock cookie");
3111 DEBUG_REQ(D_DLMTRACE, req,
3112 "restoring lock cookie "LPX64,
3113 lh->mlh_reg_lh.cookie);
3115 *old_lock = LDLM_LOCK_GET(lock);
3116 cfs_hash_put(exp->exp_lock_hash,
3118 unlock_res_and_lock(lock);
3121 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3122 unlock_res_and_lock(lock);
3126 * If the xid matches, then we know this is a resent request, and allow
3127 * it. (It's probably an OPEN, for which we don't send a lock.
3129 if (req_xid_is_last(req))
3133 * This remote handle isn't enqueued, so we never received or processed
3134 * this request. Clear MSG_RESENT, because it can be handled like any
3135 * normal request now.
3137 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3139 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3143 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3144 struct mdt_thread_info *info,
3145 struct ldlm_lock **lockp,
3148 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3149 struct ldlm_reply *ldlm_rep = NULL;
3153 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3154 * (for the resend case) or a new lock. Below we will use it to
3155 * replace the original lock.
3157 mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3158 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3159 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3160 rc = mdt_object_lock(info, info->mti_object, lhc,
3161 MDS_INODELOCK_XATTR,
3167 grc = mdt_getxattr(info);
3169 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3171 if (mdt_info_req(info)->rq_repmsg != NULL)
3172 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3173 if (ldlm_rep == NULL)
3174 RETURN(err_serious(-EFAULT));
3176 ldlm_rep->lock_policy_res2 = grc;
3181 static int mdt_intent_getattr(enum mdt_it_code opcode,
3182 struct mdt_thread_info *info,
3183 struct ldlm_lock **lockp,
3186 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3187 struct ldlm_lock *new_lock = NULL;
3189 struct ldlm_reply *ldlm_rep;
3190 struct mdt_body *reqbody;
3191 struct mdt_body *repbody;
3195 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3198 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3201 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3202 repbody->mbo_eadatasize = 0;
3203 repbody->mbo_aclsize = 0;
3207 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3209 case MDT_IT_GETATTR:
3210 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3214 CERROR("Unsupported intent (%d)\n", opcode);
3215 GOTO(out_shrink, rc = -EINVAL);
3218 rc = mdt_init_ucred(info, reqbody);
3220 GOTO(out_shrink, rc);
3222 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3223 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3225 /* Get lock from request for possible resent case. */
3226 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3228 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3229 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3231 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3232 ldlm_rep->lock_policy_res2 = 0;
3233 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3234 ldlm_rep->lock_policy_res2) {
3235 lhc->mlh_reg_lh.cookie = 0ull;
3236 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3239 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3242 mdt_exit_ucred(info);
3244 mdt_client_compatibility(info);
3245 rc2 = mdt_fix_reply(info);
3251 static int mdt_intent_layout(enum mdt_it_code opcode,
3252 struct mdt_thread_info *info,
3253 struct ldlm_lock **lockp,
3256 struct layout_intent *layout;
3258 struct mdt_object *obj = NULL;
3262 if (opcode != MDT_IT_LAYOUT) {
3263 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3268 fid = &info->mti_tmp_fid2;
3269 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3271 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3273 RETURN(PTR_ERR(obj));
3275 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3276 /* get the length of lsm */
3277 rc = mdt_attr_get_eabuf_size(info, obj);
3281 if (rc > info->mti_mdt->mdt_max_mdsize)
3282 info->mti_mdt->mdt_max_mdsize = rc;
3285 mdt_object_put(info->mti_env, obj);
3287 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3288 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3289 rc = req_capsule_server_pack(info->mti_pill);
3293 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3294 LASSERT(layout != NULL);
3295 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3296 /* return to normal ldlm handling */
3299 CERROR("%s: Unsupported layout intent (%d)\n",
3300 mdt_obd_name(info->mti_mdt), layout->li_opc);
3304 static int mdt_intent_reint(enum mdt_it_code opcode,
3305 struct mdt_thread_info *info,
3306 struct ldlm_lock **lockp,
3309 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3310 struct ldlm_reply *rep = NULL;
3314 static const struct req_format *intent_fmts[REINT_MAX] = {
3315 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3316 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3321 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3325 if (mdt_it_flavor[opcode].it_reint != opc) {
3326 CERROR("Reint code %ld doesn't match intent: %d\n",
3328 RETURN(err_serious(-EPROTO));
3331 /* Get lock from request for possible resent case. */
3332 mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3334 rc = mdt_reint_internal(info, lhc, opc);
3336 /* Check whether the reply has been packed successfully. */
3337 if (mdt_info_req(info)->rq_repmsg != NULL)
3338 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3340 RETURN(err_serious(-EFAULT));
3342 /* MDC expects this in any case */
3344 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3346 /* the open lock or the lock for cross-ref object should be
3347 * returned to the client */
3348 if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3349 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3350 rep->lock_policy_res2 = 0;
3351 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3355 rep->lock_policy_res2 = clear_serious(rc);
3357 if (rep->lock_policy_res2 == -ENOENT &&
3358 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3359 rep->lock_policy_res2 = 0;
3361 if (rc == -ENOTCONN || rc == -ENODEV ||
3362 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3364 * If it is the disconnect error (ENODEV & ENOCONN), the error
3365 * will be returned by rq_status, and client at ptlrpc layer
3366 * will detect this, then disconnect, reconnect the import
3367 * immediately, instead of impacting the following the rpc.
3369 lhc->mlh_reg_lh.cookie = 0ull;
3373 * For other cases, the error will be returned by intent.
3374 * and client will retrieve the result from intent.
3377 * FIXME: when open lock is finished, that should be
3380 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3381 LASSERTF(rc == 0, "Error occurred but lock handle "
3382 "is still in use, rc = %d\n", rc);
3383 rep->lock_policy_res2 = 0;
3384 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3387 lhc->mlh_reg_lh.cookie = 0ull;
3388 RETURN(ELDLM_LOCK_ABORTED);
3393 static int mdt_intent_code(long itcode)
3401 case IT_OPEN|IT_CREAT:
3408 rc = MDT_IT_READDIR;
3411 rc = MDT_IT_GETATTR;
3423 rc = MDT_IT_GETXATTR;
3428 case IT_QUOTA_DQACQ:
3433 CERROR("Unknown intent opcode: %ld\n", itcode);
3440 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3441 struct ldlm_lock **lockp, __u64 flags)
3443 struct req_capsule *pill;
3444 struct mdt_it_flavor *flv;
3449 opc = mdt_intent_code(itopc);
3453 pill = info->mti_pill;
3455 if (opc == MDT_IT_QUOTA) {
3456 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3459 RETURN(-EOPNOTSUPP);
3461 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3462 /* pass the request to quota master */
3463 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3464 mdt_info_req(info), lockp,
3469 flv = &mdt_it_flavor[opc];
3470 if (flv->it_fmt != NULL)
3471 req_capsule_extend(pill, flv->it_fmt);
3473 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3475 struct ptlrpc_request *req = mdt_info_req(info);
3476 if (flv->it_flags & MUTABOR &&
3477 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3480 if (rc == 0 && flv->it_act != NULL) {
3481 struct ldlm_reply *rep;
3483 /* execute policy */
3484 rc = flv->it_act(opc, info, lockp, flags);
3486 /* Check whether the reply has been packed successfully. */
3487 if (mdt_info_req(info)->rq_repmsg != NULL) {
3488 rep = req_capsule_server_get(info->mti_pill,
3490 rep->lock_policy_res2 =
3491 ptlrpc_status_hton(rep->lock_policy_res2);
3499 static int mdt_intent_policy(struct ldlm_namespace *ns,
3500 struct ldlm_lock **lockp, void *req_cookie,
3501 ldlm_mode_t mode, __u64 flags, void *data)
3503 struct tgt_session_info *tsi;
3504 struct mdt_thread_info *info;
3505 struct ptlrpc_request *req = req_cookie;
3506 struct ldlm_intent *it;
3507 struct req_capsule *pill;
3512 LASSERT(req != NULL);
3514 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3516 info = tsi2mdt_info(tsi);
3517 LASSERT(info != NULL);
3518 pill = info->mti_pill;
3519 LASSERT(pill->rc_req == req);
3521 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3522 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3523 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3525 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3529 /* Lock without inodebits makes no sense and will oops
3530 * later in ldlm. Let's check it now to see if we have
3531 * ibits corrupted somewhere in mdt_intent_opc().
3532 * The case for client miss to set ibits has been
3533 * processed by others. */
3534 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3535 lr_type == LDLM_IBITS,
3536 info->mti_dlm_req->lock_desc.\
3537 l_policy_data.l_inodebits.bits != 0));
3539 rc = err_serious(-EFAULT);
3541 /* No intent was provided */
3542 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3543 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3544 rc = req_capsule_server_pack(pill);
3546 rc = err_serious(rc);
3548 mdt_thread_info_fini(info);
3552 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3554 struct seq_server_site *ss = mdt_seq_site(mdt);
3556 if (ss->ss_node_id == 0)
3559 if (ss->ss_client_seq != NULL) {
3560 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3561 ss->ss_client_seq->lcs_exp = NULL;
3564 if (ss->ss_server_fld != NULL) {
3565 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3566 ss->ss_server_fld->lsf_control_exp = NULL;
3570 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3572 struct seq_server_site *ss = mdt_seq_site(mdt);
3577 if (ss->ss_server_seq != NULL)
3578 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3581 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3583 mdt_seq_fini_cli(mdt);
3584 mdt_deregister_seq_exp(mdt);
3586 return seq_site_fini(env, mdt_seq_site(mdt));
3590 * It will retrieve its FLDB entries from MDT0, and it only happens
3591 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3592 * and it needs to refresh FLDB from the MDT0.
3594 static int mdt_register_lwp_callback(void *data)
3597 struct mdt_device *mdt = data;
3598 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3602 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3604 if (!likely(fld->lsf_new))
3607 rc = lu_env_init(&env, LCT_MD_THREAD);
3609 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3613 rc = fld_update_from_controller(&env, fld);
3615 CERROR("%s: cannot update controller: rc = %d\n",
3616 mdt_obd_name(mdt), rc);
3624 static int mdt_register_seq_exp(struct mdt_device *mdt)
3626 struct seq_server_site *ss = mdt_seq_site(mdt);
3627 char *lwp_name = NULL;
3630 if (ss->ss_node_id == 0)
3633 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3634 if (lwp_name == NULL)
3635 GOTO(out_free, rc = -ENOMEM);
3637 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3641 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3646 rc = lustre_register_lwp_item(lwp_name,
3647 &ss->ss_server_fld->lsf_control_exp,
3648 mdt_register_lwp_callback, mdt);
3650 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3651 ss->ss_client_seq->lcs_exp = NULL;
3655 if (lwp_name != NULL)
3656 OBD_FREE(lwp_name, MAX_OBD_NAME);
3662 * Init client sequence manager which is used by local MDS to talk to sequence
3663 * controller on remote node.
3665 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3667 struct seq_server_site *ss = mdt_seq_site(mdt);
3672 /* check if this is adding the first MDC and controller is not yet
3674 OBD_ALLOC_PTR(ss->ss_client_seq);
3675 if (ss->ss_client_seq == NULL)
3678 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3679 if (prefix == NULL) {
3680 OBD_FREE_PTR(ss->ss_client_seq);
3681 ss->ss_client_seq = NULL;
3685 /* Note: seq_client_fini will be called in seq_site_fini */
3686 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3687 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3688 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3690 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3692 OBD_FREE_PTR(ss->ss_client_seq);
3693 ss->ss_client_seq = NULL;
3697 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3702 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3704 struct seq_server_site *ss;
3708 ss = mdt_seq_site(mdt);
3709 /* init sequence controller server(MDT0) */
3710 if (ss->ss_node_id == 0) {
3711 OBD_ALLOC_PTR(ss->ss_control_seq);
3712 if (ss->ss_control_seq == NULL)
3715 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3716 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3719 GOTO(out_seq_fini, rc);
3722 /* Init normal sequence server */
3723 OBD_ALLOC_PTR(ss->ss_server_seq);
3724 if (ss->ss_server_seq == NULL)
3725 GOTO(out_seq_fini, rc = -ENOMEM);
3727 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3728 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3730 GOTO(out_seq_fini, rc);
3732 /* init seq client for seq server to talk to seq controller(MDT0) */
3733 rc = mdt_seq_init_cli(env, mdt);
3735 GOTO(out_seq_fini, rc);
3737 if (ss->ss_node_id != 0)
3738 /* register controler export through lwp */
3739 rc = mdt_register_seq_exp(mdt);
3744 mdt_seq_fini(env, mdt);
3752 static int mdt_fld_fini(const struct lu_env *env,
3753 struct mdt_device *m)
3755 struct seq_server_site *ss = mdt_seq_site(m);
3758 if (ss && ss->ss_server_fld) {
3759 fld_server_fini(env, ss->ss_server_fld);
3760 OBD_FREE_PTR(ss->ss_server_fld);
3761 ss->ss_server_fld = NULL;
3767 static int mdt_fld_init(const struct lu_env *env,
3769 struct mdt_device *m)
3771 struct seq_server_site *ss;
3775 ss = mdt_seq_site(m);
3777 OBD_ALLOC_PTR(ss->ss_server_fld);
3778 if (ss->ss_server_fld == NULL)
3779 RETURN(rc = -ENOMEM);
3781 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3784 OBD_FREE_PTR(ss->ss_server_fld);
3785 ss->ss_server_fld = NULL;
3792 static void mdt_stack_pre_fini(const struct lu_env *env,
3793 struct mdt_device *m, struct lu_device *top)
3795 struct lustre_cfg_bufs *bufs;
3796 struct lustre_cfg *lcfg;
3797 struct mdt_thread_info *info;
3802 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3803 LASSERT(info != NULL);
3805 bufs = &info->mti_u.bufs;
3807 LASSERT(m->mdt_child_exp);
3808 LASSERT(m->mdt_child_exp->exp_obd);
3810 /* process cleanup, pass mdt obd name to get obd umount flags */
3811 /* XXX: this is needed because all layers are referenced by
3812 * objects (some of them are pinned by osd, for example *
3813 * the proper solution should be a model where object used
3814 * by osd only doesn't have mdt/mdd slices -bzzz */
3815 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3816 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3817 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3819 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3822 top->ld_ops->ldo_process_config(env, top, lcfg);
3823 lustre_cfg_free(lcfg);
3827 static void mdt_stack_fini(const struct lu_env *env,
3828 struct mdt_device *m, struct lu_device *top)
3830 struct obd_device *obd = mdt2obd_dev(m);
3831 struct lustre_cfg_bufs *bufs;
3832 struct lustre_cfg *lcfg;
3833 struct mdt_thread_info *info;
3837 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3838 LASSERT(info != NULL);
3840 lu_dev_del_linkage(top->ld_site, top);
3842 lu_site_purge(env, top->ld_site, -1);
3844 bufs = &info->mti_u.bufs;
3845 /* process cleanup, pass mdt obd name to get obd umount flags */
3846 /* another purpose is to let all layers to release their objects */
3847 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3852 lustre_cfg_bufs_set_string(bufs, 1, flags);
3853 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3855 CERROR("Cannot alloc lcfg!\n");
3859 top->ld_ops->ldo_process_config(env, top, lcfg);
3860 lustre_cfg_free(lcfg);
3862 lu_site_purge(env, top->ld_site, -1);
3864 m->mdt_child = NULL;
3865 m->mdt_bottom = NULL;
3867 obd_disconnect(m->mdt_child_exp);
3868 m->mdt_child_exp = NULL;
3870 obd_disconnect(m->mdt_bottom_exp);
3871 m->mdt_child_exp = NULL;
3874 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3875 const char *next, struct obd_export **exp)
3877 struct obd_connect_data *data = NULL;
3878 struct obd_device *obd;
3882 OBD_ALLOC_PTR(data);
3884 GOTO(out, rc = -ENOMEM);
3886 obd = class_name2obd(next);
3888 CERROR("%s: can't locate next device: %s\n",
3889 mdt_obd_name(m), next);
3890 GOTO(out, rc = -ENOTCONN);
3893 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3894 data->ocd_version = LUSTRE_VERSION_CODE;
3896 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3898 CERROR("%s: cannot connect to next dev %s (%d)\n",
3899 mdt_obd_name(m), next, rc);
3909 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3910 struct lustre_cfg *cfg)
3912 char *dev = lustre_cfg_string(cfg, 0);
3913 int rc, name_size, uuid_size;
3914 char *name, *uuid, *p;
3915 struct lustre_cfg_bufs *bufs;
3916 struct lustre_cfg *lcfg;
3917 struct obd_device *obd;
3918 struct lustre_profile *lprof;
3919 struct lu_site *site;
3922 /* in 1.8 we had the only device in the stack - MDS.
3923 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3924 * in 2.3 OSD is instantiated by obd_mount.c, so we need
3925 * to generate names and setup MDT, MDD. MDT will be using
3926 * generated name to connect to MDD. for MDD the next device
3927 * will be LOD with name taken from so called "profile" which
3928 * is generated by mount_option line
3930 * 1.8 MGS generates config. commands like this:
3931 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
3932 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
3933 * 2.0 MGS generates config. commands like this:
3934 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
3935 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3936 * 3:lustre-MDT0000-mdtlov 4:f
3938 * we generate MDD name from MDT one, just replacing T with D
3940 * after all the preparations, the logical equivalent will be
3941 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
3942 * 3:lustre-MDT0000-mdtlov 4:f
3943 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3944 * 3:lustre-MDD0000 4:f
3946 * notice we build the stack from down to top: MDD first, then MDT */
3948 name_size = MAX_OBD_NAME;
3949 uuid_size = MAX_OBD_NAME;
3951 OBD_ALLOC(name, name_size);
3952 OBD_ALLOC(uuid, uuid_size);
3953 if (name == NULL || uuid == NULL)
3954 GOTO(cleanup_mem, rc = -ENOMEM);
3956 OBD_ALLOC_PTR(bufs);
3958 GOTO(cleanup_mem, rc = -ENOMEM);
3961 p = strstr(name, "-MDT");
3963 GOTO(free_bufs, rc = -ENOMEM);
3966 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3968 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3969 if (lprof == NULL || lprof->lp_dt == NULL) {
3970 CERROR("can't find the profile: %s\n",
3971 lustre_cfg_string(cfg, 0));
3972 GOTO(free_bufs, rc = -EINVAL);
3975 lustre_cfg_bufs_reset(bufs, name);
3976 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3977 lustre_cfg_bufs_set_string(bufs, 2, uuid);
3978 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3980 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3982 GOTO(free_bufs, rc = -ENOMEM);
3984 rc = class_attach(lcfg);
3986 GOTO(lcfg_cleanup, rc);
3988 obd = class_name2obd(name);
3990 CERROR("Can not find obd %s (%s in config)\n",
3991 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3992 GOTO(class_detach, rc = -EINVAL);
3995 lustre_cfg_free(lcfg);
3997 lustre_cfg_bufs_reset(bufs, name);
3998 lustre_cfg_bufs_set_string(bufs, 1, uuid);
3999 lustre_cfg_bufs_set_string(bufs, 2, dev);
4000 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4002 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4004 rc = class_setup(obd, lcfg);
4006 GOTO(class_detach, rc);
4008 /* connect to MDD we just setup */
4009 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4011 GOTO(class_detach, rc);
4013 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4015 LASSERT(mdt_lu_site(mdt) == NULL);
4016 mdt->mdt_lu_dev.ld_site = site;
4017 site->ls_top_dev = &mdt->mdt_lu_dev;
4018 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4020 /* now connect to bottom OSD */
4021 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4022 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4024 GOTO(class_detach, rc);
4026 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4028 rc = lu_env_refill((struct lu_env *)env);
4030 CERROR("Failure to refill session: '%d'\n", rc);
4032 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4037 class_detach(obd, lcfg);
4039 lustre_cfg_free(lcfg);
4044 OBD_FREE(name, name_size);
4046 OBD_FREE(uuid, uuid_size);
4050 /* setup quota master target on MDT0 */
4051 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4052 struct lustre_cfg *cfg)
4054 struct obd_device *obd;
4055 char *dev = lustre_cfg_string(cfg, 0);
4056 char *qmtname, *uuid, *p;
4057 struct lustre_cfg_bufs *bufs;
4058 struct lustre_cfg *lcfg;
4059 struct lustre_profile *lprof;
4060 struct obd_connect_data *data;
4064 LASSERT(mdt->mdt_qmt_exp == NULL);
4065 LASSERT(mdt->mdt_qmt_dev == NULL);
4067 /* quota master is on MDT0 only for now */
4068 if (mdt->mdt_seq_site.ss_node_id != 0)
4071 /* MGS generates config commands which look as follows:
4072 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4073 * 3:lustre-MDT0000-mdtlov 4:f
4075 * We generate the QMT name from the MDT one, just replacing MD with QM
4076 * after all the preparations, the logical equivalent will be:
4077 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4078 * 3:lustre-MDT0000-osd 4:f */
4079 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4080 OBD_ALLOC(uuid, UUID_MAX);
4081 OBD_ALLOC_PTR(bufs);
4082 OBD_ALLOC_PTR(data);
4083 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4084 GOTO(cleanup_mem, rc = -ENOMEM);
4086 strcpy(qmtname, dev);
4087 p = strstr(qmtname, "-MDT");
4089 GOTO(cleanup_mem, rc = -ENOMEM);
4090 /* replace MD with QM */
4094 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4096 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4097 if (lprof == NULL || lprof->lp_dt == NULL) {
4098 CERROR("can't find profile for %s\n",
4099 lustre_cfg_string(cfg, 0));
4100 GOTO(cleanup_mem, rc = -EINVAL);
4103 lustre_cfg_bufs_reset(bufs, qmtname);
4104 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4105 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4106 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4108 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4110 GOTO(cleanup_mem, rc = -ENOMEM);
4112 rc = class_attach(lcfg);
4114 GOTO(lcfg_cleanup, rc);
4116 obd = class_name2obd(qmtname);
4118 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4119 lustre_cfg_string(cfg, 0));
4120 GOTO(class_detach, rc = -EINVAL);
4123 lustre_cfg_free(lcfg);
4125 lustre_cfg_bufs_reset(bufs, qmtname);
4126 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4127 lustre_cfg_bufs_set_string(bufs, 2, dev);
4129 /* for quota, the next device should be the OSD device */
4130 lustre_cfg_bufs_set_string(bufs, 3,
4131 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4133 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4135 rc = class_setup(obd, lcfg);
4137 GOTO(class_detach, rc);
4139 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4141 /* configure local quota objects */
4142 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4146 GOTO(class_cleanup, rc);
4148 /* connect to quota master target */
4149 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4150 data->ocd_version = LUSTRE_VERSION_CODE;
4151 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4154 CERROR("cannot connect to quota master device %s (%d)\n",
4156 GOTO(class_cleanup, rc);
4162 class_manual_cleanup(obd);
4163 mdt->mdt_qmt_dev = NULL;
4167 class_detach(obd, lcfg);
4169 lustre_cfg_free(lcfg);
4174 OBD_FREE(qmtname, MAX_OBD_NAME);
4176 OBD_FREE(uuid, UUID_MAX);
4182 /* Shutdown quota master target associated with mdt */
4183 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4187 if (mdt->mdt_qmt_exp == NULL)
4189 LASSERT(mdt->mdt_qmt_dev != NULL);
4191 /* the qmt automatically shuts down when the mdt disconnects */
4192 obd_disconnect(mdt->mdt_qmt_exp);
4193 mdt->mdt_qmt_exp = NULL;
4194 mdt->mdt_qmt_dev = NULL;
4198 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4199 * for now. This will be removed along with converting rest of MDT code
4200 * to use tgt_session_info */
4201 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4203 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4206 rc = mdt_getxattr(info);
4208 mdt_thread_info_fini(info);
4212 static struct tgt_handler mdt_tgt_handlers[] = {
4213 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4214 0, MDS_CONNECT, mdt_tgt_connect,
4215 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4216 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4217 0, MDS_DISCONNECT, tgt_disconnect,
4218 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4219 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4220 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4221 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4222 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4223 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus),
4224 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4225 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4227 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4228 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4229 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4230 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4231 TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING,
4233 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4234 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4235 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4236 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4238 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4239 mdt_hsm_ct_register),
4240 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4241 mdt_hsm_ct_unregister),
4242 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4244 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4246 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4247 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4249 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4254 static struct tgt_handler mdt_sec_ctx_ops[] = {
4255 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4256 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4257 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4260 static struct tgt_handler mdt_quota_ops[] = {
4261 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4264 static struct tgt_opc_slice mdt_common_slice[] = {
4266 .tos_opc_start = MDS_FIRST_OPC,
4267 .tos_opc_end = MDS_LAST_OPC,
4268 .tos_hs = mdt_tgt_handlers
4271 .tos_opc_start = OBD_FIRST_OPC,
4272 .tos_opc_end = OBD_LAST_OPC,
4273 .tos_hs = tgt_obd_handlers
4276 .tos_opc_start = LDLM_FIRST_OPC,
4277 .tos_opc_end = LDLM_LAST_OPC,
4278 .tos_hs = tgt_dlm_handlers
4281 .tos_opc_start = SEC_FIRST_OPC,
4282 .tos_opc_end = SEC_LAST_OPC,
4283 .tos_hs = mdt_sec_ctx_ops
4286 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4287 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4288 .tos_hs = tgt_out_handlers
4291 .tos_opc_start = FLD_FIRST_OPC,
4292 .tos_opc_end = FLD_LAST_OPC,
4293 .tos_hs = fld_handlers
4296 .tos_opc_start = SEQ_FIRST_OPC,
4297 .tos_opc_end = SEQ_LAST_OPC,
4298 .tos_hs = seq_handlers
4301 .tos_opc_start = QUOTA_DQACQ,
4302 .tos_opc_end = QUOTA_LAST_OPC,
4303 .tos_hs = mdt_quota_ops
4306 .tos_opc_start = LLOG_FIRST_OPC,
4307 .tos_opc_end = LLOG_LAST_OPC,
4308 .tos_hs = tgt_llog_handlers
4311 .tos_opc_start = LFSCK_FIRST_OPC,
4312 .tos_opc_end = LFSCK_LAST_OPC,
4313 .tos_hs = tgt_lfsck_handlers
4321 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4323 struct md_device *next = m->mdt_child;
4324 struct lu_device *d = &m->mdt_lu_dev;
4325 struct obd_device *obd = mdt2obd_dev(m);
4326 struct lfsck_stop stop;
4329 stop.ls_status = LS_PAUSED;
4331 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4333 target_recovery_fini(obd);
4334 ping_evictor_stop();
4335 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4337 if (m->mdt_opts.mo_coordinator)
4338 mdt_hsm_cdt_stop(m);
4340 mdt_hsm_cdt_fini(m);
4342 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4343 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4344 obd_exports_barrier(obd);
4345 obd_zombie_barrier();
4349 tgt_fini(env, &m->mdt_lut);
4350 mdt_fs_cleanup(env, m);
4351 upcall_cache_cleanup(m->mdt_identity_cache);
4352 m->mdt_identity_cache = NULL;
4354 if (m->mdt_namespace != NULL) {
4355 ldlm_namespace_free(m->mdt_namespace, NULL,
4356 d->ld_obd->obd_force);
4357 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4360 mdt_quota_fini(env, m);
4362 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4364 mdt_seq_fini(env, m);
4365 mdt_fld_fini(env, m);
4367 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4368 cfs_timer_disarm(&m->mdt_ck_timer);
4369 mdt_ck_thread_stop(m);
4374 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4376 LASSERT(atomic_read(&d->ld_ref) == 0);
4378 server_put_mount(mdt_obd_name(m), true);
4383 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4385 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4386 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4388 struct mdt_thread_info *info;
4389 struct obd_device *obd;
4390 const char *dev = lustre_cfg_string(cfg, 0);
4391 const char *num = lustre_cfg_string(cfg, 2);
4392 struct lustre_mount_info *lmi = NULL;
4393 struct lustre_sb_info *lsi;
4395 struct seq_server_site *ss_site;
4396 const char *identity_upcall = "NONE";
4397 struct md_device *next;
4403 lu_device_init(&m->mdt_lu_dev, ldt);
4405 * Environment (env) might be missing mdt_thread_key values at that
4406 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4409 * Usually device allocation path doesn't use module key values, but
4410 * mdt has to do a lot of work here, so allocate key value.
4412 rc = lu_env_refill((struct lu_env *)env);
4416 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4417 LASSERT(info != NULL);
4419 obd = class_name2obd(dev);
4420 LASSERT(obd != NULL);
4422 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4424 m->mdt_som_conf = 0;
4426 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4428 /* default is coordinator off, it is started through conf_param
4430 m->mdt_opts.mo_coordinator = 0;
4432 lmi = server_get_mount(dev);
4434 CERROR("Cannot get mount info for %s!\n", dev);
4437 lsi = s2lsi(lmi->lmi_sb);
4438 /* CMD is supported only in IAM mode */
4440 node_id = simple_strtol(num, NULL, 10);
4441 obd->u.obt.obt_magic = OBT_MAGIC;
4444 spin_lock_init(&m->mdt_ioepoch_lock);
4445 m->mdt_capa_timeout = CAPA_TIMEOUT;
4446 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4447 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4448 m->mdt_squash.rsi_uid = 0;
4449 m->mdt_squash.rsi_gid = 0;
4450 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4451 init_rwsem(&m->mdt_squash.rsi_sem);
4452 spin_lock_init(&m->mdt_osfs_lock);
4453 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4454 m->mdt_enable_remote_dir = 0;
4455 m->mdt_enable_remote_dir_gid = 0;
4457 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4458 m->mdt_lu_dev.ld_obd = obd;
4459 /* Set this lu_device to obd for error handling purposes. */
4460 obd->obd_lu_dev = &m->mdt_lu_dev;
4462 /* init the stack */
4463 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4465 CERROR("%s: Can't init device stack, rc %d\n",
4466 mdt_obd_name(m), rc);
4471 ss_site = mdt_seq_site(m);
4472 s->ld_seq_site = ss_site;
4475 /* set server index */
4476 ss_site->ss_node_id = node_id;
4478 /* failover is the default
4479 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4480 * assumed whose ss_node_id == 0 XXX
4482 obd->obd_replayable = 1;
4483 /* No connection accepted until configurations will finish */
4484 obd->obd_no_conn = 1;
4486 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4487 char *str = lustre_cfg_string(cfg, 4);
4488 if (strchr(str, 'n')) {
4489 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4490 obd->obd_replayable = 0;
4494 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4496 GOTO(err_fini_stack, rc);
4498 rc = mdt_seq_init(env, m);
4500 GOTO(err_fini_fld, rc);
4502 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4503 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4504 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4505 LDLM_NAMESPACE_SERVER,
4506 LDLM_NAMESPACE_GREEDY,
4508 if (m->mdt_namespace == NULL)
4509 GOTO(err_fini_seq, rc = -ENOMEM);
4511 m->mdt_namespace->ns_lvbp = m;
4512 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4514 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4515 /* set obd_namespace for compatibility with old code */
4516 obd->obd_namespace = m->mdt_namespace;
4518 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4520 rc = mdt_hsm_cdt_init(m);
4522 CERROR("%s: error initializing coordinator, rc %d\n",
4523 mdt_obd_name(m), rc);
4524 GOTO(err_free_ns, rc);
4527 rc = mdt_ck_thread_start(m);
4529 GOTO(err_free_hsm, rc);
4531 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4532 OBD_FAIL_MDS_ALL_REQUEST_NET,
4533 OBD_FAIL_MDS_ALL_REPLY_NET);
4537 rc = mdt_fs_setup(env, m, obd, lsi);
4541 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4543 next = m->mdt_child;
4544 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4547 GOTO(err_fs_cleanup, rc);
4549 if (mntopts & MNTOPT_USERXATTR)
4550 m->mdt_opts.mo_user_xattr = 1;
4552 m->mdt_opts.mo_user_xattr = 0;
4554 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4556 GOTO(err_fs_cleanup, rc);
4558 if (mntopts & MNTOPT_ACL)
4559 m->mdt_opts.mo_acl = 1;
4561 m->mdt_opts.mo_acl = 0;
4563 /* XXX: to support suppgid for ACL, we enable identity_upcall
4564 * by default, otherwise, maybe got unexpected -EACCESS. */
4565 if (m->mdt_opts.mo_acl)
4566 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4568 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4570 &mdt_identity_upcall_cache_ops);
4571 if (IS_ERR(m->mdt_identity_cache)) {
4572 rc = PTR_ERR(m->mdt_identity_cache);
4573 m->mdt_identity_cache = NULL;
4574 GOTO(err_fs_cleanup, rc);
4577 rc = mdt_procfs_init(m, dev);
4579 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4580 GOTO(err_recovery, rc);
4583 rc = mdt_quota_init(env, m, cfg);
4585 GOTO(err_procfs, rc);
4587 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4588 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4589 "mdt_ldlm_client", m->mdt_ldlm_client);
4591 ping_evictor_start();
4593 /* recovery will be started upon mdt_prepare()
4594 * when the whole stack is complete and ready
4595 * to serve the requests */
4597 mdt_init_capa_ctxt(env, m);
4599 /* Reduce the initial timeout on an MDS because it doesn't need such
4600 * a long timeout as an OST does. Adaptive timeouts will adjust this
4601 * value appropriately. */
4602 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4603 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4609 target_recovery_fini(obd);
4610 upcall_cache_cleanup(m->mdt_identity_cache);
4611 m->mdt_identity_cache = NULL;
4613 mdt_fs_cleanup(env, m);
4615 tgt_fini(env, &m->mdt_lut);
4617 cfs_timer_disarm(&m->mdt_ck_timer);
4618 mdt_ck_thread_stop(m);
4620 mdt_hsm_cdt_fini(m);
4622 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4623 obd->obd_namespace = m->mdt_namespace = NULL;
4625 mdt_seq_fini(env, m);
4627 mdt_fld_fini(env, m);
4629 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4632 server_put_mount(dev, true);
4636 /* For interoperability, the left element is old parameter, the right one
4637 * is the new version of the parameter, if some parameter is deprecated,
4638 * the new version should be set as NULL. */
4639 static struct cfg_interop_param mdt_interop_param[] = {
4640 { "mdt.group_upcall", NULL },
4641 { "mdt.quota_type", NULL },
4642 { "mdd.quota_type", NULL },
4643 { "mdt.rootsquash", "mdt.root_squash" },
4644 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4648 /* used by MGS to process specific configurations */
4649 static int mdt_process_config(const struct lu_env *env,
4650 struct lu_device *d, struct lustre_cfg *cfg)
4652 struct mdt_device *m = mdt_dev(d);
4653 struct md_device *md_next = m->mdt_child;
4654 struct lu_device *next = md2lu_dev(md_next);
4658 switch (cfg->lcfg_command) {
4660 struct obd_device *obd = d->ld_obd;
4662 /* For interoperability */
4663 struct cfg_interop_param *ptr = NULL;
4664 struct lustre_cfg *old_cfg = NULL;
4667 param = lustre_cfg_string(cfg, 1);
4668 if (param == NULL) {
4669 CERROR("param is empty\n");
4674 ptr = class_find_old_param(param, mdt_interop_param);
4676 if (ptr->new_param == NULL) {
4678 CWARN("For interoperability, skip this %s."
4679 " It is obsolete.\n", ptr->old_param);
4683 CWARN("Found old param %s, changed it to %s.\n",
4684 ptr->old_param, ptr->new_param);
4687 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4694 rc = class_process_proc_seq_param(PARAM_MDT, obd->obd_vars,
4696 if (rc > 0 || rc == -ENOSYS) {
4697 /* is it an HSM var ? */
4698 rc = class_process_proc_seq_param(PARAM_HSM,
4699 hsm_cdt_get_proc_vars(),
4701 if (rc > 0 || rc == -ENOSYS)
4702 /* we don't understand; pass it on */
4703 rc = next->ld_ops->ldo_process_config(env, next,
4707 if (old_cfg != NULL)
4708 lustre_cfg_free(cfg);
4713 /* others are passed further */
4714 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4720 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4721 const struct lu_object_header *hdr,
4722 struct lu_device *d)
4724 struct mdt_object *mo;
4728 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4730 struct lu_object *o;
4731 struct lu_object_header *h;
4734 h = &mo->mot_header;
4735 lu_object_header_init(h);
4736 lu_object_init(o, h, d);
4737 lu_object_add_top(h, o);
4738 o->lo_ops = &mdt_obj_ops;
4739 mutex_init(&mo->mot_ioepoch_mutex);
4740 mutex_init(&mo->mot_lov_mutex);
4741 init_rwsem(&mo->mot_open_sem);
4747 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4748 const struct lu_object_conf *unused)
4750 struct mdt_device *d = mdt_dev(o->lo_dev);
4751 struct lu_device *under;
4752 struct lu_object *below;
4756 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4757 PFID(lu_object_fid(o)));
4759 under = &d->mdt_child->md_lu_dev;
4760 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4761 if (below != NULL) {
4762 lu_object_add(o, below);
4769 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4771 struct mdt_object *mo = mdt_obj(o);
4772 struct lu_object_header *h;
4776 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4777 PFID(lu_object_fid(o)));
4779 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4780 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4783 lu_object_header_fini(h);
4784 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4789 static int mdt_object_print(const struct lu_env *env, void *cookie,
4790 lu_printer_t p, const struct lu_object *o)
4792 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4793 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4794 "flags="LPX64", epochcount=%d, writecount=%d)",
4795 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4796 mdto->mot_ioepoch_count, mdto->mot_writecount);
4799 static int mdt_prepare(const struct lu_env *env,
4800 struct lu_device *pdev,
4801 struct lu_device *cdev)
4803 struct mdt_device *mdt = mdt_dev(cdev);
4804 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4805 struct obd_device *obd = cdev->ld_obd;
4812 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4816 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4820 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4824 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4825 /* The LFSCK instance is registered just now, so it must be there when
4826 * register the namespace to such instance. */
4827 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4829 if (mdt->mdt_seq_site.ss_node_id == 0) {
4830 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4831 &mdt->mdt_md_root_fid);
4836 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4837 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4838 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4839 LASSERT(obd->obd_no_conn);
4840 spin_lock(&obd->obd_dev_lock);
4841 obd->obd_no_conn = 0;
4842 spin_unlock(&obd->obd_dev_lock);
4844 if (obd->obd_recovering == 0)
4845 mdt_postrecov(env, mdt);
4850 const struct lu_device_operations mdt_lu_ops = {
4851 .ldo_object_alloc = mdt_object_alloc,
4852 .ldo_process_config = mdt_process_config,
4853 .ldo_prepare = mdt_prepare,
4856 static const struct lu_object_operations mdt_obj_ops = {
4857 .loo_object_init = mdt_object_init,
4858 .loo_object_free = mdt_object_free,
4859 .loo_object_print = mdt_object_print
4862 static int mdt_obd_set_info_async(const struct lu_env *env,
4863 struct obd_export *exp,
4864 __u32 keylen, void *key,
4865 __u32 vallen, void *val,
4866 struct ptlrpc_request_set *set)
4872 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4873 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4881 * Match client and server connection feature flags.
4883 * Compute the compatibility flags for a connection request based on
4884 * features mutually supported by client and server.
4886 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4887 * must not be updated here, otherwise a partially initialized value may
4888 * be exposed. After the connection request is successfully processed,
4889 * the top-level MDT connect request handler atomically updates the export
4890 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4891 * reply. \see mdt_connect().
4893 * \param exp the obd_export associated with this client/target pair
4894 * \param mdt the target device for the connection
4895 * \param data stores data for this connect request
4898 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4899 * \retval -EBADE client and server feature requirements are incompatible
4901 static int mdt_connect_internal(struct obd_export *exp,
4902 struct mdt_device *mdt,
4903 struct obd_connect_data *data)
4905 LASSERT(data != NULL);
4907 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4908 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4910 /* If no known bits (which should not happen, probably,
4911 as everybody should support LOOKUP and UPDATE bits at least)
4912 revert to compat mode with plain locks. */
4913 if (!data->ocd_ibits_known &&
4914 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4915 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4917 if (!mdt->mdt_opts.mo_acl)
4918 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4920 if (!mdt->mdt_opts.mo_user_xattr)
4921 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4923 if (!mdt->mdt_som_conf)
4924 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4926 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4927 data->ocd_brw_size = min(data->ocd_brw_size,
4928 (__u32)MD_MAX_BRW_SIZE);
4929 if (data->ocd_brw_size == 0) {
4930 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4931 " ocd_version: %x ocd_grant: %d "
4932 "ocd_index: %u ocd_brw_size is "
4933 "unexpectedly zero, network data "
4934 "corruption? Refusing connection of this"
4937 exp->exp_client_uuid.uuid,
4938 exp, data->ocd_connect_flags, data->ocd_version,
4939 data->ocd_grant, data->ocd_index);
4944 /* NB: Disregard the rule against updating
4945 * exp_connect_data.ocd_connect_flags in this case, since
4946 * tgt_client_new() needs to know if this is a lightweight
4947 * connection, and it is safe to expose this flag before
4948 * connection processing completes. */
4949 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4950 spin_lock(&exp->exp_lock);
4951 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4952 spin_unlock(&exp->exp_lock);
4955 data->ocd_version = LUSTRE_VERSION_CODE;
4957 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4958 CWARN("%s: MDS requires FID support, but client not\n",
4963 if (mdt->mdt_som_conf &&
4964 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4965 OBD_CONNECT_MDS_MDS |
4966 OBD_CONNECT_SOM))) {
4967 CWARN("%s: MDS has SOM enabled, but client does not support "
4968 "it\n", mdt_obd_name(mdt));
4972 if (OCD_HAS_FLAG(data, PINGLESS)) {
4973 if (ptlrpc_pinger_suppress_pings()) {
4974 spin_lock(&exp->exp_obd->obd_dev_lock);
4975 list_del_init(&exp->exp_obd_chain_timed);
4976 spin_unlock(&exp->exp_obd->obd_dev_lock);
4978 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4982 data->ocd_max_easize = mdt->mdt_max_ea_size;
4987 /* mds_connect copy */
4988 static int mdt_obd_connect(const struct lu_env *env,
4989 struct obd_export **exp, struct obd_device *obd,
4990 struct obd_uuid *cluuid,
4991 struct obd_connect_data *data,
4994 struct obd_export *lexp;
4995 struct lustre_handle conn = { 0 };
4996 struct mdt_device *mdt;
5000 LASSERT(env != NULL);
5001 if (!exp || !obd || !cluuid)
5004 mdt = mdt_dev(obd->obd_lu_dev);
5007 * first, check whether the stack is ready to handle requests
5008 * XXX: probably not very appropriate method is used now
5009 * at some point we should find a better one
5011 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5012 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5013 rc = obd_get_info(env, mdt->mdt_child_exp,
5014 sizeof(KEY_OSP_CONNECTED),
5015 KEY_OSP_CONNECTED, NULL, NULL, NULL);
5018 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5021 rc = class_connect(&conn, obd, cluuid);
5025 lexp = class_conn2export(&conn);
5026 LASSERT(lexp != NULL);
5028 rc = mdt_connect_internal(lexp, mdt, data);
5030 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5033 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5034 rc = tgt_client_new(env, lexp);
5036 mdt_export_stats_init(obd, lexp, localdata);
5038 /* For phase I, sync for cross-ref operation. */
5039 lexp->exp_keep_sync = 1;
5043 class_disconnect(lexp);
5052 static int mdt_obd_reconnect(const struct lu_env *env,
5053 struct obd_export *exp, struct obd_device *obd,
5054 struct obd_uuid *cluuid,
5055 struct obd_connect_data *data,
5061 if (exp == NULL || obd == NULL || cluuid == NULL)
5064 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5066 mdt_export_stats_init(obd, exp, localdata);
5071 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5072 struct mdt_thread_info *info,
5073 struct mdt_file_data *mfd)
5075 struct lu_context ses;
5079 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5084 lu_context_enter(&ses);
5086 mdt_ucred(info)->uc_valid = UCRED_OLD;
5087 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5089 lu_context_exit(&ses);
5090 lu_context_fini(&ses);
5096 static int mdt_export_cleanup(struct obd_export *exp)
5098 struct mdt_export_data *med = &exp->exp_mdt_data;
5099 struct obd_device *obd = exp->exp_obd;
5100 struct mdt_device *mdt;
5101 struct mdt_thread_info *info;
5103 CFS_LIST_HEAD(closing_list);
5104 struct mdt_file_data *mfd, *n;
5108 spin_lock(&med->med_open_lock);
5109 while (!cfs_list_empty(&med->med_open_head)) {
5110 cfs_list_t *tmp = med->med_open_head.next;
5111 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5113 /* Remove mfd handle so it can't be found again.
5114 * We are consuming the mfd_list reference here. */
5115 class_handle_unhash(&mfd->mfd_handle);
5116 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5118 spin_unlock(&med->med_open_lock);
5119 mdt = mdt_dev(obd->obd_lu_dev);
5120 LASSERT(mdt != NULL);
5122 rc = lu_env_init(&env, LCT_MD_THREAD);
5126 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5127 LASSERT(info != NULL);
5128 memset(info, 0, sizeof *info);
5129 info->mti_env = &env;
5130 info->mti_mdt = mdt;
5131 info->mti_exp = exp;
5133 if (!cfs_list_empty(&closing_list)) {
5134 struct md_attr *ma = &info->mti_attr;
5136 /* Close any open files (which may also cause orphan unlinking). */
5137 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5138 cfs_list_del_init(&mfd->mfd_list);
5139 ma->ma_need = ma->ma_valid = 0;
5141 /* This file is being closed due to an eviction, it
5142 * could have been modified and now dirty regarding to
5143 * HSM archive, check this!
5144 * The logic here is to mark a file dirty if there's a
5145 * chance it was dirtied before the client was evicted,
5146 * so that we don't have to wait for a release attempt
5147 * before finding out the file was actually dirty and
5148 * fail the release. Aggressively marking it dirty here
5149 * will cause the policy engine to attempt to
5150 * re-archive it; when rearchiving, we can compare the
5151 * current version to the HSM data_version and make the
5152 * archive request into a noop if it's not actually
5155 if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5156 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5158 /* Don't unlink orphan on failover umount, LU-184 */
5159 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5160 ma->ma_valid = MA_FLAGS;
5161 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5163 mdt_mfd_close(info, mfd);
5166 info->mti_mdt = NULL;
5167 /* cleanup client slot early */
5168 /* Do not erase record for recoverable client. */
5169 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5170 tgt_client_del(&env, exp);
5176 static int mdt_obd_disconnect(struct obd_export *exp)
5182 class_export_get(exp);
5184 rc = server_disconnect_export(exp);
5186 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5188 rc = mdt_export_cleanup(exp);
5189 class_export_put(exp);
5193 /* FIXME: Can we avoid using these two interfaces? */
5194 static int mdt_init_export(struct obd_export *exp)
5196 struct mdt_export_data *med = &exp->exp_mdt_data;
5200 CFS_INIT_LIST_HEAD(&med->med_open_head);
5201 spin_lock_init(&med->med_open_lock);
5202 mutex_init(&med->med_idmap_mutex);
5203 med->med_idmap = NULL;
5204 spin_lock(&exp->exp_lock);
5205 exp->exp_connecting = 1;
5206 spin_unlock(&exp->exp_lock);
5208 /* self-export doesn't need client data and ldlm initialization */
5209 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5210 &exp->exp_client_uuid)))
5213 rc = tgt_client_alloc(exp);
5217 rc = ldlm_init_export(exp);
5224 tgt_client_free(exp);
5226 CERROR("%s: Failed to initialize export: rc = %d\n",
5227 exp->exp_obd->obd_name, rc);
5231 static int mdt_destroy_export(struct obd_export *exp)
5235 if (exp_connect_rmtclient(exp))
5236 mdt_cleanup_idmap(&exp->exp_mdt_data);
5238 target_destroy_export(exp);
5239 /* destroy can be called from failed obd_setup, so
5240 * checking uuid is safer than obd_self_export */
5241 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5242 &exp->exp_client_uuid)))
5245 ldlm_destroy_export(exp);
5246 tgt_client_free(exp);
5248 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5249 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5254 /** The maximum depth that fid2path() will search.
5255 * This is limited only because we want to store the fids for
5256 * historical path lookup purposes.
5258 #define MAX_PATH_DEPTH 100
5260 /** mdt_path() lookup structure. */
5261 struct path_lookup_info {
5262 __u64 pli_recno; /**< history point */
5263 __u64 pli_currec; /**< current record */
5264 struct lu_fid pli_fid;
5265 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5266 struct mdt_object *pli_mdt_obj;
5267 char *pli_path; /**< full path */
5269 int pli_linkno; /**< which hardlink to follow */
5270 int pli_fidcount; /**< number of \a pli_fids */
5273 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5274 struct linkea_data *ldata)
5278 LASSERT(ldata->ld_buf->lb_buf != NULL);
5280 if (!mdt_object_exists(mdt_obj))
5283 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5284 ldata->ld_buf, XATTR_NAME_LINK);
5285 if (rc == -ERANGE) {
5286 /* Buf was too small, figure out what we need. */
5287 lu_buf_free(ldata->ld_buf);
5288 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5289 ldata->ld_buf, XATTR_NAME_LINK);
5292 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5293 if (ldata->ld_buf->lb_buf == NULL)
5295 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5296 ldata->ld_buf, XATTR_NAME_LINK);
5301 return linkea_init(ldata);
5304 static int mdt_path_current(struct mdt_thread_info *info,
5305 struct path_lookup_info *pli)
5307 struct mdt_device *mdt = info->mti_mdt;
5308 struct mdt_object *mdt_obj;
5309 struct link_ea_header *leh;
5310 struct link_ea_entry *lee;
5311 struct lu_name *tmpname = &info->mti_name;
5312 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5313 struct lu_buf *buf = &info->mti_big_buf;
5316 struct linkea_data ldata = { 0 };
5320 /* temp buffer for path element, the buffer will be finally freed
5321 * in mdt_thread_info_fini */
5322 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5323 if (buf->lb_buf == NULL)
5327 ptr = pli->pli_path + pli->pli_pathlen - 1;
5330 pli->pli_fidcount = 0;
5331 pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5332 *tmpfid = pli->pli_fids[0];
5333 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5334 * so checking root_fid can only happen on MDT0. */
5335 while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5336 &pli->pli_fids[pli->pli_fidcount])) {
5337 struct lu_buf lmv_buf;
5339 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5340 if (IS_ERR(mdt_obj))
5341 GOTO(out, rc = PTR_ERR(mdt_obj));
5343 if (mdt_object_remote(mdt_obj)) {
5344 mdt_object_put(info->mti_env, mdt_obj);
5345 GOTO(remote_out, rc = -EREMOTE);
5348 lmv_buf.lb_buf = info->mti_xattr_buf;
5349 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5351 /* Check if it is slave stripes */
5352 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5353 &lmv_buf, XATTR_NAME_LMV);
5355 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5357 /* For slave stripes, get its master */
5358 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5359 struct lmv_mds_md_v1 *lmm1 = &lmm->lmv_md_v1;
5361 fid_le_to_cpu(tmpfid, &lmm1->lmv_master_fid);
5362 if (!fid_is_sane(tmpfid)) {
5363 mdt_object_put(info->mti_env, mdt_obj);
5364 GOTO(out, rc = -EINVAL);
5366 mdt_object_put(info->mti_env, mdt_obj);
5367 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5372 if (!mdt_object_exists(mdt_obj)) {
5373 mdt_object_put(info->mti_env, mdt_obj);
5374 GOTO(out, rc = -ENOENT);
5377 rc = mdt_links_read(info, mdt_obj, &ldata);
5378 mdt_object_put(info->mti_env, mdt_obj);
5383 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5384 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5385 /* If set, use link #linkno for path lookup, otherwise use
5386 link #0. Only do this for the final path element. */
5387 if (pli->pli_fidcount == 0 &&
5388 pli->pli_linkno < leh->leh_reccount) {
5390 for (count = 0; count < pli->pli_linkno; count++) {
5391 lee = (struct link_ea_entry *)
5392 ((char *)lee + reclen);
5393 linkea_entry_unpack(lee, &reclen, tmpname,
5396 if (pli->pli_linkno < leh->leh_reccount - 1)
5397 /* indicate to user there are more links */
5401 /* Pack the name in the end of the buffer */
5402 ptr -= tmpname->ln_namelen;
5403 if (ptr - 1 <= pli->pli_path)
5404 GOTO(out, rc = -EOVERFLOW);
5405 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5408 /* Store the parent fid for historic lookup */
5409 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5410 GOTO(out, rc = -EOVERFLOW);
5411 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5415 ptr++; /* skip leading / */
5416 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5423 /* Returns the full path to this fid, as of changelog record recno. */
5424 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5425 char *path, int pathlen, __u64 *recno, int *linkno,
5428 struct mdt_device *mdt = info->mti_mdt;
5429 struct path_lookup_info *pli;
5437 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5446 pli->pli_mdt_obj = obj;
5447 pli->pli_recno = *recno;
5448 pli->pli_path = path;
5449 pli->pli_pathlen = pathlen;
5450 pli->pli_linkno = *linkno;
5452 /* Retry multiple times in case file is being moved */
5453 while (tries-- && rc == -EAGAIN)
5454 rc = mdt_path_current(info, pli);
5456 /* return the last resolved fids to the client, so the client will
5457 * build the left path on another MDT for remote object */
5458 *fid = pli->pli_fids[pli->pli_fidcount];
5460 *recno = pli->pli_currec;
5461 /* Return next link index to caller */
5462 *linkno = pli->pli_linkno;
5469 static int mdt_fid2path(struct mdt_thread_info *info,
5470 struct getinfo_fid2path *fp)
5472 struct mdt_device *mdt = info->mti_mdt;
5473 struct mdt_object *obj;
5477 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5478 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5480 if (!fid_is_sane(&fp->gf_fid))
5483 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5484 CWARN("%s: "DFID" is invalid, sequence should be "
5485 ">= "LPX64"\n", mdt_obd_name(mdt),
5486 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5490 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5491 if (obj == NULL || IS_ERR(obj)) {
5492 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5497 if (mdt_object_remote(obj))
5499 else if (!mdt_object_exists(obj))
5505 mdt_object_put(info->mti_env, obj);
5506 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5507 PFID(&fp->gf_fid), rc);
5511 rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5512 &fp->gf_linkno, &fp->gf_fid);
5514 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5515 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5517 mdt_object_put(info->mti_env, obj);
5522 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5523 void *val, int vallen)
5525 struct getinfo_fid2path *fpout, *fpin;
5528 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5531 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5532 lustre_swab_fid2path(fpin);
5534 memcpy(fpout, fpin, sizeof(*fpin));
5535 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5538 rc = mdt_fid2path(info, fpout);
5542 int mdt_get_info(struct tgt_session_info *tsi)
5552 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5554 CDEBUG(D_IOCTL, "No GETINFO key");
5555 RETURN(err_serious(-EFAULT));
5557 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5560 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5561 if (vallen == NULL) {
5562 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5563 RETURN(err_serious(-EFAULT));
5566 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5568 rc = req_capsule_server_pack(tsi->tsi_pill);
5570 RETURN(err_serious(rc));
5572 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5573 if (valout == NULL) {
5574 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5575 RETURN(err_serious(-EFAULT));
5578 if (KEY_IS(KEY_FID2PATH)) {
5579 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5581 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5582 mdt_thread_info_fini(info);
5589 /* Pass the ioc down */
5590 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5591 unsigned int cmd, int len, void *data)
5593 struct lu_context ioctl_session;
5594 struct md_device *next = mdt->mdt_child;
5598 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5601 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5602 lu_context_enter(&ioctl_session);
5603 env->le_ses = &ioctl_session;
5605 LASSERT(next->md_ops->mdo_iocontrol);
5606 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5608 lu_context_exit(&ioctl_session);
5609 lu_context_fini(&ioctl_session);
5613 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5615 struct obd_ioctl_data *data = karg;
5618 struct mdt_object *obj;
5619 struct mdt_lock_handle *lh;
5623 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5624 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5627 fid = (struct lu_fid *)data->ioc_inlbuf1;
5629 if (!fid_is_sane(fid))
5632 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5634 lh = &mti->mti_lh[MDT_LH_PARENT];
5635 mdt_lock_reg_init(lh, LCK_CR);
5637 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5639 RETURN(PTR_ERR(obj));
5641 if (mdt_object_remote(obj)) {
5644 * before calling version get the correct MDS should be
5645 * fid, this is error to find remote object here
5647 CERROR("nonlocal object "DFID"\n", PFID(fid));
5648 } else if (!mdt_object_exists(obj)) {
5649 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5652 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5653 *(__u64 *)data->ioc_inlbuf2 = version;
5656 mdt_object_unlock_put(mti, obj, lh, 1);
5660 /* ioctls on obd dev */
5661 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5662 void *karg, void *uarg)
5665 struct obd_device *obd = exp->exp_obd;
5666 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5667 struct dt_device *dt = mdt->mdt_bottom;
5671 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5672 rc = lu_env_init(&env, LCT_MD_THREAD);
5678 rc = mdt_device_sync(&env, mdt);
5680 case OBD_IOC_SET_READONLY:
5681 rc = dt->dd_ops->dt_ro(&env, dt);
5683 case OBD_IOC_ABORT_RECOVERY:
5684 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5685 target_stop_recovery_thread(obd);
5688 case OBD_IOC_CHANGELOG_REG:
5689 case OBD_IOC_CHANGELOG_DEREG:
5690 case OBD_IOC_CHANGELOG_CLEAR:
5691 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5693 case OBD_IOC_START_LFSCK: {
5694 struct md_device *next = mdt->mdt_child;
5695 struct obd_ioctl_data *data = karg;
5696 struct lfsck_start_param lsp;
5698 if (unlikely(data == NULL)) {
5703 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5704 lsp.lsp_index_valid = 0;
5705 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5708 case OBD_IOC_STOP_LFSCK: {
5709 struct md_device *next = mdt->mdt_child;
5710 struct obd_ioctl_data *data = karg;
5711 struct lfsck_stop stop;
5713 stop.ls_status = LS_STOPPED;
5714 /* Old lfsck utils may pass NULL @stop. */
5715 if (data->ioc_inlbuf1 == NULL)
5719 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5721 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5724 case OBD_IOC_GET_OBJ_VERSION: {
5725 struct mdt_thread_info *mti;
5726 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5727 memset(mti, 0, sizeof *mti);
5728 mti->mti_env = &env;
5732 rc = mdt_ioc_version_get(mti, karg);
5735 case OBD_IOC_CATLOGLIST: {
5736 struct mdt_thread_info *mti;
5738 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5739 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5740 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5741 &mti->mti_tmp_fid1);
5746 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5747 mdt_obd_name(mdt), cmd, rc);
5754 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5756 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5757 struct lfsck_start_param lsp;
5761 lsp.lsp_start = NULL;
5762 lsp.lsp_index_valid = 0;
5763 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5764 OBD_IOC_START_LFSCK,
5766 if (rc != 0 && rc != -EALREADY)
5767 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5768 mdt_obd_name(mdt), rc);
5770 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5774 static int mdt_obd_postrecov(struct obd_device *obd)
5779 rc = lu_env_init(&env, LCT_MD_THREAD);
5782 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5787 static struct obd_ops mdt_obd_device_ops = {
5788 .o_owner = THIS_MODULE,
5789 .o_set_info_async = mdt_obd_set_info_async,
5790 .o_connect = mdt_obd_connect,
5791 .o_reconnect = mdt_obd_reconnect,
5792 .o_disconnect = mdt_obd_disconnect,
5793 .o_init_export = mdt_init_export,
5794 .o_destroy_export = mdt_destroy_export,
5795 .o_iocontrol = mdt_iocontrol,
5796 .o_postrecov = mdt_obd_postrecov,
5799 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5800 struct lu_device *d)
5802 struct mdt_device *m = mdt_dev(d);
5809 static struct lu_device *mdt_device_free(const struct lu_env *env,
5810 struct lu_device *d)
5812 struct mdt_device *m = mdt_dev(d);
5815 lu_device_fini(&m->mdt_lu_dev);
5821 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5822 struct lu_device_type *t,
5823 struct lustre_cfg *cfg)
5825 struct lu_device *l;
5826 struct mdt_device *m;
5833 rc = mdt_init0(env, m, t, cfg);
5835 mdt_device_free(env, l);
5840 l = ERR_PTR(-ENOMEM);
5844 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5845 LU_KEY_INIT(mdt, struct mdt_thread_info);
5847 static void mdt_key_fini(const struct lu_context *ctx,
5848 struct lu_context_key *key, void* data)
5850 struct mdt_thread_info *info = data;
5852 if (info->mti_big_lmm) {
5853 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5854 info->mti_big_lmm = NULL;
5855 info->mti_big_lmmsize = 0;
5860 /* context key: mdt_thread_key */
5861 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5863 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5865 return lu_ucred(info->mti_env);
5868 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5870 return lu_ucred_check(info->mti_env);
5874 * Enable/disable COS (Commit On Sharing).
5876 * Set/Clear the COS flag in mdt options.
5878 * \param mdt mdt device
5879 * \param val 0 disables COS, other values enable COS
5881 void mdt_enable_cos(struct mdt_device *mdt, int val)
5886 mdt->mdt_opts.mo_cos = !!val;
5887 rc = lu_env_init(&env, LCT_LOCAL);
5888 if (unlikely(rc != 0)) {
5889 CWARN("lu_env initialization failed with rc = %d,"
5890 "cannot sync\n", rc);
5893 mdt_device_sync(&env, mdt);
5898 * Check COS (Commit On Sharing) status.
5900 * Return COS flag status.
5902 * \param mdt mdt device
5904 int mdt_cos_is_enabled(struct mdt_device *mdt)
5906 return mdt->mdt_opts.mo_cos != 0;
5909 static struct lu_device_type_operations mdt_device_type_ops = {
5910 .ldto_device_alloc = mdt_device_alloc,
5911 .ldto_device_free = mdt_device_free,
5912 .ldto_device_fini = mdt_device_fini
5915 static struct lu_device_type mdt_device_type = {
5916 .ldt_tags = LU_DEVICE_MD,
5917 .ldt_name = LUSTRE_MDT_NAME,
5918 .ldt_ops = &mdt_device_type_ops,
5919 .ldt_ctx_tags = LCT_MD_THREAD
5922 static int __init mdt_mod_init(void)
5926 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5927 FID_NOBRACE_LEN + 1);
5928 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5930 rc = lu_kmem_init(mdt_caches);
5934 rc = mds_mod_init();
5938 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5939 #ifndef HAVE_ONLY_PROCFS_SEQ
5942 LUSTRE_MDT_NAME, &mdt_device_type);
5947 lu_kmem_fini(mdt_caches);
5954 static void __exit mdt_mod_exit(void)
5956 class_unregister_type(LUSTRE_MDT_NAME);
5958 lu_kmem_fini(mdt_caches);
5961 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5962 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5963 MODULE_LICENSE("GPL");
5965 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);