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>
72 #include <lustre_nodemap.h>
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75 [LCK_MINMODE] = MDL_MINMODE,
82 [LCK_GROUP] = MDL_GROUP
85 ldlm_mode_t mdt_dlm_lock_modes[] = {
86 [MDL_MINMODE] = LCK_MINMODE,
93 [MDL_GROUP] = LCK_GROUP
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
99 static const struct lu_object_operations mdt_obj_ops;
101 /* Slab for MDT object allocation */
102 static struct kmem_cache *mdt_object_kmem;
104 /* For HSM restore handles */
105 struct kmem_cache *mdt_hsm_cdt_kmem;
107 /* For HSM request handles */
108 struct kmem_cache *mdt_hsm_car_kmem;
110 static struct lu_kmem_descr mdt_caches[] = {
112 .ckd_cache = &mdt_object_kmem,
113 .ckd_name = "mdt_obj",
114 .ckd_size = sizeof(struct mdt_object)
117 .ckd_cache = &mdt_hsm_cdt_kmem,
118 .ckd_name = "mdt_cdt_restore_handle",
119 .ckd_size = sizeof(struct cdt_restore_handle)
122 .ckd_cache = &mdt_hsm_car_kmem,
123 .ckd_name = "mdt_cdt_agent_req",
124 .ckd_size = sizeof(struct cdt_agent_req)
131 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
135 return (rep->lock_policy_res1 & flag);
138 void mdt_clear_disposition(struct mdt_thread_info *info,
139 struct ldlm_reply *rep, int flag)
142 info->mti_opdata &= ~flag;
143 tgt_opdata_clear(info->mti_env, flag);
146 rep->lock_policy_res1 &= ~flag;
149 void mdt_set_disposition(struct mdt_thread_info *info,
150 struct ldlm_reply *rep, int flag)
153 info->mti_opdata |= flag;
154 tgt_opdata_set(info->mti_env, flag);
157 rep->lock_policy_res1 |= flag;
160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
162 lh->mlh_pdo_hash = 0;
163 lh->mlh_reg_mode = lm;
164 lh->mlh_rreg_mode = lm;
165 lh->mlh_type = MDT_REG_LOCK;
168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
169 const struct lu_name *lname)
171 lh->mlh_reg_mode = lock_mode;
172 lh->mlh_rreg_mode = lock_mode;
173 lh->mlh_type = MDT_PDO_LOCK;
175 if (lu_name_is_valid(lname)) {
176 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
178 /* XXX Workaround for LU-2856
180 * Zero is a valid return value of full_name_hash, but
181 * several users of mlh_pdo_hash assume a non-zero
182 * hash value. We therefore map zero onto an
183 * arbitrary, but consistent value (1) to avoid
184 * problems further down the road. */
185 if (unlikely(lh->mlh_pdo_hash == 0))
186 lh->mlh_pdo_hash = 1;
188 lh->mlh_pdo_hash = 0;
192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
193 struct mdt_lock_handle *lh)
199 * Any dir access needs couple of locks:
201 * 1) on part of dir we gonna take lookup/modify;
203 * 2) on whole dir to protect it from concurrent splitting and/or to
204 * flush client's cache for readdir().
206 * so, for a given mode and object this routine decides what lock mode
207 * to use for lock #2:
209 * 1) if caller's gonna lookup in dir then we need to protect dir from
210 * being splitted only - LCK_CR
212 * 2) if caller's gonna modify dir then we need to protect dir from
213 * being splitted and to flush cache - LCK_CW
215 * 3) if caller's gonna modify dir and that dir seems ready for
216 * splitting then we need to protect it from any type of access
217 * (lookup/modify/split) - LCK_EX --bzzz
220 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
221 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
224 * Ask underlaying level its opinion about preferable PDO lock mode
225 * having access type passed as regular lock mode:
227 * - MDL_MINMODE means that lower layer does not want to specify lock
230 * - MDL_NL means that no PDO lock should be taken. This is used in some
231 * cases. Say, for non-splittable directories no need to use PDO locks
234 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
235 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
237 if (mode != MDL_MINMODE) {
238 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
241 * Lower layer does not want to specify locking mode. We do it
242 * our selves. No special protection is needed, just flush
243 * client's cache on modification and allow concurrent
246 switch (lh->mlh_reg_mode) {
248 lh->mlh_pdo_mode = LCK_EX;
251 lh->mlh_pdo_mode = LCK_CR;
254 lh->mlh_pdo_mode = LCK_CW;
257 CERROR("Not expected lock type (0x%x)\n",
258 (int)lh->mlh_reg_mode);
263 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
267 static int mdt_getstatus(struct tgt_session_info *tsi)
269 struct mdt_thread_info *info = tsi2mdt_info(tsi);
270 struct mdt_device *mdt = info->mti_mdt;
271 struct mdt_body *repbody;
276 rc = mdt_check_ucred(info);
278 GOTO(out, rc = err_serious(rc));
280 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
281 GOTO(out, rc = err_serious(-ENOMEM));
283 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
284 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
285 repbody->mbo_valid |= OBD_MD_FLID;
287 if (tsi->tsi_tgt->lut_mds_capa &&
288 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
289 struct mdt_object *root;
290 struct lustre_capa *capa;
292 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
294 GOTO(out, rc = PTR_ERR(root));
296 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
298 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
299 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
301 mdt_object_put(info->mti_env, root);
303 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
307 mdt_thread_info_fini(info);
311 static int mdt_statfs(struct tgt_session_info *tsi)
313 struct ptlrpc_request *req = tgt_ses_req(tsi);
314 struct mdt_thread_info *info = tsi2mdt_info(tsi);
315 struct md_device *next = info->mti_mdt->mdt_child;
316 struct ptlrpc_service_part *svcpt;
317 struct obd_statfs *osfs;
322 svcpt = req->rq_rqbd->rqbd_svcpt;
324 /* This will trigger a watchdog timeout */
325 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
326 (MDT_SERVICE_WATCHDOG_FACTOR *
327 at_get(&svcpt->scp_at_estimate)) + 1);
329 rc = mdt_check_ucred(info);
331 GOTO(out, rc = err_serious(rc));
333 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
334 GOTO(out, rc = err_serious(-ENOMEM));
336 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
338 GOTO(out, rc = -EPROTO);
340 /** statfs information are cached in the mdt_device */
341 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
342 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
343 /** statfs data is too old, get up-to-date one */
344 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
347 spin_lock(&info->mti_mdt->mdt_osfs_lock);
348 info->mti_mdt->mdt_osfs = *osfs;
349 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
350 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
352 /** use cached statfs data */
353 spin_lock(&info->mti_mdt->mdt_osfs_lock);
354 *osfs = info->mti_mdt->mdt_osfs;
355 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
359 mdt_counter_incr(req, LPROC_MDT_STATFS);
361 mdt_thread_info_fini(info);
366 * Pack SOM attributes into the reply.
367 * Call under a DLM UPDATE lock.
369 static void mdt_pack_size2body(struct mdt_thread_info *info,
370 struct mdt_object *mo)
373 struct md_attr *ma = &info->mti_attr;
375 LASSERT(ma->ma_attr.la_valid & LA_MODE);
376 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
378 /* Check if Size-on-MDS is supported, if this is a regular file,
379 * if SOM is enabled on the object and if SOM cache exists and valid.
380 * Otherwise do not pack Size-on-MDS attributes to the reply. */
381 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
382 !S_ISREG(ma->ma_attr.la_mode) ||
383 !mdt_object_is_som_enabled(mo) ||
384 !(ma->ma_valid & MA_SOM))
387 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
388 b->mbo_size = ma->ma_som->msd_size;
389 b->mbo_blocks = ma->ma_som->msd_blocks;
392 #ifdef CONFIG_FS_POSIX_ACL
394 * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
396 * \param info thread info object
397 * \param repbody reply to pack ACLs into
398 * \param o mdt object of file to examine
399 * \param nodemap nodemap of client to reply to
401 * \retval -errno error getting or parsing ACL from disk
403 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
404 struct mdt_object *o, struct lu_nodemap *nodemap)
406 const struct lu_env *env = info->mti_env;
407 struct md_object *next = mdt_object_child(o);
408 struct lu_buf *buf = &info->mti_buf;
411 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
412 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
414 if (buf->lb_len == 0)
417 rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
419 if (rc == -ENODATA) {
420 repbody->mbo_aclsize = 0;
421 repbody->mbo_valid |= OBD_MD_FLACL;
423 } else if (rc == -EOPNOTSUPP) {
426 CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
427 mdt_obd_name(info->mti_mdt),
428 PFID(mdt_object_fid(o)), rc);
431 rc = nodemap_map_acl(nodemap, buf->lb_buf,
432 rc, NODEMAP_FS_TO_CLIENT);
433 /* if all ACLs mapped out, rc is still >= 0 */
435 CERROR("%s: nodemap_map_acl unable to parse "DFID
436 " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
437 PFID(mdt_object_fid(o)), rc);
439 repbody->mbo_aclsize = rc;
440 repbody->mbo_valid |= OBD_MD_FLACL;
448 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
449 const struct lu_attr *attr, const struct lu_fid *fid)
451 struct md_attr *ma = &info->mti_attr;
452 struct obd_export *exp = info->mti_exp;
453 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
455 LASSERT(ma->ma_valid & MA_INODE);
457 b->mbo_atime = attr->la_atime;
458 b->mbo_mtime = attr->la_mtime;
459 b->mbo_ctime = attr->la_ctime;
460 b->mbo_mode = attr->la_mode;
461 b->mbo_size = attr->la_size;
462 b->mbo_blocks = attr->la_blocks;
463 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
464 NODEMAP_FS_TO_CLIENT,
466 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
467 NODEMAP_FS_TO_CLIENT,
469 b->mbo_flags = attr->la_flags;
470 b->mbo_nlink = attr->la_nlink;
471 b->mbo_rdev = attr->la_rdev;
473 /* XXX should pack the reply body according to lu_valid */
474 b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
475 OBD_MD_FLGID | OBD_MD_FLTYPE |
476 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
477 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
479 if (!S_ISREG(attr->la_mode)) {
480 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
481 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
482 /* means no objects are allocated on osts. */
483 LASSERT(!(ma->ma_valid & MA_LOV));
484 /* just ignore blocks occupied by extend attributes on MDS */
486 /* if no object is allocated on osts, the size on mds is valid.
488 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
489 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
490 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
491 /* A released file stores its size on MDS. */
492 /* But return 1 block for released file, unless tools like tar
493 * will consider it fully sparse. (LU-3864)
495 if (unlikely(b->mbo_size == 0))
499 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
504 b->mbo_valid |= OBD_MD_FLID;
505 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
506 PFID(fid), b->mbo_nlink,
507 b->mbo_mode, b->mbo_size);
511 mdt_body_reverse_idmap(info, b);
513 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
514 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
515 PFID(fid), (unsigned long long)b->mbo_size);
518 static inline int mdt_body_has_lov(const struct lu_attr *la,
519 const struct mdt_body *body)
521 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
522 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
525 void mdt_client_compatibility(struct mdt_thread_info *info)
527 struct mdt_body *body;
528 struct ptlrpc_request *req = mdt_info_req(info);
529 struct obd_export *exp = req->rq_export;
530 struct md_attr *ma = &info->mti_attr;
531 struct lu_attr *la = &ma->ma_attr;
534 if (exp_connect_layout(exp))
535 /* the client can deal with 16-bit lmm_stripe_count */
538 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
540 if (!mdt_body_has_lov(la, body))
543 /* now we have a reply with a lov for a client not compatible with the
544 * layout lock so we have to clean the layout generation number */
545 if (S_ISREG(la->la_mode))
546 ma->ma_lmm->lmm_layout_gen = 0;
550 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
552 const struct lu_env *env = info->mti_env;
555 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
564 /* Is it a directory? Let's check for the LMV as well */
565 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
566 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
570 rc2 = mo_xattr_get(env, mdt_object_child(o),
572 XATTR_NAME_DEFAULT_LMV);
574 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
582 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
585 const struct lu_env *env = info->mti_env;
589 LASSERT(info->mti_big_lmm_used == 0);
590 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
594 /* big_lmm may need to be grown */
595 if (info->mti_big_lmmsize < rc) {
596 int size = size_roundup_power2(rc);
598 if (info->mti_big_lmmsize > 0) {
599 /* free old buffer */
600 LASSERT(info->mti_big_lmm);
601 OBD_FREE_LARGE(info->mti_big_lmm,
602 info->mti_big_lmmsize);
603 info->mti_big_lmm = NULL;
604 info->mti_big_lmmsize = 0;
607 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
608 if (info->mti_big_lmm == NULL)
610 info->mti_big_lmmsize = size;
612 LASSERT(info->mti_big_lmmsize >= rc);
614 info->mti_buf.lb_buf = info->mti_big_lmm;
615 info->mti_buf.lb_len = info->mti_big_lmmsize;
616 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
621 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
622 struct md_attr *ma, const char *name)
624 struct md_object *next = mdt_object_child(o);
625 struct lu_buf *buf = &info->mti_buf;
628 if (strcmp(name, XATTR_NAME_LOV) == 0) {
629 buf->lb_buf = ma->ma_lmm;
630 buf->lb_len = ma->ma_lmm_size;
631 LASSERT(!(ma->ma_valid & MA_LOV));
632 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
633 buf->lb_buf = ma->ma_lmv;
634 buf->lb_len = ma->ma_lmv_size;
635 LASSERT(!(ma->ma_valid & MA_LMV));
636 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
637 buf->lb_buf = ma->ma_lmv;
638 buf->lb_len = ma->ma_lmv_size;
639 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
644 rc = mo_xattr_get(info->mti_env, next, buf, name);
648 if (strcmp(name, XATTR_NAME_LOV) == 0) {
649 if (info->mti_big_lmm_used)
650 ma->ma_lmm = info->mti_big_lmm;
652 /* NOT return LOV EA with hole to old client. */
653 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
654 LOV_PATTERN_F_HOLE) &&
655 !(exp_connect_flags(info->mti_exp) &
656 OBD_CONNECT_LFSCK)) {
659 ma->ma_lmm_size = rc;
660 ma->ma_valid |= MA_LOV;
662 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
663 if (info->mti_big_lmm_used)
664 ma->ma_lmv = info->mti_big_lmm;
666 ma->ma_lmv_size = rc;
667 ma->ma_valid |= MA_LMV;
668 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
669 ma->ma_lmv_size = rc;
670 ma->ma_valid |= MA_LMV_DEF;
673 /* Update mdt_max_mdsize so all clients will be aware that */
674 if (info->mti_mdt->mdt_max_mdsize < rc)
675 info->mti_mdt->mdt_max_mdsize = rc;
678 } else if (rc == -ENODATA) {
681 } else if (rc == -ERANGE) {
682 /* Default LMV has fixed size, so it must be able to fit
683 * in the original buffer */
684 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
686 rc = mdt_big_xattr_get(info, o, name);
688 info->mti_big_lmm_used = 1;
696 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
697 struct mdt_object *o, struct lu_fid *pfid)
699 struct lu_buf *buf = &info->mti_buf;
700 struct link_ea_header *leh;
701 struct link_ea_entry *lee;
705 buf->lb_buf = info->mti_big_lmm;
706 buf->lb_len = info->mti_big_lmmsize;
707 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
708 buf, XATTR_NAME_LINK);
709 /* ignore errors, MA_PFID won't be set and it is
710 * up to the caller to treat this as an error */
711 if (rc == -ERANGE || buf->lb_len == 0) {
712 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
713 buf->lb_buf = info->mti_big_lmm;
714 buf->lb_len = info->mti_big_lmmsize;
719 if (rc < sizeof(*leh)) {
720 CERROR("short LinkEA on "DFID": rc = %d\n",
721 PFID(mdt_object_fid(o)), rc);
725 leh = (struct link_ea_header *) buf->lb_buf;
726 lee = (struct link_ea_entry *)(leh + 1);
727 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
728 leh->leh_magic = LINK_EA_MAGIC;
729 leh->leh_reccount = __swab32(leh->leh_reccount);
730 leh->leh_len = __swab64(leh->leh_len);
732 if (leh->leh_magic != LINK_EA_MAGIC)
734 if (leh->leh_reccount == 0)
737 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
738 fid_be_to_cpu(pfid, pfid);
743 int mdt_attr_get_complex(struct mdt_thread_info *info,
744 struct mdt_object *o, struct md_attr *ma)
746 const struct lu_env *env = info->mti_env;
747 struct md_object *next = mdt_object_child(o);
748 struct lu_buf *buf = &info->mti_buf;
749 int need = ma->ma_need;
756 if (mdt_object_exists(o) == 0)
757 GOTO(out, rc = -ENOENT);
758 mode = lu_object_attr(&next->mo_lu);
760 if (need & MA_INODE) {
761 ma->ma_need = MA_INODE;
762 rc = mo_attr_get(env, next, ma);
765 ma->ma_valid |= MA_INODE;
768 if (need & MA_PFID) {
769 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
771 ma->ma_valid |= MA_PFID;
772 /* ignore this error, parent fid is not mandatory */
776 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
777 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
782 if (need & MA_LMV && S_ISDIR(mode)) {
783 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
788 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
789 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
794 if (need & MA_SOM && S_ISREG(mode)) {
795 buf->lb_buf = info->mti_xattr_buf;
796 buf->lb_len = sizeof(info->mti_xattr_buf);
797 CLASSERT(sizeof(struct som_attrs) <=
798 sizeof(info->mti_xattr_buf));
799 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
800 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
802 ma->ma_valid |= MA_SOM;
803 else if (rc2 < 0 && rc2 != -ENODATA)
807 if (need & MA_HSM && S_ISREG(mode)) {
808 buf->lb_buf = info->mti_xattr_buf;
809 buf->lb_len = sizeof(info->mti_xattr_buf);
810 CLASSERT(sizeof(struct hsm_attrs) <=
811 sizeof(info->mti_xattr_buf));
812 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
813 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
815 ma->ma_valid |= MA_HSM;
816 else if (rc2 < 0 && rc2 != -ENODATA)
820 #ifdef CONFIG_FS_POSIX_ACL
821 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
822 buf->lb_buf = ma->ma_acl;
823 buf->lb_len = ma->ma_acl_size;
824 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
826 ma->ma_acl_size = rc2;
827 ma->ma_valid |= MA_ACL_DEF;
828 } else if (rc2 == -ENODATA) {
837 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
838 rc, ma->ma_valid, ma->ma_lmm);
842 static int mdt_getattr_internal(struct mdt_thread_info *info,
843 struct mdt_object *o, int ma_need)
845 struct md_object *next = mdt_object_child(o);
846 const struct mdt_body *reqbody = info->mti_body;
847 struct ptlrpc_request *req = mdt_info_req(info);
848 struct md_attr *ma = &info->mti_attr;
849 struct lu_attr *la = &ma->ma_attr;
850 struct req_capsule *pill = info->mti_pill;
851 const struct lu_env *env = info->mti_env;
852 struct mdt_body *repbody;
853 struct lu_buf *buffer = &info->mti_buf;
854 struct obd_export *exp = info->mti_exp;
855 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
860 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
861 RETURN(err_serious(-ENOMEM));
863 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
867 if (mdt_object_remote(o)) {
868 /* This object is located on remote node.*/
869 /* Return -ENOTSUPP for old client */
870 if (!mdt_is_dne_client(req->rq_export))
871 GOTO(out, rc = -ENOTSUPP);
873 repbody->mbo_fid1 = *mdt_object_fid(o);
874 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
878 if (reqbody->mbo_eadatasize > 0) {
879 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
880 if (buffer->lb_buf == NULL)
881 GOTO(out, rc = -EPROTO);
882 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
885 buffer->lb_buf = NULL;
887 ma_need &= ~(MA_LOV | MA_LMV);
888 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
889 mdt_obd_name(info->mti_mdt),
890 req->rq_export->exp_client_uuid.uuid);
893 /* If it is dir object and client require MEA, then we got MEA */
894 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
895 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
896 /* Assumption: MDT_MD size is enough for lmv size. */
897 ma->ma_lmv = buffer->lb_buf;
898 ma->ma_lmv_size = buffer->lb_len;
899 ma->ma_need = MA_INODE;
900 if (ma->ma_lmv_size > 0) {
901 if (reqbody->mbo_valid & OBD_MD_MEA)
902 ma->ma_need |= MA_LMV;
903 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
904 ma->ma_need |= MA_LMV_DEF;
907 ma->ma_lmm = buffer->lb_buf;
908 ma->ma_lmm_size = buffer->lb_len;
909 ma->ma_need = MA_INODE | MA_HSM;
910 if (ma->ma_lmm_size > 0)
911 ma->ma_need |= MA_LOV;
914 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
915 reqbody->mbo_valid & OBD_MD_FLDIREA &&
916 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
917 /* get default stripe info for this dir. */
918 ma->ma_need |= MA_LOV_DEF;
920 ma->ma_need |= ma_need;
921 if (ma->ma_need & MA_SOM)
922 ma->ma_som = &info->mti_u.som.data;
924 rc = mdt_attr_get_complex(info, o, ma);
926 CERROR("%s: getattr error for "DFID": rc = %d\n",
927 mdt_obd_name(info->mti_mdt),
928 PFID(mdt_object_fid(o)), rc);
932 /* if file is released, check if a restore is running */
933 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
934 mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
935 repbody->mbo_t_state = MS_RESTORE;
936 repbody->mbo_valid |= OBD_MD_TSTATE;
939 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
941 /* the Lustre protocol supposes to return default striping
942 * on the user-visible root if explicitly requested */
943 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
944 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
945 struct lu_fid rootfid;
946 struct mdt_object *root;
947 struct mdt_device *mdt = info->mti_mdt;
949 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
952 root = mdt_object_find(env, mdt, &rootfid);
954 RETURN(PTR_ERR(root));
955 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
956 mdt_object_put(info->mti_env, root);
958 CERROR("%s: getattr error for "DFID": rc = %d\n",
959 mdt_obd_name(info->mti_mdt),
960 PFID(mdt_object_fid(o)), rc);
965 if (likely(ma->ma_valid & MA_INODE))
966 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
970 if (mdt_body_has_lov(la, reqbody)) {
971 if (ma->ma_valid & MA_LOV) {
972 LASSERT(ma->ma_lmm_size);
973 repbody->mbo_eadatasize = ma->ma_lmm_size;
974 if (S_ISDIR(la->la_mode))
975 repbody->mbo_valid |= OBD_MD_FLDIREA;
977 repbody->mbo_valid |= OBD_MD_FLEASIZE;
978 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
980 if (ma->ma_valid & MA_LMV) {
981 /* Return -ENOTSUPP for old client */
982 if (!mdt_is_striped_client(req->rq_export))
985 LASSERT(S_ISDIR(la->la_mode));
986 mdt_dump_lmv(D_INFO, ma->ma_lmv);
987 repbody->mbo_eadatasize = ma->ma_lmv_size;
988 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
990 if (ma->ma_valid & MA_LMV_DEF) {
991 /* Return -ENOTSUPP for old client */
992 if (!mdt_is_striped_client(req->rq_export))
994 LASSERT(S_ISDIR(la->la_mode));
995 repbody->mbo_eadatasize = ma->ma_lmv_size;
996 repbody->mbo_valid |= (OBD_MD_FLDIREA |
999 } else if (S_ISLNK(la->la_mode) &&
1000 reqbody->mbo_valid & OBD_MD_LINKNAME) {
1001 buffer->lb_buf = ma->ma_lmm;
1002 /* eadatasize from client includes NULL-terminator, so
1003 * there is no need to read it */
1004 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1005 rc = mo_readlink(env, next, buffer);
1006 if (unlikely(rc <= 0)) {
1007 CERROR("%s: readlink failed for "DFID": rc = %d\n",
1008 mdt_obd_name(info->mti_mdt),
1009 PFID(mdt_object_fid(o)), rc);
1012 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1014 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1016 repbody->mbo_valid |= OBD_MD_LINKNAME;
1017 /* we need to report back size with NULL-terminator
1018 * because client expects that */
1019 repbody->mbo_eadatasize = rc + 1;
1020 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1021 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1022 "on "DFID ", expected %d\n",
1023 mdt_obd_name(info->mti_mdt),
1024 rc, PFID(mdt_object_fid(o)),
1025 reqbody->mbo_eadatasize - 1);
1026 /* NULL terminate */
1027 ((char *)ma->ma_lmm)[rc] = 0;
1029 /* If the total CDEBUG() size is larger than a page, it
1030 * will print a warning to the console, avoid this by
1031 * printing just the last part of the symlink. */
1032 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1033 print_limit < rc ? "..." : "", print_limit,
1034 (char *)ma->ma_lmm + rc - print_limit, rc);
1039 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1040 repbody->mbo_max_cookiesize = 0;
1041 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1042 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1043 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
1044 "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
1045 repbody->mbo_max_cookiesize);
1048 if (exp_connect_rmtclient(info->mti_exp) &&
1049 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1050 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1052 /* mdt_getattr_lock only */
1053 rc = mdt_pack_remote_perm(info, o, buf);
1055 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1056 repbody->mbo_aclsize = 0;
1059 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1060 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1063 #ifdef CONFIG_FS_POSIX_ACL
1064 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1065 (reqbody->mbo_valid & OBD_MD_FLACL))
1066 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1069 if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1070 info->mti_mdt->mdt_lut.lut_mds_capa &&
1071 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1072 struct lustre_capa *capa;
1074 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1076 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1077 rc = mo_capa_get(env, next, capa, 0);
1080 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1085 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1090 static int mdt_renew_capa(struct mdt_thread_info *info)
1092 struct mdt_object *obj = info->mti_object;
1093 struct mdt_body *body;
1094 struct lustre_capa *capa, *c;
1098 /* if object doesn't exist, or server has disabled capability,
1099 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1102 if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1103 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1106 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1107 LASSERT(body != NULL);
1109 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1112 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1116 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1118 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1122 static int mdt_getattr(struct tgt_session_info *tsi)
1124 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1125 struct mdt_object *obj = info->mti_object;
1126 struct req_capsule *pill = info->mti_pill;
1127 struct mdt_body *reqbody;
1128 struct mdt_body *repbody;
1132 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1135 if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1136 rc = req_capsule_server_pack(pill);
1138 RETURN(err_serious(rc));
1139 rc = mdt_renew_capa(info);
1140 GOTO(out_shrink, rc);
1143 LASSERT(obj != NULL);
1144 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1146 /* Unlike intent case where we need to pre-fill out buffers early on
1147 * in intent policy for ldlm reasons, here we can have a much better
1148 * guess at EA size by just reading it from disk.
1149 * Exceptions are readdir and (missing) directory striping */
1151 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1152 /* No easy way to know how long is the symlink, but it cannot
1153 * be more than PATH_MAX, so we allocate +1 */
1156 /* A special case for fs ROOT: getattr there might fetch
1157 * default EA for entire fs, not just for this dir!
1159 } else if (lu_fid_eq(mdt_object_fid(obj),
1160 &info->mti_mdt->mdt_md_root_fid) &&
1161 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1162 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1164 /* Should the default strping be bigger, mdt_fix_reply
1165 * will reallocate */
1166 rc = DEF_REP_MD_SIZE;
1168 /* Read the actual EA size from disk */
1169 rc = mdt_attr_get_eabuf_size(info, obj);
1173 GOTO(out, rc = err_serious(rc));
1175 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1177 rc = req_capsule_server_pack(pill);
1178 if (unlikely(rc != 0))
1179 GOTO(out, rc = err_serious(rc));
1181 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1182 LASSERT(repbody != NULL);
1183 repbody->mbo_eadatasize = 0;
1184 repbody->mbo_aclsize = 0;
1186 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1187 rc = mdt_init_ucred(info, reqbody);
1189 rc = mdt_check_ucred(info);
1191 GOTO(out_shrink, rc);
1193 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1196 * Don't check capability at all, because rename might getattr for
1197 * remote obj, and at that time no capability is available.
1199 mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1200 rc = mdt_getattr_internal(info, obj, 0);
1201 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1202 mdt_exit_ucred(info);
1205 mdt_client_compatibility(info);
1206 rc2 = mdt_fix_reply(info);
1210 mdt_thread_info_fini(info);
1214 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1216 struct mdt_thread_info *info;
1217 struct ptlrpc_request *req = tgt_ses_req(tsi);
1218 struct obd_export *exp = req->rq_export;
1219 struct mdt_object *o1, *o2, *o;
1220 struct mdt_lock_handle *lh1, *lh2;
1221 struct mdc_swap_layouts *msl;
1225 /* client does not support layout lock, so layout swaping
1227 * FIXME: there is a problem for old clients which don't support
1228 * layout lock yet. If those clients have already opened the file
1229 * they won't be notified at all so that old layout may still be
1230 * used to do IO. This can be fixed after file release is landed by
1231 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1232 if (!exp_connect_layout(exp))
1233 RETURN(-EOPNOTSUPP);
1235 info = tsi2mdt_info(tsi);
1237 if (info->mti_dlm_req != NULL)
1238 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1240 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1241 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1242 req_capsule_client_get(info->mti_pill,
1245 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1246 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1247 req_capsule_client_get(info->mti_pill,
1250 o1 = info->mti_object;
1251 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1252 &info->mti_body->mbo_fid2);
1254 GOTO(out, rc = PTR_ERR(o));
1256 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1257 GOTO(put, rc = -ENOENT);
1259 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1260 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1266 /* permission check. Make sure the calling process having permission
1267 * to write both files. */
1268 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1273 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1278 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1280 GOTO(put, rc = -EPROTO);
1282 lh1 = &info->mti_lh[MDT_LH_NEW];
1283 mdt_lock_reg_init(lh1, LCK_EX);
1284 lh2 = &info->mti_lh[MDT_LH_OLD];
1285 mdt_lock_reg_init(lh2, LCK_EX);
1287 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1288 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1292 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1293 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1297 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1298 mdt_object_child(o2), msl->msl_flags);
1301 mdt_object_unlock(info, o2, lh2, rc);
1303 mdt_object_unlock(info, o1, lh1, rc);
1305 mdt_object_put(info->mti_env, o);
1307 mdt_thread_info_fini(info);
1311 static int mdt_raw_lookup(struct mdt_thread_info *info,
1312 struct mdt_object *parent,
1313 const struct lu_name *lname,
1314 struct ldlm_reply *ldlm_rep)
1316 struct md_object *next = mdt_object_child(info->mti_object);
1317 const struct mdt_body *reqbody = info->mti_body;
1318 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1319 struct mdt_body *repbody;
1323 if (reqbody->mbo_valid != OBD_MD_FLID)
1326 LASSERT(!info->mti_cross_ref);
1328 /* Only got the fid of this obj by name */
1329 fid_zero(child_fid);
1330 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1333 /* XXX is raw_lookup possible as intent operation? */
1336 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1339 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1341 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1344 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1345 repbody->mbo_fid1 = *child_fid;
1346 repbody->mbo_valid = OBD_MD_FLID;
1352 * UPDATE lock should be taken against parent, and be release before exit;
1353 * child_bits lock should be taken against child, and be returned back:
1354 * (1)normal request should release the child lock;
1355 * (2)intent request will grant the lock to client.
1357 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1358 struct mdt_lock_handle *lhc,
1360 struct ldlm_reply *ldlm_rep)
1362 struct ptlrpc_request *req = mdt_info_req(info);
1363 struct mdt_body *reqbody = NULL;
1364 struct mdt_object *parent = info->mti_object;
1365 struct mdt_object *child;
1366 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1367 struct lu_name *lname = NULL;
1368 struct mdt_lock_handle *lhp = NULL;
1369 struct ldlm_lock *lock;
1376 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1377 LASSERT(ergo(is_resent,
1378 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1383 if (info->mti_cross_ref) {
1384 /* Only getattr on the child. Parent is on another node. */
1385 mdt_set_disposition(info, ldlm_rep,
1386 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1388 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1390 PFID(mdt_object_fid(child)), ldlm_rep);
1392 rc = mdt_check_resent_lock(info, child, lhc);
1395 } else if (rc > 0) {
1396 mdt_lock_handle_init(lhc);
1397 mdt_lock_reg_init(lhc, LCK_PR);
1400 * Object's name is on another MDS, no lookup or layout
1401 * lock is needed here but update lock is.
1403 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1404 MDS_INODELOCK_LAYOUT);
1405 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1407 rc = mdt_object_lock(info, child, lhc, child_bits,
1413 /* Finally, we can get attr for child. */
1414 if (!mdt_object_exists(child)) {
1415 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1417 "remote object doesn't exist.\n");
1418 mdt_object_unlock(info, child, lhc, 1);
1422 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1423 rc = mdt_getattr_internal(info, child, 0);
1424 if (unlikely(rc != 0))
1425 mdt_object_unlock(info, child, lhc, 1);
1430 lname = &info->mti_name;
1431 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1433 if (lu_name_is_valid(lname)) {
1434 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1435 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1436 PNAME(lname), ldlm_rep);
1438 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1439 if (unlikely(reqbody == NULL))
1440 RETURN(err_serious(-EPROTO));
1442 *child_fid = reqbody->mbo_fid2;
1444 if (unlikely(!fid_is_sane(child_fid)))
1445 RETURN(err_serious(-EINVAL));
1447 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1449 PFID(mdt_object_fid(parent)),
1450 PFID(&reqbody->mbo_fid2), ldlm_rep);
1453 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1455 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1456 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1458 "Parent doesn't exist!\n");
1462 if (mdt_object_remote(parent)) {
1463 CERROR("%s: parent "DFID" is on remote target\n",
1464 mdt_obd_name(info->mti_mdt),
1465 PFID(mdt_object_fid(parent)));
1469 if (lu_name_is_valid(lname)) {
1470 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1477 /* step 1: lock parent only if parent is a directory */
1478 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1479 lhp = &info->mti_lh[MDT_LH_PARENT];
1480 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1481 rc = mdt_object_lock(info, parent, lhp,
1482 MDS_INODELOCK_UPDATE,
1484 if (unlikely(rc != 0))
1488 /* step 2: lookup child's fid by name */
1489 fid_zero(child_fid);
1490 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1491 child_fid, &info->mti_spec);
1493 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1496 GOTO(out_parent, rc);
1499 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1502 *step 3: find the child object by fid & lock it.
1503 * regardless if it is local or remote.
1505 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1506 * set parent dir fid the same as child fid in getattr by fid case
1507 * we should not lu_object_find() the object again, could lead
1508 * to hung if there is a concurrent unlink destroyed the object.
1510 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1511 mdt_object_get(info->mti_env, parent);
1514 child = mdt_object_find(info->mti_env, info->mti_mdt,
1518 if (unlikely(IS_ERR(child)))
1519 GOTO(out_parent, rc = PTR_ERR(child));
1521 rc = mdt_check_resent_lock(info, child, lhc);
1523 GOTO(out_child, rc);
1524 } else if (rc > 0) {
1525 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1526 mdt_lock_handle_init(lhc);
1527 mdt_lock_reg_init(lhc, LCK_PR);
1530 if (!mdt_object_exists(child)) {
1531 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1533 "Object doesn't exist!\n");
1534 GOTO(out_child, rc = -ENOENT);
1537 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1538 mdt_object_exists(child) && !mdt_object_remote(child)) {
1539 struct md_attr *ma = &info->mti_attr;
1542 ma->ma_need = MA_INODE;
1543 rc = mdt_attr_get_complex(info, child, ma);
1544 if (unlikely(rc != 0))
1545 GOTO(out_child, rc);
1547 /* If the file has not been changed for some time, we
1548 * return not only a LOOKUP lock, but also an UPDATE
1549 * lock and this might save us RPC on later STAT. For
1550 * directories, it also let negative dentry cache start
1551 * working for this dir. */
1552 if (ma->ma_valid & MA_INODE &&
1553 ma->ma_attr.la_valid & LA_CTIME &&
1554 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1555 ma->ma_attr.la_ctime < cfs_time_current_sec())
1556 child_bits |= MDS_INODELOCK_UPDATE;
1559 /* layout lock must be granted in a best-effort way
1560 * for IT operations */
1561 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1562 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1563 exp_connect_layout(info->mti_exp) &&
1564 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1565 !mdt_object_remote(child) && ldlm_rep != NULL) {
1566 /* try to grant layout lock for regular file. */
1572 child_bits |= MDS_INODELOCK_LAYOUT;
1573 /* try layout lock, it may fail to be granted due to
1574 * contention at LOOKUP or UPDATE */
1575 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1577 child_bits &= ~MDS_INODELOCK_LAYOUT;
1578 LASSERT(child_bits != 0);
1579 rc = mdt_object_lock(info, child, lhc,
1580 child_bits, MDT_CROSS_LOCK);
1585 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1586 * client will enqueue the lock to the remote MDT */
1587 if (mdt_object_remote(child))
1588 child_bits &= ~MDS_INODELOCK_UPDATE;
1589 rc = mdt_object_lock(info, child, lhc, child_bits,
1592 if (unlikely(rc != 0))
1593 GOTO(out_child, rc);
1596 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1597 /* Get MA_SOM attributes if update lock is given. */
1599 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1600 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1603 /* finally, we can get attr for child. */
1604 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1605 rc = mdt_getattr_internal(info, child, ma_need);
1606 if (unlikely(rc != 0)) {
1607 mdt_object_unlock(info, child, lhc, 1);
1609 /* Debugging code. */
1610 LDLM_DEBUG(lock, "Returning lock to client");
1611 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1612 &lock->l_resource->lr_name),
1613 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1614 PLDLMRES(lock->l_resource),
1615 PFID(mdt_object_fid(child)));
1616 if (mdt_object_exists(child) && !mdt_object_remote(child))
1617 mdt_pack_size2body(info, child);
1620 LDLM_LOCK_PUT(lock);
1624 mdt_object_put(info->mti_env, child);
1627 mdt_object_unlock(info, parent, lhp, 1);
1631 /* normal handler: should release the child lock */
1632 static int mdt_getattr_name(struct tgt_session_info *tsi)
1634 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1635 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1636 struct mdt_body *reqbody;
1637 struct mdt_body *repbody;
1641 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1642 LASSERT(reqbody != NULL);
1643 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1644 LASSERT(repbody != NULL);
1646 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1647 repbody->mbo_eadatasize = 0;
1648 repbody->mbo_aclsize = 0;
1650 rc = mdt_init_ucred(info, reqbody);
1652 GOTO(out_shrink, rc);
1654 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1655 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1656 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1657 lhc->mlh_reg_lh.cookie = 0;
1659 mdt_exit_ucred(info);
1662 mdt_client_compatibility(info);
1663 rc2 = mdt_fix_reply(info);
1666 mdt_thread_info_fini(info);
1670 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1671 void *karg, void *uarg);
1673 static int mdt_set_info(struct tgt_session_info *tsi)
1675 struct ptlrpc_request *req = tgt_ses_req(tsi);
1678 int keylen, vallen, rc = 0;
1682 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1684 DEBUG_REQ(D_HA, req, "no set_info key");
1685 RETURN(err_serious(-EFAULT));
1688 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1691 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1693 DEBUG_REQ(D_HA, req, "no set_info val");
1694 RETURN(err_serious(-EFAULT));
1697 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1700 /* Swab any part of val you need to here */
1701 if (KEY_IS(KEY_READ_ONLY)) {
1702 spin_lock(&req->rq_export->exp_lock);
1704 *exp_connect_flags_ptr(req->rq_export) |=
1707 *exp_connect_flags_ptr(req->rq_export) &=
1708 ~OBD_CONNECT_RDONLY;
1709 spin_unlock(&req->rq_export->exp_lock);
1710 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1711 struct changelog_setinfo *cs = val;
1713 if (vallen != sizeof(*cs)) {
1714 CERROR("%s: bad changelog_clear setinfo size %d\n",
1715 tgt_name(tsi->tsi_tgt), vallen);
1718 if (ptlrpc_req_need_swab(req)) {
1719 __swab64s(&cs->cs_recno);
1720 __swab32s(&cs->cs_id);
1723 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1731 static int mdt_readpage(struct tgt_session_info *tsi)
1733 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1734 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1735 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1736 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1737 struct mdt_body *repbody;
1743 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1744 RETURN(err_serious(-ENOMEM));
1746 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1747 if (repbody == NULL || reqbody == NULL)
1748 RETURN(err_serious(-EFAULT));
1751 * prepare @rdpg before calling lower layers and transfer itself. Here
1752 * reqbody->size contains offset of where to start to read and
1753 * reqbody->nlink contains number bytes to read.
1755 rdpg->rp_hash = reqbody->mbo_size;
1756 if (rdpg->rp_hash != reqbody->mbo_size) {
1757 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1758 rdpg->rp_hash, reqbody->mbo_size);
1762 rdpg->rp_attrs = reqbody->mbo_mode;
1763 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1764 rdpg->rp_attrs |= LUDA_64BITHASH;
1765 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1766 exp_max_brw_size(tsi->tsi_exp));
1767 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1769 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1770 if (rdpg->rp_pages == NULL)
1773 for (i = 0; i < rdpg->rp_npages; ++i) {
1774 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1775 if (rdpg->rp_pages[i] == NULL)
1776 GOTO(free_rdpg, rc = -ENOMEM);
1779 /* call lower layers to fill allocated pages with directory data */
1780 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1782 GOTO(free_rdpg, rc);
1784 /* send pages to client */
1785 rc = tgt_sendpage(tsi, rdpg, rc);
1790 for (i = 0; i < rdpg->rp_npages; i++)
1791 if (rdpg->rp_pages[i] != NULL)
1792 __free_page(rdpg->rp_pages[i]);
1793 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1795 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1801 static int mdt_reint_internal(struct mdt_thread_info *info,
1802 struct mdt_lock_handle *lhc,
1805 struct req_capsule *pill = info->mti_pill;
1806 struct mdt_body *repbody;
1811 rc = mdt_reint_unpack(info, op);
1813 CERROR("Can't unpack reint, rc %d\n", rc);
1814 RETURN(err_serious(rc));
1817 /* for replay (no_create) lmm is not needed, client has it already */
1818 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1819 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1822 /* llog cookies are always 0, the field is kept for compatibility */
1823 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1824 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1826 rc = req_capsule_server_pack(pill);
1828 CERROR("Can't pack response, rc %d\n", rc);
1829 RETURN(err_serious(rc));
1832 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1833 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1835 repbody->mbo_eadatasize = 0;
1836 repbody->mbo_aclsize = 0;
1839 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1841 /* for replay no cookkie / lmm need, because client have this already */
1842 if (info->mti_spec.no_create)
1843 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1844 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1846 rc = mdt_init_ucred_reint(info);
1848 GOTO(out_shrink, rc);
1850 rc = mdt_fix_attr_ucred(info, op);
1852 GOTO(out_ucred, rc = err_serious(rc));
1854 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1855 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1856 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1857 GOTO(out_ucred, rc);
1859 rc = mdt_reint_rec(info, lhc);
1862 mdt_exit_ucred(info);
1864 mdt_client_compatibility(info);
1865 rc2 = mdt_fix_reply(info);
1871 static long mdt_reint_opcode(struct ptlrpc_request *req,
1872 const struct req_format **fmt)
1874 struct mdt_device *mdt;
1875 struct mdt_rec_reint *rec;
1878 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1880 opc = rec->rr_opcode;
1881 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1882 if (opc < REINT_MAX && fmt[opc] != NULL)
1883 req_capsule_extend(&req->rq_pill, fmt[opc]);
1885 mdt = mdt_exp2dev(req->rq_export);
1886 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1887 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1888 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1889 opc = err_serious(-EFAULT);
1892 opc = err_serious(-EFAULT);
1897 static int mdt_reint(struct tgt_session_info *tsi)
1901 static const struct req_format *reint_fmts[REINT_MAX] = {
1902 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1903 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1904 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1905 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1906 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1907 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1908 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1909 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1910 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1915 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1917 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1919 * No lock possible here from client to pass it to reint code
1922 rc = mdt_reint_internal(info, NULL, opc);
1923 mdt_thread_info_fini(info);
1928 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1932 /* this should sync the whole device */
1933 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1935 struct dt_device *dt = mdt->mdt_bottom;
1939 rc = dt->dd_ops->dt_sync(env, dt);
1943 /* this should sync this object */
1944 static int mdt_object_sync(struct mdt_thread_info *info)
1946 struct md_object *next;
1950 if (!mdt_object_exists(info->mti_object)) {
1951 CWARN("%s: non existing object "DFID": rc = %d\n",
1952 mdt_obd_name(info->mti_mdt),
1953 PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1956 next = mdt_object_child(info->mti_object);
1957 rc = mo_object_sync(info->mti_env, next);
1962 static int mdt_sync(struct tgt_session_info *tsi)
1964 struct ptlrpc_request *req = tgt_ses_req(tsi);
1965 struct req_capsule *pill = tsi->tsi_pill;
1966 struct mdt_body *body;
1971 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1972 RETURN(err_serious(-ENOMEM));
1974 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1975 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1977 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1979 /* sync an object */
1980 rc = mdt_object_sync(info);
1982 const struct lu_fid *fid;
1983 struct lu_attr *la = &info->mti_attr.ma_attr;
1985 info->mti_attr.ma_need = MA_INODE;
1986 info->mti_attr.ma_valid = 0;
1987 rc = mdt_attr_get_complex(info, info->mti_object,
1990 body = req_capsule_server_get(pill,
1992 fid = mdt_object_fid(info->mti_object);
1993 mdt_pack_attr2body(info, body, la, fid);
1996 mdt_thread_info_fini(info);
1999 mdt_counter_incr(req, LPROC_MDT_SYNC);
2005 * Handle quota control requests to consult current usage/limit, but also
2006 * to configure quota enforcement
2008 static int mdt_quotactl(struct tgt_session_info *tsi)
2010 struct obd_export *exp = tsi->tsi_exp;
2011 struct req_capsule *pill = tsi->tsi_pill;
2012 struct obd_quotactl *oqctl, *repoqc;
2014 struct mdt_device *mdt = mdt_exp2dev(exp);
2015 struct lu_device *qmt = mdt->mdt_qmt_dev;
2016 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
2019 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2021 RETURN(err_serious(-EPROTO));
2023 rc = req_capsule_server_pack(pill);
2025 RETURN(err_serious(rc));
2027 switch (oqctl->qc_cmd) {
2029 case LUSTRE_Q_INVALIDATE:
2030 case LUSTRE_Q_FINVALIDATE:
2034 /* deprecated, not used any more */
2035 RETURN(-EOPNOTSUPP);
2036 /* master quotactl */
2039 if (!nodemap_can_setquota(nodemap))
2044 RETURN(-EOPNOTSUPP);
2045 /* slave quotactl */
2050 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2054 /* map uid/gid for remote client */
2056 if (exp_connect_rmtclient(exp)) {
2057 struct lustre_idmap_table *idmap;
2059 idmap = exp->exp_mdt_data.med_idmap;
2061 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2062 oqctl->qc_cmd != Q_GETINFO))
2065 if (oqctl->qc_type == USRQUOTA)
2066 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2068 else if (oqctl->qc_type == GRPQUOTA)
2069 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2074 if (id == CFS_IDMAP_NOTFOUND) {
2075 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2080 if (oqctl->qc_type == USRQUOTA)
2081 id = nodemap_map_id(nodemap, NODEMAP_UID,
2082 NODEMAP_CLIENT_TO_FS, id);
2083 else if (oqctl->qc_type == GRPQUOTA)
2084 id = nodemap_map_id(nodemap, NODEMAP_UID,
2085 NODEMAP_CLIENT_TO_FS, id);
2087 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2089 RETURN(err_serious(-EFAULT));
2091 if (oqctl->qc_id != id)
2092 swap(oqctl->qc_id, id);
2094 switch (oqctl->qc_cmd) {
2100 /* forward quotactl request to QMT */
2101 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2106 /* slave quotactl */
2107 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2112 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2116 if (oqctl->qc_id != id)
2117 swap(oqctl->qc_id, id);
2123 /** clone llog ctxt from child (mdd)
2124 * This allows remote llog (replicator) access.
2125 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2126 * context was originally set up, or we can handle them directly.
2127 * I choose the latter, but that means I need any llog
2128 * contexts set up by child to be accessable by the mdt. So we clone the
2129 * context into our context list here.
2131 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2134 struct md_device *next = mdt->mdt_child;
2135 struct llog_ctxt *ctxt;
2138 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2141 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2142 if (rc || ctxt == NULL) {
2146 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2148 CERROR("Can't set mdt ctxt %d\n", rc);
2153 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2154 struct mdt_device *mdt, int idx)
2156 struct llog_ctxt *ctxt;
2158 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2161 /* Put once for the get we just did, and once for the clone */
2162 llog_ctxt_put(ctxt);
2163 llog_ctxt_put(ctxt);
2168 * sec context handlers
2170 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2174 rc = mdt_handle_idmap(tsi);
2176 struct ptlrpc_request *req = tgt_ses_req(tsi);
2179 opc = lustre_msg_get_opc(req->rq_reqmsg);
2180 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2181 sptlrpc_svc_ctx_invalidate(req);
2184 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2190 * quota request handlers
2192 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2194 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2195 struct lu_device *qmt = mdt->mdt_qmt_dev;
2200 RETURN(err_serious(-EOPNOTSUPP));
2202 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2206 struct mdt_object *mdt_object_new(const struct lu_env *env,
2207 struct mdt_device *d,
2208 const struct lu_fid *f)
2210 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2211 struct lu_object *o;
2212 struct mdt_object *m;
2215 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2216 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2217 if (unlikely(IS_ERR(o)))
2218 m = (struct mdt_object *)o;
2224 struct mdt_object *mdt_object_find(const struct lu_env *env,
2225 struct mdt_device *d,
2226 const struct lu_fid *f)
2228 struct lu_object *o;
2229 struct mdt_object *m;
2232 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2233 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2234 if (unlikely(IS_ERR(o)))
2235 m = (struct mdt_object *)o;
2243 * Asyncronous commit for mdt device.
2245 * Pass asynchonous commit call down the MDS stack.
2247 * \param env environment
2248 * \param mdt the mdt device
2250 static void mdt_device_commit_async(const struct lu_env *env,
2251 struct mdt_device *mdt)
2253 struct dt_device *dt = mdt->mdt_bottom;
2256 rc = dt->dd_ops->dt_commit_async(env, dt);
2257 if (unlikely(rc != 0))
2258 CWARN("%s: async commit start failed: rc = %d\n",
2259 mdt_obd_name(mdt), rc);
2263 * Mark the lock as "synchonous".
2265 * Mark the lock to deffer transaction commit to the unlock time.
2267 * \param lock the lock to mark as "synchonous"
2269 * \see mdt_is_lock_sync
2270 * \see mdt_save_lock
2272 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2274 lock->l_ast_data = (void*)1;
2278 * Check whehter the lock "synchonous" or not.
2280 * \param lock the lock to check
2281 * \retval 1 the lock is "synchonous"
2282 * \retval 0 the lock isn't "synchronous"
2284 * \see mdt_set_lock_sync
2285 * \see mdt_save_lock
2287 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2289 return lock->l_ast_data != NULL;
2293 * Blocking AST for mdt locks.
2295 * Starts transaction commit if in case of COS lock conflict or
2296 * deffers such a commit to the mdt_save_lock.
2298 * \param lock the lock which blocks a request or cancelling lock
2299 * \param desc unused
2300 * \param data unused
2301 * \param flag indicates whether this cancelling or blocking callback
2303 * \see ldlm_blocking_ast_nocheck
2305 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2306 void *data, int flag)
2308 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2309 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2313 if (flag == LDLM_CB_CANCELING)
2315 lock_res_and_lock(lock);
2316 if (lock->l_blocking_ast != mdt_blocking_ast) {
2317 unlock_res_and_lock(lock);
2320 if (mdt_cos_is_enabled(mdt) &&
2321 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2322 lock->l_blocking_lock != NULL &&
2323 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2324 mdt_set_lock_sync(lock);
2326 rc = ldlm_blocking_ast_nocheck(lock);
2328 /* There is no lock conflict if l_blocking_lock == NULL,
2329 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2330 * when the last reference to a local lock was released */
2331 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2334 rc = lu_env_init(&env, LCT_LOCAL);
2335 if (unlikely(rc != 0))
2336 CWARN("%s: lu_env initialization failed, cannot "
2337 "start asynchronous commit: rc = %d\n",
2340 mdt_device_commit_async(&env, mdt);
2346 /* Used for cross-MDT lock */
2347 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2348 void *data, int flag)
2350 struct lustre_handle lockh;
2354 case LDLM_CB_BLOCKING:
2355 ldlm_lock2handle(lock, &lockh);
2356 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2358 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2362 case LDLM_CB_CANCELING:
2363 LDLM_DEBUG(lock, "Revoke remote lock\n");
2371 int mdt_check_resent_lock(struct mdt_thread_info *info,
2372 struct mdt_object *mo,
2373 struct mdt_lock_handle *lhc)
2375 /* the lock might already be gotten in ldlm_handle_enqueue() */
2376 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2377 struct ptlrpc_request *req = mdt_info_req(info);
2378 struct ldlm_lock *lock;
2380 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2381 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2383 /* Lock is pinned by ldlm_handle_enqueue0() as it is
2384 * a resend case, however, it could be already destroyed
2385 * due to client eviction or a raced cancel RPC. */
2386 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2387 lhc->mlh_reg_lh.cookie);
2391 if (!fid_res_name_eq(mdt_object_fid(mo),
2392 &lock->l_resource->lr_name)) {
2393 CWARN("%s: Although resent, but still not "
2394 "get child lock:"DFID"\n",
2395 info->mti_exp->exp_obd->obd_name,
2396 PFID(mdt_object_fid(mo)));
2397 LDLM_LOCK_PUT(lock);
2400 LDLM_LOCK_PUT(lock);
2406 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2407 struct mdt_object *o, const struct lu_fid *fid,
2408 struct lustre_handle *lh, ldlm_mode_t mode,
2411 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2412 ldlm_policy_data_t *policy = &mti->mti_policy;
2413 struct ldlm_res_id *res_id = &mti->mti_res_id;
2417 LASSERT(mdt_object_remote(o));
2419 LASSERT(ibits == MDS_INODELOCK_UPDATE);
2421 fid_build_reg_res_name(fid, res_id);
2423 memset(einfo, 0, sizeof(*einfo));
2424 einfo->ei_type = LDLM_IBITS;
2425 einfo->ei_mode = mode;
2426 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2427 einfo->ei_cb_cp = ldlm_completion_ast;
2428 einfo->ei_enq_slave = 0;
2429 einfo->ei_res_id = res_id;
2431 memset(policy, 0, sizeof(*policy));
2432 policy->l_inodebits.bits = ibits;
2434 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2439 static int mdt_object_local_lock(struct mdt_thread_info *info,
2440 struct mdt_object *o,
2441 struct mdt_lock_handle *lh, __u64 ibits,
2442 bool nonblock, int locality)
2444 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2445 ldlm_policy_data_t *policy = &info->mti_policy;
2446 struct ldlm_res_id *res_id = &info->mti_res_id;
2451 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2452 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2453 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2454 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2456 /* Only enqueue LOOKUP lock for remote object */
2457 if (mdt_object_remote(o))
2458 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2460 if (lh->mlh_type == MDT_PDO_LOCK) {
2461 /* check for exists after object is locked */
2462 if (mdt_object_exists(o) == 0) {
2463 /* Non-existent object shouldn't have PDO lock */
2466 /* Non-dir object shouldn't have PDO lock */
2467 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2472 memset(policy, 0, sizeof(*policy));
2473 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2475 dlmflags = LDLM_FL_ATOMIC_CB;
2477 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2480 * Take PDO lock on whole directory and build correct @res_id for lock
2481 * on part of directory.
2483 if (lh->mlh_pdo_hash != 0) {
2484 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2485 mdt_lock_pdo_mode(info, o, lh);
2486 if (lh->mlh_pdo_mode != LCK_NL) {
2488 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2489 * is never going to be sent to client and we do not
2490 * want it slowed down due to possible cancels.
2492 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2493 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2494 policy, res_id, dlmflags,
2495 info->mti_exp == NULL ? NULL :
2496 &info->mti_exp->exp_handle.h_cookie);
2497 if (unlikely(rc != 0))
2498 GOTO(out_unlock, rc);
2502 * Finish res_id initializing by name hash marking part of
2503 * directory which is taking modification.
2505 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2508 policy->l_inodebits.bits = ibits;
2511 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2512 * going to be sent to client. If it is - mdt_intent_policy() path will
2513 * fix it up and turn FL_LOCAL flag off.
2515 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2516 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2517 info->mti_exp == NULL ? NULL :
2518 &info->mti_exp->exp_handle.h_cookie);
2521 mdt_object_unlock(info, o, lh, 1);
2522 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2523 lh->mlh_pdo_hash != 0 &&
2524 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2525 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2531 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2532 struct mdt_lock_handle *lh, __u64 ibits,
2533 bool nonblock, int locality)
2538 if (!mdt_object_remote(o))
2539 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2542 if (locality == MDT_LOCAL_LOCK) {
2543 CERROR("%s: try to get local lock for remote object"
2544 DFID".\n", mdt_obd_name(info->mti_mdt),
2545 PFID(mdt_object_fid(o)));
2549 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2550 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2551 MDS_INODELOCK_XATTR);
2552 if (ibits & MDS_INODELOCK_UPDATE) {
2553 /* Sigh, PDO needs to enqueue 2 locks right now, but
2554 * enqueue RPC can only request 1 lock, to avoid extra
2555 * RPC, so it will instead enqueue EX lock for remote
2556 * object anyway XXX*/
2557 if (lh->mlh_type == MDT_PDO_LOCK &&
2558 lh->mlh_pdo_hash != 0) {
2559 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2560 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2561 PFID(mdt_object_fid(o)));
2562 lh->mlh_pdo_hash = 0;
2563 lh->mlh_rreg_mode = LCK_EX;
2564 lh->mlh_type = MDT_REG_LOCK;
2566 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2569 MDS_INODELOCK_UPDATE);
2574 /* Only enqueue LOOKUP lock for remote object */
2575 if (ibits & MDS_INODELOCK_LOOKUP) {
2576 rc = mdt_object_local_lock(info, o, lh,
2577 MDS_INODELOCK_LOOKUP,
2578 nonblock, locality);
2586 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2587 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2589 return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2592 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2593 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2595 struct mdt_lock_handle tmp = *lh;
2598 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2606 * Save a lock within request object.
2608 * Keep the lock referenced until whether client ACK or transaction
2609 * commit happens or release the lock immediately depending on input
2610 * parameters. If COS is ON, a write lock is converted to COS lock
2613 * \param info thead info object
2614 * \param h lock handle
2615 * \param mode lock mode
2616 * \param decref force immediate lock releasing
2619 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2620 ldlm_mode_t mode, int decref)
2624 if (lustre_handle_is_used(h)) {
2625 if (decref || !info->mti_has_trans ||
2626 !(mode & (LCK_PW | LCK_EX))){
2627 mdt_fid_unlock(h, mode);
2629 struct mdt_device *mdt = info->mti_mdt;
2630 struct ldlm_lock *lock = ldlm_handle2lock(h);
2631 struct ptlrpc_request *req = mdt_info_req(info);
2634 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2636 /* there is no request if mdt_object_unlock() is called
2637 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2638 if (likely(req != NULL)) {
2639 CDEBUG(D_HA, "request = %p reply state = %p"
2640 " transno = "LPD64"\n", req,
2641 req->rq_reply_state, req->rq_transno);
2642 if (mdt_cos_is_enabled(mdt)) {
2644 ldlm_lock_downgrade(lock, LCK_COS);
2647 ptlrpc_save_lock(req, h, mode, no_ack);
2649 ldlm_lock_decref(h, mode);
2651 if (mdt_is_lock_sync(lock)) {
2652 CDEBUG(D_HA, "found sync-lock,"
2653 " async commit started\n");
2654 mdt_device_commit_async(info->mti_env,
2657 LDLM_LOCK_PUT(lock);
2666 * Unlock mdt object.
2668 * Immeditely release the regular lock and the PDO lock or save the
2669 * lock in reqeuest and keep them referenced until client ACK or
2670 * transaction commit.
2672 * \param info thread info object
2673 * \param o mdt object
2674 * \param lh mdt lock handle referencing regular and PDO locks
2675 * \param decref force immediate lock releasing
2677 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2678 struct mdt_lock_handle *lh, int decref)
2682 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2683 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2685 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2686 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2691 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2692 const struct lu_fid *f,
2693 struct mdt_lock_handle *lh,
2696 struct mdt_object *o;
2698 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2702 rc = mdt_object_lock(info, o, lh, ibits,
2705 mdt_object_put(info->mti_env, o);
2712 void mdt_object_unlock_put(struct mdt_thread_info * info,
2713 struct mdt_object * o,
2714 struct mdt_lock_handle *lh,
2717 mdt_object_unlock(info, o, lh, decref);
2718 mdt_object_put(info->mti_env, o);
2722 * Generic code handling requests that have struct mdt_body passed in:
2724 * - extract mdt_body from request and save it in @info, if present;
2726 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2729 * - if HABEO_CORPUS flag is set for this request type check whether object
2730 * actually exists on storage (lu_object_exists()).
2733 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2735 const struct mdt_body *body;
2736 struct mdt_object *obj;
2737 const struct lu_env *env;
2738 struct req_capsule *pill;
2742 env = info->mti_env;
2743 pill = info->mti_pill;
2745 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2749 if (!(body->mbo_valid & OBD_MD_FLID))
2752 if (!fid_is_sane(&body->mbo_fid1)) {
2753 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2758 * Do not get size or any capa fields before we check that request
2759 * contains capa actually. There are some requests which do not, for
2760 * instance MDS_IS_SUBDIR.
2762 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2763 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2764 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2765 req_capsule_client_get(pill, &RMF_CAPA1));
2767 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2769 if ((flags & HABEO_CORPUS) &&
2770 !mdt_object_exists(obj)) {
2771 mdt_object_put(env, obj);
2772 /* for capability renew ENOENT will be handled in
2774 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2779 info->mti_object = obj;
2788 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2790 struct req_capsule *pill = info->mti_pill;
2794 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2795 rc = mdt_body_unpack(info, flags);
2799 if (rc == 0 && (flags & HABEO_REFERO)) {
2801 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2802 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2804 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2805 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2808 rc = req_capsule_server_pack(pill);
2813 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2815 struct md_device *next = m->mdt_child;
2817 return next->md_ops->mdo_init_capa_ctxt(env, next,
2818 m->mdt_lut.lut_mds_capa,
2819 m->mdt_capa_timeout,
2824 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2826 lh->mlh_type = MDT_NUL_LOCK;
2827 lh->mlh_reg_lh.cookie = 0ull;
2828 lh->mlh_reg_mode = LCK_MINMODE;
2829 lh->mlh_pdo_lh.cookie = 0ull;
2830 lh->mlh_pdo_mode = LCK_MINMODE;
2831 lh->mlh_rreg_lh.cookie = 0ull;
2832 lh->mlh_rreg_mode = LCK_MINMODE;
2835 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2837 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2838 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2842 * Initialize fields of struct mdt_thread_info. Other fields are left in
2843 * uninitialized state, because it's too expensive to zero out whole
2844 * mdt_thread_info (> 1K) on each request arrival.
2846 void mdt_thread_info_init(struct ptlrpc_request *req,
2847 struct mdt_thread_info *info)
2851 info->mti_pill = &req->rq_pill;
2854 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2855 mdt_lock_handle_init(&info->mti_lh[i]);
2857 /* mdt device: it can be NULL while CONNECT */
2858 if (req->rq_export) {
2859 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2860 info->mti_exp = req->rq_export;
2862 info->mti_mdt = NULL;
2863 info->mti_env = req->rq_svc_thread->t_env;
2864 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2866 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2867 info->mti_big_buf = LU_BUF_NULL;
2868 info->mti_body = NULL;
2869 info->mti_object = NULL;
2870 info->mti_dlm_req = NULL;
2871 info->mti_has_trans = 0;
2872 info->mti_cross_ref = 0;
2873 info->mti_opdata = 0;
2874 info->mti_big_lmm_used = 0;
2876 info->mti_spec.no_create = 0;
2877 info->mti_spec.sp_rm_entry = 0;
2879 info->mti_spec.u.sp_ea.eadata = NULL;
2880 info->mti_spec.u.sp_ea.eadatalen = 0;
2883 void mdt_thread_info_fini(struct mdt_thread_info *info)
2887 if (info->mti_object != NULL) {
2888 mdt_object_put(info->mti_env, info->mti_object);
2889 info->mti_object = NULL;
2892 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2893 mdt_lock_handle_fini(&info->mti_lh[i]);
2894 info->mti_env = NULL;
2895 info->mti_pill = NULL;
2896 info->mti_exp = NULL;
2898 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2899 lu_buf_free(&info->mti_big_buf);
2902 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2904 struct mdt_thread_info *mti;
2905 struct lustre_capa *lc;
2907 mti = mdt_th_info(tsi->tsi_env);
2908 LASSERT(mti != NULL);
2910 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2911 if (tsi->tsi_corpus != NULL) {
2912 struct req_capsule *pill = tsi->tsi_pill;
2914 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2915 lu_object_get(tsi->tsi_corpus);
2918 * XXX: must be part of tgt_mdt_body_unpack but moved here
2919 * due to mdt_set_capainfo().
2921 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2922 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2923 lc = req_capsule_client_get(pill, &RMF_CAPA1);
2924 mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2928 mti->mti_body = tsi->tsi_mdt_body;
2929 mti->mti_dlm_req = tsi->tsi_dlm_req;
2934 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2936 struct ptlrpc_request *req = tgt_ses_req(tsi);
2941 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2943 tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
2944 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
2945 msecs_to_jiffies(3 * MSEC_PER_SEC));
2947 rc = tgt_connect(tsi);
2951 rc = mdt_init_idmap(tsi);
2956 obd_disconnect(class_export_get(req->rq_export));
2975 static int mdt_intent_getattr(enum mdt_it_code opcode,
2976 struct mdt_thread_info *info,
2977 struct ldlm_lock **,
2980 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2981 struct mdt_thread_info *info,
2982 struct ldlm_lock **lockp,
2985 static int mdt_intent_layout(enum mdt_it_code opcode,
2986 struct mdt_thread_info *info,
2987 struct ldlm_lock **,
2989 static int mdt_intent_reint(enum mdt_it_code opcode,
2990 struct mdt_thread_info *info,
2991 struct ldlm_lock **,
2994 static struct mdt_it_flavor {
2995 const struct req_format *it_fmt;
2997 int (*it_act)(enum mdt_it_code ,
2998 struct mdt_thread_info *,
2999 struct ldlm_lock **,
3002 } mdt_it_flavor[] = {
3004 .it_fmt = &RQF_LDLM_INTENT,
3005 /*.it_flags = HABEO_REFERO,*/
3007 .it_act = mdt_intent_reint,
3008 .it_reint = REINT_OPEN
3011 .it_fmt = &RQF_LDLM_INTENT,
3013 * OCREAT is not a MUTABOR request as if the file
3015 * We do the extra check of OBD_CONNECT_RDONLY in
3016 * mdt_reint_open() when we really need to create
3020 .it_act = mdt_intent_reint,
3021 .it_reint = REINT_OPEN
3024 .it_fmt = &RQF_LDLM_INTENT,
3025 .it_flags = MUTABOR,
3026 .it_act = mdt_intent_reint,
3027 .it_reint = REINT_CREATE
3029 [MDT_IT_GETATTR] = {
3030 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3031 .it_flags = HABEO_REFERO,
3032 .it_act = mdt_intent_getattr
3034 [MDT_IT_READDIR] = {
3040 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3041 .it_flags = HABEO_REFERO,
3042 .it_act = mdt_intent_getattr
3045 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3046 .it_flags = MUTABOR,
3048 .it_reint = REINT_UNLINK
3052 .it_flags = MUTABOR,
3055 [MDT_IT_GETXATTR] = {
3056 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
3057 .it_flags = HABEO_CORPUS,
3058 .it_act = mdt_intent_getxattr
3061 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3063 .it_act = mdt_intent_layout
3068 mdt_intent_lock_replace(struct mdt_thread_info *info,
3069 struct ldlm_lock **lockp,
3070 struct mdt_lock_handle *lh,
3073 struct ptlrpc_request *req = mdt_info_req(info);
3074 struct ldlm_lock *lock = *lockp;
3075 struct ldlm_lock *new_lock;
3077 /* If possible resent found a lock, @lh is set to its handle */
3078 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3080 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3081 lh->mlh_reg_lh.cookie = 0;
3085 LASSERTF(new_lock != NULL,
3086 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3089 * If we've already given this lock to a client once, then we should
3090 * have no readers or writers. Otherwise, we should have one reader
3091 * _or_ writer ref (which will be zeroed below) before returning the
3094 if (new_lock->l_export == req->rq_export) {
3095 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3097 LASSERT(new_lock->l_export == NULL);
3098 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3103 if (new_lock->l_export == req->rq_export) {
3105 * Already gave this to the client, which means that we
3106 * reconstructed a reply.
3108 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3111 LDLM_LOCK_RELEASE(new_lock);
3112 lh->mlh_reg_lh.cookie = 0;
3113 RETURN(ELDLM_LOCK_REPLACED);
3117 * Fixup the lock to be given to the client.
3119 lock_res_and_lock(new_lock);
3120 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3121 * possible blocking AST. */
3122 while (new_lock->l_readers > 0) {
3123 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3124 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3125 new_lock->l_readers--;
3127 while (new_lock->l_writers > 0) {
3128 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3129 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3130 new_lock->l_writers--;
3133 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3134 new_lock->l_blocking_ast = lock->l_blocking_ast;
3135 new_lock->l_completion_ast = lock->l_completion_ast;
3136 new_lock->l_remote_handle = lock->l_remote_handle;
3137 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3139 unlock_res_and_lock(new_lock);
3141 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3142 &new_lock->l_remote_handle,
3143 &new_lock->l_exp_hash);
3145 LDLM_LOCK_RELEASE(new_lock);
3146 lh->mlh_reg_lh.cookie = 0;
3148 RETURN(ELDLM_LOCK_REPLACED);
3151 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3152 struct ldlm_lock *new_lock,
3153 struct mdt_lock_handle *lh,
3156 struct ptlrpc_request *req = mdt_info_req(info);
3157 struct ldlm_request *dlmreq;
3159 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3162 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3164 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3165 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3167 if (flags & LDLM_FL_RESENT) {
3168 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3169 lh->mlh_reg_mode = new_lock->l_granted_mode;
3171 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3172 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3173 lh->mlh_reg_lh.cookie);
3178 * If the xid matches, then we know this is a resent request, and allow
3179 * it. (It's probably an OPEN, for which we don't send a lock.
3181 if (req_xid_is_last(req))
3185 * This remote handle isn't enqueued, so we never received or processed
3186 * this request. Clear MSG_RESENT, because it can be handled like any
3187 * normal request now.
3189 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3191 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3192 dlmreq->lock_handle[0].cookie);
3195 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3196 struct mdt_thread_info *info,
3197 struct ldlm_lock **lockp,
3200 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3201 struct ldlm_reply *ldlm_rep = NULL;
3205 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3206 * (for the resend case) or a new lock. Below we will use it to
3207 * replace the original lock.
3209 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3210 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3211 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3212 rc = mdt_object_lock(info, info->mti_object, lhc,
3213 MDS_INODELOCK_XATTR,
3219 grc = mdt_getxattr(info);
3221 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3223 if (mdt_info_req(info)->rq_repmsg != NULL)
3224 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3225 if (ldlm_rep == NULL)
3226 RETURN(err_serious(-EFAULT));
3228 ldlm_rep->lock_policy_res2 = grc;
3233 static int mdt_intent_getattr(enum mdt_it_code opcode,
3234 struct mdt_thread_info *info,
3235 struct ldlm_lock **lockp,
3238 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3240 struct ldlm_reply *ldlm_rep;
3241 struct mdt_body *reqbody;
3242 struct mdt_body *repbody;
3246 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3249 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3252 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3253 repbody->mbo_eadatasize = 0;
3254 repbody->mbo_aclsize = 0;
3258 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3260 case MDT_IT_GETATTR:
3261 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3265 CERROR("Unsupported intent (%d)\n", opcode);
3266 GOTO(out_shrink, rc = -EINVAL);
3269 rc = mdt_init_ucred(info, reqbody);
3271 GOTO(out_shrink, rc);
3273 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3274 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3276 /* Get lock from request for possible resent case. */
3277 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3279 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3280 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3282 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3283 ldlm_rep->lock_policy_res2 = 0;
3284 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3285 ldlm_rep->lock_policy_res2) {
3286 lhc->mlh_reg_lh.cookie = 0ull;
3287 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3290 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3293 mdt_exit_ucred(info);
3295 mdt_client_compatibility(info);
3296 rc2 = mdt_fix_reply(info);
3302 static int mdt_intent_layout(enum mdt_it_code opcode,
3303 struct mdt_thread_info *info,
3304 struct ldlm_lock **lockp,
3307 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3308 struct layout_intent *layout;
3310 struct mdt_object *obj = NULL;
3314 if (opcode != MDT_IT_LAYOUT) {
3315 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3320 fid = &info->mti_tmp_fid2;
3321 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3323 /* Get lock from request for possible resent case. */
3324 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3326 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3328 RETURN(PTR_ERR(obj));
3330 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3331 /* get the length of lsm */
3332 rc = mdt_attr_get_eabuf_size(info, obj);
3334 mdt_object_put(info->mti_env, obj);
3338 if (rc > info->mti_mdt->mdt_max_mdsize)
3339 info->mti_mdt->mdt_max_mdsize = rc;
3342 mdt_object_put(info->mti_env, obj);
3344 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3345 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3346 rc = req_capsule_server_pack(info->mti_pill);
3350 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3351 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3353 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3354 LASSERT(layout != NULL);
3355 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3356 /* return to normal/resent ldlm handling */
3359 CERROR("%s: Unsupported layout intent (%d)\n",
3360 mdt_obd_name(info->mti_mdt), layout->li_opc);
3364 static int mdt_intent_reint(enum mdt_it_code opcode,
3365 struct mdt_thread_info *info,
3366 struct ldlm_lock **lockp,
3369 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3370 struct ldlm_reply *rep = NULL;
3374 static const struct req_format *intent_fmts[REINT_MAX] = {
3375 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3376 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3381 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3385 if (mdt_it_flavor[opcode].it_reint != opc) {
3386 CERROR("Reint code %ld doesn't match intent: %d\n",
3388 RETURN(err_serious(-EPROTO));
3391 /* Get lock from request for possible resent case. */
3392 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3394 rc = mdt_reint_internal(info, lhc, opc);
3396 /* Check whether the reply has been packed successfully. */
3397 if (mdt_info_req(info)->rq_repmsg != NULL)
3398 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3400 RETURN(err_serious(-EFAULT));
3402 /* MDC expects this in any case */
3404 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3406 /* the open lock or the lock for cross-ref object should be
3407 * returned to the client */
3408 if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3409 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3410 rep->lock_policy_res2 = 0;
3411 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3415 rep->lock_policy_res2 = clear_serious(rc);
3417 if (rep->lock_policy_res2 == -ENOENT &&
3418 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3419 rep->lock_policy_res2 = 0;
3421 if (rc == -ENOTCONN || rc == -ENODEV ||
3422 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3424 * If it is the disconnect error (ENODEV & ENOCONN), the error
3425 * will be returned by rq_status, and client at ptlrpc layer
3426 * will detect this, then disconnect, reconnect the import
3427 * immediately, instead of impacting the following the rpc.
3429 lhc->mlh_reg_lh.cookie = 0ull;
3433 * For other cases, the error will be returned by intent.
3434 * and client will retrieve the result from intent.
3437 * FIXME: when open lock is finished, that should be
3440 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3441 LASSERTF(rc == 0, "Error occurred but lock handle "
3442 "is still in use, rc = %d\n", rc);
3443 rep->lock_policy_res2 = 0;
3444 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3447 lhc->mlh_reg_lh.cookie = 0ull;
3448 RETURN(ELDLM_LOCK_ABORTED);
3453 static int mdt_intent_code(long itcode)
3461 case IT_OPEN|IT_CREAT:
3468 rc = MDT_IT_READDIR;
3471 rc = MDT_IT_GETATTR;
3483 rc = MDT_IT_GETXATTR;
3488 case IT_QUOTA_DQACQ:
3493 CERROR("Unknown intent opcode: %ld\n", itcode);
3500 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3501 struct ldlm_lock **lockp, __u64 flags)
3503 struct req_capsule *pill = info->mti_pill;
3504 struct ptlrpc_request *req = mdt_info_req(info);
3505 struct mdt_it_flavor *flv;
3510 opc = mdt_intent_code(itopc);
3514 if (opc == MDT_IT_QUOTA) {
3515 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3518 RETURN(-EOPNOTSUPP);
3520 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3521 /* pass the request to quota master */
3522 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3523 mdt_info_req(info), lockp,
3528 flv = &mdt_it_flavor[opc];
3529 if (flv->it_fmt != NULL)
3530 req_capsule_extend(pill, flv->it_fmt);
3532 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3536 if (flv->it_flags & MUTABOR &&
3537 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3540 if (flv->it_act != NULL) {
3541 struct ldlm_reply *rep;
3543 /* execute policy */
3544 rc = flv->it_act(opc, info, lockp, flags);
3546 /* Check whether the reply has been packed successfully. */
3547 if (req->rq_repmsg != NULL) {
3548 rep = req_capsule_server_get(info->mti_pill,
3550 rep->lock_policy_res2 =
3551 ptlrpc_status_hton(rep->lock_policy_res2);
3558 static int mdt_intent_policy(struct ldlm_namespace *ns,
3559 struct ldlm_lock **lockp, void *req_cookie,
3560 ldlm_mode_t mode, __u64 flags, void *data)
3562 struct tgt_session_info *tsi;
3563 struct mdt_thread_info *info;
3564 struct ptlrpc_request *req = req_cookie;
3565 struct ldlm_intent *it;
3566 struct req_capsule *pill;
3571 LASSERT(req != NULL);
3573 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3575 info = tsi2mdt_info(tsi);
3576 LASSERT(info != NULL);
3577 pill = info->mti_pill;
3578 LASSERT(pill->rc_req == req);
3580 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3581 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3582 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3584 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3588 /* Lock without inodebits makes no sense and will oops
3589 * later in ldlm. Let's check it now to see if we have
3590 * ibits corrupted somewhere in mdt_intent_opc().
3591 * The case for client miss to set ibits has been
3592 * processed by others. */
3593 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3594 lr_type == LDLM_IBITS,
3595 info->mti_dlm_req->lock_desc.\
3596 l_policy_data.l_inodebits.bits != 0));
3598 rc = err_serious(-EFAULT);
3600 /* No intent was provided */
3601 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3602 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3603 rc = req_capsule_server_pack(pill);
3605 rc = err_serious(rc);
3607 mdt_thread_info_fini(info);
3611 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3613 struct seq_server_site *ss = mdt_seq_site(mdt);
3615 if (ss->ss_node_id == 0)
3618 if (ss->ss_client_seq != NULL) {
3619 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3620 ss->ss_client_seq->lcs_exp = NULL;
3623 if (ss->ss_server_fld != NULL) {
3624 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3625 ss->ss_server_fld->lsf_control_exp = NULL;
3629 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3631 struct seq_server_site *ss = mdt_seq_site(mdt);
3636 if (ss->ss_server_seq != NULL)
3637 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3640 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3642 mdt_seq_fini_cli(mdt);
3643 mdt_deregister_seq_exp(mdt);
3645 return seq_site_fini(env, mdt_seq_site(mdt));
3649 * It will retrieve its FLDB entries from MDT0, and it only happens
3650 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3651 * and it needs to refresh FLDB from the MDT0.
3653 static int mdt_register_lwp_callback(void *data)
3656 struct mdt_device *mdt = data;
3657 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3661 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3663 if (!likely(fld->lsf_new))
3666 rc = lu_env_init(&env, LCT_MD_THREAD);
3668 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3672 rc = fld_update_from_controller(&env, fld);
3674 CERROR("%s: cannot update controller: rc = %d\n",
3675 mdt_obd_name(mdt), rc);
3683 static int mdt_register_seq_exp(struct mdt_device *mdt)
3685 struct seq_server_site *ss = mdt_seq_site(mdt);
3686 char *lwp_name = NULL;
3689 if (ss->ss_node_id == 0)
3692 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3693 if (lwp_name == NULL)
3694 GOTO(out_free, rc = -ENOMEM);
3696 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3700 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3705 rc = lustre_register_lwp_item(lwp_name,
3706 &ss->ss_server_fld->lsf_control_exp,
3707 mdt_register_lwp_callback, mdt);
3709 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3710 ss->ss_client_seq->lcs_exp = NULL;
3714 if (lwp_name != NULL)
3715 OBD_FREE(lwp_name, MAX_OBD_NAME);
3721 * Init client sequence manager which is used by local MDS to talk to sequence
3722 * controller on remote node.
3724 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3726 struct seq_server_site *ss = mdt_seq_site(mdt);
3731 /* check if this is adding the first MDC and controller is not yet
3733 OBD_ALLOC_PTR(ss->ss_client_seq);
3734 if (ss->ss_client_seq == NULL)
3737 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3738 if (prefix == NULL) {
3739 OBD_FREE_PTR(ss->ss_client_seq);
3740 ss->ss_client_seq = NULL;
3744 /* Note: seq_client_fini will be called in seq_site_fini */
3745 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3746 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3747 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3749 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3751 OBD_FREE_PTR(ss->ss_client_seq);
3752 ss->ss_client_seq = NULL;
3756 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3761 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3763 struct seq_server_site *ss;
3767 ss = mdt_seq_site(mdt);
3768 /* init sequence controller server(MDT0) */
3769 if (ss->ss_node_id == 0) {
3770 OBD_ALLOC_PTR(ss->ss_control_seq);
3771 if (ss->ss_control_seq == NULL)
3774 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3775 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3778 GOTO(out_seq_fini, rc);
3781 /* Init normal sequence server */
3782 OBD_ALLOC_PTR(ss->ss_server_seq);
3783 if (ss->ss_server_seq == NULL)
3784 GOTO(out_seq_fini, rc = -ENOMEM);
3786 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3787 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3789 GOTO(out_seq_fini, rc);
3791 /* init seq client for seq server to talk to seq controller(MDT0) */
3792 rc = mdt_seq_init_cli(env, mdt);
3794 GOTO(out_seq_fini, rc);
3796 if (ss->ss_node_id != 0)
3797 /* register controler export through lwp */
3798 rc = mdt_register_seq_exp(mdt);
3803 mdt_seq_fini(env, mdt);
3811 static int mdt_fld_fini(const struct lu_env *env,
3812 struct mdt_device *m)
3814 struct seq_server_site *ss = mdt_seq_site(m);
3817 if (ss && ss->ss_server_fld) {
3818 fld_server_fini(env, ss->ss_server_fld);
3819 OBD_FREE_PTR(ss->ss_server_fld);
3820 ss->ss_server_fld = NULL;
3826 static int mdt_fld_init(const struct lu_env *env,
3828 struct mdt_device *m)
3830 struct seq_server_site *ss;
3834 ss = mdt_seq_site(m);
3836 OBD_ALLOC_PTR(ss->ss_server_fld);
3837 if (ss->ss_server_fld == NULL)
3838 RETURN(rc = -ENOMEM);
3840 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3843 OBD_FREE_PTR(ss->ss_server_fld);
3844 ss->ss_server_fld = NULL;
3851 static void mdt_stack_pre_fini(const struct lu_env *env,
3852 struct mdt_device *m, struct lu_device *top)
3854 struct lustre_cfg_bufs *bufs;
3855 struct lustre_cfg *lcfg;
3856 struct mdt_thread_info *info;
3861 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3862 LASSERT(info != NULL);
3864 bufs = &info->mti_u.bufs;
3866 LASSERT(m->mdt_child_exp);
3867 LASSERT(m->mdt_child_exp->exp_obd);
3869 /* process cleanup, pass mdt obd name to get obd umount flags */
3870 /* XXX: this is needed because all layers are referenced by
3871 * objects (some of them are pinned by osd, for example *
3872 * the proper solution should be a model where object used
3873 * by osd only doesn't have mdt/mdd slices -bzzz */
3874 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3875 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3876 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3878 CERROR("%s: cannot alloc lcfg\n", mdt_obd_name(m));
3881 top->ld_ops->ldo_process_config(env, top, lcfg);
3882 lustre_cfg_free(lcfg);
3886 static void mdt_stack_fini(const struct lu_env *env,
3887 struct mdt_device *m, struct lu_device *top)
3889 struct obd_device *obd = mdt2obd_dev(m);
3890 struct lustre_cfg_bufs *bufs;
3891 struct lustre_cfg *lcfg;
3892 struct mdt_thread_info *info;
3896 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3897 LASSERT(info != NULL);
3899 lu_dev_del_linkage(top->ld_site, top);
3901 lu_site_purge(env, top->ld_site, -1);
3903 bufs = &info->mti_u.bufs;
3904 /* process cleanup, pass mdt obd name to get obd umount flags */
3905 /* another purpose is to let all layers to release their objects */
3906 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3911 lustre_cfg_bufs_set_string(bufs, 1, flags);
3912 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3914 CERROR("Cannot alloc lcfg!\n");
3918 top->ld_ops->ldo_process_config(env, top, lcfg);
3919 lustre_cfg_free(lcfg);
3921 lu_site_purge(env, top->ld_site, -1);
3923 m->mdt_child = NULL;
3924 m->mdt_bottom = NULL;
3926 obd_disconnect(m->mdt_child_exp);
3927 m->mdt_child_exp = NULL;
3929 obd_disconnect(m->mdt_bottom_exp);
3930 m->mdt_child_exp = NULL;
3933 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3934 const char *next, struct obd_export **exp)
3936 struct obd_connect_data *data = NULL;
3937 struct obd_device *obd;
3941 OBD_ALLOC_PTR(data);
3943 GOTO(out, rc = -ENOMEM);
3945 obd = class_name2obd(next);
3947 CERROR("%s: can't locate next device: %s\n",
3948 mdt_obd_name(m), next);
3949 GOTO(out, rc = -ENOTCONN);
3952 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3953 data->ocd_version = LUSTRE_VERSION_CODE;
3955 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3957 CERROR("%s: cannot connect to next dev %s (%d)\n",
3958 mdt_obd_name(m), next, rc);
3968 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3969 struct lustre_cfg *cfg)
3971 char *dev = lustre_cfg_string(cfg, 0);
3972 int rc, name_size, uuid_size;
3973 char *name, *uuid, *p;
3974 struct lustre_cfg_bufs *bufs;
3975 struct lustre_cfg *lcfg;
3976 struct obd_device *obd;
3977 struct lustre_profile *lprof;
3978 struct lu_site *site;
3981 /* in 1.8 we had the only device in the stack - MDS.
3982 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3983 * in 2.3 OSD is instantiated by obd_mount.c, so we need
3984 * to generate names and setup MDT, MDD. MDT will be using
3985 * generated name to connect to MDD. for MDD the next device
3986 * will be LOD with name taken from so called "profile" which
3987 * is generated by mount_option line
3989 * 1.8 MGS generates config. commands like this:
3990 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
3991 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
3992 * 2.0 MGS generates config. commands like this:
3993 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
3994 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3995 * 3:lustre-MDT0000-mdtlov 4:f
3997 * we generate MDD name from MDT one, just replacing T with D
3999 * after all the preparations, the logical equivalent will be
4000 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
4001 * 3:lustre-MDT0000-mdtlov 4:f
4002 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4003 * 3:lustre-MDD0000 4:f
4005 * notice we build the stack from down to top: MDD first, then MDT */
4007 name_size = MAX_OBD_NAME;
4008 uuid_size = MAX_OBD_NAME;
4010 OBD_ALLOC(name, name_size);
4011 OBD_ALLOC(uuid, uuid_size);
4012 if (name == NULL || uuid == NULL)
4013 GOTO(cleanup_mem, rc = -ENOMEM);
4015 OBD_ALLOC_PTR(bufs);
4017 GOTO(cleanup_mem, rc = -ENOMEM);
4020 p = strstr(name, "-MDT");
4022 GOTO(free_bufs, rc = -ENOMEM);
4025 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4027 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4028 if (lprof == NULL || lprof->lp_dt == NULL) {
4029 CERROR("can't find the profile: %s\n",
4030 lustre_cfg_string(cfg, 0));
4031 GOTO(free_bufs, rc = -EINVAL);
4034 lustre_cfg_bufs_reset(bufs, name);
4035 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4036 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4037 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4039 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4041 GOTO(free_bufs, rc = -ENOMEM);
4043 rc = class_attach(lcfg);
4045 GOTO(lcfg_cleanup, rc);
4047 obd = class_name2obd(name);
4049 CERROR("Can not find obd %s (%s in config)\n",
4050 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4051 GOTO(lcfg_cleanup, rc = -EINVAL);
4054 lustre_cfg_free(lcfg);
4056 lustre_cfg_bufs_reset(bufs, name);
4057 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4058 lustre_cfg_bufs_set_string(bufs, 2, dev);
4059 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4061 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4063 rc = class_setup(obd, lcfg);
4065 GOTO(class_detach, rc);
4067 /* connect to MDD we just setup */
4068 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4070 GOTO(class_detach, rc);
4072 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4074 LASSERT(mdt_lu_site(mdt) == NULL);
4075 mdt->mdt_lu_dev.ld_site = site;
4076 site->ls_top_dev = &mdt->mdt_lu_dev;
4077 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4079 /* now connect to bottom OSD */
4080 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4081 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4083 GOTO(class_detach, rc);
4085 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4087 rc = lu_env_refill((struct lu_env *)env);
4089 CERROR("Failure to refill session: '%d'\n", rc);
4091 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4096 class_detach(obd, lcfg);
4098 lustre_cfg_free(lcfg);
4103 OBD_FREE(name, name_size);
4105 OBD_FREE(uuid, uuid_size);
4109 /* setup quota master target on MDT0 */
4110 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4111 struct lustre_cfg *cfg)
4113 struct obd_device *obd;
4114 char *dev = lustre_cfg_string(cfg, 0);
4115 char *qmtname, *uuid, *p;
4116 struct lustre_cfg_bufs *bufs;
4117 struct lustre_cfg *lcfg;
4118 struct lustre_profile *lprof;
4119 struct obd_connect_data *data;
4123 LASSERT(mdt->mdt_qmt_exp == NULL);
4124 LASSERT(mdt->mdt_qmt_dev == NULL);
4126 /* quota master is on MDT0 only for now */
4127 if (mdt->mdt_seq_site.ss_node_id != 0)
4130 /* MGS generates config commands which look as follows:
4131 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4132 * 3:lustre-MDT0000-mdtlov 4:f
4134 * We generate the QMT name from the MDT one, just replacing MD with QM
4135 * after all the preparations, the logical equivalent will be:
4136 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4137 * 3:lustre-MDT0000-osd 4:f */
4138 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4139 OBD_ALLOC(uuid, UUID_MAX);
4140 OBD_ALLOC_PTR(bufs);
4141 OBD_ALLOC_PTR(data);
4142 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4143 GOTO(cleanup_mem, rc = -ENOMEM);
4145 strcpy(qmtname, dev);
4146 p = strstr(qmtname, "-MDT");
4148 GOTO(cleanup_mem, rc = -ENOMEM);
4149 /* replace MD with QM */
4153 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4155 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4156 if (lprof == NULL || lprof->lp_dt == NULL) {
4157 CERROR("can't find profile for %s\n",
4158 lustre_cfg_string(cfg, 0));
4159 GOTO(cleanup_mem, rc = -EINVAL);
4162 lustre_cfg_bufs_reset(bufs, qmtname);
4163 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4164 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4165 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4167 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4169 GOTO(cleanup_mem, rc = -ENOMEM);
4171 rc = class_attach(lcfg);
4173 GOTO(lcfg_cleanup, rc);
4175 obd = class_name2obd(qmtname);
4177 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4178 lustre_cfg_string(cfg, 0));
4179 GOTO(lcfg_cleanup, rc = -EINVAL);
4182 lustre_cfg_free(lcfg);
4184 lustre_cfg_bufs_reset(bufs, qmtname);
4185 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4186 lustre_cfg_bufs_set_string(bufs, 2, dev);
4188 /* for quota, the next device should be the OSD device */
4189 lustre_cfg_bufs_set_string(bufs, 3,
4190 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4192 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4194 rc = class_setup(obd, lcfg);
4196 GOTO(class_detach, rc);
4198 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4200 /* configure local quota objects */
4201 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4205 GOTO(class_cleanup, rc);
4207 /* connect to quota master target */
4208 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4209 data->ocd_version = LUSTRE_VERSION_CODE;
4210 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4213 CERROR("cannot connect to quota master device %s (%d)\n",
4215 GOTO(class_cleanup, rc);
4221 class_manual_cleanup(obd);
4222 mdt->mdt_qmt_dev = NULL;
4226 class_detach(obd, lcfg);
4228 lustre_cfg_free(lcfg);
4233 OBD_FREE(qmtname, MAX_OBD_NAME);
4235 OBD_FREE(uuid, UUID_MAX);
4241 /* Shutdown quota master target associated with mdt */
4242 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4246 if (mdt->mdt_qmt_exp == NULL)
4248 LASSERT(mdt->mdt_qmt_dev != NULL);
4250 /* the qmt automatically shuts down when the mdt disconnects */
4251 obd_disconnect(mdt->mdt_qmt_exp);
4252 mdt->mdt_qmt_exp = NULL;
4253 mdt->mdt_qmt_dev = NULL;
4257 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4258 * for now. This will be removed along with converting rest of MDT code
4259 * to use tgt_session_info */
4260 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4262 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4265 rc = mdt_getxattr(info);
4267 mdt_thread_info_fini(info);
4271 static struct tgt_handler mdt_tgt_handlers[] = {
4272 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4273 0, MDS_CONNECT, mdt_tgt_connect,
4274 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4275 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4276 0, MDS_DISCONNECT, tgt_disconnect,
4277 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4278 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4279 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4280 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4281 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4282 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus),
4283 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4284 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4286 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4287 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4288 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4289 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4290 TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING,
4292 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4293 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4294 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4295 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4297 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4298 mdt_hsm_ct_register),
4299 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4300 mdt_hsm_ct_unregister),
4301 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4303 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4305 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4306 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4308 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4313 static struct tgt_handler mdt_sec_ctx_ops[] = {
4314 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4315 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4316 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4319 static struct tgt_handler mdt_quota_ops[] = {
4320 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4323 static struct tgt_opc_slice mdt_common_slice[] = {
4325 .tos_opc_start = MDS_FIRST_OPC,
4326 .tos_opc_end = MDS_LAST_OPC,
4327 .tos_hs = mdt_tgt_handlers
4330 .tos_opc_start = OBD_FIRST_OPC,
4331 .tos_opc_end = OBD_LAST_OPC,
4332 .tos_hs = tgt_obd_handlers
4335 .tos_opc_start = LDLM_FIRST_OPC,
4336 .tos_opc_end = LDLM_LAST_OPC,
4337 .tos_hs = tgt_dlm_handlers
4340 .tos_opc_start = SEC_FIRST_OPC,
4341 .tos_opc_end = SEC_LAST_OPC,
4342 .tos_hs = mdt_sec_ctx_ops
4345 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4346 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4347 .tos_hs = tgt_out_handlers
4350 .tos_opc_start = FLD_FIRST_OPC,
4351 .tos_opc_end = FLD_LAST_OPC,
4352 .tos_hs = fld_handlers
4355 .tos_opc_start = SEQ_FIRST_OPC,
4356 .tos_opc_end = SEQ_LAST_OPC,
4357 .tos_hs = seq_handlers
4360 .tos_opc_start = QUOTA_DQACQ,
4361 .tos_opc_end = QUOTA_LAST_OPC,
4362 .tos_hs = mdt_quota_ops
4365 .tos_opc_start = LLOG_FIRST_OPC,
4366 .tos_opc_end = LLOG_LAST_OPC,
4367 .tos_hs = tgt_llog_handlers
4370 .tos_opc_start = LFSCK_FIRST_OPC,
4371 .tos_opc_end = LFSCK_LAST_OPC,
4372 .tos_hs = tgt_lfsck_handlers
4380 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4382 struct md_device *next = m->mdt_child;
4383 struct lu_device *d = &m->mdt_lu_dev;
4384 struct obd_device *obd = mdt2obd_dev(m);
4385 struct lfsck_stop stop;
4388 stop.ls_status = LS_PAUSED;
4390 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4392 target_recovery_fini(obd);
4393 ping_evictor_stop();
4394 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4396 if (m->mdt_opts.mo_coordinator)
4397 mdt_hsm_cdt_stop(m);
4399 mdt_hsm_cdt_fini(m);
4401 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4402 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4403 obd_exports_barrier(obd);
4404 obd_zombie_barrier();
4408 tgt_fini(env, &m->mdt_lut);
4409 mdt_fs_cleanup(env, m);
4410 upcall_cache_cleanup(m->mdt_identity_cache);
4411 m->mdt_identity_cache = NULL;
4413 if (m->mdt_namespace != NULL) {
4414 ldlm_namespace_free(m->mdt_namespace, NULL,
4415 d->ld_obd->obd_force);
4416 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4419 mdt_quota_fini(env, m);
4421 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4423 mdt_seq_fini(env, m);
4424 mdt_fld_fini(env, m);
4426 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4427 cfs_timer_disarm(&m->mdt_ck_timer);
4428 mdt_ck_thread_stop(m);
4433 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4435 LASSERT(atomic_read(&d->ld_ref) == 0);
4437 server_put_mount(mdt_obd_name(m), true);
4442 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4444 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4445 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4447 struct mdt_thread_info *info;
4448 struct obd_device *obd;
4449 const char *dev = lustre_cfg_string(cfg, 0);
4450 const char *num = lustre_cfg_string(cfg, 2);
4451 struct lustre_mount_info *lmi = NULL;
4452 struct lustre_sb_info *lsi;
4454 struct seq_server_site *ss_site;
4455 const char *identity_upcall = "NONE";
4456 struct md_device *next;
4462 lu_device_init(&m->mdt_lu_dev, ldt);
4464 * Environment (env) might be missing mdt_thread_key values at that
4465 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4468 * Usually device allocation path doesn't use module key values, but
4469 * mdt has to do a lot of work here, so allocate key value.
4471 rc = lu_env_refill((struct lu_env *)env);
4475 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4476 LASSERT(info != NULL);
4478 obd = class_name2obd(dev);
4479 LASSERT(obd != NULL);
4481 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4483 m->mdt_som_conf = 0;
4485 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4487 /* default is coordinator off, it is started through conf_param
4489 m->mdt_opts.mo_coordinator = 0;
4491 lmi = server_get_mount(dev);
4493 CERROR("Cannot get mount info for %s!\n", dev);
4496 lsi = s2lsi(lmi->lmi_sb);
4497 /* CMD is supported only in IAM mode */
4499 node_id = simple_strtol(num, NULL, 10);
4500 obd->u.obt.obt_magic = OBT_MAGIC;
4503 spin_lock_init(&m->mdt_ioepoch_lock);
4504 m->mdt_capa_timeout = CAPA_TIMEOUT;
4505 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4506 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4507 m->mdt_squash.rsi_uid = 0;
4508 m->mdt_squash.rsi_gid = 0;
4509 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4510 init_rwsem(&m->mdt_squash.rsi_sem);
4511 spin_lock_init(&m->mdt_osfs_lock);
4512 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4513 m->mdt_enable_remote_dir = 0;
4514 m->mdt_enable_remote_dir_gid = 0;
4516 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4517 m->mdt_lu_dev.ld_obd = obd;
4518 /* Set this lu_device to obd for error handling purposes. */
4519 obd->obd_lu_dev = &m->mdt_lu_dev;
4521 /* init the stack */
4522 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4524 CERROR("%s: Can't init device stack, rc %d\n",
4525 mdt_obd_name(m), rc);
4530 ss_site = mdt_seq_site(m);
4531 s->ld_seq_site = ss_site;
4534 /* set server index */
4535 ss_site->ss_node_id = node_id;
4537 /* failover is the default
4538 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4539 * assumed whose ss_node_id == 0 XXX
4541 obd->obd_replayable = 1;
4542 /* No connection accepted until configurations will finish */
4543 obd->obd_no_conn = 1;
4545 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4546 char *str = lustre_cfg_string(cfg, 4);
4547 if (strchr(str, 'n')) {
4548 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4549 obd->obd_replayable = 0;
4553 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4555 GOTO(err_fini_stack, rc);
4557 rc = mdt_seq_init(env, m);
4559 GOTO(err_fini_fld, rc);
4561 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4562 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4563 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4564 LDLM_NAMESPACE_SERVER,
4565 LDLM_NAMESPACE_GREEDY,
4567 if (m->mdt_namespace == NULL)
4568 GOTO(err_fini_seq, rc = -ENOMEM);
4570 m->mdt_namespace->ns_lvbp = m;
4571 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4573 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4574 /* set obd_namespace for compatibility with old code */
4575 obd->obd_namespace = m->mdt_namespace;
4577 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4579 rc = mdt_hsm_cdt_init(m);
4581 CERROR("%s: error initializing coordinator, rc %d\n",
4582 mdt_obd_name(m), rc);
4583 GOTO(err_free_ns, rc);
4586 rc = mdt_ck_thread_start(m);
4588 GOTO(err_free_hsm, rc);
4590 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4591 OBD_FAIL_MDS_ALL_REQUEST_NET,
4592 OBD_FAIL_MDS_ALL_REPLY_NET);
4596 rc = mdt_fs_setup(env, m, obd, lsi);
4600 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4602 next = m->mdt_child;
4603 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4606 GOTO(err_fs_cleanup, rc);
4608 if (mntopts & MNTOPT_USERXATTR)
4609 m->mdt_opts.mo_user_xattr = 1;
4611 m->mdt_opts.mo_user_xattr = 0;
4613 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4615 GOTO(err_fs_cleanup, rc);
4617 if (mntopts & MNTOPT_ACL)
4618 m->mdt_opts.mo_acl = 1;
4620 m->mdt_opts.mo_acl = 0;
4622 /* XXX: to support suppgid for ACL, we enable identity_upcall
4623 * by default, otherwise, maybe got unexpected -EACCESS. */
4624 if (m->mdt_opts.mo_acl)
4625 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4627 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4629 &mdt_identity_upcall_cache_ops);
4630 if (IS_ERR(m->mdt_identity_cache)) {
4631 rc = PTR_ERR(m->mdt_identity_cache);
4632 m->mdt_identity_cache = NULL;
4633 GOTO(err_fs_cleanup, rc);
4636 rc = mdt_procfs_init(m, dev);
4638 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4639 GOTO(err_recovery, rc);
4642 rc = mdt_quota_init(env, m, cfg);
4644 GOTO(err_procfs, rc);
4646 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4647 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4648 "mdt_ldlm_client", m->mdt_ldlm_client);
4650 ping_evictor_start();
4652 /* recovery will be started upon mdt_prepare()
4653 * when the whole stack is complete and ready
4654 * to serve the requests */
4656 mdt_init_capa_ctxt(env, m);
4658 /* Reduce the initial timeout on an MDS because it doesn't need such
4659 * a long timeout as an OST does. Adaptive timeouts will adjust this
4660 * value appropriately. */
4661 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4662 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4668 target_recovery_fini(obd);
4669 upcall_cache_cleanup(m->mdt_identity_cache);
4670 m->mdt_identity_cache = NULL;
4672 mdt_fs_cleanup(env, m);
4674 tgt_fini(env, &m->mdt_lut);
4676 cfs_timer_disarm(&m->mdt_ck_timer);
4677 mdt_ck_thread_stop(m);
4679 mdt_hsm_cdt_fini(m);
4681 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4682 obd->obd_namespace = m->mdt_namespace = NULL;
4684 mdt_seq_fini(env, m);
4686 mdt_fld_fini(env, m);
4688 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4691 server_put_mount(dev, true);
4695 /* For interoperability, the left element is old parameter, the right one
4696 * is the new version of the parameter, if some parameter is deprecated,
4697 * the new version should be set as NULL. */
4698 static struct cfg_interop_param mdt_interop_param[] = {
4699 { "mdt.group_upcall", NULL },
4700 { "mdt.quota_type", NULL },
4701 { "mdd.quota_type", NULL },
4702 { "mdt.rootsquash", "mdt.root_squash" },
4703 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4707 /* used by MGS to process specific configurations */
4708 static int mdt_process_config(const struct lu_env *env,
4709 struct lu_device *d, struct lustre_cfg *cfg)
4711 struct mdt_device *m = mdt_dev(d);
4712 struct md_device *md_next = m->mdt_child;
4713 struct lu_device *next = md2lu_dev(md_next);
4717 switch (cfg->lcfg_command) {
4719 struct obd_device *obd = d->ld_obd;
4721 /* For interoperability */
4722 struct cfg_interop_param *ptr = NULL;
4723 struct lustre_cfg *old_cfg = NULL;
4726 param = lustre_cfg_string(cfg, 1);
4727 if (param == NULL) {
4728 CERROR("param is empty\n");
4733 ptr = class_find_old_param(param, mdt_interop_param);
4735 if (ptr->new_param == NULL) {
4737 CWARN("For interoperability, skip this %s."
4738 " It is obsolete.\n", ptr->old_param);
4742 CWARN("Found old param %s, changed it to %s.\n",
4743 ptr->old_param, ptr->new_param);
4746 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4753 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4755 if (rc > 0 || rc == -ENOSYS) {
4756 /* is it an HSM var ? */
4757 rc = class_process_proc_param(PARAM_HSM,
4758 hsm_cdt_get_proc_vars(),
4760 if (rc > 0 || rc == -ENOSYS)
4761 /* we don't understand; pass it on */
4762 rc = next->ld_ops->ldo_process_config(env, next,
4766 if (old_cfg != NULL)
4767 lustre_cfg_free(cfg);
4772 /* others are passed further */
4773 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4779 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4780 const struct lu_object_header *hdr,
4781 struct lu_device *d)
4783 struct mdt_object *mo;
4787 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4789 struct lu_object *o;
4790 struct lu_object_header *h;
4793 h = &mo->mot_header;
4794 lu_object_header_init(h);
4795 lu_object_init(o, h, d);
4796 lu_object_add_top(h, o);
4797 o->lo_ops = &mdt_obj_ops;
4798 mutex_init(&mo->mot_ioepoch_mutex);
4799 mutex_init(&mo->mot_lov_mutex);
4800 init_rwsem(&mo->mot_open_sem);
4806 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4807 const struct lu_object_conf *unused)
4809 struct mdt_device *d = mdt_dev(o->lo_dev);
4810 struct lu_device *under;
4811 struct lu_object *below;
4815 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4816 PFID(lu_object_fid(o)));
4818 under = &d->mdt_child->md_lu_dev;
4819 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4820 if (below != NULL) {
4821 lu_object_add(o, below);
4828 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4830 struct mdt_object *mo = mdt_obj(o);
4831 struct lu_object_header *h;
4835 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4836 PFID(lu_object_fid(o)));
4838 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4839 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4842 lu_object_header_fini(h);
4843 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4848 static int mdt_object_print(const struct lu_env *env, void *cookie,
4849 lu_printer_t p, const struct lu_object *o)
4851 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4852 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4853 "flags="LPX64", epochcount=%d, writecount=%d)",
4854 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4855 mdto->mot_ioepoch_count, mdto->mot_writecount);
4858 static int mdt_prepare(const struct lu_env *env,
4859 struct lu_device *pdev,
4860 struct lu_device *cdev)
4862 struct mdt_device *mdt = mdt_dev(cdev);
4863 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4864 struct obd_device *obd = cdev->ld_obd;
4871 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4875 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4879 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4883 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4884 /* The LFSCK instance is registered just now, so it must be there when
4885 * register the namespace to such instance. */
4886 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4888 if (mdt->mdt_seq_site.ss_node_id == 0) {
4889 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4890 &mdt->mdt_md_root_fid);
4895 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4896 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4897 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4898 LASSERT(obd->obd_no_conn);
4899 spin_lock(&obd->obd_dev_lock);
4900 obd->obd_no_conn = 0;
4901 spin_unlock(&obd->obd_dev_lock);
4903 if (obd->obd_recovering == 0)
4904 mdt_postrecov(env, mdt);
4909 const struct lu_device_operations mdt_lu_ops = {
4910 .ldo_object_alloc = mdt_object_alloc,
4911 .ldo_process_config = mdt_process_config,
4912 .ldo_prepare = mdt_prepare,
4915 static const struct lu_object_operations mdt_obj_ops = {
4916 .loo_object_init = mdt_object_init,
4917 .loo_object_free = mdt_object_free,
4918 .loo_object_print = mdt_object_print
4921 static int mdt_obd_set_info_async(const struct lu_env *env,
4922 struct obd_export *exp,
4923 __u32 keylen, void *key,
4924 __u32 vallen, void *val,
4925 struct ptlrpc_request_set *set)
4931 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4932 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4940 * Match client and server connection feature flags.
4942 * Compute the compatibility flags for a connection request based on
4943 * features mutually supported by client and server.
4945 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4946 * must not be updated here, otherwise a partially initialized value may
4947 * be exposed. After the connection request is successfully processed,
4948 * the top-level MDT connect request handler atomically updates the export
4949 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4950 * reply. \see mdt_connect().
4952 * \param exp the obd_export associated with this client/target pair
4953 * \param mdt the target device for the connection
4954 * \param data stores data for this connect request
4957 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4958 * \retval -EBADE client and server feature requirements are incompatible
4960 static int mdt_connect_internal(struct obd_export *exp,
4961 struct mdt_device *mdt,
4962 struct obd_connect_data *data)
4964 LASSERT(data != NULL);
4966 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4967 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4969 if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4970 !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4971 CWARN("%s: client %s does not support ibits lock, either "
4972 "very old or an invalid client: flags "LPX64"\n",
4973 mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4974 data->ocd_connect_flags);
4978 if (!mdt->mdt_opts.mo_acl)
4979 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4981 if (!mdt->mdt_opts.mo_user_xattr)
4982 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4984 if (!mdt->mdt_som_conf)
4985 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4987 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4988 data->ocd_brw_size = min(data->ocd_brw_size,
4989 (__u32)MD_MAX_BRW_SIZE);
4990 if (data->ocd_brw_size == 0) {
4991 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4992 " ocd_version: %x ocd_grant: %d "
4993 "ocd_index: %u ocd_brw_size is "
4994 "unexpectedly zero, network data "
4995 "corruption? Refusing connection of this"
4998 exp->exp_client_uuid.uuid,
4999 exp, data->ocd_connect_flags, data->ocd_version,
5000 data->ocd_grant, data->ocd_index);
5005 /* NB: Disregard the rule against updating
5006 * exp_connect_data.ocd_connect_flags in this case, since
5007 * tgt_client_new() needs to know if this is a lightweight
5008 * connection, and it is safe to expose this flag before
5009 * connection processing completes. */
5010 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5011 spin_lock(&exp->exp_lock);
5012 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5013 spin_unlock(&exp->exp_lock);
5016 data->ocd_version = LUSTRE_VERSION_CODE;
5018 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5019 CWARN("%s: MDS requires FID support, but client not\n",
5024 if (mdt->mdt_som_conf &&
5025 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5026 OBD_CONNECT_MDS_MDS |
5027 OBD_CONNECT_SOM))) {
5028 CWARN("%s: MDS has SOM enabled, but client does not support "
5029 "it\n", mdt_obd_name(mdt));
5033 if (OCD_HAS_FLAG(data, PINGLESS)) {
5034 if (ptlrpc_pinger_suppress_pings()) {
5035 spin_lock(&exp->exp_obd->obd_dev_lock);
5036 list_del_init(&exp->exp_obd_chain_timed);
5037 spin_unlock(&exp->exp_obd->obd_dev_lock);
5039 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5043 data->ocd_max_easize = mdt->mdt_max_ea_size;
5048 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5049 struct mdt_thread_info *info,
5050 struct mdt_file_data *mfd)
5052 struct lu_context ses;
5056 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5061 lu_context_enter(&ses);
5063 mdt_ucred(info)->uc_valid = UCRED_OLD;
5064 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5066 lu_context_exit(&ses);
5067 lu_context_fini(&ses);
5073 static int mdt_export_cleanup(struct obd_export *exp)
5075 struct list_head closing_list;
5076 struct mdt_export_data *med = &exp->exp_mdt_data;
5077 struct obd_device *obd = exp->exp_obd;
5078 struct mdt_device *mdt;
5079 struct mdt_thread_info *info;
5081 struct mdt_file_data *mfd, *n;
5085 INIT_LIST_HEAD(&closing_list);
5086 spin_lock(&med->med_open_lock);
5087 while (!list_empty(&med->med_open_head)) {
5088 struct list_head *tmp = med->med_open_head.next;
5089 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5091 /* Remove mfd handle so it can't be found again.
5092 * We are consuming the mfd_list reference here. */
5093 class_handle_unhash(&mfd->mfd_handle);
5094 list_move_tail(&mfd->mfd_list, &closing_list);
5096 spin_unlock(&med->med_open_lock);
5097 mdt = mdt_dev(obd->obd_lu_dev);
5098 LASSERT(mdt != NULL);
5100 rc = lu_env_init(&env, LCT_MD_THREAD);
5104 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5105 LASSERT(info != NULL);
5106 memset(info, 0, sizeof *info);
5107 info->mti_env = &env;
5108 info->mti_mdt = mdt;
5109 info->mti_exp = exp;
5111 if (!list_empty(&closing_list)) {
5112 struct md_attr *ma = &info->mti_attr;
5114 /* Close any open files (which may also cause orphan
5116 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5117 list_del_init(&mfd->mfd_list);
5118 ma->ma_need = ma->ma_valid = 0;
5120 /* This file is being closed due to an eviction, it
5121 * could have been modified and now dirty regarding to
5122 * HSM archive, check this!
5123 * The logic here is to mark a file dirty if there's a
5124 * chance it was dirtied before the client was evicted,
5125 * so that we don't have to wait for a release attempt
5126 * before finding out the file was actually dirty and
5127 * fail the release. Aggressively marking it dirty here
5128 * will cause the policy engine to attempt to
5129 * re-archive it; when rearchiving, we can compare the
5130 * current version to the HSM data_version and make the
5131 * archive request into a noop if it's not actually
5134 if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5135 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5137 /* Don't unlink orphan on failover umount, LU-184 */
5138 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5139 ma->ma_valid = MA_FLAGS;
5140 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5142 mdt_mfd_close(info, mfd);
5145 info->mti_mdt = NULL;
5146 /* cleanup client slot early */
5147 /* Do not erase record for recoverable client. */
5148 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5149 tgt_client_del(&env, exp);
5155 static int mdt_obd_disconnect(struct obd_export *exp)
5161 class_export_get(exp);
5163 nodemap_del_member(exp);
5164 rc = server_disconnect_export(exp);
5166 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5168 rc = mdt_export_cleanup(exp);
5169 class_export_put(exp);
5173 /* mds_connect copy */
5174 static int mdt_obd_connect(const struct lu_env *env,
5175 struct obd_export **exp, struct obd_device *obd,
5176 struct obd_uuid *cluuid,
5177 struct obd_connect_data *data,
5180 struct obd_export *lexp;
5181 struct lustre_handle conn = { 0 };
5182 struct mdt_device *mdt;
5184 lnet_nid_t *client_nid = localdata;
5187 LASSERT(env != NULL);
5188 if (!exp || !obd || !cluuid)
5191 mdt = mdt_dev(obd->obd_lu_dev);
5194 * first, check whether the stack is ready to handle requests
5195 * XXX: probably not very appropriate method is used now
5196 * at some point we should find a better one
5198 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5199 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5200 rc = obd_get_info(env, mdt->mdt_child_exp,
5201 sizeof(KEY_OSP_CONNECTED),
5202 KEY_OSP_CONNECTED, NULL, NULL, NULL);
5205 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5208 rc = class_connect(&conn, obd, cluuid);
5212 lexp = class_conn2export(&conn);
5213 LASSERT(lexp != NULL);
5215 rc = mdt_connect_internal(lexp, mdt, data);
5217 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5220 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5221 rc = tgt_client_new(env, lexp);
5223 rc = nodemap_add_member(*client_nid, lexp);
5224 if (rc != 0 && rc != -EEXIST)
5227 mdt_export_stats_init(obd, lexp, localdata);
5230 /* For phase I, sync for cross-ref operation. */
5231 spin_lock(&lexp->exp_lock);
5232 lexp->exp_keep_sync = 1;
5233 spin_unlock(&lexp->exp_lock);
5237 class_disconnect(lexp);
5246 static int mdt_obd_reconnect(const struct lu_env *env,
5247 struct obd_export *exp, struct obd_device *obd,
5248 struct obd_uuid *cluuid,
5249 struct obd_connect_data *data,
5252 lnet_nid_t *client_nid = localdata;
5256 if (exp == NULL || obd == NULL || cluuid == NULL)
5259 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5261 rc = nodemap_add_member(*client_nid, exp);
5262 if (rc == 0 || rc == -EEXIST)
5263 mdt_export_stats_init(obd, exp, localdata);
5269 /* FIXME: Can we avoid using these two interfaces? */
5270 static int mdt_init_export(struct obd_export *exp)
5272 struct mdt_export_data *med = &exp->exp_mdt_data;
5276 INIT_LIST_HEAD(&med->med_open_head);
5277 spin_lock_init(&med->med_open_lock);
5278 mutex_init(&med->med_idmap_mutex);
5279 med->med_idmap = NULL;
5280 spin_lock(&exp->exp_lock);
5281 exp->exp_connecting = 1;
5282 spin_unlock(&exp->exp_lock);
5284 /* self-export doesn't need client data and ldlm initialization */
5285 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5286 &exp->exp_client_uuid)))
5289 rc = tgt_client_alloc(exp);
5293 rc = ldlm_init_export(exp);
5300 tgt_client_free(exp);
5302 CERROR("%s: Failed to initialize export: rc = %d\n",
5303 exp->exp_obd->obd_name, rc);
5307 static int mdt_destroy_export(struct obd_export *exp)
5311 if (exp_connect_rmtclient(exp))
5312 mdt_cleanup_idmap(&exp->exp_mdt_data);
5314 target_destroy_export(exp);
5315 /* destroy can be called from failed obd_setup, so
5316 * checking uuid is safer than obd_self_export */
5317 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5318 &exp->exp_client_uuid)))
5321 ldlm_destroy_export(exp);
5322 tgt_client_free(exp);
5324 LASSERT(list_empty(&exp->exp_outstanding_replies));
5325 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5330 /** The maximum depth that fid2path() will search.
5331 * This is limited only because we want to store the fids for
5332 * historical path lookup purposes.
5334 #define MAX_PATH_DEPTH 100
5336 /** mdt_path() lookup structure. */
5337 struct path_lookup_info {
5338 __u64 pli_recno; /**< history point */
5339 __u64 pli_currec; /**< current record */
5340 struct lu_fid pli_fid;
5341 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5342 struct mdt_object *pli_mdt_obj;
5343 char *pli_path; /**< full path */
5345 int pli_linkno; /**< which hardlink to follow */
5346 int pli_fidcount; /**< number of \a pli_fids */
5349 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5350 struct linkea_data *ldata)
5354 LASSERT(ldata->ld_buf->lb_buf != NULL);
5356 if (!mdt_object_exists(mdt_obj))
5359 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5360 ldata->ld_buf, XATTR_NAME_LINK);
5361 if (rc == -ERANGE) {
5362 /* Buf was too small, figure out what we need. */
5363 lu_buf_free(ldata->ld_buf);
5364 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5365 ldata->ld_buf, XATTR_NAME_LINK);
5368 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5369 if (ldata->ld_buf->lb_buf == NULL)
5371 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5372 ldata->ld_buf, XATTR_NAME_LINK);
5377 return linkea_init(ldata);
5380 static int mdt_path_current(struct mdt_thread_info *info,
5381 struct path_lookup_info *pli)
5383 struct mdt_device *mdt = info->mti_mdt;
5384 struct mdt_object *mdt_obj;
5385 struct link_ea_header *leh;
5386 struct link_ea_entry *lee;
5387 struct lu_name *tmpname = &info->mti_name;
5388 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5389 struct lu_buf *buf = &info->mti_big_buf;
5392 struct linkea_data ldata = { 0 };
5396 /* temp buffer for path element, the buffer will be finally freed
5397 * in mdt_thread_info_fini */
5398 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5399 if (buf->lb_buf == NULL)
5403 ptr = pli->pli_path + pli->pli_pathlen - 1;
5406 pli->pli_fidcount = 0;
5407 pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5408 *tmpfid = pli->pli_fids[0];
5409 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5410 * so checking root_fid can only happen on MDT0. */
5411 while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5412 &pli->pli_fids[pli->pli_fidcount])) {
5413 struct lu_buf lmv_buf;
5415 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5416 if (IS_ERR(mdt_obj))
5417 GOTO(out, rc = PTR_ERR(mdt_obj));
5419 if (!mdt_object_exists(mdt_obj)) {
5420 mdt_object_put(info->mti_env, mdt_obj);
5421 GOTO(out, rc = -ENOENT);
5424 if (mdt_object_remote(mdt_obj)) {
5425 mdt_object_put(info->mti_env, mdt_obj);
5426 GOTO(remote_out, rc = -EREMOTE);
5429 rc = mdt_links_read(info, mdt_obj, &ldata);
5431 mdt_object_put(info->mti_env, mdt_obj);
5436 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5437 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5438 /* If set, use link #linkno for path lookup, otherwise use
5439 link #0. Only do this for the final path element. */
5440 if (pli->pli_fidcount == 0 &&
5441 pli->pli_linkno < leh->leh_reccount) {
5443 for (count = 0; count < pli->pli_linkno; count++) {
5444 lee = (struct link_ea_entry *)
5445 ((char *)lee + reclen);
5446 linkea_entry_unpack(lee, &reclen, tmpname,
5449 if (pli->pli_linkno < leh->leh_reccount - 1)
5450 /* indicate to user there are more links */
5454 lmv_buf.lb_buf = info->mti_xattr_buf;
5455 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5456 /* Check if it is slave stripes */
5457 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5458 &lmv_buf, XATTR_NAME_LMV);
5459 mdt_object_put(info->mti_env, mdt_obj);
5461 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5463 /* For slave stripes, get its master */
5464 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5465 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5468 } else if (rc < 0 && rc != -ENODATA) {
5474 /* Pack the name in the end of the buffer */
5475 ptr -= tmpname->ln_namelen;
5476 if (ptr - 1 <= pli->pli_path)
5477 GOTO(out, rc = -EOVERFLOW);
5478 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5481 /* Store the parent fid for historic lookup */
5482 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5483 GOTO(out, rc = -EOVERFLOW);
5484 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5488 ptr++; /* skip leading / */
5489 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5496 /* Returns the full path to this fid, as of changelog record recno. */
5497 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5498 char *path, int pathlen, __u64 *recno, int *linkno,
5501 struct mdt_device *mdt = info->mti_mdt;
5502 struct path_lookup_info *pli;
5510 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5519 pli->pli_mdt_obj = obj;
5520 pli->pli_recno = *recno;
5521 pli->pli_path = path;
5522 pli->pli_pathlen = pathlen;
5523 pli->pli_linkno = *linkno;
5525 /* Retry multiple times in case file is being moved */
5526 while (tries-- && rc == -EAGAIN)
5527 rc = mdt_path_current(info, pli);
5529 /* return the last resolved fids to the client, so the client will
5530 * build the left path on another MDT for remote object */
5531 *fid = pli->pli_fids[pli->pli_fidcount];
5533 *recno = pli->pli_currec;
5534 /* Return next link index to caller */
5535 *linkno = pli->pli_linkno;
5542 static int mdt_fid2path(struct mdt_thread_info *info,
5543 struct getinfo_fid2path *fp)
5545 struct mdt_device *mdt = info->mti_mdt;
5546 struct mdt_object *obj;
5550 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5551 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5553 if (!fid_is_sane(&fp->gf_fid))
5556 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5557 CWARN("%s: "DFID" is invalid, sequence should be "
5558 ">= "LPX64"\n", mdt_obd_name(mdt),
5559 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5563 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5564 if (obj == NULL || IS_ERR(obj)) {
5565 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5570 if (mdt_object_remote(obj))
5572 else if (!mdt_object_exists(obj))
5578 mdt_object_put(info->mti_env, obj);
5579 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5580 PFID(&fp->gf_fid), rc);
5584 rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5585 &fp->gf_linkno, &fp->gf_fid);
5587 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5588 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5590 mdt_object_put(info->mti_env, obj);
5595 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5596 void *val, int vallen)
5598 struct getinfo_fid2path *fpout, *fpin;
5601 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5604 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5605 lustre_swab_fid2path(fpin);
5607 memcpy(fpout, fpin, sizeof(*fpin));
5608 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5611 rc = mdt_fid2path(info, fpout);
5615 int mdt_get_info(struct tgt_session_info *tsi)
5625 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5627 CDEBUG(D_IOCTL, "No GETINFO key\n");
5628 RETURN(err_serious(-EFAULT));
5630 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5633 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5634 if (vallen == NULL) {
5635 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5636 tgt_name(tsi->tsi_tgt));
5637 RETURN(err_serious(-EFAULT));
5640 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5642 rc = req_capsule_server_pack(tsi->tsi_pill);
5644 RETURN(err_serious(rc));
5646 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5647 if (valout == NULL) {
5648 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5649 tgt_name(tsi->tsi_tgt));
5650 RETURN(err_serious(-EFAULT));
5653 if (KEY_IS(KEY_FID2PATH)) {
5654 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5656 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5657 mdt_thread_info_fini(info);
5664 /* Pass the ioc down */
5665 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5666 unsigned int cmd, int len, void *data)
5668 struct lu_context ioctl_session;
5669 struct md_device *next = mdt->mdt_child;
5673 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5676 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5677 lu_context_enter(&ioctl_session);
5678 env->le_ses = &ioctl_session;
5680 LASSERT(next->md_ops->mdo_iocontrol);
5681 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5683 lu_context_exit(&ioctl_session);
5684 lu_context_fini(&ioctl_session);
5688 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5690 struct obd_ioctl_data *data = karg;
5693 struct mdt_object *obj;
5694 struct mdt_lock_handle *lh;
5698 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5699 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5702 fid = (struct lu_fid *)data->ioc_inlbuf1;
5704 if (!fid_is_sane(fid))
5707 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5709 lh = &mti->mti_lh[MDT_LH_PARENT];
5710 mdt_lock_reg_init(lh, LCK_CR);
5712 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5714 RETURN(PTR_ERR(obj));
5716 if (mdt_object_remote(obj)) {
5719 * before calling version get the correct MDS should be
5720 * fid, this is error to find remote object here
5722 CERROR("nonlocal object "DFID"\n", PFID(fid));
5723 } else if (!mdt_object_exists(obj)) {
5724 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5727 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5728 *(__u64 *)data->ioc_inlbuf2 = version;
5731 mdt_object_unlock_put(mti, obj, lh, 1);
5735 /* ioctls on obd dev */
5736 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5737 void *karg, void *uarg)
5740 struct obd_device *obd = exp->exp_obd;
5741 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5742 struct dt_device *dt = mdt->mdt_bottom;
5746 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5747 rc = lu_env_init(&env, LCT_MD_THREAD);
5753 rc = mdt_device_sync(&env, mdt);
5755 case OBD_IOC_SET_READONLY:
5756 rc = dt->dd_ops->dt_ro(&env, dt);
5758 case OBD_IOC_ABORT_RECOVERY:
5759 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5760 target_stop_recovery_thread(obd);
5763 case OBD_IOC_CHANGELOG_REG:
5764 case OBD_IOC_CHANGELOG_DEREG:
5765 case OBD_IOC_CHANGELOG_CLEAR:
5766 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5768 case OBD_IOC_START_LFSCK: {
5769 struct md_device *next = mdt->mdt_child;
5770 struct obd_ioctl_data *data = karg;
5771 struct lfsck_start_param lsp;
5773 if (unlikely(data == NULL)) {
5778 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5779 lsp.lsp_index_valid = 0;
5780 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5783 case OBD_IOC_STOP_LFSCK: {
5784 struct md_device *next = mdt->mdt_child;
5785 struct obd_ioctl_data *data = karg;
5786 struct lfsck_stop stop;
5788 stop.ls_status = LS_STOPPED;
5789 /* Old lfsck utils may pass NULL @stop. */
5790 if (data->ioc_inlbuf1 == NULL)
5794 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5796 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5799 case OBD_IOC_GET_OBJ_VERSION: {
5800 struct mdt_thread_info *mti;
5801 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5802 memset(mti, 0, sizeof *mti);
5803 mti->mti_env = &env;
5807 rc = mdt_ioc_version_get(mti, karg);
5810 case OBD_IOC_CATLOGLIST: {
5811 struct mdt_thread_info *mti;
5813 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5814 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5815 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5816 &mti->mti_tmp_fid1);
5821 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5822 mdt_obd_name(mdt), cmd, rc);
5829 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5831 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5832 struct lfsck_start_param lsp;
5836 lsp.lsp_start = NULL;
5837 lsp.lsp_index_valid = 0;
5838 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5839 OBD_IOC_START_LFSCK,
5841 if (rc != 0 && rc != -EALREADY)
5842 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5843 mdt_obd_name(mdt), rc);
5845 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5849 static int mdt_obd_postrecov(struct obd_device *obd)
5854 rc = lu_env_init(&env, LCT_MD_THREAD);
5857 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5862 static struct obd_ops mdt_obd_device_ops = {
5863 .o_owner = THIS_MODULE,
5864 .o_set_info_async = mdt_obd_set_info_async,
5865 .o_connect = mdt_obd_connect,
5866 .o_reconnect = mdt_obd_reconnect,
5867 .o_disconnect = mdt_obd_disconnect,
5868 .o_init_export = mdt_init_export,
5869 .o_destroy_export = mdt_destroy_export,
5870 .o_iocontrol = mdt_iocontrol,
5871 .o_postrecov = mdt_obd_postrecov,
5874 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5875 struct lu_device *d)
5877 struct mdt_device *m = mdt_dev(d);
5884 static struct lu_device *mdt_device_free(const struct lu_env *env,
5885 struct lu_device *d)
5887 struct mdt_device *m = mdt_dev(d);
5890 lu_device_fini(&m->mdt_lu_dev);
5896 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5897 struct lu_device_type *t,
5898 struct lustre_cfg *cfg)
5900 struct lu_device *l;
5901 struct mdt_device *m;
5908 rc = mdt_init0(env, m, t, cfg);
5910 mdt_device_free(env, l);
5915 l = ERR_PTR(-ENOMEM);
5919 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5920 LU_KEY_INIT(mdt, struct mdt_thread_info);
5922 static void mdt_key_fini(const struct lu_context *ctx,
5923 struct lu_context_key *key, void* data)
5925 struct mdt_thread_info *info = data;
5927 if (info->mti_big_lmm) {
5928 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5929 info->mti_big_lmm = NULL;
5930 info->mti_big_lmmsize = 0;
5935 /* context key: mdt_thread_key */
5936 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5938 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5940 return lu_ucred(info->mti_env);
5943 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5945 return lu_ucred_check(info->mti_env);
5949 * Enable/disable COS (Commit On Sharing).
5951 * Set/Clear the COS flag in mdt options.
5953 * \param mdt mdt device
5954 * \param val 0 disables COS, other values enable COS
5956 void mdt_enable_cos(struct mdt_device *mdt, int val)
5961 mdt->mdt_opts.mo_cos = !!val;
5962 rc = lu_env_init(&env, LCT_LOCAL);
5963 if (unlikely(rc != 0)) {
5964 CWARN("%s: lu_env initialization failed, cannot "
5965 "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5968 mdt_device_sync(&env, mdt);
5973 * Check COS (Commit On Sharing) status.
5975 * Return COS flag status.
5977 * \param mdt mdt device
5979 int mdt_cos_is_enabled(struct mdt_device *mdt)
5981 return mdt->mdt_opts.mo_cos != 0;
5984 static struct lu_device_type_operations mdt_device_type_ops = {
5985 .ldto_device_alloc = mdt_device_alloc,
5986 .ldto_device_free = mdt_device_free,
5987 .ldto_device_fini = mdt_device_fini
5990 static struct lu_device_type mdt_device_type = {
5991 .ldt_tags = LU_DEVICE_MD,
5992 .ldt_name = LUSTRE_MDT_NAME,
5993 .ldt_ops = &mdt_device_type_ops,
5994 .ldt_ctx_tags = LCT_MD_THREAD
5997 static int __init mdt_mod_init(void)
6001 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6002 FID_NOBRACE_LEN + 1);
6003 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6005 rc = lu_kmem_init(mdt_caches);
6009 rc = mds_mod_init();
6013 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6014 LUSTRE_MDT_NAME, &mdt_device_type);
6019 lu_kmem_fini(mdt_caches);
6026 static void __exit mdt_mod_exit(void)
6028 class_unregister_type(LUSTRE_MDT_NAME);
6030 lu_kmem_fini(mdt_caches);
6033 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6034 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6035 MODULE_LICENSE("GPL");
6037 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);