1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
50 # define EXPORT_SYMTAB
52 #define DEBUG_SUBSYSTEM S_MDS
54 #include <linux/module.h>
56 * struct OBD_{ALLOC,FREE}*()
58 #include <obd_support.h>
59 /* struct ptlrpc_request */
60 #include <lustre_net.h>
61 /* struct obd_export */
62 #include <lustre_export.h>
63 /* struct obd_device */
66 #include <dt_object.h>
67 #include <lustre_mds.h>
68 #include <lustre_mdt.h>
69 #include "mdt_internal.h"
70 #ifdef HAVE_QUOTA_SUPPORT
71 # include <lustre_quota.h>
73 #include <lustre_acl.h>
74 #include <lustre_param.h>
75 #include <lustre_fsfilt.h>
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78 [LCK_MINMODE] = MDL_MINMODE,
85 [LCK_GROUP] = MDL_GROUP
88 ldlm_mode_t mdt_dlm_lock_modes[] = {
89 [MDL_MINMODE] = LCK_MINMODE,
96 [MDL_GROUP] = LCK_GROUP
100 * Initialized in mdt_mod_init().
102 static unsigned long mdt_num_threads;
103 static unsigned long mdt_min_threads;
104 static unsigned long mdt_max_threads;
106 /* ptlrpc request handler for MDT. All handlers are
107 * grouped into several slices - struct mdt_opc_slice,
108 * and stored in an array - mdt_handlers[].
111 /* The name of this handler. */
113 /* Fail id for this handler, checked at the beginning of this handler*/
115 /* Operation code for this handler */
117 /* flags are listed in enum mdt_handler_flags below. */
119 /* The actual handler function to execute. */
120 int (*mh_act)(struct mdt_thread_info *info);
121 /* Request format for this request. */
122 const struct req_format *mh_fmt;
125 enum mdt_handler_flags {
127 * struct mdt_body is passed in the incoming message, and object
128 * identified by this fid exists on disk.
130 * "habeo corpus" == "I have a body"
132 HABEO_CORPUS = (1 << 0),
134 * struct ldlm_request is passed in the incoming message.
136 * "habeo clavis" == "I have a key"
138 HABEO_CLAVIS = (1 << 1),
140 * this request has fixed reply format, so that reply message can be
141 * packed by generic code.
143 * "habeo refero" == "I have a reply"
145 HABEO_REFERO = (1 << 2),
147 * this request will modify something, so check whether the filesystem
148 * is readonly or not, then return -EROFS to client asap if necessary.
150 * "mutabor" == "I shall modify"
155 struct mdt_opc_slice {
158 struct mdt_handler *mos_hs;
161 static struct mdt_opc_slice mdt_regular_handlers[];
162 static struct mdt_opc_slice mdt_readpage_handlers[];
163 static struct mdt_opc_slice mdt_xmds_handlers[];
164 static struct mdt_opc_slice mdt_seq_handlers[];
165 static struct mdt_opc_slice mdt_fld_handlers[];
167 static struct mdt_device *mdt_dev(struct lu_device *d);
168 static int mdt_regular_handle(struct ptlrpc_request *req);
169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
171 struct getinfo_fid2path *fp);
173 static const struct lu_object_operations mdt_obj_ops;
175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
179 return (rep->lock_policy_res1 & flag);
182 void mdt_clear_disposition(struct mdt_thread_info *info,
183 struct ldlm_reply *rep, int flag)
186 info->mti_opdata &= ~flag;
188 rep->lock_policy_res1 &= ~flag;
191 void mdt_set_disposition(struct mdt_thread_info *info,
192 struct ldlm_reply *rep, int flag)
195 info->mti_opdata |= flag;
197 rep->lock_policy_res1 |= flag;
200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
202 lh->mlh_pdo_hash = 0;
203 lh->mlh_reg_mode = lm;
204 lh->mlh_type = MDT_REG_LOCK;
207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
208 const char *name, int namelen)
210 lh->mlh_reg_mode = lm;
211 lh->mlh_type = MDT_PDO_LOCK;
214 LASSERT(namelen > 0);
215 lh->mlh_pdo_hash = full_name_hash(name, namelen);
217 LASSERT(namelen == 0);
218 lh->mlh_pdo_hash = 0ull;
222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
223 struct mdt_lock_handle *lh)
229 * Any dir access needs couple of locks:
231 * 1) on part of dir we gonna take lookup/modify;
233 * 2) on whole dir to protect it from concurrent splitting and/or to
234 * flush client's cache for readdir().
236 * so, for a given mode and object this routine decides what lock mode
237 * to use for lock #2:
239 * 1) if caller's gonna lookup in dir then we need to protect dir from
240 * being splitted only - LCK_CR
242 * 2) if caller's gonna modify dir then we need to protect dir from
243 * being splitted and to flush cache - LCK_CW
245 * 3) if caller's gonna modify dir and that dir seems ready for
246 * splitting then we need to protect it from any type of access
247 * (lookup/modify/split) - LCK_EX --bzzz
250 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
251 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
254 * Ask underlaying level its opinion about preferable PDO lock mode
255 * having access type passed as regular lock mode:
257 * - MDL_MINMODE means that lower layer does not want to specify lock
260 * - MDL_NL means that no PDO lock should be taken. This is used in some
261 * cases. Say, for non-splittable directories no need to use PDO locks
264 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
265 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
267 if (mode != MDL_MINMODE) {
268 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
271 * Lower layer does not want to specify locking mode. We do it
272 * our selves. No special protection is needed, just flush
273 * client's cache on modification and allow concurrent
276 switch (lh->mlh_reg_mode) {
278 lh->mlh_pdo_mode = LCK_EX;
281 lh->mlh_pdo_mode = LCK_CR;
284 lh->mlh_pdo_mode = LCK_CW;
287 CERROR("Not expected lock type (0x%x)\n",
288 (int)lh->mlh_reg_mode);
293 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
297 static int mdt_getstatus(struct mdt_thread_info *info)
299 struct mdt_device *mdt = info->mti_mdt;
300 struct md_device *next = mdt->mdt_child;
301 struct mdt_body *repbody;
306 rc = mdt_check_ucred(info);
308 RETURN(err_serious(rc));
310 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
311 RETURN(err_serious(-ENOMEM));
313 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
314 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
318 repbody->valid |= OBD_MD_FLID;
320 if (mdt->mdt_opts.mo_mds_capa &&
321 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
322 struct mdt_object *root;
323 struct lustre_capa *capa;
325 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
327 RETURN(PTR_ERR(root));
329 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
331 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
332 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
334 mdt_object_put(info->mti_env, root);
336 repbody->valid |= OBD_MD_FLMDSCAPA;
342 static int mdt_statfs(struct mdt_thread_info *info)
344 struct md_device *next = info->mti_mdt->mdt_child;
345 struct ptlrpc_service *svc;
346 struct obd_statfs *osfs;
351 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
353 /* This will trigger a watchdog timeout */
354 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
355 (MDT_SERVICE_WATCHDOG_FACTOR *
356 at_get(&svc->srv_at_estimate)) + 1);
358 rc = mdt_check_ucred(info);
360 RETURN(err_serious(rc));
362 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
363 rc = err_serious(-ENOMEM);
365 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
366 rc = next->md_ops->mdo_statfs(info->mti_env, next,
368 statfs_pack(osfs, &info->mti_u.ksfs);
374 * Pack SOM attributes into the reply.
375 * Call under a DLM UPDATE lock.
377 static void mdt_pack_size2body(struct mdt_thread_info *info,
378 struct mdt_object *mo)
381 struct md_attr *ma = &info->mti_attr;
383 LASSERT(ma->ma_attr.la_valid & LA_MODE);
384 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
386 /* Check if Size-on-MDS is supported, if this is a regular file,
387 * if SOM is enabled on the object and if SOM cache exists and valid.
388 * Otherwise do not pack Size-on-MDS attributes to the reply. */
389 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
390 !S_ISREG(ma->ma_attr.la_mode) ||
391 !mdt_object_is_som_enabled(mo) ||
392 !(ma->ma_valid & MA_SOM))
395 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
396 b->size = ma->ma_som->msd_size;
397 b->blocks = ma->ma_som->msd_blocks;
400 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
401 const struct lu_attr *attr, const struct lu_fid *fid)
403 struct md_attr *ma = &info->mti_attr;
405 /*XXX should pack the reply body according to lu_valid*/
406 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
407 OBD_MD_FLGID | OBD_MD_FLTYPE |
408 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
409 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
411 if (!S_ISREG(attr->la_mode))
412 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
414 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
415 if (ma->ma_lmm_size == 0)
416 b->valid |= OBD_MD_FLSIZE;
418 b->atime = attr->la_atime;
419 b->mtime = attr->la_mtime;
420 b->ctime = attr->la_ctime;
421 b->mode = attr->la_mode;
422 b->size = attr->la_size;
423 b->blocks = attr->la_blocks;
424 b->uid = attr->la_uid;
425 b->gid = attr->la_gid;
426 b->flags = attr->la_flags;
427 b->nlink = attr->la_nlink;
428 b->rdev = attr->la_rdev;
432 b->valid |= OBD_MD_FLID;
434 /* FIXME: these should be fixed when new igif ready.*/
435 b->ino = fid_oid(fid); /* 1.6 compatibility */
436 b->generation = fid_ver(fid); /* 1.6 compatibility */
437 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
439 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
440 PFID(fid), b->nlink, b->mode, b->size);
444 mdt_body_reverse_idmap(info, b);
447 static inline int mdt_body_has_lov(const struct lu_attr *la,
448 const struct mdt_body *body)
450 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
451 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
454 static int mdt_getattr_internal(struct mdt_thread_info *info,
455 struct mdt_object *o, int ma_need)
457 struct md_object *next = mdt_object_child(o);
458 const struct mdt_body *reqbody = info->mti_body;
459 struct ptlrpc_request *req = mdt_info_req(info);
460 struct md_attr *ma = &info->mti_attr;
461 struct lu_attr *la = &ma->ma_attr;
462 struct req_capsule *pill = info->mti_pill;
463 const struct lu_env *env = info->mti_env;
464 struct mdt_body *repbody;
465 struct lu_buf *buffer = &info->mti_buf;
469 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
470 RETURN(err_serious(-ENOMEM));
472 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
476 rc = mdt_object_exists(o);
478 /* This object is located on remote node.*/
479 repbody->fid1 = *mdt_object_fid(o);
480 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
484 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
485 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
487 /* If it is dir object and client require MEA, then we got MEA */
488 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
489 reqbody->valid & OBD_MD_MEA) {
490 /* Assumption: MDT_MD size is enough for lmv size. */
491 ma->ma_lmv = buffer->lb_buf;
492 ma->ma_lmv_size = buffer->lb_len;
493 ma->ma_need = MA_LMV | MA_INODE;
495 ma->ma_lmm = buffer->lb_buf;
496 ma->ma_lmm_size = buffer->lb_len;
497 ma->ma_need = MA_LOV | MA_INODE;
500 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
501 reqbody->valid & OBD_MD_FLDIREA &&
502 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
503 /* get default stripe info for this dir. */
504 ma->ma_need |= MA_LOV_DEF;
506 ma->ma_need |= ma_need;
507 if (ma->ma_need & MA_SOM)
508 ma->ma_som = &info->mti_u.som.data;
510 rc = mo_attr_get(env, next, ma);
512 CERROR("getattr error for "DFID": %d\n",
513 PFID(mdt_object_fid(o)), rc);
517 if (likely(ma->ma_valid & MA_INODE))
518 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
522 if (mdt_body_has_lov(la, reqbody)) {
523 if (ma->ma_valid & MA_LOV) {
524 LASSERT(ma->ma_lmm_size);
525 mdt_dump_lmm(D_INFO, ma->ma_lmm);
526 repbody->eadatasize = ma->ma_lmm_size;
527 if (S_ISDIR(la->la_mode))
528 repbody->valid |= OBD_MD_FLDIREA;
530 repbody->valid |= OBD_MD_FLEASIZE;
532 if (ma->ma_valid & MA_LMV) {
533 LASSERT(S_ISDIR(la->la_mode));
534 repbody->eadatasize = ma->ma_lmv_size;
535 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
537 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
538 repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
540 } else if (S_ISLNK(la->la_mode) &&
541 reqbody->valid & OBD_MD_LINKNAME) {
542 buffer->lb_buf = ma->ma_lmm;
543 buffer->lb_len = reqbody->eadatasize;
544 rc = mo_readlink(env, next, buffer);
545 if (unlikely(rc <= 0)) {
546 CERROR("readlink failed: %d\n", rc);
549 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
551 repbody->valid |= OBD_MD_LINKNAME;
552 repbody->eadatasize = rc;
554 ((char*)ma->ma_lmm)[rc - 1] = 0;
555 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
556 (char*)ma->ma_lmm, rc);
561 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
562 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
563 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
564 repbody->valid |= OBD_MD_FLMODEASIZE;
565 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
566 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
567 repbody->max_cookiesize);
570 if (exp_connect_rmtclient(info->mti_exp) &&
571 reqbody->valid & OBD_MD_FLRMTPERM) {
572 void *buf = req_capsule_server_get(pill, &RMF_ACL);
574 /* mdt_getattr_lock only */
575 rc = mdt_pack_remote_perm(info, o, buf);
577 repbody->valid &= ~OBD_MD_FLRMTPERM;
578 repbody->aclsize = 0;
581 repbody->valid |= OBD_MD_FLRMTPERM;
582 repbody->aclsize = sizeof(struct mdt_remote_perm);
585 #ifdef CONFIG_FS_POSIX_ACL
586 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
587 (reqbody->valid & OBD_MD_FLACL)) {
588 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
589 buffer->lb_len = req_capsule_get_size(pill,
590 &RMF_ACL, RCL_SERVER);
591 if (buffer->lb_len > 0) {
592 rc = mo_xattr_get(env, next, buffer,
593 XATTR_NAME_ACL_ACCESS);
595 if (rc == -ENODATA) {
596 repbody->aclsize = 0;
597 repbody->valid |= OBD_MD_FLACL;
599 } else if (rc == -EOPNOTSUPP) {
602 CERROR("got acl size: %d\n", rc);
605 repbody->aclsize = rc;
606 repbody->valid |= OBD_MD_FLACL;
613 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
614 info->mti_mdt->mdt_opts.mo_mds_capa &&
615 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
616 struct lustre_capa *capa;
618 capa = req_capsule_server_get(pill, &RMF_CAPA1);
620 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
621 rc = mo_capa_get(env, next, capa, 0);
624 repbody->valid |= OBD_MD_FLMDSCAPA;
629 static int mdt_renew_capa(struct mdt_thread_info *info)
631 struct mdt_object *obj = info->mti_object;
632 struct mdt_body *body;
633 struct lustre_capa *capa, *c;
637 /* if object doesn't exist, or server has disabled capability,
638 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
641 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
642 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
645 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
646 LASSERT(body != NULL);
648 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
651 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
655 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
657 body->valid |= OBD_MD_FLOSSCAPA;
661 static int mdt_getattr(struct mdt_thread_info *info)
663 struct mdt_object *obj = info->mti_object;
664 struct req_capsule *pill = info->mti_pill;
665 struct mdt_body *reqbody;
666 struct mdt_body *repbody;
672 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
675 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
676 rc = req_capsule_server_pack(pill);
678 RETURN(err_serious(rc));
679 rc = mdt_renew_capa(info);
680 GOTO(out_shrink, rc);
683 LASSERT(obj != NULL);
684 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
686 mode = lu_object_attr(&obj->mot_obj.mo_lu);
687 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
688 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
689 md_size = reqbody->eadatasize;
691 md_size = info->mti_mdt->mdt_max_mdsize;
693 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
695 rc = req_capsule_server_pack(pill);
696 if (unlikely(rc != 0))
697 RETURN(err_serious(rc));
699 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
700 LASSERT(repbody != NULL);
701 repbody->eadatasize = 0;
702 repbody->aclsize = 0;
704 if (reqbody->valid & OBD_MD_FLRMTPERM)
705 rc = mdt_init_ucred(info, reqbody);
707 rc = mdt_check_ucred(info);
709 GOTO(out_shrink, rc);
711 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
712 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
715 * Don't check capability at all, because rename might getattr for
716 * remote obj, and at that time no capability is available.
718 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
719 rc = mdt_getattr_internal(info, obj, 0);
720 if (reqbody->valid & OBD_MD_FLRMTPERM)
721 mdt_exit_ucred(info);
724 mdt_shrink_reply(info);
728 static int mdt_is_subdir(struct mdt_thread_info *info)
730 struct mdt_object *o = info->mti_object;
731 struct req_capsule *pill = info->mti_pill;
732 const struct mdt_body *body = info->mti_body;
733 struct mdt_body *repbody;
739 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
742 * We save last checked parent fid to @repbody->fid1 for remote
745 LASSERT(fid_is_sane(&body->fid2));
746 LASSERT(mdt_object_exists(o) > 0);
747 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
748 &body->fid2, &repbody->fid1);
749 if (rc == 0 || rc == -EREMOTE)
750 repbody->valid |= OBD_MD_FLID;
755 static int mdt_raw_lookup(struct mdt_thread_info *info,
756 struct mdt_object *parent,
757 const struct lu_name *lname,
758 struct ldlm_reply *ldlm_rep)
760 struct md_object *next = mdt_object_child(info->mti_object);
761 const struct mdt_body *reqbody = info->mti_body;
762 struct lu_fid *child_fid = &info->mti_tmp_fid1;
763 struct mdt_body *repbody;
767 if (reqbody->valid != OBD_MD_FLID)
770 LASSERT(!info->mti_cross_ref);
772 /* Only got the fid of this obj by name */
773 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
776 /* XXX is raw_lookup possible as intent operation? */
779 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
782 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
784 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
787 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
788 repbody->fid1 = *child_fid;
789 repbody->valid = OBD_MD_FLID;
795 * UPDATE lock should be taken against parent, and be release before exit;
796 * child_bits lock should be taken against child, and be returned back:
797 * (1)normal request should release the child lock;
798 * (2)intent request will grant the lock to client.
800 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
801 struct mdt_lock_handle *lhc,
803 struct ldlm_reply *ldlm_rep)
805 struct ptlrpc_request *req = mdt_info_req(info);
806 struct mdt_body *reqbody = NULL;
807 struct mdt_object *parent = info->mti_object;
808 struct mdt_object *child;
809 struct md_object *next = mdt_object_child(parent);
810 struct lu_fid *child_fid = &info->mti_tmp_fid1;
811 struct lu_name *lname = NULL;
812 const char *name = NULL;
814 struct mdt_lock_handle *lhp;
815 struct ldlm_lock *lock;
816 struct ldlm_res_id *res_id;
823 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
824 LASSERT(ergo(is_resent,
825 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
827 LASSERT(parent != NULL);
828 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
830 RETURN(err_serious(-EFAULT));
832 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
834 if (!info->mti_cross_ref) {
836 * XXX: Check for "namelen == 0" is for getattr by fid
837 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
838 * that is the name must contain at least one character and
839 * the terminating '\0'
842 reqbody = req_capsule_client_get(info->mti_pill,
844 LASSERT(fid_is_sane(&reqbody->fid2));
847 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
849 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
852 lname = mdt_name(info->mti_env, (char *)name, namelen);
853 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
854 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
858 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
860 rc = mdt_object_exists(parent);
861 if (unlikely(rc == 0)) {
862 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
863 &parent->mot_obj.mo_lu,
864 "Parent doesn't exist!\n");
866 } else if (!info->mti_cross_ref) {
867 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
868 PFID(mdt_object_fid(parent)));
871 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
879 if (info->mti_cross_ref) {
880 /* Only getattr on the child. Parent is on another node. */
881 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
883 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
884 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
887 /* Do not take lock for resent case. */
888 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
889 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
890 lhc->mlh_reg_lh.cookie);
891 LASSERT(fid_res_name_eq(mdt_object_fid(child),
892 &lock->l_resource->lr_name));
896 mdt_lock_handle_init(lhc);
897 mdt_lock_reg_init(lhc, LCK_PR);
900 * Object's name is on another MDS, no lookup lock is
901 * needed here but update is.
903 child_bits &= ~MDS_INODELOCK_LOOKUP;
904 child_bits |= MDS_INODELOCK_UPDATE;
906 rc = mdt_object_lock(info, child, lhc, child_bits,
910 /* Finally, we can get attr for child. */
911 mdt_set_capainfo(info, 0, mdt_object_fid(child),
913 rc = mdt_getattr_internal(info, child, 0);
914 if (unlikely(rc != 0))
915 mdt_object_unlock(info, child, lhc, 1);
920 /* step 1: lock parent */
921 lhp = &info->mti_lh[MDT_LH_PARENT];
922 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
923 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
926 if (unlikely(rc != 0))
930 /* step 2: lookup child's fid by name */
931 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
936 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
937 GOTO(out_parent, rc);
939 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
941 *child_fid = reqbody->fid2;
942 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
946 *step 3: find the child object by fid & lock it.
947 * regardless if it is local or remote.
949 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
951 if (unlikely(IS_ERR(child)))
952 GOTO(out_parent, rc = PTR_ERR(child));
954 /* Do not take lock for resent case. */
955 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
956 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
957 lhc->mlh_reg_lh.cookie);
959 res_id = &lock->l_resource->lr_name;
960 if (!fid_res_name_eq(mdt_object_fid(child),
961 &lock->l_resource->lr_name)) {
962 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
963 &lock->l_resource->lr_name),
964 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
965 (unsigned long)res_id->name[0],
966 (unsigned long)res_id->name[1],
967 (unsigned long)res_id->name[2],
968 PFID(mdt_object_fid(parent)));
969 CWARN("Although resent, but still not get child lock"
970 "parent:"DFID" child:"DFID"\n",
971 PFID(mdt_object_fid(parent)),
972 PFID(mdt_object_fid(child)));
973 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
982 ma = &info->mti_attr;
984 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
985 mdt_lock_handle_init(lhc);
986 mdt_lock_reg_init(lhc, LCK_PR);
988 if (mdt_object_exists(child) == 0) {
989 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
990 &child->mot_obj.mo_lu,
991 "Object doesn't exist!\n");
992 GOTO(out_child, rc = -ESTALE);
996 ma->ma_need = MA_INODE;
997 rc = mo_attr_get(info->mti_env, next, ma);
998 if (unlikely(rc != 0))
1001 /* If the file has not been changed for some time, we return
1002 * not only a LOOKUP lock, but also an UPDATE lock and this
1003 * might save us RPC on later STAT. For directories, it also
1004 * let negative dentry starts working for this dir. */
1005 if (ma->ma_valid & MA_INODE &&
1006 ma->ma_attr.la_valid & LA_CTIME &&
1007 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1008 ma->ma_attr.la_ctime < cfs_time_current_sec())
1009 child_bits |= MDS_INODELOCK_UPDATE;
1011 rc = mdt_object_lock(info, child, lhc, child_bits,
1014 if (unlikely(rc != 0))
1015 GOTO(out_child, rc);
1018 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1019 /* Get MA_SOM attributes if update lock is given. */
1021 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1022 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1025 /* finally, we can get attr for child. */
1026 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1027 rc = mdt_getattr_internal(info, child, ma_need);
1028 if (unlikely(rc != 0)) {
1029 mdt_object_unlock(info, child, lhc, 1);
1031 /* Debugging code. */
1032 res_id = &lock->l_resource->lr_name;
1033 LDLM_DEBUG(lock, "Returning lock to client");
1034 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1035 &lock->l_resource->lr_name),
1036 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1037 (unsigned long)res_id->name[0],
1038 (unsigned long)res_id->name[1],
1039 (unsigned long)res_id->name[2],
1040 PFID(mdt_object_fid(child)));
1041 mdt_pack_size2body(info, child);
1044 LDLM_LOCK_PUT(lock);
1048 mdt_object_put(info->mti_env, child);
1050 mdt_object_unlock(info, parent, lhp, 1);
1054 /* normal handler: should release the child lock */
1055 static int mdt_getattr_name(struct mdt_thread_info *info)
1057 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1058 struct mdt_body *reqbody;
1059 struct mdt_body *repbody;
1063 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1064 LASSERT(reqbody != NULL);
1065 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1066 LASSERT(repbody != NULL);
1068 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1069 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1070 repbody->eadatasize = 0;
1071 repbody->aclsize = 0;
1073 rc = mdt_init_ucred(info, reqbody);
1075 GOTO(out_shrink, rc);
1077 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1078 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1079 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1080 lhc->mlh_reg_lh.cookie = 0;
1082 mdt_exit_ucred(info);
1085 mdt_shrink_reply(info);
1089 static const struct lu_device_operations mdt_lu_ops;
1091 static int lu_device_is_mdt(struct lu_device *d)
1093 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1096 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1097 void *karg, void *uarg);
1099 static int mdt_set_info(struct mdt_thread_info *info)
1101 struct ptlrpc_request *req = mdt_info_req(info);
1104 int keylen, vallen, rc = 0;
1107 rc = req_capsule_server_pack(info->mti_pill);
1111 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1113 DEBUG_REQ(D_HA, req, "no set_info key");
1117 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1120 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1122 DEBUG_REQ(D_HA, req, "no set_info val");
1126 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1129 /* Swab any part of val you need to here */
1130 if (KEY_IS(KEY_READ_ONLY)) {
1132 lustre_msg_set_status(req->rq_repmsg, 0);
1134 cfs_spin_lock(&req->rq_export->exp_lock);
1136 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1138 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1139 cfs_spin_unlock(&req->rq_export->exp_lock);
1141 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1142 struct changelog_setinfo *cs =
1143 (struct changelog_setinfo *)val;
1144 if (vallen != sizeof(*cs)) {
1145 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1148 if (ptlrpc_req_need_swab(req)) {
1149 __swab64s(&cs->cs_recno);
1150 __swab32s(&cs->cs_id);
1153 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1155 lustre_msg_set_status(req->rq_repmsg, rc);
1163 static int mdt_connect(struct mdt_thread_info *info)
1166 struct ptlrpc_request *req;
1168 req = mdt_info_req(info);
1169 rc = target_handle_connect(req);
1171 LASSERT(req->rq_export != NULL);
1172 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1173 rc = mdt_init_sec_level(info);
1175 rc = mdt_init_idmap(info);
1177 obd_disconnect(class_export_get(req->rq_export));
1179 rc = err_serious(rc);
1184 static int mdt_disconnect(struct mdt_thread_info *info)
1189 rc = target_handle_disconnect(mdt_info_req(info));
1191 rc = err_serious(rc);
1195 static int mdt_sendpage(struct mdt_thread_info *info,
1196 struct lu_rdpg *rdpg)
1198 struct ptlrpc_request *req = mdt_info_req(info);
1199 struct obd_export *exp = req->rq_export;
1200 struct ptlrpc_bulk_desc *desc;
1201 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1209 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1214 for (i = 0, tmpcount = rdpg->rp_count;
1215 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1216 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1217 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1220 LASSERT(desc->bd_nob == rdpg->rp_count);
1221 rc = sptlrpc_svc_wrap_bulk(req, desc);
1223 GOTO(free_desc, rc);
1225 rc = ptlrpc_start_bulk_transfer(desc);
1227 GOTO(free_desc, rc);
1229 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1230 GOTO(abort_bulk, rc = 0);
1232 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1234 CERROR("Req deadline already passed %lu (now: %lu)\n",
1235 req->rq_deadline, cfs_time_current_sec());
1236 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1237 cfs_time_seconds(1), NULL, NULL);
1238 rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
1239 exp->exp_failed || exp->exp_abort_active_req, lwi);
1240 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1243 if (desc->bd_success &&
1244 desc->bd_nob_transferred == rdpg->rp_count)
1245 GOTO(free_desc, rc);
1248 if (exp->exp_abort_active_req || exp->exp_failed)
1249 GOTO(abort_bulk, rc);
1252 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1253 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1254 desc->bd_nob_transferred, rdpg->rp_count,
1255 exp->exp_client_uuid.uuid,
1256 exp->exp_connection->c_remote_uuid.uuid);
1258 class_fail_export(exp);
1262 ptlrpc_abort_bulk(desc);
1264 ptlrpc_free_bulk(desc);
1268 #ifdef HAVE_SPLIT_SUPPORT
1270 * Retrieve dir entry from the page and insert it to the slave object, actually,
1271 * this should be in osd layer, but since it will not in the final product, so
1272 * just do it here and do not define more moo api anymore for this.
1274 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1277 struct mdt_object *object = info->mti_object;
1278 struct lu_fid *lf = &info->mti_tmp_fid2;
1279 struct md_attr *ma = &info->mti_attr;
1280 struct lu_dirpage *dp;
1281 struct lu_dirent *ent;
1282 int rc = 0, offset = 0;
1285 /* Make sure we have at least one entry. */
1290 * Disable trans for this name insert, since it will include many trans
1293 info->mti_no_need_trans = 1;
1295 * When write_dir_page, no need update parent's ctime,
1296 * and no permission check for name_insert.
1298 ma->ma_attr.la_ctime = 0;
1299 ma->ma_attr.la_valid = LA_MODE;
1300 ma->ma_valid = MA_INODE;
1303 dp = page_address(page);
1304 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1306 for (ent = lu_dirent_start(dp); ent != NULL;
1307 ent = lu_dirent_next(ent)) {
1308 struct lu_name *lname;
1311 if (le16_to_cpu(ent->lde_namelen) == 0)
1314 fid_le_to_cpu(lf, &ent->lde_fid);
1315 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1316 ma->ma_attr.la_mode = S_IFDIR;
1318 ma->ma_attr.la_mode = 0;
1319 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1321 GOTO(out, rc = -ENOMEM);
1323 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1324 lname = mdt_name(info->mti_env, name,
1325 le16_to_cpu(ent->lde_namelen));
1326 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1327 rc = mdo_name_insert(info->mti_env,
1328 md_object_next(&object->mot_obj),
1330 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1332 CERROR("Can't insert %*.*s, rc %d\n",
1333 le16_to_cpu(ent->lde_namelen),
1334 le16_to_cpu(ent->lde_namelen),
1339 offset += lu_dirent_size(ent);
1349 static int mdt_bulk_timeout(void *data)
1353 CERROR("mdt bulk transfer timeout \n");
1358 static int mdt_writepage(struct mdt_thread_info *info)
1360 struct ptlrpc_request *req = mdt_info_req(info);
1361 struct mdt_body *reqbody;
1362 struct l_wait_info *lwi;
1363 struct ptlrpc_bulk_desc *desc;
1369 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1370 if (reqbody == NULL)
1371 RETURN(err_serious(-EFAULT));
1373 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1375 RETURN(err_serious(-ENOMEM));
1377 /* allocate the page for the desc */
1378 page = cfs_alloc_page(CFS_ALLOC_STD);
1380 GOTO(desc_cleanup, rc = -ENOMEM);
1382 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1383 (int)reqbody->size, (int)reqbody->nlink);
1385 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1386 (int)reqbody->nlink);
1388 rc = sptlrpc_svc_prep_bulk(req, desc);
1390 GOTO(cleanup_page, rc);
1392 * Check if client was evicted while we were doing i/o before touching
1397 GOTO(cleanup_page, rc = -ENOMEM);
1399 if (desc->bd_export->exp_failed)
1402 rc = ptlrpc_start_bulk_transfer (desc);
1404 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1405 mdt_bulk_timeout, desc);
1406 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1407 desc->bd_export->exp_failed, lwi);
1408 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1409 if (rc == -ETIMEDOUT) {
1410 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1411 ptlrpc_abort_bulk(desc);
1412 } else if (desc->bd_export->exp_failed) {
1413 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1415 ptlrpc_abort_bulk(desc);
1416 } else if (!desc->bd_success ||
1417 desc->bd_nob_transferred != desc->bd_nob) {
1418 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1420 "truncated" : "network error on",
1421 desc->bd_nob_transferred, desc->bd_nob);
1422 /* XXX should this be a different errno? */
1426 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1429 GOTO(cleanup_lwi, rc);
1430 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1435 cfs_free_page(page);
1437 ptlrpc_free_bulk(desc);
1442 static int mdt_readpage(struct mdt_thread_info *info)
1444 struct mdt_object *object = info->mti_object;
1445 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1446 struct mdt_body *reqbody;
1447 struct mdt_body *repbody;
1452 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1453 RETURN(err_serious(-ENOMEM));
1455 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1456 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1457 if (reqbody == NULL || repbody == NULL)
1458 RETURN(err_serious(-EFAULT));
1461 * prepare @rdpg before calling lower layers and transfer itself. Here
1462 * reqbody->size contains offset of where to start to read and
1463 * reqbody->nlink contains number bytes to read.
1465 rdpg->rp_hash = reqbody->size;
1466 if (rdpg->rp_hash != reqbody->size) {
1467 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1468 rdpg->rp_hash, reqbody->size);
1472 rdpg->rp_attrs = reqbody->mode;
1473 rdpg->rp_count = reqbody->nlink;
1474 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1475 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1476 if (rdpg->rp_pages == NULL)
1479 for (i = 0; i < rdpg->rp_npages; ++i) {
1480 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1481 if (rdpg->rp_pages[i] == NULL)
1482 GOTO(free_rdpg, rc = -ENOMEM);
1485 /* call lower layers to fill allocated pages with directory data */
1486 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1488 GOTO(free_rdpg, rc);
1490 /* send pages to client */
1491 rc = mdt_sendpage(info, rdpg);
1496 for (i = 0; i < rdpg->rp_npages; i++)
1497 if (rdpg->rp_pages[i] != NULL)
1498 cfs_free_page(rdpg->rp_pages[i]);
1499 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1501 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1507 static int mdt_reint_internal(struct mdt_thread_info *info,
1508 struct mdt_lock_handle *lhc,
1511 struct req_capsule *pill = info->mti_pill;
1512 struct mdt_device *mdt = info->mti_mdt;
1513 struct md_quota *mq = md_quota(info->mti_env);
1514 struct mdt_body *repbody;
1519 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1520 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1521 mdt->mdt_max_mdsize);
1522 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1523 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1524 mdt->mdt_max_cookiesize);
1526 rc = req_capsule_server_pack(pill);
1528 CERROR("Can't pack response, rc %d\n", rc);
1529 RETURN(err_serious(rc));
1532 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1533 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1535 repbody->eadatasize = 0;
1536 repbody->aclsize = 0;
1539 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1540 GOTO(out_shrink, rc = err_serious(-EFAULT));
1542 rc = mdt_reint_unpack(info, op);
1544 CERROR("Can't unpack reint, rc %d\n", rc);
1545 GOTO(out_shrink, rc = err_serious(rc));
1548 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1550 /* for replay no cookkie / lmm need, because client have this already */
1551 if (info->mti_spec.no_create == 1) {
1552 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1553 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1555 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1556 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1560 rc = mdt_init_ucred_reint(info);
1562 GOTO(out_shrink, rc);
1564 rc = mdt_fix_attr_ucred(info, op);
1566 GOTO(out_ucred, rc = err_serious(rc));
1568 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1569 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1570 GOTO(out_ucred, rc);
1572 mq->mq_exp = info->mti_exp;
1573 rc = mdt_reint_rec(info, lhc);
1576 mdt_exit_ucred(info);
1578 mdt_shrink_reply(info);
1582 static long mdt_reint_opcode(struct mdt_thread_info *info,
1583 const struct req_format **fmt)
1585 struct mdt_rec_reint *rec;
1588 opc = err_serious(-EFAULT);
1589 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1591 opc = rec->rr_opcode;
1592 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1593 if (opc < REINT_MAX && fmt[opc] != NULL)
1594 req_capsule_extend(info->mti_pill, fmt[opc]);
1596 CERROR("Unsupported opc: %ld\n", opc);
1597 opc = err_serious(opc);
1603 static int mdt_reint(struct mdt_thread_info *info)
1608 static const struct req_format *reint_fmts[REINT_MAX] = {
1609 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1610 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1611 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1612 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1613 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1614 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1615 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1620 opc = mdt_reint_opcode(info, reint_fmts);
1623 * No lock possible here from client to pass it to reint code
1626 rc = mdt_reint_internal(info, NULL, opc);
1631 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1635 /* this should sync the whole device */
1636 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1638 struct dt_device *dt = mdt->mdt_bottom;
1642 rc = dt->dd_ops->dt_sync(env, dt);
1646 /* this should sync this object */
1647 static int mdt_object_sync(struct mdt_thread_info *info)
1649 struct md_object *next;
1653 if (!mdt_object_exists(info->mti_object)) {
1654 CWARN("Non existing object "DFID"!\n",
1655 PFID(mdt_object_fid(info->mti_object)));
1658 next = mdt_object_child(info->mti_object);
1659 rc = mo_object_sync(info->mti_env, next);
1664 static int mdt_sync(struct mdt_thread_info *info)
1666 struct req_capsule *pill = info->mti_pill;
1667 struct mdt_body *body;
1671 /* The fid may be zero, so we req_capsule_set manually */
1672 req_capsule_set(pill, &RQF_MDS_SYNC);
1674 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1676 RETURN(err_serious(-EINVAL));
1678 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1679 RETURN(err_serious(-ENOMEM));
1681 if (fid_seq(&body->fid1) == 0) {
1682 /* sync the whole device */
1683 rc = req_capsule_server_pack(pill);
1685 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1687 rc = err_serious(rc);
1689 /* sync an object */
1690 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1692 rc = mdt_object_sync(info);
1694 struct md_object *next;
1695 const struct lu_fid *fid;
1696 struct lu_attr *la = &info->mti_attr.ma_attr;
1698 next = mdt_object_child(info->mti_object);
1699 info->mti_attr.ma_need = MA_INODE;
1700 info->mti_attr.ma_valid = 0;
1701 rc = mo_attr_get(info->mti_env, next,
1704 body = req_capsule_server_get(pill,
1706 fid = mdt_object_fid(info->mti_object);
1707 mdt_pack_attr2body(info, body, la, fid);
1711 rc = err_serious(rc);
1716 #ifdef HAVE_QUOTA_SUPPORT
1717 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1719 struct obd_quotactl *oqctl;
1720 struct req_capsule *pill = info->mti_pill;
1721 struct obd_export *exp = info->mti_exp;
1722 struct md_quota *mq = md_quota(info->mti_env);
1723 struct md_device *next = info->mti_mdt->mdt_child;
1727 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1731 /* remote client has no permission for quotacheck */
1732 if (unlikely(exp_connect_rmtclient(exp)))
1735 rc = req_capsule_server_pack(pill);
1740 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1745 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1747 struct obd_quotactl *oqctl, *repoqc;
1748 struct req_capsule *pill = info->mti_pill;
1749 struct obd_export *exp = info->mti_exp;
1750 struct md_quota *mq = md_quota(info->mti_env);
1751 struct md_device *next = info->mti_mdt->mdt_child;
1752 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1756 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1761 if (exp_connect_rmtclient(exp)) {
1762 struct ptlrpc_request *req = mdt_info_req(info);
1763 struct mdt_export_data *med = mdt_req2med(req);
1764 struct lustre_idmap_table *idmap = med->med_idmap;
1766 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1767 oqctl->qc_cmd != Q_GETINFO))
1771 if (oqctl->qc_type == USRQUOTA)
1772 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1774 else if (oqctl->qc_type == GRPQUOTA)
1775 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1780 if (id == CFS_IDMAP_NOTFOUND) {
1781 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1787 rc = req_capsule_server_pack(pill);
1791 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1792 LASSERT(repoqc != NULL);
1795 switch (oqctl->qc_cmd) {
1797 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1800 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1803 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1807 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1811 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1815 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1819 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1823 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1826 case LUSTRE_Q_INVALIDATE:
1827 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1829 case LUSTRE_Q_FINVALIDATE:
1830 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1833 CERROR("unsupported mdt_quotactl command: %d\n",
1845 * OBD PING and other handlers.
1847 static int mdt_obd_ping(struct mdt_thread_info *info)
1852 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1854 rc = target_handle_ping(mdt_info_req(info));
1856 rc = err_serious(rc);
1860 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1862 return err_serious(-EOPNOTSUPP);
1865 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1867 return err_serious(-EOPNOTSUPP);
1875 /** clone llog ctxt from child (mdd)
1876 * This allows remote llog (replicator) access.
1877 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1878 * context was originally set up, or we can handle them directly.
1879 * I choose the latter, but that means I need any llog
1880 * contexts set up by child to be accessable by the mdt. So we clone the
1881 * context into our context list here.
1883 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1886 struct md_device *next = mdt->mdt_child;
1887 struct llog_ctxt *ctxt;
1890 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1893 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1894 if (rc || ctxt == NULL) {
1895 CERROR("Can't get mdd ctxt %d\n", rc);
1899 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1901 CERROR("Can't set mdt ctxt %d\n", rc);
1906 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1907 struct mdt_device *mdt, int idx)
1909 struct llog_ctxt *ctxt;
1911 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1914 /* Put once for the get we just did, and once for the clone */
1915 llog_ctxt_put(ctxt);
1916 llog_ctxt_put(ctxt);
1920 static int mdt_llog_create(struct mdt_thread_info *info)
1924 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1925 rc = llog_origin_handle_create(mdt_info_req(info));
1926 return (rc < 0 ? err_serious(rc) : rc);
1929 static int mdt_llog_destroy(struct mdt_thread_info *info)
1933 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1934 rc = llog_origin_handle_destroy(mdt_info_req(info));
1935 return (rc < 0 ? err_serious(rc) : rc);
1938 static int mdt_llog_read_header(struct mdt_thread_info *info)
1942 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1943 rc = llog_origin_handle_read_header(mdt_info_req(info));
1944 return (rc < 0 ? err_serious(rc) : rc);
1947 static int mdt_llog_next_block(struct mdt_thread_info *info)
1951 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1952 rc = llog_origin_handle_next_block(mdt_info_req(info));
1953 return (rc < 0 ? err_serious(rc) : rc);
1956 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1960 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1961 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1962 return (rc < 0 ? err_serious(rc) : rc);
1969 static struct ldlm_callback_suite cbs = {
1970 .lcs_completion = ldlm_server_completion_ast,
1971 .lcs_blocking = ldlm_server_blocking_ast,
1975 static int mdt_enqueue(struct mdt_thread_info *info)
1977 struct ptlrpc_request *req;
1981 * info->mti_dlm_req already contains swapped and (if necessary)
1982 * converted dlm request.
1984 LASSERT(info->mti_dlm_req != NULL);
1986 req = mdt_info_req(info);
1987 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1988 req, info->mti_dlm_req, &cbs);
1989 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1990 return rc ? err_serious(rc) : req->rq_status;
1993 static int mdt_convert(struct mdt_thread_info *info)
1996 struct ptlrpc_request *req;
1998 LASSERT(info->mti_dlm_req);
1999 req = mdt_info_req(info);
2000 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2001 return rc ? err_serious(rc) : req->rq_status;
2004 static int mdt_bl_callback(struct mdt_thread_info *info)
2006 CERROR("bl callbacks should not happen on MDS\n");
2008 return err_serious(-EOPNOTSUPP);
2011 static int mdt_cp_callback(struct mdt_thread_info *info)
2013 CERROR("cp callbacks should not happen on MDS\n");
2015 return err_serious(-EOPNOTSUPP);
2019 * sec context handlers
2021 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2025 rc = mdt_handle_idmap(info);
2028 struct ptlrpc_request *req = mdt_info_req(info);
2031 opc = lustre_msg_get_opc(req->rq_reqmsg);
2032 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2033 sptlrpc_svc_ctx_invalidate(req);
2036 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2041 static struct mdt_object *mdt_obj(struct lu_object *o)
2043 LASSERT(lu_device_is_mdt(o->lo_dev));
2044 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2047 struct mdt_object *mdt_object_find(const struct lu_env *env,
2048 struct mdt_device *d,
2049 const struct lu_fid *f)
2051 struct lu_object *o;
2052 struct mdt_object *m;
2055 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2056 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2057 if (unlikely(IS_ERR(o)))
2058 m = (struct mdt_object *)o;
2065 * Asyncronous commit for mdt device.
2067 * Pass asynchonous commit call down the MDS stack.
2069 * \param env environment
2070 * \param mdt the mdt device
2072 static void mdt_device_commit_async(const struct lu_env *env,
2073 struct mdt_device *mdt)
2075 struct dt_device *dt = mdt->mdt_bottom;
2078 rc = dt->dd_ops->dt_commit_async(env, dt);
2079 if (unlikely(rc != 0))
2080 CWARN("async commit start failed with rc = %d", rc);
2084 * Mark the lock as "synchonous".
2086 * Mark the lock to deffer transaction commit to the unlock time.
2088 * \param lock the lock to mark as "synchonous"
2090 * \see mdt_is_lock_sync
2091 * \see mdt_save_lock
2093 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2095 lock->l_ast_data = (void*)1;
2099 * Check whehter the lock "synchonous" or not.
2101 * \param lock the lock to check
2102 * \retval 1 the lock is "synchonous"
2103 * \retval 0 the lock isn't "synchronous"
2105 * \see mdt_set_lock_sync
2106 * \see mdt_save_lock
2108 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2110 return lock->l_ast_data != NULL;
2114 * Blocking AST for mdt locks.
2116 * Starts transaction commit if in case of COS lock conflict or
2117 * deffers such a commit to the mdt_save_lock.
2119 * \param lock the lock which blocks a request or cancelling lock
2120 * \param desc unused
2121 * \param data unused
2122 * \param flag indicates whether this cancelling or blocking callback
2124 * \see ldlm_blocking_ast_nocheck
2126 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2127 void *data, int flag)
2129 struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2130 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2134 if (flag == LDLM_CB_CANCELING)
2136 lock_res_and_lock(lock);
2137 if (lock->l_blocking_ast != mdt_blocking_ast) {
2138 unlock_res_and_lock(lock);
2141 if (mdt_cos_is_enabled(mdt) &&
2142 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2143 lock->l_blocking_lock != NULL &&
2144 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2145 mdt_set_lock_sync(lock);
2147 rc = ldlm_blocking_ast_nocheck(lock);
2149 /* There is no lock conflict if l_blocking_lock == NULL,
2150 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2151 * when the last reference to a local lock was released */
2152 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2155 rc = lu_env_init(&env, LCT_MD_THREAD);
2156 if (unlikely(rc != 0))
2157 CWARN("lu_env initialization failed with rc = %d,"
2158 "cannot start asynchronous commit\n", rc);
2160 mdt_device_commit_async(&env, mdt);
2166 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2167 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2169 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2170 ldlm_policy_data_t *policy = &info->mti_policy;
2171 struct ldlm_res_id *res_id = &info->mti_res_id;
2175 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2176 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2177 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2178 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2180 if (mdt_object_exists(o) < 0) {
2181 if (locality == MDT_CROSS_LOCK) {
2182 /* cross-ref object fix */
2183 ibits &= ~MDS_INODELOCK_UPDATE;
2184 ibits |= MDS_INODELOCK_LOOKUP;
2186 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2187 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2189 /* No PDO lock on remote object */
2190 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2193 if (lh->mlh_type == MDT_PDO_LOCK) {
2194 /* check for exists after object is locked */
2195 if (mdt_object_exists(o) == 0) {
2196 /* Non-existent object shouldn't have PDO lock */
2199 /* Non-dir object shouldn't have PDO lock */
2200 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2204 memset(policy, 0, sizeof(*policy));
2205 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2208 * Take PDO lock on whole directory and build correct @res_id for lock
2209 * on part of directory.
2211 if (lh->mlh_pdo_hash != 0) {
2212 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2213 mdt_lock_pdo_mode(info, o, lh);
2214 if (lh->mlh_pdo_mode != LCK_NL) {
2216 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2217 * is never going to be sent to client and we do not
2218 * want it slowed down due to possible cancels.
2220 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2221 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2222 policy, res_id, LDLM_FL_ATOMIC_CB,
2223 &info->mti_exp->exp_handle.h_cookie);
2229 * Finish res_id initializing by name hash marking part of
2230 * directory which is taking modification.
2232 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2235 policy->l_inodebits.bits = ibits;
2238 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2239 * going to be sent to client. If it is - mdt_intent_policy() path will
2240 * fix it up and turn FL_LOCAL flag off.
2242 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2243 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2244 &info->mti_exp->exp_handle.h_cookie);
2246 mdt_object_unlock(info, o, lh, 1);
2247 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2248 lh->mlh_pdo_hash != 0 &&
2249 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2250 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2257 * Save a lock within request object.
2259 * Keep the lock referenced until whether client ACK or transaction
2260 * commit happens or release the lock immediately depending on input
2261 * parameters. If COS is ON, a write lock is converted to COS lock
2264 * \param info thead info object
2265 * \param h lock handle
2266 * \param mode lock mode
2267 * \param decref force immediate lock releasing
2270 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2271 ldlm_mode_t mode, int decref)
2275 if (lustre_handle_is_used(h)) {
2276 if (decref || !info->mti_has_trans ||
2277 !(mode & (LCK_PW | LCK_EX))){
2278 mdt_fid_unlock(h, mode);
2280 struct mdt_device *mdt = info->mti_mdt;
2281 struct ldlm_lock *lock = ldlm_handle2lock(h);
2282 struct ptlrpc_request *req = mdt_info_req(info);
2285 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2287 CDEBUG(D_HA, "request = %p reply state = %p"
2288 " transno = "LPD64"\n",
2289 req, req->rq_reply_state, req->rq_transno);
2290 if (mdt_cos_is_enabled(mdt)) {
2292 ldlm_lock_downgrade(lock, LCK_COS);
2295 ptlrpc_save_lock(req, h, mode, no_ack);
2296 if (mdt_is_lock_sync(lock)) {
2297 CDEBUG(D_HA, "found sync-lock,"
2298 " async commit started\n");
2299 mdt_device_commit_async(info->mti_env,
2302 LDLM_LOCK_PUT(lock);
2311 * Unlock mdt object.
2313 * Immeditely release the regular lock and the PDO lock or save the
2314 * lock in reqeuest and keep them referenced until client ACK or
2315 * transaction commit.
2317 * \param info thread info object
2318 * \param o mdt object
2319 * \param lh mdt lock handle referencing regular and PDO locks
2320 * \param decref force immediate lock releasing
2322 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2323 struct mdt_lock_handle *lh, int decref)
2327 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2328 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2333 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2334 const struct lu_fid *f,
2335 struct mdt_lock_handle *lh,
2338 struct mdt_object *o;
2340 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2344 rc = mdt_object_lock(info, o, lh, ibits,
2347 mdt_object_put(info->mti_env, o);
2354 void mdt_object_unlock_put(struct mdt_thread_info * info,
2355 struct mdt_object * o,
2356 struct mdt_lock_handle *lh,
2359 mdt_object_unlock(info, o, lh, decref);
2360 mdt_object_put(info->mti_env, o);
2363 static struct mdt_handler *mdt_handler_find(__u32 opc,
2364 struct mdt_opc_slice *supported)
2366 struct mdt_opc_slice *s;
2367 struct mdt_handler *h;
2370 for (s = supported; s->mos_hs != NULL; s++) {
2371 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2372 h = s->mos_hs + (opc - s->mos_opc_start);
2373 if (likely(h->mh_opc != 0))
2374 LASSERTF(h->mh_opc == opc,
2375 "opcode mismatch %d != %d\n",
2378 h = NULL; /* unsupported opc */
2385 static int mdt_lock_resname_compat(struct mdt_device *m,
2386 struct ldlm_request *req)
2388 /* XXX something... later. */
2392 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2394 /* XXX something... later. */
2399 * Generic code handling requests that have struct mdt_body passed in:
2401 * - extract mdt_body from request and save it in @info, if present;
2403 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2406 * - if HABEO_CORPUS flag is set for this request type check whether object
2407 * actually exists on storage (lu_object_exists()).
2410 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2412 const struct mdt_body *body;
2413 struct mdt_object *obj;
2414 const struct lu_env *env;
2415 struct req_capsule *pill;
2419 env = info->mti_env;
2420 pill = info->mti_pill;
2422 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2426 if (!(body->valid & OBD_MD_FLID))
2429 if (!fid_is_sane(&body->fid1)) {
2430 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2435 * Do not get size or any capa fields before we check that request
2436 * contains capa actually. There are some requests which do not, for
2437 * instance MDS_IS_SUBDIR.
2439 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2440 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2441 mdt_set_capainfo(info, 0, &body->fid1,
2442 req_capsule_client_get(pill, &RMF_CAPA1));
2444 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2446 if ((flags & HABEO_CORPUS) &&
2447 !mdt_object_exists(obj)) {
2448 mdt_object_put(env, obj);
2449 /* for capability renew ENOENT will be handled in
2451 if (body->valid & OBD_MD_FLOSSCAPA)
2456 info->mti_object = obj;
2465 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2467 struct req_capsule *pill = info->mti_pill;
2471 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2472 rc = mdt_body_unpack(info, flags);
2476 if (rc == 0 && (flags & HABEO_REFERO)) {
2477 struct mdt_device *mdt = info->mti_mdt;
2481 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2482 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2483 mdt->mdt_max_mdsize);
2484 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2485 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2486 mdt->mdt_max_cookiesize);
2488 rc = req_capsule_server_pack(pill);
2493 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2495 struct md_device *next = m->mdt_child;
2497 return next->md_ops->mdo_init_capa_ctxt(env, next,
2498 m->mdt_opts.mo_mds_capa,
2499 m->mdt_capa_timeout,
2505 * Invoke handler for this request opc. Also do necessary preprocessing
2506 * (according to handler ->mh_flags), and post-processing (setting of
2507 * ->last_{xid,committed}).
2509 static int mdt_req_handle(struct mdt_thread_info *info,
2510 struct mdt_handler *h, struct ptlrpc_request *req)
2512 int rc, serious = 0;
2517 LASSERT(h->mh_act != NULL);
2518 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2519 LASSERT(current->journal_info == NULL);
2522 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2523 * to put same checks into handlers like mdt_close(), mdt_reint(),
2524 * etc., without talking to mdt authors first. Checking same thing
2525 * there again is useless and returning 0 error without packing reply
2526 * is buggy! Handlers either pack reply or return error.
2528 * We return 0 here and do not send any reply in order to emulate
2529 * network failure. Do not send any reply in case any of NET related
2530 * fail_id has occured.
2532 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2536 flags = h->mh_flags;
2537 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2539 if (h->mh_fmt != NULL) {
2540 req_capsule_set(info->mti_pill, h->mh_fmt);
2541 rc = mdt_unpack_req_pack_rep(info, flags);
2544 if (rc == 0 && flags & MUTABOR &&
2545 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2546 /* should it be rq_status? */
2549 if (rc == 0 && flags & HABEO_CLAVIS) {
2550 struct ldlm_request *dlm_req;
2552 LASSERT(h->mh_fmt != NULL);
2554 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2555 if (dlm_req != NULL) {
2556 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2558 dlm_req->lock_desc.l_policy_data.\
2559 l_inodebits.bits == 0)) {
2561 * Lock without inodebits makes no sense and
2562 * will oops later in ldlm. If client miss to
2563 * set such bits, do not trigger ASSERTION.
2565 * For liblustre flock case, it maybe zero.
2569 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2570 rc = mdt_lock_resname_compat(
2573 info->mti_dlm_req = dlm_req;
2580 /* capability setting changed via /proc, needs reinitialize ctxt */
2581 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2582 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2583 info->mti_mdt->mdt_capa_conf = 0;
2586 if (likely(rc == 0)) {
2588 * Process request, there can be two types of rc:
2589 * 1) errors with msg unpack/pack, other failures outside the
2590 * operation itself. This is counted as serious errors;
2591 * 2) errors during fs operation, should be placed in rq_status
2594 rc = h->mh_act(info);
2596 !req->rq_no_reply && req->rq_reply_state == NULL) {
2597 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2598 "pack reply and returned 0 error\n",
2602 serious = is_serious(rc);
2603 rc = clear_serious(rc);
2607 req->rq_status = rc;
2610 * ELDLM_* codes which > 0 should be in rq_status only as well as
2611 * all non-serious errors.
2613 if (rc > 0 || !serious)
2616 LASSERT(current->journal_info == NULL);
2618 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2619 info->mti_mdt->mdt_opts.mo_compat_resname) {
2620 struct ldlm_reply *dlmrep;
2622 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2624 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2627 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2628 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2629 target_committed_to_req(req);
2631 if (unlikely(req_is_replay(req) &&
2632 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2633 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2637 target_send_reply(req, rc, info->mti_fail_id);
2641 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2643 lh->mlh_type = MDT_NUL_LOCK;
2644 lh->mlh_reg_lh.cookie = 0ull;
2645 lh->mlh_reg_mode = LCK_MINMODE;
2646 lh->mlh_pdo_lh.cookie = 0ull;
2647 lh->mlh_pdo_mode = LCK_MINMODE;
2650 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2652 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2653 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2657 * Initialize fields of struct mdt_thread_info. Other fields are left in
2658 * uninitialized state, because it's too expensive to zero out whole
2659 * mdt_thread_info (> 1K) on each request arrival.
2661 static void mdt_thread_info_init(struct ptlrpc_request *req,
2662 struct mdt_thread_info *info)
2665 struct md_capainfo *ci;
2667 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2668 info->mti_pill = &req->rq_pill;
2671 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2672 mdt_lock_handle_init(&info->mti_lh[i]);
2674 /* mdt device: it can be NULL while CONNECT */
2675 if (req->rq_export) {
2676 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2677 info->mti_exp = req->rq_export;
2679 info->mti_mdt = NULL;
2680 info->mti_env = req->rq_svc_thread->t_env;
2681 ci = md_capainfo(info->mti_env);
2682 memset(ci, 0, sizeof *ci);
2683 if (req->rq_export) {
2684 if (exp_connect_rmtclient(req->rq_export))
2685 ci->mc_auth = LC_ID_CONVERT;
2686 else if (req->rq_export->exp_connect_flags &
2687 OBD_CONNECT_MDS_CAPA)
2688 ci->mc_auth = LC_ID_PLAIN;
2690 ci->mc_auth = LC_ID_NONE;
2693 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2694 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2695 info->mti_mos = NULL;
2697 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2698 info->mti_body = NULL;
2699 info->mti_object = NULL;
2700 info->mti_dlm_req = NULL;
2701 info->mti_has_trans = 0;
2702 info->mti_no_need_trans = 0;
2703 info->mti_cross_ref = 0;
2704 info->mti_opdata = 0;
2706 /* To not check for split by default. */
2707 info->mti_spec.sp_ck_split = 0;
2708 info->mti_spec.no_create = 0;
2711 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2715 req_capsule_fini(info->mti_pill);
2716 if (info->mti_object != NULL) {
2718 * freeing an object may lead to OSD level transaction, do not
2719 * let it mess with MDT. bz19385.
2721 info->mti_no_need_trans = 1;
2722 mdt_object_put(info->mti_env, info->mti_object);
2723 info->mti_object = NULL;
2725 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2726 mdt_lock_handle_fini(&info->mti_lh[i]);
2727 info->mti_env = NULL;
2730 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2731 struct obd_device *obd, int *process)
2733 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2734 case MDS_CONNECT: /* This will never get here, but for completeness. */
2735 case OST_CONNECT: /* This will never get here, but for completeness. */
2736 case MDS_DISCONNECT:
2737 case OST_DISCONNECT:
2742 case MDS_DONE_WRITING:
2743 case MDS_SYNC: /* used in unmounting */
2749 *process = target_queue_recovery_request(req, obd);
2753 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2760 * Handle recovery. Return:
2761 * +1: continue request processing;
2762 * -ve: abort immediately with the given error code;
2763 * 0: send reply with error code in req->rq_status;
2765 static int mdt_recovery(struct mdt_thread_info *info)
2767 struct ptlrpc_request *req = mdt_info_req(info);
2768 struct obd_device *obd;
2772 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2775 case SEC_CTX_INIT_CONT:
2781 rc = mdt_handle_idmap(info);
2790 if (unlikely(!class_connected_export(req->rq_export))) {
2791 CERROR("operation %d on unconnected MDS from %s\n",
2792 lustre_msg_get_opc(req->rq_reqmsg),
2793 libcfs_id2str(req->rq_peer));
2794 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2795 * mds_A will get -ENOTCONN(especially for ping req),
2796 * which will cause that mds_A deactive timeout, then when
2797 * mds_A cleanup, the cleanup process will be suspended since
2798 * deactive timeout is not zero.
2800 req->rq_status = -ENOTCONN;
2801 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2805 /* sanity check: if the xid matches, the request must be marked as a
2806 * resent or replayed */
2807 if (req_xid_is_last(req)) {
2808 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2809 (MSG_RESENT | MSG_REPLAY))) {
2810 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2811 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2812 lustre_msg_get_flags(req->rq_reqmsg));
2814 req->rq_status = -ENOTCONN;
2819 /* else: note the opposite is not always true; a RESENT req after a
2820 * failover will usually not match the last_xid, since it was likely
2821 * never committed. A REPLAYed request will almost never match the
2822 * last xid, however it could for a committed, but still retained,
2825 obd = req->rq_export->exp_obd;
2827 /* Check for aborted recovery... */
2828 if (unlikely(obd->obd_recovering)) {
2831 DEBUG_REQ(D_INFO, req, "Got new replay");
2832 rc = mdt_filter_recovery_request(req, obd, &should_process);
2833 if (rc != 0 || !should_process)
2835 else if (should_process < 0) {
2836 req->rq_status = should_process;
2837 rc = ptlrpc_error(req);
2844 static int mdt_msg_check_version(struct lustre_msg *msg)
2848 switch (lustre_msg_get_opc(msg)) {
2850 case MDS_DISCONNECT:
2853 case SEC_CTX_INIT_CONT:
2855 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2857 CERROR("bad opc %u version %08x, expecting %08x\n",
2858 lustre_msg_get_opc(msg),
2859 lustre_msg_get_version(msg),
2860 LUSTRE_OBD_VERSION);
2864 case MDS_GETATTR_NAME:
2871 case MDS_DONE_WRITING:
2878 case MDS_QUOTACHECK:
2884 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2886 CERROR("bad opc %u version %08x, expecting %08x\n",
2887 lustre_msg_get_opc(msg),
2888 lustre_msg_get_version(msg),
2889 LUSTRE_MDS_VERSION);
2893 case LDLM_BL_CALLBACK:
2894 case LDLM_CP_CALLBACK:
2895 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2897 CERROR("bad opc %u version %08x, expecting %08x\n",
2898 lustre_msg_get_opc(msg),
2899 lustre_msg_get_version(msg),
2900 LUSTRE_DLM_VERSION);
2902 case OBD_LOG_CANCEL:
2903 case LLOG_ORIGIN_HANDLE_CREATE:
2904 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2905 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2906 case LLOG_ORIGIN_HANDLE_CLOSE:
2907 case LLOG_ORIGIN_HANDLE_DESTROY:
2908 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2910 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2912 CERROR("bad opc %u version %08x, expecting %08x\n",
2913 lustre_msg_get_opc(msg),
2914 lustre_msg_get_version(msg),
2915 LUSTRE_LOG_VERSION);
2918 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2924 static int mdt_handle0(struct ptlrpc_request *req,
2925 struct mdt_thread_info *info,
2926 struct mdt_opc_slice *supported)
2928 struct mdt_handler *h;
2929 struct lustre_msg *msg;
2934 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2937 LASSERT(current->journal_info == NULL);
2939 msg = req->rq_reqmsg;
2940 rc = mdt_msg_check_version(msg);
2941 if (likely(rc == 0)) {
2942 rc = mdt_recovery(info);
2943 if (likely(rc == +1)) {
2944 h = mdt_handler_find(lustre_msg_get_opc(msg),
2946 if (likely(h != NULL)) {
2947 rc = mdt_req_handle(info, h, req);
2949 CERROR("The unsupported opc: 0x%x\n",
2950 lustre_msg_get_opc(msg) );
2951 req->rq_status = -ENOTSUPP;
2952 rc = ptlrpc_error(req);
2957 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2962 * MDT handler function called by ptlrpc service thread when request comes.
2964 * XXX common "target" functionality should be factored into separate module
2965 * shared by mdt, ost and stand-alone services like fld.
2967 static int mdt_handle_common(struct ptlrpc_request *req,
2968 struct mdt_opc_slice *supported)
2971 struct mdt_thread_info *info;
2975 env = req->rq_svc_thread->t_env;
2976 LASSERT(env != NULL);
2977 LASSERT(env->le_ses != NULL);
2978 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2979 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2980 LASSERT(info != NULL);
2982 mdt_thread_info_init(req, info);
2984 rc = mdt_handle0(req, info, supported);
2986 mdt_thread_info_fini(info);
2991 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2994 int mdt_recovery_handle(struct ptlrpc_request *req)
2999 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3001 rc = mdt_handle_common(req, mdt_fld_handlers);
3004 rc = mdt_handle_common(req, mdt_seq_handlers);
3007 rc = mdt_handle_common(req, mdt_regular_handlers);
3014 static int mdt_regular_handle(struct ptlrpc_request *req)
3016 return mdt_handle_common(req, mdt_regular_handlers);
3019 static int mdt_readpage_handle(struct ptlrpc_request *req)
3021 return mdt_handle_common(req, mdt_readpage_handlers);
3024 static int mdt_xmds_handle(struct ptlrpc_request *req)
3026 return mdt_handle_common(req, mdt_xmds_handlers);
3029 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3031 return mdt_handle_common(req, mdt_seq_handlers);
3034 static int mdt_mdss_handle(struct ptlrpc_request *req)
3036 return mdt_handle_common(req, mdt_seq_handlers);
3039 static int mdt_dtss_handle(struct ptlrpc_request *req)
3041 return mdt_handle_common(req, mdt_seq_handlers);
3044 static int mdt_fld_handle(struct ptlrpc_request *req)
3046 return mdt_handle_common(req, mdt_fld_handlers);
3062 static int mdt_intent_getattr(enum mdt_it_code opcode,
3063 struct mdt_thread_info *info,
3064 struct ldlm_lock **,
3066 static int mdt_intent_reint(enum mdt_it_code opcode,
3067 struct mdt_thread_info *info,
3068 struct ldlm_lock **,
3071 static struct mdt_it_flavor {
3072 const struct req_format *it_fmt;
3074 int (*it_act)(enum mdt_it_code ,
3075 struct mdt_thread_info *,
3076 struct ldlm_lock **,
3079 } mdt_it_flavor[] = {
3081 .it_fmt = &RQF_LDLM_INTENT,
3082 /*.it_flags = HABEO_REFERO,*/
3084 .it_act = mdt_intent_reint,
3085 .it_reint = REINT_OPEN
3088 .it_fmt = &RQF_LDLM_INTENT,
3089 .it_flags = MUTABOR,
3090 .it_act = mdt_intent_reint,
3091 .it_reint = REINT_OPEN
3094 .it_fmt = &RQF_LDLM_INTENT,
3095 .it_flags = MUTABOR,
3096 .it_act = mdt_intent_reint,
3097 .it_reint = REINT_CREATE
3099 [MDT_IT_GETATTR] = {
3100 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3101 .it_flags = HABEO_REFERO,
3102 .it_act = mdt_intent_getattr
3104 [MDT_IT_READDIR] = {
3110 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3111 .it_flags = HABEO_REFERO,
3112 .it_act = mdt_intent_getattr
3115 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3116 .it_flags = MUTABOR,
3118 .it_reint = REINT_UNLINK
3122 .it_flags = MUTABOR,
3125 [MDT_IT_GETXATTR] = {
3132 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3133 struct ldlm_lock **lockp,
3134 struct ldlm_lock *new_lock,
3135 struct mdt_lock_handle *lh,
3138 struct ptlrpc_request *req = mdt_info_req(info);
3139 struct ldlm_lock *lock = *lockp;
3142 * Get new lock only for cases when possible resent did not find any
3145 if (new_lock == NULL)
3146 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3148 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3149 lh->mlh_reg_lh.cookie = 0;
3153 LASSERTF(new_lock != NULL,
3154 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3157 * If we've already given this lock to a client once, then we should
3158 * have no readers or writers. Otherwise, we should have one reader
3159 * _or_ writer ref (which will be zeroed below) before returning the
3162 if (new_lock->l_export == req->rq_export) {
3163 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3165 LASSERT(new_lock->l_export == NULL);
3166 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3171 if (new_lock->l_export == req->rq_export) {
3173 * Already gave this to the client, which means that we
3174 * reconstructed a reply.
3176 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3178 lh->mlh_reg_lh.cookie = 0;
3179 RETURN(ELDLM_LOCK_REPLACED);
3183 * Fixup the lock to be given to the client.
3185 lock_res_and_lock(new_lock);
3186 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3187 * possible blocking AST. */
3188 while (new_lock->l_readers > 0) {
3189 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3190 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3191 new_lock->l_readers--;
3193 while (new_lock->l_writers > 0) {
3194 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3195 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3196 new_lock->l_writers--;
3199 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3200 new_lock->l_blocking_ast = lock->l_blocking_ast;
3201 new_lock->l_completion_ast = lock->l_completion_ast;
3202 new_lock->l_remote_handle = lock->l_remote_handle;
3203 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3205 unlock_res_and_lock(new_lock);
3207 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3208 &new_lock->l_remote_handle,
3209 &new_lock->l_exp_hash);
3211 LDLM_LOCK_RELEASE(new_lock);
3212 lh->mlh_reg_lh.cookie = 0;
3214 RETURN(ELDLM_LOCK_REPLACED);
3217 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3218 struct ldlm_lock *new_lock,
3219 struct ldlm_lock **old_lock,
3220 struct mdt_lock_handle *lh)
3222 struct ptlrpc_request *req = mdt_info_req(info);
3223 struct obd_export *exp = req->rq_export;
3224 struct lustre_handle remote_hdl;
3225 struct ldlm_request *dlmreq;
3226 struct ldlm_lock *lock;
3228 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3231 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3232 remote_hdl = dlmreq->lock_handle[0];
3234 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3236 if (lock != new_lock) {
3237 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3238 lh->mlh_reg_mode = lock->l_granted_mode;
3240 LDLM_DEBUG(lock, "Restoring lock cookie");
3241 DEBUG_REQ(D_DLMTRACE, req,
3242 "restoring lock cookie "LPX64,
3243 lh->mlh_reg_lh.cookie);