1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mds) request handler
7 * Copyright (c) 2001-2005 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
9 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * Author: Phil Schwan <phil@clusterfs.com>
11 * Author: Mike Shaver <shaver@clusterfs.com>
13 * This file is part of the Lustre file system, http://www.lustre.org
14 * Lustre is a trademark of Cluster File Systems, Inc.
16 * You may have signed or agreed to another license before downloading
17 * this software. If so, you are bound by the terms and conditions
18 * of that agreement, and the following does not apply to you. See the
19 * LICENSE file included with this distribution for more information.
21 * If you did not agree to a different license, then this copy of Lustre
22 * is open source software; you can redistribute it and/or modify it
23 * under the terms of version 2 of the GNU General Public License as
24 * published by the Free Software Foundation.
26 * In either case, Lustre is distributed in the hope that it will be
27 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
28 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * license text for more details.
33 # define EXPORT_SYMTAB
35 #define DEBUG_SUBSYSTEM S_MDS
37 #include <lustre_mds.h>
38 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/random.h>
42 #include <linux/jbd.h>
43 #include <linux/smp_lock.h>
44 #include <linux/buffer_head.h>
45 #include <linux/workqueue.h>
46 #include <linux/mount.h>
48 #include <linux/lustre_acl.h>
49 #include <obd_class.h>
50 #include <lustre_dlm.h>
52 #include <lustre_fsfilt.h>
53 #include <lprocfs_status.h>
54 #include <lustre_commit_confd.h>
55 #include <lustre_quota.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
59 #include "mds_internal.h"
62 CFS_MODULE_PARM(mds_num_threads, "i", int, 0444,
63 "number of MDS service threads to start");
65 __u32 mds_max_ost_index=0xFFFF;
66 CFS_MODULE_PARM(mds_max_ost_index, "i", int, 0444,
69 static int mds_intent_policy(struct ldlm_namespace *ns,
70 struct ldlm_lock **lockp, void *req_cookie,
71 ldlm_mode_t mode, int flags, void *data);
72 static int mds_postsetup(struct obd_device *obd);
73 static int mds_cleanup(struct obd_device *obd);
75 /* Assumes caller has already pushed into the kernel filesystem context */
76 static int mds_sendpage(struct ptlrpc_request *req, struct file *file,
77 loff_t offset, int count)
79 struct ptlrpc_bulk_desc *desc;
80 struct l_wait_info lwi;
82 int rc = 0, npages, i, tmpcount, tmpsize = 0;
85 LASSERT((offset & ~CFS_PAGE_MASK) == 0); /* I'm dubious about this */
87 npages = (count + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
88 OBD_ALLOC(pages, sizeof(*pages) * npages);
90 GOTO(out, rc = -ENOMEM);
92 desc = ptlrpc_prep_bulk_exp(req, npages, BULK_PUT_SOURCE,
95 GOTO(out_free, rc = -ENOMEM);
97 for (i = 0, tmpcount = count; i < npages; i++, tmpcount -= tmpsize) {
98 tmpsize = tmpcount > CFS_PAGE_SIZE ? CFS_PAGE_SIZE : tmpcount;
100 OBD_PAGE_ALLOC(pages[i], CFS_ALLOC_STD);
101 if (pages[i] == NULL)
102 GOTO(cleanup_buf, rc = -ENOMEM);
104 ptlrpc_prep_bulk_page(desc, pages[i], 0, tmpsize);
107 for (i = 0, tmpcount = count; i < npages; i++, tmpcount -= tmpsize) {
108 tmpsize = tmpcount > CFS_PAGE_SIZE ? CFS_PAGE_SIZE : tmpcount;
109 CDEBUG(D_EXT2, "reading %u@%llu from dir %lu (size %llu)\n",
110 tmpsize, offset, file->f_dentry->d_inode->i_ino,
111 i_size_read(file->f_dentry->d_inode));
113 rc = fsfilt_readpage(req->rq_export->exp_obd, file,
114 kmap(pages[i]), tmpsize, &offset);
118 GOTO(cleanup_buf, rc = -EIO);
121 LASSERT(desc->bd_nob == count);
123 rc = ptlrpc_start_bulk_transfer(desc);
125 GOTO(cleanup_buf, rc);
127 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
128 CERROR("obd_fail_loc=%x, fail operation rc=%d\n",
129 OBD_FAIL_MDS_SENDPAGE, rc);
130 GOTO(abort_bulk, rc);
133 lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
134 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), &lwi);
135 LASSERT (rc == 0 || rc == -ETIMEDOUT);
138 if (desc->bd_success &&
139 desc->bd_nob_transferred == count)
140 GOTO(cleanup_buf, rc);
142 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
145 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
146 (rc == -ETIMEDOUT) ? "timeout" : "network error",
147 desc->bd_nob_transferred, count,
148 req->rq_export->exp_client_uuid.uuid,
149 req->rq_export->exp_connection->c_remote_uuid.uuid);
151 class_fail_export(req->rq_export);
155 ptlrpc_abort_bulk (desc);
157 for (i = 0; i < npages; i++)
159 OBD_PAGE_FREE(pages[i]);
161 ptlrpc_free_bulk(desc);
163 OBD_FREE(pages, sizeof(*pages) * npages);
168 /* only valid locked dentries or errors should be returned */
169 struct dentry *mds_fid2locked_dentry(struct obd_device *obd, struct ll_fid *fid,
170 struct vfsmount **mnt, int lock_mode,
171 struct lustre_handle *lockh,
174 struct mds_obd *mds = &obd->u.mds;
175 struct dentry *de = mds_fid2dentry(mds, fid, mnt), *retval = de;
176 struct ldlm_res_id res_id = { .name = {0} };
177 int flags = LDLM_FL_ATOMIC_CB, rc;
178 ldlm_policy_data_t policy = { .l_inodebits = { lockpart} };
184 res_id.name[0] = de->d_inode->i_ino;
185 res_id.name[1] = de->d_inode->i_generation;
186 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id,
187 LDLM_IBITS, &policy, lock_mode, &flags,
188 ldlm_blocking_ast, ldlm_completion_ast,
189 NULL, NULL, 0, NULL, lockh);
190 if (rc != ELDLM_OK) {
192 retval = ERR_PTR(-EIO); /* XXX translate ldlm code */
198 /* Look up an entry by inode number. */
199 /* this function ONLY returns valid dget'd dentries with an initialized inode
201 struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
202 struct vfsmount **mnt)
205 unsigned long ino = fid->id;
206 __u32 generation = fid->generation;
208 struct dentry *result;
211 RETURN(ERR_PTR(-ESTALE));
213 snprintf(fid_name, sizeof(fid_name), "0x%lx", ino);
215 CDEBUG(D_DENTRY, "--> mds_fid2dentry: ino/gen %lu/%u, sb %p\n",
216 ino, generation, mds->mds_obt.obt_sb);
218 /* under ext3 this is neither supposed to return bad inodes
220 result = ll_lookup_one_len(fid_name, mds->mds_fid_de, strlen(fid_name));
224 inode = result->d_inode;
226 RETURN(ERR_PTR(-ENOENT));
228 if (inode->i_generation == 0 || inode->i_nlink == 0) {
229 LCONSOLE_WARN("Found inode with zero generation or link -- this"
230 " may indicate disk corruption (inode: %lu/%u, "
231 "link %lu, count %d)\n", inode->i_ino,
232 inode->i_generation,(unsigned long)inode->i_nlink,
233 atomic_read(&inode->i_count));
235 RETURN(ERR_PTR(-ENOENT));
238 if (generation && inode->i_generation != generation) {
239 /* we didn't find the right inode.. */
240 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
241 "count: %d, generation %u/%u\n", inode->i_ino,
242 (unsigned long)inode->i_nlink,
243 atomic_read(&inode->i_count), inode->i_generation,
246 RETURN(ERR_PTR(-ENOENT));
250 *mnt = mds->mds_vfsmnt;
257 static int mds_connect_internal(struct obd_export *exp,
258 struct obd_connect_data *data)
260 struct obd_device *obd = exp->exp_obd;
262 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
263 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
265 /* If no known bits (which should not happen, probably,
266 as everybody should support LOOKUP and UPDATE bits at least)
267 revert to compat mode with plain locks. */
268 if (!data->ocd_ibits_known &&
269 data->ocd_connect_flags & OBD_CONNECT_IBITS)
270 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
272 if (!obd->u.mds.mds_fl_acl)
273 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
275 if (!obd->u.mds.mds_fl_user_xattr)
276 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
278 exp->exp_connect_flags = data->ocd_connect_flags;
279 data->ocd_version = LUSTRE_VERSION_CODE;
280 exp->exp_mds_data.med_ibits_known = data->ocd_ibits_known;
283 if (obd->u.mds.mds_fl_acl &&
284 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
285 CWARN("%s: MDS requires ACL support but client does not\n",
292 static int mds_reconnect(const struct lu_env *env,
293 struct obd_export *exp, struct obd_device *obd,
294 struct obd_uuid *cluuid,
295 struct obd_connect_data *data)
300 if (exp == NULL || obd == NULL || cluuid == NULL)
303 rc = mds_connect_internal(exp, data);
308 /* Establish a connection to the MDS.
310 * This will set up an export structure for the client to hold state data
311 * about that client, like open files, the last operation number it did
312 * on the server, etc.
314 static int mds_connect(const struct lu_env *env,
315 struct lustre_handle *conn, struct obd_device *obd,
316 struct obd_uuid *cluuid, struct obd_connect_data *data,
319 struct obd_export *exp;
320 struct mds_export_data *med;
321 struct mds_client_data *mcd = NULL;
325 if (!conn || !obd || !cluuid)
328 /* XXX There is a small race between checking the list and adding a
329 * new connection for the same UUID, but the real threat (list
330 * corruption when multiple different clients connect) is solved.
332 * There is a second race between adding the export to the list,
333 * and filling in the client data below. Hence skipping the case
334 * of NULL mcd above. We should already be controlling multiple
335 * connects at the client, and we can't hold the spinlock over
336 * memory allocations without risk of deadlocking.
338 rc = class_connect(conn, obd, cluuid);
341 exp = class_conn2export(conn);
343 med = &exp->exp_mds_data;
345 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_NULL;
347 rc = mds_connect_internal(exp, data);
351 OBD_ALLOC(mcd, sizeof(*mcd));
353 GOTO(out, rc = -ENOMEM);
355 memcpy(mcd->mcd_uuid, cluuid, sizeof(mcd->mcd_uuid));
358 rc = mds_client_add(obd, exp, -1, localdata);
364 OBD_FREE(mcd, sizeof(*mcd));
367 class_disconnect(exp);
369 class_export_put(exp);
375 int mds_init_export(struct obd_export *exp)
377 struct mds_export_data *med = &exp->exp_mds_data;
379 CFS_INIT_LIST_HEAD(&med->med_open_head);
380 spin_lock_init(&med->med_open_lock);
382 spin_lock(&exp->exp_lock);
383 exp->exp_connecting = 1;
384 spin_unlock(&exp->exp_lock);
389 static int mds_destroy_export(struct obd_export *export)
391 struct mds_export_data *med;
392 struct obd_device *obd = export->exp_obd;
393 struct mds_obd *mds = &obd->u.mds;
394 struct lvfs_run_ctxt saved;
395 struct lov_mds_md *lmm;
396 struct llog_cookie *logcookies;
400 med = &export->exp_mds_data;
401 target_destroy_export(export);
403 if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
406 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
407 /* Close any open files (which may also cause orphan unlinking). */
409 OBD_ALLOC(lmm, mds->mds_max_mdsize);
411 CWARN("%s: allocation failure during cleanup; can not force "
412 "close file handles on this service.\n", obd->obd_name);
413 GOTO(out, rc = -ENOMEM);
416 OBD_ALLOC(logcookies, mds->mds_max_cookiesize);
417 if (logcookies == NULL) {
418 CWARN("%s: allocation failure during cleanup; can not force "
419 "close file handles on this service.\n", obd->obd_name);
420 OBD_FREE(lmm, mds->mds_max_mdsize);
421 GOTO(out_lmm, rc = -ENOMEM);
424 spin_lock(&med->med_open_lock);
425 while (!list_empty(&med->med_open_head)) {
426 struct list_head *tmp = med->med_open_head.next;
427 struct mds_file_data *mfd =
428 list_entry(tmp, struct mds_file_data, mfd_list);
429 int lmm_size = mds->mds_max_mdsize;
430 umode_t mode = mfd->mfd_dentry->d_inode->i_mode;
433 /* Remove mfd handle so it can't be found again.
434 * We are consuming the mfd_list reference here. */
435 mds_mfd_unlink(mfd, 0);
436 spin_unlock(&med->med_open_lock);
438 /* If you change this message, be sure to update
439 * replay_single:test_46 */
440 CDEBUG(D_INODE|D_IOCTL, "%s: force closing file handle for "
441 "%.*s (ino %lu)\n", obd->obd_name,
442 mfd->mfd_dentry->d_name.len,mfd->mfd_dentry->d_name.name,
443 mfd->mfd_dentry->d_inode->i_ino);
445 rc = mds_get_md(obd, mfd->mfd_dentry->d_inode, lmm, &lmm_size, 1);
447 CWARN("mds_get_md failure, rc=%d\n", rc);
449 valid |= OBD_MD_FLEASIZE;
451 /* child orphan sem protects orphan_dec_test and
452 * is_orphan race, mds_mfd_close drops it */
453 MDS_DOWN_WRITE_ORPHAN_SEM(mfd->mfd_dentry->d_inode);
454 rc = mds_mfd_close(NULL, REQ_REC_OFF, obd, mfd,
455 !(export->exp_flags & OBD_OPT_FAILOVER),
456 lmm, lmm_size, logcookies,
457 mds->mds_max_cookiesize,
461 CDEBUG(D_INODE|D_IOCTL, "Error closing file: %d\n", rc);
463 if (valid & OBD_MD_FLCOOKIE) {
464 rc = mds_osc_destroy_orphan(obd, mode, lmm,
465 lmm_size, logcookies, 1);
467 CDEBUG(D_INODE, "%s: destroy of orphan failed,"
468 " rc = %d\n", obd->obd_name, rc);
471 valid &= ~OBD_MD_FLCOOKIE;
474 spin_lock(&med->med_open_lock);
476 spin_unlock(&med->med_open_lock);
478 OBD_FREE(logcookies, mds->mds_max_cookiesize);
480 OBD_FREE(lmm, mds->mds_max_mdsize);
482 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
483 mds_client_free(export);
487 static int mds_disconnect(struct obd_export *exp)
493 class_export_get(exp);
495 /* Disconnect early so that clients can't keep using export */
496 rc = class_disconnect(exp);
497 if (exp->exp_obd->obd_namespace != NULL)
498 ldlm_cancel_locks_for_export(exp);
500 /* complete all outstanding replies */
501 spin_lock(&exp->exp_lock);
502 while (!list_empty(&exp->exp_outstanding_replies)) {
503 struct ptlrpc_reply_state *rs =
504 list_entry(exp->exp_outstanding_replies.next,
505 struct ptlrpc_reply_state, rs_exp_list);
506 struct ptlrpc_service *svc = rs->rs_service;
508 spin_lock(&svc->srv_lock);
509 list_del_init(&rs->rs_exp_list);
510 ptlrpc_schedule_difficult_reply(rs);
511 spin_unlock(&svc->srv_lock);
513 spin_unlock(&exp->exp_lock);
515 class_export_put(exp);
519 static int mds_getstatus(struct ptlrpc_request *req)
521 struct mds_obd *mds = mds_req2mds(req);
522 struct mds_body *body;
523 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
526 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
527 RETURN(req->rq_status = -ENOMEM);
528 rc = lustre_pack_reply(req, 2, size, NULL);
530 RETURN(req->rq_status = rc);
532 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
533 memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
535 /* the last_committed and last_xid fields are filled in for all
536 * replies already - no need to do so here also.
541 /* get the LOV EA from @inode and store it into @md. It can be at most
542 * @size bytes, and @size is updated with the actual EA size.
543 * The EA size is also returned on success, and -ve errno on failure.
544 * If there is no EA then 0 is returned. */
545 int mds_get_md(struct obd_device *obd, struct inode *inode, void *md,
552 LOCK_INODE_MUTEX(inode);
553 rc = fsfilt_get_md(obd, inode, md, *size, "lov");
556 CERROR("Error %d reading eadata for ino %lu\n",
560 rc = mds_convert_lov_ea(obd, inode, md, lmm_size);
572 UNLOCK_INODE_MUTEX(inode);
578 /* Call with lock=1 if you want mds_pack_md to take the i_mutex.
579 * Call with lock=0 if the caller has already taken the i_mutex. */
580 int mds_pack_md(struct obd_device *obd, struct lustre_msg *msg, int offset,
581 struct mds_body *body, struct inode *inode, int lock)
583 struct mds_obd *mds = &obd->u.mds;
589 lmm = lustre_msg_buf(msg, offset, 0);
591 /* Some problem with getting eadata when I sized the reply
593 CDEBUG(D_INFO, "no space reserved for inode %lu MD\n",
597 lmm_size = lustre_msg_buflen(msg, offset);
599 /* I don't really like this, but it is a sanity check on the client
600 * MD request. However, if the client doesn't know how much space
601 * to reserve for the MD, it shouldn't be bad to have too much space.
603 if (lmm_size > mds->mds_max_mdsize) {
604 CWARN("Reading MD for inode %lu of %d bytes > max %d\n",
605 inode->i_ino, lmm_size, mds->mds_max_mdsize);
609 rc = mds_get_md(obd, inode, lmm, &lmm_size, lock);
611 if (S_ISDIR(inode->i_mode))
612 body->valid |= OBD_MD_FLDIREA;
614 body->valid |= OBD_MD_FLEASIZE;
615 body->eadatasize = lmm_size;
622 #ifdef CONFIG_FS_POSIX_ACL
624 int mds_pack_posix_acl(struct inode *inode, struct lustre_msg *repmsg,
625 struct mds_body *repbody, int repoff)
627 struct dentry de = { .d_inode = inode };
631 LASSERT(repbody->aclsize == 0);
632 LASSERT(lustre_msg_bufcount(repmsg) > repoff);
634 buflen = lustre_msg_buflen(repmsg, repoff);
638 if (!inode->i_op || !inode->i_op->getxattr)
641 rc = inode->i_op->getxattr(&de, MDS_XATTR_NAME_ACL_ACCESS,
642 lustre_msg_buf(repmsg, repoff, buflen),
646 repbody->aclsize = rc;
647 else if (rc != -ENODATA) {
648 CERROR("buflen %d, get acl: %d\n", buflen, rc);
653 repbody->valid |= OBD_MD_FLACL;
657 #define mds_pack_posix_acl(inode, repmsg, repbody, repoff) 0
660 int mds_pack_acl(struct mds_export_data *med, struct inode *inode,
661 struct lustre_msg *repmsg, struct mds_body *repbody,
664 return mds_pack_posix_acl(inode, repmsg, repbody, repoff);
667 static int mds_getattr_internal(struct obd_device *obd, struct dentry *dentry,
668 struct ptlrpc_request *req,
669 struct mds_body *reqbody, int reply_off)
671 struct mds_body *body;
672 struct inode *inode = dentry->d_inode;
679 body = lustre_msg_buf(req->rq_repmsg, reply_off, sizeof(*body));
680 LASSERT(body != NULL); /* caller prepped reply */
682 mds_pack_inode2fid(&body->fid1, inode);
683 body->flags = reqbody->flags; /* copy MDS_BFLAG_EXT_FLAGS if present */
684 mds_pack_inode2body(body, inode);
687 if ((S_ISREG(inode->i_mode) && (reqbody->valid & OBD_MD_FLEASIZE)) ||
688 (S_ISDIR(inode->i_mode) && (reqbody->valid & OBD_MD_FLDIREA))) {
689 rc = mds_pack_md(obd, req->rq_repmsg, reply_off, body,
692 /* If we have LOV EA data, the OST holds size, atime, mtime */
693 if (!(body->valid & OBD_MD_FLEASIZE) &&
694 !(body->valid & OBD_MD_FLDIREA))
695 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
696 OBD_MD_FLATIME | OBD_MD_FLMTIME);
698 lustre_shrink_reply(req, reply_off, body->eadatasize, 0);
699 if (body->eadatasize)
701 } else if (S_ISLNK(inode->i_mode) &&
702 (reqbody->valid & OBD_MD_LINKNAME) != 0) {
703 char *symname = lustre_msg_buf(req->rq_repmsg, reply_off, 0);
706 LASSERT (symname != NULL); /* caller prepped reply */
707 len = lustre_msg_buflen(req->rq_repmsg, reply_off);
709 rc = inode->i_op->readlink(dentry, symname, len);
711 CERROR("readlink failed: %d\n", rc);
712 } else if (rc != len - 1) {
713 CERROR ("Unexpected readlink rc %d: expecting %d\n",
717 CDEBUG(D_INODE, "read symlink dest %s\n", symname);
718 body->valid |= OBD_MD_LINKNAME;
719 body->eadatasize = rc + 1;
720 symname[rc] = 0; /* NULL terminate */
724 } else if (reqbody->valid == OBD_MD_FLFLAGS &&
725 reqbody->flags & MDS_BFLAG_EXT_FLAGS) {
728 /* We only return the full set of flags on ioctl, otherwise we
729 * get enough flags from the inode in mds_pack_inode2body(). */
730 rc = fsfilt_iocontrol(obd, inode, NULL, EXT3_IOC_GETFLAGS,
733 body->flags = flags | MDS_BFLAG_EXT_FLAGS;
736 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
737 struct mds_obd *mds = mds_req2mds(req);
738 body->max_cookiesize = mds->mds_max_cookiesize;
739 body->max_mdsize = mds->mds_max_mdsize;
740 body->valid |= OBD_MD_FLMODEASIZE;
746 #ifdef CONFIG_FS_POSIX_ACL
747 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
748 (reqbody->valid & OBD_MD_FLACL)) {
749 rc = mds_pack_acl(&req->rq_export->exp_mds_data,
750 inode, req->rq_repmsg,
753 lustre_shrink_reply(req, reply_off, body->aclsize, 0);
762 static int mds_getattr_pack_msg(struct ptlrpc_request *req, struct inode *inode,
765 struct mds_obd *mds = mds_req2mds(req);
766 struct mds_body *body;
767 int rc, bufcount = 2;
768 int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body) };
771 LASSERT(offset == REQ_REC_OFF); /* non-intent */
773 body = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*body));
774 LASSERT(body != NULL); /* checked by caller */
775 LASSERT(lustre_req_swabbed(req, offset)); /* swabbed by caller */
777 if ((S_ISREG(inode->i_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
778 (S_ISDIR(inode->i_mode) && (body->valid & OBD_MD_FLDIREA))) {
779 LOCK_INODE_MUTEX(inode);
780 rc = fsfilt_get_md(req->rq_export->exp_obd, inode, NULL, 0,
782 UNLOCK_INODE_MUTEX(inode);
783 CDEBUG(D_INODE, "got %d bytes MD data for inode %lu\n",
786 if (rc != -ENODATA) {
787 CERROR("error getting inode %lu MD: rc = %d\n",
792 } else if (rc > mds->mds_max_mdsize) {
794 CERROR("MD size %d larger than maximum possible %u\n",
795 rc, mds->mds_max_mdsize);
800 } else if (S_ISLNK(inode->i_mode) && (body->valid & OBD_MD_LINKNAME)) {
801 if (i_size_read(inode) + 1 != body->eadatasize)
802 CERROR("symlink size: %Lu, reply space: %d\n",
803 i_size_read(inode) + 1, body->eadatasize);
804 size[bufcount] = min_t(int, i_size_read(inode) + 1,
807 CDEBUG(D_INODE, "symlink size: %Lu, reply space: %d\n",
808 i_size_read(inode) + 1, body->eadatasize);
811 #ifdef CONFIG_FS_POSIX_ACL
812 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
813 (body->valid & OBD_MD_FLACL)) {
814 struct dentry de = { .d_inode = inode };
817 if (inode->i_op && inode->i_op->getxattr) {
818 rc = inode->i_op->getxattr(&de, MDS_XATTR_NAME_ACL_ACCESS,
822 if (rc != -ENODATA) {
823 CERROR("got acl size: %d\n", rc);
833 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) {
834 CERROR("failed MDS_GETATTR_PACK test\n");
835 req->rq_status = -ENOMEM;
839 rc = lustre_pack_reply(req, bufcount, size, NULL);
848 static int mds_getattr_lock(struct ptlrpc_request *req, int offset,
849 int child_part, struct lustre_handle *child_lockh)
851 struct obd_device *obd = req->rq_export->exp_obd;
852 struct mds_obd *mds = &obd->u.mds;
853 struct ldlm_reply *rep = NULL;
854 struct lvfs_run_ctxt saved;
855 struct mds_body *body;
856 struct dentry *dparent = NULL, *dchild = NULL;
857 struct lvfs_ucred uc = {0,};
858 struct lustre_handle parent_lockh;
860 int rc = 0, cleanup_phase = 0, resent_req = 0;
864 LASSERT(!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME));
866 /* Swab now, before anyone looks inside the request */
867 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
868 lustre_swab_mds_body);
870 CERROR("Can't swab mds_body\n");
874 lustre_set_req_swabbed(req, offset + 1);
875 name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
877 CERROR("Can't unpack name\n");
880 namesize = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
881 /* namesize less than 2 means we have empty name, probably came from
882 revalidate by cfid, so no point in having name to be set */
886 rc = mds_init_ucred(&uc, req, offset);
890 LASSERT(offset == REQ_REC_OFF || offset == DLM_INTENT_REC_OFF);
891 /* if requests were at offset 2, the getattr reply goes back at 1 */
892 if (offset == DLM_INTENT_REC_OFF) {
893 rep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF,
895 offset = DLM_REPLY_REC_OFF;
898 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
899 cleanup_phase = 1; /* kernel context */
900 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
902 /* FIXME: handle raw lookup */
904 if (body->valid == OBD_MD_FLID) {
905 struct mds_body *mds_reply;
906 int size = sizeof(*mds_reply);
908 // The user requested ONLY the inode number, so do a raw lookup
909 rc = lustre_pack_reply(req, 1, &size, NULL);
911 CERROR("out of memory\n");
915 rc = dir->i_op->lookup_raw(dir, name, namesize - 1, &inum);
917 mds_reply = lustre_msg_buf(req->rq_repmsg, offset,
919 mds_reply->fid1.id = inum;
920 mds_reply->valid = OBD_MD_FLID;
925 if (lustre_handle_is_used(child_lockh)) {
926 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
930 if (resent_req == 0) {
932 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
933 rc = mds_get_parent_child_locked(obd, &obd->u.mds,
937 MDS_INODELOCK_UPDATE,
939 child_lockh, &dchild,
942 /* For revalidate by fid we always take UPDATE lock */
943 dchild = mds_fid2locked_dentry(obd, &body->fid2, NULL,
948 rc = PTR_ERR(dchild);
953 struct ldlm_lock *granted_lock;
954 struct ll_fid child_fid;
955 struct ldlm_resource *res;
956 DEBUG_REQ(D_DLMTRACE, req, "resent, not enqueuing new locks");
957 granted_lock = ldlm_handle2lock(child_lockh);
958 LASSERTF(granted_lock != NULL, LPU64"/%u lockh "LPX64"\n",
959 body->fid1.id, body->fid1.generation,
960 child_lockh->cookie);
963 res = granted_lock->l_resource;
964 child_fid.id = res->lr_name.name[0];
965 child_fid.generation = res->lr_name.name[1];
966 dchild = mds_fid2dentry(&obd->u.mds, &child_fid, NULL);
967 LASSERT(!IS_ERR(dchild));
968 LDLM_LOCK_PUT(granted_lock);
971 cleanup_phase = 2; /* dchild, dparent, locks */
973 if (dchild->d_inode == NULL) {
974 intent_set_disposition(rep, DISP_LOOKUP_NEG);
975 /* in the intent case, the policy clears this error:
976 the disposition is enough */
977 GOTO(cleanup, rc = -ENOENT);
979 intent_set_disposition(rep, DISP_LOOKUP_POS);
982 if (req->rq_repmsg == NULL) {
983 rc = mds_getattr_pack_msg(req, dchild->d_inode, offset);
985 CERROR ("mds_getattr_pack_msg: %d\n", rc);
990 rc = mds_getattr_internal(obd, dchild, req, body, offset);
991 GOTO(cleanup, rc); /* returns the lock to the client */
994 switch (cleanup_phase) {
996 if (resent_req == 0) {
997 if (rc && dchild->d_inode)
998 ldlm_lock_decref(child_lockh, LCK_CR);
1000 ldlm_lock_decref(&parent_lockh, LCK_CR);
1006 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1008 mds_exit_ucred(&uc, mds);
1009 if (req->rq_reply_state == NULL) {
1010 int rc2 = lustre_pack_reply(req, 1, NULL, NULL);
1013 req->rq_status = rc;
1019 static int mds_getattr(struct ptlrpc_request *req, int offset)
1021 struct mds_obd *mds = mds_req2mds(req);
1022 struct obd_device *obd = req->rq_export->exp_obd;
1023 struct lvfs_run_ctxt saved;
1025 struct mds_body *body;
1026 struct lvfs_ucred uc = {0,};
1030 OBD_COUNTER_INCREMENT(obd, getattr);
1032 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1033 lustre_swab_mds_body);
1037 rc = mds_init_ucred(&uc, req, offset);
1039 GOTO(out_ucred, rc);
1041 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1042 de = mds_fid2dentry(mds, &body->fid1, NULL);
1044 rc = req->rq_status = PTR_ERR(de);
1048 rc = mds_getattr_pack_msg(req, de->d_inode, offset);
1050 CERROR("mds_getattr_pack_msg: %d\n", rc);
1054 req->rq_status = mds_getattr_internal(obd, de, req, body,
1060 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1062 if (req->rq_reply_state == NULL) {
1063 int rc2 = lustre_pack_reply(req, 1, NULL, NULL);
1066 req->rq_status = rc;
1068 mds_exit_ucred(&uc, mds);
1072 static int mds_obd_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1073 __u64 max_age, __u32 flags)
1077 spin_lock(&obd->obd_osfs_lock);
1078 rc = fsfilt_statfs(obd, obd->u.obt.obt_sb, max_age);
1080 memcpy(osfs, &obd->obd_osfs, sizeof(*osfs));
1081 spin_unlock(&obd->obd_osfs_lock);
1086 static int mds_statfs(struct ptlrpc_request *req)
1088 struct obd_device *obd = req->rq_export->exp_obd;
1089 int rc, size[2] = { sizeof(struct ptlrpc_body),
1090 sizeof(struct obd_statfs) };
1093 /* This will trigger a watchdog timeout */
1094 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
1095 (MDS_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
1096 OBD_COUNTER_INCREMENT(obd, statfs);
1098 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
1099 GOTO(out, rc = -ENOMEM);
1100 rc = lustre_pack_reply(req, 2, size, NULL);
1104 /* We call this so that we can cache a bit - 1 jiffie worth */
1105 rc = mds_obd_statfs(obd, lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1106 size[REPLY_REC_OFF]),
1107 cfs_time_current_64() - HZ, 0);
1109 CERROR("mds_obd_statfs failed: rc %d\n", rc);
1115 req->rq_status = rc;
1119 static int mds_sync(struct ptlrpc_request *req, int offset)
1121 struct obd_device *obd = req->rq_export->exp_obd;
1122 struct mds_obd *mds = &obd->u.mds;
1123 struct mds_body *body;
1124 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1127 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1128 lustre_swab_mds_body);
1130 GOTO(out, rc = -EFAULT);
1132 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1133 GOTO(out, rc = -ENOMEM);
1134 rc = lustre_pack_reply(req, 2, size, NULL);
1138 rc = fsfilt_sync(obd, obd->u.obt.obt_sb);
1139 if (rc == 0 && body->fid1.id != 0) {
1142 de = mds_fid2dentry(mds, &body->fid1, NULL);
1144 GOTO(out, rc = PTR_ERR(de));
1146 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1148 mds_pack_inode2fid(&body->fid1, de->d_inode);
1149 mds_pack_inode2body(body, de->d_inode);
1155 req->rq_status = rc;
1159 /* mds_readpage does not take a DLM lock on the inode, because the client must
1160 * already have a PR lock.
1162 * If we were to take another one here, a deadlock will result, if another
1163 * thread is already waiting for a PW lock. */
1164 static int mds_readpage(struct ptlrpc_request *req, int offset)
1166 struct obd_device *obd = req->rq_export->exp_obd;
1167 struct mds_obd *mds = &obd->u.mds;
1168 struct vfsmount *mnt;
1171 struct mds_body *body, *repbody;
1172 struct lvfs_run_ctxt saved;
1173 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
1174 struct lvfs_ucred uc = {0,};
1177 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1179 rc = lustre_pack_reply(req, 2, size, NULL);
1183 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1184 lustre_swab_mds_body);
1186 GOTO (out, rc = -EFAULT);
1188 rc = mds_init_ucred(&uc, req, offset);
1192 push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1193 de = mds_fid2dentry(&obd->u.mds, &body->fid1, &mnt);
1195 GOTO(out_pop, rc = PTR_ERR(de));
1197 CDEBUG(D_INODE, "ino %lu\n", de->d_inode->i_ino);
1199 file = dentry_open(de, mnt, O_RDONLY | O_LARGEFILE);
1200 /* note: in case of an error, dentry_open puts dentry */
1202 GOTO(out_pop, rc = PTR_ERR(file));
1204 /* body->size is actually the offset -eeb */
1205 if ((body->size & (de->d_inode->i_sb->s_blocksize - 1)) != 0) {
1206 CERROR("offset "LPU64" not on a block boundary of %lu\n",
1207 body->size, de->d_inode->i_sb->s_blocksize);
1208 GOTO(out_file, rc = -EFAULT);
1211 /* body->nlink is actually the #bytes to read -eeb */
1212 if (body->nlink & (de->d_inode->i_sb->s_blocksize - 1)) {
1213 CERROR("size %u is not multiple of blocksize %lu\n",
1214 body->nlink, de->d_inode->i_sb->s_blocksize);
1215 GOTO(out_file, rc = -EFAULT);
1218 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1220 repbody->size = i_size_read(file->f_dentry->d_inode);
1221 repbody->valid = OBD_MD_FLSIZE;
1223 /* to make this asynchronous make sure that the handling function
1224 doesn't send a reply when this function completes. Instead a
1225 callback function would send the reply */
1226 /* body->size is actually the offset -eeb */
1227 rc = mds_sendpage(req, file, body->size, body->nlink);
1230 filp_close(file, 0);
1232 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
1234 mds_exit_ucred(&uc, mds);
1235 req->rq_status = rc;
1239 int mds_reint(struct ptlrpc_request *req, int offset,
1240 struct lustre_handle *lockh)
1242 struct mds_update_record *rec; /* 116 bytes on the stack? no sir! */
1245 OBD_ALLOC(rec, sizeof(*rec));
1249 rc = mds_update_unpack(req, offset, rec);
1250 if (rc || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) {
1251 CERROR("invalid record\n");
1252 GOTO(out, req->rq_status = -EINVAL);
1255 /* rc will be used to interrupt a for loop over multiple records */
1256 rc = mds_reint_rec(rec, offset, req, lockh);
1258 OBD_FREE(rec, sizeof(*rec));
1262 int mds_filter_recovery_request(struct ptlrpc_request *req,
1263 struct obd_device *obd, int *process)
1265 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1266 case MDS_CONNECT: /* This will never get here, but for completeness. */
1267 case OST_CONNECT: /* This will never get here, but for completeness. */
1268 case MDS_DISCONNECT:
1269 case OST_DISCONNECT:
1274 case MDS_DONE_WRITING:
1275 case MDS_SYNC: /* used in unmounting */
1281 *process = target_queue_recovery_request(req, obd);
1285 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1290 EXPORT_SYMBOL(mds_filter_recovery_request);
1292 static char *reint_names[] = {
1293 [REINT_SETATTR] "setattr",
1294 [REINT_CREATE] "create",
1295 [REINT_LINK] "link",
1296 [REINT_UNLINK] "unlink",
1297 [REINT_RENAME] "rename",
1298 [REINT_OPEN] "open",
1301 static int mds_set_info_rpc(struct obd_export *exp, struct ptlrpc_request *req)
1304 int keylen, vallen, rc = 0;
1307 key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1309 DEBUG_REQ(D_HA, req, "no set_info key");
1312 keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1314 val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0);
1315 vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1317 rc = lustre_pack_reply(req, 1, NULL, NULL);
1321 lustre_msg_set_status(req->rq_repmsg, 0);
1323 if (KEY_IS("read-only")) {
1324 if (val == NULL || vallen < sizeof(__u32)) {
1325 DEBUG_REQ(D_HA, req, "no set_info val");
1330 exp->exp_connect_flags |= OBD_CONNECT_RDONLY;
1332 exp->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1340 static int mds_handle_quotacheck(struct ptlrpc_request *req)
1342 struct obd_quotactl *oqctl;
1346 oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1347 lustre_swab_obd_quotactl);
1351 rc = lustre_pack_reply(req, 1, NULL, NULL);
1355 req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1359 static int mds_handle_quotactl(struct ptlrpc_request *req)
1361 struct obd_quotactl *oqctl, *repoqc;
1362 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
1365 oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
1366 lustre_swab_obd_quotactl);
1370 rc = lustre_pack_reply(req, 2, size, NULL);
1374 repoqc = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repoqc));
1376 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1381 int mds_msg_check_version(struct lustre_msg *msg)
1385 switch (lustre_msg_get_opc(msg)) {
1387 case MDS_DISCONNECT:
1390 case SEC_CTX_INIT_CONT:
1392 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1394 CERROR("bad opc %u version %08x, expecting %08x\n",
1395 lustre_msg_get_opc(msg),
1396 lustre_msg_get_version(msg),
1397 LUSTRE_OBD_VERSION);
1401 case MDS_GETATTR_NAME:
1408 case MDS_DONE_WRITING:
1414 case MDS_QUOTACHECK:
1420 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
1422 CERROR("bad opc %u version %08x, expecting %08x\n",
1423 lustre_msg_get_opc(msg),
1424 lustre_msg_get_version(msg),
1425 LUSTRE_MDS_VERSION);
1429 case LDLM_BL_CALLBACK:
1430 case LDLM_CP_CALLBACK:
1431 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1433 CERROR("bad opc %u version %08x, expecting %08x\n",
1434 lustre_msg_get_opc(msg),
1435 lustre_msg_get_version(msg),
1436 LUSTRE_DLM_VERSION);
1438 case OBD_LOG_CANCEL:
1439 case LLOG_ORIGIN_HANDLE_CREATE:
1440 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
1441 case LLOG_ORIGIN_HANDLE_READ_HEADER:
1442 case LLOG_ORIGIN_HANDLE_CLOSE:
1443 case LLOG_ORIGIN_HANDLE_DESTROY:
1444 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
1446 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1448 CERROR("bad opc %u version %08x, expecting %08x\n",
1449 lustre_msg_get_opc(msg),
1450 lustre_msg_get_version(msg),
1451 LUSTRE_LOG_VERSION);
1454 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
1459 EXPORT_SYMBOL(mds_msg_check_version);
1461 int mds_handle(struct ptlrpc_request *req)
1463 int should_process, fail = OBD_FAIL_MDS_ALL_REPLY_NET;
1465 struct mds_obd *mds = NULL; /* quell gcc overwarning */
1466 struct obd_device *obd = NULL;
1469 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
1472 LASSERT(current->journal_info == NULL);
1474 rc = mds_msg_check_version(req->rq_reqmsg);
1476 CERROR("MDS drop mal-formed request\n");
1480 /* XXX identical to OST */
1481 if (lustre_msg_get_opc(req->rq_reqmsg) != MDS_CONNECT) {
1482 struct mds_export_data *med;
1485 if (req->rq_export == NULL) {
1486 CERROR("operation %d on unconnected MDS from %s\n",
1487 lustre_msg_get_opc(req->rq_reqmsg),
1488 libcfs_id2str(req->rq_peer));
1489 req->rq_status = -ENOTCONN;
1490 GOTO(out, rc = -ENOTCONN);
1493 med = &req->rq_export->exp_mds_data;
1494 obd = req->rq_export->exp_obd;
1495 mds = mds_req2mds(req);
1497 /* sanity check: if the xid matches, the request must
1498 * be marked as a resent or replayed */
1499 if (req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_xid) ||
1500 req->rq_xid == le64_to_cpu(med->med_mcd->mcd_last_close_xid))
1501 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
1502 (MSG_RESENT | MSG_REPLAY))) {
1503 CERROR("rq_xid "LPU64" matches last_xid, "
1504 "expected RESENT flag\n",
1506 req->rq_status = -ENOTCONN;
1507 GOTO(out, rc = -EFAULT);
1509 /* else: note the opposite is not always true; a
1510 * RESENT req after a failover will usually not match
1511 * the last_xid, since it was likely never
1512 * committed. A REPLAYed request will almost never
1513 * match the last xid, however it could for a
1514 * committed, but still retained, open. */
1516 /* Check for aborted recovery. */
1517 spin_lock_bh(&obd->obd_processing_task_lock);
1518 recovering = obd->obd_recovering;
1519 spin_unlock_bh(&obd->obd_processing_task_lock);
1521 rc = mds_filter_recovery_request(req, obd,
1523 if (rc || !should_process)
1525 else if (should_process < 0) {
1526 req->rq_status = should_process;
1527 rc = ptlrpc_error(req);
1533 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1535 DEBUG_REQ(D_INODE, req, "connect");
1536 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CONNECT_NET))
1538 rc = target_handle_connect(req);
1540 /* Now that we have an export, set mds. */
1542 * XXX nikita: these assignments are useless: mds is
1543 * never used below, and obd is only used for
1544 * MSG_LAST_REPLAY case, which never happens for
1547 obd = req->rq_export->exp_obd;
1548 mds = mds_req2mds(req);
1552 case MDS_DISCONNECT:
1553 DEBUG_REQ(D_INODE, req, "disconnect");
1554 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DISCONNECT_NET))
1556 rc = target_handle_disconnect(req);
1557 req->rq_status = rc; /* superfluous? */
1561 DEBUG_REQ(D_INODE, req, "getstatus");
1562 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_NET))
1564 rc = mds_getstatus(req);
1568 DEBUG_REQ(D_INODE, req, "getattr");
1569 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_NET))
1571 rc = mds_getattr(req, REQ_REC_OFF);
1575 DEBUG_REQ(D_INODE, req, "setxattr");
1576 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SETXATTR_NET))
1578 rc = mds_setxattr(req);
1582 DEBUG_REQ(D_INODE, req, "getxattr");
1583 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETXATTR_NET))
1585 rc = mds_getxattr(req);
1588 case MDS_GETATTR_NAME: {
1589 struct lustre_handle lockh = { 0 };
1590 DEBUG_REQ(D_INODE, req, "getattr_name");
1591 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_NAME_NET))
1594 /* If this request gets a reconstructed reply, we won't be
1595 * acquiring any new locks in mds_getattr_lock, so we don't
1598 rc = mds_getattr_lock(req, REQ_REC_OFF, MDS_INODELOCK_UPDATE,
1600 /* this non-intent call (from an ioctl) is special */
1601 req->rq_status = rc;
1602 if (rc == 0 && lustre_handle_is_used(&lockh))
1603 ldlm_lock_decref(&lockh, LCK_CR);
1607 DEBUG_REQ(D_INODE, req, "statfs");
1608 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_NET))
1610 rc = mds_statfs(req);
1614 DEBUG_REQ(D_INODE, req, "readpage");
1615 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_NET))
1617 rc = mds_readpage(req, REQ_REC_OFF);
1619 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) {
1626 __u32 *opcp = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
1630 int size[4] = { sizeof(struct ptlrpc_body),
1631 sizeof(struct mds_body),
1632 mds->mds_max_mdsize,
1633 mds->mds_max_cookiesize };
1636 /* NB only peek inside req now; mds_reint() will swab it */
1638 CERROR ("Can't inspect opcode\n");
1643 if (lustre_msg_swabbed(req->rq_reqmsg))
1646 DEBUG_REQ(D_INODE, req, "reint %d (%s)", opc,
1647 (opc < sizeof(reint_names) / sizeof(reint_names[0]) ||
1648 reint_names[opc] == NULL) ? reint_names[opc] :
1652 op = PTLRPC_LAST_CNTR + MDS_REINT_CREATE;
1655 op = PTLRPC_LAST_CNTR + MDS_REINT_LINK;
1658 op = PTLRPC_LAST_CNTR + MDS_REINT_OPEN;
1661 op = PTLRPC_LAST_CNTR + MDS_REINT_SETATTR;
1664 op = PTLRPC_LAST_CNTR + MDS_REINT_RENAME;
1667 op = PTLRPC_LAST_CNTR + MDS_REINT_UNLINK;
1674 if (op && req->rq_rqbd->rqbd_service->srv_stats)
1675 lprocfs_counter_incr(
1676 req->rq_rqbd->rqbd_service->srv_stats, op);
1678 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_NET))
1681 if (opc == REINT_UNLINK || opc == REINT_RENAME)
1683 else if (opc == REINT_OPEN)
1688 rc = lustre_pack_reply(req, bufcount, size, NULL);
1692 rc = mds_reint(req, REQ_REC_OFF, NULL);
1693 fail = OBD_FAIL_MDS_REINT_NET_REP;
1698 DEBUG_REQ(D_INODE, req, "close");
1699 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_NET))
1701 rc = mds_close(req, REQ_REC_OFF);
1702 fail = OBD_FAIL_MDS_CLOSE_NET_REP;
1705 case MDS_DONE_WRITING:
1706 DEBUG_REQ(D_INODE, req, "done_writing");
1707 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DONE_WRITING_NET))
1709 rc = mds_done_writing(req, REQ_REC_OFF);
1713 DEBUG_REQ(D_INODE, req, "pin");
1714 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_PIN_NET))
1716 rc = mds_pin(req, REQ_REC_OFF);
1720 DEBUG_REQ(D_INODE, req, "sync");
1721 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_NET))
1723 rc = mds_sync(req, REQ_REC_OFF);
1727 DEBUG_REQ(D_INODE, req, "set_info");
1728 rc = mds_set_info_rpc(req->rq_export, req);
1731 case MDS_QUOTACHECK:
1732 DEBUG_REQ(D_INODE, req, "quotacheck");
1733 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET))
1735 rc = mds_handle_quotacheck(req);
1739 DEBUG_REQ(D_INODE, req, "quotactl");
1740 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET))
1742 rc = mds_handle_quotactl(req);
1746 DEBUG_REQ(D_INODE, req, "ping");
1747 rc = target_handle_ping(req);
1750 case OBD_LOG_CANCEL:
1751 CDEBUG(D_INODE, "log cancel\n");
1752 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
1754 rc = -ENOTSUPP; /* la la la */
1758 DEBUG_REQ(D_INODE, req, "enqueue");
1759 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
1761 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1762 ldlm_server_blocking_ast, NULL);
1763 fail = OBD_FAIL_LDLM_REPLY;
1766 DEBUG_REQ(D_INODE, req, "convert");
1767 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
1769 rc = ldlm_handle_convert(req);
1771 case LDLM_BL_CALLBACK:
1772 case LDLM_CP_CALLBACK:
1773 DEBUG_REQ(D_INODE, req, "callback");
1774 CERROR("callbacks should not happen on MDS\n");
1776 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK))
1779 case LLOG_ORIGIN_HANDLE_CREATE:
1780 DEBUG_REQ(D_INODE, req, "llog_init");
1781 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1783 rc = llog_origin_handle_create(req);
1785 case LLOG_ORIGIN_HANDLE_DESTROY:
1786 DEBUG_REQ(D_INODE, req, "llog_init");
1787 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1789 rc = llog_origin_handle_destroy(req);
1791 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
1792 DEBUG_REQ(D_INODE, req, "llog next block");
1793 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1795 rc = llog_origin_handle_next_block(req);
1797 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
1798 DEBUG_REQ(D_INODE, req, "llog prev block");
1799 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1801 rc = llog_origin_handle_prev_block(req);
1803 case LLOG_ORIGIN_HANDLE_READ_HEADER:
1804 DEBUG_REQ(D_INODE, req, "llog read header");
1805 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1807 rc = llog_origin_handle_read_header(req);
1809 case LLOG_ORIGIN_HANDLE_CLOSE:
1810 DEBUG_REQ(D_INODE, req, "llog close");
1811 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1813 rc = llog_origin_handle_close(req);
1816 DEBUG_REQ(D_INODE, req, "llog catinfo");
1817 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET))
1819 rc = llog_catinfo(req);
1822 req->rq_status = -ENOTSUPP;
1823 rc = ptlrpc_error(req);
1827 LASSERT(current->journal_info == NULL);
1829 /* If we're DISCONNECTing, the mds_export_data is already freed */
1830 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != MDS_DISCONNECT) {
1831 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1833 /* I don't think last_xid is used for anyway, so I'm not sure
1834 if we need to care about last_close_xid here.*/
1835 lustre_msg_set_last_xid(req->rq_repmsg,
1836 le64_to_cpu(med->med_mcd->mcd_last_xid));
1838 target_committed_to_req(req);
1844 target_send_reply(req, rc, fail);
1848 /* Update the server data on disk. This stores the new mount_count and
1849 * also the last_rcvd value to disk. If we don't have a clean shutdown,
1850 * then the server last_rcvd value may be less than that of the clients.
1851 * This will alert us that we may need to do client recovery.
1853 * Also assumes for mds_last_transno that we are not modifying it (no locking).
1855 int mds_update_server_data(struct obd_device *obd, int force_sync)
1857 struct mds_obd *mds = &obd->u.mds;
1858 struct lr_server_data *lsd = mds->mds_server_data;
1859 struct file *filp = mds->mds_rcvd_filp;
1860 struct lvfs_run_ctxt saved;
1865 CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
1866 mds->mds_mount_count, mds->mds_last_transno);
1868 spin_lock(&mds->mds_transno_lock);
1869 lsd->lsd_last_transno = cpu_to_le64(mds->mds_last_transno);
1870 spin_unlock(&mds->mds_transno_lock);
1872 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1873 rc = fsfilt_write_record(obd, filp, lsd, sizeof(*lsd), &off,force_sync);
1874 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1876 CERROR("error writing MDS server data: rc = %d\n", rc);
1880 static void fsoptions_to_mds_flags(struct mds_obd *mds, char *options)
1890 while (*p && *p != ',')
1894 if (len == sizeof("user_xattr") - 1 &&
1895 memcmp(options, "user_xattr", len) == 0) {
1896 mds->mds_fl_user_xattr = 1;
1897 LCONSOLE_INFO("Enabling user_xattr\n");
1898 } else if (len == sizeof("nouser_xattr") - 1 &&
1899 memcmp(options, "nouser_xattr", len) == 0) {
1900 mds->mds_fl_user_xattr = 0;
1901 LCONSOLE_INFO("Disabling user_xattr\n");
1902 } else if (len == sizeof("acl") - 1 &&
1903 memcmp(options, "acl", len) == 0) {
1904 #ifdef CONFIG_FS_POSIX_ACL
1905 mds->mds_fl_acl = 1;
1906 LCONSOLE_INFO("Enabling ACL\n");
1908 CWARN("ignoring unsupported acl mount option\n");
1910 } else if (len == sizeof("noacl") - 1 &&
1911 memcmp(options, "noacl", len) == 0) {
1912 #ifdef CONFIG_FS_POSIX_ACL
1913 mds->mds_fl_acl = 0;
1914 LCONSOLE_INFO("Disabling ACL\n");
1921 static int mds_lov_presetup (struct mds_obd *mds, struct lustre_cfg *lcfg)
1926 if (lcfg->lcfg_bufcount >= 4 && LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
1929 ll_generate_random_uuid(uuid);
1930 class_uuid_unparse(uuid, &mds->mds_lov_uuid);
1932 OBD_ALLOC(mds->mds_profile, LUSTRE_CFG_BUFLEN(lcfg, 3));
1933 if (mds->mds_profile == NULL)
1936 strncpy(mds->mds_profile, lustre_cfg_string(lcfg, 3),
1937 LUSTRE_CFG_BUFLEN(lcfg, 3));
1942 /* mount the file system (secretly). lustre_cfg parameters are:
1948 static int mds_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
1950 struct lprocfs_static_vars lvars;
1951 struct mds_obd *mds = &obd->u.mds;
1952 struct lustre_mount_info *lmi;
1953 struct vfsmount *mnt;
1954 struct lustre_sb_info *lsi;
1955 struct obd_uuid uuid;
1962 /* setup 1:/dev/loop/0 2:ext3 3:mdsA 4:errors=remount-ro,iopen_nopriv */
1964 CLASSERT(offsetof(struct obd_device, u.obt) ==
1965 offsetof(struct obd_device, u.mds.mds_obt));
1967 if (lcfg->lcfg_bufcount < 3)
1970 if (LUSTRE_CFG_BUFLEN(lcfg, 1) == 0 || LUSTRE_CFG_BUFLEN(lcfg, 2) == 0)
1973 lmi = server_get_mount(obd->obd_name);
1975 CERROR("Not mounted in lustre_fill_super?\n");
1979 /* We mounted in lustre_fill_super.
1980 lcfg bufs 1, 2, 4 (device, fstype, mount opts) are ignored.*/
1982 lsi = s2lsi(lmi->lmi_sb);
1983 fsoptions_to_mds_flags(mds, lsi->lsi_ldd->ldd_mount_opts);
1984 fsoptions_to_mds_flags(mds, lsi->lsi_lmd->lmd_opts);
1986 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
1987 if (IS_ERR(obd->obd_fsops))
1988 GOTO(err_put, rc = PTR_ERR(obd->obd_fsops));
1990 CDEBUG(D_SUPER, "%s: mnt = %p\n", lustre_cfg_string(lcfg, 1), mnt);
1992 LASSERT(!lvfs_check_rdonly(lvfs_sbdev(mnt->mnt_sb)));
1994 sema_init(&mds->mds_epoch_sem, 1);
1995 spin_lock_init(&mds->mds_transno_lock);
1996 mds->mds_max_mdsize = sizeof(struct lov_mds_md);
1997 mds->mds_max_cookiesize = sizeof(struct llog_cookie);
1998 mds->mds_atime_diff = MAX_ATIME_DIFF;
1999 mds->mds_evict_ost_nids = 1;
2001 sprintf(ns_name, "mds-%s", obd->obd_uuid.uuid);
2002 obd->obd_namespace = ldlm_namespace_new(ns_name, LDLM_NAMESPACE_SERVER,
2003 LDLM_NAMESPACE_GREEDY);
2004 if (obd->obd_namespace == NULL) {
2006 GOTO(err_ops, rc = -ENOMEM);
2008 ldlm_register_intent(obd->obd_namespace, mds_intent_policy);
2010 lprocfs_mds_init_vars(&lvars);
2011 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
2012 lprocfs_alloc_obd_stats(obd, LPROC_MDS_LAST) == 0) {
2013 /* Init private stats here */
2014 mds_stats_counter_init(obd->obd_stats);
2015 obd->obd_proc_exports_entry = lprocfs_register("exports",
2016 obd->obd_proc_entry,
2018 if (IS_ERR(obd->obd_proc_exports_entry)) {
2019 rc = PTR_ERR(obd->obd_proc_exports_entry);
2020 CERROR("error %d setting up lprocfs for %s\n",
2022 obd->obd_proc_exports_entry = NULL;
2026 rc = mds_fs_setup(obd, mnt);
2028 CERROR("%s: MDS filesystem method init failed: rc = %d\n",
2033 if (obd->obd_proc_exports_entry)
2034 lprocfs_add_simple(obd->obd_proc_exports_entry,
2035 "clear", lprocfs_nid_stats_clear_read,
2036 lprocfs_nid_stats_clear_write, obd);
2038 rc = mds_lov_presetup(mds, lcfg);
2042 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
2043 "mds_ldlm_client", &obd->obd_ldlm_client);
2044 obd->obd_replayable = 1;
2046 rc = lquota_setup(mds_quota_interface_ref, obd);
2051 mds->mds_group_hash = upcall_cache_init(obd->obd_name);
2052 if (IS_ERR(mds->mds_group_hash)) {
2053 rc = PTR_ERR(mds->mds_group_hash);
2054 mds->mds_group_hash = NULL;
2055 GOTO(err_qctxt, rc);
2059 /* Don't wait for mds_postrecov trying to clear orphans */
2060 obd->obd_async_recov = 1;
2061 rc = mds_postsetup(obd);
2062 /* Bug 11557 - allow async abort_recov start
2063 FIXME can remove most of this obd_async_recov plumbing
2064 obd->obd_async_recov = 0;
2067 GOTO(err_qctxt, rc);
2069 uuid_ptr = fsfilt_uuid(obd, obd->u.obt.obt_sb);
2070 if (uuid_ptr != NULL) {
2071 class_uuid_unparse(uuid_ptr, &uuid);
2077 label = fsfilt_get_label(obd, obd->u.obt.obt_sb);
2078 if (obd->obd_recovering) {
2079 LCONSOLE_WARN("MDT %s now serving %s (%s%s%s), but will be in "
2080 "recovery until %d %s reconnect, or if no clients"
2081 " reconnect for %d:%.02d; during that time new "
2082 "clients will not be allowed to connect. "
2083 "Recovery progress can be monitored by watching "
2084 "/proc/fs/lustre/mds/%s/recovery_status.\n",
2085 obd->obd_name, lustre_cfg_string(lcfg, 1),
2086 label ?: "", label ? "/" : "", str,
2087 obd->obd_max_recoverable_clients,
2088 (obd->obd_max_recoverable_clients == 1) ?
2089 "client" : "clients",
2090 (int)(OBD_RECOVERY_TIMEOUT) / 60,
2091 (int)(OBD_RECOVERY_TIMEOUT) % 60,
2094 LCONSOLE_INFO("MDT %s now serving %s (%s%s%s) with recovery "
2095 "%s\n", obd->obd_name, lustre_cfg_string(lcfg, 1),
2096 label ?: "", label ? "/" : "", str,
2097 obd->obd_replayable ? "enabled" : "disabled");
2100 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
2106 lquota_cleanup(mds_quota_interface_ref, obd);
2108 /* No extra cleanup needed for llog_init_commit_thread() */
2109 mds_fs_cleanup(obd);
2111 upcall_cache_cleanup(mds->mds_group_hash);
2112 mds->mds_group_hash = NULL;
2115 lprocfs_free_obd_stats(obd);
2116 lprocfs_obd_cleanup(obd);
2117 ldlm_namespace_free(obd->obd_namespace, 0);
2118 obd->obd_namespace = NULL;
2120 fsfilt_put_ops(obd->obd_fsops);
2122 server_put_mount(obd->obd_name, mnt);
2123 obd->u.obt.obt_sb = NULL;
2127 static int mds_lov_clean(struct obd_device *obd)
2129 struct mds_obd *mds = &obd->u.mds;
2130 struct obd_device *osc = mds->mds_osc_obd;
2133 if (mds->mds_profile) {
2134 class_del_profile(mds->mds_profile);
2135 OBD_FREE(mds->mds_profile, strlen(mds->mds_profile) + 1);
2136 mds->mds_profile = NULL;
2139 /* There better be a lov */
2143 RETURN(PTR_ERR(osc));
2145 obd_register_observer(osc, NULL);
2147 /* Give lov our same shutdown flags */
2148 osc->obd_force = obd->obd_force;
2149 osc->obd_fail = obd->obd_fail;
2151 /* Cleanup the lov */
2152 obd_disconnect(mds->mds_osc_exp);
2153 class_manual_cleanup(osc);
2154 mds->mds_osc_exp = NULL;
2159 static int mds_postsetup(struct obd_device *obd)
2161 struct mds_obd *mds = &obd->u.mds;
2165 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd, 0, NULL,
2170 rc = llog_setup(obd, &obd->obd_olg, LLOG_LOVEA_ORIG_CTXT, obd, 0, NULL,
2175 if (mds->mds_profile) {
2176 struct lustre_profile *lprof;
2177 /* The profile defines which osc and mdc to connect to, for a
2178 client. We reuse that here to figure out the name of the
2179 lov to use (and ignore lprof->lp_md).
2180 The profile was set in the config log with
2181 LCFG_MOUNTOPT profilenm oscnm mdcnm */
2182 lprof = class_get_profile(mds->mds_profile);
2183 if (lprof == NULL) {
2184 CERROR("No profile found: %s\n", mds->mds_profile);
2185 GOTO(err_cleanup, rc = -ENOENT);
2187 rc = mds_lov_connect(obd, lprof->lp_dt);
2189 GOTO(err_cleanup, rc);
2196 llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
2197 llog_cleanup(llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT));
2201 int mds_postrecov(struct obd_device *obd)
2209 LASSERT(!obd->obd_recovering);
2210 LASSERT(!llog_ctxt_null(obd, LLOG_MDS_OST_ORIG_CTXT));
2212 /* clean PENDING dir */
2213 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
2214 rc = mds_cleanup_pending(obd);
2218 /* FIXME Does target_finish_recovery really need this to block? */
2219 /* Notify the LOV, which will in turn call mds_notify for each tgt */
2220 /* This means that we have to hack obd_notify to think we're obd_set_up
2221 during mds_lov_connect. */
2222 obd_notify(obd->u.mds.mds_osc_obd, NULL,
2223 obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK :
2224 OBD_NOTIFY_SYNC, NULL);
2226 /* quota recovery */
2227 lquota_recovery(mds_quota_interface_ref, obd);
2233 /* We need to be able to stop an mds_lov_synchronize */
2234 static int mds_lov_early_clean(struct obd_device *obd)
2236 struct mds_obd *mds = &obd->u.mds;
2237 struct obd_device *osc = mds->mds_osc_obd;
2239 if (!osc || (!obd->obd_force && !obd->obd_fail))
2242 CDEBUG(D_HA, "abort inflight\n");
2243 return (obd_precleanup(osc, OBD_CLEANUP_EARLY));
2246 static int mds_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2252 case OBD_CLEANUP_EARLY:
2254 case OBD_CLEANUP_EXPORTS:
2255 /*XXX Use this for mdd mds cleanup, so comment out
2256 *this target_cleanup_recovery for this tmp MDD MDS
2258 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
2259 target_cleanup_recovery(obd);
2260 mds_lov_early_clean(obd);
2262 case OBD_CLEANUP_SELF_EXP:
2263 mds_lov_disconnect(obd);
2265 llog_cleanup(llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT));
2266 llog_cleanup(llog_get_context(obd, LLOG_LOVEA_ORIG_CTXT));
2267 rc = obd_llog_finish(obd, 0);
2269 case OBD_CLEANUP_OBD:
2275 static int mds_cleanup(struct obd_device *obd)
2277 struct mds_obd *mds = &obd->u.mds;
2278 lvfs_sbdev_type save_dev;
2281 if (obd->u.obt.obt_sb == NULL)
2283 save_dev = lvfs_sbdev(obd->u.obt.obt_sb);
2285 if (mds->mds_osc_exp)
2286 /* lov export was disconnected by mds_lov_clean;
2287 we just need to drop our ref */
2288 class_export_put(mds->mds_osc_exp);
2290 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
2291 lprocfs_free_per_client_stats(obd);
2292 lprocfs_free_obd_stats(obd);
2293 lprocfs_obd_cleanup(obd);
2295 lquota_cleanup(mds_quota_interface_ref, obd);
2297 mds_update_server_data(obd, 1);
2299 mds_lov_destroy_objids(obd);
2301 mds_fs_cleanup(obd);
2304 upcall_cache_cleanup(mds->mds_group_hash);
2305 mds->mds_group_hash = NULL;
2308 server_put_mount(obd->obd_name, mds->mds_vfsmnt);
2309 obd->u.obt.obt_sb = NULL;
2311 ldlm_namespace_free(obd->obd_namespace, obd->obd_force);
2313 spin_lock_bh(&obd->obd_processing_task_lock);
2314 if (obd->obd_recovering) {
2315 target_cancel_recovery_timer(obd);
2316 obd->obd_recovering = 0;
2318 spin_unlock_bh(&obd->obd_processing_task_lock);
2320 fsfilt_put_ops(obd->obd_fsops);
2322 LCONSOLE_INFO("MDT %s has stopped.\n", obd->obd_name);
2327 static void fixup_handle_for_resent_req(struct ptlrpc_request *req, int offset,
2328 struct ldlm_lock *new_lock,
2329 struct ldlm_lock **old_lock,
2330 struct lustre_handle *lockh)
2332 struct obd_export *exp = req->rq_export;
2333 struct ldlm_request *dlmreq =
2334 lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*dlmreq));
2335 struct lustre_handle remote_hdl = dlmreq->lock_handle[0];
2336 struct list_head *iter;
2338 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2341 spin_lock(&exp->exp_ldlm_data.led_lock);
2342 list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2343 struct ldlm_lock *lock;
2344 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2345 if (lock == new_lock)
2347 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2348 lockh->cookie = lock->l_handle.h_cookie;
2349 LDLM_DEBUG(lock, "restoring lock cookie");
2350 DEBUG_REQ(D_DLMTRACE, req,"restoring lock cookie "LPX64,
2353 *old_lock = LDLM_LOCK_GET(lock);
2354 spin_unlock(&exp->exp_ldlm_data.led_lock);
2358 spin_unlock(&exp->exp_ldlm_data.led_lock);
2360 /* If the xid matches, then we know this is a resent request,
2361 * and allow it. (It's probably an OPEN, for which we don't
2364 le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_xid))
2368 le64_to_cpu(exp->exp_mds_data.med_mcd->mcd_last_close_xid))
2371 /* This remote handle isn't enqueued, so we never received or
2372 * processed this request. Clear MSG_RESENT, because it can
2373 * be handled like any normal request now. */
2375 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2377 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2381 int intent_disposition(struct ldlm_reply *rep, int flag)
2385 return (rep->lock_policy_res1 & flag);
2388 void intent_set_disposition(struct ldlm_reply *rep, int flag)
2392 rep->lock_policy_res1 |= flag;
2395 static int mds_intent_policy(struct ldlm_namespace *ns,
2396 struct ldlm_lock **lockp, void *req_cookie,
2397 ldlm_mode_t mode, int flags, void *data)
2399 struct ptlrpc_request *req = req_cookie;
2400 struct ldlm_lock *lock = *lockp;
2401 struct ldlm_intent *it;
2402 struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
2403 struct ldlm_reply *rep;
2404 struct lustre_handle lockh = { 0 };
2405 struct ldlm_lock *new_lock = NULL;
2406 int getattr_part = MDS_INODELOCK_UPDATE;
2407 int repsize[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
2408 [DLM_LOCKREPLY_OFF] = sizeof(struct ldlm_reply),
2409 [DLM_REPLY_REC_OFF] = sizeof(struct mds_body),
2410 [DLM_REPLY_REC_OFF+1] = mds->mds_max_mdsize };
2411 int repbufcnt = 4, rc;
2414 LASSERT(req != NULL);
2416 if (lustre_msg_bufcount(req->rq_reqmsg) <= DLM_INTENT_IT_OFF) {
2417 /* No intent was provided */
2418 rc = lustre_pack_reply(req, 2, repsize, NULL);
2424 it = lustre_swab_reqbuf(req, DLM_INTENT_IT_OFF, sizeof(*it),
2425 lustre_swab_ldlm_intent);
2427 CERROR("Intent missing\n");
2428 RETURN(req->rq_status = -EFAULT);
2431 LDLM_DEBUG(lock, "intent policy, opc: %s", ldlm_it2str(it->opc));
2433 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
2434 (it->opc & (IT_OPEN | IT_GETATTR | IT_LOOKUP)))
2435 /* we should never allow OBD_CONNECT_ACL if not configured */
2436 repsize[repbufcnt++] = LUSTRE_POSIX_ACL_MAX_SIZE;
2437 else if (it->opc & IT_UNLINK)
2438 repsize[repbufcnt++] = mds->mds_max_cookiesize;
2440 rc = lustre_pack_reply(req, repbufcnt, repsize, NULL);
2442 RETURN(req->rq_status = rc);
2444 rep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF, sizeof(*rep));
2445 intent_set_disposition(rep, DISP_IT_EXECD);
2448 /* execute policy */
2449 switch ((long)it->opc) {
2451 case IT_CREAT|IT_OPEN:
2452 mds_counter_incr(req->rq_export, LPROC_MDS_OPEN);
2453 fixup_handle_for_resent_req(req, DLM_LOCKREQ_OFF, lock, NULL,
2455 /* XXX swab here to assert that an mds_open reint
2456 * packet is following */
2457 rep->lock_policy_res2 = mds_reint(req, DLM_INTENT_REC_OFF,
2460 /* We abort the lock if the lookup was negative and
2461 * we did not make it to the OPEN portion */
2462 if (!intent_disposition(rep, DISP_LOOKUP_EXECD))
2463 RETURN(ELDLM_LOCK_ABORTED);
2464 if (intent_disposition(rep, DISP_LOOKUP_NEG) &&
2465 !intent_disposition(rep, DISP_OPEN_OPEN))
2468 /* If there was an error of some sort or if we are not
2469 * returning any locks */
2470 if (rep->lock_policy_res2 ||
2471 !intent_disposition(rep, DISP_OPEN_LOCK))
2472 RETURN(ELDLM_LOCK_ABORTED);
2475 getattr_part = MDS_INODELOCK_LOOKUP;
2477 getattr_part |= MDS_INODELOCK_LOOKUP;
2478 OBD_COUNTER_INCREMENT(req->rq_export->exp_obd, getattr);
2480 fixup_handle_for_resent_req(req, DLM_LOCKREQ_OFF, lock,
2483 /* INODEBITS_INTEROP: if this lock was converted from a
2484 * plain lock (client does not support inodebits), then
2485 * child lock must be taken with both lookup and update
2486 * bits set for all operations.
2488 if (!(req->rq_export->exp_connect_flags & OBD_CONNECT_IBITS))
2489 getattr_part = MDS_INODELOCK_LOOKUP |
2490 MDS_INODELOCK_UPDATE;
2492 rep->lock_policy_res2 = mds_getattr_lock(req,DLM_INTENT_REC_OFF,
2493 getattr_part, &lockh);
2494 /* FIXME: LDLM can set req->rq_status. MDS sets
2495 policy_res{1,2} with disposition and status.
2496 - replay: returns 0 & req->status is old status
2497 - otherwise: returns req->status */
2498 if (intent_disposition(rep, DISP_LOOKUP_NEG))
2499 rep->lock_policy_res2 = 0;
2500 if (!intent_disposition(rep, DISP_LOOKUP_POS) ||
2501 rep->lock_policy_res2)
2502 RETURN(ELDLM_LOCK_ABORTED);
2503 if (req->rq_status != 0) {
2505 rep->lock_policy_res2 = req->rq_status;
2506 RETURN(ELDLM_LOCK_ABORTED);
2510 CERROR("Unhandled intent "LPD64"\n", it->opc);
2514 /* By this point, whatever function we called above must have either
2515 * filled in 'lockh', been an intent replay, or returned an error. We
2516 * want to allow replayed RPCs to not get a lock, since we would just
2517 * drop it below anyways because lock replay is done separately by the
2518 * client afterwards. For regular RPCs we want to give the new lock to
2519 * the client instead of whatever lock it was about to get. */
2520 if (new_lock == NULL)
2521 new_lock = ldlm_handle2lock(&lockh);
2522 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY))
2525 LASSERTF(new_lock != NULL, "op "LPX64" lockh "LPX64"\n",
2526 it->opc, lockh.cookie);
2528 /* If we've already given this lock to a client once, then we should
2529 * have no readers or writers. Otherwise, we should have one reader
2530 * _or_ writer ref (which will be zeroed below) before returning the
2531 * lock to a client. */
2532 if (new_lock->l_export == req->rq_export) {
2533 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2535 LASSERT(new_lock->l_export == NULL);
2536 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2541 if (new_lock->l_export == req->rq_export) {
2542 /* Already gave this to the client, which means that we
2543 * reconstructed a reply. */
2544 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2546 RETURN(ELDLM_LOCK_REPLACED);
2549 /* Fixup the lock to be given to the client */
2550 lock_res_and_lock(new_lock);
2551 new_lock->l_readers = 0;
2552 new_lock->l_writers = 0;
2554 new_lock->l_export = class_export_get(req->rq_export);
2555 spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2556 list_add(&new_lock->l_export_chain,
2557 &new_lock->l_export->exp_ldlm_data.led_held_locks);
2558 spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2560 new_lock->l_blocking_ast = lock->l_blocking_ast;
2561 new_lock->l_completion_ast = lock->l_completion_ast;
2563 memcpy(&new_lock->l_remote_handle, &lock->l_remote_handle,
2564 sizeof(lock->l_remote_handle));
2566 new_lock->l_flags &= ~LDLM_FL_LOCAL;
2568 unlock_res_and_lock(new_lock);
2569 LDLM_LOCK_PUT(new_lock);
2571 RETURN(ELDLM_LOCK_REPLACED);
2574 static int mdt_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
2576 struct mds_obd *mds = &obd->u.mds;
2577 struct lprocfs_static_vars lvars;
2578 int mds_min_threads;
2579 int mds_max_threads;
2583 lprocfs_mdt_init_vars(&lvars);
2584 lprocfs_obd_setup(obd, lvars.obd_vars);
2586 sema_init(&mds->mds_health_sem, 1);
2588 if (mds_num_threads) {
2589 /* If mds_num_threads is set, it is the min and the max. */
2590 if (mds_num_threads > MDS_THREADS_MAX)
2591 mds_num_threads = MDS_THREADS_MAX;
2592 if (mds_num_threads < MDS_THREADS_MIN)
2593 mds_num_threads = MDS_THREADS_MIN;
2594 mds_max_threads = mds_min_threads = mds_num_threads;
2596 /* Base min threads on memory and cpus */
2597 mds_min_threads = num_possible_cpus() * num_physpages >>
2598 (27 - CFS_PAGE_SHIFT);
2599 if (mds_min_threads < MDS_THREADS_MIN)
2600 mds_min_threads = MDS_THREADS_MIN;
2601 /* Largest auto threads start value */
2602 if (mds_min_threads > 32)
2603 mds_min_threads = 32;
2604 mds_max_threads = min(MDS_THREADS_MAX, mds_min_threads * 4);
2608 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
2609 MDS_MAXREPSIZE, MDS_REQUEST_PORTAL,
2610 MDC_REPLY_PORTAL, MDS_SERVICE_WATCHDOG_TIMEOUT,
2611 mds_handle, LUSTRE_MDS_NAME,
2612 obd->obd_proc_entry, NULL,
2613 mds_min_threads, mds_max_threads, "ll_mdt", 0);
2615 if (!mds->mds_service) {
2616 CERROR("failed to start service\n");
2617 GOTO(err_lprocfs, rc = -ENOMEM);
2620 rc = ptlrpc_start_threads(obd, mds->mds_service);
2622 GOTO(err_thread, rc);
2624 mds->mds_setattr_service =
2625 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
2626 MDS_MAXREPSIZE, MDS_SETATTR_PORTAL,
2627 MDC_REPLY_PORTAL, MDS_SERVICE_WATCHDOG_TIMEOUT,
2628 mds_handle, "mds_setattr",
2629 obd->obd_proc_entry, NULL,
2630 mds_min_threads, mds_max_threads,
2632 if (!mds->mds_setattr_service) {
2633 CERROR("failed to start getattr service\n");
2634 GOTO(err_thread, rc = -ENOMEM);
2637 rc = ptlrpc_start_threads(obd, mds->mds_setattr_service);
2639 GOTO(err_thread2, rc);
2641 mds->mds_readpage_service =
2642 ptlrpc_init_svc(MDS_NBUFS, MDS_BUFSIZE, MDS_MAXREQSIZE,
2643 MDS_MAXREPSIZE, MDS_READPAGE_PORTAL,
2644 MDC_REPLY_PORTAL, MDS_SERVICE_WATCHDOG_TIMEOUT,
2645 mds_handle, "mds_readpage",
2646 obd->obd_proc_entry, NULL,
2647 MDS_THREADS_MIN_READPAGE, mds_max_threads,
2649 if (!mds->mds_readpage_service) {
2650 CERROR("failed to start readpage service\n");
2651 GOTO(err_thread2, rc = -ENOMEM);
2654 rc = ptlrpc_start_threads(obd, mds->mds_readpage_service);
2657 GOTO(err_thread3, rc);
2659 ping_evictor_start();
2664 ptlrpc_unregister_service(mds->mds_readpage_service);
2665 mds->mds_readpage_service = NULL;
2667 ptlrpc_unregister_service(mds->mds_setattr_service);
2668 mds->mds_setattr_service = NULL;
2670 ptlrpc_unregister_service(mds->mds_service);
2671 mds->mds_service = NULL;
2673 lprocfs_obd_cleanup(obd);
2677 static int mdt_cleanup(struct obd_device *obd)
2679 struct mds_obd *mds = &obd->u.mds;
2682 ping_evictor_stop();
2684 down(&mds->mds_health_sem);
2685 ptlrpc_unregister_service(mds->mds_readpage_service);
2686 ptlrpc_unregister_service(mds->mds_setattr_service);
2687 ptlrpc_unregister_service(mds->mds_service);
2688 mds->mds_readpage_service = NULL;
2689 mds->mds_setattr_service = NULL;
2690 mds->mds_service = NULL;
2691 up(&mds->mds_health_sem);
2693 lprocfs_obd_cleanup(obd);
2698 static int mdt_health_check(struct obd_device *obd)
2700 struct mds_obd *mds = &obd->u.mds;
2703 down(&mds->mds_health_sem);
2704 rc |= ptlrpc_service_health_check(mds->mds_readpage_service);
2705 rc |= ptlrpc_service_health_check(mds->mds_setattr_service);
2706 rc |= ptlrpc_service_health_check(mds->mds_service);
2707 up(&mds->mds_health_sem);
2710 * health_check to return 0 on healthy
2711 * and 1 on unhealthy.
2719 static struct dentry *mds_lvfs_fid2dentry(__u64 id, __u32 gen, __u64 gr,
2722 struct obd_device *obd = data;
2725 fid.generation = gen;
2726 return mds_fid2dentry(&obd->u.mds, &fid, NULL);
2729 static int mds_health_check(struct obd_device *obd)
2731 struct obd_device_target *odt = &obd->u.obt;
2732 #ifdef USE_HEALTH_CHECK_WRITE
2733 struct mds_obd *mds = &obd->u.mds;
2737 if (odt->obt_sb->s_flags & MS_RDONLY)
2740 #ifdef USE_HEALTH_CHECK_WRITE
2741 LASSERT(mds->mds_health_check_filp != NULL);
2742 rc |= !!lvfs_check_io_health(obd, mds->mds_health_check_filp);
2747 static int mds_process_config(struct obd_device *obd, obd_count len, void *buf)
2749 struct lustre_cfg *lcfg = buf;
2750 struct lprocfs_static_vars lvars;
2753 lprocfs_mds_init_vars(&lvars);
2755 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, lcfg, obd);
2759 struct lvfs_callback_ops mds_lvfs_ops = {
2760 l_fid2dentry: mds_lvfs_fid2dentry,
2763 /* use obd ops to offer management infrastructure */
2764 static struct obd_ops mds_obd_ops = {
2765 .o_owner = THIS_MODULE,
2766 .o_connect = mds_connect,
2767 .o_reconnect = mds_reconnect,
2768 .o_init_export = mds_init_export,
2769 .o_destroy_export = mds_destroy_export,
2770 .o_disconnect = mds_disconnect,
2771 .o_setup = mds_setup,
2772 .o_precleanup = mds_precleanup,
2773 .o_cleanup = mds_cleanup,
2774 .o_postrecov = mds_postrecov,
2775 .o_statfs = mds_obd_statfs,
2776 .o_iocontrol = mds_iocontrol,
2777 .o_create = mds_obd_create,
2778 .o_destroy = mds_obd_destroy,
2779 .o_llog_init = mds_llog_init,
2780 .o_llog_finish = mds_llog_finish,
2781 .o_notify = mds_notify,
2782 .o_health_check = mds_health_check,
2783 .o_process_config = mds_process_config,
2786 static struct obd_ops mdt_obd_ops = {
2787 .o_owner = THIS_MODULE,
2788 .o_setup = mdt_setup,
2789 .o_cleanup = mdt_cleanup,
2790 .o_health_check = mdt_health_check,
2793 quota_interface_t *mds_quota_interface_ref;
2794 extern quota_interface_t mds_quota_interface;
2796 static __attribute__((unused)) int __init mds_init(void)
2799 struct lprocfs_static_vars lvars;
2801 request_module("lquota");
2802 mds_quota_interface_ref = PORTAL_SYMBOL_GET(mds_quota_interface);
2803 rc = lquota_init(mds_quota_interface_ref);
2805 if (mds_quota_interface_ref)
2806 PORTAL_SYMBOL_PUT(mds_quota_interface);
2809 init_obd_quota_ops(mds_quota_interface_ref, &mds_obd_ops);
2811 lprocfs_mds_init_vars(&lvars);
2812 class_register_type(&mds_obd_ops, NULL,
2813 lvars.module_vars, LUSTRE_MDS_NAME, NULL);
2814 lprocfs_mds_init_vars(&lvars);
2815 mdt_obd_ops = mdt_obd_ops; //make compiler happy
2816 // class_register_type(&mdt_obd_ops, NULL,
2817 // lvars.module_vars, LUSTRE_MDT_NAME, NULL);
2822 static __attribute__((unused)) void /*__exit*/ mds_exit(void)
2824 lquota_exit(mds_quota_interface_ref);
2825 if (mds_quota_interface_ref)
2826 PORTAL_SYMBOL_PUT(mds_quota_interface);
2828 class_unregister_type(LUSTRE_MDS_NAME);
2829 // class_unregister_type(LUSTRE_MDT_NAME);
2831 /*mds still need lov setup here*/
2832 static int mds_cmd_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
2834 struct mds_obd *mds = &obd->u.mds;
2835 struct lvfs_run_ctxt saved;
2837 struct vfsmount *mnt;
2838 struct lustre_sb_info *lsi;
2839 struct lustre_mount_info *lmi;
2840 struct dentry *dentry;
2844 CDEBUG(D_INFO, "obd %s setup \n", obd->obd_name);
2845 if (strncmp(obd->obd_name, MDD_OBD_NAME, strlen(MDD_OBD_NAME)))
2848 if (lcfg->lcfg_bufcount < 5) {
2849 CERROR("invalid arg for setup %s\n", MDD_OBD_NAME);
2852 dev = lustre_cfg_string(lcfg, 4);
2853 lmi = server_get_mount(dev);
2854 LASSERT(lmi != NULL);
2856 lsi = s2lsi(lmi->lmi_sb);
2858 /* FIXME: MDD LOV initialize objects.
2859 * we need only lmi here but not get mount
2860 * OSD did mount already, so put mount back
2862 atomic_dec(&lsi->lsi_mounts);
2865 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
2866 mds_init_ctxt(obd, mnt);
2868 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2869 dentry = simple_mkdir(current->fs->pwd, "OBJECTS", 0777, 1);
2870 if (IS_ERR(dentry)) {
2871 rc = PTR_ERR(dentry);
2872 CERROR("cannot create OBJECTS directory: rc = %d\n", rc);
2873 GOTO(err_putfs, rc);
2875 mds->mds_objects_dir = dentry;
2877 dentry = lookup_one_len("__iopen__", current->fs->pwd,
2878 strlen("__iopen__"));
2879 if (IS_ERR(dentry)) {
2880 rc = PTR_ERR(dentry);
2881 CERROR("cannot lookup __iopen__ directory: rc = %d\n", rc);
2882 GOTO(err_objects, rc);
2885 mds->mds_fid_de = dentry;
2886 if (!dentry->d_inode || is_bad_inode(dentry->d_inode)) {
2888 CERROR("__iopen__ directory has no inode? rc = %d\n", rc);
2891 rc = mds_lov_init_objids(obd);
2893 CERROR("cannot init lov objid rc = %d\n", rc);
2897 rc = mds_lov_presetup(mds, lcfg);
2899 GOTO(err_objects, rc);
2901 /* Don't wait for mds_postrecov trying to clear orphans */
2902 obd->obd_async_recov = 1;
2903 rc = mds_postsetup(obd);
2904 /* Bug 11557 - allow async abort_recov start
2905 FIXME can remove most of this obd_async_recov plumbing
2906 obd->obd_async_recov = 0;
2910 GOTO(err_objects, rc);
2912 mds->mds_max_mdsize = sizeof(struct lov_mds_md);
2913 mds->mds_max_cookiesize = sizeof(struct llog_cookie);
2916 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2919 dput(mds->mds_fid_de);
2921 dput(mds->mds_objects_dir);
2923 fsfilt_put_ops(obd->obd_fsops);
2927 static int mds_cmd_cleanup(struct obd_device *obd)
2929 struct mds_obd *mds = &obd->u.mds;
2930 struct lvfs_run_ctxt saved;
2935 LCONSOLE_WARN("%s: shutting down for failover; client state "
2936 "will be preserved.\n", obd->obd_name);
2938 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2940 mds_lov_destroy_objids(obd);
2942 if (mds->mds_objects_dir != NULL) {
2943 l_dput(mds->mds_objects_dir);
2944 mds->mds_objects_dir = NULL;
2947 shrink_dcache_parent(mds->mds_fid_de);
2948 dput(mds->mds_fid_de);
2949 LL_DQUOT_OFF(obd->u.obt.obt_sb);
2950 fsfilt_put_ops(obd->obd_fsops);
2952 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
2957 static int mds_cmd_health_check(struct obd_device *obd)
2962 static struct obd_ops mds_cmd_obd_ops = {
2963 .o_owner = THIS_MODULE,
2964 .o_setup = mds_cmd_setup,
2965 .o_cleanup = mds_cmd_cleanup,
2966 .o_precleanup = mds_precleanup,
2967 .o_create = mds_obd_create,
2968 .o_destroy = mds_obd_destroy,
2969 .o_llog_init = mds_llog_init,
2970 .o_llog_finish = mds_llog_finish,
2971 .o_notify = mds_notify,
2972 .o_postrecov = mds_postrecov,
2973 // .o_health_check = mds_cmd_health_check,
2976 static int __init mds_cmd_init(void)
2978 struct lprocfs_static_vars lvars;
2980 lprocfs_mds_init_vars(&lvars);
2981 class_register_type(&mds_cmd_obd_ops, NULL, lvars.module_vars,
2982 LUSTRE_MDS_NAME, NULL);
2987 static void /*__exit*/ mds_cmd_exit(void)
2989 class_unregister_type(LUSTRE_MDS_NAME);
2992 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2993 MODULE_DESCRIPTION("Lustre Metadata Server (MDS)");
2994 MODULE_LICENSE("GPL");
2996 module_init(mds_cmd_init);
2997 module_exit(mds_cmd_exit);