*
* Copyright (C) 2001-2003 Cluster File Systems, Inc.
*
- * This file is part of Lustre, http://www.sf.net/projects/lustre/
+ * This file is part of the Lustre file system, http://www.lustre.org
+ * Lustre is a trademark of Cluster File Systems, Inc.
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * You may have signed or agreed to another license before downloading
+ * this software. If so, you are bound by the terms and conditions
+ * of that agreement, and the following does not apply to you. See the
+ * LICENSE file included with this distribution for more information.
*
- * Lustre is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * GNU General Public License for more details.
+ * If you did not agree to a different license, then this copy of Lustre
+ * is open source software; you can redistribute it and/or modify it
+ * under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * In either case, Lustre is distributed in the hope that it will be
+ * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * license text for more details.
*/
#ifndef EXPORT_SYMTAB
# include <liblustre.h>
#endif
-#include <linux/obd_class.h>
-#include <linux/lustre_mds.h>
-#include <linux/lustre_dlm.h>
-#include <linux/lprocfs_status.h>
+#include <linux/lustre_acl.h>
+#include <obd_class.h>
+#include <lustre_dlm.h>
+/* fid_res_name_eq() */
+#include <lustre_fid.h>
+#include <lprocfs_status.h>
#include "mdc_internal.h"
int it_disposition(struct lookup_intent *it, int flag)
}
EXPORT_SYMBOL(it_set_disposition);
-static void mdc_fid2mdc_op_data(struct mdc_op_data *data, struct ll_uctxt *ctxt,
- struct ll_fid *f1, struct ll_fid *f2,
- const char *name, int namelen, int mode)
+void it_clear_disposition(struct lookup_intent *it, int flag)
{
- LASSERT(data);
- LASSERT(ctxt);
- LASSERT(f1);
-
- data->ctxt = *ctxt;
- data->fid1 = *f1;
- if (f2)
- data->fid2 = *f2;
- else
- memset(&data->fid2, 0, sizeof(data->fid2));
- data->name = name;
- data->namelen = namelen;
- data->create_mode = mode;
- data->mod_time = LTIME_S(CURRENT_TIME);
+ it->d.lustre.it_disposition &= ~flag;
}
+EXPORT_SYMBOL(it_clear_disposition);
static int it_to_lock_mode(struct lookup_intent *it)
{
+ ENTRY;
+
/* CREAT needs to be tested before open (both could be set) */
if (it->it_op & IT_CREAT)
return LCK_PW;
int it_open_error(int phase, struct lookup_intent *it)
{
if (it_disposition(it, DISP_OPEN_OPEN)) {
- if (phase == DISP_OPEN_OPEN)
+ if (phase >= DISP_OPEN_OPEN)
return it->d.lustre.it_status;
else
return 0;
}
if (it_disposition(it, DISP_OPEN_CREATE)) {
- if (phase == DISP_OPEN_CREATE)
+ if (phase >= DISP_OPEN_CREATE)
return it->d.lustre.it_status;
else
return 0;
}
if (it_disposition(it, DISP_LOOKUP_EXECD)) {
- if (phase == DISP_LOOKUP_EXECD)
+ if (phase >= DISP_LOOKUP_EXECD)
return it->d.lustre.it_status;
else
return 0;
}
if (it_disposition(it, DISP_IT_EXECD)) {
- if (phase == DISP_IT_EXECD)
+ if (phase >= DISP_IT_EXECD)
return it->d.lustre.it_status;
else
return 0;
EXPORT_SYMBOL(it_open_error);
/* this must be called on a lockh that is known to have a referenced lock */
-void mdc_set_lock_data(__u64 *l, void *data)
+int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
{
struct ldlm_lock *lock;
- struct lustre_handle *lockh = (struct lustre_handle *)l;
ENTRY;
- if (!*l) {
+ if (!*lockh) {
EXIT;
- return;
+ RETURN(0);
}
- lock = ldlm_handle2lock(lockh);
+ lock = ldlm_handle2lock((struct lustre_handle *)lockh);
LASSERT(lock != NULL);
- l_lock(&lock->l_resource->lr_namespace->ns_lock);
+ lock_res_and_lock(lock);
#ifdef __KERNEL__
if (lock->l_ast_data && lock->l_ast_data != data) {
struct inode *new_inode = data;
struct inode *old_inode = lock->l_ast_data;
- unsigned long state = old_inode->i_state & I_FREEING;
- CERROR("Found existing inode %p/%lu/%u state %lu in lock: "
- "setting data to %p/%lu/%u\n", old_inode,
- old_inode->i_ino, old_inode->i_generation, state,
- new_inode, new_inode->i_ino, new_inode->i_generation);
- LASSERT(state);
+ LASSERTF(old_inode->i_state & I_FREEING,
+ "Found existing inode %p/%lu/%u state %lu in lock: "
+ "setting data to %p/%lu/%u\n", old_inode,
+ old_inode->i_ino, old_inode->i_generation,
+ old_inode->i_state,
+ new_inode, new_inode->i_ino, new_inode->i_generation);
}
#endif
lock->l_ast_data = data;
- l_unlock(&lock->l_resource->lr_namespace->ns_lock);
+ unlock_res_and_lock(lock);
LDLM_LOCK_PUT(lock);
- EXIT;
+ RETURN(0);
}
-EXPORT_SYMBOL(mdc_set_lock_data);
-int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
+ldlm_mode_t mdc_lock_match(struct obd_export *exp, int flags,
+ const struct lu_fid *fid, ldlm_type_t type,
+ ldlm_policy_data_t *policy, ldlm_mode_t mode,
+ struct lustre_handle *lockh)
+{
+ struct ldlm_res_id res_id =
+ { .name = {fid_seq(fid),
+ fid_oid(fid),
+ fid_ver(fid)} };
+ ldlm_mode_t rc;
+ ENTRY;
+
+ rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
+ &res_id, type, policy, mode, lockh);
+ RETURN(rc);
+}
+
+int mdc_cancel_unused(struct obd_export *exp,
+ const struct lu_fid *fid,
+ ldlm_policy_data_t *policy,
+ ldlm_mode_t mode, int flags, void *opaque)
+{
+ struct ldlm_res_id res_id =
+ { .name = {fid_seq(fid),
+ fid_oid(fid),
+ fid_ver(fid)} };
+ struct obd_device *obd = class_exp2obd(exp);
+ int rc;
+
+ ENTRY;
+
+ rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
+ policy, mode, flags, opaque);
+ RETURN(rc);
+}
+
+int mdc_change_cbdata(struct obd_export *exp,
+ const struct lu_fid *fid,
ldlm_iterator_t it, void *data)
{
struct ldlm_res_id res_id = { .name = {0} };
ENTRY;
- res_id.name[0] = fid->id;
- res_id.name[1] = fid->generation;
+ res_id.name[0] = fid_seq(fid);
+ res_id.name[1] = fid_oid(fid);
+ res_id.name[2] = fid_ver(fid);
- ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace, &res_id, it,
- data);
+ ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace,
+ &res_id, it, data);
EXIT;
return 0;
}
+static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
+{
+ /* Don't hold error requests for replay. */
+ if (req->rq_replay) {
+ spin_lock(&req->rq_lock);
+ req->rq_replay = 0;
+ spin_unlock(&req->rq_lock);
+ }
+ if (rc && req->rq_transno != 0) {
+ DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
+ LBUG();
+ }
+}
+
+/* Save a large LOV EA into the request buffer so that it is available
+ * for replay. We don't do this in the initial request because the
+ * original request doesn't need this buffer (at most it sends just the
+ * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
+ * buffer and may also be difficult to allocate and save a very large
+ * request buffer for each open. (bug 5707)
+ *
+ * OOM here may cause recovery failure if lmm is needed (only for the
+ * original open if the MDS crashed just when this client also OOM'd)
+ * but this is incredibly unlikely, and questionable whether the client
+ * could do MDS recovery under OOM anyways... */
+static void mdc_realloc_openmsg(struct ptlrpc_request *req,
+ struct mdt_body *body, int size[9])
+{
+ int rc;
+ ENTRY;
+ rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
+ body->eadatasize);
+ if (rc) {
+ CERROR("Can't enlarge segment %d size to %d\n",
+ DLM_INTENT_REC_OFF + 4, body->eadatasize);
+ body->valid &= ~OBD_MD_FLEASIZE;
+ body->eadatasize = 0;
+ }
+ EXIT;
+}
/* We always reserve enough space in the reply packet for a stripe MD, because
* we don't know in advance the file type. */
-int mdc_enqueue(struct obd_export *exp,
- int lock_type,
- struct lookup_intent *it,
- int lock_mode,
- struct mdc_op_data *data,
- struct lustre_handle *lockh,
- void *lmm,
- int lmmsize,
- ldlm_completion_callback cb_completion,
- ldlm_blocking_callback cb_blocking,
- void *cb_data)
+int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
+ struct lookup_intent *it, struct md_op_data *op_data,
+ struct lustre_handle *lockh, void *lmm, int lmmsize,
+ int extra_lock_flags)
{
struct ptlrpc_request *req;
struct obd_device *obddev = class_exp2obd(exp);
struct ldlm_res_id res_id =
- { .name = {data->fid1.id, data->fid1.generation} };
+ { .name = {fid_seq(&op_data->op_fid1),
+ fid_oid(&op_data->op_fid1),
+ fid_ver(&op_data->op_fid1)} };
ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
- int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
- int rc, flags = LDLM_FL_HAS_INTENT;
- int repsize[4] = {sizeof(struct ldlm_reply),
- sizeof(struct mds_body),
- obddev->u.cli.cl_max_mds_easize,
- obddev->u.cli.cl_max_mds_cookiesize};
- struct ldlm_reply *dlm_rep;
- struct ldlm_intent *lit;
struct ldlm_request *lockreq;
- void *eadata;
- unsigned long irqflags;
- int reply_buffers = 0;
+ struct ldlm_intent *lit;
+ struct ldlm_reply *lockrep;
+ int size[9] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREQ_OFF] = sizeof(*lockreq),
+ [DLM_INTENT_IT_OFF] = sizeof(*lit),
+ 0, 0, 0, 0, 0, 0 };
+ int repsize[7] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREPLY_OFF] = sizeof(*lockrep),
+ [DLM_REPLY_REC_OFF] = sizeof(struct mdt_body),
+ [DLM_REPLY_REC_OFF+1] = obddev->u.cli.
+ cl_max_mds_easize,
+ 0, 0, 0 };
+ int flags = extra_lock_flags | LDLM_FL_HAS_INTENT;
+ int repbufcnt = 4, rc;
ENTRY;
-// LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
-// ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
+ LASSERTF(einfo->ei_type == LDLM_IBITS,"lock type %d\n", einfo->ei_type);
if (it->it_op & IT_OPEN) {
- it->it_create_mode |= S_IFREG;
- it->it_create_mode &= ~current->fs->umask;
-
- size[2] = sizeof(struct mds_rec_create);
- size[3] = data->namelen + 1;
- size[4] = obddev->u.cli.cl_max_mds_easize;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE,
- 5, size, NULL);
+ int do_join = !!(it->it_flags & O_JOIN_FILE);
+ CFS_LIST_HEAD(cancels);
+ int count = 0;
+ int mode;
+
+ it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
+
+ size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_rec_create);
+ /* parent capability */
+ size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
+ sizeof(struct lustre_capa) : 0;
+ /* child capability, used for replay only */
+ size[DLM_INTENT_REC_OFF + 2] = sizeof(struct lustre_capa);
+ size[DLM_INTENT_REC_OFF + 3] = op_data->op_namelen + 1;
+ /* As an optimization, we allocate an RPC request buffer for
+ * at least a default-sized LOV EA even if we aren't sending
+ * one.
+ */
+ size[DLM_INTENT_REC_OFF + 4] = max(lmmsize,
+ obddev->u.cli.cl_default_mds_easize);
+
+ /* XXX: openlock is not cancelled for cross-refs. */
+ /* If inode is known, cancel conflicting OPEN locks. */
+ if (fid_is_sane(&op_data->op_fid2)) {
+ if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
+ mode = LCK_CW;
+#ifdef FMODE_EXEC
+ else if (it->it_flags & FMODE_EXEC)
+ mode = LCK_PR;
+#endif
+ else
+ mode = LCK_CR;
+ count = mdc_resource_get_unused(exp, &op_data->op_fid2,
+ &cancels, mode,
+ MDS_INODELOCK_OPEN);
+ }
+
+ /* If CREATE or JOIN_FILE, cancel parent's UPDATE lock. */
+ if (it->it_op & IT_CREAT || it->it_flags & O_JOIN_FILE)
+ mode = LCK_EX;
+ else
+ mode = LCK_CR;
+ count += mdc_resource_get_unused(exp, &op_data->op_fid1,
+ &cancels, mode,
+ MDS_INODELOCK_UPDATE);
+
+ if (do_join)
+ size[DLM_INTENT_REC_OFF + 5] =
+ sizeof(struct mdt_rec_join);
+
+ req = ldlm_prep_enqueue_req(exp, 8 + do_join, size, &cancels,
+ count);
if (!req)
RETURN(-ENOMEM);
- spin_lock_irqsave (&req->rq_lock, irqflags);
+ if (do_join) {
+ /* join is like an unlink of the tail */
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ mdc_join_pack(req, DLM_INTENT_REC_OFF + 5, op_data,
+ (*(__u64 *)op_data->op_data));
+ }
+
+ spin_lock(&req->rq_lock);
req->rq_replay = 1;
- spin_unlock_irqrestore (&req->rq_lock, irqflags);
+ spin_unlock(&req->rq_lock);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
+ sizeof(*lit));
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_open_pack(req, 2, data, it->it_create_mode, 0,
- it->it_flags, lmm, lmmsize);
- /* get ready for the reply */
- reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
+ mdc_open_pack(req, DLM_INTENT_REC_OFF, op_data,
+ it->it_create_mode, 0, it->it_flags,
+ lmm, lmmsize);
+
+ /* for remote client, fetch remote perm for current user */
+ repsize[repbufcnt++] = client_is_remote(exp) ?
+ sizeof(struct mdt_remote_perm) :
+ LUSTRE_POSIX_ACL_MAX_SIZE;
+ repsize[repbufcnt++] = sizeof(struct lustre_capa);
+ repsize[repbufcnt++] = sizeof(struct lustre_capa);
} else if (it->it_op & IT_UNLINK) {
- size[2] = sizeof(struct mds_rec_unlink);
- size[3] = data->namelen + 1;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 4,
- size, NULL);
+ size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_rec_unlink);
+ size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
+ sizeof(struct lustre_capa) : 0;
+ size[DLM_INTENT_REC_OFF + 2] = op_data->op_namelen + 1;
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ req = ldlm_prep_enqueue_req(exp, 6, size, NULL, 0);
if (!req)
RETURN(-ENOMEM);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
+ sizeof(*lit));
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_unlink_pack(req, 2, data);
- /* get ready for the reply */
- reply_buffers = 4;
- req->rq_replen = lustre_msg_size(4, repsize);
+ mdc_unlink_pack(req, DLM_INTENT_REC_OFF, op_data);
+
+ repsize[repbufcnt++] = obddev->u.cli.cl_max_mds_cookiesize;
} else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
- int valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
- size[2] = sizeof(struct mds_body);
- size[3] = data->namelen + 1;
+ obd_valid valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
+ OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
+ OBD_MD_FLMDSCAPA | OBD_MD_MEA;
+ valid |= client_is_remote(exp) ? OBD_MD_FLRMTPERM :
+ OBD_MD_FLACL;
+ size[DLM_INTENT_REC_OFF] = sizeof(struct mdt_body);
+ size[DLM_INTENT_REC_OFF + 1] = op_data->op_capa1 ?
+ sizeof(struct lustre_capa) : 0;
+ size[DLM_INTENT_REC_OFF + 2] = op_data->op_namelen + 1;
if (it->it_op & IT_GETATTR)
policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 4,
- size, NULL);
+ req = ldlm_prep_enqueue_req(exp, 6, size, NULL, 0);
if (!req)
RETURN(-ENOMEM);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
+ sizeof(*lit));
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_getattr_pack(req, valid, 2, it->it_flags, data);
- /* get ready for the reply */
- reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
+ mdc_getattr_pack(req, DLM_INTENT_REC_OFF, valid,
+ it->it_flags, op_data);
+
+ repsize[repbufcnt++] = client_is_remote(exp) ?
+ sizeof(struct mdt_remote_perm) :
+ LUSTRE_POSIX_ACL_MAX_SIZE;
+ repsize[repbufcnt++] = sizeof(struct lustre_capa);
} else if (it->it_op == IT_READDIR) {
- policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 1,
- size, NULL);
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ req = ldlm_prep_enqueue_req(exp, 2, size, NULL, 0);
if (!req)
RETURN(-ENOMEM);
- /* get ready for the reply */
- reply_buffers = 1;
- req->rq_replen = lustre_msg_size(1, repsize);
- } else {
+ repbufcnt = 2;
+ } else {
LBUG();
RETURN(-EINVAL);
}
+
+ /* get ready for the reply */
+ ptlrpc_req_set_repsize(req, repbufcnt, repsize);
+
+ /* It is important to obtain rpc_lock first (if applicable), so that
+ * threads that are serialised with rpc_lock are not polluting our
+ * rpcs in flight counter */
mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
- rc = ldlm_cli_enqueue(exp, req, obddev->obd_namespace, res_id,
- lock_type, &policy, lock_mode, &flags,cb_blocking,
- cb_completion, NULL, cb_data, NULL, 0, NULL,
- lockh);
+ mdc_enter_request(&obddev->u.cli);
+ rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
+ 0, NULL, lockh, 0);
+ mdc_exit_request(&obddev->u.cli);
mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
/* Similarly, if we're going to replay this request, we don't want to
* actually get a lock, just perform the intent. */
if (req->rq_transno || req->rq_replay) {
- lockreq = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*lockreq));
+ lockreq = lustre_msg_buf(req->rq_reqmsg, DLM_LOCKREQ_OFF,
+ sizeof(*lockreq));
lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
}
- /* This can go when we're sure that this can never happen */
- LASSERT(rc != -ENOENT);
if (rc == ELDLM_LOCK_ABORTED) {
- lock_mode = 0;
+ einfo->ei_mode = 0;
memset(lockh, 0, sizeof(*lockh));
rc = 0;
} else if (rc != 0) {
CERROR("ldlm_cli_enqueue: %d\n", rc);
- LASSERT (rc < 0);
+ LASSERTF(rc < 0, "rc %d\n", rc);
+ mdc_clear_replay_flag(req, rc);
ptlrpc_req_finished(req);
RETURN(rc);
} else { /* rc = 0 */
/* If the server gave us back a different lock mode, we should
* fix up our variables. */
- if (lock->l_req_mode != lock_mode) {
+ if (lock->l_req_mode != einfo->ei_mode) {
ldlm_lock_addref(lockh, lock->l_req_mode);
- ldlm_lock_decref(lockh, lock_mode);
- lock_mode = lock->l_req_mode;
+ ldlm_lock_decref(lockh, einfo->ei_mode);
+ einfo->ei_mode = lock->l_req_mode;
}
-
- ldlm_lock_allow_match(lock);
LDLM_LOCK_PUT(lock);
}
- dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
- LASSERT(dlm_rep != NULL); /* checked by ldlm_cli_enqueue() */
- LASSERT_REPSWABBED(req, 0); /* swabbed by ldlm_cli_enqueue() */
+ lockrep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF,
+ sizeof(*lockrep));
+ LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
+ /* swabbed by ldlm_cli_enqueue() */
+ LASSERT(lustre_rep_swabbed(req, DLM_LOCKREPLY_OFF));
- it->d.lustre.it_disposition = (int) dlm_rep->lock_policy_res1;
- it->d.lustre.it_status = (int) dlm_rep->lock_policy_res2;
- it->d.lustre.it_lock_mode = lock_mode;
+ it->d.lustre.it_disposition = (int)lockrep->lock_policy_res1;
+ it->d.lustre.it_status = (int)lockrep->lock_policy_res2;
+ it->d.lustre.it_lock_mode = einfo->ei_mode;
it->d.lustre.it_data = req;
- if (it->d.lustre.it_status < 0 && req->rq_replay) {
- LASSERT(req->rq_transno == 0);
- /* Don't hold error requests for replay. */
- spin_lock(&req->rq_lock);
- req->rq_replay = 0;
- spin_unlock(&req->rq_lock);
- }
+ if (it->d.lustre.it_status < 0 && req->rq_replay)
+ mdc_clear_replay_flag(req, it->d.lustre.it_status);
+
+ /* If we're doing an IT_OPEN which did not result in an actual
+ * successful open, then we need to remove the bit which saves
+ * this request for unconditional replay.
+ *
+ * It's important that we do this first! Otherwise we might exit the
+ * function without doing so, and try to replay a failed create
+ * (bug 3440) */
+ if (it->it_op & IT_OPEN && req->rq_replay &&
+ (!it_disposition(it, DISP_OPEN_OPEN) ||it->d.lustre.it_status != 0))
+ mdc_clear_replay_flag(req, it->d.lustre.it_status);
+
+ DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
+ it->it_op,it->d.lustre.it_disposition,it->d.lustre.it_status);
/* We know what to expect, so we do any byte flipping required here */
- LASSERT(reply_buffers == 4 || reply_buffers == 3 || reply_buffers == 1);
- if (reply_buffers >= 3) {
- struct mds_body *body;
+ LASSERT(repbufcnt == 7 || repbufcnt == 6 || repbufcnt == 2);
+ if (repbufcnt >= 6) {
+ int reply_off = DLM_REPLY_REC_OFF;
+ struct mdt_body *body;
- body = lustre_swab_repbuf(req, 1, sizeof (*body),
- lustre_swab_mds_body);
+ body = lustre_swab_repbuf(req, reply_off++, sizeof(*body),
+ lustre_swab_mdt_body);
if (body == NULL) {
- CERROR ("Can't swab mds_body\n");
+ CERROR ("Can't swab mdt_body\n");
RETURN (-EPROTO);
}
- if ((body->valid & OBD_MD_FLEASIZE) != 0) {
- void *replayea;
- /* The eadata is opaque; just check that it is
- * there. Eventually, obd_unpackmd() will check
- * the contents */
- eadata = lustre_swab_repbuf(req, 2, body->eadatasize,
- NULL);
+ if (req->rq_replay && it_disposition(it, DISP_OPEN_OPEN) &&
+ !it_open_error(DISP_OPEN_OPEN, it)) {
+ /*
+ * If this is a successful OPEN request, we need to set
+ * replay handler and data early, so that if replay
+ * happens immediately after swabbing below, new reply
+ * is swabbed by that handler correctly.
+ */
+ mdc_set_open_replay_data(NULL, NULL, req);
+ }
+
+ if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
+ void *eadata;
+
+ /*
+ * The eadata is opaque; just check that it is there.
+ * Eventually, obd_unpackmd() will check the contents.
+ */
+ eadata = lustre_swab_repbuf(req, reply_off++,
+ body->eadatasize, NULL);
if (eadata == NULL) {
- CERROR ("Missing/short eadata\n");
- RETURN (-EPROTO);
+ CERROR("Missing/short eadata\n");
+ RETURN(-EPROTO);
+ }
+ if (body->valid & OBD_MD_FLMODEASIZE) {
+ if (obddev->u.cli.cl_max_mds_easize <
+ body->max_mdsize) {
+ obddev->u.cli.cl_max_mds_easize =
+ body->max_mdsize;
+ CDEBUG(D_INFO, "maxeasize become %d\n",
+ body->max_mdsize);
+ }
+ if (obddev->u.cli.cl_max_mds_cookiesize <
+ body->max_cookiesize) {
+ obddev->u.cli.cl_max_mds_cookiesize =
+ body->max_cookiesize;
+ CDEBUG(D_INFO, "cookiesize become %d\n",
+ body->max_cookiesize);
+ }
+ }
+
+ /*
+ * We save the reply LOV EA in case we have to replay a
+ * create for recovery. If we didn't allocate a large
+ * enough request buffer above we need to reallocate it
+ * here to hold the actual LOV EA.
+ *
+ * To not save LOV EA if request is not going to replay
+ * (for example error one).
+ */
+ if ((it->it_op & IT_OPEN) && req->rq_replay) {
+ if (lustre_msg_buflen(req->rq_reqmsg,
+ DLM_INTENT_REC_OFF + 4) <
+ body->eadatasize)
+ mdc_realloc_openmsg(req, body, size);
+
+ lmm = lustre_msg_buf(req->rq_reqmsg,
+ DLM_INTENT_REC_OFF + 4,
+ body->eadatasize);
+ if (lmm)
+ memcpy(lmm, eadata, body->eadatasize);
+ }
+ }
+ if (body->valid & OBD_MD_FLRMTPERM) {
+ struct mdt_remote_perm *perm;
+
+ LASSERT(client_is_remote(exp));
+ perm = lustre_swab_repbuf(req, reply_off++,
+ sizeof(*perm),
+ lustre_swab_mdt_remote_perm);
+ if (perm == NULL) {
+ CERROR("missing remote permission!\n");
+ RETURN(-EPROTO);
}
+ } else if ((body->valid & OBD_MD_FLACL) && body->aclsize) {
+ reply_off++;
+ }
+ if (body->valid & OBD_MD_FLMDSCAPA) {
+ struct lustre_capa *capa, *p;
+
+ capa = lustre_unpack_capa(req->rq_repmsg, reply_off++);
+ if (capa == NULL) {
+ CERROR("Missing/short MDS capability\n");
+ RETURN(-EPROTO);
+ }
+
if (it->it_op & IT_OPEN) {
- replayea = lustre_msg_buf(req->rq_reqmsg, 4,
- obddev->u.cli.cl_max_mds_easize);
- LASSERT(replayea);
- memcpy(replayea, eadata, body->eadatasize);
+ /* client fid capa will be checked in replay */
+ p = lustre_msg_buf(req->rq_reqmsg,
+ DLM_INTENT_REC_OFF + 2,
+ sizeof(*p));
+ LASSERT(p);
+ *p = *capa;
+ }
+ }
+ if (body->valid & OBD_MD_FLOSSCAPA) {
+ struct lustre_capa *capa;
+
+ capa = lustre_unpack_capa(req->rq_repmsg, reply_off++);
+ if (capa == NULL) {
+ CERROR("Missing/short OSS capability\n");
+ RETURN(-EPROTO);
}
}
}
RETURN(rc);
}
-EXPORT_SYMBOL(mdc_enqueue);
-
-/*
+/*
* This long block is all about fixing up the lock and request state
* so that it is correct as of the moment _before_ the operation was
* applied; that way, the VFS will think that everything is normal and
* Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
* child lookup.
*/
-int mdc_intent_lock(struct obd_export *exp, struct ll_uctxt *uctxt,
- struct ll_fid *pfid, const char *name, int len,
- void *lmm, int lmmsize, struct ll_fid *cfid,
- struct lookup_intent *it, int lookup_flags,
- struct ptlrpc_request **reqp,
- ldlm_blocking_callback cb_blocking)
+int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
+ void *lmm, int lmmsize, struct lookup_intent *it,
+ int lookup_flags, struct ptlrpc_request **reqp,
+ ldlm_blocking_callback cb_blocking,
+ int extra_lock_flags)
{
- struct lustre_handle lockh;
struct ptlrpc_request *request;
- int rc = 0;
- struct mds_body *mds_body;
struct lustre_handle old_lock;
+ struct lustre_handle lockh;
+ struct mdt_body *mdt_body;
struct ldlm_lock *lock;
+ int rc = 0;
ENTRY;
LASSERT(it);
- CDEBUG(D_DLMTRACE, "name: %*s in %ld, intent: %s\n", len, name,
- pfid ? (unsigned long) pfid->id : 0 , ldlm_it2str(it->it_op));
+ CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
+ ", intent: %s flags %#o\n", op_data->op_namelen,
+ op_data->op_name, PFID(&op_data->op_fid2),
+ PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
+ it->it_flags);
- if ((it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR) && cfid) {
+ if (fid_is_sane(&op_data->op_fid2) &&
+ (it->it_op & (IT_LOOKUP | IT_GETATTR))) {
/* We could just return 1 immediately, but since we should only
* be called in revalidate_it if we already have a lock, let's
* verify that. */
- struct ldlm_res_id res_id ={.name = {cfid->id,
- cfid->generation}};
- struct lustre_handle lockh;
ldlm_policy_data_t policy;
- int mode = LCK_PR;
-
- /* For the GETATTR case, ll_revalidate_it issues two separate
- queries - for LOOKUP and for UPDATE lock because if cannot
- check them together - we might have those two bits to be
- present in two separate granted locks */
- policy.l_inodebits.bits =
- (it->it_op == IT_GETATTR)?MDS_INODELOCK_UPDATE:
- MDS_INODELOCK_LOOKUP;
- mode = LCK_PR;
- rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, LCK_PR, &lockh);
- if (!rc) {
- mode = LCK_PW;
- rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, LCK_PW,
- &lockh);
- }
- if (rc) {
+ ldlm_mode_t mode;
+
+ /* As not all attributes are kept under update lock, e.g.
+ owner/group/acls are under lookup lock, we need both
+ ibits for GETATTR. */
+
+ /* For CMD, UPDATE lock and LOOKUP lock can not be got
+ * at the same for cross-object, so we can not match
+ * the 2 lock at the same time FIXME: but how to handle
+ * the above situation */
+ policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
+ MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
+
+ mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED,
+ &op_data->op_fid2, LDLM_IBITS, &policy,
+ LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh);
+ if (mode) {
memcpy(&it->d.lustre.it_lock_handle, &lockh,
sizeof(lockh));
it->d.lustre.it_lock_mode = mode;
}
- RETURN(rc);
+
+ /* Only return failure if it was not GETATTR by cfid
+ (from inode_revalidate) */
+ if (mode || op_data->op_namelen != 0)
+ RETURN(!!mode);
}
/* lookup_it may be called only after revalidate_it has run, because
* this and use the request from revalidate. In this case, revalidate
* never dropped its reference, so the refcounts are all OK */
if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
- struct mdc_op_data op_data;
- mdc_fid2mdc_op_data(&op_data, uctxt, pfid, cfid, name, len, 0);
-
- rc = mdc_enqueue(exp, LDLM_IBITS, it, it_to_lock_mode(it),
- &op_data, &lockh, lmm, lmmsize,
- ldlm_completion_ast, cb_blocking, NULL);
+ struct ldlm_enqueue_info einfo =
+ { LDLM_IBITS, it_to_lock_mode(it), cb_blocking,
+ ldlm_completion_ast, NULL, NULL };
+
+ /* For case if upper layer did not alloc fid, do it now. */
+ if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
+ rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
+ if (rc < 0) {
+ CERROR("Can't alloc new fid, rc %d\n", rc);
+ RETURN(rc);
+ }
+ }
+ rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh,
+ lmm, lmmsize, extra_lock_flags);
if (rc < 0)
RETURN(rc);
memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
+ } else if (!fid_is_sane(&op_data->op_fid2) ||
+ !(it->it_flags & O_CHECK_STALE)) {
+ /* DISP_ENQ_COMPLETE set means there is extra reference on
+ * request referenced from this intent, saved for subsequent
+ * lookup. This path is executed when we proceed to this
+ * lookup, so we clear DISP_ENQ_COMPLETE */
+ it_clear_disposition(it, DISP_ENQ_COMPLETE);
}
request = *reqp = it->d.lustre.it_data;
LASSERT(request != NULL);
+ LASSERT(request != LP_POISON);
+ LASSERT(request->rq_repmsg != LP_POISON);
if (!it_disposition(it, DISP_IT_EXECD)) {
/* The server failed before it even started executing the
if (rc)
RETURN(rc);
- mds_body = lustre_msg_buf(request->rq_repmsg, 1, sizeof(*mds_body));
- LASSERT(mds_body != NULL); /* mdc_enqueue checked */
- LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
+ mdt_body = lustre_msg_buf(request->rq_repmsg, DLM_REPLY_REC_OFF,
+ sizeof(*mdt_body));
+ /* mdc_enqueue checked */
+ LASSERT(mdt_body != NULL);
+ /* mdc_enqueue swabbed */
+ LASSERT(lustre_rep_swabbed(request, 1));
/* If we were revalidating a fid/name pair, mark the intent in
* case we fail and get called again from lookup */
- if (cfid != NULL) {
+ if (fid_is_sane(&op_data->op_fid2) && (it->it_flags & O_CHECK_STALE) &&
+ (it->it_op != IT_GETATTR)) {
it_set_disposition(it, DISP_ENQ_COMPLETE);
+
/* Also: did we find the same inode? */
- /* we have to compare all the fields but type, because
- * MDS can return mds/ino/generation triple if inode
- * lives on another MDS -bzzz */
- if (cfid->generation != mds_body->fid1.generation ||
- cfid->id != mds_body->fid1.id ||
- cfid->mds != mds_body->fid1.mds)
+ if (!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1))
RETURN(-ESTALE);
}
- /* If we're doing an IT_OPEN which did not result in an actual
- * successful open, then we need to remove the bit which saves
- * this request for unconditional replay. */
- if (it->it_op & IT_OPEN) {
- if (!it_disposition(it, DISP_OPEN_OPEN) ||
- it->d.lustre.it_status != 0) {
- unsigned long irqflags;
-
- spin_lock_irqsave(&request->rq_lock, irqflags);
- request->rq_replay = 0;
- spin_unlock_irqrestore(&request->rq_lock, irqflags);
- }
- }
-
rc = it_open_error(DISP_LOOKUP_EXECD, it);
if (rc)
RETURN(rc);
/* keep requests around for the multiple phases of the call
* this shows the DISP_XX must guarantee we make it into the call
*/
- if (it_disposition(it, DISP_OPEN_CREATE) &&
- !it_open_error(DISP_OPEN_CREATE, it))
- ptlrpc_request_addref(request);
- if (it_disposition(it, DISP_OPEN_OPEN) &&
- !it_open_error(DISP_OPEN_OPEN, it))
- ptlrpc_request_addref(request);
+ if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
+ it_disposition(it, DISP_OPEN_CREATE) &&
+ !it_open_error(DISP_OPEN_CREATE, it)) {
+ it_set_disposition(it, DISP_ENQ_CREATE_REF);
+ ptlrpc_request_addref(request); /* balanced in ll_create_node */
+ }
+ if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
+ it_disposition(it, DISP_OPEN_OPEN) &&
+ !it_open_error(DISP_OPEN_OPEN, it)) {
+ it_set_disposition(it, DISP_ENQ_OPEN_REF);
+ ptlrpc_request_addref(request); /* balanced in ll_file_open */
+ /* BUG 11546 - eviction in the middle of open rpc processing */
+ OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
+ }
if (it->it_op & IT_CREAT) {
- /* XXX this belongs in ll_create_iit */
+ /* XXX this belongs in ll_create_it */
} else if (it->it_op == IT_OPEN) {
LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
} else {
if (lock) {
ldlm_policy_data_t policy = lock->l_policy_data;
LDLM_DEBUG(lock, "matching against this");
+
+ LASSERTF(fid_res_name_eq(&mdt_body->fid1,
+ &lock->l_resource->lr_name),
+ "Lock res_id: %lu/%lu/%lu, fid: %lu/%lu/%lu.\n",
+ (unsigned long)lock->l_resource->lr_name.name[0],
+ (unsigned long)lock->l_resource->lr_name.name[1],
+ (unsigned long)lock->l_resource->lr_name.name[2],
+ (unsigned long)fid_seq(&mdt_body->fid1),
+ (unsigned long)fid_oid(&mdt_body->fid1),
+ (unsigned long)fid_ver(&mdt_body->fid1));
LDLM_LOCK_PUT(lock);
+
memcpy(&old_lock, &lockh, sizeof(lockh));
if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
sizeof(lockh));
}
}
- CDEBUG(D_DENTRY, "D_IT dentry %*s intent: %s status %d disp %x rc %d\n",
- len, name, ldlm_it2str(it->it_op), it->d.lustre.it_status,
- it->d.lustre.it_disposition, rc);
+ CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
+ op_data->op_namelen, op_data->op_name, ldlm_it2str(it->it_op),
+ it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
RETURN(rc);
}
-EXPORT_SYMBOL(mdc_intent_lock);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff