*
* 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/lustre_sec.h>
-#include <linux/lprocfs_status.h>
-#include <linux/lustre_acl.h>
-#include <linux/lustre_gs.h>
-#include <linux/lustre_lite.h>
+#include <obd_class.h>
+#include <lustre_dlm.h>
+#include <lprocfs_status.h>
#include "mdc_internal.h"
int it_disposition(struct lookup_intent *it, int flag)
{
- return LUSTRE_IT(it)->it_disposition & flag;
+ return it->d.lustre.it_disposition & flag;
}
EXPORT_SYMBOL(it_disposition);
void it_set_disposition(struct lookup_intent *it, int flag)
{
- LUSTRE_IT(it)->it_disposition |= flag;
+ it->d.lustre.it_disposition |= flag;
}
EXPORT_SYMBOL(it_set_disposition);
-static void mdc_id2mdc_data(struct mdc_op_data *data,
- struct lustre_id *f1,
- struct lustre_id *f2,
- const char *name,
- int namelen, int mode)
+void it_clear_disposition(struct lookup_intent *it, int flag)
{
- LASSERT(data);
- LASSERT(f1);
-
- data->id1 = *f1;
- if (f2)
- data->id2 = *f2;
-
- data->valid = 0;
- data->name = name;
- data->namelen = namelen;
- data->create_mode = mode;
- data->mod_time = LTIME_S(CURRENT_TIME);
-}
-
-static int it_to_lock_mode(struct lookup_intent *it)
-{
- /* CREAT needs to be tested before open (both could be set) */
- if (it->it_op & IT_CREAT)
- return LCK_PW;
- else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
- IT_CHDIR))
- return LCK_PR;
-
- LBUG();
- RETURN(-EINVAL);
+ it->d.lustre.it_disposition &= ~flag;
}
+EXPORT_SYMBOL(it_clear_disposition);
int it_open_error(int phase, struct lookup_intent *it)
{
if (it_disposition(it, DISP_OPEN_OPEN)) {
- if (phase == DISP_OPEN_OPEN)
- return LUSTRE_IT(it)->it_status;
+ 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)
- return LUSTRE_IT(it)->it_status;
+ 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)
- return LUSTRE_IT(it)->it_status;
+ 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)
- return LUSTRE_IT(it)->it_status;
+ if (phase >= DISP_IT_EXECD)
+ return it->d.lustre.it_status;
else
return 0;
}
- CERROR("it disp: %X, status: %d\n", LUSTRE_IT(it)->it_disposition,
- LUSTRE_IT(it)->it_status);
+ CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
+ it->d.lustre.it_status);
LBUG();
return 0;
}
EXPORT_SYMBOL(it_open_error);
/* this must be called on a lockh that is known to have a referenced lock */
-int mdc_set_lock_data(struct obd_export *exp, __u64 *l, void *data)
+void mdc_set_lock_data(__u64 *l, void *data)
{
struct ldlm_lock *lock;
struct lustre_handle *lockh = (struct lustre_handle *)l;
if (!*l) {
EXIT;
- return 0;
+ return;
}
lock = ldlm_handle2lock(lockh);
lock_res_and_lock(lock);
#ifdef __KERNEL__
if (lock->l_ast_data && lock->l_ast_data != data) {
- struct ll_inode_info *lli = ll_i2info(data);
- struct inode *old_inode = lock->l_ast_data;
struct inode *new_inode = data;
- LASSERT(lli->lli_inode_magic == LLI_INODE_MAGIC);
- if (!(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, old_inode->i_state,
- new_inode, new_inode->i_ino,
- new_inode->i_generation);
- unlock_res_and_lock(lock);
- LBUG();
- }
+ struct inode *old_inode = lock->l_ast_data;
+ 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;
LDLM_LOCK_PUT(lock);
EXIT;
- return 0;
}
EXPORT_SYMBOL(mdc_set_lock_data);
-int mdc_change_cbdata(struct obd_export *exp, struct lustre_id *id,
+int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
ldlm_iterator_t it, void *data)
{
struct ldlm_res_id res_id = { .name = {0} };
ENTRY;
- res_id.name[0] = id_fid(id);
- res_id.name[1] = id_group(id);
+ res_id.name[0] = fid->id;
+ res_id.name[1] = fid->generation;
- 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)
+static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
{
/* Don't hold error requests for replay. */
if (req->rq_replay) {
- unsigned long irqflags;
- spin_lock_irqsave(&req->rq_lock, irqflags);
+ spin_lock(&req->rq_lock);
req->rq_replay = 0;
- spin_unlock_irqrestore(&req->rq_lock, irqflags);
+ spin_unlock(&req->rq_lock);
}
if (rc && req->rq_transno != 0) {
DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
}
}
-/* 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)
+static int round_up(int val)
+{
+ int ret = 1;
+ while (val) {
+ val >>= 1;
+ ret <<= 1;
+ }
+ return ret;
+}
+
+/* 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 mds_body *body)
+{
+ int old_len, new_size, old_size;
+ struct lustre_msg *old_msg = req->rq_reqmsg;
+ struct lustre_msg *new_msg;
+
+ old_len = lustre_msg_buflen(old_msg, DLM_INTENT_REC_OFF + 2);
+ old_size = lustre_packed_msg_size(old_msg);
+ lustre_msg_set_buflen(old_msg, DLM_INTENT_REC_OFF + 2,
+ body->eadatasize);
+ new_size = lustre_packed_msg_size(old_msg);
+
+ OBD_ALLOC(new_msg, new_size);
+ if (new_msg != NULL) {
+ DEBUG_REQ(D_INFO, req, "replace reqmsg for larger EA %u\n",
+ body->eadatasize);
+ memcpy(new_msg, old_msg, old_size);
+
+ spin_lock(&req->rq_lock);
+ req->rq_reqmsg = new_msg;
+ req->rq_reqlen = new_size;
+ spin_unlock(&req->rq_lock);
+
+ OBD_FREE(old_msg, old_size);
+ } else {
+ lustre_msg_set_buflen(old_msg, DLM_INTENT_REC_OFF + 2, old_len);
+ body->valid &= ~OBD_MD_FLEASIZE;
+ body->eadatasize = 0;
+ }
+}
+
+static struct ptlrpc_request *mdc_intent_open_pack(struct obd_export *exp,
+ struct lookup_intent *it,
+ struct mdc_op_data *data,
+ void *lmm, int lmmsize)
{
struct ptlrpc_request *req;
- struct ldlm_res_id res_id = {
- .name = {id_fid(&data->id1), id_group(&data->id1)}
- };
- struct obd_device *obddev = class_exp2obd(exp);
- ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
struct ldlm_intent *lit;
- struct ldlm_request *lockreq;
- int reqsize[6] = {[MDS_REQ_SECDESC_OFF] = 0,
- [MDS_REQ_INTENT_LOCKREQ_OFF] = sizeof(*lockreq),
- [MDS_REQ_INTENT_IT_OFF] = sizeof(*lit)};
- int repsize[8] = {sizeof(struct ldlm_reply),
- sizeof(struct mds_body),
- obddev->u.cli.cl_max_mds_easize};
- int req_buffers = 3, reply_buffers = 0;
- int rc, flags = LDLM_FL_HAS_INTENT;
- struct ldlm_reply *dlm_rep = NULL;
- void *eadata;
- unsigned long irqflags;
- ENTRY;
-
-// LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
-// ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
+ struct obd_device *obddev = class_exp2obd(exp);
+ int size[7] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREQ_OFF] = sizeof(struct ldlm_request),
+ [DLM_INTENT_IT_OFF] = sizeof(*lit),
+ [DLM_INTENT_REC_OFF] = sizeof(struct mds_rec_create),
+ [DLM_INTENT_REC_OFF+1]= data->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. We grow the whole request to the next
+ * power-of-two size since we get that much from a slab
+ * allocation anyways. This avoids an allocation below
+ * in the common case where we need to save a
+ * default-sized LOV EA for open replay. */
+ [DLM_INTENT_REC_OFF+2]= max(lmmsize,
+ obddev->u.cli.cl_default_mds_easize) };
+ int repsize[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREPLY_OFF] = sizeof(struct ldlm_reply),
+ [DLM_REPLY_REC_OFF] = sizeof(struct mds_body),
+ [DLM_REPLY_REC_OFF+1] = obddev->u.cli.
+ cl_max_mds_easize,
+ [DLM_REPLY_REC_OFF+2] = LUSTRE_POSIX_ACL_MAX_SIZE };
+ CFS_LIST_HEAD(cancels);
+ int do_join = (it->it_flags & O_JOIN_FILE) && data->data;
+ int count = 0;
+ int mode;
+ int rc;
+
+ it->it_create_mode |= S_IFREG;
+
+ rc = lustre_msg_size(class_exp2cliimp(exp)->imp_msg_magic, 6, size);
+ if (rc & (rc - 1))
+ size[DLM_INTENT_REC_OFF + 2] =
+ min(size[DLM_INTENT_REC_OFF + 2] + round_up(rc) - rc,
+ obddev->u.cli.cl_max_mds_easize);
+
+ /* If inode is known, cancel conflicting OPEN locks. */
+ if (data->fid2.id) {
+ 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, &data->fid2, &cancels,
+ mode, MDS_INODELOCK_OPEN);
+ }
- reqsize[0] = lustre_secdesc_size();
+ /* If CREATE or JOIN_FILE, cancel parent's UPDATE lock. */
+ if (it->it_op & IT_CREAT || do_join)
+ mode = LCK_EX;
+ else
+ mode = LCK_CR;
+ count += mdc_resource_get_unused(exp, &data->fid1, &cancels, mode,
+ MDS_INODELOCK_UPDATE);
+ if (do_join) {
+ __u64 head_size = (*(__u64 *)data->data);
+ /* join is like an unlink of the tail */
+ size[DLM_INTENT_REC_OFF + 3] = sizeof(struct mds_rec_join);
+ req = ldlm_prep_enqueue_req(exp, 7, size, &cancels, count);
+ if (req)
+ mdc_join_pack(req, DLM_INTENT_REC_OFF + 3, data,
+ head_size);
+ } else {
+ req = ldlm_prep_enqueue_req(exp, 6, size, &cancels, count);
+ it->it_flags &= ~O_JOIN_FILE;
+ }
- if (it->it_op & IT_OPEN) {
- struct lustre_intent_data *lustre_data =
- (struct lustre_intent_data *) it->d.fs_data;
- it->it_create_mode |= S_IFREG;
- it->it_create_mode &= ~current->fs->umask;
-
- reqsize[req_buffers++] = sizeof(struct mds_rec_create);
- reqsize[req_buffers++] = data->namelen + 1;
- reqsize[req_buffers++] = obddev->u.cli.cl_max_mds_easize;
- /*pack the lustre key*/
- if (lustre_data->it_key_size > 0 && lustre_data->it_key)
- reqsize[req_buffers++] = lustre_data->it_key_size;
-
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
- LDLM_ENQUEUE, req_buffers, reqsize, NULL);
- if (!req)
- RETURN(-ENOMEM);
-
- spin_lock_irqsave (&req->rq_lock, irqflags);
+ if (req) {
+ 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, MDS_REQ_INTENT_IT_OFF,
- 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->rq_reqmsg, MDS_REQ_INTENT_REC_OFF, data,
- it->it_create_mode, 0, it->it_flags, lmm, lmmsize,
- lustre_data->it_key, lustre_data->it_key_size);
- /* get ready for the reply */
- reply_buffers = 3;
- repsize[reply_buffers++] = sizeof(int);
- repsize[reply_buffers++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
- repsize[reply_buffers++] = sizeof(int);
- repsize[reply_buffers++] = sizeof(struct crypto_key);
- repsize[reply_buffers++] = sizeof(struct lustre_capa);
- req->rq_replen = lustre_msg_size(reply_buffers, repsize);
- } else if (it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR)) {
- __u64 valid = data->valid | OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE |
- OBD_MD_FLACL | OBD_MD_FLKEY;
-
- /* we don't expect xattr retrieve could reach here */
- LASSERT(!(valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLIST)));
-
- reqsize[req_buffers++] = sizeof(struct mds_body);
- reqsize[req_buffers++] = data->namelen + 1;
-
- if (it->it_op & IT_GETATTR)
- policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
- LDLM_ENQUEUE, req_buffers, reqsize, NULL);
+ mdc_open_pack(req, DLM_INTENT_REC_OFF, data,
+ it->it_create_mode, 0, it->it_flags,
+ lmm, lmmsize);
- if (!req)
- RETURN(-ENOMEM);
+ ptlrpc_req_set_repsize(req, 5, repsize);
+ }
+ return req;
+}
+static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
+ struct lookup_intent *it,
+ struct mdc_op_data *data)
+{
+ struct ptlrpc_request *req;
+ struct ldlm_intent *lit;
+ struct obd_device *obddev = class_exp2obd(exp);
+ int size[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREQ_OFF] = sizeof(struct ldlm_request),
+ [DLM_INTENT_IT_OFF] = sizeof(*lit),
+ [DLM_INTENT_REC_OFF] = sizeof(struct mds_rec_unlink),
+ [DLM_INTENT_REC_OFF+1]= data->namelen + 1 };
+ int repsize[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREPLY_OFF] = sizeof(struct ldlm_reply),
+ [DLM_REPLY_REC_OFF] = sizeof(struct mds_body),
+ [DLM_REPLY_REC_OFF+1] = obddev->u.cli.
+ cl_max_mds_easize,
+ [DLM_REPLY_REC_OFF+2] = obddev->u.cli.
+ cl_max_mds_cookiesize };
+
+ req = ldlm_prep_enqueue_req(exp, 5, size, NULL, 0);
+ if (req) {
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_INTENT_IT_OFF,
- 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->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
- valid, it->it_flags, data);
-
- /* get ready for the reply */
- reply_buffers = 3;
- repsize[reply_buffers++] = sizeof(int);
- repsize[reply_buffers++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
- repsize[reply_buffers++] = sizeof(int);
- repsize[reply_buffers++] = sizeof(struct crypto_key);
- req->rq_replen = lustre_msg_size(reply_buffers, repsize);
- } else if (it->it_op == IT_READDIR) {
- policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
- LDLM_ENQUEUE, 2, reqsize, NULL);
-
- if (!req)
- RETURN(-ENOMEM);
- /* get ready for the reply */
- reply_buffers = 1;
- req->rq_replen = lustre_msg_size(1, repsize);
- } else if (it->it_op == IT_UNLINK) {
- reqsize[req_buffers++] = sizeof(struct mds_body);
- policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
- LDLM_ENQUEUE, req_buffers, reqsize, NULL);
- if (!req)
- RETURN(-ENOMEM);
+ mdc_unlink_pack(req, DLM_INTENT_REC_OFF, data);
- /* pack the intended request */
- mdc_getattr_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
- 0, 0, data);
+ ptlrpc_req_set_repsize(req, 5, repsize);
+ }
+ return req;
+}
+static struct ptlrpc_request *mdc_intent_lookup_pack(struct obd_export *exp,
+ struct lookup_intent *it,
+ struct mdc_op_data *data)
+{
+ struct ptlrpc_request *req;
+ struct ldlm_intent *lit;
+ struct obd_device *obddev = class_exp2obd(exp);
+ int size[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREQ_OFF] = sizeof(struct ldlm_request),
+ [DLM_INTENT_IT_OFF] = sizeof(*lit),
+ [DLM_INTENT_REC_OFF] = sizeof(struct mds_body),
+ [DLM_INTENT_REC_OFF+1]= data->namelen + 1 };
+ int repsize[5] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREPLY_OFF] = sizeof(struct ldlm_reply),
+ [DLM_REPLY_REC_OFF] = sizeof(struct mds_body),
+ [DLM_REPLY_REC_OFF+1] = obddev->u.cli.
+ cl_max_mds_easize,
+ [DLM_REPLY_REC_OFF+2] = LUSTRE_POSIX_ACL_MAX_SIZE };
+ obd_valid valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE | OBD_MD_FLACL |
+ OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA;
+
+ req = ldlm_prep_enqueue_req(exp, 5, size, NULL, 0);
+ if (req) {
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_INTENT_IT_OFF,
- sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, DLM_INTENT_IT_OFF,
+ sizeof(*lit));
lit->opc = (__u64)it->it_op;
- /* get ready for the reply */
- reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
- } else {
- LBUG();
- RETURN(-EINVAL);
+ /* pack the intended request */
+ mdc_getattr_pack(req, DLM_INTENT_REC_OFF, valid, it->it_flags,
+ data);
+ ptlrpc_req_set_repsize(req, 5, repsize);
}
+ return req;
+}
- lustre_pack_secdesc(req, reqsize[0]);
+static struct ptlrpc_request *mdc_intent_readdir_pack(struct obd_export *exp)
+{
+ struct ptlrpc_request *req;
+ int size[2] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREQ_OFF] = sizeof(struct ldlm_request) };
+ int repsize[2] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body),
+ [DLM_LOCKREPLY_OFF] = sizeof(struct ldlm_reply) };
+
+ req = ldlm_prep_enqueue_req(exp, 2, size, NULL, 0);
+ if (req)
+ ptlrpc_req_set_repsize(req, 2, repsize);
+ return req;
+}
- 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_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+static int mdc_finish_enqueue(struct obd_export *exp,
+ struct ptlrpc_request *req,
+ struct ldlm_enqueue_info *einfo,
+ struct lookup_intent *it,
+ struct lustre_handle *lockh,
+ int rc)
+{
+ struct ldlm_request *lockreq;
+ struct ldlm_reply *lockrep;
+ ENTRY;
/* 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,
- MDS_REQ_INTENT_LOCKREQ_OFF,
- 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);
- /* We need dlm_rep to be assigned this early, to check lock mode of
- returned lock from request to avoid possible race with lock
- conversion */
- if (rc == ELDLM_LOCK_ABORTED || !rc) {
- dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
- LASSERT(dlm_rep != NULL); /* checked by ldlm_cli_enqueue() */
- }
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);
- LASSERTF(rc < 0, "rc = %d\n", rc);
+ LASSERTF(rc < 0, "rc %d\n", rc);
mdc_clear_replay_flag(req, rc);
ptlrpc_req_finished(req);
RETURN(rc);
/* If the server gave us back a different lock mode, we should
* fix up our variables. */
- if (dlm_rep->lock_desc.l_req_mode != lock_mode) {
- ldlm_lock_addref(lockh, dlm_rep->lock_desc.l_req_mode);
- ldlm_lock_decref(lockh, lock_mode);
- lock_mode = dlm_rep->lock_desc.l_req_mode;
+ if (lock->l_req_mode != einfo->ei_mode) {
+ ldlm_lock_addref(lockh, 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);
}
- 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));
- LUSTRE_IT(it)->it_disposition = (int) dlm_rep->lock_policy_res1;
- LUSTRE_IT(it)->it_status = (int) dlm_rep->lock_policy_res2;
- LUSTRE_IT(it)->it_lock_mode = lock_mode;
- LUSTRE_IT(it)->it_data = req;
+ 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 (LUSTRE_IT(it)->it_status < 0 && req->rq_replay)
- mdc_clear_replay_flag(req, LUSTRE_IT(it)->it_status);
+ if (it->d.lustre.it_status < 0 && req->rq_replay)
+ mdc_clear_replay_flag(req, it->d.lustre.it_status);
- DEBUG_REQ(D_DLMTRACE, req, "disposition: %x, status: %d",
- LUSTRE_IT(it)->it_disposition, LUSTRE_IT(it)->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 == 1 || reply_buffers == 3 ||
- reply_buffers == 5 || reply_buffers == 7 ||
- reply_buffers == 8);
- if (reply_buffers >= 3) {
+ if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
struct mds_body *body;
- body = lustre_swab_repbuf(req, 1, sizeof (*body),
- lustre_swab_mds_body);
+ body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mds_body);
if (body == NULL) {
CERROR ("Can't swab mds_body\n");
RETURN (-EPROTO);
}
- if ((it->it_op & IT_OPEN) && (LUSTRE_IT(it)->it_status >= 0))
- mdc_store_inode_generation(exp, req, MDS_REQ_INTENT_REC_OFF, 1);
+ /* 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 */
+ if (it_disposition(it, DISP_OPEN_OPEN) &&
+ !it_open_error(DISP_OPEN_OPEN, it))
+ mdc_set_open_replay_data(NULL, req);
if ((body->valid & 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, 2, body->eadatasize,
- NULL);
+ eadata = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF + 1,
+ body->eadatasize, NULL);
if (eadata == NULL) {
CERROR ("Missing/short eadata\n");
RETURN (-EPROTO);
}
+ if (body->valid & OBD_MD_FLMODEASIZE) {
+ struct obd_device *obddev = class_exp2obd(exp);
+
+ 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. */
if (it->it_op & IT_OPEN) {
- void *replayea;
-
- replayea = lustre_msg_buf(req->rq_reqmsg,
- MDS_REQ_INTENT_REC_OFF + 2,
- body->eadatasize);
- LASSERT(replayea);
- memcpy(replayea, eadata, body->eadatasize);
-
- LASSERT(req->rq_reqmsg->bufcount == 6 ||
- req->rq_reqmsg->bufcount == 7);
- req->rq_reqmsg->buflens[5] = body->eadatasize;
- /* If this isn't the last buffer, we might
- * have to shift other data around. */
+ int offset = DLM_INTENT_REC_OFF + 2;
+ void *lmm;
+
+ if (lustre_msg_buflen(req->rq_reqmsg, offset) <
+ body->eadatasize)
+ mdc_realloc_openmsg(req, body);
+
+ lmm = lustre_msg_buf(req->rq_reqmsg, offset,
+ body->eadatasize);
+ if (lmm)
+ memcpy(lmm, eadata, body->eadatasize);
}
}
+ }
+
+ RETURN(rc);
+}
- /* just swab out capa to check here, and for future use */
- if (body->valid & OBD_MD_CAPA) {
- struct lustre_capa *capa;
+/* 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, struct ldlm_enqueue_info *einfo,
+ struct lookup_intent *it, struct mdc_op_data *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} };
+ ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
+ int flags = extra_lock_flags | LDLM_FL_HAS_INTENT;
+ int rc;
+ ENTRY;
- LASSERT(it->it_op & IT_OPEN);
- capa = lustre_swab_repbuf(req, 7, sizeof(*capa),
- lustre_swab_lustre_capa);
- LASSERT(capa);
+ LASSERTF(einfo->ei_type == LDLM_IBITS,"lock type %d\n", einfo->ei_type);
+ if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+
+ if (it->it_op & IT_OPEN) {
+ req = mdc_intent_open_pack(exp, it, data, lmm, lmmsize);
+ if (it->it_flags & O_JOIN_FILE) {
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
}
+ } else if (it->it_op & IT_UNLINK) {
+ req = mdc_intent_unlink_pack(exp, it, data);
+ } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
+ req = mdc_intent_lookup_pack(exp, it, data);
+ } else if (it->it_op == IT_READDIR) {
+ req = mdc_intent_readdir_pack(exp);
+ } else {
+ CERROR("bad it_op %x\n", it->it_op);
+ RETURN(-EINVAL);
}
+ if (!req)
+ RETURN(-ENOMEM);
+
+ /* 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);
+ 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);
+
+ rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
+
RETURN(rc);
}
EXPORT_SYMBOL(mdc_enqueue);
+int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
+ struct ll_fid *fid)
+{
+ /* 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 ={fid->id, fid->generation}};
+ struct lustre_handle lockh;
+ ldlm_policy_data_t policy;
+ 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. */
+ policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
+ MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP :
+ MDS_INODELOCK_LOOKUP;
+
+ mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
+ LDLM_FL_BLOCK_GRANTED, &res_id, 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 !!mode;
+}
+EXPORT_SYMBOL(mdc_revalidate_lock);
+
+static int mdc_finish_intent_lock(struct obd_export *exp,
+ struct ptlrpc_request *req,
+ struct mdc_op_data *data,
+ struct lookup_intent *it,
+ struct lustre_handle *lockh)
+{
+ struct mds_body *mds_body;
+ struct lustre_handle old_lock;
+ struct ldlm_lock *lock;
+ int rc;
+ ENTRY;
+
+ LASSERT(req != NULL);
+ LASSERT(req != LP_POISON);
+ LASSERT(req->rq_repmsg != LP_POISON);
+
+ if (!it_disposition(it, DISP_IT_EXECD)) {
+ /* The server failed before it even started executing the
+ * intent, i.e. because it couldn't unpack the request. */
+ LASSERT(it->d.lustre.it_status != 0);
+ RETURN(it->d.lustre.it_status);
+ }
+ rc = it_open_error(DISP_IT_EXECD, it);
+ if (rc)
+ RETURN(rc);
+
+ mds_body = lustre_msg_buf(req->rq_repmsg, DLM_REPLY_REC_OFF,
+ sizeof(*mds_body));
+ /* mdc_enqueue checked */
+ LASSERT(mds_body != NULL);
+ /* mdc_enqueue swabbed */
+ LASSERT(lustre_rep_swabbed(req, DLM_REPLY_REC_OFF));
+
+ /* If we were revalidating a fid/name pair, mark the intent in
+ * case we fail and get called again from lookup */
+ if (data->fid2.id && (it->it_op != IT_GETATTR)) {
+ it_set_disposition(it, DISP_ENQ_COMPLETE);
+ /* Also: did we find the same inode? */
+ if (memcmp(&data->fid2, &mds_body->fid1, sizeof(data->fid2)))
+ RETURN(-ESTALE);
+ }
+
+ 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_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(req); /* 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(req); /* 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_it */
+ } else if (it->it_op == IT_OPEN) {
+ LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
+ } else {
+ LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
+ }
+
+ /* If we already have a matching lock, then cancel the new
+ * one. We have to set the data here instead of in
+ * mdc_enqueue, because we need to use the child's inode as
+ * the l_ast_data to match, and that's not available until
+ * intent_finish has performed the iget().) */
+ lock = ldlm_handle2lock(lockh);
+ if (lock) {
+ ldlm_policy_data_t policy = lock->l_policy_data;
+
+ LDLM_DEBUG(lock, "matching against this");
+ 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)) {
+ ldlm_lock_decref_and_cancel(lockh,
+ it->d.lustre.it_lock_mode);
+ memcpy(lockh, &old_lock, sizeof(old_lock));
+ memcpy(&it->d.lustre.it_lock_handle, lockh,
+ sizeof(*lockh));
+ }
+ }
+
+ CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
+ data->namelen, data->name, ldlm_it2str(it->it_op),
+ it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
+ RETURN(rc);
+}
+
/*
* 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
* ll_create/ll_open gets called.
*
* The server will return to us, in it_disposition, an indication of
- * exactly what d.lustre->it_status refers to.
+ * exactly what d.lustre.it_status refers to.
*
- * If DISP_OPEN_OPEN is set, then d.lustre->it_status refers to the open() call,
+ * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
* otherwise if DISP_OPEN_CREATE is set, then it status is the
* creation failure mode. In either case, one of DISP_LOOKUP_NEG or
* DISP_LOOKUP_POS will be set, indicating whether the child lookup
* was successful.
*
- * Else, if DISP_LOOKUP_EXECD then d.lustre->it_status is the rc of the
+ * 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 lustre_id *pid,
- const char *name, int len, void *lmm, int lmmsize,
- struct lustre_id *cid, struct lookup_intent *it,
+int mdc_intent_lock(struct obd_export *exp, struct mdc_op_data *op_data,
+ void *lmm, int lmmsize, struct lookup_intent *it,
int lookup_flags, struct ptlrpc_request **reqp,
- ldlm_blocking_callback cb_blocking)
+ ldlm_blocking_callback cb_blocking, int extra_lock_flags)
{
struct lustre_handle lockh;
- struct ptlrpc_request *request;
- struct mds_body *mds_body;
- struct lustre_handle old_lock;
- struct ldlm_lock *lock;
- int rc = 0;
+ int rc;
ENTRY;
- LASSERT(it);
- CDEBUG(D_DLMTRACE, "name: %*s in obj "DLID4", intent: %s flags %#o\n",
- len, name, OLID4(pid), ldlm_it2str(it->it_op), it->it_flags);
+ LASSERT(it);
- if (cid && (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
- it->it_op == IT_CHDIR)) {
- /* 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 = {id_fid(cid),
- id_group(cid)}};
- struct lustre_handle lockh;
- ldlm_policy_data_t policy;
- int mode;
-
- /* For the GETATTR case, ll_revalidate_it issues two separate
- queries - for LOOKUP and for UPDATE lock because it 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, mode,
- &lockh);
-
- if (!rc) {
- mode = LCK_CR;
- rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, mode,
- &lockh);
- }
- if (!rc) {
- mode = LCK_PW;
- rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, mode,
- &lockh);
- }
- if (!rc) {
- mode = LCK_CW;
- rc = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id,
- LDLM_IBITS, &policy, mode,
- &lockh);
- }
- if (rc) {
- if (ptlrpcs_check_cred(exp->exp_obd->u.cli.cl_import)) {
- /* return immediately if no credential held */
- ldlm_lock_decref(&lockh, mode);
- RETURN(-EACCES);
- }
- memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
- sizeof(lockh));
- LUSTRE_IT(it)->it_lock_mode = mode;
- }
+ CDEBUG(D_DLMTRACE,"name: %.*s in inode "LPU64", intent: %s flags %#o\n",
+ op_data->namelen, op_data->name, op_data->fid1.id,
+ ldlm_it2str(it->it_op), it->it_flags);
- /* Only return failure if it was not GETATTR by cid (from
- inode_revalidate) */
- if (rc || name)
+ if (op_data->fid2.id &&
+ (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR)) {
+ rc = mdc_revalidate_lock(exp, it, &op_data->fid2);
+ /* Only return failure if it was not GETATTR by cfid
+ (from inode_revalidate) */
+ if (rc || op_data->namelen != 0)
RETURN(rc);
}
* 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;
+ struct ldlm_enqueue_info einfo =
+ { LDLM_IBITS, it_to_lock_mode(it), cb_blocking,
+ ldlm_completion_ast, NULL, NULL };
- OBD_ALLOC(op_data, sizeof(*op_data));
- if (op_data == NULL)
- RETURN(-ENOMEM);
-
- mdc_id2mdc_data(op_data, pid, cid, name, len, 0);
-
- if (name != NULL)
- op_data->valid |= OBD_MD_FID;
-
- rc = mdc_enqueue(exp, LDLM_IBITS, it, it_to_lock_mode(it),
- op_data, &lockh, lmm, lmmsize,
- ldlm_completion_ast, cb_blocking, NULL);
- OBD_FREE(op_data, sizeof(*op_data));
+ rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh,
+ lmm, lmmsize, extra_lock_flags);
if (rc < 0)
RETURN(rc);
-
- memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh, sizeof(lockh));
+ memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
+ } else if (!op_data->fid2.id) {
+ /* 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 = LUSTRE_IT(it)->it_data;
- LASSERT(request != NULL);
-
- /* 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 && request->rq_replay &&
- (!it_disposition(it, DISP_OPEN_OPEN) || LUSTRE_IT(it)->it_status != 0))
- mdc_clear_replay_flag(request, LUSTRE_IT(it)->it_status);
-
- if (!it_disposition(it, DISP_IT_EXECD)) {
- /* The server failed before it even started executing the
- * intent, i.e. because it couldn't unpack the request. */
- LASSERT(LUSTRE_IT(it)->it_status != 0);
- RETURN(LUSTRE_IT(it)->it_status);
- }
- rc = it_open_error(DISP_IT_EXECD, it);
- 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 */
+ *reqp = it->d.lustre.it_data;
+ rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
- /* If we were revalidating a fid/name pair, mark the intent in case we
- * fail and get called again from lookup */
- if (cid != NULL) {
- 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 fid/mds/ino/gen if inode lives on another MDS -bzzz */
- if (!(lookup_flags & LOOKUP_COBD) && !id_equal(cid, &mds_body->id1))
- RETURN(-ESTALE);
- }
+ RETURN(rc);
+}
+EXPORT_SYMBOL(mdc_intent_lock);
- rc = it_open_error(DISP_LOOKUP_EXECD, it);
+static int mdc_intent_getattr_async_interpret(struct ptlrpc_request *req,
+ void *unused, int rc)
+{
+ struct mdc_enqueue_args *ma;
+ struct md_enqueue_info *minfo;
+ struct ldlm_enqueue_info *einfo;
+ struct obd_export *exp;
+ struct lookup_intent *it;
+ struct lustre_handle *lockh;
+ struct obd_device *obddev;
+ int flags = LDLM_FL_HAS_INTENT;
+ ENTRY;
+
+ ma = (struct mdc_enqueue_args *)&req->rq_async_args;
+ minfo = ma->ma_mi;
+ einfo = ma->ma_ei;
+
+ exp = minfo->mi_exp;
+ it = &minfo->mi_it;
+ lockh = &minfo->mi_lockh;
+
+ obddev = class_exp2obd(exp);
+
+ mdc_exit_request(&obddev->u.cli);
+
+ rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
+ &flags, NULL, 0, NULL, lockh, rc);
+
+ rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
if (rc)
- RETURN(rc);
+ GOTO(out, 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); /* balanced in ll_create_node */
- if (it_disposition(it, DISP_OPEN_OPEN) &&
- !it_open_error(DISP_OPEN_OPEN, it))
- ptlrpc_request_addref(request); /* balanced in ll_file_open */
+ memcpy(&it->d.lustre.it_lock_handle, lockh, sizeof(*lockh));
- if (it->it_op & IT_CREAT) {
- /* XXX this belongs in ll_create_it */
- } else if (it->it_op == IT_OPEN) {
- LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
- } else {
- LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR));
- }
+ rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
+ GOTO(out, rc);
+out:
+ OBD_FREE_PTR(einfo);
+ minfo->mi_cb(exp, req, minfo, rc);
- /*
- * if we already have a matching lock, then cancel the new one. We have
- * to set the data here instead of in mdc_enqueue, because we need to
- * use the child's inode as the l_ast_data to match, and that's not
- * available until intent_finish has performed the iget().)
- */
- lock = ldlm_handle2lock(&lockh);
- if (lock) {
- ldlm_policy_data_t policy = lock->l_policy_data;
- LDLM_DEBUG(lock, "matching against this");
- LDLM_LOCK_PUT(lock);
-
- LASSERTF(id_fid(&mds_body->id1) == lock->l_resource->lr_name.name[0] &&
- id_group(&mds_body->id1) == lock->l_resource->lr_name.name[1],
- "Invalid lock is returned to client. Lock res_is: %lu/%lu, "
- "response res_id: %lu/%lu.\n",
- (unsigned long)lock->l_resource->lr_name.name[0],
- (unsigned long)lock->l_resource->lr_name.name[1],
- (unsigned long)id_fid(&mds_body->id1),
- (unsigned long)id_group(&mds_body->id1));
-
- memcpy(&old_lock, &lockh, sizeof(lockh));
- if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
- LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
- ldlm_lock_decref_and_cancel(&lockh,
- LUSTRE_IT(it)->it_lock_mode);
- memcpy(&lockh, &old_lock, sizeof(old_lock));
- memcpy(&LUSTRE_IT(it)->it_lock_handle, &lockh,
- sizeof(lockh));
- }
+ return 0;
+}
+
+int mdc_intent_getattr_async(struct obd_export *exp,
+ struct md_enqueue_info *minfo,
+ struct ldlm_enqueue_info *einfo)
+{
+ struct mdc_op_data *op_data = &minfo->mi_data;
+ struct lookup_intent *it = &minfo->mi_it;
+ struct ptlrpc_request *req;
+ struct obd_device *obddev = class_exp2obd(exp);
+ struct ldlm_res_id res_id = {
+ .name = {op_data->fid1.id,
+ op_data->fid1.generation}
+ };
+ ldlm_policy_data_t policy = {
+ .l_inodebits = { MDS_INODELOCK_LOOKUP }
+ };
+ struct mdc_enqueue_args *aa;
+ int rc;
+ int flags = LDLM_FL_HAS_INTENT;
+ ENTRY;
+
+ CDEBUG(D_DLMTRACE,"name: %.*s in inode "LPU64", intent: %s flags %#o\n",
+ op_data->namelen, op_data->name, op_data->fid1.id,
+ ldlm_it2str(it->it_op), it->it_flags);
+
+ req = mdc_intent_lookup_pack(exp, it, op_data);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ mdc_enter_request(&obddev->u.cli);
+ rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, &policy, &flags, NULL,
+ 0, NULL, &minfo->mi_lockh, 1);
+ if (rc < 0) {
+ mdc_exit_request(&obddev->u.cli);
+ RETURN(rc);
}
- CDEBUG(D_DENTRY, "D_IT dentry %*s intent: %s status %d disp %x rc %d\n",
- len, name, ldlm_it2str(it->it_op), LUSTRE_IT(it)->it_status,
- LUSTRE_IT(it)->it_disposition, rc);
- RETURN(rc);
+ CLASSERT(sizeof(*aa) < sizeof(req->rq_async_args));
+ aa = (struct mdc_enqueue_args *)&req->rq_async_args;
+ aa->ma_mi = minfo;
+ aa->ma_ei = einfo;
+ req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
+ ptlrpcd_add_req(req);
+
+ RETURN(0);
}
-EXPORT_SYMBOL(mdc_intent_lock);
+EXPORT_SYMBOL(mdc_intent_getattr_async);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff