#include <linux/obd_class.h>
#include <linux/lustre_mds.h>
#include <linux/lustre_dlm.h>
-#include <linux/lustre_snap.h>
#include <linux/lprocfs_status.h>
#include "mdc_internal.h"
}
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)
+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)
{
LASSERT(data);
- LASSERT(ctxt);
LASSERT(f1);
- data->ctxt = *ctxt;
- data->fid1 = *f1;
+ data->id1 = *f1;
if (f2)
- data->fid2 = *f2;
- else
- memset(&data->fid2, 0, sizeof(data->fid2));
+ data->id2 = *f2;
+
+ data->valid = 0;
data->name = name;
data->namelen = namelen;
data->create_mode = mode;
"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);
+ old_inode->i_state, new_inode, new_inode->i_ino,
+ new_inode->i_generation);
}
#endif
lock->l_ast_data = data;
}
EXPORT_SYMBOL(mdc_set_lock_data);
-int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
+int mdc_change_cbdata(struct obd_export *exp, struct lustre_id *id,
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] = id_fid(id);
+ res_id.name[1] = id_group(id);
- ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace, &res_id, it,
- data);
+ ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace,
+ &res_id, it, data);
EXIT;
return 0;
}
-#if CONFIG_SNAPFS
-int mdc_set_clone_info(struct obd_import *climp, struct lustre_msg *msg,
- int offset)
-{
- struct client_obd *cli_obd = &climp->imp_obd->u.cli;
- struct clonefs_info *cl_info;
- ENTRY;
-
- cl_info = (struct clonefs_info *)lustre_msg_buf(msg, offset,
- sizeof (*cl_info));
- memcpy(cl_info, cli_obd->cl_clone_info, sizeof(*cl_info));
- RETURN(0);
-}
-#endif
-
/* 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,
void *cb_data)
{
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);
- struct ldlm_res_id res_id =
- { .name = {data->fid1.id, data->fid1.generation} };
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;
+ struct ldlm_intent *lit;
+ struct ldlm_request *lockreq;
+ struct ldlm_reply *dlm_rep;
+ int reqsize[6] = {[MDS_REQ_SECDESC_OFF] = 0,
+ [MDS_REQ_INTENT_LOCKREQ_OFF] = sizeof(*lockreq),
+ [MDS_REQ_INTENT_IT_OFF] = sizeof(*lit)};
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;
+ int req_buffers = 3, reply_buffers = 0;
+ int rc, flags = LDLM_FL_HAS_INTENT;
void *eadata;
unsigned long irqflags;
- int reply_buffers = 0;
ENTRY;
// LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
// ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
+ reqsize[0] = mdc_get_secdesc_size();
+
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);
+ reqsize[req_buffers++] = sizeof(struct mds_rec_create);
+ reqsize[req_buffers++] = data->namelen + 1;
+ reqsize[req_buffers++] = obddev->u.cli.cl_max_mds_easize;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
+ LDLM_ENQUEUE, req_buffers, reqsize, NULL);
if (!req)
RETURN(-ENOMEM);
spin_unlock_irqrestore (&req->rq_lock, irqflags);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_INTENT_IT_OFF,
+ sizeof (*lit));
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_open_pack(req->rq_reqmsg, 2, data, it->it_create_mode, 0,
- it->it_flags, lmm, lmmsize);
+ mdc_open_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF, 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);
} else if (it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR)) {
- int valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
- size[2] = sizeof(struct mds_body);
- size[3] = data->namelen + 1;
+ __u64 valid = data->valid | OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
+
+ 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;
-#if CONFIG_SNAPFS
- size[4] = sizeof(struct clonefs_info);
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 5,
- size, NULL);
-#else
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 4,
- size, NULL);
-#endif
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
+ LDLM_ENQUEUE, req_buffers, reqsize, NULL);
+
if (!req)
RETURN(-ENOMEM);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_INTENT_IT_OFF,
+ sizeof (*lit));
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_getattr_pack(req->rq_reqmsg, valid, 2, it->it_flags, data);
-#if CONFIG_SNAPFS
- mdc_set_clone_info(class_exp2cliimp(exp), req->rq_reqmsg, 4);
-#endif
+ mdc_getattr_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
+ valid, it->it_flags, data);
+
/* get ready for the reply */
reply_buffers = 3;
req->rq_replen = lustre_msg_size(3, repsize);
} else if (it->it_op == IT_READDIR) {
policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-#if CONFIG_SNAPFS
- size[1] = sizeof(struct clonefs_info);
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 2,
- size, NULL);
-#else
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 1,
- size, NULL);
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
+ LDLM_ENQUEUE, 2, reqsize, NULL);
-#endif
if (!req)
RETURN(-ENOMEM);
-#if CONFIG_SNAPFS
- mdc_set_clone_info(class_exp2cliimp(exp), req->rq_reqmsg, 1);
-#endif
/* get ready for the reply */
reply_buffers = 1;
req->rq_replen = lustre_msg_size(1, repsize);
} else if (it->it_op == IT_UNLINK) {
- size[2] = sizeof(struct mds_body);
+ reqsize[req_buffers++] = sizeof(struct mds_body);
policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 3,
- size, NULL);
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_DLM_VERSION,
+ LDLM_ENQUEUE, req_buffers, reqsize, NULL);
if (!req)
RETURN(-ENOMEM);
/* pack the intended request */
- mdc_getattr_pack(req->rq_reqmsg, 0, 2, 0, data);
+ mdc_getattr_pack(req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
+ 0, 0, data);
/* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
+ lit = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_INTENT_IT_OFF,
+ sizeof (*lit));
lit->opc = (__u64)it->it_op;
/* get ready for the reply */
LBUG();
RETURN(-EINVAL);
}
+
+ mdc_pack_secdesc(req, reqsize[0]);
+
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,
/* 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,
+ MDS_REQ_INTENT_LOCKREQ_OFF,
+ sizeof (*lockreq));
lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
}
struct mds_body *body;
body = lustre_swab_repbuf(req, 1, sizeof (*body),
- lustre_swab_mds_body);
+ lustre_swab_mds_body);
if (body == NULL) {
CERROR ("Can't swab mds_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 */
+ /* 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 (eadata == NULL) {
RETURN (-EPROTO);
}
if (it->it_op & IT_OPEN) {
- replayea = lustre_msg_buf(req->rq_reqmsg, 4,
- obddev->u.cli.cl_max_mds_easize);
+ 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->buflens[5] = body->eadatasize;
+ /* If this isn't the last buffer, we might
+ * have to shift other data around. */
}
}
}
* 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,
+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 lookup_flags, struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking)
{
struct lustre_handle lockh;
struct ptlrpc_request *request;
- int rc = 0;
struct mds_body *mds_body;
struct lustre_handle old_lock;
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 in obj "DLID4", intent: %s flags %#o\n",
+ len, name, OLID4(pid), ldlm_it2str(it->it_op), it->it_flags);
- if (cfid && (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR ||
- it->it_op == IT_CHDIR)) {
+ 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 = {cfid->id,
- cfid->generation}};
+ struct ldlm_res_id res_id = {.name = {id_fid(cid),
+ id_group(cid)}};
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
+ 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;
+ 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,
sizeof(lockh));
it->d.lustre.it_lock_mode = mode;
}
- RETURN(rc);
+
+ /* Only return failure if it was not GETATTR by cid (from
+ inode_revalidate) */
+ if (rc || name)
+ RETURN(rc);
}
/* lookup_it may be called only after revalidate_it has run, because
* 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);
+
+ mdc_id2mdc_data(&op_data, pid, cid, name, len, 0);
+
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+ /*
+ * This is optimization. Now fid will not be obtained from
+ * server if client inode already exists. This flag is set in
+ * ll_revalidate_it() if it finds that passed dentry contains
+ * inode.
+ */
+ if (!(it->d.lustre.it_int_flags && LL_IT_EXIST)) {
+#endif
+ /*
+ * if we get inode by name (ll_lookup_it() case), we
+ * always should ask for fid, as we will not be able to
+ * take locks, revalidate dentry, etc. later with
+ * invalid fid in inode.
+ */
+ if (cid == NULL && name != NULL)
+ op_data.valid |= OBD_MD_FID;
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+ }
+#endif
rc = mdc_enqueue(exp, LDLM_IBITS, it, it_to_lock_mode(it),
&op_data, &lockh, lmm, lmmsize,
ldlm_completion_ast, cb_blocking, NULL);
if (rc < 0)
RETURN(rc);
+
memcpy(&it->d.lustre.it_lock_handle, &lockh, sizeof(lockh));
}
request = *reqp = it->d.lustre.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) {
+ 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);
+ }
+ }
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. */
RETURN(rc);
mds_body = lustre_msg_buf(request->rq_repmsg, 1, sizeof(*mds_body));
- LASSERT(mds_body != NULL); /* mdc_enqueue checked */
+ LASSERT(mds_body != NULL); /* mdc_enqueue checked */
LASSERT_REPSWABBED(request, 1); /* mdc_enqueue swabbed */
- /* 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 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 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)
+
+ /* 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 (!id_equal(cid, &mds_body->id1))
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
+ /*
+ * 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);
+ 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);
+ ptlrpc_request_addref(request); /* balanced in ll_file_open */
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 {
LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR));
}
- /* 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().) */
+ /*
+ * 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)) {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff