#include <linux/obd_class.h>
#include <linux/lustre_mds.h>
#include <linux/lustre_dlm.h>
+//#include <linux/lustre_smfs.h>
+//#include <linux/lustre_snap.h>
#include <linux/lprocfs_status.h>
#include "mdc_internal.h"
/* 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))
+ else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
+ IT_CHDIR))
return LCK_PR;
LBUG();
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 *l, void *data)
{
struct ldlm_lock *lock;
struct lustre_handle *lockh = (struct lustre_handle *)l;
if (!*l) {
EXIT;
- return;
+ return 0;
}
lock = ldlm_handle2lock(lockh);
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;
LDLM_LOCK_PUT(lock);
EXIT;
+ return 0;
}
EXPORT_SYMBOL(mdc_set_lock_data);
ldlm_change_cbdata(class_exp2obd(exp)->obd_namespace, &res_id, it,
data);
+
EXIT;
return 0;
}
-
-
/* 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 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;
int repsize[4] = {sizeof(struct ldlm_reply),
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_open_pack(req, 2, data, it->it_create_mode, 0,
+ mdc_open_pack(req->rq_reqmsg, 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);
- } 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);
- if (!req)
- RETURN(-ENOMEM);
-
- /* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, 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);
- } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
+ } 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;
+ 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);
+
if (!req)
RETURN(-ENOMEM);
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_getattr_pack(req, valid, 2, it->it_flags, data);
+ mdc_getattr_pack(req->rq_reqmsg, valid, 2, 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;
req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 1,
size, NULL);
+
if (!req)
RETURN(-ENOMEM);
-
/* get ready for the reply */
reply_buffers = 1;
req->rq_replen = lustre_msg_size(1, repsize);
- } else {
+ } else if (it->it_op == IT_UNLINK) {
+ size[2] = sizeof(struct mds_body);
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LDLM_ENQUEUE, 3,
+ size, NULL);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ /* pack the intended request */
+ mdc_getattr_pack(req->rq_reqmsg, 0, 2, 0, data);
+
+ /* pack the intent */
+ lit = lustre_msg_buf(req->rq_reqmsg, 1, 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);
}
-
mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
rc = ldlm_cli_enqueue(exp, req, obddev->obd_namespace, res_id,
- lock_type, NULL, lock_mode, &flags, cb_blocking,
+ 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);
it->d.lustre.it_lock_mode = lock_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);
+ }
+
+ DEBUG_REQ(D_RPCTRACE, req, "disposition: %x, status: %d",
+ 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) {
}
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, 4,
+ body->eadatasize);
LASSERT(replayea);
memcpy(replayea, eadata, body->eadatasize);
+
+ LASSERT(req->rq_reqmsg->bufcount == 5);
+ req->rq_reqmsg->buflens[4] = body->eadatasize;
+ /* If this isn't the last buffer, we might
+ * have to shift other data around. */
}
}
}
*/
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 flags,
+ void *lmm, int lmmsize, struct ll_fid *cfid,
+ struct lookup_intent *it, int lookup_flags,
struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking)
{
LASSERT(it);
CDEBUG(D_DLMTRACE, "name: %*s in %ld, intent: %s\n", len, name,
- (unsigned long) pfid->id, ldlm_it2str(it->it_op));
+ pfid ? (unsigned long) pfid->id : 0 , ldlm_it2str(it->it_op));
- if (cfid && (it->it_op == IT_LOOKUP || it->it_op == IT_GETATTR)) {
+ if (cfid && (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,
+ struct ldlm_res_id res_id ={.name = {cfid->id,
cfid->generation}};
struct lustre_handle lockh;
- int mode, flags = LDLM_FL_BLOCK_GRANTED;
-
+ 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, flags,
- &res_id, LDLM_PLAIN, NULL, LCK_PR, &lockh);
+ 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, flags,
- &res_id, LDLM_PLAIN, NULL, 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) {
- memcpy(&it->d.lustre.it_lock_handle, &lockh,
+ memcpy(&it->d.lustre.it_lock_handle, &lockh,
sizeof(lockh));
it->d.lustre.it_lock_mode = mode;
}
struct mdc_op_data op_data;
mdc_fid2mdc_op_data(&op_data, uctxt, pfid, cfid, name, len, 0);
- rc = mdc_enqueue(exp, LDLM_PLAIN, it, it_to_lock_mode(it),
+ 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)
if (cfid != NULL) {
it_set_disposition(it, DISP_ENQ_COMPLETE);
/* Also: did we find the same inode? */
- if (memcmp(cfid, &mds_body->fid1, sizeof(*cfid)))
+ /* 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)
RETURN(-ESTALE);
}
if (it->it_op & IT_OPEN) {
if (!it_disposition(it, DISP_OPEN_OPEN) ||
it->d.lustre.it_status != 0) {
- unsigned long flags;
+ unsigned long irqflags;
- spin_lock_irqsave(&request->rq_lock, flags);
+ spin_lock_irqsave(&request->rq_lock, irqflags);
request->rq_replay = 0;
- spin_unlock_irqrestore(&request->rq_lock, flags);
+ spin_unlock_irqrestore(&request->rq_lock, irqflags);
}
}
} else if (it->it_op == IT_OPEN) {
LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
} else {
- LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
+ LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_CHDIR));
}
/* If we already have a matching lock, then cancel the new
* 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_PLAIN, NULL, LCK_NL, &old_lock)) {
+ 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));
git://git.whamcloud.com / fs / lustre-release.git / blobdiff