Whamcloud - gitweb
1)add .snap namespace to smfs
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 8f7b3b6..09a709c 100644 (file)
@@ -36,6 +36,8 @@
 #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"
 
@@ -76,7 +78,8 @@ 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))
+        else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
+                              IT_CHDIR))
                 return LCK_PR;
 
         LBUG();
@@ -120,7 +123,7 @@ int it_open_error(int phase, struct lookup_intent *it)
 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;
@@ -128,7 +131,7 @@ void mdc_set_lock_data(__u64 *l, void *data)
 
         if (!*l) {
                 EXIT;
-                return;
+                return 0;
         }
 
         lock = ldlm_handle2lock(lockh);
@@ -152,6 +155,7 @@ void mdc_set_lock_data(__u64 *l, void *data)
         LDLM_LOCK_PUT(lock);
 
         EXIT;
+        return 0;
 }
 EXPORT_SYMBOL(mdc_set_lock_data);
 
@@ -171,8 +175,6 @@ int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
         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,
@@ -230,38 +232,21 @@ int mdc_enqueue(struct obd_export *exp,
                 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);
 
@@ -270,21 +255,39 @@ int mdc_enqueue(struct obd_export *exp,
                 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;
+                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);
         }
@@ -346,6 +349,9 @@ int mdc_enqueue(struct obd_export *exp,
                 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) {
@@ -359,10 +365,8 @@ int mdc_enqueue(struct obd_export *exp,
                 }
 
                 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) {
@@ -370,10 +374,17 @@ int mdc_enqueue(struct obd_export *exp,
                                 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. */
                         }
                 }
         }
@@ -411,9 +422,9 @@ EXPORT_SYMBOL(mdc_enqueue);
  */
 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,
+                    void *lmm, int lmmsize, struct ll_fid *cfid,
+                    struct lookup_intent *it, int lookup_flags,
+                    struct ptlrpc_request **reqp,
                     ldlm_blocking_callback cb_blocking)
 {
         struct lustre_handle lockh;
@@ -426,9 +437,10 @@ int mdc_intent_lock(struct obd_export *exp, struct ll_uctxt *uctxt,
         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. */
@@ -505,7 +517,12 @@ int mdc_intent_lock(struct obd_export *exp, struct ll_uctxt *uctxt,
         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);
         }
 
@@ -542,7 +559,7 @@ int mdc_intent_lock(struct obd_export *exp, struct ll_uctxt *uctxt,
         } 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