long code, code1;
cm_scache_t *scp = vscp;
afs_int32 nbytes;
+#ifdef AFS_LARGEFILES
+ afs_int32 save_nbytes;
+#endif
long temp;
AFSFetchStatus outStatus;
AFSStoreStatus inStatus;
lock_ReleaseWrite(&scp->rw);
/* now we're ready to do the store operation */
+#ifdef AFS_LARGEFILES
+ save_nbytes = nbytes;
+#endif
do {
code = cm_ConnFromFID(&scp->fid, userp, reqp, &connp);
if (code)
/* write the data from the the list of buffers */
qdp = NULL;
while(nbytes > 0) {
- if (qdp == NULL)
+ afs_uint32 buf_offset;
+ if (qdp == NULL) {
qdp = biod.bufListEndp;
- else
+ buf_offset = offsetp->LowPart % cm_data.buf_blockSize;
+ } else {
qdp = (osi_queueData_t *) osi_QPrev(&qdp->q);
+ buf_offset = 0;
+ }
osi_assertx(qdp != NULL, "null osi_queueData_t");
bufp = osi_GetQData(qdp);
- bufferp = bufp->datap;
+ bufferp = bufp->datap + buf_offset;
wbytes = nbytes;
- if (wbytes > cm_data.buf_blockSize)
- wbytes = cm_data.buf_blockSize;
+ if (wbytes > cm_data.buf_blockSize - buf_offset)
+ wbytes = cm_data.buf_blockSize - buf_offset;
/* write out wbytes of data from bufferp */
temp = rx_Write(rxcallp, bufferp, wbytes);
if (temp != wbytes) {
osi_Log3(afsd_logp, "rx_Write failed bp 0x%p, %d != %d",bufp,temp,wbytes);
- code = -1;
+ code = (rxcallp->error < 0) ? rxcallp->error : RX_PROTOCOL_ERROR;
break;
} else {
osi_Log2(afsd_logp, "rx_Write succeeded bp 0x%p, %d",bufp,temp);
#ifdef AFS_LARGEFILES
if ((code == RXGEN_OPCODE || code1 == RXGEN_OPCODE) && SERVERHAS64BIT(connp)) {
SET_SERVERHASNO64BIT(connp);
+ qdp = NULL;
+ nbytes = save_nbytes;
goto retry;
}
#endif
/* Prefer StoreData error over rx_EndCall error */
- if (code == 0 && code1 != 0)
+ if (code1 != 0)
code = code1;
} while (cm_Analyze(connp, userp, reqp, &scp->fid, &volSync, NULL, NULL, code));
/* release a bulk I/O structure that was setup by cm_SetupFetchBIOD or by
* cm_SetupStoreBIOD
*/
-void cm_ReleaseBIOD(cm_bulkIO_t *biop, int isStore, int failed, int scp_locked)
+void cm_ReleaseBIOD(cm_bulkIO_t *biop, int isStore, long code, int scp_locked)
{
cm_scache_t *scp; /* do not release; not held in biop */
cm_buf_t *bufp;
osi_Log2(afsd_logp, "cm_ReleaseBIOD Waking [scp 0x%p] bp 0x%p", scp, bufp);
osi_Wakeup((LONG_PTR) bufp);
}
- if (failed)
+ if (code) {
bufp->flags &= ~CM_BUF_WRITING;
- else {
+ switch (code) {
+ case CM_ERROR_NOSUCHFILE:
+ case CM_ERROR_BADFD:
+ case CM_ERROR_NOACCESS:
+ case CM_ERROR_QUOTA:
+ case CM_ERROR_SPACE:
+ case CM_ERROR_TOOBIG:
+ case CM_ERROR_READONLY:
+ case CM_ERROR_NOSUCHPATH:
+ /*
+ * Apply the fatal error to this buffer.
+ */
+ bufp->flags &= ~CM_BUF_DIRTY;
+ bufp->flags |= CM_BUF_ERROR;
+ bufp->dirty_offset = 0;
+ bufp->dirty_length = 0;
+ bufp->error = code;
+ bufp->dataVersion = CM_BUF_VERSION_BAD;
+ bufp->dirtyCounter++;
+ break;
+ case CM_ERROR_TIMEDOUT:
+ case CM_ERROR_ALLDOWN:
+ case CM_ERROR_ALLBUSY:
+ case CM_ERROR_ALLOFFLINE:
+ case CM_ERROR_CLOCKSKEW:
+ default:
+ /* do not mark the buffer in error state but do
+ * not attempt to complete the rest either.
+ */
+ break;
+ }
+ } else {
bufp->flags &= ~(CM_BUF_WRITING | CM_BUF_DIRTY);
bufp->dirty_offset = bufp->dirty_length = 0;
}
biop->bufListEndp = NULL;
}
+static int
+cm_CloneStatus(cm_scache_t *scp, cm_user_t *userp, int scp_locked,
+ AFSFetchStatus *afsStatusp, AFSVolSync *volSyncp)
+{
+ // setup the status based upon the scp data
+ afsStatusp->InterfaceVersion = 0x1;
+ switch (scp->fileType) {
+ case CM_SCACHETYPE_FILE:
+ afsStatusp->FileType = File;
+ break;
+ case CM_SCACHETYPE_DIRECTORY:
+ afsStatusp->FileType = Directory;
+ break;
+ case CM_SCACHETYPE_MOUNTPOINT:
+ afsStatusp->FileType = SymbolicLink;
+ break;
+ case CM_SCACHETYPE_SYMLINK:
+ case CM_SCACHETYPE_DFSLINK:
+ afsStatusp->FileType = SymbolicLink;
+ break;
+ default:
+ afsStatusp->FileType = -1; /* an invalid value */
+ }
+ afsStatusp->LinkCount = scp->linkCount;
+ afsStatusp->Length = scp->length.LowPart;
+ afsStatusp->DataVersion = (afs_uint32)(scp->dataVersion & MAX_AFS_UINT32);
+ afsStatusp->Author = 0x1;
+ afsStatusp->Owner = scp->owner;
+ if (!scp_locked) {
+ lock_ObtainWrite(&scp->rw);
+ scp_locked = 1;
+ }
+ if (cm_FindACLCache(scp, userp, &afsStatusp->CallerAccess))
+ afsStatusp->CallerAccess = scp->anyAccess;
+ afsStatusp->AnonymousAccess = scp->anyAccess;
+ afsStatusp->UnixModeBits = scp->unixModeBits;
+ afsStatusp->ParentVnode = scp->parentVnode;
+ afsStatusp->ParentUnique = scp->parentUnique;
+ afsStatusp->ResidencyMask = 0;
+ afsStatusp->ClientModTime = scp->clientModTime;
+ afsStatusp->ServerModTime = scp->serverModTime;
+ afsStatusp->Group = scp->group;
+ afsStatusp->SyncCounter = 0;
+ afsStatusp->dataVersionHigh = (afs_uint32)(scp->dataVersion >> 32);
+ afsStatusp->lockCount = 0;
+ afsStatusp->Length_hi = scp->length.HighPart;
+ afsStatusp->errorCode = 0;
+
+ memset(volSyncp, 0, sizeof(AFSVolSync));
+ volSyncp->spare1 = scp->lastUpdateRO;
+
+ return scp_locked;
+}
+
/* Fetch a buffer. Called with scp locked.
* The scp is locked on return.
*/
long cm_GetBuffer(cm_scache_t *scp, cm_buf_t *bufp, int *cpffp, cm_user_t *userp,
cm_req_t *reqp)
{
- long code, code1;
- afs_int32 nbytes; /* bytes in transfer */
- afs_int32 nbytes_hi = 0; /* high-order 32 bits of bytes in transfer */
- afs_int64 length_found = 0;
+ long code=0, code1=0;
+ afs_uint32 nbytes; /* bytes in transfer */
+ afs_uint32 nbytes_hi = 0; /* high-order 32 bits of bytes in transfer */
+ afs_uint64 length_found = 0;
long rbytes; /* bytes in rx_Read call */
long temp;
AFSFetchStatus afsStatus;
afs_int32 t1,t2;
int require_64bit_ops = 0;
int call_was_64bit = 0;
+ int fs_fetchdata_offset_bug = 0;
+ int first_read = 1;
+ int scp_locked = 1;
/* now, the buffer may or may not be filled with good data (buf_GetNew
* drops lots of locks, and may indeed return a properly initialized
}
lock_ReleaseWrite(&scp->rw);
+ scp_locked = 0;
if (LargeIntegerGreaterThan(LargeIntegerAdd(biod.offset,
ConvertLongToLargeInteger(biod.length)),
afsStatus.lockCount = 0;
afsStatus.Length_hi = 0;
afsStatus.errorCode = 0;
-
+ memset(&volSync, 0, sizeof(volSync));
+
// once we're done setting up the status info,
// we just fill the buffer pages with fakedata
// from cm_FakeRootDir. Extra pages are set to
#endif /* AFS_FREELANCE_CLIENT */
- /* now make the call */
+ /*
+ * if the requested offset is greater than the file length,
+ * the file server will return zero bytes of data and the
+ * current status for the file which we already have since
+ * we have just obtained a callback. Instead, we can avoid
+ * the network round trip by allocating zeroed buffers and
+ * faking the status info.
+ */
+ if (biod.offset.QuadPart >= scp->length.QuadPart) {
+ osi_Log5(afsd_logp, "SKIP FetchData64 scp 0x%p, off 0x%x:%08x > length 0x%x:%08x",
+ scp, biod.offset.HighPart, biod.offset.LowPart,
+ scp->length.HighPart, scp->length.LowPart);
+
+ /* Clone the current status info */
+ scp_locked = cm_CloneStatus(scp, userp, scp_locked, &afsStatus, &volSync);
+
+ /* status info complete, fill pages with zeros */
+ for (qdp = biod.bufListEndp;
+ qdp;
+ qdp = (osi_queueData_t *) osi_QPrev(&qdp->q)) {
+ tbufp = osi_GetQData(qdp);
+ bufferp=tbufp->datap;
+ memset(bufferp, 0, cm_data.buf_blockSize);
+ }
+
+ /* no need to contact the file server */
+ goto fetchingcompleted;
+ }
+
+ if (scp_locked) {
+ lock_ReleaseWrite(&scp->rw);
+ scp_locked = 0;
+ }
+
+ /* now make the call */
do {
code = cm_ConnFromFID(&scp->fid, userp, reqp, &connp);
if (code)
if (temp == sizeof(afs_int32)) {
nbytes = ntohl(nbytes);
FillInt64(length_found, nbytes_hi, nbytes);
- if (length_found > biod.length)
- code = (rxcallp->error < 0) ? rxcallp->error : -1;
+ if (length_found > biod.length) {
+ /*
+ * prior to 1.4.12 and 1.5.65 the file server would return
+ * (filesize - offset) if the requested offset was greater than
+ * the filesize. The correct return value would have been zero.
+ * Force a retry by returning an RX_PROTOCOL_ERROR. If the cause
+ * is a race between two RPCs issues by this cache manager, the
+ * correct thing will happen the second time.
+ */
+ osi_Log0(afsd_logp, "cm_GetBuffer length_found > biod.length");
+ fs_fetchdata_offset_bug = 1;
+ }
} else {
- code = (rxcallp->error < 0) ? rxcallp->error : -1;
+ osi_Log1(afsd_logp, "cm_GetBuffer rx_Read32 returns %d != 4", temp);
+ code = (rxcallp->error < 0) ? rxcallp->error : RX_PROTOCOL_ERROR;
}
}
/* for the moment, nbytes_hi will always be 0 if code == 0
if (code == 0) {
temp = rx_Read32(rxcallp, &nbytes);
if (temp == sizeof(afs_int32)) {
- nbytes = ntohl(nbytes);
- if (nbytes > biod.length)
- code = (rxcallp->error < 0) ? rxcallp->error : -1;
+ length_found = ntohl(nbytes);
+ if (length_found > biod.length) {
+ /*
+ * prior to 1.4.12 and 1.5.65 the file server would return
+ * (filesize - offset) if the requested offset was greater than
+ * the filesize. The correct return value would have been zero.
+ * Force a retry by returning an RX_PROTOCOL_ERROR. If the cause
+ * is a race between two RPCs issues by this cache manager, the
+ * correct thing will happen the second time.
+ */
+ osi_Log0(afsd_logp, "cm_GetBuffer length_found > biod.length");
+ fs_fetchdata_offset_bug = 1;
+ }
+ }
+ else {
+ osi_Log1(afsd_logp, "cm_GetBuffer rx_Read32 returns %d != 4", temp);
+ code = (rxcallp->error < 0) ? rxcallp->error : RX_PROTOCOL_ERROR;
}
- else
- code = (rxcallp->error < 0) ? rxcallp->error : -1;
}
#endif
}
else
bufferp = NULL;
- /* fill nbytes of data from the pipe into the pages.
+ /* fill length_found of data from the pipe into the pages.
* When we stop, qdp will point at the last page we're
* dealing with, and bufferp will tell us where we
* stopped. We'll need this info below when we clear
* the remainder of the last page out (and potentially
* clear later pages out, if we fetch past EOF).
*/
- while (nbytes > 0) {
+ while (length_found > 0) {
/* assert that there are still more buffers;
- * our check above for nbytes being less than
+ * our check above for length_found being less than
* biod.length should ensure this.
*/
osi_assertx(bufferp != NULL, "null cm_buf_t");
/* read rbytes of data */
- rbytes = (nbytes > cm_data.buf_blockSize? cm_data.buf_blockSize : nbytes);
+ rbytes = (afs_uint32)(length_found > cm_data.buf_blockSize ? cm_data.buf_blockSize : length_found);
temp = rx_Read(rxcallp, bufferp, rbytes);
if (temp < rbytes) {
- code = (rxcallp->error < 0) ? rxcallp->error : -1;
+ /*
+ * If the file server returned (filesize - offset),
+ * then the first rx_Read will return zero octets of data.
+ * If it does, do not treat it as an error. Correct the
+ * length_found and continue as if the file server said
+ * it was sending us zero octets of data.
+ */
+ if (fs_fetchdata_offset_bug && first_read)
+ length_found = 0;
+ else
+ code = (rxcallp->error < 0) ? rxcallp->error : RX_PROTOCOL_ERROR;
break;
}
+ first_read = 0;
/* allow read-while-fetching.
* if this is the last buffer, clear the
lock_ReleaseWrite(&scp->rw);
/* and adjust counters */
- nbytes -= temp;
+ length_found -= temp;
/* and move to the next buffer */
- if (nbytes != 0) {
+ if (length_found != 0) {
qdp = (osi_queueData_t *) osi_QPrev(&qdp->q);
if (qdp) {
tbufp = osi_GetQData(qdp);
if (code1 == RXKADUNKNOWNKEY)
osi_Log0(afsd_logp, "CALL EndCall returns RXKADUNKNOWNKEY");
+ /* If we are avoiding a file server bug, ignore the error state */
+ if (fs_fetchdata_offset_bug && first_read && length_found == 0 && code == -451) {
+ /* Clone the current status info and clear the error state */
+ scp_locked = cm_CloneStatus(scp, userp, scp_locked, &afsStatus, &volSync);
+ if (scp_locked) {
+ lock_ReleaseWrite(&scp->rw);
+ scp_locked = 0;
+ }
+ code = 0;
/* Prefer the error value from FetchData over rx_EndCall */
- if (code == 0 && code1 != 0)
+ } else if (code == 0 && code1 != 0)
code = code1;
osi_Log0(afsd_logp, "CALL FetchData DONE");
fetchingcompleted:
code = cm_MapRPCError(code, reqp);
- lock_ObtainWrite(&scp->rw);
+ if (!scp_locked)
+ lock_ObtainWrite(&scp->rw);
/* we know that no one else has changed the buffer, since we still have
* the fetching flag on the buffers, and we have the scp locked again.