Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

338 lines
13 KiB

This is a list of code "concerns". these have been removed from the code and
placed here so that the code will not contain untoward comments.
rx\read.c and write.c
tomm claims that the memory manager has been fixed before nt4.0 shipped
so that it is no longer necessary to synchronize flushers by acquiring and
releasing the pagingio resource. the is in the check for noncachedio when there
is a private cachemap. we should make sure that everyone else removes their
synchronization or we should put ours back.
rx\read.c and write.c
the pipe read/write paths should be completely separated in the wrapper
and the opportunity to do so should be presented in the mini.
rx\read.c
there is a place where we set status pipe empty; at that point we should
also backoff a la rdr1
rx\lock.c
there should be backoff logic
rx\strtstop.c
the call to rxforcenettablefinalization should be per-minirdr and should
be moved out of the iftest.
rx\strtstop
acquireing the netname table lock is was only required becauseof
precious servers. now that that's gone. we should remove the lock. but, it's
safe, of course.
mrx in general
we should review the throttling constants....seem pretty conservative to
me; we should consider being substantially more aggressive.
mrx
we should have precious server code.
rx\lockctrl.c
we should only allocate a lock structure if someone does a lock! same
with the oplock structure. ntfs does this. obviously, we would have a change in
fcb.h to go with this.
rx\lockctrl.c
the last little bit of getting the refcounts right was to account for
the fact that the fspdispatch does not remove a reference if pending is
returned. this is probably wrong....and we should map over to a scheme where the
fsd and fsp dispatches are the ones that do this instead of having every common
routine do it. this would be a major change but it would really make things more
consistent.
rx\lockctrl.c
it is stupid that we will have to lookup the lock again if it fails
(async completion path). we should ask for an fsrtl interface that removes a
lock that i provide.
rx\create.c
in RxCreateFromNetRoot, after the call to
rx\ntinit.c
in RxGetRegistryParameters, we should allocate space for Storage
instead. really what we should do is to implement and export a routine that both
allocates and reads a string........for Ulong no allocation should be necessary
rx\ntfsp.c
in fspdispatch and also addtoworkq, we use the wrapper's device object
for throttling instead of the minirdr's device object. this will change.
rx\ntfsp.c
in fspdispatch, if status_retry is returned then we will have already
completed the packet.
rx\ntfsp.c
prepostirp should be made pageable; also, i'm not too happy
rx\ntfsd.c
initialization of prefixtable should be in RxInitializeDispatchVectors
rx\ntfsd.c
RxFsdCommonDispatch in an excellent canididate for inlining
rx\ntfsd.c
we shouldn't be checking everywhere for null dispatch points....rather
we should fill them in with a good default
rx\ntfsd.c
we should remove the async code from everywhere else and have it only in
fsddispatch and fspdispatch. same with retry. this would get rid of the
"should i remove a reference or not" decision.
rx\??
the code that chains together pagingio reads should be removed
rx\ntfastio.c
in fastiocheckifpossible...return the error directly. also, we should
the cached netroottype as well as nodetype to determine if it's a
diskfile. also in markoncleanup.
rx\ntexcept.c
i removed this from the header
Original Comments
This routine process an exception. It either completes the request
with the saved exception status or it sends it off to IoRaiseHardError()
My New Improved Version
This routine processes an exception from the top level FSD routine. You can tell because it has
an Irp w/o an RxContext. The original code did all kinds of stuff like raising hard errors and
posting and such. My take is different...if you get here, things are hosed. So we either complete
or we bugcheck. PERIOD. we can always be more liberal later...........
rx\ntfsd.c
in RxDevFcbQueryDeviceInfo, the special handling for pipes should be
removed.
rx\lockctrl.c
don't allocate a lock structure unless you need one. also in fcb.h and
fcbstruc.c
rx\lockctrl.c
at rxunlock, we could be called from the cleanup routine on the last
handle. in such a case, we should just swallow the unlocks. this might be a
minirdr thing instead.
rx\fsctrl.c
in RxLowIoFsCtlShell, we should have to switch out again. we should have
handled the peeknamepipe throttling directly in the entry routine.
rx\flush.c
maybe we should try to route for filetypes that we don't understand. or
at least to have the minirdr signal that we should doit thru some stateful call.
rx\fileinfo.c
i remove the following
//CODE.IMPROVEMENT having a separate wrapper for every routine is stupid. we should (a) only
// have it if we need it and (b) preinitialize the buffer and lengthremaining so we don't always
// have to pass them.
[considering this we should fix the trace as well........]
and
//CODE.IMPROVEMENT probably there's a better way than this to figure out whether
// we need the resource at all or need it shared or need it excl.
// for name, we don't need it at all since the name never changes
and from allinfo handling
//CODE.IMPROVEMENT.ASHAMED either here or in the minirdr we should glob up call so that multiple trips
// are not taken
rx\fileinfo.c
you shouldn't do the fsrtloplock check for pipes. this keeps us from not
allocating the struct
rx\fileinfo.c
should we really change the name in the fcb after a rename. it
complicates locking a bit for querys
rx\dirctrl.c
no need to allocate templace buffer if it's small enough. most templates
are probably less than 8.3 in size
rx\create.c
the writeserialization field in the fobx is used by spoolfiles as well
as pipes. maybe a name change to the specific field could reflect this dual
usage.
rx\create.c
perhaps we should wait to finalize on a failed open.
rx\create.c
it would be good to "summarize" the fcb names that are logged
rx\create.c
we should have a flag on fcb/netroot somewhere to denote that only
32-bit operations are possible instead of having every mini have to check.
rxce\scavengr.c
in scavenge-fobxs, starting at the beginning of the list each time is
bad. what if there is some problem then we'd never get off the first guy.
rxce\prefix.c
i removed this comment from the lookup
// CODE.IMPROVEMENT we need code to lookup vnetroots MUCH faster....like a 26way cache........
// failing that, we could at least not do the lookups (and hash computation)
// until we're on the third component, i.e. \;m:\server\share. the cache is better.
rxce\fcbstruc.c
in finishfcbinitialization, we should not initialize the lock structure
for pipes. of course, we shouldn't even have a lock structure yet but that's a
different c.i.
rxce\buffring.c
i removed the following comment from RxChangeBufferingState
//CODE.IMPROVEMENT we sure use up a lot of log entries in here! maybe we should put this state info
// on the fcb.
BUGBUG
rxce\rxcontx.c
the priority boost is set to disk....probably should be net. actually
should be mini parameter.
in the place where we check for advance-only on setfileinfo, we'd need a
bit different for locals. there are a couple of places.
rxce\rxconnct.c
i removed the following from RxConstructSrvCall
//BUGBUG we should have some sort of refcount machanism to prevent a minirdr from
// unregistering while he's part of a calldown
//also, we shouldn't necessarily be calling down if this guy is not a UNC provider
rxce\fcbstruc.c
in RxWaitForStableCondition, you need to be able to cancel.
rxce\fcbstruc.c
in RxCreateNetRoot,
//TRUE means case insensitive compares BUGBUG get this from srvcall
turns out, the whole issue of casing is incorrectly handled because
there is no way to get casing handled on a per-call basis in case the
casesensitive flag is set on open.
rxce\fcbstruc.c
in RxCreateNetFcb,
//BUGBUG: this should be on the fobx......not on the FCB. it should then be used correctly
// in the querynameinfo code. same with stripped backslash
the flag in question is RX_CONTEXT_CREATE_FLAG_ADDEDBACKSLASH.
rx\strtstop.c
in RxStopMinirdr, the comment was removed from the following line
RxTerminateScavenging(RxContext); //BUGBUG separate into the two parts and make the purge depend on the DObj
rx\lockctrl.c
in RxUnlockOperation, the following comment was removed where we try to
allocate
//bugbug we cannot fail........i guess what we'd have to do is to send the unlock
//from here!
rx\fileinfo.c
in RxQueryStandardInfo, i removed this about the fact that links are
returned incorrectly
//BUGBUG there is a problem here with number of links...basically, the deal is that numberoflinks is
//dynamic whereas everything else in the structure is knowable by the client side under oplock.
rx\create.c
in , i removed this about downlevel canonicalization.
DOWNLEVEL.BUGBUG we don't really do it yet but we do
componentize and check for .. and .; for names that come down from win32
this is fine.....for direct names going to downlevel servers we should do
more checking maybe.
rx\create.c
in RxCanonicalizeFileNameByServerSpecs, i removed the comment on this
line both places
&& RX2C_IS_DOT(Buffer[firstchar+2])) //??BUGBUG what about a alternate stream :.
rx\read.c
need to implement a policy for controlling readahead differently for
different customers
rx\dirctrl.c
i removed this at the place when the uniarg template is comared
// then match all names. joejoe....do i really need all this?? ntfs doesn't!
rx\create.c
this from the header comments in RxCreateFromNetRoot:
joejoe There is some possibility that we should just keep shared on the tablelock until we
dispatch to the minirdr. That would save one release/acquire per open.
rx\cleanup.c
removed the joejoe from this
if (NeedDelete || NeedPurge) {
//joejoe need to put in logic to keep a new read/write from
//reinitializing
RxDbgTrace(0, Dbg, ("CleanupPurge:MmFlushImage\n", 0));
MmFlushImageSection(&capFcb->NonPaged->SectionObjectPointers,
MmFlushForWrite);
rx\cleanup.c
and from this.....
KeQuerySystemTime( &CurrentTime );
//
// Note that we HAVE to use BooleanFlagOn() here because
// FO_FILE_SIZE_CHANGED > 0x80 (i.e., not in the first byte).
//
//joejoe we need to probably do changetime independently
//perhaps the smartest thing would just to put a basicinfo
//struc in the fcb
rxce\rxconnct.c
from RxConstructSrvCall, removed the joejoe
//joejoe we should have a structure tag on this
pSrvCalldownStructure = RxAllocatePoolWithTag(
NonPagedPool,
sizeof(MRX_SRVCALLDOWN_STRUCTURE) +
(sizeof(MRX_SRVCALL_CALLBACK_CONTEXT) * NumberOfMinirdrs),
'CSxR' );
if (pSrvCalldownStructure == NULL) {
ExReleaseFastMutexUnsafe(&RxData.MinirdrRegistrationMutex);
return STATUS_INSUFFICIENT_RESOURCES;
}
rxce\fcbstruc.c
from RxFinalizeSrvOpen, removed the joejoe remark
if (ThisSrvOpen->Condition == Condition_Good) {
RxPurgeChangeBufferingStateRequestsForSrvOpen(ThisSrvOpen);
}
//joejoe we could put sme more asserts in here about empty srvfobxlist etc.
// close the file.
MINIRDR_CALL_THROUGH(Status,Fcb->MRxDispatch,MRxForceClosed,((PMRX_SRV_OPEN)ThisSrvOpen));
}
rxce\buffring.c
from RxChangeBufferingState, removed joejoe
MINIRDR_CALL_THROUGH(
Status,
Fcb->MRxDispatch,
MRxCompleteBufferingStateChangeRequest,
(RxContext,(PMRX_SRV_OPEN)SrvOpen,Context));
}
//joejoe this could fail!!!! we should place the fcb into a
// "failed oplock break" state.
RxDereferenceAndDeleteRxContext(RxContext);
}
AND
} finally {
ClearFlag(Fcb->FcbState,FCB_STATE_BUFFERSTATE_CHANGING); //this is informational for error recovery
//joejoe there should be a callout here for a readpolicy...not vanilla..........
}