This patch adds large file support (> 2GB) to the CIFS server and
the cache manager.
It also corrects a significant oversight where the definition of
struct rx_call was inconsistent due to a failure to define AFS_PTHREAD_ENV
prior to including rx.h in all source files. This prevented all errors
during rx_Read and rx_Write to be ignored.
When Freelance mode is not in use the cache manager must load the root.afs
volume of the default cell in order to be able to start. Unfortunately,
if the first vlserver fails the request to obtain the volume information
will timeout and no retries will be performed. This patch performs up
to 10 retries in the initialization code in order to enable the vlservers
to be marked down so that failover can occur.
"the non AFS_GLOBAL_RXLOCK_KERNEL version of
rxi_ReceiveAckPacket() wasnt counting 'hard' acks and not opening the
congestion window. using rx_Readv() directs the incoming packets into
the target iov, and sends a hard ack (skipping soft acks). "
A deadlock was detected when performing "fs flushall" if the file
server reports VNOVNODE. The scp->createBufferLock is already
held by the current thread and there is no mechanism to propagate
the knowledge. Therefore, an alternate mechanism for clearing the
cache must be developed. A new function cm_RecycleSCache(scp,flags)
has been extracted from cm_GetNewSCache(). This function performs
the task of recycling an cm_scache_t object. When called from
cm_GetNewSCache() with no flags the expectation is that there are
no associated buffers that are queued to be read or written. When
called from cm_Analyze() with the CM_SCACHE_RECYCLEFLAG_DESTROY_BUFFERS
flag, any queued buffers will be de-queued and marked as if the operations
were performed so the data can be discarded.
This patch also addresses the following issues:
* CM_ERROR codes were logged as "unknown" by cm_Analyze.
* In response to VNOVNODE, the parent is only discarded if the current
cm_scache_t is not a directory.
* In the Ioctl FlushFile and FlushVolume functions, there were no
checks to protect against flushing the Freelance SCache entries.
* In FlushFile, the wrong cm_scache_t object was being released.
* In cm_GetNewSCache, do not allow Freelance SCache entries to be
recycled. Choose a new entry if cm_RecycleSCache fails.
It is possible to have a valid callback but not to have the
required ACL info for the current user. Force acquisition of
a FetchStatus but do so without making multiple calls.
Dynamically adjust the priority of server threads based upon the age
of the cifs request that is being processed. Bump the priority one
level for each 15 seconds of age.
while investigating the cause of the delayed write errors it was observed
that all of the sleep queues are LIFO. This has the side effect of
encouraging starvation. Changing the queues to FIFOs revealed a serious
problem affecting the use of all queues which use both head and tail
pointers. The removal function osi_QRemove does not take a tail pointer
and therefore the pointer is always left hanging. If the number of elements
ever drops to zero the queue becomes corrupted.
Added osi_QRemoveHT to be used whenever head and tail pointers are used.
Updated all callers in afsd.
Do not delete an old host entry because the UUID because the UUIDs
of the clients at the old addresses are wrong. Remove the addresses.
Do not allow the same host,addr,port combination to be added into
the hash table multiple times.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
Add some more logging
Do not allow 'code' to be re-used for probing the uuid of the old host
primary address because we are relying on the value of 'code' to indicate
whether or not RCallBackConnectBack failed. If it did, we mark the
client as VENUSDOWN, if not we marked it as RESETDONE.
Suppress warnings from prototype changes.
Trial code for OpenBSD support.
Fix missing piece from UKERNEL.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
IPv6 support for Windows requires the use of ws2tcpip.h and associated
restructing of include file order.
Large series of changes to replace the use of afs_uint32 for IP address
with struct sockaddr_storage in Rx. May even work with IPv6, but only
tested with IPv4. Significant change to internal APIs; existing external
APIs kept, new ones added that take struct sockaddr_storage arguments.
According to src/libacl/prs_fs.h the PRSFS_LOCK privilege is only supposed
to control read-locks. The PRSFS_WRITE and PRSFS_INSERT privileges are
supposed to control write-locks.
A number of crash reports have been filed with Microsoft in afspthread.dll.
The crashes are occuring as the processes are terminating. In order to
attempt to address this problem this patch adds support for a DllMain()
entry point that is used to cleanup Thread Local Storage and the various
RX queues. One theory is that processes are loading and unloading DLLs
that are linked to afspthread.dll leaving the pthread data in a very
inconsistent state after each unload. By cleaning up the static data
during the unload, if afspthread is loading again it will be forced to
initialize the data once again.
This patch disables the lookup of the user's AFSID and instead solves
the problem by tracking which cm_user_t created the file. If the user
is unable to obtain a write lock and is the creator, then we check for
the Insert bit. Otherwise, we deny access.
Vista has a bug in the Netbios ADAPTER_STATUS output in which the
first four bytes of the MAC address with 127.0.0.1. We must check
for that when searching for the Loopback Adapter.
when the cache manager is being frequently restarted (especially during
debugging) the 7001/udp port is often not freed appropriately. This
prevents the cache manager from running. This patch allows the cache
manager to retry with a random port number.
In order to make the client side locking more usable, if the client
is denied a lock by the server and the user only has read privileges,
then we will allocate a local lock. Local locks are not allocated
for write locks.
There is a bug in the file server locking dating back to at least
AFS 3.1 in which the lock privilege is used for read, write and insert
locks. According to the docs, the lock privilege is only supposed to
control read locks. Write locks on new and existing files are supposed
to be controlled by the Write and Insert privileges. This will be fixed
in the file server by another commit.
In order to properly handle the local allocation of locks
the cache manager must enforce the locking rules associated
with PRSFS_WRITE and PRSFS_INSERT and PRSFS_LOCK. Insert
affects "new files" which in AFS are defined as any file that
is owned by the user. Therefore, we must know the afsid of the
user so that it can be compared to owner of the file.
This commit includes a query using PR_SNameToId where the name
is the name specified by the entity setting the token. There
does not currently exist a Protection Service RPC to return the
AFSID of the entity that the rx connection was authenticated as.
Such a function will have to be added in order to make this
functionality secure.
====================
This delta was composed from multiple commits as part of the CVS->Git migration.
The checkin message with each commit was inconsistent.
The following are the additional commit messages.
====================
Export functions missing from the UNIX version and the PR_ functions
required by the Windows Cache Manager.
This patch adds code to perform return code error checking for all calls
to the ReadXyz() routines. If the rx connection was lost or if the dump
format being parsed was incorrect there was an opportunity for data to
the processed after an error had already been detected. In the best case
this would have simply resulted in additional work being performed when
it was not necessary but there was also the potential for modifying return
values which would later be used as input to later operations. Better
to simply catch the error when it initially occurs and be done with it.
The ReadString() function had the potential to write into one byte prior
to the start of the input array.
ProcessIndex() would not properly close Fdhandle nor the StreamHandle
if an error was detected.
RestoreVolume() had a problem with the processing of incremental dumps.
Originally AFS supported a concept of incrementals dumps which when
restored would simply add their contents to the existing volume. One
of the distinctions between the incremental and full dumps is that full
dump processing is responsible for removing any vnode entries from the
volume that are not present in the dump file headers. When the incremental
dump support was removed, the RestoreVolume() function was broken. If
a dump file contains multiple Dump Headers, then only the vnode entries
in the first Dump Header would be marked as being restored. Any vnode
entries restored from subsequent Dump Headers would be deleted when the
RestoreVolume() function reached the end. By removing the assignment
of 'tdelo = -1' within the loop that processes the Dump Headers, this
bug is resolved. Although RestoreVolume is only called with the
'incremental' parameter == 0 at the current time, the check for non-zero
values is restored and the "remove unrestored vnodes" functionality is
disabled when a non-zero value is provided.
The ReadVnodes() function did not perform error checking.
The 'active_vcp' variable is no longer being used for any useful
purpose and is the center of a race condition that could lead to
an undercount of references to 'vcp' objects. Remove it.
An undercount has been detected of the smb_vc_t objects stored
in the smb_allVCsp list. Unfortunately, we have yet to be able
to find the cause of the undercount so this patch adds logic to
protect against the side effects until such time as the cause
can be identified.
The old StoreData RPC is only supposed to be valid for files
smaller than 2GB. When StoreData64 was added, StoreData and StoreData64
were implemented as calls to common_StoreData64. This removed the bounds
checking on the old StoreData RPC making it possible for operations
beyond two 2GB to be requested even if the file server cannot support
them. This patch adds a validity check to ensure that the requested
file operations remain below 2GB. Failures return E2BIG.
projects / packages / o / openafs.git / log