Commit e86eb73e (obsd-vattrs-20040125) introduced an XBSD-specific
check to detect some unchanged attributes. But the #ifdef for XBSD for
the va_uid section was added in the middle of an HPUX-specific block
by mistake.
Move this #ifdef one level higher, so it's actually used on BSD
platforms.
Change-Id: I606f87f21d6c4830ed8bcf50abd6fb5807868ff5
Reviewed-on: https://gerrit.openafs.org/14473
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Tim Creech <tcreech@tcreech.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Normally, a client starts a new Rx call by sending DATA packets for
that call to a server, and rxi_ReceiveServerCall on the server creates
a new call struct for that call (since we don't recognize it as an
existing call).
Under certain circumstances, it's possible for a server to see a
non-DATA packet as the first packet for a call, and currently
rxi_ReceiveServerCall will create a new server call for any packet
type. The call cannot actually proceed until the server receives data
from the client (and goes through the challenge/response auth
handshake, if needed), but usually this is harmless, since the
existence of any packets for a particular call channel indicate that
the client is trying to run such a call. The server will respond to
the client with ACKs to indicate that it is missing the needed DATA
packet(s), and the client will send them and the call can proceed.
However, if a call is in the middle of running when the server is
restarted, the client may be sending ACKs for a pre-existing call that
the server doesn't know about. In this case, the server generates ACKs
that indicate the server has not received any DATA packets, which may
appear to violate the protocol, depending on the prior state of the
call (e.g. the server appears to try to move the window backwards).
Clients should be able to detect this and kill the call, but many do
not. For many OpenAFS releases before commit 7b204946 (rx: Avoid
lastReceiveTime update for invalid ACKs), the client will get confused
in this situation and will keep the call open forever, never making
progress.
There isn't any benefit to creating a new server call in these
situations, so just ignore non-DATA packets for unrecognized calls, to
avoid stalled calls from such clients. Those clients will not get a
response from the server, and so the call will eventually die from the
normal Rx call timeout.
Change-Id: I565625ba8b6901f9b745124a8816a9ba816c0264
Reviewed-on: https://gerrit.openafs.org/13758
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Currently, we ignore ACK packets in a few cases:
- If the ACK appears to move the window backwards (if firstPacket is
smaller than call->tfirst).
- If the ACK appears to have been received out of order (if
previousPacket is smaller than call->tprev).
- If the ACK packet appears truncated.
In all of these cases, we ignore the ACK packet completely in our ACK
processing code (rxi_ReceiveAckPacket), but we still process the
packet at higher levels (rxi_ReceivePacket). Notably, this means we
update call->lastReceiveTime after rxi_ReceiveAckPacket returns, even
for ACK packets we haven't really looked at.
Normally this does not cause any noticeable problems, because such
packets should either never be encountered, or only consist of a small
number of packets that are mixed in with valid packets.
However, if our peer is a server, and it is restarted in the middle of
a call, our peer may exclusively send us packets that fall into the
above categories. (This does not happen if our peer is a client,
because clients just ignore packets for calls they do not recognize.)
For example:
Consider a call where a client is sending data to a server, and the
server restarts after the client has sent a DATA packet with sequence
number 1000. The server may then start responding to the client with ACKs
with firstPacket set to 1, since the restarted server has no knowledge
of the call's state.
In this case, a firstPacket of 1 is well below where our window was,
so all of the ACKs from the server are ignored. But we keep updating
call->lastReceiveTime for all of these packets, and so the call stays
alive forever until an idle-dead or hard-dead timeout activates (if
any are set).
As another example, consider the case where a client is sending data
to a server, and the server receives a full window of packets (say, 16
packets), has not yet passed any data to the application yet, and the
server restarts. The restarted server then starts responding to the
client with ACKs with firstPacket set to 1, and previousPacket set to
0. We also ignore all of the ACKs from the server in this case,
because even though firstPacket looks sane, it looks like
previousPacket has gone backwards. We still update
call->lastReceiveTime for each ignored ACK we get, keeping the call
alive.
Before commit 4e71409f (Rx: Reject out of order ACK packets) was
introduced in 1.6.0, neither of these issues could occur. That commit
introduced the issue specifically if previousPacket goes backwards;
that is, if the server restarts before firstPacket moves forwards.
Commit 8d359e6d (rx: Remove duplicate out of order ACK check) in 1.8.0
introduced the issue when 'firstPacket' goes backwards, since
previously the FIRSTACKOFFSET-based check caused us to ignore those
packets without updating call->lastReceiveTime. That is, if the server
restarts after firstPacket moves forwards.
In this commit, we still ignore packets in the above cases, but we
also avoid updating lastReceiveTime when we update such packets, to
make sure that we do not keep a call alive solely from receiving these
invalid packets.
Alternatively, we could change our logic to immediately abort calls
where firstPacket moves backwards (since this violates the Rx
protocol), or to not ignore some packets where previousPacket goes
backwards (since these calls may be recoverable). And we could also
skip updating lastReceiveTime for invalid packets of other types. But
for now, this commit just avoids updating lastReceiveTime for invalid
ACK packets, in order to just try to restore our behavior before
1.6.0, while still retaining the benefits of ignoring out-of-order
ACKs. Further changes in this area can potentially be handled
separately by future commits.
Also increment the spuriousPacketsRead counter for packets that we
ignore in this way (which we used to do for some packets before commit
8d359e6d), so we are not entirely silent about ignoring them.
Written in collaboration with mvitale@sinenomine.net.
Change-Id: Ibf11bcb2417d481ab80cf4104f2862d1d6502bf4
Reviewed-on: https://gerrit.openafs.org/13875
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Take some of our existing logic for ignoring invalid ACK packets and
split it out into a separate function, ack_is_valid. This just makes
it easier to add more complex logic in here and write longer comments
explaining the decisions.
Note that the bug mentioned regarding the previousPacket field was
introduced in IBM AFS 3.5, and was fixed in OpenAFS in commit bbf92017
(rx: rxi_ReceiveDataPacket do not set rprev on drop), included in
OpenAFS 1.6.23.
This commit incurs no functional change; it is just code
reorganization.
Change-Id: Idd569c6bc0c475e700935cf86780a04ab24102f4
Reviewed-on: https://gerrit.openafs.org/13874
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
When AFS_RXERRQ_ENV is defined, we currently end up doing something
like this for our sendmsg abstractions:
if (sendmsg(...) < 0) {
while (rxi_HandleSocketError(sock))
;
return error;
}
return success;
This means that when sendmsg() returns an error, we process the socket
error queue before returning an error.
The problem with this is that when we receive an ICMP error on our
socket, it creates a pending socket error that is returned for any
operation on the socket. So, if we receive an ICMP error after trying
to contact any peer, sendmsg() could return an error when trying to
send for any other peer. Even though there is no issue preventing us
from sending the packet, we'll fail to actually send the packet
because sendmsg() returned an error. This effectively causes an extra
outgoing packet drop, possibly delaying the related RPC.
To avoid this, change Rx to retry the sendmsg call when it returns an
error, since the error may be due to an unrelated ICMP error.
To avoid needing to implement this retry loop in multiple places, move
around our sendmsg code for AFS_RXERRQ_ENV, so that the higher-level
function rxi_NetSend performs the retry and checks for socket errors
(instead of the lower-level rxi_Sendmsg or osi_NetSend). Also change
our functions to process socket errors to be more consistent between
kernel and userspace: now we always have rxi_HandleSocketErrors, which
runs a loop around the platform-specific osi_HandleSocketError.
With this commit, osi_HandleSocketError is now required to be
implemented when AFS_RXERRQ_ENV is defined. We hadn't been
implementing this for UKERNEL, so just turn off AFS_RXERRQ_ENV for
UKERNEL.
Thanks to mbarbosa@sinenomine.net for discovering and providing
information about the relevant issue.
Change-Id: Iccceddcd2d28992ed7a00dc308816a0cb1a0195f
Reviewed-on: https://gerrit.openafs.org/14424
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Currently, our pthread version of rxi_Sendmsg uses 'errno' in some
logic if sendmsg fails, but we do so after calling functions that
might alter errno (e.g. fflush).
To make sure we get the correct errno value, save the value of errno
right after sendmsg returns an error. Reorganize this function a bit
to help make the logic easier to follow.
Change-Id: I6bf284bd75edb5404bb6771bb99a9381b0f8654d
Reviewed-on: https://gerrit.openafs.org/14423
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Add the readdir/fidreaddir sub-commands to afsio dump AFS3 directory
objects. This command dumps the raw directory object to stdout. Pipe
the output to a program, such as the afsdump_dirlist program (from the
CMU dumpscan tool kit), to parse the directory object.
Example usage:
afsio readdir -dir /afs/mycell/mypath/somedir | afsdump_dirlist
Change-Id: Ief181b432cdea6a11bbe61e781686ade2795faad
Reviewed-on: https://gerrit.openafs.org/12381
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
In order to avoid future bit-rot, always build vol-bless. Also add it
to the clean rule. However, continue to leave it undistributed and
uninstalled by default.
Change-Id: I3d2dc94c28a7feeb20167223655e97538e807ce6
Reviewed-on: https://gerrit.openafs.org/14464
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Stephan Wiesand <stephan.wiesand@desy.de>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
A comment in rx_packet.h referred to the size of struct rx_ackpacket,
but the actual structure is spelled with a majuscule 'P'.
Change-Id: Iaf57f098b2e818fe0d492a89347a0a14bc3eb392
Reviewed-on: https://gerrit.openafs.org/14468
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Our LWP version of rxi_Sendmsg can allocate an fd_set, but we don't
free the fd_set if sendmsg() returns certain errors afterwards.
To make sure we go through the same cleanup code for the different
possible error code paths, reorganize the function to go through a
'goto error'-style destructor. This also makes our return codes a bit
more consistent; we should always return -errno now for errors.
Change-Id: I5eaeb7f4ea1d76acc3bd9c52dc258f53f59f631e
Reviewed-on: https://gerrit.openafs.org/14422
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
In commit 9ebff4c6 (OPENAFS-SA-2018-001 audit: support butc types),
several new butc-related audit data types were added. In the
AIX-specific audmakebuf() function, the case for the AUD_TSTT type is
missing the actual "case" clause in the code, causing AUD_TSTT types
to be treated as invalid (and so falling through to the
"AFS_Aud_EINVAL" case).
Add the "case" for AUD_TSTT, so it's treated properly on AIX. Note
that the non-AIX printbuf() already handled this properly, so no
changes are needed there.
Change-Id: Ic46c18b503bacb0901ff0a60534f6c45ce3c9a75
Reviewed-on: https://gerrit.openafs.org/14466
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
No functional change is incurred by this commit.
Change-Id: If84ba946d43d67eb6c253462f5826f9a45a2df46
Reviewed-on: https://gerrit.openafs.org/14463
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
The vol-bless utility is not built by default and so is subject to
bit-rot. Thus commit 170dbb3ce301329ff127bb23fb588db31439ae8d 'rx: Use
opr queues' overlooked vol-bless.c when adding includes for users of
struct rx_queue.
Add the required #include <rx/rx_queue.h> so vol-bless builds again.
Note to maintainers: this change is only required for 1.8.x and later;
vol-bless builds fine in 1.6.x and earlier releases.
Change-Id: Ia0bb78e3e7dd74b2f65ac07707aced2c81aaa5d9
Reviewed-on: https://gerrit.openafs.org/14462
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Add counters for cache too full and waiting to drain occurrences. These
will be used in later commits to indicate how often the cache truncation
is required and how often the cache manager is waiting for cache
truncation to complete.
Change-Id: I4aa802729f0910dff1fb3e90b2d44d36df8bf8f3
Reviewed-on: https://gerrit.openafs.org/13168
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
New functionality was added to the audit facility that allows multiple
audit logs. The updated audit interfaces require a specific calling
sequence even if multiple audit logs are not used.
Support for multiple auditlogs is not supported for kauth. Since kauth
does not use libcmd for processing the command line, and adding support
for multiple audit log instances requires additional effort, that is not
warranted.
Update kauth to follow the proper calling sequences for the audit
facility.
Update help message and manpage entries for -auditlog and
-audit-interface. Make note that multiple -auditlogs are not supported.
Change-Id: I98111b1e399e6687fde235bc2eadf0a28fa8acf4
Reviewed-on: https://gerrit.openafs.org/13782
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Gerrits #13774 (audit: Support multiple audit interfaces and interface
options) and #13775 (audit: Add cmd helper for processing audit options)
added support in the audit facility for multiple audit logs.
Add command line support to use multiple audit logs for daemons that use
libcmd for command line processing: bosserver, buserver, butc,
fileserver, volserver, ptserver, and vlserver.
Update the daemons to add a call to audit_open, and where possible add a
call to audit_close when shutting down the daemon.
Update help message and manpage entries for -auditlog and
-audit-interface
Change-Id: I4356e1aa84f580897a0e788e2a2829685be891aa
Reviewed-on: https://gerrit.openafs.org/13776
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
osi_audit_cmd_Options will handle the processing for the
-audit-interface and -audit-log command line options.
The auditlog / audit-interface options are used by several services;
this new helper routine provides a simple method to process the audit
related command line options in a consistent fashion.
Change-Id: I5acd12062dbfec23c1cbb0b2cdfc2d224354eed9
Reviewed-on: https://gerrit.openafs.org/13775
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Currently, the audit subsystem only allows for one audit log to exist
for the entire process. This can make it cumbersome to use for sites
that have multiple tools or destinations that want to read the audit
data. For example, to feed the audit data to two separate scripts, one
script needs to read the data, and retransmit the data to the second
script.
To make such a setup easier, change the audit system to allow for
multiple audit logs to exist at once. To allow callers to associate
each audit log with an interface, we change the syntax for the value to
the -auditlog parameter to the following:
[interface:]filespec[:options]
For example:
-auditlog sysvmq:/tmp/msgqueue
To accommodate the existing -audit-interface parameter, change the
behavior of -audit-interface so that it sets the default audit interface
if none is specified for -auditlog. This allows existing users of
-audit-interface to experience the same behavior as before.
In order to implement this, change the audit API and all existing audit
interfaces to avoid using per-interface globals, and instead allocate
per-instance contexts during startup. Also change the code so the audit
message is constructed inside audit.c, instead of via a per-interface
callback, which eliminates the duplicated logic in each interface's
append_msg(), and lets us avoid holding 'audit_lock' during message
construction.
While we're changing the audit API, also introduce a few new operations:
open_interface, close_interface and set_options. This commit and the
existing interfaces do not make use of these new functions, but future
commits will do so.
This commit also only changes the audit subsystem itself to be able to
handle multiple audit logs, and doesn't change any command-line parsing
logic. Future commits will add the command-line parsing logic changes
required so daemons can actually configure multiple interfaces.
Thanks to Andrew Deason (adeason@sinenomine.net) for providing the
changes needed to reduce holding the 'audit_lock' and improve
performance as well as providing input during the development of this
change.
Change-Id: I1311ea417fdd0ba38d2206083cd65bd7a054d017
Reviewed-on: https://gerrit.openafs.org/13774
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
In our d_revalidate callback (afs_linux_dentry_revalidate), we
currently 'goto bad_dentry' when we encounter any error. This can
happen if we can't allocate memory or some other internal errors, or
if the relevant afs_lookup call fails just due to plain network
errors.
For any of these cases, we'll treat the dentry as if it's no longer
valid, so we'll return '0' and call d_invalidate() on the dentry.
However, the behavior of d_invalidate changed, as mentioned in commit
afbc199f1 (LINUX: Avoid d_invalidate() during
afs_ShakeLooseVCaches()). After a certain point in the Linux kernel,
d_invalidate() will also effectively d_drop() the given dentry,
unhashing it. This can cause getcwd() calls to fail with ENOENT for
those directories (as mentioned in afbc199f1), and can cause
bind-mount calls to fail similarly during a small window.
To avoid all of this, when we encounter an error that prevents us from
checking if the dentry is valid or not, we need to return an error,
instead of saying 'yes' or 'no'. So, change
afs_linux_dentry_revalidate to jump to the 'done' label when we
encounter such errors, and avoid calling d_drop/d_invalidate in such
cases. This also lets us remove the 'lookup_good' variable and
consolidate some of the related logic.
Important note: in older Linux kernels, d_revalidate cannot return
errors; callers just interpreted its return value as either 'valid'
(non-zero) or 'not valid' (zero). The treatment of negative values as
errors was introduced in Linux commit
bcdc5e019d9f525a9f181a7de642d3a9c27c7610, which was included in
2.6.19. This is very old, but technically still above our stated
requirements for the Linux kernel, so try to handle this case, by
jumping to 'bad_dentry' still for those old kernels. Just do this with
a version check, since no configure check can detect this (no function
signatures changed), and the only Linux versions that are a concern
are quite old.
Change-Id: Ie530ce08463cf6b6899f056cb76ae4047c989ef2
Reviewed-on: https://gerrit.openafs.org/14417
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Verify the a lock timestamp is set if, and only if, a lock volume
operation flag is also set.
When running vldb_check with the -fix option, fix the inconsistent
entries by setting the lock timestamp to the current time if a lock flag
is set, or by setting the VLOP_DELETE flag if the lock timestamp is set
but no lock flags are set. (The VLOP_DELETE flag is the flag set by the
'vos lock command, and is shown in vos output as "delete/misc".)
Volume lock fields can be put into an inconsistent state, at least, by
interupted vos rename operations, due to bugs in vos rename. When the
volume lock timestamp and lock flags are in this inconsistent state, the
volume is locked, but that is not indicated by 'vos listvldb'. The
volume can be unlocked by issuing 'vos unlock'.
Change-Id: Idc4f821a9eb7675edd78a8547fdfe46e838b0c89
Reviewed-on: https://gerrit.openafs.org/14307
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Remove the dead code in UV_VolumeMove() commented out with the macro
ENABLE_BUGFIX_1165.
Remove two commented out lines of code in UV_ConvertRO().
Change-Id: Ic628c74df011b0f09be6b03f72ab1baac5e59caf
Reviewed-on: https://gerrit.openafs.org/14004
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
The functions defined within vos.c are not referenced outside of vos.c
but are not declared as static.
Convert the functions within vos.c to static declarations.
Change-Id: Ia684e698adc53ced964e10ee0496cb52a3af564e
Reviewed-on: https://gerrit.openafs.org/14009
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Clean out dead code from vos.c
GetVolumeType - not referenced anywhere
CompareVLDBEntry - commented out since 1st git commit
osi_audit - Comment indicates this might have been needed at one point.
Builds without it. Does not look like the vos executable
is pulling in any of the audit code.
RestoreVolume - remove stale comment about typo previous to openafs 1.0
RemoveSite - remove commented out partition check
Change-Id: I9c0b59d5c37d403610c7a904717ac9765598fc99
Reviewed-on: https://gerrit.openafs.org/14008
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
The vos convertROtoRW command converts a RO volume into a RW volume.
Unfortunately, the RO volume is not checked out from the fileserver
during this process. As a result, accesses to the volume being converted
can leave volume objects in an inconsistent state.
Moreover, consider the following scenario:
1. Create a volume on host_b and add replicas on host_a and host_b.
$ vos create host_b a vol_1
$ vos addsite host_b a vol_1
$ vos addiste host_a a vol_1
2. Mount the volume:
$ fs mkmount /afs/.mycell/vol_1 vol_1
$ vos release vol_1
$ vos release root.cell
3. Shutdown dafs on host_b:
$ bos shutdown host_b dafs
4. Remove RO reference to host_b from the vldb:
$ vos remsite host_b a vol_1
5. Attach the RO copy by touching it:
$ fs flushall
$ ls /afs/mycell/vol_1
6. Convert RO copy to RW:
$ vos convertROtoRW host_a a vol_1
Notice that FSYNC_com_VolDone fails silently (FSYNC_BAD_STATE), leaving
the volume object for the RO copy set as VOL_STATE_ATTACHED (on success,
this volume should be set as VOL_STATE_DELETED).
7. Add replica on host_a:
$ vos addsite host_a a vol_1
8. Wait until the "inUse" flag of the RO entry is cleared (or force this
to happen by attaching multiple volumes).
9. Release the volume:
$ vos release vol_1
Failed to start transaction on volume 536870922
Volume not attached, does not exist, or not on line
Error in vos release command.
Volume not attached, does not exist, or not on line
Notice that this happens because we cannot mark an attached volume as
destroyed (FSYNC_com_VolDone).
To avoid the problem mentioned above and to prevent accesses to the
volume being converted, take the RO volume offline before converting it
to RW.
Change-Id: Ifd342e1f420dc42e5da49242a7aa70db7d97a884
Reviewed-on: https://gerrit.openafs.org/14340
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Fix the indentation whitespace in vos.c, and remove double blank
lines. No functional change.
Change-Id: I97587779d6d2c131b5eac98bbee49efae73fafe9
Reviewed-on: https://gerrit.openafs.org/14007
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Change the GetServerAndPart() function to return true when a volume site
in the vldb entry is found. Do not change the output arguments unless
the site is found. Also, add a function comment header and fix some
comment typos in this function.
Change-Id: I10b43054b1bf9e6757ccdc95cb4559ab8b6dc013
Reviewed-on: https://gerrit.openafs.org/14006
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
The `vos remove` command was missing a check for the -server option when
the -partition option is given. This command requires the -server option
when the -partition is given, as documented in the man page.
The `vos syncvldb` command performed the check for the -server option
when the -partition option is given, but in the wrong location.
As documented, the `vos unlockvldb` command permits the -partition
option without a -server option, in which case all of the volumes listed
in the VLDB with sites on the specified partition are unlocked.
However, this command incorrectly issued an RPC to a volume server at
address 0.0.0.0 when only the partition is given.
Change-Id: I6b878678e28b34250e63d2d082747f6fd416972d
Reviewed-on: https://gerrit.openafs.org/14005
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
To update a volume entry in the VLDB, vos commands typically lock the
volume entry via VL_SetLock, then call VL_UpdateEntryN, then release the
lock via VL_ReleaseLock. However, some vos commands exploit the
optional lock release flags of VL_UpdateEntryN to combine the update and
unlock operations into a single RPC. This approach requires extra care
to ensure that VL_ReleaseLock is issued for a failed VL_UpdateEntryN,
but NOT for a successful VL_UpdateEntryN.
Unfortunately, the following commands have success paths that fall
through to the error path, resulting in a double release of the volume
lock:
- vos convertROtoRW
- vos release
A second VL_ReleaseLock of a volume entry that has already been unlocked
via VL_UpdateEntryN is essentially a harmless no-op (other than negating
any benefit of exploiting the VL_UpdateEntryN lock flags). However, if
there is a race with another volume operation on the same volume, it is
possible for this bug to release the volume lock of a different volume
operation.
This problem has been present in 'vos release' since OpenAFS 1.0. This
problem has been present in 'vos convertROtoRW' since the command's
introduction in commit 8af8241e94284522feb77d75aee8ea3deb73f3cc
vol-ro-to-rw-tool-20030314.
Properly maintain state to avoid unlocking a volume (with
VL_ReleaseLock) that has already been unlocked (via VL_UpdateEntryN).
Thanks to Andrew Deason for discovering the issue and suggesting the
fix.
Change-Id: I757b4619b9431d1ca980f755349806993add14a5
Reviewed-on: https://gerrit.openafs.org/14426
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Commit 0c03f8607e15 vos-command-enhancements-20011008 introduced the
'vos restore' -readonly option, which allows the restored volume to be
RO instead of the default RW. The commit message documents the
following restriction:
- ... This option causes the restored volume to be an RO volume. It is
not permitted to restore an RO volume when the associated RW volume
already exists. While it is possible to restore an RW volume where an
RO volume exists, caution should be used to avoid doing this with VLDB
entries created by 'vos restore -readonly', since such entries have
their ROVOL and RWVOL ID's set to the same thing.
Document this restriction in the 'vos restore' man page, and in a code
comment.
No functional change is incurred by this commit.
Change-Id: I34f6c5434b82da538a38a9d219207b33dcf62b17
Reviewed-on: https://gerrit.openafs.org/14348
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
UV_RestoreVolume2 calls VLDB_GetEntryByName to obtain information for
sanity checking, but only checks for a VL_NOENT error code; other codes
are thus ignored, which may lead to confusing results.
Add an additional error check for 'vos restore' (and other callers of
UV_RestoreVolume2) to stop and issue an error message if a non-VL_NOENT
error code is received from VLDB_GetEntryByName.
Change-Id: Idf41965fdd84fa282a3397215ec393ae10f72018
Reviewed-on: https://gerrit.openafs.org/14347
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Correctly spell "can't" in a log message and a comment.
Change-Id: I9d5c667d9c5ea3c5b726f958431c497353433239
Reviewed-on: https://gerrit.openafs.org/14346
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
afs_WriteDCache may fail for an IO error, or if interrupted (EINTR).
Unfortunately, DNew will panic in this case, crashing the entire
machine.
In order to avoid an outage in this case, don't panic. Instead, reflect
the error back to the caller of DNew.
While here, add Doxygen comments to DNew.
Change-Id: I27a8f89bab979c5691dded70e8b9eacbe8aff4fd
Reviewed-on: https://gerrit.openafs.org/13804
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
DRelease has two assignments for tp = entry->buffer; remove the second
(redundant) one.
Introduced with 0284e65f97861e888d95576f22a93cd681813c39 'dir: Explicitly
state buffer locations for data'.
No functional change should be incurred by this commit.
Change-Id: If4a17862f451973075fa3fa267b5139046d97ede
Reviewed-on: https://gerrit.openafs.org/13802
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Commit 0284e65f97861e888d95576f22a93cd681813c39 'dir: Explicitly state
buffer locations for data' changed DNew and DRead to return a return
code. However, the callers of DNew were not modified to check the new
return code. (This commit applied only to the implementations dealing
with AFS directories, in afs/afs_buffer.c and dir/dir.c. The ubik
implmentations of DNew and DRead, dealing with ubik databases, were not
modified.)
Modify all (non-ubik) callers of DNew to check the return code. In
addition, modify code as needed so return codes are properly propagated
to the callers.
While here, add Doxygen comments for AddPage and FindBlobs.
Change-Id: Iabde6499745dd351f3fcda73c9f52c440a36490e
Reviewed-on: https://gerrit.openafs.org/13801
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Numerous types and constants are defined in our various RPC-L files
that are never used or referenced by anything. Remove them.
Change-Id: I0b03be1ce0e186a88f80d2f3f7a66a1e25965ff3
Reviewed-on: https://gerrit.openafs.org/14404
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
The function declaration was already marked as static; mark the
definition as well for consistency (and consistency with the other
helpers in this file).
Change-Id: I642db1d27efd34ab2a09f7299791c19d07b1f923
Reviewed-on: https://gerrit.openafs.org/13321
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
afs_dir_MakeDir() ignores the return code from afs_dir_Create() for the
'.' and '..' ("dot" and "dotdot") directories. This has been the case
from the earliest implementation (MakeDir() calling Create()) in the
original IBM import.
Instead, check the return codes to prevent the possibility of creating
malformed directories.
Change-Id: I60179488429dfa9afe60c4862c5e42b41f1e0048
Reviewed-on: https://gerrit.openafs.org/13800
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
These helper function names alias the names of public RPCs and can
cause confusion when grepping the code. Rename them in a different
style to provide greater hamming distance between the various
functions involved in handling these RPCs.
Change-Id: I0e2c7997bc145888affdac28716293ff820756c7
Reviewed-on: https://gerrit.openafs.org/13320
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Since the original IBM import, DirSalvage() has ignored the return code
from afs_dir_MakeDir() (f.k.a. MakeDir). This has been safe because, as
the comment states, afs_dir_MakeDir returns no (non-zero) error code.
In preparation for a future commit, add a check for the return from
afs_dir_MakeDir and remove the comment.
Change-Id: Ibb259a7aaeeb21ef70a7794143a0dadb2a75725d
Reviewed-on: https://gerrit.openafs.org/13799
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
The directory package (src/dir) salvage routines DirOK and DirSalvage
check a global variable 'DErrno' to distinguish logical errors (e.g.
short read) from physical errors (e.g. EIO). However, since the
original IBM import, this logic has not worked correctly because there
is no longer any code that sets the value of DErrno - its value is
always zero.
Instead, modify all implementations of ReallyRead to optionally return
the errno for low-level IO errors.
Also, create a new userspace-only variant - DReadWithErrno() - of the
src/dir/buffer.c version of DRead (the version called by DirOK and
DirSalvage, and the only caller of ReallyRead) to return the ReallyRead
errno upon request.
Also create an analogous variant of afs_dir_GetBlobs,
afs_dir_GetBlobsWithErrno().
Finally, convert DirOK and DirSalvage to use the new variants and
replace DErrno with equivalent logic. Remove all other references to
DErrno.
Change-Id: I3de182ce49c1682572142da594af5dc2c00ede74
Reviewed-on: https://gerrit.openafs.org/13798
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Log the current pid (and procname) when we complain about an error
when reading from CacheItems in afs_UFSGetDSlot. These errors can
result in confusing situations, so it can be helpful to know at least
what process saw the error.
Our logic for logging this information is getting a bit large, so also
move this to a new function, LogCacheError.
Change-Id: I3427e736458784df0d516f4182684605e930e128
Reviewed-on: https://gerrit.openafs.org/14416
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Reviewed-by: Cheyenne Wills <cwills@sinenomine.net>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Windows standard library doesn't provide strtok_r. Use the strtok_r
that is provided from roken.
Change-Id: I1bccb9a306c9dd1963f044127fb5dfe4da5728cc
Reviewed-on: https://gerrit.openafs.org/13891
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
This commit updates the code imported from heimdal to
5dfaa0d10b8320293e85387778adcdd043dfc1fe (git2svn-syncpoint-master-311-g5dfaa0d10)
New files are:
roken/strtok_r.c
Change-Id: I27042f614c7d6ce9a95a80d01474e8bf401e4760
Reviewed-on: https://gerrit.openafs.org/13890
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Import the strtok_r function which is needed by audit for parsing
command line options.
Change-Id: I8412c5a663dc3315c4146665edb72d9a6b8df5be
Reviewed-on: https://gerrit.openafs.org/13889
Reviewed-by: Andrew Deason <adeason@sinenomine.net>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
While reviewing other commits, a call to realloc() was discovered that
would leak memory on failure (by virtue of always assigning the realloc()
return value to the pointer holding the input address, even when the
return value is NULL). Check for failure and return early in that case
(giving an incomplete list of events).
Change-Id: Ic6e889f1d990bd289812ce4bf8e9cd4ebce488ec
Reviewed-on: https://gerrit.openafs.org/13313
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
These two helpers are only used in implementing server-side RPC handlers,
and having to track the codeflow across files is unhelpful. Move them
into the file where they're used, make them static, and remove the
prototypes from ptrototypes.h (which is not an installed header, so
there is no API/ABI breakage).
Change-Id: I236d17865a296933f41aaee206535d341c3a955d
Reviewed-on: https://gerrit.openafs.org/13319
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
This should prevent inadvertent reassignment if additional RPCs
are introduced in the future.
Change-Id: I5645ca478d2ecef9962f4bde04ab8f9895dd9497
Reviewed-on: https://gerrit.openafs.org/13317
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Michael Meffie <mmeffie@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
If we try to delete a vlentry, and the vlentry cannot be found on one
of its hash chains, we cannot unhash the vlentry properly and the
operation fails with VL_NOENT. This results in the following error
messages to the user:
$ vos delentry 123456
Could not delete entry for volume 123456
You must specify a RW volume name or ID (the entire VLDB entry will be deleted)
VLDB: no such entry
Deleted 0 VLDB entries
This is confusing, because VL_NOENT can also occur if the user
specifies a volume that does actually not exist. This situation is
indicative of database corruption, usually because of a ubik
transaction that was only half-applied, or because of other ubik bugs
in the past.
The situation can only really be fixed by repairing the database, so
return VL_DBBAD in this case instead, to more clearly indicate that
something is wrong with the database, and not a problem with the
arguments the caller provided.
Change-Id: I6fc275c3ad05c108778f36687227b0a927cca5da
Reviewed-on: https://gerrit.openafs.org/13384
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
Tested-by: BuildBot <buildbot@rampaginggeek.com>
The VL_ error table currently doesn't have an error code to indicate
that an operation cannot succeed because the database is corrupted.
There are a few error codes for specific cases of errors that are
probably the result of corruption (like VL_IDALREADYHASHED, or
VL_EMPTY), but these are only for specific cases and indicate rather
low-level internal problems.
There are some instances where the real problem preventing an
operation from succeeding is that the database is just corrupt or
inconsistent in some way, and the administrator must repair the
database before it can succeed. And we currently don't have any way of
indicating that situation via an error code.
So, introduce the VL_DBBAD code, to indicate this situation. Error
codes already exist in other tables for similar situations, such as
PRDBBAD, and KADATABASEINCONSISTENT.
This commit does not use the new error code anywhere; we just
introduce it into the VL_ error table, so comerr-using applications
will be able to interpret it.
Note that the VL_DBBAD error code has been recognized by the AFS
Assigned Numbers Registry as recorded in the ticket history of
<https://rt.central.org/rt/Ticket/Display.html?id=134817>
Change-Id: I8fea356a4e0db907ec8418efe6ef35d547be0a63
Reviewed-on: https://gerrit.openafs.org/13383
Tested-by: BuildBot <buildbot@rampaginggeek.com>
Reviewed-by: Mark Vitale <mvitale@sinenomine.net>
Reviewed-by: Benjamin Kaduk <kaduk@mit.edu>
