| Age | Commit message (Collapse) | Author |
|---|
|
Change-Id: I15b21230d52479d008a00d9e2191dda020f00925
Signed-off-by: Greg Hackmann <ghackmann@google.com>
|
|
Stop using obsolete procfs api.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
|
|
- Stop using obsolete create_proc_entry api.
- Use proc_set_user instead of directly accessing the private structure.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
|
|
Convert to use seq_read
Signed-off-by: Arve Hjønnevåg <arve@android.com>
|
|
Since (41063e9 ipv4: Early TCP socket demux), skb's can have an sk which
is not a struct sock but the smaller struct inet_timewait_sock without an
sk->sk_socket. Now we bypass sk_state == TCP_TIME_WAIT
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Avoid waking up every thread sleeping in read call on an AF_UNIX
socket during suspend and resume by calling a freezable blocking
call. Previous patches modified the freezer to avoid sending
wakeups to threads that are blocked in freezable blocking calls.
This call was selected to be converted to a freezable call because
it doesn't hold any locks or release any resources when interrupted
that might be needed by another freezing task or a kernel driver
during suspend, and is a common site where idle userspace tasks are
blocked.
Change-Id: I788246a76780ea892659526e70be018b18f646c4
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Colin Cross <ccross@android.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
|
|
NFS calls the freezable helpers with locks held, which is unsafe
and will cause lockdep warnings when 6aa9707 "lockdep: check
that no locks held at freeze time" is reapplied (it was reverted
in dbf520a). NFS shouldn't be doing this, but it has
long-running syscalls that must hold a lock but also shouldn't
block suspend. Until NFS freeze handling is rewritten to use a
signal to exit out of the critical section, add new *_unsafe
versions of the helpers that will not run the lockdep test when
6aa9707 is reapplied, and call them from NFS.
In practice the likley result of holding the lock while freezing
is that a second task blocked on the lock will never freeze,
aborting suspend, but it is possible to manufacture a case using
the cgroup freezer, the lock, and the suspend freezer to create
a deadlock. Silencing the lockdep warning here will allow
problems to be found in other drivers that may have a more
serious deadlock risk, and prevent new problems from being added.
Change-Id: Ia17d32cdd013a6517bdd5759da900970a4427170
Signed-off-by: Colin Cross <ccross@android.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
|
|
Some of the printks are in the packet handling path.
We now ratelimit the very unlikely errors to avoid
kmsg spamming.
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Removed bluetooth checks for AID_NET_BT and AID_NET_BT_ADMIN
which are not useful anymore.
This is in preparation for getting rid of all the AID_* gids.
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
In the past it would always ignore interfaces with loopback addresses.
Now we just treat them like any other.
This also helps with writing tests that check for the presence
of the qtaguid module.
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
In the past the iface_stat_fmt would only show global bytes/packets
for the skb-based numbers.
For stall detection in userspace, distinguishing tcp vs other protocols
makes it easier.
Now we report
ifname total_skb_rx_bytes total_skb_rx_packets total_skb_tx_bytes
total_skb_tx_packets {rx,tx}_{tcp,udp,ohter}_{bytes,packets}
Bug: 6818637
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
qtaguid limits what can be done with /ctrl and /stats based on group
membership.
This changes removes AID_NET_BW_STATS and AID_NET_BW_ACCT, and picks
up the groups from the gid of the matching proc entry files.
Signed-off-by: JP Abgrall <jpa@google.com>
Change-Id: I42e477adde78a12ed5eb58fbc0b277cdaadb6f94
|
|
If create_if_tag_stat fails to allocate memory (GFP_ATOMIC) the
following will happen:
qtaguid: iface_stat: tag stat alloc failed
...
kernel BUG at xt_qtaguid.c:1482!
Signed-off-by: Pontus Fuchs <pontus.fuchs@gmail.com>
|
|
qtudev_open() could return with a uid_tag_data_tree_lock held
when an kzalloc(..., GFP_ATOMIC) would fail.
Very unlikely to get triggered AND survive the mayhem of running out of mem.
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
In the past, a process could only see its own stats (uid-based summary,
and details).
Now we allow any process to see other UIDs uid-based stats, but still
hide the detailed stats.
Change-Id: I7666961ed244ac1d9359c339b048799e5db9facc
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Change-Id: Iaeca5dd2d7878c0733923ae03309a2a7b86979ca
Signed-off-by: Ashish Sharma <ashishsharma@google.com>
|
|
qtaguid tracks the device stats by monitoring when it goes up and down,
then it gets the dev_stats().
But devs don't correctly report stats (either they don't count headers
symmetrically between rx/tx, or they count internal control messages).
Now qtaguid counts the rx/tx bytes/packets during raw:prerouting and
mangle:postrouting (nat is not available in ipv6).
The results are in
/proc/net/xt_qtaguid/iface_stat_fmt
which outputs a format line (bash expansion):
ifname total_skb_{rx,tx}_{bytes,packets}
Added event counters for pre/post handling.
Added extra ctrl_*() pid/uid debugging.
Change-Id: Id84345d544ad1dd5f63e3842cab229e71d339297
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Send notifications when the label becomes active after an idle period.
Send netlink message notifications in addition to sysfs notifications.
Using a uevent with
subsystem=xt_idletimer
INTERFACE=...
STATE={active,inactive}
This is backport from common android-3.0
commit: beb914e987cbbd368988d2b94a6661cb907c4d5a
with uevent support instead of a new netlink message type.
Change-Id: I31677ef00c94b5f82c8457e5bf9e5e584c23c523
Signed-off-by: Ashish Sharma <ashishsharma@google.com>
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
When updating the stats for a given uid it would incorrectly assume
IPV4 and pick up the wrong protocol when IPV6.
Change-Id: Iea4a635012b4123bf7aa93809011b7b2040bb3d5
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
There was a case that might have seemed like new_tag_stat was not
initialized and actually used.
Added comment explaining why it was impossible, and a BUG()
in case the logic gets changed.
Change-Id: I1eddd1b6f754c08a3bf89f7e9427e5dce1dfb081
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Since rx_bytes accounting does not include Ethernet Headers in
br_input.c, excluding ETH_HLEN on the transmit path for consistent
measurement of packet length on both the Tx and Rx chains.
The clean way would be for Rx to include the eth header, but the
skb len has already been adjusted by the time the br code sees the skb.
This is only a temporary workaround until we can completely ignore or
cleanly fix the skb->len handling.
Change-Id: I910de95a4686b2119da7f1f326e2154ef31f9972
Signed-off-by: Ashish Sharma <ashishsharma@google.com>
|
|
When calling:
ipv6_find_hdr(skb, &thoff, -1, NULL)
on a fragmented packet, thoff would be left with a random
value causing callers to read random memory offsets with:
skb_header_pointer(skb, thoff, ...)
Now we force ipv6_find_hdr() to return a failure in this case.
Calling:
ipv6_find_hdr(skb, &thoff, -1, &fragoff)
will set fragoff as expected, and not return a failure.
Change-Id: Ib474e8a4267dd2b300feca325811330329684a88
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
The xt_quota2 came from
http://sourceforge.net/projects/xtables-addons/develop
It needed tweaking for it to compile within the kernel tree.
Fixed kmalloc() and create_proc_entry() invocations within
a non-interruptible context.
Removed useless copying of current quota back to the iptable's
struct matchinfo:
- those are per CPU: they will change randomly based on which
cpu gets to update the value.
- they prevent matching a rule: e.g.
-A chain -m quota2 --name q1 --quota 123
can't be followed by
-D chain -m quota2 --name q1 --quota 123
as the 123 will be compared to the struct matchinfo's quota member.
Use the NETLINK NETLINK_NFLOG family to log a single message
when the quota limit is reached.
It uses the same packet type as ipt_ULOG, but
- never copies skb data,
- uses 112 as the event number (ULOG's +1)
It doesn't log if the module param "event_num" is 0.
Change-Id: I021d3b743db3b22158cc49acb5c94d905b501492
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
The original xt_quota in the kernel is plain broken:
- counts quota at a per CPU level
(was written back when ubiquitous SMP was just a dream)
- provides no way to count across IPV4/IPV6.
This patch is the original unaltered code from:
http://sourceforge.net/projects/xtables-addons
at commit e84391ce665cef046967f796dd91026851d6bbf3
Change-Id: I19d49858840effee9ecf6cff03c23b45a97efdeb
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Allow the REJECT --reject-with icmp*blabla to also set the matching error
locally on the socket affected by the reject.
This allows the process to see an error almost as if it received it
via ICMP.
It avoids the local process who's ingress packet is rejected to have to
wait for a pseudo-eternity until some timeout kicks in.
Ideally, this should be enabled with a new iptables flag similar to
--reject-with-sock-err
For now it is enabled with CONFIG_IP*_NF_TARGET_REJECT_SKERR option.
Change-Id: I649a4fd5940029ec0b3233e5abb205da6984891e
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
This module allows tracking stats at the socket level for given UIDs.
It replaces xt_owner.
If the --uid-owner is not specified, it will just count stats based on
who the skb belongs to. This will even happen on incoming skbs as it
looks into the skb via xt_socket magic to see who owns it.
If an skb is lost, it will be assigned to uid=0.
To control what sockets of what UIDs are tagged by what, one uses:
echo t $sock_fd $accounting_tag $the_billed_uid \
> /proc/net/xt_qtaguid/ctrl
So whenever an skb belongs to a sock_fd, it will be accounted against
$the_billed_uid
and matching stats will show up under the uid with the given
$accounting_tag.
Because the number of allocations for the stats structs is not that big:
~500 apps * 32 per app
we'll just do it atomic. This avoids walking lists many times, and
the fancy worker thread handling. Slabs will grow when needed later.
It use netdevice and inetaddr notifications instead of hooks in the core dev
code to track when a device comes and goes. This removes the need for
exposed iface_stat.h.
Put procfs dirs in /proc/net/xt_qtaguid/
ctrl
stats
iface_stat/<iface>/...
The uid stats are obtainable in ./stats.
Change-Id: I01af4fd91c8de651668d3decb76d9bdc1e343919
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
The socket matching function has some nifty logic to get the struct sock
from the skb or from the connection tracker.
We export this so other xt_* can use it, similarly to ho how
xt_socket uses nf_tproxy_get_sock.
Change-Id: I11c58f59087e7f7ae09e4abd4b937cd3370fa2fd
Signed-off-by: JP Abgrall <jpa@google.com>
|
|
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
|
|
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
|
|
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
|
|
Change-Id: I6e8d838d91bebc28f4cd09dcb8b9f1de775be13d
Signed-off-by: jun.ho.lee <jun.ho.lee@samsung.com>
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
|
|
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
|
|
When enabled, tracks the frequency of network transmissions
(inbound and outbound) and buckets them accordingly.
Buckets are determined by time between network activity.
Each bucket represents the number of network transmisions that were
N sec or longer apart. Where N is defined as 1 << bucket index.
This network pattern tracking is particularly useful for wireless
networks (ie: 3G) where batching network activity closely together
is more power efficient than far apart.
New file: /proc/net/stat/activity
output:
Min Bucket(sec) Count
1 7
2 0
4 1
8 0
16 0
32 2
64 1
128 0
Change-Id: I4c4cd8627b872a55f326b1715c51bc3bdd6e8d92
Signed-off-by: Mike Chan <mike@android.com>
|
|
This improves compatbility with a lot of headset / chipset
combinations. Ideally this should not be needed.
Change-Id: I8b676701e12e416aa7d60801b9d353b15d102709
Signed-off-by: hyungseoung.yoo <hyungseoung.yoo@samsung.com>
Signed-off-by: Jaikumar Ganesh <jaikumarg@android.com>
|
|
__u16 sco_pkt_type is introduced to struct sockaddr_sco. It allows bitwise
selection of SCO/eSCO packet types. Currently those bits are:
0x0001 HV1 may be used.
0x0002 HV2 may be used.
0x0004 HV3 may be used.
0x0008 EV3 may be used.
0x0010 EV4 may be used.
0x0020 EV5 may be used.
0x0040 2-EV3 may be used.
0x0080 3-EV3 may be used.
0x0100 2-EV5 may be used.
0x0200 3-EV5 may be used.
This is similar to the Packet Type parameter in the HCI Setup Synchronous
Connection Command, except that we are not reversing the logic on the EDR bits.
This makes the use of sco_pkt_tpye forward portable for the use case of
white-listing packet types, which we expect will be the primary use case.
If sco_pkt_type is zero, or userspace uses the old struct sockaddr_sco,
then the default behavior is to allow all packet types.
Packet type selection is just a request made to the Bluetooth chipset, and
it is up to the link manager on the chipset to negiotiate and decide on the
actual packet types used. Furthermore, when a SCO/eSCO connection is eventually
made there is no way for the host stack to determine which packet type was used
(however it is possible to get the link type of SCO or eSCO).
sco_pkt_type is ignored for incoming SCO connections. It is possible
to add this in the future as a parameter to the Accept Synchronous Connection
Command, however its a little trickier because the kernel does not
currently preserve sockaddr_sco data between userspace calls to accept().
The most common use for sco_pkt_type will be to white-list only SCO packets,
which can be done with the hci.h constant SCO_ESCO_MASK.
This patch is motivated by broken Bluetooth carkits such as the Motorolo
HF850 (it claims to support eSCO, but will actually reject eSCO connections
after 5 seconds) and the 2007/2008 Infiniti G35/37 (fails to route audio
if a 2-EV5 packet type is negiotiated). With this patch userspace can maintain
a list of compatible packet types to workaround remote devices such as these.
Based on a patch by Marcel Holtmann.
Rebased to 2.6.39.
Change-Id: Ide1c89574fa4f6f1b9218282e1af17051eb86315
Signed-off-by: Nick Pelly <npelly@google.com>
|
|
This provides userspace debugging tools access to ACL flow control state.
Signed-off-by: Nick Pelly <npelly@google.com>
|
|
This is to support the Motorola HF850 carkit which reports the error code 0x10
for an eSCO attempt, even though it advertises eSCO support. Here is the
hcidump:
2009-09-22 15:40:24.492391 < HCI Command: Setup Synchronous Connection
(0x01|0x0028) plen 17
handle 1 voice setting 0x0060
2009-09-22 15:40:24.493002 > HCI Event: Command Status (0x0f) plen 4
Setup Synchronous Connection (0x01|0x0028) status 0x00 ncmd 1
2009-09-22 15:40:30.594869 > HCI Event: Synchronous Connect Complete (0x2c)
plen 17
status 0x10 handle 257 bdaddr 00:50:CD:20:C6:84 type eSCO
Error: Connection Accept Timeout Exceeded
With this patch we will retry with a SCO connection, which succeeds.
Unfortunately the Moto HF850 also takes 5 seconds to return the error for the
eSCO attempt, so it will still take 5 seconds to fallback to SCO with this
patch.
Signed-off-by: Nick Pelly <npelly@google.com>
|
|
This fixes a bug where shutdown() and close() on a rfcomm socket during ACL
connection would not cause HCI Create Connection Cancel.
Signed-off-by: Nick Pelly <npelly@google.com>
|
|
Some platforms do not want to power down rfkill devices on suspend.
Change-Id: I62a11630521c636d54a4a02ab9037a43435925f5
Signed-off-by: Nick Pelly <npelly@google.com>
|
|
Signed-off-by: Chia-chi Yeh <chiachi@android.com>
|
|
Signed-off-by: Mike Chan <mike@android.com>
|
|
Add a family of knobs to /sys/kernel/ipv4 for controlling the TCP window size:
tcp_wmem_min
tcp_wmem_def
tcp_wmem_max
tcp_rmem_min
tcp_rmem_def
tcp_rmem_max
This six values mirror the sysctl knobs in /proc/sys/net/ipv4/tcp_wmem and
/proc/sys/net/ipv4/tcp_rmem.
Sysfs, unlike sysctl, allows us to set and manage the files' permissions and
owners.
Signed-off-by: Robert Love <rlove@google.com>
|
|
Introduce a new socket ioctl, SIOCKILLADDR, that nukes all sockets
bound to the same local address. This is useful in situations with
dynamic IPs, to kill stuck connections.
Signed-off-by: Brian Swetland <swetland@google.com>
net: fix tcp_v4_nuke_addr
Signed-off-by: Dima Zavin <dima@android.com>
net: ipv4: Fix a spinlock recursion bug in tcp_v4_nuke.
We can't hold the lock while calling to tcp_done(), so we drop
it before calling. We then have to start at the top of the chain again.
Signed-off-by: Dima Zavin <dima@android.com>
net: ipv4: Fix race in tcp_v4_nuke_addr().
To fix a recursive deadlock in 2.6.29, we stopped holding the hash table lock
across tcp_done() calls. This fixed the deadlock, but introduced a race where
the socket could die or change state.
Fix: Before unlocking the hash table, we grab a reference to the socket. We
can then unlock the hash table without risk of the socket going away. We then
lock the socket, which is safe because it is pinned. We can then call
tcp_done() without recursive deadlock and without race. Upon return, we unlock
the socket and then unpin it, killing it.
Change-Id: Idcdae072b48238b01bdbc8823b60310f1976e045
Signed-off-by: Robert Love <rlove@google.com>
Acked-by: Dima Zavin <dima@android.com>
ipv4: disable bottom halves around call to tcp_done().
Signed-off-by: Robert Love <rlove@google.com>
Signed-off-by: Colin Cross <ccross@android.com>
ipv4: Move sk_error_report inside bh_lock_sock in tcp_v4_nuke_addr
When sk_error_report is called, it wakes up the user-space thread, which then
calls tcp_close. When the tcp_close is interrupted by the tcp_v4_nuke_addr
ioctl thread running tcp_done, it leaks 392 bytes and triggers a WARN_ON.
This patch moves the call to sk_error_report inside the bh_lock_sock, which
matches the locking used in tcp_v4_err.
Signed-off-by: Colin Cross <ccross@android.com>
|
|
With CONFIG_ANDROID_PARANOID_NETWORK, require specific uids/gids to instantiate
network sockets.
Signed-off-by: Robert Love <rlove@google.com>
paranoid networking: Use in_egroup_p() to check group membership
The previous group_search() caused trouble for partners with module builds.
in_egroup_p() is also cleaner.
Signed-off-by: Nick Pelly <npelly@google.com>
Fix 2.6.29 build.
Signed-off-by: Arve Hjønnevåg <arve@android.com>
net: Fix compilation of the IPv6 module
Fix compilation of the IPv6 module -- current->euid does not exist anymore,
current_euid() is what needs to be used.
Signed-off-by: Steinar H. Gunderson <sesse@google.com>
|
|
key_notify_sa_flush() and key_notify_policy_flush() miss to initialize
the sadb_msg_reserved member of the broadcasted message and thereby
leak 2 bytes of heap memory to listeners. Fix that.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
It's possible to use AF_INET6 sockets and to connect to an IPv4
destination. After this, socket dst cache is a pointer to a rtable,
not rt6_info.
ip6_sk_dst_check() should check the socket dst cache is IPv6, or else
various corruptions/crashes can happen.
Dave Jones can reproduce immediate crash with
trinity -q -l off -n -c sendmsg -c connect
With help from Hannes Frederic Sowa
Reported-by: Dave Jones <davej@redhat.com>
Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When the kernel (compiled with CONFIG_PREEMPT=n) is performing the
rename of a network interface, it can end up waiting for a workqueue
to complete. If userland is able to invoke a SIOCGIFNAME ioctl or a
SO_BINDTODEVICE getsockopt in between, the kernel will deadlock due to
the fact that read_secklock_begin() will spin forever waiting for the
writer process (the one doing the interface rename) to update the
devnet_rename_seq sequence.
This patch fixes the problem by adding a helper (netdev_get_name())
and using it in the code handling the SIOCGIFNAME ioctl and
SO_BINDTODEVICE setsockopt.
The netdev_get_name() helper uses raw_seqcount_begin() to avoid
spinning forever, waiting for devnet_rename_seq->sequence to become
even. cond_resched() is used in the contended case, before retrying
the access to give the writer process a chance to finish.
The use of raw_seqcount_begin() will incur some unneeded work in the
reader process in the contended case, but this is better than
deadlocking the system.
Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We should check the return value of ipv6_get_lladdr in inet6_set_iftoken.
A possible situation, which could leave ll_addr unassigned is, when
the user removed her link-local address but a global scoped address was
already set. In this case the interface would still be IF_READY and not
dead. In that case the RS source address is some value from the stack.
v2: Daniel Borkmann noted a small indent inconstancy; no semantic
changes.
Cc: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Daniel Borkmann <dborkman@redhat.com>
Reviewed-by: Flavio Leitner <fbl@redhat.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
commit 68c331631143 ("v4 GRE: Add TCP segmentation offload for GRE")
added a possible skb leak, because it frees only the head of segment
list, in case a skb_linearize() call fails.
This patch adds a kfree_skb_list() helper to fix the bug.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Pravin B Shelar <pshelar@nicira.com>
Cc: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless
John W. Linville says:
====================
A few more late-breaking fixes hoping for 3.10...
Regarding the Bluetooth fix, Gustavo says:
"A important fix to 3.10, this patch fixes an issues that was preventing
the l2cap info response command to be handled properly."
Also for that Bluetooth fix, Johan adds:
"Once the code gives up parsing this PDU it also gives up essential
parts of the L2CAP connection creation process, i.e. without this
patch the stack will fail to establish connections properly."
Moving onto ath9k, Felix Fietkau fixes an RCU locking issue in
the transmit path. As for ath9k_htc, Sujith Manoharan fixes some
authentication timeouts by ensuring that a chip reset is done when
IDLE is turned off.
I think these are all micro-fixes that shouldn't cause any trouble.
Please let me know if there are problems!
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
