freebsd.git
9 years agoRemove old comment, we already do 802.1q tagging.
araujo [Thu, 18 Sep 2014 03:09:34 +0000 (03:09 +0000)]
Remove old comment, we already do 802.1q tagging.

Phabric: D797
Reviewed by: kevlo
Approved by: kevlo
Sponsored by: QNAP Systems Inc.

9 years agoThe lagg(4) interface is based on trunk(4) interface from OpenBSD.
araujo [Thu, 18 Sep 2014 02:22:02 +0000 (02:22 +0000)]
The lagg(4) interface is based on trunk(4) interface from OpenBSD.
The FreeBSD is the only system that has the FEC protocol, that is a simple alias
to loadbalance protocol and does not implement the ancient Cisco FEC standard.

From now on, we remove the fec protocol from the documentation and keep the FEC
code only for compatibility.

Phabric: D539
Reviewed by: glebius, thompsa
Approved by: glebius
Sponsored by: QNAP Systems Inc.

9 years agoAdd laggproto broadcast, it allows sends frames to all ports of the lagg(4) group
araujo [Thu, 18 Sep 2014 02:12:48 +0000 (02:12 +0000)]
Add laggproto broadcast, it allows sends frames to all ports of the lagg(4) group
and receives frames on any port of the lagg(4).

Phabric: D549
Reviewed by: glebius, thompsa
Approved by: glebius
Obtained from: OpenBSD
Sponsored by: QNAP Systems Inc.

9 years agoFix a kernel panic when unloading isp(4).
will [Thu, 18 Sep 2014 02:01:36 +0000 (02:01 +0000)]
Fix a kernel panic when unloading isp(4).

In the current implementation, the isp_kthread() threads never exit.

The target threads do have an exit mode from isp_attach(), but it is
not invoked from isp_detach().

Ensure isp_detach() notifies threads started for each channel, such
that they exit before their parent device softc detaches, and thus
before the module does.  Otherwise, a page fault panic occurs later in:

sysctl_kern_proc
  sysctl_out_proc
    kern_proc_out
      fill_kinfo_proc
        fill_kinfo_thread
          strlcpy(kp->ki_wmesg, td->td_wmesg, sizeof(kp->ki_wmesg));

For isp_kthread() (and isp(4) target threads), td->td_wmesg references
now-unmapped memory after the module has been unloaded.  These threads
are typically msleep()ing at the time of unload, but they could also
attempt to execute now-unmapped code segments.

MFC after: 1 month
Sponsored by: Spectra Logic
MFSpectraBSD: r1070921 on 2014/06/22 13:01:17

9 years agoRoot the lib32 object tree under the overall object tree.
will [Thu, 18 Sep 2014 01:57:36 +0000 (01:57 +0000)]
Root the lib32 object tree under the overall object tree.

This enables a common root directory for all object files for a given tree,
which eases sharing a common MAKEOBJDIRPREFIX, and cleaning up of object trees.

In particular, one can simply (from the source directory) rm -rf /usr/obj$(pwd)
to destroy all object files for it.  Or to copy/sync files, etc.

Reviewed by: bdrewery
CR: https://reviews.freebsd.org/D796
MFC after: 1 month
Sponsored by: Spectra Logic

9 years agoFor ECORE_DBG_BREAK_IF() ECORE_BUG() ECORE_BUG_ON() check bxe_debug flag before
davidcs [Wed, 17 Sep 2014 22:49:29 +0000 (22:49 +0000)]
For ECORE_DBG_BREAK_IF() ECORE_BUG() ECORE_BUG_ON() check bxe_debug flag before
printing error message.

MFC after:5 days

9 years agoRemove clean option
davidcs [Wed, 17 Sep 2014 22:26:01 +0000 (22:26 +0000)]
Remove clean option

MFC after:5 days

9 years agoRemove clean option
davidcs [Wed, 17 Sep 2014 22:24:51 +0000 (22:24 +0000)]
Remove clean option

MFC after:5 days

9 years agoThe vm_mmap_cdev() explicitely converts absence of both MAP_SHARED and
kib [Wed, 17 Sep 2014 21:04:50 +0000 (21:04 +0000)]
The vm_mmap_cdev() explicitely converts absence of both MAP_SHARED and
MAP_PRIVATE flags to MAP_SHARED.  Apparently, some code in tree, in
particular, libgeom, relied on this behaviour, see r271721.  For
regular file types, the absence of the flags is interpreted as
MAP_PRIVATE, and libc nlist used this (fixed in r271723).

Allow the implicit flags for legacy binaries.  Bump __FreeBSD_version
to get the ABI note on new binaries to check for in mmap code.

Remove the test for presence of one of the MAP_ANON, MAP_SHARED or
MAP_PRIVATE flags before fget_mmap().  For MAP_ANON, we already verify
that passed fd == -1.  For fd != -1, test after fget_mmap() (for newer
binaries) covers the case.

Reported by: bdrewery, pho
Reviewed by: jhb
Sponsored by: The FreeBSD Foundation

9 years agoExplicitly set MAP_PRIVATE to avoid [EINVAL] after r271635.
bdrewery [Wed, 17 Sep 2014 20:26:27 +0000 (20:26 +0000)]
Explicitly set MAP_PRIVATE to avoid [EINVAL] after r271635.

X-MFC-With: r271635
Reviewed by: jhb

9 years ago- Remove mention of MAP_INHERIT. It hasn't been implemented for thirteen
jhb [Wed, 17 Sep 2014 19:45:34 +0000 (19:45 +0000)]
- Remove mention of MAP_INHERIT.  It hasn't been implemented for thirteen
  years.
- Remove mention of unimplemented MAP_SWAP.  There are no future plans to
  implement it.

Submitted by: alc (2)

9 years agoExplicitly specify MAP_SHARED when mapping the stats file descriptor.
jhb [Wed, 17 Sep 2014 19:37:58 +0000 (19:37 +0000)]
Explicitly specify MAP_SHARED when mapping the stats file descriptor.

Reviewed by: kib
MFC after: 1 week

9 years agoIf fgets(3) fails in getbounds(), show strerror(3) if not an EOF. Also fix
bdrewery [Wed, 17 Sep 2014 19:09:58 +0000 (19:09 +0000)]
If fgets(3) fails in getbounds(), show strerror(3) if not an EOF. Also fix
a FILE* leak in getbounds().

Submitted by: Conrad Meyer <conrad.meyer@isilon.com>
PR: 192032
Sponsored by: EMC / Isilon Storage Division
MFC after: 1 week

9 years agoFor targets that have a signed zero, lgamma_r(-0, &signgamp) should
kargl [Wed, 17 Sep 2014 19:01:22 +0000 (19:01 +0000)]
For targets that have a signed zero, lgamma_r(-0, &signgamp) should
set signgamp = -1.

Submitted by: enh at google dot com (e_lgamma[f]_r.c)

9 years agoCorrect a comment
bdrewery [Wed, 17 Sep 2014 18:59:25 +0000 (18:59 +0000)]
Correct a comment

9 years agoRevert unrelated changes accidentally committed in r271192.
jhb [Wed, 17 Sep 2014 18:55:39 +0000 (18:55 +0000)]
Revert unrelated changes accidentally committed in r271192.

9 years agoPresence of any VM_PROT bits in the permission argument on x86 implies
kib [Wed, 17 Sep 2014 18:49:57 +0000 (18:49 +0000)]
Presence of any VM_PROT bits in the permission argument on x86 implies
that the entry is readable and valid.

Reported by: markj
Submitted by: alc
Tested by: pho (previous version), markj
MFC after: 3 days

9 years agoReprase r271616 comments.
mav [Wed, 17 Sep 2014 17:43:32 +0000 (17:43 +0000)]
Reprase r271616 comments.

Submitted by: alc
MFC after: 1 month

9 years agoMigrate ie->ie_assign_cpu and associated code to use an int for CPU rather
adrian [Wed, 17 Sep 2014 17:33:22 +0000 (17:33 +0000)]
Migrate ie->ie_assign_cpu and associated code to use an int for CPU rather
than u_char.

Migrate post_filter to use an int for a CPU rather than u_char.

Change intr_event_bind() to use an int for CPU rather than u_char.

It touches the ppc, sparc64, arm and mips machdep code but it should
(hah!) be a no-op.

Tested:

* i386, AMD64 laptops

Reviewed by: jhb

9 years agoUpdate the bhyve(8) manual to reflect that it is no
gjb [Wed, 17 Sep 2014 16:45:20 +0000 (16:45 +0000)]
Update the bhyve(8) manual to reflect that it is no
longer considered 'experimental.'

Reviewed by: grehan
MFC after: 3 days
Sponsored by: The FreeBSD Foundation

9 years agodrm/i915: Add HW context support
dumbbell [Wed, 17 Sep 2014 08:28:50 +0000 (08:28 +0000)]
drm/i915: Add HW context support

This feature is required by Mesa 9.2+. Without this, a GL application
crashes with the following message:
    # glxinfo
    name of display: :0.0
    Gen6+ requires Kernel 3.6 or later.
    Assertion failed: (ctx->Version > 0), function handle_first_current,
      file ../../src/mesa/main/context.c, line 1498.
    Abort (core dumped)

Now, Mesa 10.2.4 and 10.3-rc3 works fine:
    # glxinfo
    name of display: :0
    display: :0  screen: 0
    direct rendering: Yes
    ...
    OpenGL renderer string: Mesa DRI Intel(R) 965GM
    OpenGL version string: 2.1 Mesa 10.2.4
    ...

The code was imported from Linux 3.8.13.

Reviewed by: kib@
Tested by: kwm@, danfe@, Henry Hu,
Lundberg, Johannes <johannes@brilliantservice.co.jp>,
Johannes Dieterich <dieterich.joh@gmail.com>,
Lutz Bichler <lutz.bichler@gmail.com>,
MFC after: 3 days
Relnotes: yes

9 years agoFix LUN discovery for targets that don't support REPORT_LUNS, broken
trasz [Wed, 17 Sep 2014 07:55:23 +0000 (07:55 +0000)]
Fix LUN discovery for targets that don't support REPORT_LUNS, broken
in r263741.  At least with CTL (slightly modified to report SPC2) there
is still some problem: it doesn't seem to find LUNs higher than 7.

Sponsored by: The FreeBSD Foundation

9 years agoFix tpc_create_token() introduced in r269497 to encode CREATOR LOGICAL UNIT
mav [Wed, 17 Sep 2014 07:08:59 +0000 (07:08 +0000)]
Fix tpc_create_token() introduced in r269497 to encode CREATOR LOGICAL UNIT
DESCRIPTOR field as Identification Descriptor CSCD descriptor, not just as
Identification Descriptor.

MFC after: 3 days

9 years agoFix typo in defined ROD types in r269497.
mav [Wed, 17 Sep 2014 06:46:37 +0000 (06:46 +0000)]
Fix typo in defined ROD types in r269497.

MFC after: 3 days

9 years agoImplement a workaround to allow this test program to be compiled with clang.
markj [Wed, 17 Sep 2014 04:02:56 +0000 (04:02 +0000)]
Implement a workaround to allow this test program to be compiled with clang.
It seems that if a pragma is used to define a weak alias for a local
function, the pragma must appear after the function is defined.

PR: 193056
MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division

9 years agoFix a number of typos and programming errors in the userland CTF tests. It
markj [Wed, 17 Sep 2014 03:40:49 +0000 (03:40 +0000)]
Fix a number of typos and programming errors in the userland CTF tests. It
seems that they would only pass by chance on illumos; on FreeBSD, they still
fail since userland CTF is not yet supported.

MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division

9 years agoFix the stack tracing for dtrace/powerpc.
jhibbits [Wed, 17 Sep 2014 02:43:47 +0000 (02:43 +0000)]
Fix the stack tracing for dtrace/powerpc.

Summary:
Fix the stack tracing for dtrace/powerpc by using the trapexit/asttrapexit
return address sentinels instead of checking within the kernel address space.

As part of this, I had to add new inline functions.  FBT traces the kernel, so
we have to have special case handling for this, since a trap will create a full
new trap frame, and there's no way to pass around the 'real' stack.  I handle
this by special-casing 'aframes == 0' with the trap frame.  If aframes counts
out to the trap frame, then assume we're looking for the full kernel trap frame,
so switch to the real stack pointer.

Test Plan: Tested on powerpc64

Reviewers: rpaulo, markj, nwhitehorn

Reviewed By: markj, nwhitehorn

Differential Revision: https://reviews.freebsd.org/D788

MFC after: 3 week
Relnotes: Yes

9 years agoUse a devd event to start hv_kvpd instead of doing so in rc.d script.
delphij [Wed, 17 Sep 2014 02:32:22 +0000 (02:32 +0000)]
Use a devd event to start hv_kvpd instead of doing so in rc.d script.
This is cleaner and eliminates the unneeded startup of KVP daemon on
systems that do not run as a Hyper-V guest.

Submitted by: hrs
X-MFC-with: 271493, 271688, 271699

9 years agoFix some incorrect endianness checks.
markj [Wed, 17 Sep 2014 00:54:00 +0000 (00:54 +0000)]
Fix some incorrect endianness checks.

MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division

9 years agoFix elfdump(1) usage in the userland CTF tests.
markj [Tue, 16 Sep 2014 22:54:15 +0000 (22:54 +0000)]
Fix elfdump(1) usage in the userland CTF tests.

PR: 193110
MFC after: 1 week
Sponsored by: EMC / Isilon Storage Division

9 years agoFix a typo.
markj [Tue, 16 Sep 2014 22:26:01 +0000 (22:26 +0000)]
Fix a typo.

MFC after: 3 days

9 years ago* Fix if_omcast handling
melifaro [Tue, 16 Sep 2014 21:48:48 +0000 (21:48 +0000)]
* Fix if_omcast handling
* Convert if_oerrors to pcpu.

Suggested by: glebius
MFC after: 2 weeks

9 years agoSuch typo, many email, very spelling. wow.
sbruno [Tue, 16 Sep 2014 20:48:13 +0000 (20:48 +0000)]
Such typo, many email, very spelling.  wow.

MFC with 271688 271493
Submitted by: ngie ian
MFC after: 2 weeks
Relnotes: yes

9 years agoAdd proper disable/enable hooks to the default scripts so that this is only
sbruno [Tue, 16 Sep 2014 20:02:16 +0000 (20:02 +0000)]
Add proper disable/enable hooks to the default scripts so that this is only
run when asked for by the user.  Right now, hv_kvpd is run on every boot.

Don't do that.

Add hv_kvpd_enable= for this script to be run.

MFC with 271493

MFC after: 2 weeks
Relnotes: yes

9 years agoBump __FreeBSD_version after SA-14:19.tcp
gjb [Tue, 16 Sep 2014 19:20:06 +0000 (19:20 +0000)]
Bump __FreeBSD_version after SA-14:19.tcp

Approved by: re (implicit)
Sponsored by: The FreeBSD Foundation

9 years agovt(4): Use vt_fb_drawrect() and vt_fb_setpixel() in all vt_fb-derivative
dumbbell [Tue, 16 Sep 2014 18:02:24 +0000 (18:02 +0000)]
vt(4): Use vt_fb_drawrect() and vt_fb_setpixel() in all vt_fb-derivative

Review: https://reviews.freebsd.org/D789
Reviewed by: nwhitehorn
Approved by: nwhitehorn
MFC after: 2 days

9 years agovt(4): Fix a LOR which occurs during a call to vt_upgrade()
dumbbell [Tue, 16 Sep 2014 17:42:37 +0000 (17:42 +0000)]
vt(4): Fix a LOR which occurs during a call to vt_upgrade()

Reported by: kib@
Review: https://reviews.freebsd.org/D785
Reviewed by: ray@
Approved by: ray@
MFC after: 2 days

9 years agoPermit MAP_RENAME and MAP_NORESERVE for now. These flags should be removed, but...
jhb [Tue, 16 Sep 2014 17:21:06 +0000 (17:21 +0000)]
Permit MAP_RENAME and MAP_NORESERVE for now.  These flags should be removed, but at least
Chromium and OpenJDK use MAP_NORESERVE.

9 years agoFix some extra whitespace noticed when reviewing git diffs.
brooks [Tue, 16 Sep 2014 17:07:32 +0000 (17:07 +0000)]
Fix some extra whitespace noticed when reviewing git diffs.

9 years agoMerge atse(4) interrupt handling and race condition fixes from cheribsd:
bz [Tue, 16 Sep 2014 15:45:53 +0000 (15:45 +0000)]
Merge atse(4) interrupt handling and race condition fixes from cheribsd:

commit 8bd88585ed8e3f7def0d780a1bc30d96fe642b9c

    Rework atse_rx_cycles handling: count packets instead of fills, and use the
    limit only when polling, not when in interrupt mode.  Otherwise, we may
    stop reading the FIFO midpacket and clear the event mask even though the
    FIFO still has data to read, which could stall receive when a large packet
    arrives.  Add a comment about races in the Altera FIFO interface: we may
    need to do a little more work to handle races than we are.

commit 20b39086cc612f8874dc9e6ef4c0c2eb777ba92a

    Use 'sizeof(data)' rather than '4' when checking an mbuf bound, as is the
    case for adjusting length/etc.

commit e18953174a265f40e9ba60d76af7d288927f5382

    Break out atse_intr() into two separate routines, one for each of the two
    interrupt sources: receive and transmit.

commit 6deedb43246ab3f9f597918361831fbab7fac4ce

    For the RX interrupt, take interest only in ALMOSTEMPTY and OVERFLOW.
    For the TX interrupt, take interest only in ALMOSTFULL and UNDERFLOW.

    Perform TX atse_start_locked() once rather than twice in TX interrupt
    handling -- and only if !FULL, rather than unconditionally.

commit 12601972ba08d4380201a74f5b967bdaeb23092c

    Experimentation suggests that the Altera Triple-Speed Ethernet documentation
    is incorrect and bits in the event and interrupt-enable registers are not
    irrationally rearranged relative to the status register.

commit 3cff2ffad769289fce3a728152e7be09405385d8

    Substantially rework interrupt handling in the atse(4) driver:

    - Introduce a new macro ATSE_TX_PENDING() which checks whether there is
      any pending data to transmit, either in an in-progress packet or in
      the TX queue.
    - Introduce new ATSE_RX_STATUS_READ() and ATSE_TX_STAUTS_WRITE() macros
      that query the FIFO status registers rather than event registers,
      offering level- rather than edge-triggered FIFO conditions.
    - For RX, interrupt only on full/overflow/underflow; for TX, interrupt
      only on empty/overflow/underflow.
    - Add new ATSE_RX_INTR_READ() and ATSE_RX_INTR_WRITE() macros useful for
      debugging interrupt behaviour.
    - Add a debug.atse_intr_debug_enable sysctl that causes various pieces
      of FIFO state to be printed out on each RX or TX interrupt.  This is
      disabled by default but good to turn on if the interface appears to
      wedge.  Also print debugging information when polling.
    - In the watchdog handler, do receive, not just transmit, processing, to
      ensure that the rx, not just tx, queue is being handled -- and, in
      particular, will be drained such that interrupts can resume.
    - Rework both atse_rx_intr() and atse_tx_intr() to eliminate many race
      conditions, and add comments on why various things are in various
      orders.  Interactions between modifications to the event and interrupt
      masks are quite subtle indeed, and we must actively check for a number
      of races (e.g., event mask cleared; packet arrives; interrupts enabled).
      We also now use the status registers rather than event registers for
      FIFO status checks to avoid other races; we continue to use event
      registers for underflow/overflow.

    With this change, interrupt-driven operation of atse appears (for the
    time being) robust.

commit 3393bbff5c68a4e61699f9b4a62af5d2a5f918f8

    atse: Fix build after 3cff2ffa

Obtained from: cheribsd
Submitted by: rwatson, emaste
Sponsored by: DARPA/AFRL
MFC after: 3 days

9 years agoSet a static revision of the Crochet checkout to avoid
gjb [Tue, 16 Sep 2014 15:35:13 +0000 (15:35 +0000)]
Set a static revision of the Crochet checkout to avoid
surprise build failures.

MFC after: 3 days
X-10.1-MFC: yes
X-MFC-With: r271676, r271677
Sponsored by: The FreeBSD Foundation

9 years agoCatch up with Crochet changes to fix the WANDBOARD-QUAD
gjb [Tue, 16 Sep 2014 15:31:50 +0000 (15:31 +0000)]
Catch up with Crochet changes to fix the WANDBOARD-QUAD
build.

Tested on: stable/10@r271618
MFC after: 3 days
X-10.1-MFC: yes
X-MFC-With: r271676
Sponsored by: The FreeBSD Foundation

9 years agoCatch up with Crochet changes to fix the BEAGLEBONE
gjb [Tue, 16 Sep 2014 15:31:12 +0000 (15:31 +0000)]
Catch up with Crochet changes to fix the BEAGLEBONE
build.

Tested on: stable/10@r271618
MFC after: 3 days
X-10.1-MFC: yes
Sponsored by: The FreeBSD Foundation

9 years agoFix source address selection on unbound sockets in the presence of multiple
asomers [Tue, 16 Sep 2014 15:28:19 +0000 (15:28 +0000)]
Fix source address selection on unbound sockets in the presence of multiple
fibs. Use the mbuf's or the socket's fib instead of RT_ALL_FIBS. Fixes PR
187553. Also fixes netperf's UDP_STREAM test on a nondefault fib.

sys/netinet/ip_output.c
In ip_output, lookup the source address using the mbuf's fib instead
of RT_ALL_FIBS.

sys/netinet/in_pcb.c
in in_pcbladdr, lookup the source address using the socket's fib,
because we don't seem to have the mbuf fib. They should be the same,
though.

tests/sys/net/fibs_test.sh
Clear the expected failure on udp_dontroute.

PR: 187553
CR: https://reviews.freebsd.org/D772
MFC after: 3 weeks
Sponsored by: Spectra Logic

9 years agoAdd a explict cast to silence a warning when building
tuexen [Tue, 16 Sep 2014 14:39:24 +0000 (14:39 +0000)]
Add a explict cast to silence a warning when building
the userland stack on Windows.
This issue was reported by Peter Kasting from Google.

MFC after: 3 days

9 years agoUse a consistent type for the number of HMAC algorithms.
tuexen [Tue, 16 Sep 2014 14:20:33 +0000 (14:20 +0000)]
Use a consistent type for the number of HMAC algorithms.
This fixes a bug which resulted in a warning on the userland
stack, when compiled on Windows.
Thanks to Peter Kasting from Google for reporting the issue and
provinding a potential fix.

MFC after: 3 days

9 years agoSmall cleanup which addresses a warning regaring the truncation
tuexen [Tue, 16 Sep 2014 13:48:46 +0000 (13:48 +0000)]
Small cleanup which addresses a warning regaring the truncation
of a 64-bit entity to a 32-bit entity. This issue was reported by
Peter Kasting from Google.

MFC after: 3 days

9 years agoFreeBSD-SA-14:19.tcp raised attention to the state of our stack
glebius [Tue, 16 Sep 2014 11:07:25 +0000 (11:07 +0000)]
FreeBSD-SA-14:19.tcp raised attention to the state of our stack
towards blind SYN/RST spoofed attack.

Originally our stack used in-window checks for incoming SYN/RST
as proposed by RFC793. Later, circa 2003 the RST attack was
mitigated using the technique described in P. Watson
"Slipping in the window" paper [1].

After that, the checks were only relaxed for the sake of
compatibility with some buggy TCP stacks. First, r192912
introduced the vulnerability, just fixed by aforementioned SA.
Second, r167310 had slightly relaxed the default RST checks,
instead of utilizing net.inet.tcp.insecure_rst sysctl.

In 2010 a new technique for mitigation of these attacks was
proposed in RFC5961 [2]. The idea is to send a "challenge ACK"
packet to the peer, to verify that packet arrived isn't spoofed.
If peer receives challenge ACK it should regenerate its RST or
SYN with correct sequence number. This should not only protect
against attacks, but also improve communication with broken
stacks, so authors of reverted r167310 and r192912 won't be
disappointed.

[1] http://bandwidthco.com/whitepapers/netforensics/tcpip/TCP Reset Attacks.pdf
[2] http://www.rfc-editor.org/rfc/rfc5961.txt

Changes made:

o Revert r167310.
o Implement "challenge ACK" protection as specificed in RFC5961
  against RST attack. On by default.
  - Carefully preserve r138098, which handles empty window edge
    case, not described by the RFC.
  - Update net.inet.tcp.insecure_rst description.
o Implement "challenge ACK" protection as specificed in RFC5961
  against SYN attack. On by default.
  - Provide net.inet.tcp.insecure_syn sysctl, to turn off
    RFC5961 protection.

The changes were tested at Netflix. The tested box didn't show
any anomalies compared to control box, except slightly increased
number of TCP connection in LAST_ACK state.

Reviewed by: rrs
Sponsored by: Netflix
Sponsored by: Nginx, Inc.

9 years agoMake a type conversion explicit. When compiling this code on
tuexen [Tue, 16 Sep 2014 10:57:55 +0000 (10:57 +0000)]
Make a type conversion explicit. When compiling this code on
Windows as part of the SCTP userland stack, this fixes a
warning reported by Peter Kasting from Google.

MFC after: 3 days

9 years agoFix Denial of Service in TCP packet processing.
delphij [Tue, 16 Sep 2014 09:48:24 +0000 (09:48 +0000)]
Fix Denial of Service in TCP packet processing.

Submitted by: glebius
Security: FreeBSD-SA-14:19.tcp

9 years agoThe MTU is handled as a 32-bit entity within the SCTP stack.
tuexen [Tue, 16 Sep 2014 09:22:43 +0000 (09:22 +0000)]
The MTU is handled as a 32-bit entity within the SCTP stack.
This was reported by Peter Kasting from Google.

MFC after: 3 days

9 years agoCache GELI passphrases entered at the console during the boot process,
cperciva [Tue, 16 Sep 2014 08:40:52 +0000 (08:40 +0000)]
Cache GELI passphrases entered at the console during the boot process,
in order to improve user-friendliness when a system has multiple disks
encrypted using the same passphrase.

When examining a new GELI provider, the most recently used passphrase
will be attempted before prompting for a passphrase; and whenever a
passphrase is entered, it is cached for later reference.  When the root
disk is mounted, the cached passphrase is zeroed (triggered by the
"mountroot" event), in order to minimize the possibility of leakage
of passphrases.  (After root is mounted, the "taste and prompt for
passphrases on the console" code path is disabled, so there is no
potential for a passphrase to be stored after the zeroing takes place.)

This behaviour can be disabled by setting kern.geom.eli.boot_passcache=0.

Reviewed by: pjd, dteske, allanjude
MFC after: 7 days

9 years agoFix a typo; master server for iprop service should be singular.
hrs [Tue, 16 Sep 2014 05:45:38 +0000 (05:45 +0000)]
Fix a typo; master server for iprop service should be singular.

9 years agoEnsure the correct software IPv4 hash is done based on the configured
adrian [Tue, 16 Sep 2014 03:26:42 +0000 (03:26 +0000)]
Ensure the correct software IPv4 hash is done based on the configured
RSS parameters, rather than assuming we're hashing IPv4+UDP and IPv4+TCP.

9 years agoModify cpuset_setithread() to take a CPU ID as an integer, not a char.
adrian [Tue, 16 Sep 2014 01:21:47 +0000 (01:21 +0000)]
Modify cpuset_setithread() to take a CPU ID as an integer, not a char.

We're going to end up having > 254 CPUs at some point.

9 years agoBump __FreeBSD_version for the change made in r271655
ngie [Tue, 16 Sep 2014 00:59:56 +0000 (00:59 +0000)]
Bump __FreeBSD_version for the change made in r271655

PR: 181155
Sponsored by: EMC / Isilon Storage Division

9 years agoBump .Dd for the content change done to access(2) in r271655
ngie [Tue, 16 Sep 2014 00:59:08 +0000 (00:59 +0000)]
Bump .Dd for the content change done to access(2) in r271655

PR: 181155
Sponsored by: EMC / Isilon Storage Division

9 years agoValidate the mode argument in access, eaccess, and faccessat for optional
ngie [Tue, 16 Sep 2014 00:56:47 +0000 (00:56 +0000)]
Validate the mode argument in access, eaccess, and faccessat for optional
POSIX compliance and to improve compatibility with Linux and NetBSD

The issue was identified with lib/libc/sys/t_access:access_inval from
NetBSD

Update the manpage accordingly

PR: 181155
Reviewed by: jilles (code), jmmv (code), wblock (manpage), wollman (code)
MFC after: 4 weeks
Phabric: D678 (code), D786 (manpage)
Sponsored by: EMC / Isilon Storage Division

9 years ago* Makefile:
kargl [Mon, 15 Sep 2014 23:21:57 +0000 (23:21 +0000)]
* Makefile:
  . Hook e_lgammal[_r].c to the build.
  . Create man page links for lgammal[-r].3.

* Symbol.map:
  . Sort lgammal to its rightful place.
  . Add FBSD_1.4 section for the new lgamal_r symbol.

* ld128/e_lgammal_r.c:
  . 128-bit implementataion of lgammal_r().

* ld80/e_lgammal_r.c:
  . Intel 80-bit format implementation of lgammal_r().

* src/e_lgamma.c:
  . Expose lgammal as a weak reference to lgamma for platforms
    where long double is mapped to double.

* src/e_lgamma_r.c:
  . Use integer literal constants instead of real literal constants.
    Let compiler(s) do the job of conversion to the appropriate type.
  . Expose lgammal_r as a weak reference to lgamma_r for platforms
    where long double is mapped to double.

* src/e_lgammaf_r.c:
  . Fixed the Cygnus Support conversion of e_lgamma_r.c to float.
    This includes the generation of new polynomial and rational
    approximations with fewer terms.  For each approximation, include
    a comment on an estimate of the accuracy over the relevant domain.
  . Use integer literal constants instead of real literal constants.
    Let compiler(s) do the job of conversion to the appropriate type.
    This allows the removal of several explicit casts of double values
    to float.

* src/e_lgammal.c:
  . Wrapper for lgammal() about lgammal_r().

* src/imprecise.c:
  . Remove the lgamma.

* src/math.h:
  . Add a prototype for lgammal_r().

* man/lgamma.3:
  . Document the new functions.

Reviewed by: bde

9 years agodocument mqueuefs is required for mq_open...
jmg [Mon, 15 Sep 2014 22:32:35 +0000 (22:32 +0000)]
document mqueuefs is required for mq_open...

9 years agoDisable flow-director support until it's been debugged and verified.
adrian [Mon, 15 Sep 2014 21:09:19 +0000 (21:09 +0000)]
Disable flow-director support until it's been debugged and verified.

The flowdirector feature shares on-chip memory with other things
such as the RX buffers.  In theory it should be configured in a way
that doesn't interfere with the rest of operation.  In practice,
the RX buffer calculation didn't take the flow-director allocation
into account and there'd be overlap.  This lead to various garbage
frames being received containing what looks like internal NIC state.

What _I_ saw was traffic ending up in the wrong RX queues.
If I was doing a UDP traffic test with only one NIC ring receiving
traffic, everything is fine.  If I fired up a second UDP stream
which came in on another ring, there'd be a few percent of traffic
from both rings ending up in the wrong ring.  Ie, the RSS hash would
indicate it was supposed to come in ring X, but it'd come in ring Y.

However, when the allocation was fixed up, the developers at Verisign
still saw traffic stalls.

The flowdirector feature ends up fiddling with the NIC to do various
attempts at load balancing connections by populating flow table rules
based on sampled traffic.  It's likely that all of that has to be
carefully reviewed and made less "magic".

So for now the flow director feature is disabled (which fixes both
what I was seeing and what they were seeing) until it's all much
more debugged and verified.

Tested:

* (me) 82599EB 2x10G NIC, RSS UDP testing.
* (verisign) not sure on the NIC (but likely 82599), 100k-200k/sec TCP
  transaction tests.

Submitted by: Marc De La Gueronniere <mdelagueronniere@verisign.com>
MFC after: 1 week
Sponsored by: Verisign, Inc.

9 years agoThe error bits are not valid with EOP=0; so intermediary fragments should
adrian [Mon, 15 Sep 2014 20:54:12 +0000 (20:54 +0000)]
The error bits are not valid with EOP=0; so intermediary fragments should
not be discarded.

Submitted by: Marc De La Gueronniere <mdelagueronniere@verisign.com>
MFC after: 1 week
Sponsored by: Verisign, Inc.

9 years agoFix a double-free of mbufs in rx_ixgbe_discard().
adrian [Mon, 15 Sep 2014 20:50:26 +0000 (20:50 +0000)]
Fix a double-free of mbufs in rx_ixgbe_discard().

fmp->buf at the free point is already part of the chain being freed,
so double-freeing is counter-productive.

Submitted by: Marc De La Gueronniere <mdelagueronniere@verisign.com>
MFC after: 1 week
Sponsored by: Verisign, Inc.

9 years agoSet DROP_EN on each RX queue if transmit flow-control is disabled.
adrian [Mon, 15 Sep 2014 19:53:49 +0000 (19:53 +0000)]
Set DROP_EN on each RX queue if transmit flow-control is disabled.

This allows the NIC to drop frames on the receive queue and not
cause the MAC to block on receiving to _any_ queue.

Tested:

igb0@pci0:5:0:0:        class=0x020000 card=0x152115d9 chip=0x15218086 rev=0x01 hdr=0x00
    vendor     = 'Intel Corporation'
    device     = 'I350 Gigabit Network Connection'
    class      = network
    subclass   = ethernet

Discussed with: Eric Joyner <eric.joyner@intel.com>

MFC after: 1 week
Sponsored by: Norse Corp, Inc.

9 years agoAdd quirks to disable READ CAPACITY (16) for PNY USB 3.0 Flash Drives.
mav [Mon, 15 Sep 2014 19:48:27 +0000 (19:48 +0000)]
Add quirks to disable READ CAPACITY (16) for PNY USB 3.0 Flash Drives.

Submitted by: Sean Fagan <sef@ixsystems.com>
MFC after: 3 days

9 years agoChunk IDs are 8 bit entities, not 16 bit.
tuexen [Mon, 15 Sep 2014 19:38:34 +0000 (19:38 +0000)]
Chunk IDs are 8 bit entities, not 16 bit.
Thanks to Peter Kasting from Google for drawing
my attention to it.

MFC after: 3 days

9 years agoAdd stricter checking of some mmap() arguments:
jhb [Mon, 15 Sep 2014 17:20:13 +0000 (17:20 +0000)]
Add stricter checking of some mmap() arguments:
- Fail with EINVAL if an invalid protection mask is passed to mmap().
- Fail with EINVAL if an unknown flag is passed to mmap().
- Fail with EINVAL if both MAP_PRIVATE and MAP_SHARED are passed to mmap().
- Require one of either MAP_PRIVATE or MAP_SHARED for non-anonymous
  mappings.

Reviewed by: alc, kib
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D698

9 years agoThree improvements to vnode_pager_generic_getpages():
alc [Mon, 15 Sep 2014 17:14:09 +0000 (17:14 +0000)]
Three improvements to vnode_pager_generic_getpages():

Eliminate an exclusive object lock acquisition and release on the expected
execution path.

Do page zeroing before the object lock is acquired rather than during the
time that the object lock is held.

Use vm_pager_free_nonreq() to eliminate duplicated code.

Reviewed by: kib
MFC after: 6 weeks
Sponsored by: EMC / Isilon Storage Division

9 years agoRemove redundant declaration. vnode.h should be included before vnode_pager.h.
glebius [Mon, 15 Sep 2014 15:49:29 +0000 (15:49 +0000)]
Remove redundant declaration. vnode.h should be included before vnode_pager.h.

9 years agoThe private peripheral interrupts start at offset 16, not 0. Also, use
ian [Mon, 15 Sep 2014 15:36:00 +0000 (15:36 +0000)]
The private peripheral interrupts start at offset 16, not 0.  Also, use
names rather than inline mystery constants for these offsets.

Pointed out by: andrew

9 years agoxen: don't set suspend/resume methods for the PIRQ PIC
royger [Mon, 15 Sep 2014 15:15:52 +0000 (15:15 +0000)]
xen: don't set suspend/resume methods for the PIRQ PIC

The suspend/resume of event channels is already handled by the xen_intr_pic.
If those methods are set on the PIRQ PIC they are just called twice, which
breaks proper resume. This fix restores migration of FreeBSD guests to a
working state.

Sponsored by: Citrix Systems R&D

9 years agoUse generic SYSCTL_* macro instead of deprecated SYSCTL_VNET_*.
hrs [Mon, 15 Sep 2014 14:43:58 +0000 (14:43 +0000)]
Use generic SYSCTL_* macro instead of deprecated SYSCTL_VNET_*.

Suggested by: glebius

9 years agoUpgrade to OpenPAM Ourouparia.
des [Mon, 15 Sep 2014 13:40:09 +0000 (13:40 +0000)]
Upgrade to OpenPAM Ourouparia.

9 years agoProvide the unique implementation for the VOP_GETPAGES() method used
kib [Mon, 15 Sep 2014 12:28:29 +0000 (12:28 +0000)]
Provide the unique implementation for the VOP_GETPAGES() method used
by ffs and ext2fs.  Remove duplicated call to vm_page_zero_invalid(),
done by VOP and by vm_pager_getpages().  Use vm_pager_free_nonreq().

Reviewed by: alc (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 6 weeks (after r271596)

9 years agor271256 fixed one segfault condition but introduced another due to the
des [Mon, 15 Sep 2014 11:32:08 +0000 (11:32 +0000)]
r271256 fixed one segfault condition but introduced another due to the
wrong operator being used in the tty check.

Reported by: avg@
MFH: 3 days

9 years agoAdd comments describing r271604 change.
mav [Mon, 15 Sep 2014 11:17:36 +0000 (11:17 +0000)]
Add comments describing r271604 change.

MFC after: 3 days

9 years agoAdd a brief description of CTLFLAG_VNET flag. Since the VIMAGE is
glebius [Mon, 15 Sep 2014 11:14:23 +0000 (11:14 +0000)]
Add a brief description of CTLFLAG_VNET flag. Since the VIMAGE is
absolutely not documented, I see no reason in long descriptions
here.

9 years agoRemove incorrect props
des [Mon, 15 Sep 2014 09:45:10 +0000 (09:45 +0000)]
Remove incorrect props

9 years agoVendor import of OpenPAM Ourouparia.
des [Mon, 15 Sep 2014 09:40:30 +0000 (09:40 +0000)]
Vendor import of OpenPAM Ourouparia.

9 years agoMake net.inet.ip.sourceroute, net.inet.ip.accept_sourceroute, and
hrs [Mon, 15 Sep 2014 07:20:40 +0000 (07:20 +0000)]
Make net.inet.ip.sourceroute, net.inet.ip.accept_sourceroute, and
net.inet.ip.process_options vnet-aware.  Revert changes in r271545.

Suggested by: bz

9 years agoadd gptzfsboot.8, zfsboot.8 and zfsloader.8 manual pages
avg [Mon, 15 Sep 2014 06:21:28 +0000 (06:21 +0000)]
add gptzfsboot.8, zfsboot.8 and zfsloader.8 manual pages

Many thanks to Warren Block for his reviews, corrections and additions.

Reviewed by: Warren Block <wblock@FreeBSD.org>
MFC after: 1 week

9 years agoUse gic_decode_fdt() rather than a local routine to parse fdt interrupt
ian [Sun, 14 Sep 2014 23:48:18 +0000 (23:48 +0000)]
Use gic_decode_fdt() rather than a local routine to parse fdt interrupt
properties.  Move fdt_pic_table and fdt_fixup_table into imx6_machdep.c,
which means imx6 doesn't need imx_common.c anymore.

9 years agoAlways report that we support REPORT TARGET PORT GROUPS command.
mav [Sun, 14 Sep 2014 23:39:13 +0000 (23:39 +0000)]
Always report that we support REPORT TARGET PORT GROUPS command.

Without clustering support we any way have only one group of permanently
active ports, but that gives us one more supported VMWare feature. ;)

Solaris' Comstar also reports it even when only one port is present.

9 years agoAdd couple memory barries to serialize tdq_cpu_idle and tdq_load accesses.
mav [Sun, 14 Sep 2014 22:13:19 +0000 (22:13 +0000)]
Add couple memory barries to serialize tdq_cpu_idle and tdq_load accesses.

This change fixes transient performance drops in some of my benchmarks,
vanishing as soon as I am trying to collect any stats from the scheduler.
It looks like reordered access to those variables sometimes caused loss of
IPI_PREEMPT, that delayed thread execution until some later interrupt.

MFC after: 3 days

9 years agoAdd DEF_CLONE_CMD_ARG2
bryanv [Sun, 14 Sep 2014 22:10:35 +0000 (22:10 +0000)]
Add DEF_CLONE_CMD_ARG2

This will be used in the forthcoming vxlan import.

Reviewed by: gnn
Phabric: https://reviews.freebsd.org/D382

9 years agoFix PowerPC backtraces. Since kernel and user have completely separate address
jhibbits [Sun, 14 Sep 2014 22:03:41 +0000 (22:03 +0000)]
Fix PowerPC backtraces.  Since kernel and user have completely separate address
spaces, rather than a split address, we actually can't check for being within
the kernel's address range.  Instead, do what other backtraces do, and use
trapexit()/asttrapexit() as the stack sentinel.

MFC after: 3 weeks

9 years agoAdd a common routine for parsing FDT data describing an ARM GIC interrupt.
ian [Sun, 14 Sep 2014 21:21:03 +0000 (21:21 +0000)]
Add a common routine for parsing FDT data describing an ARM GIC interrupt.

In the fdt data we've written for ourselves, the interrupt properties
for GIC interrupts have just been a bare interrupt number.  In standard
data that conforms to the published bindings, GIC interrupt properties
contain 3-tuples that describe the interrupt as shared vs private, the
interrupt number within the shared/private address space, and configuration
info such as level vs edge triggered.

The new gic_decode_fdt() function parses both types of data, based on the
#interrupt-cells property.  Previously, each platform implemented a decode
routine and put a pointer to it into fdt_pic_table.  Now they can just
list this function in their table instead if they use arm/gic.c.

9 years agoPull in r217410 from upstream llvm trunk (by Bob Wilson):
dim [Sun, 14 Sep 2014 18:50:38 +0000 (18:50 +0000)]
Pull in r217410 from upstream llvm trunk (by Bob Wilson):

  Set trunc store action to Expand for all X86 targets.

  When compiling without SSE2, isTruncStoreLegal(F64, F32) would return
  Legal, whereas with SSE2 it would return Expand. And since the Target
  doesn't seem to actually handle a truncstore for double -> float, it
  would just output a store of a full double in the space for a float
  hence overwriting other bits on the stack.

  Patch by Luqman Aden!

This should fix clang -O0 on i386 assigning garbage to floats, in
certain scenarios.

PR: 187437
Submitted by: cebd@gmail.com
Obtained from: http://llvm.org/viewvc/llvm-project?rev=217410&view=rev
MFC after: 3 days

9 years agoAvoid an exclusive acquisition of the object lock on the expected execution
alc [Sun, 14 Sep 2014 18:07:55 +0000 (18:07 +0000)]
Avoid an exclusive acquisition of the object lock on the expected execution
path through the NFS clients' getpages functions.

Introduce vm_pager_free_nonreq().  This function can be used to eliminate
code that is duplicated in many getpages functions.  Also, in contrast to
the code that currently appears in those getpages functions,
vm_pager_free_nonreq() avoids acquiring an exclusive object lock in one
case.

Reviewed by: kib
MFC after: 6 weeks
Sponsored by: EMC / Isilon Storage Division

9 years agoAdd compat strings for all the flavors of GIC this driver should support.
ian [Sun, 14 Sep 2014 17:47:04 +0000 (17:47 +0000)]
Add compat strings for all the flavors of GIC this driver should support.
Also allow the driver to attach to ofwbus as well as simplebus, some FDT
data puts the root interrupt controller on the root bus.

9 years agoFix an undefined variable that was accidentally not causing an error.
ian [Sun, 14 Sep 2014 17:36:57 +0000 (17:36 +0000)]
Fix an undefined variable that was accidentally not causing an error.

The code had references to both intr_offset and intr_parent variable names
as referring to the parent interrupt node.  The intr_parent variable
wasn't actually defined anywhere, but the only references to it were as
an argument to a macro that didn't use that argument in expansion, so
the undefined variable accidentally didn't cause an error.

The intr_parent name makes more sense in context, so change all occurrances
of intr_offset to intr_parent.

9 years agosh: Remove arbitrary length limit on << EOF markers.
jilles [Sun, 14 Sep 2014 16:46:30 +0000 (16:46 +0000)]
sh: Remove arbitrary length limit on << EOF markers.

This also simplifies the code.

9 years agosh: Make checkend() a real function instead of an emulated nested function.
jilles [Sun, 14 Sep 2014 16:27:49 +0000 (16:27 +0000)]
sh: Make checkend() a real function instead of an emulated nested function.

No functional change is intended, but the generated code is slightly
different.

9 years agoAdd a comment giving an overview of the driver. Remove leftover debugging.
ian [Sun, 14 Sep 2014 16:12:43 +0000 (16:12 +0000)]
Add a comment giving an overview of the driver.  Remove leftover debugging.

9 years agosh: Add some const keywords.
jilles [Sun, 14 Sep 2014 15:59:15 +0000 (15:59 +0000)]
sh: Add some const keywords.

9 years agoAdded missing ZFS sysctls
smh [Sun, 14 Sep 2014 12:23:00 +0000 (12:23 +0000)]
Added missing ZFS sysctls
* vfs.zfs.vdev.async_write_active_min_dirty_percent
* vfs.zfs.vdev.async_write_active_max_dirty_percent

Added validation of min / max for ZFS sysctl
* vfs.zfs.dirty_data_max_percent

MFC after: 3 days

9 years agoUpdate CAM CCB accounting for the new status quo.
mav [Sun, 14 Sep 2014 11:59:49 +0000 (11:59 +0000)]
Update CAM CCB accounting for the new status quo.

devq_openings counter lost its meaning after allocation queues has gone.
held counter is still meaningful, but problematic to update due to separate
locking of CCB allocation and queuing.

To fix that replace devq_openings counter with allocated counter.  held is
now calculated on request as difference between number of allocated, queued
and active CCBs.

MFC after: 1 month

9 years agoFix mis-spelling of bits and types names in the vnode_pager_putpages().
kib [Sun, 14 Sep 2014 10:27:36 +0000 (10:27 +0000)]
Fix mis-spelling of bits and types names in the vnode_pager_putpages().
The changes should not modify the generated code.

The pager->pgo_putpages() method takes int flags as its fourth
argument, while vnode_pager_putpages() used boolean_t (which is
typedef'ed to int).  The flags are from VM_PAGER_* namespace, while
vnode_pager_putpages() passed TRUE and OBJPC_SYNC to VOP_PUTPAGES(),
which both are numerically equal to VM_PAGER_PUT_SYNC.

Noted and reviewed by: alc (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 1 week

9 years agoUse non-recursive algorithm for traversing nvlists. This also removes
pjd [Sun, 14 Sep 2014 09:30:09 +0000 (09:30 +0000)]
Use non-recursive algorithm for traversing nvlists. This also removes
the limit on number of nested nvlists.

Submitted by: Mariusz Zaborski

9 years agoRemove the limit on descriptors that can be send in one nvlist.
pjd [Sun, 14 Sep 2014 09:27:12 +0000 (09:27 +0000)]
Remove the limit on descriptors that can be send in one nvlist.

Submitted by: Mariusz Zaborski