mav [Thu, 18 Dec 2014 08:32:06 +0000 (08:32 +0000)]
MFC r275458:
Do not pre-allocate UNIT ATTENTIONs storage for every possible initiator.
Abusing ability of major UAs cover minor ones we may not account UAs for
inactive ports. Allocate UAs storage for port and start accounting only
after some initiator from that port fetched its first POWER ON OCCURRED.
This reduces per-LUN CTL memory usage from >1MB to less then 100K.
mav [Thu, 18 Dec 2014 08:30:28 +0000 (08:30 +0000)]
MFC r275447:
Do not pre-allocate reservation keys memory for every possible initiator.
In configurations with many ports, like iSCSI, each LUN is typically
accessed only by limited subset of ports. Allocating that memory on
demand allows to reduce CTL memory usage from 5.3MB/LUN to 1.3MB/LUN.
mav [Thu, 18 Dec 2014 08:25:00 +0000 (08:25 +0000)]
MFC r275058: Coalesce last data move and command status for read commands.
Make CTL core and block backend set success status before initiating last
data move for read commands. Make CAM target and iSCSI frontends detect
such condition and send command status together with data. New I/O flag
allows to skip duplicate status sending on later fe_done() call.
For Fibre Channel this change saves one of three interrupts per read command,
increasing performance from 126K to 160K IOPS. For iSCSI this change saves
one of three PDUs per read command, increasing performance from 1M to 1.2M
IOPS.
mav [Thu, 18 Dec 2014 08:22:16 +0000 (08:22 +0000)]
MFC r274962: Replace home-grown CTL IO allocator with UMA.
Old allocator created significant lock congestion protecting its lists
of preallocated I/Os, while UMA provides much better SMP scalability.
The downside of UMA is lack of reliable preallocation, that could guarantee
successful allocation in non-sleepable environments. But careful code
review shown, that only CAM target frontend really has that requirement.
Fix that making that frontend preallocate and statically bind CTL I/O for
every ATIO/INOT it preallocates any way. That allows to avoid allocations
in hot I/O path. Other frontends either may sleep in allocation context
or can properly handle allocation errors.
On 40-core server with 6 ZVOL-backed LUNs and 7 iSCSI client connections
this change increases peak performance from ~700K to >1M IOPS! Yay! :)
pfg [Tue, 16 Dec 2014 21:13:55 +0000 (21:13 +0000)]
MFC r275553, r275612;
patch(1): Bring fixes from OpenBSD
Check fstat return value. Use off_t for file size and offsets.
Avoid iterating over end of string.
Introduce strtolinenum to properly check line numbers while parsing:
no signs, no spaces, just digits, 0 <= x <= LONG_MAX
Properly validate line ranges supplied in diff file to prevent overflows.
Also fixes an out of boundary memory access because the resulting values
are used as array indices.
jhb [Tue, 16 Dec 2014 20:05:10 +0000 (20:05 +0000)]
MFC 271635,271722:
- Only the manpage updates from 271635 are merged to give additional
heads up for the stricter checks in 11, but the kernel in 10 remains
permissive.
- 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.
- 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.
jhb [Tue, 16 Dec 2014 19:45:56 +0000 (19:45 +0000)]
MFC 272897:
Various fixes to stats:
- Read the counts of received, dropped, and transmitted management
packets and add sysctl nodes for them.
- Fix the total octets received/transmitted to read all 64 bits of
the counters.
- Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot.
- Remove spurious spaces.
pfg [Tue, 16 Dec 2014 18:45:31 +0000 (18:45 +0000)]
MFC r275645;
ext2fs: Fix old out-of-bounds access.
Overrunning buffer pointed to by (caddr_t)&oip->i_db[0] of 48 bytes by
passing it to a function which accesses it at byte offset 59 using
argument 60UL.
The issue was inherited from an older FFS implementation and
fixed there with by merging UFS2 in r98542. We follow the
FFS fix.
kib [Mon, 15 Dec 2014 11:00:56 +0000 (11:00 +0000)]
MFC r275619:
Check for bo_bufobj->bo_object for NULL and cache the value in local
variable to avoid NULL dereference in getnewbuf_reuse_bp(). The vnode
owning the buffer is not locked there.
Currently, there can be no more than INT_MAX positional parameters. Make
sure to treat all higher ones as unset to avoid incorrect results and
crashes.
On 64-bit systems, our atoi() takes the low 32 bits of the strtol() and
sign-extends them.
On 32-bit systems, the call to atoi() returned INT_MAX for too high values
and there is not enough address space for so many positional parameters, so
there was no issue.
andrew [Sun, 14 Dec 2014 17:38:43 +0000 (17:38 +0000)]
MFC 275415:
Allow the UAL APSR_nzcv format for the mrc and mrc2 instructions. The clang
integrated assembler only allows these forms so binutils will need to
support them.
Sponsored by: ABT Systems Ltd
MFC 275584:
Fix mrc and mrc2 with APSR_nzcv. Binutils encodes it internally as 0 where
we need it to be set to 15 for it to be equivalent to r15.
andrew [Sun, 14 Dec 2014 16:28:53 +0000 (16:28 +0000)]
Clean up our ARM assembly:
MFC 275256:
Switch to the ARM unified assembly language as the clang integrated as only
supports it. Binutils supports it when the ".syntax unified" directive
is set.
Sponsored by: ABT Systems Ltd
MFC 275264:
Update _ENTRY to use _EENTRY to reduce the common code.
MFC 275321:
Remove extra labels, ENTRY_NP already provides them.
Sponsored by: ABT Systems Ltd
MFC 275322:
Correctly a few incorrect uses of ENTRY/EENTRY and END/EEND
Sponsored by: ABT Systems Ltd
MFC 275416:
Fix the name of the coprocessor to include the "p" prefix, the clang
integrated assembler expects this.
Sponsored by: ABT Systems Ltd
MFC 275418:
Switch to unified syntax so these can be built with clang 3.5.
Sponsored by: ABT Systems Ltd
MFC 275519:
Add missing END macros to some of the xscale functions.
Sponsored by: ABT Systems Ltd
MFC 275520:
Use the unified syntax in a few more assembly files
Sponsored by: ABT Systems Ltd
MFC 275521:
Set the alignment to 4-bytes after a string as clang 3.5 can switch to
thumb mode if this is incorrect.
Sponsored by: ABT Systems Ltd
MFC 275522:
Place the literal pool after a RET otherwise clang 3.5 tries to put it too
far away from a ldr psuedo instruction. With this clang will place the
literal value here where it's close enough to be loaded.
Sponsored by: ABT Systems Ltd
MFC 275523:
Switch to an armv6k cpu, without this clang 3.5 complains "bx lr" is
unsupported as it needs a newer cpu.
Sponsored by: ABT Systems Ltd
MFC 275524:
Switch to a .cpu directive. These will work when clang 3.5 is imported
where the .arch directive is a nop.
andrew [Sun, 14 Dec 2014 15:33:45 +0000 (15:33 +0000)]
MFC 273927:
Move the definitions of the fdt functions from a uboot header to a new fdt
header. There is nothing in the fdt spec that ties it to U-Boot.
While here sort and fix the signature of fdt_setup_fdtp.
MFC 273934:
Start to allow platforms other than U-Boot to use the FDT code in loader by
moving U-Boot specific code from libfdt.a to a new libuboot_fdt.a. This
needs to be a new library for linking to work correctly.
This fixes a bug in the llvm vectorizer, which could sometimes cause
vectorized loops to perform an additional iteration, leading to possible
buffer overruns. Symptoms of this, which are usually segfaults, were
first noticed when building gcc ports, here:
dim [Tue, 9 Dec 2014 07:15:53 +0000 (07:15 +0000)]
MFC r275386:
Let GNU ld be less obscure about missing symbols and DSOs. If the BFD
object looks like a typical shared library, suggest adding '-l<foo>',
where <foo> has the 'lib' prefix and '.so<bar>' or '.a' suffix removed.
Otherwise, suggest adding '-l:<foo>', where <foo> is the full DT_SONAME.
delphij [Tue, 9 Dec 2014 00:47:02 +0000 (00:47 +0000)]
MFC r275071:
Reinstitate send() after syslogd restarts.
In r228193 the test of CONNPRIV have been moved to before the _usleep
and send in vsyslog(). When syslogd restarts, this would prevent the
message being logged after the disconnect/connect dance for
scenario #1.
mav [Mon, 8 Dec 2014 09:47:49 +0000 (09:47 +0000)]
MFC r275368:
When passing LUN IDs through treat ASCII values as fixed-length, not
interpreating NULLs as EOLs, but converting them to spaces.
SPC-4 does not tell that T10-based IDs should be NULL-terminated/padded.
And while it tells that it should include only ASCII chars (0x20-0x7F),
there are some USB sticks (SanDisk Ultra Fit), that have NULLs inside
the value. Treating NULLs as EOLs there made those LUN IDs non-unique.
mav [Sat, 6 Dec 2014 07:30:08 +0000 (07:30 +0000)]
MFC r274843, r274845:
Move icl_pdu_get_data() and xpt_done() out of initiator's session lock.
During heavy reads data copying in icl_pdu_get_data() may consume large
percent of CPU time. Moving it out of the lock significantly reduces
lock hold time and respectively lock congestion on read operations.
ngie [Fri, 5 Dec 2014 12:51:27 +0000 (12:51 +0000)]
MFC r275170:
Convert sbin/mdconfig/tests from prove format tests to ATF format tests
As a side effect...
1. The tests now checks for the root user before continuing with kyua, which is
more visible than the test being skipped with the TAP protocol
2. The tests work with devices that aren't /dev/md0 by caching the device
attached during the test to a file, and later use the cached information to
detach the device in the cleanup routine
3. The tests no longer require perl to run
- Remove tools/regression/pjdfstest
- Add upgrade directions for contrib/pjdfstest
- Add a note to UPDATING for the move (the reachover Makefiles are coming
soon)
Functional differences:
- ftruncate testcases are added from upstream (github)
Non-functional differences:
- The copyright for the project has been updated to 2012
- pjd's contact information has been updated
Expect ELOOP on Darwin/Linux with "O_NOFOLLOW was specified and the target is a
symbolic link" case. Assume EMLINK on the rest of the OSes (FreeBSD, Solaris,
etc)
Sponsored by: EMC / Isilon Storage Division
r272084:
Fix the executed testplan count
Sponsored by: EMC / Isilon Storage Division
r272087:
Increase the memory disk size in the following testcases to avoid mount
failures, which would cause cascade failures in the rest of the test
run:
Integrate pjdfstest test suite execution into kyua
pjdfstest execution is opt-in and must be done as root due to some of the
assumptions made by the test suite and lack of error checking in the non-root
case
A description of how to execute pjdfstest with kyua is provided in
share/pjdfstest/README
mav [Fri, 5 Dec 2014 07:25:02 +0000 (07:25 +0000)]
MFC r274805:
Make cfiscsi_offline() synchronous, waiting for connections termination
before return. This should make ctld restart more clean and predictable.
In this mode one head is in Active state, supporting all commands, while
another is in Standby state, supporting only minimal LUN discovery subset.
It is still incomplete since Standby state requires reservation support,
which is impossible to do right without having interlink between heads.
But it allows to run some basic experiments.
delphij [Fri, 5 Dec 2014 00:32:33 +0000 (00:32 +0000)]
MFC r274172 (avg)
fix l2arc compression buffers leak
We have observed that arc_release() can be called concurrently with a
l2arc in-flight write.
Also, we have observed that arc_hdr_destroy() can be called from
arc_write_done() for a zio with ZIO_FLAG_IO_REWRITE flag in similar
circumstances.
Previously the l2arc headers would be freed while leaking their
associated compression buffers. Now the buffers are placed on
l2arc_free_on_write list for delayed freeing. This is similar to what
was already done to arc buffers that were supposed to be freed
concurrently with in-flight writes of those buffers.
In addition to fixing the discovered leaks this change also adds some
protective code to assert that a compression buffer associated with a
l2arc header is never leaked.
A new kstat l2_cdata_free_on_write is added. It keeps a count of
delayed compression buffer frees which previously would have been leaks.
Tested by: Vitalij Satanivskij <satan@ukr.net> et al
Requested by: many
Sponsored by: HybridCluster / ClusterHQ
delphij [Thu, 4 Dec 2014 23:20:44 +0000 (23:20 +0000)]
MFC r274276: MFV r274271:
Improve zdb -b performance:
- Reduce gethrtime() call to 1/100th of blkptr's;
- Skip manipulating the size-ordered tree;
- Issue more (10, previously 3) async reads;
- Use lighter weight testing in traverse_visitbp();
Initialize tqent_flags in the userland taskq implementation. Without
this the assertion of tq->tq_freelist != NULL may fail in taskq_destroy.
The problem is that tqent_flags is never initialized in the userland
implementation while the kernel one does initialize it. Without proper
initialization, the flag may have its lowest bit set, making it treated
as TQENT_FLAG_PREALLOC and never removing taskq_ent_t from tq_freelist.
dim [Thu, 4 Dec 2014 01:21:56 +0000 (01:21 +0000)]
MFC r275366:
Pull in r209785 from upstream libc++ trunk (by Marshall Clow):
Fix a problem exposed by r208825, which caused bind (and other bits of
libc++) to stop working. And tests
This fix is needed to support clang 3.5.0 and higher, which are more
strict about forming pointer-to-function types with cv-qualifiers or
ref-qualifiers. See also the upstream PR <http://llvm.org/PR19742> and
<http://llvm.org/viewvc/llvm-project?rev=208825&view=rev>
dim [Thu, 4 Dec 2014 01:10:50 +0000 (01:10 +0000)]
MFC r274922:
Fix the following -Werror warning from clang 3.5.0, while building the
ath kernel module:
sys/dev/ath/ath_hal/ar5212/ar5212_reset.c:2642:7: error: taking the absolute value of unsigned type 'unsigned int' has no effect [-Werror,-Wabsolute-value]
if (abs(lp[0] * EEP_SCALE - target) < EEP_DELTA) {
^
sys/dev/ath/ah_osdep.h:74:18: note: expanded from macro 'abs'
#define abs(_a) __builtin_abs(_a)
^
sys/dev/ath/ath_hal/ar5212/ar5212_reset.c:2642:7: note: remove the call to '__builtin_abs' since unsigned values cannot be negative
sys/dev/ath/ah_osdep.h:74:18: note: expanded from macro 'abs'
#define abs(_a) __builtin_abs(_a)
^
1 error generated.
This warning occurs because both lp[0] and target are unsigned, so the
subtraction expression is also unsigned, and calling abs() is a no-op.
However, the intention was to look at the absolute difference between
the two unsigned quantities. Introduce a small static function to
clarify what we're doing, and call that instead.
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D1212
hselasky [Wed, 3 Dec 2014 09:08:04 +0000 (09:08 +0000)]
MFC r275328 and r275330:
Add missing libraries when linking and fix building of some infiniband
utilities by updating some header file locations and compiler include
directives.