When bus_alloc_resource_any() fails, dc_detach() is called and it
attempts to deallocate busdma tags and resources that haven't been
allocated yet, causing a panic every time a dc interface fails to
attach. Fix by checking that we really have something to dealloc
before calling bus_dma*() functions.
Explicitly switch to the new TSS by updating the current CPU's TSS selector
and reloading it in i386_extend_pcb() rather than trying to force a context
switch to reload the TSS via the TDF_NEEDRESCHED flag. Optimizations to
avoid calling cpu_switch() when the new thread was identical to the old
thread defeated the attempt to force a TSS reload. Explicitly loading the
new TSS is what we really want to do anyway.
PR: i386/84842
Reported by: Alexander Best arundel at h3c dot de
MFC after: 1 week
Reviewed by: bde (mostly)
Cosmetic fixes to prev. commit.
Change first MAXPATHLEN to more standard PATH_MAX
Change second MAXPATHLEN to 1024 (it is temp buffer not related)
Change comment to reflect that.
Add "-q" argument to sysctl(8), which suppresses a limited set of warnings/
errors generated. In particular, it suppresses "unknown oid" when
attempting to get or set a sysctl not present in the kernel.
Improve the MP safeness associated with the creation of symbolic
links and the execution of ELF binaries. Two problems were found:
1) The link path wasn't tagged as being MP safe and thus was not properly
protected.
2) The ELF interpreter vnode wasnt being locked in namei(9) and thus was
insufficiently protected.
This commit makes the following changes:
-Sets the MPSAFE flag in NDINIT for symbolic link paths
-Sets the MPSAFE flag in NDINIT and introduce a vfslocked variable which
will be used to instruct VFS_UNLOCK_GIANT to unlock Giant if it has been
picked up.
-Drop in an assertion into vfs_lookup which ensures that if the MPSAFE
flag is NOT set, that we have picked up giant. If not panic (if WITNESS
compiled into the kernel). This should help us find conditions where vnode
operations are in-sufficiently protected.
Portsnap uses host(1) to search for mirrors, but this is only available
when the base system is not compiled with NO_BIND set. Before we start
searching for mirrors, make sure that host(1) can be found, and if it
doesn't exist then fallback to the A record instead of the SRV records.
Fix system shutdown timeout handling by again supporting longer running
shutdown procedures (which have a duration of more than 120 seconds).
We have two user-space affecting shutdown timeouts: a "soft" one in
/etc/rc.shutdown and a "hard" one in init(8). The first one can be
configured via /etc/rc.conf variable "rcshutdown_timeout" and defaults
to 30 seconds. The second one was originally (in 1998) intended to be
configured via sysctl(8) variable "kern.shutdown_timeout" and defaults
to 120 seconds.
Unfortunately, the "kern.shutdown_timeout" was declared "unused" in 1999
(as it obviously is actually not used within the kernel itself) and
hence was intentionally but misleadingly removed in revision 1.107 from
init_main.c. Kernel sysctl(8) variables are certainly a wrong way to
control user-space processes in general, but in this particular case the
sysctl(8) variable should have remained as it supports init(8), which
isn't passed command line flags (which in turn could have been set via
/etc/rc.conf), etc.
As there is already a similar "kern.init_path" sysctl(8) variable which
directly affects init(8), resurrect the init(8) shutdown timeout under
sysctl(8) variable "kern.init_shutdown_timeout". But this time document
it as being intentionally unused within the kernel and used by init(8).
Also document it in the manpages init(8) and rc.conf(5).
Protect the devfs rule internal global lists with a sx lock, the per
mount locks are not enough. Finer granularity (x)locking could be
implemented, but I prefer to keep it simple for now.
Close a race which could result in unwarranted "ruleset %d already
running" panics.
Previously, recursion through the "include" feature was prevented by
marking each ruleset as "running" when applied. This doesn't work for
the case where two DEVFS instances try to apply the same ruleset at
the same time.
Instead introduce the sysctl vfs.devfs.rule_depth (default == 1) which
limits how many levels of "include" we will traverse.
Be aware that traversal of "include" is recursive and kernel stack
size is limited.
When stopping the card, and returning to page 0, it is best if you do
that with the NIC set of registers rather than the ASIC registers. I
believe this was a harmless oversight, since we set ED_P0_CR to the
same value 5ms later, but just to be safe...
Fix so that when a slice or a partition is removed through g_slice_config(),
it is destroyed in GEOM, in addition to being removed from /dev.
Before this patch, if you applied a new MBR which deleted a slice,
the deleted slice would not be in /dev, but it would still appear
in kern.geom.conftxt and kern.geom.confxml, which would confused
the diskPartitionEditor in sysinstall.
Determine where the runtime loader got mapped to when a dynamically
linked process was exec()'ed, and map samples in that address range
into the appropriate buckets in the runtime loader's profile.
Improve a few comments.
Remove any hardcoded assumptions about malloc's way of allocating,
just use MAXPATHLEN. It prevents potential buffer overflow with other
malloc implementations.
(this change based on submitted patch)
Use time_uptime for timing commands instead of time_second. This is a bit
risky because the "current time" is supposed to be fed to the card during
initialization, and the current time is supposed to be put into each command
that is sent to the card. Hopefully either the card doesn't actually care
about the timestamps, or it doesn't care about the absolute values so long
and the relative values are consistent. Not an MFC candidate until more
thorough testing can be done.
o Attach AX88x90's MII bus to system, and require its presence.
o Reorg the mii code a little, and move more of it into pccard attachment.
o Eliminate ed_pccard_{read,write}_attrmem in favor of a more appropriate
function in the pccard layer.
o Update comments to reflect knowledge gained.
o Update how re recognize a NE-2000 ROM. I found a couple of different
datasheets that define the structure of the PROM data, so the code's
old heuristics have been removed, and comments updated to reflect the
structure.
o Eliminate work around for EC2T. It is no longer needed, and was wrong
headed since the EC2T has a Winbound 82C926C in it, not a AX88x90.
o Add copyright to if_ed_pccard.c, since I believe I've re-written more than
3/4 of it.
# With these changes, all of my 20-odd ed based cards work, except for the
# NetGear FA-410, and I'm pretty sure that's a MII/PHY problem.
Don't consider being unable to open the bounds file worthy of printing
at LOG_WARNING by default; instead, consider it something to be printed
to the tty when 'verbose' mode is set. This avoids printing out extra
lines at every boot on a system with crash dumps enabled, but that has
not yet had to generate a crashdump.
Use kenv -q to extract dumpdev rather than kenv, in order to avoid
spamming the console in the event that a loader tunable 'dumpdev'
isn't defined, which is not a relevant failure to report.
Use smaller names. The "NE2000 PCI Ethernet" is largerly redundant
with edX at pciY. This removes 8*22 bytes from the strings. Not a
huge win, but a cheap one.
Add the olicom OC2231 and OC2232 to the list of cards that we should
attach to. These cards are combo cards (in that they have a modem
inside of them), but not true MFC cards. Full support of these cards
will have to wait until we can pick the config to use and for the PFC
support that I have brewing.
Trim down now (believed to be) unused fifo_ioctl() and
fifo_kqfilter() VOP implementations, since they in theory are used
only on open file descriptors, in which case the ioctls are via
fifo_ioctl_f() and kqueue requests are via fifo_kqfilter_f().
Generate warnings if they are entered for now. These printf()
calls should become panic() calls.
Annotate and re-implement fifo_ioctl_f(): don't arbitrarily
forward ioctls to the socket layer, only forward the ones we
explicitly support for fifos. In the case of FIONREAD, don't
forward the request to the write socket on a read-write fifo, or
the read result is overwritten. Annotate a nasty case for the
undefined POSIX O_RDWR on fifos, in which failure of the second
ioctl will result in the socket pair being in an inconsistent
state.
Assert copyright as I find myself rewriting non-trivial parts of
fifofs.
Restructure fifo_io event-related tests in order to support test cases
relating to O_RDWR file descriptors, which while not defined in POSIX,
are in fact used:
(1) Revise assumption that we have two file descriptors when testing I/O
operations on a fifo. Provide cleanup routines that handle either
two or three file descriptors (including a kqueue descriptor).
(2) Add an openfifo_rw() routine to supplement openfifo().
(3) kqueue_setup() now configures an existing kqueue to monitor a new
file descriptor, rather than allocating a new kqueue to monitor two
existing file descriptors.
(4) Wrap all direct poll/select/kqueue/FIONREAD interactions in a single
function, assert_status(), which takes a file descriptor, kqueue
descriptor, assertion of read/writable/exception states, and
test description, then tests the assertion. This reduces the bulk
of calls in many of the tests, making them shorter, more readable,
and easier to determine correct.
(5) Add a new test_events_rdwr() function, which performs a basic create/
write/read event test on a O_RDWR fifo file descriptor. This is
currently failed by our fifo code in HEAD due to a bug in FIONREAD
handling. Fix to be merged shortly.
Teach fifo_io how to use ioctl(FIONREAD) to determine if a fifo is
readable, and use that in addition to poll+select+kqueue to verify
appropriate readability.
Various and sundry improvements:
- Replace 'process' with 'thread' everywhere.
- Update several places to note that that the fact that default mutexes
may adaptively spin isn't necessarily MD, but is just part of the
implementation as a whole.
- Clarify the text about MTX_SPIN mutexes only being appropriate for
INTR_FAST interrupts or other low level scheduler code to make the
jargon more FreeBSD-ish rather than BSD/OS-ish.
- Also, note that it is possible that interrupts aren't blocked but just
deferred when a spin lock is held (the whole blocked vs. deferred bit is
an MD implementation detail).
- Remove statements saying that spin locks must be released in the exact
opposite order that they were acquired. This stopped being true several
years ago when we first added critical sections that stored their state
in the current thread rather than in struct mtx.
- Note that a mutex must be initialized before it is passed to any other
mutex function, not just mtx_lock.
- Clarify that mtx_trylock() only operates on MTX_DEF mutexes.
- Simplify the text about possible preemption during a mtx_unlock().
- Use complete English sentences in place of phrases in a few places.
- Clarify that it isn't ever safe to sleep with a mutex held. The kernel
tends to panic when you do that.
As a result of kqueue locking work, socket buffer locks will always
be held when entering a kqueue filter for fifos via a socket buffer
event: as such, assert the lock unconditionally rather than acquiring
it conditionall.
1) fifo_kqfilter() is not actually ever used, it likely should be GC'd.
2) fifo_kqfilter_f() doesn't implement EVFILT_VNODE, so detecting events
on the underlying vnode for a fifo no longer works (it did in 4.x).
Likely, fifo_kqfilter_f() should forward the request to the VFS using
fp->f_vnode, which would work once fifo_kqfilter() was detached from
the vnode operation vector (removing the fifo override).
Introduce no-op nosup fifo kqueue filter and detach routine, which are
used when a read filter is requested on a write-only fifo descriptor, or
a write filter is requested on a read-only fifo descriptor. This
permits the filters to be registered, but never raises the event, which
causes kqueue behavior for fifos to more closely match similar semantics
for poll and select, which permit testing for the condition even though
the condition will never be raised, and is consistent with POSIX's notion
that a fifo has identical semantics to a one-way IPC channel created
using pipe() on most operating systems.
The fifo regression test suite can now run to completion on HEAD without
errors.
Add test_kqueue(), which registers and unregisters various kqueue filter
types on a fifo in order to make sure that EVFILT_READ, EVFILT_WRITE can
be registered, and that EVFILT_NETDEV can't be registered. For now, we
don't test that EVFILT_VNODE can be registered on fifos, as that has been
broken at some point.
- Fix the locking in dsp.c to prevent a LOR (AFAIK not on the LOR page).
- Remove an assertion in sound.c, it's not needed (and causes a panic now).
From the conversation via mail between glebius and Ariff:
---snip---
> Well, but which mutex protects now? Do we own anything else
> in pcm_chnalloc()? I see some queue(4) macros in pcm_chnalloc(),
> they should be protected, shouldn't they?
Queue insertion/removal occur during
1) driver loading (which is pretty much single thread /
sequential) or unloading (mutex protected, bail out if there is
any channel with refcount > 0 or busy).
2) vchan_create()/destroy(), (which is *sigh* quite complicated), but
somehow protected by 'master'/parent channel mutex. Other
thread cannot add/remove vchan (or even continue traversing
that queue) unless it can acquire parent channel mutex.
---snip---
Fix the locking in dsp.c to prevent a LOR (AFAIK not on the LOR page).
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested with: INVARIANTS[1] and DIAGNOSTICS[2]
Tested by: netchild [1,2], David Reid <david@jetnet.co.uk> [1]
When a request is made to register a filter on a fifo that doesn't
apply to the fifo (i.e., not EVFILT_READ or EVFILT_WRITE), reject
it as EINVAL, not by returning 1 (EPERM).
- Teach fifo_io about kqueue monitoring of fifo file descriptor status,
and add test cases for kqueue to match existing case for poll and
select. Add a new cleanup routine, cleanfifokq(), for use in tests that
use kqueues. kqueue_setup() sets up kqueue sessions, and kqueue_status()
returns file descriptor status.
- Correct a bug in select handling relating to the nfds argument, which
was incorrect so resulted in select occuring on the wrong file descriptor,
and possible false positive/negative results.
- Clarify error reporting in one byte write+read tests to distinguish
errors in the after case from the before case.
Fix the prototypes for devname() and devname_r(), the first two argument
types are supposed to be dev_t and mode_t (prefixed with __ due to
namespace reasons).