reliability when tracing fast-moving processes or writing traces to
slow file systems by avoiding unbounded queueuing and dropped records.
Record loss was previously possible when the global pool of records
become depleted as a result of record generation outstripping record
commit, which occurred quickly in many common situations.
These changes partially restore the 4.x model of committing ktrace
records at the point of trace generation (synchronous), but maintain
the 5.x deferred record commit behavior (asynchronous) for situations
where entering VFS and sleeping is not possible (i.e., in the
scheduler). Records are now queued per-process as opposed to
globally, with processes responsible for committing records from their
own context as required.
- Eliminate the ktrace worker thread and global record queue, as they
are no longer used. Keep the global free record list, as records
are still used.
- Add a per-process record queue, which will hold any asynchronously
generated records, such as from context switches. This replaces the
global queue as the place to submit asynchronous records to.
- When a record is committed asynchronously, simply queue it to the
process.
- When a record is committed synchronously, first drain any pending
per-process records in order to maintain ordering as best we can.
Currently ordering between competing threads is provided via a global
ktrace_sx, but a per-process flag or lock may be desirable in the
future.
- When a process returns to user space following a system call, trap,
signal delivery, etc, flush any pending records.
- When a process exits, flush any pending records.
- Assert on process tear-down that there are no pending records.
- Slightly abstract the notion of being "in ktrace", which is used to
prevent the recursive generation of records, as well as generating
traces for ktrace events.
Future work here might look at changing the set of events marked for
synchronous and asynchronous record generation, re-balancing queue
depth, timeliness of commit to disk, and so on. I.e., performing a
drain every (n) records.
MFC after: 1 month
Discussed with: jhb
Requested by: Marc Olzheim <marcolz at stack dot nl>
- Introducing the possibility of using locks different than mutexes
for the knlist locking. In order to do this, we add three arguments to
knlist_init() to specify the functions to use to lock, unlock and
check if the lock is owned. If these arguments are NULL, we assume
mtx_lock, mtx_unlock and mtx_owned, respectively.
- Using the vnode lock for the knlist locking, when doing kqueue operations
on a vnode. This way, we don't have to lock the vnode while holding a
mutex, in filt_vfsread.
Reviewed by: jmg
Approved by: re (scottl), scottl (mentor override)
Pointyhat to: ssouhlal
Will be happy: everyone
critical_enter() and critical_exit() are now solely a mechanism for
deferring kernel preemptions. They no longer have any affect on
interrupts. This means that standalone critical sections are now very
cheap as they are simply unlocked integer increments and decrements for the
common case.
Spin mutexes now use a separate KPI implemented in MD code: spinlock_enter()
and spinlock_exit(). This KPI is responsible for providing whatever MD
guarantees are needed to ensure that a thread holding a spin lock won't
be preempted by any other code that will try to lock the same lock. For
now all archs continue to block interrupts in a "spinlock section" as they
did formerly in all critical sections. Note that I've also taken this
opportunity to push a few things into MD code rather than MI. For example,
critical_fork_exit() no longer exists. Instead, MD code ensures that new
threads have the correct state when they are created. Also, we no longer
try to fixup the idlethreads for APs in MI code. Instead, each arch sets
the initial curthread and adjusts the state of the idle thread it borrows
in order to perform the initial context switch.
This change is largely a big NOP, but the cleaner separation it provides
will allow for more efficient alternative locking schemes in other parts
of the kernel (bare critical sections rather than per-CPU spin mutexes
for per-CPU data for example).
Reviewed by: grehan, cognet, arch@, others
Tested on: i386, alpha, sparc64, powerpc, arm, possibly more
for ensuring that a process' filedesc is not shared with anybody.
Use it in the two places which previously had private implmentations.
This collects all fd_refcnt handling in kern_descrip.c
Use this in all the places where sleeping with the lock held is not
an issue.
The distinction will become significant once we finalize the exact
lock-type to use for this kind of case.
but with slightly cleaned up interfaces.
The KSE structure has become the same as the "per thread scheduler
private data" structure. In order to not make the diffs too great
one is #defined as the other at this time.
The KSE (or td_sched) structure is now allocated per thread and has no
allocation code of its own.
Concurrency for a KSEGRP is now kept track of via a simple pair of counters
rather than using KSE structures as tokens.
Since the KSE structure is different in each scheduler, kern_switch.c
is now included at the end of each scheduler. Nothing outside the
scheduler knows the contents of the KSE (aka td_sched) structure.
The fields in the ksegrp structure that are to do with the scheduler's
queueing mechanisms are now moved to the kg_sched structure.
(per ksegrp scheduler private data structure). In other words how the
scheduler queues and keeps track of threads is no-one's business except
the scheduler's. This should allow people to write experimental
schedulers with completely different internal structuring.
A scheduler call sched_set_concurrency(kg, N) has been added that
notifies teh scheduler that no more than N threads from that ksegrp
should be allowed to be on concurrently scheduled. This is also
used to enforce 'fainess' at this time so that a ksegrp with
10000 threads can not swamp a the run queue and force out a process
with 1 thread, since the current code will not set the concurrency above
NCPU, and both schedulers will not allow more than that many
onto the system run queue at a time. Each scheduler should eventualy develop
their own methods to do this now that they are effectively separated.
Rejig libthr's kernel interface to follow the same code paths as
linkse for scope system threads. This has slightly hurt libthr's performance
but I will work to recover as much of it as I can.
Thread exit code has been cleaned up greatly.
exit and exec code now transitions a process back to
'standard non-threaded mode' before taking the next step.
Reviewed by: scottl, peter
MFC after: 1 week
in diagnostics. It has outlived its usefulness and has started
causing panics for people who turn on DIAGNOSTIC, in what is otherwise
good code.
MFC after: 2 days
a more complete subsystem, and removes the knowlege of how things are
implemented from the drivers. Include locking around filter ops, so a
module like aio will know when not to be unloaded if there are outstanding
knotes using it's filter ops.
Currently, it uses the MTX_DUPOK even though it is not always safe to
aquire duplicate locks. Witness currently doesn't support the ability
to discover if a dup lock is ok (in some cases).
Reviewed by: green, rwatson (both earlier versions)
switch in fork_exit() to before anything else is done (but keep
schedlock for the deadthread check). This means one less
nasty bug if ever in the future whatever might have been called
before the update played with schedlock or critical sections.
Discussed with: tjr
the system" resource limit code: When checking if the caller has superuser
privileges, we should be checking the *real* user, not the *effective*
user. (In general, resource limiting is done based on the real user, in
order to avoid resource-exhaustion-by-setuid-program attacks.)
Now that a SUSER_RUID flag to suser_cred exists, use it here to return
this code to its correct behaviour.
Pointed out by: rwatson
specify "us" as the thread not the process/ksegrp/kse.
You can always find the others from the thread but the converse is not true.
Theorotically this would lead to runtime being allocated to the wrong
entity in some cases though it is not clear how often this actually happenned.
(would only affect threaded processes and would probably be pretty benign,
but it WAS a bug..)
Reviewed by: peter
The overhead of unconditionally allocating TIDs (and likewise,
unconditionally deallocating them), is amortized across multiple
thread creations by the way UMA makes it possible to have type-stable
storage.
Previously the cost was kept down by having threads created as part
of a fork operation use the process' PID as the TID. While this had
some nice properties, it also introduced complexity in the way TIDs
were allocated. Most importantly, by using the type-stable storage
that UMA gives us this was also unnecessary.
This change affects how core dumps are created and in particular how
the PRSTATUS notes are dumped. Since we don't have a thread with a
TID equalling the PID, we now need a different way to preserve the
old and previous behavior. We do this by having the given thread (i.e.
the thread passed to the core dump code in td) dump it's state first
and fill in pr_pid with the actual PID. All other threads will have
pr_pid contain their TIDs. The upshot of all this is that the debugger
will now likely select the right LWP (=TID) as the initial thread.
Credits to: julian@ for spotting how we can utilize UMA.
Thanks to: all who provided julian@ with test results.
is twofold:
1. When a 1:1 or M:N threaded process dumps core, we need to put the
register state of each of its kernel threads in the core file.
This can only be done by differentiating the pid field in the
respective note. For this we need the tid.
2. When thread support is present for remote debugging the kernel
with gdb(1), threads need to be identified by an integer due to
limitations in the remote protocol. This requires having a tid.
To minimize the impact of having thread IDs, threads that are created
as part of a fork (i.e. the initial thread in a process) will inherit
the process ID (i.e. tid=pid). Subsequent threads will have IDs larger
than PID_MAX to avoid interference with the pid allocation algorithm.
The assignment of tids is handled by thread_new_tid().
The thread ID allocation algorithm has been written with 3 assumptions
in mind:
1. IDs need to be created as fast a possible,
2. Reuse of IDs may happen instantaneously,
3. Someone else will write a better algorithm.
in the two consumers that need it.. processes using AIO and netncp.
Update docs. Say that process_exec is called with Giant, but not to
depend on it. All our consumers can handle it without Giant.
- no longer serialize on Giant for thread_single*() and family in fork,
exit and exec
- thread_wait() is mpsafe, assert no Giant
- reduce scope of Giant in exit to not cover thread_wait and just do
vm_waitproc().
- assert that thread_single() family are not called with Giant
- remove the DROP/PICKUP_GIANT macros from thread_single() family
- assert that thread_suspend_check() s not called with Giant
- remove manual drop_giant hack in thread_suspend_check since we know it
isn't held.
- remove the DROP/PICKUP_GIANT macros from thread_suspend_check() family
- mark kse_create() mpsafe
that we (p1) are currently running, we hold a reference on p_textvp which
means the vnode cannot go away. p2 cannot run yet (and hence cannot exit)
so this should be safe to do at this point. As a bonus, it removes a
block of under-Giant code that was there to support the vref.
process group and session dereferences. Also, check that p_pgrp and
p_sesssion are NULL before dereferencing them.
- Push down Giant in fork1().
Requested by: peter
the syscall arguments and does the suser() permission check, and
kern_mlock(), which does the resource limit checking and calls
vm_map_wire(). Split munlock() in a similar way.
Enable the RLIMIT_MEMLOCK checking code in kern_mlock().
Replace calls to vslock() and vsunlock() in the sysctl code with
calls to kern_mlock() and kern_munlock() so that the sysctl code
will obey the wired memory limits.
Nuke the vslock() and vsunlock() implementations, which are no
longer used.
Add a member to struct sysctl_req to track the amount of memory
that is wired to handle the request.
Modify sysctl_wire_old_buffer() to return an error if its call to
kern_mlock() fails. Only wire the minimum of the length specified
in the sysctl request and the length specified in its argument list.
It is recommended that sysctl handlers that use sysctl_wire_old_buffer()
should specify reasonable estimates for the amount of data they
want to return so that only the minimum amount of memory is wired
no matter what length has been specified by the request.
Modify the callers of sysctl_wire_old_buffer() to look for the
error return.
Modify sysctl_old_user to obey the wired buffer length and clean up
its implementation.
Reviewed by: bms
- struct plimit includes a mutex to protect a reference count. The plimit
structure is treated similarly to struct ucred in that is is always copy
on write, so having a reference to a structure is sufficient to read from
it without needing a further lock.
- The proc lock protects the p_limit pointer and must be held while reading
limits from a process to keep the limit structure from changing out from
under you while reading from it.
- Various global limits that are ints are not protected by a lock since
int writes are atomic on all the archs we support and thus a lock
wouldn't buy us anything.
- All accesses to individual resource limits from a process are abstracted
behind a simple lim_rlimit(), lim_max(), and lim_cur() API that return
either an rlimit, or the current or max individual limit of the specified
resource from a process.
- dosetrlimit() was renamed to kern_setrlimit() to match existing style of
other similar syscall helper functions.
- The alpha OSF/1 compat layer no longer calls getrlimit() and setrlimit()
(it didn't used the stackgap when it should have) but uses lim_rlimit()
and kern_setrlimit() instead.
- The svr4 compat no longer uses the stackgap for resource limits calls,
but uses lim_rlimit() and kern_setrlimit() instead.
- The ibcs2 compat no longer uses the stackgap for resource limits. It
also no longer uses the stackgap for accessing sysctl's for the
ibcs2_sysconf() syscall but uses kernel_sysctl() instead. As a result,
ibcs2_sysconf() no longer needs Giant.
- The p_rlimit macro no longer exists.
Submitted by: mtm (mostly, I only did a few cleanups and catchups)
Tested on: i386
Compiled on: alpha, amd64
Presumably, at some point, you had to include jail.h if you included
proc.h, but that is no longer required.
Result of: self injury involving adding something to struct prison
setting the new process' p_pgrp again before inserting it in the p_pglist.
Without it we can get the new process to be inserted in a different p_pglist
than the one p2->p_pgrp points to, and this is not something we want to happen.
This is not a fix, merely a bandaid, but it will work until someone finds a
better way to do it.
Discussed with: jhb (a long time ago)
is useless for threaded programs, multiple threads can not share same
stack.
The alternative signal stack is private for thread, no lock is needed,
the orignal P_ALTSTACK is now moved into td_pflags and renamed to
TDP_ALTSTACK.
For single thread or Linux clone() based threaded program, there is no
semantic changed, because those programs only have one kernel thread
in every process.
Reviewed by: deischen, dfr
clobbers this variable. Long ago, when the idle loop wasn't in a
process, it set switchtime.tv_sec to zero to indicate that the time
needs to be read after the idle loop finishes. The special case for
this isn't needed now that there is an idle process (for each CPU).
The time is read in the normal way when the idle process is switched
away from. The seconds component of the time is only zero for the
first second after the uptime is set, and the mostly-dead code was only
executed during this time. (This was slightly broken by using uptimes
instead of times relative to the Epoch -- in the original version the
seconds component of the time was only 0 for the first second after
the Epoch.)
In mi_switch(), moved the setting of switchticks to just after the
first (and now only) setting of switchtime. This setting used to be
delayed since a late setting was needed for the idle case and an early
setting was not needed. Now the early setting is needed so that
fork_exit() doesn't need to set either switchtime or switchticks.
Removed now-completely-rotted comment attached to this. Most of the
code described by the comment had already moved to sched_switch().
begin with sched_lock held but not recursed, so this variable was
always 0.
Removed fixup of sched_lock.mtx_recurse after context switches in
sched_switch(). Context switches always end with this variable in the
same state that it began in, so there is no need to fix it up. Only
sched_lock.mtx_lock really needs a fixup.
Replaced fixup of sched_lock.mtx_recurse in fork_exit() by an assertion
that sched_lock is owned and not recursed after it is fixed up. This
assertion much match the one in mi_switch(), and if sched_lock were
recursed then a non-null fixup of sched_lock.mtx_recurse would probably
be needed again, unlike in sched_switch(), since fork_exit() doesn't
return to its caller in the normal way.
