Currently, the trim module uses the same algorithm for data and cache
devices when deciding to issue TRIM requests, based on how far in the
past the TXG is.
Unfortunately, this is not ideal for cache devices, because the L2ARC
doesn't use the concept of TXGs at all. In fact, when using a pool for
reading only, the L2ARC is written but the TXG counter doesn't
increase, and so no new TRIM requests are issued to the cache device.
This patch fixes the issue by using time instead of the TXG number as
the criteria for trimming on cache devices. The basic delay principle
stays the same, but parameters are expressed in seconds instead of
TXGs. The new parameters are named trim_l2arc_limit and
trim_l2arc_batch, and both default to 30 second.
Reviewed by: pjd (mentor)
Approved by: pjd (mentor)
Obtained from: 17122c31ac
MFC after: 2 weeks
This patch adds some improvements to the way the trim module considers
TXGs:
- Free ZIOs are registered with the TXG from the ZIO itself, not the
current SPA syncing TXG (which may be out of date);
- L2ARC are registered with a zero TXG number, as L2ARC has no concept
of TXGs;
- The TXG limit for issuing TRIMs is now computed from the last synced
TXG, not the currently syncing TXG. Indeed, under extremely unlikely
race conditions, there is a risk we could trim blocks which have been
freed in a TXG that has not finished syncing, resulting in potential
data corruption in case of a crash.
Reviewed by: pjd (mentor)
Approved by: pjd (mentor)
Obtained from: 5b46ad40d9
MFC after: 2 weeks
The trim map inflight writes tree assumes non-conflicting writes, i.e.
that there will never be two simultaneous write I/Os to the same range
on the same vdev. This seemed like a sane assumption; however, in
actual testing, it appears that repair I/Os can very well conflict
with "normal" writes.
I'm not quite sure if these conflicting writes are supposed to happen
or not, but in the mean time, let's ignore repair writes for now. This
should be safe considering that, by definition, we never repair blocks
that are freed.
Reviewed by: pjd (mentor)
Approved by: pjd (mentor)
Obtained from: Source: 6a3cebaf7c
This adds TRIM support to cache vdevs. When ARC buffers are removed
from the L2ARC in arc_hdr_destroy(), arc_release() or l2arc_evict(),
the size previously occupied by the buffer gets scheduled for TRIMming.
As always, actual TRIMs are only issued to the L2ARC after
txg_trim_limit.
Reviewed by: pjd (mentor)
Approved by: pjd (mentor)
Obtained from: 31aae37399
MFC after: 2 weeks
includes MFV 238590, 238592, 247580
MFV 238590, 238592:
In the first zfs ioctl restructuring phase, the libzfs_core library was
introduced. It is a new thin library that wraps around kernel ioctl's.
The idea is to provide a forward-compatible way of dealing with new
features. Arguments are passed in nvlists and not random zfs_cmd fields,
new-style ioctls are logged to pool history using a new method of
history logging.
http://blog.delphix.com/matt/2012/01/17/the-future-of-libzfs/
MFV 247580 [1]:
To address issues of several deadlocks and race conditions the locking
code around dsl_dataset was rewritten and the interface to synctasks
was changed.
User-Visible Changes:
"zfs snapshot" can create more arbitrary snapshots at once (atomically)
"zfs destroy" destroys multiple snapshots at once
"zfs recv" has improved performance
Backward Compatibility:
I have extended the compatibility layer to support full backward
compatibility by remapping or rewriting the responsible ioctl arguments.
Old utilities are fully supported by the new kernel module.
Forward Compatibility:
New utilities work with old kernels with the following restrictions:
- creating, destroying, holding and releasing of multiple snapshots
at once is not supported, this includes recursive (-r) commands
Illumos ZFS issues:
2882 implement libzfs_core
2900 "zfs snapshot" should be able to create multiple,
arbitrary snapshots at once
3464 zfs synctask code needs restructuring
References:
https://www.illumos.org/issues/2882https://www.illumos.org/issues/2900https://www.illumos.org/issues/3464 [1]
MFC after: 1 month
Sponsored by: Hybrid Logic Inc. [1]
locked. vnode_pager_setsize() might sleep waiting for the page after
EOF be unbusied.
Call vnode_pager_setsize() both for the regular and directory vnodes.
Reported by: mich
Reviewed by: rmacklem
Discussed with: avg, jhb
MFC after: 2 weeks
bufobj counter of the writes in progress is incremented. Other thread
inspecting the bufobj would consider it clean.
For the regular vnodes, the vnode lock is typically held both by the
thread performing the bufwrite() and an other thread doing syncing,
which prevents the situation. On the other hand, writes to the VCHR
vnodes are done without holding vnode lock.
Increment the write ref counter for the buffer object before calling
bundirty().
Sponsored by: The FreeBSD Foundation
Tested by: pho
MFC after: 2 weeks
becoming too low, the softdep flush thread processes the workitems,
which frees the space in journal, and then unsuspends the fs. The
softdep_flush() and other workitem processing functions busy the
filesystem before iterating over the worklist, to prevent the parallel
unmount from freeing the mount data. The vfs_busy() is called with
MBF_NOWAIT flag.
Now, if the unmount is already started and the filesystem is suspended
due to low journal space, the journal is never flushed and filesystem
is never unsuspended, because vfs_busy(MBF_NOWAIT) call cannot succeed
for the unmounting fs, and softdep_flush() does not process the
workitems. Unmount needs to write metadata, where it hangs in the
"suspfs" state.
Move the vn_start_write() call in the dounmount() before setting the
MNTK_UNMOUNT flag. This practically ensures that softdep_flush()
processed the pending journal writes by making dounmount() wait for
the lift of the suspension.
Sponsored by: The FreeBSD Foundation
Reported and tested by: pho
MFC after: 2 weeks
we need to call ufs_checkpath() to walk from our new location to
the root of the filesystem to ensure that we do not encounter
ourselves along the way. Until now, we accomplished this by reading
the ".." entries of each directory in our path until we reached
the root (or encountered an error). This change tries to avoid the
I/O of reading the ".." entries by first looking them up in the
name cache and only doing the I/O when the name cache lookup fails.
Reviewed by: kib
Tested by: Peter Holm
MFC after: 4 weeks
Setting DSCP support is done via O_SETDSCP which works for both
IPv4 and IPv6 packets. Fast checksum recalculation (RFC 1624) is done for IPv4.
Dscp can be specified by name (AFXY, CSX, BE, EF), by value
(0..63) or via tablearg.
Matching DSCP is done via another opcode (O_DSCP) which accepts several
classes at once (af11,af22,be). Classes are stored in bitmask (2 u32 words).
Many people made their variants of this patch, the ones I'm aware of are
(in alphabetic order):
Dmitrii Tejblum
Marcelo Araujo
Roman Bogorodskiy (novel)
Sergey Matveichuk (sem)
Sergey Ryabin
PR: kern/102471, kern/121122
MFC after: 2 weeks
auto-correction. This change makes re(4) establish a link with
a system using non-crossover UTP cable.
Tested by: Michael BlackHeart < amdmiek <> gmail dot com >
This change allows creating file descriptors with close-on-exec set in some
situations. SOCK_CLOEXEC and SOCK_NONBLOCK can be OR'ed in socket() and
socketpair()'s type parameter, and MSG_CMSG_CLOEXEC to recvmsg() makes file
descriptors (SCM_RIGHTS) atomically close-on-exec.
The numerical values for SOCK_CLOEXEC and SOCK_NONBLOCK are as in NetBSD.
MSG_CMSG_CLOEXEC is the first free bit for MSG_*.
The SOCK_* flags are not passed to MAC because this may cause incorrect
failures and can be done later via fcntl() anyway. On the other hand, audit
is expected to cope with the new flags.
For MSG_CMSG_CLOEXEC, unp_externalize() is extended to take a flags
argument.
Reviewed by: kib
"complete RX frames."
The 128 entry RX FIFO is really easy to fill up and miss refilling
when it's done in the ath taskq - as that gets blocked up doing
RX completion, TX completion and other random things.
So the 128 entry RX FIFO now gets emptied and refilled in the ath_intr()
task (and it grabs / releases locks, so now ath_intr() can't just be
a FAST handler yet!) but the locks aren't held for very long. The
completion part is done in the ath taskqueue context.
Details:
* Create a new completed frame list - sc->sc_rx_rxlist;
* Split the EDMA RX process queue into two halves - one that
processes the RX FIFO and refills it with new frames; another
that completes the completed frame list;
* When tearing down the driver, flush whatever is in the deferred
queue as well as what's in the FIFO;
* Create two new RX methods - one that processes all RX queues,
one that processes the given RX queue. When MSI is implemented,
we get told which RX queue the interrupt came in on so we can
specifically schedule that. (And I can do that with the non-MSI
path too; I'll figure that out later.)
* Convert the legacy code over to use these new RX methods;
* Replace all the instances of the RX taskqueue enqueue with a call
to a relevant RX method to enqueue one or all RX queues.
Tested:
* AR9380, STA
* AR9580, STA
* AR5413, STA
for the r248519:
For the cam-attached HBAs, allow the driver to specify that it accepts
the unmapped bio by the PIM_UNMAPPED flag. The CAM passes the
CAM_DATA_BIO data transfer type request for the unmapped bio, and the
driver could use the bus_dmamap_load_ccb() as a helper to
transparently handle the ccb.
Sponsored by: The FreeBSD Foundation
Reviewed by: scottl
Tested by: pho, scottl
The vnode-backed md(4) has to map the unmapped bio because VOP_READ()
and VOP_WRITE() interfaces do not allow to pass unmapped requests to
the filesystem. Vnode-backed md(4) uses pbufs instead of relying on
the bio_transient_map, to avoid usual md deadlock.
Sponsored by: The FreeBSD Foundation
Tested by: pho, scottl
The vnode-backed md(4) has to map the unmapped bio because VOP_READ()
and VOP_WRITE() interfaces do not allow to pass unmapped requests to
the filesystem. Vnode-backed md(4) uses pbufs instead of relying on
the bio_transient_map, to avoid usual md deadlock.
Sponsored by: The FreeBSD Foundation
Tested by: pho, scottl
buffer, transparently handling mapped or unmapped buffers. Its intent
is to replace the use of bzero(bp->b_data) in cases where the buffer
might be unmapped, to avoid unneeded upgrades.
Sponsored by: The FreeBSD Foundation
Tested by: pho
do not map the b_pages pages into buffer_map KVA. The use of the
unmapped buffers eliminate the need to perform TLB shootdown for
mapping on the buffer creation and reuse, greatly reducing the amount
of IPIs for shootdown on big-SMP machines and eliminating up to 25-30%
of the system time on i/o intensive workloads.
The unmapped buffer should be explicitely requested by the GB_UNMAPPED
flag by the consumer. For unmapped buffer, no KVA reservation is
performed at all. The consumer might request unmapped buffer which
does have a KVA reserve, to manually map it without recursing into
buffer cache and blocking, with the GB_KVAALLOC flag.
When the mapped buffer is requested and unmapped buffer already
exists, the cache performs an upgrade, possibly reusing the KVA
reservation.
Unmapped buffer is translated into unmapped bio in g_vfs_strategy().
Unmapped bio carry a pointer to the vm_page_t array, offset and length
instead of the data pointer. The provider which processes the bio
should explicitely specify a readiness to accept unmapped bio,
otherwise g_down geom thread performs the transient upgrade of the bio
request by mapping the pages into the new bio_transient_map KVA
submap.
The bio_transient_map submap claims up to 10% of the buffer map, and
the total buffer_map + bio_transient_map KVA usage stays the
same. Still, it could be manually tuned by kern.bio_transient_maxcnt
tunable, in the units of the transient mappings. Eventually, the
bio_transient_map could be removed after all geom classes and drivers
can accept unmapped i/o requests.
Unmapped support can be turned off by the vfs.unmapped_buf_allowed
tunable, disabling which makes the buffer (or cluster) creation
requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped
buffers are only enabled by default on the architectures where
pmap_copy_page() was implemented and tested.
In the rework, filesystem metadata is not the subject to maxbufspace
limit anymore. Since the metadata buffers are always mapped, the
buffers still have to fit into the buffer map, which provides a
reasonable (but practically unreachable) upper bound on it. The
non-metadata buffer allocations, both mapped and unmapped, is
accounted against maxbufspace, as before. Effectively, this means that
the maxbufspace is forced on mapped and unmapped buffers separately.
The pre-patch bufspace limiting code did not worked, because
buffer_map fragmentation does not allow the limit to be reached.
By Jeff Roberson request, the getnewbuf() function was split into
smaller single-purpose functions.
Sponsored by: The FreeBSD Foundation
Discussed with: jeff (previous version)
Tested by: pho, scottl (previous version), jhb, bf
MFC after: 2 weeks
breadn_flags(). Comparing with bread(), it adds an argument to pass
the flags to getblk().
Sponsored by: The FreeBSD Foundation
Tested by: pho
MFC after: 2 weeks
