This is a much shorter set of patches that were on the go but didn't make it
in to the early pull request for the merge window. It's really a set of bug
fixes plus some final cleanup work on the new tag queue API.
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAABAgAGBQJUlaYEAAoJEDeqqVYsXL0MmXAH/2UUcE11p0KBHMR4cAn76xrG
9093ZT9VZ4LH/Z7PbgwIWC4YHDqVpwA1+Trj1mh8PxiZz2SopWe27O2lQMRS5VUc
MN28kbmK3L0jQj+OUez10Da6k0hU/KL8TlWT765MxFDKCaAuPZ4u541tyZEIGmLL
olOQrn/fSlu+18QqqZ+D2rMaK7kGH6ZgbOadnRfYGkLjU4YeAMEC9L7UgnDxHiaN
gZozoARkGeAnDJERVETRTtAiOXGRH8sGCpue0yYlhZXpAQ9cFUkS/hMqDWnaVC2S
0x0w34RvbxSqO0gPT0K5XLoMiFyg04vnZ2xBVFBsANQTSEjQJO8USNAa4r74hf8=
=D3eN
-----END PGP SIGNATURE-----
Merge tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull SCSI update from James Bottomley:
"This is a much shorter set of patches that were on the go but didn't
make it in to the early pull request for the merge window. It's
really a set of bug fixes plus some final cleanup work on the new tag
queue API"
* tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
storvsc: ring buffer failures may result in I/O freeze
ipr: set scsi_level correctly for disk arrays
ipr: add support for async scanning to speed up boot
scsi_debug: fix missing "break;" in SDEBUG_UA_CAPACITY_CHANGED case
scsi_debug: take sdebug_host_list_lock when changing capacity
scsi_debug: improve driver description in Kconfig
scsi_debug: fix compare and write errors
qla2xxx: fix race in handling rport deletion during recovery causes panic
scsi: blacklist RSOC for Microsoft iSCSI target devices
scsi: fix random memory corruption with scsi-mq + T10 PI
Revert "[SCSI] mpt3sas: Remove phys on topology change"
Revert "[SCSI] mpt2sas: Remove phys on topology change."
esas2r: Correct typos of "validate" in a comment
fc: FCP_PTA_SIMPLE is 0
ibmvfc: remove unused tag variable
scsi: remove MSG_*_TAG defines
scsi: remove scsi_set_tag_type
scsi: remove scsi_get_tag_type
scsi: never drop to untagged mode during queue ramp down
scsi: remove ->change_queue_type method
Pull SCSI target fixes from Nicholas Bellinger:
"The highlights this merge window include:
- Allow target fabric drivers to function as built-in. (Roland)
- Fix tcm_loop multi-TPG endpoint nexus bug. (Hannes)
- Move per device config_item_type into se_subsystem_api, allowing
configfs attributes to be defined at module_init time. (Jerome +
nab)
- Convert existing IBLOCK/FILEIO/RAMDISK/PSCSI/TCMU drivers to use
external configfs attributes. (nab)
- A number of iser-target fixes related to active session + network
portal shutdown stability during extended stress testing. (Sagi +
Slava)
- Dynamic allocation of T10-PI contexts for iser-target, fixing a
potentially bogus iscsi_np->tpg_np pointer reference in >= v3.14
code. (Sagi)
- iser-target performance + scalability improvements. (Sagi)
- Fixes for SPC-4 Persistent Reservation AllRegistrants spec
compliance. (Ilias + James + nab)
- Avoid potential short kern_sendmsg() in iscsi-target for now until
Al's conversion to use msghdr iteration is merged post -rc1.
(Viro)
Also, Sagi has requested a number of iser-target patches (9) that
address stability issues he's encountered during extended stress
testing be considered for v3.10.y + v3.14.y code. Given the amount of
LOC involved, it will certainly require extra backporting effort.
Apologies in advance to Greg-KH & Co on this. Sagi and I will be
working post-merge to ensure they each get applied correctly"
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending: (53 commits)
target: Allow AllRegistrants to re-RESERVE existing reservation
uapi/linux/target_core_user.h: fix headers_install.sh badness
iscsi-target: Fail connection on short sendmsg writes
iscsi-target: nullify session in failed login sequence
target: Avoid dropping AllRegistrants reservation during unregister
target: Fix R_HOLDER bit usage for AllRegistrants
iscsi-target: Drop left-over bogus iscsi_np->tpg_np
iser-target: Fix wc->wr_id cast warning
iser-target: Remove code duplication
iser-target: Adjust log levels and prettify some prints
iser-target: Use debug_level parameter to control logging level
iser-target: Fix logout sequence
iser-target: Don't wait for session commands from completion context
iser-target: Reduce CQ lock contention by batch polling
iser-target: Introduce isert_poll_budget
iser-target: Remove an atomic operation from the IO path
iser-target: Remove redundant call to isert_conn_terminate
iser-target: Use single CQ for TX and RX
iser-target: Centralize completion elements to a context
iser-target: Cast wr_id with uintptr_t instead of unsinged long
...
* Implement the relevant invalidation functions (zap MTTs as needed)
* Implement interlocking (and rollback in the page fault handlers) for
cases of a racing notifier and fault.
* With this patch we can now enable the capability bits for supporting RC
send/receive/RDMA read/RDMA write, and UD send.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This patch implement a page fault handler (leaving the pages pinned as
of time being). The page fault handler handles initiator and responder
page faults for UD/RC transports, for send/receive operations, as well
as RDMA read/write initiator support.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Refactor MR registration and cleanup, and fix reg_pages accounting.
* Create a work queue to handle page fault events in a kthread context.
* Register a fault handler to get events from the core for each QP.
The registered fault handler is empty in this patch, and only a later
patch implements it.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The new function allows updating the page tables of a memory region
after it was created. This can be used to handle page faults and page
invalidations.
Since mlx5_ib_update_mtt will need to work from within page invalidation,
so it must not block on memory allocation. It employs an atomic memory
allocation mechanism that is used as a fallback when kmalloc(GFP_ATOMIC) fails.
In order to reuse code from mlx5_ib_populate_pas, the patch splits
this function and add the needed parameters.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This patch wraps together several changes needed for on-demand paging support
in the mlx5_ib_populate_pas function, and when registering memory regions.
* Instead of accepting a UMR bit telling the function to enable all
access flags, the function now accepts the access flags themselves.
* For on-demand paging memory regions, fill the memory tables from the
correct list, and enable/disable the access flags per-page according
to whether the page is present.
* A new bit is set to enable writing of access flags when using the
firmware create_mkey command.
* Disable contig pages when on-demand paging is enabled.
In addition the patch changes the UMR code to use PTR_ALIGN instead of
our own macro.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The patch adds infrastructure to query ODP capabilities in the mlx5
driver. The code will read the capabilities from the device, and
enable only those capabilities that both the driver and the device
supports. At this point ODP is not supported, so no capability is
copied from the device, but the patch exposes the global ODP device
capability bit.
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
In case the last argument of the connection string is processed as a
string (destination GID for example).
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Acked-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Add an interval tree implementation for ODP umems. Create an
interval tree for each ucontext (including a count of the number of
ODP MRs in this context, semaphore, etc.), and register ODP umems in
the interval tree.
* Add MMU notifiers handling functions, using the interval tree to
notify only the relevant umems and underlying MRs.
* Register to receive MMU notifier events from the MM subsystem upon
ODP MR registration (and unregister accordingly).
* Add a completion object to synchronize the destruction of ODP umems.
* Add mechanism to abort page faults when there's a concurrent invalidation.
The way we synchronize between concurrent invalidations and page
faults is by keeping a counter of currently running invalidations, and
a sequence number that is incremented whenever an invalidation is
caught. The page fault code checks the counter and also verifies that
the sequence number hasn't progressed before it updates the umem's
page tables. This is similar to what the kvm module does.
In order to prevent the case where we register a umem in the middle of
an ongoing notifier, we also keep a per ucontext counter of the total
number of active mmu notifiers. We only enable new umems when all the
running notifiers complete.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Yuval Dagan <yuvalda@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Extend the umem struct to keep the ODP related data.
* Allocate and initialize the ODP related information in the umem
(page_list, dma_list) and freeing as needed in the end of the run.
* Store a reference to the process PID struct in the ucontext. Used to
safely obtain the task_struct and the mm during fault handling,
without preventing the task destruction if needed.
* Add 2 helper functions: ib_umem_odp_map_dma_pages and
ib_umem_odp_unmap_dma_pages. These functions get the DMA addresses
of specific pages of the umem (and, currently, pin them).
* Support for page faults only - IB core will keep the reference on
the pages used and call put_page when freeing an ODP umem
area. Invalidations support will be added in a later patch.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
* Add a configuration option for enable on-demand paging support in
the infiniband subsystem (CONFIG_INFINIBAND_ON_DEMAND_PAGING). In a
later patch, this configuration option will select the MMU_NOTIFIER
configuration option to enable mmu notifiers.
* Add a flag for on demand paging (ODP) support in the IB device capabilities.
* Add a flag to request ODP MR in the access flags to reg_mr.
* Fail registrations done with the ODP flag when the low-level driver
doesn't support this.
* Change the conditions in which an MR will be writable to explicitly
specify the access flags. This is to avoid making an MR writable just
because it is an ODP MR.
* Add a ODP capabilities to the extended query device verb.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Add extensible query device capabilities verb to allow adding new features.
ib_uverbs_ex_query_device is added and copy_query_dev_fields is used to
copy capability fields to be used by both ib_uverbs_query_device and
ib_uverbs_ex_query_device.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Add a helper function mlx5_ib_read_user_wqe to read information from
user-space owned work queues. The function will be used in a later
patch by the page-fault handling code in mlx5_ib.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
[ Add stub for ib_umem_copy_from() for CONFIG_INFINIBAND_USER_MEM=n
- Roland ]
Signed-off-by: Roland Dreier <roland@purestorage.com>
In some drivers there's a need to read data from a user space area
that was pinned using ib_umem when running from a different process
context.
The ib_umem_copy_from function allows reading data from the physical
pages pinned in the ib_umem struct.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
In order to allow umems that do not pin memory, we need the umem to
keep track of its region's address.
This makes the offset field redundant, and so this patch removes it.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
The current UMR interface doesn't allow partial updates to a memory
region's page tables. This patch changes the interface to allow that.
It also changes the way the UMR operation validates the memory
region's state. When set, IB_SEND_UMR_FAIL_IF_FREE will cause the UMR
operation to fail if the MKEY is in the free state. When it is
unchecked the operation will check that it isn't in the free state.
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Shachar Raindel <raindel@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Since UMR code now uses its own context struct on the stack, the pas
and dma pointers for the UMR operation that remained in the mlx5_ib_mr
struct are not necessary. This patch removes them.
Fixes: a74d24168d ("IB/mlx5: Refactor UMR to have its own context struct")
Signed-off-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
For user applications that use UD QPs, always resolve destination MAC
from the GRH. This is to avoid failure due to any garbage value in
the attr->dmac.
Signed-off-by: Selvin Xavier <selvin.xavier@emulex.com>
Signed-off-by: Devesh Sharma <devesh.sharma@emulex.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
This error was detected by sparse static checker:
drivers/infiniband/hw/mlx4/mr.c:226:21: warning: symbol 'err' shadows an earlier one
drivers/infiniband/hw/mlx4/mr.c:197:13: originally declared here
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Following few recent Block integrity updates, we align the iSER data
integrity offload settings with:
- Deprecate pi_guard module param
- Expose support for DIX type 0.
- Use scsi_transfer_length for the transfer length
- Get pi_interval, ref_tag, ref_remap, bg_type and
check_mask setting from scsi_cmnd
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use likely() for wc.status == IB_WC_SUCCESS
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
And fix a checkpatch warning.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
No reason to settle with four, can use the min between device max comp
vectors and number of cores.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
It is enough to check mem_h pointer assignment, mem_h == NULL will
indicate that buffer is not registered using mr.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When closing the connection, we should first terminate the connection
(in case it was not previously terminated) to guarantee the QP is in
error state and we are done with servicing IO. Only then go ahead with
tasks cleanup via iscsi_conn_stop.
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
In certain scenarios (target kill with live IO) scsi TMFs may race
with iser RDMA teardown, which might cause NULL dereference on iser IB
device handle (which might have been freed). In this case we take a
conditional lock for TMFs and check the connection state (avoid
introducing lock contention in the IO path). This is indeed best
effort approach, but sufficient to survive multi targets sudden death
while heavy IO is inflight.
While we are on it, add a nice kernel-doc style documentation.
Reported-by: Ariel Nahum <arieln@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
If rdma_cm error event comes after ep_poll but before conn_bind, we
should protect against dereferncing the device (which may have been
terminated) in session_create and conn_create (already protected)
callbacks.
Signed-off-by: Ariel Nahum <arieln@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Use uintptr_t to handle wr_id casting, which was found by Kbuild test
robot and smatch. Also remove an internal definition of variable which
potentially shadows an external one (and make sparse happy).
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Fix a regression was introduced in commit 6df5a128f0 ("IB/iser:
Suppress scsi command send completions").
The sig_count was wrongly set to be static variable, thus it is
possible that we won't reach to (sig_count % ISER_SIGNAL_BATCH) == 0
condition (due to races) and the send queue will be overflowed.
Instead keep sig_count per connection. We don't need it to be atomic
as we are safe under the iscsi session frwd_lock taken by libiscsi on
the queuecommand path.
Fixes: 6df5a128f0 ("IB/iser: Suppress scsi command send completions")
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When creating a connection QP we choose the least used CQ and inc the
number of active QPs on that. If we fail to create the QP, we need to
decrement the active QPs counter.
Reported-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
No real need to wait for TIMEWAIT_EXIT before we destroy the RDMA
resources (also TIMEAWAIT_EXIT is not guarenteed to always arrive). As
for the cma_id, only destroy it if the state is not DOWN where in this
case, conn_release is already running and we don't want to compete.
Signed-off-by: Ariel Nahum <arieln@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
In case of the HCA going into catasrophic error flow, the
beacon post_send is likely to fail, so surely there will
be no completion for it.
In this case, use a best effort approach and don't wait for beacon
completion if we failed to post the send.
Reported-by: Alex Tabachnik <alext@mellanox.com>
Signed-off-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Re-adjust max CQEs per CQ and max send_wr per QP according
to the resource limits supported by underlying hardware.
Signed-off-by: Minh Tran <minhduc.tran@emulex.com>
Signed-off-by: Jayamohan Kallickal <jayamohan.kallickal@emulex.com>
Acked-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Various places in the IPoIB code had a deadlock related to flushing
the ipoib workqueue. Now that we have per device workqueues and a
specific flush workqueue, there is no longer a deadlock issue with
flushing the device specific workqueues and we can do so unilaterally.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
We used to pass a flush variable to mcast_stop_thread to indicate if
we should flush the workqueue or not. This was due to some code
trying to flush a workqueue that it was currently running on which is
a no-no. Now that we have per-device work queues, and now that
ipoib_mcast_restart_task has taken the fact that it is queued on a
single thread workqueue with all of the ipoib_mcast_join_task's and
therefore has no need to stop the join task while it runs, we can do
away with the flush parameter and unilaterally flush always.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
During my recent work on the rtnl lock deadlock in the IPoIB driver, I
saw that even once I fixed the apparent races for a single device, as
soon as that device had any children, new races popped up. It turns
out that this is because no matter how well we protect against races
on a single device, the fact that all devices use the same workqueue,
and flush_workqueue() flushes *everything* from that workqueue, we can
have one device in the middle of a down and holding the rtnl lock and
another totally unrelated device needing to run mcast_restart_task,
which wants the rtnl lock and will loop trying to take it unless is
sees its own FLAG_ADMIN_UP flag go away. Because the unrelated
interface will never see its own ADMIN_UP flag drop, the interface
going down will deadlock trying to flush the queue. There are several
possible solutions to this problem:
Make carrier_on_task and mcast_restart_task try to take the rtnl for
some set period of time and if they fail, then bail. This runs the
real risk of dropping work on the floor, which can end up being its
own separate kind of deadlock.
Set some global flag in the driver that says some device is in the
middle of going down, letting all tasks know to bail. Again, this can
drop work on the floor. I suppose if our own ADMIN_UP flag doesn't go
away, then maybe after a few tries on the rtnl lock we can queue our
own task back up as a delayed work and return and avoid dropping work
on the floor that way. But I'm not 100% convinced that we won't cause
other problems.
Or the method this patch attempts to use, which is when we bring an
interface up, create a workqueue specifically for that interface, so
that when we take it back down, we are flushing only those tasks
associated with our interface. In addition, keep the global
workqueue, but now limit it to only flush tasks. In this way, the
flush tasks can always flush the device specific work queues without
having deadlock issues.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
In preparation for using per device work queues, we need to move the
start of the neighbor thread task to after ipoib_ib_dev_init and move
the destruction of the neighbor task to before ipoib_ib_dev_cleanup.
Otherwise we will end up freeing our workqueue with work possibly
still on it.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Our mcast_dev_flush routine and our mcast_restart_task can race
against each other. In particular, they both hold the priv->lock
while manipulating the rbtree and while removing mcast entries from
the multicast_list and while adding entries to the remove_list, but
they also both drop their locks prior to doing the actual removes.
The mcast_dev_flush routine is run entirely under the rtnl lock and so
has at least some locking. The actual race condition is like this:
Thread 1 Thread 2
ifconfig ib0 up
start multicast join for broadcast
multicast join completes for broadcast
start to add more multicast joins
call mcast_restart_task to add new entries
ifconfig ib0 down
mcast_dev_flush
mcast_leave(mcast A)
mcast_leave(mcast A)
As mcast_leave calls ib_sa_multicast_leave, and as member in
core/multicast.c is ref counted, we run into an unbalanced refcount
issue. To avoid stomping on each others removes, take the rtnl lock
specifically when we are deleting the entries from the remove list.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Commit a9c8ba5884 ("IPoIB: Fix usage of uninitialized multicast
objects") added a new flag MCAST_JOIN_STARTED, but was not very strict
in how it was used. We didn't always initialize the completion struct
before we set the flag, and we didn't always call complete on the
completion struct from all paths that complete it. This made it less
than totally effective, and certainly made its use confusing. And in
the flush function we would use the presence of this flag to signal
that we should wait on the completion struct, but we never cleared
this flag, ever. This is further muddied by the fact that we overload
the MCAST_FLAG_BUSY flag to mean two different things: we have a join
in flight, and we have succeeded in getting an ib_sa_join_multicast.
In order to make things clearer and aid in resolving the rtnl deadlock
bug I've been chasing, I cleaned this up a bit.
1) Remove the MCAST_JOIN_STARTED flag entirely
2) Un-overload MCAST_FLAG_BUSY so it now only means a join is in-flight
3) Test on mcast->mc directly to see if we have completed
ib_sa_join_multicast (using IS_ERR_OR_NULL)
4) Make sure that before setting MCAST_FLAG_BUSY we always initialize
the mcast->done completion struct
5) Make sure that before calling complete(&mcast->done), we always clear
the MCAST_FLAG_BUSY bit
6) Take the mcast_mutex before we call ib_sa_multicast_join and also
take the mutex in our join callback. This forces
ib_sa_multicast_join to return and set mcast->mc before we process
the callback. This way, our callback can safely clear mcast->mc
if there is an error on the join and we will do the right thing as
a result in mcast_dev_flush.
7) Because we need the mutex to synchronize mcast->mc, we can no
longer call mcast_sendonly_join directly from mcast_send and
instead must add sendonly join processing to the mcast_join_task
A number of different races are resolved with these changes. These
races existed with the old MCAST_FLAG_BUSY usage, the
MCAST_JOIN_STARTED flag was an attempt to address them, and while it
helped, a determined effort could still trip things up.
One race looks something like this:
Thread 1 Thread 2
ib_sa_join_multicast (as part of running restart mcast task)
alloc member
call callback
ifconfig ib0 down
wait_for_completion
callback call completes
wait_for_completion in
mcast_dev_flush completes
mcast->mc is PTR_ERR_OR_NULL
so we skip ib_sa_leave_multicast
return from callback
return from ib_sa_join_multicast
set mcast->mc = return from ib_sa_multicast
We now have a permanently unbalanced join/leave issue that trips up the
refcounting in core/multicast.c
Another like this:
Thread 1 Thread 2 Thread 3
ib_sa_multicast_join
ifconfig ib0 down
priv->broadcast = NULL
join_complete
wait_for_completion
mcast->mc is not yet set, so don't clear
return from ib_sa_join_multicast and set mcast->mc
complete
return -EAGAIN (making mcast->mc invalid)
call ib_sa_multicast_leave
on invalid mcast->mc, hang
forever
By holding the mutex around ib_sa_multicast_join and taking the mutex
early in the callback, we force mcast->mc to be valid at the time we
run the callback. This allows us to clear mcast->mc if there is an
error and the join is going to fail. We do this before we complete
the mcast. In this way, mcast_dev_flush always sees consistent state
in regards to mcast->mc membership at the time that the
wait_for_completion() returns.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
We blindly assume that we can just take the rtnl lock and that will
prevent races with downing this interface. Unfortunately, that's not
the case. In ipoib_mcast_stop_thread() we will call flush_workqueue()
in an attempt to clear out all remaining instances of ipoib_join_task.
But, since this task is put on the same workqueue as the join task,
the flush_workqueue waits on this thread too. But this thread is
deadlocked on the rtnl lock. The better thing here is to use trylock
and loop on that until we either get the lock or we see that
FLAG_ADMIN_UP has been cleared, in which case we don't need to do
anything anyway and we just return.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
Setting the MTU can safely be moved to the carrier_on_task, which keeps
us from needing to take the rtnl lock in the join_finish section.
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
cxgb4_create_server() and cxgb4_create_server6() return NET_XMIT_*
values or a negative errno. iw_cxgb4 need to handle this correctly.
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
When transitioning into ERROR state, the QP was getting flushed after
waking up any waiters. This can cause applications to miss flushed work
requests which can stall an NFS mount.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
