Adding Direct IO Support

Adding O_DIRECT support to ZFS to bypass the ARC for writes/reads.

O_DIRECT support in ZFS will always ensure there is coherency between
buffered and O_DIRECT IO requests. This ensures that all IO requests,
whether buffered or direct, will see the same file contents at all
times. Just as in other FS's , O_DIRECT does not imply O_SYNC. While
data is written directly to VDEV disks, metadata will not be synced
until the associated  TXG is synced.
For both O_DIRECT read and write request the offset and request sizes,
at a minimum, must be PAGE_SIZE aligned. In the event they are not,
then EINVAL is returned unless the direct property is set to always (see
below).

For O_DIRECT writes:
The request also must be block aligned (recordsize) or the write
request will take the normal (buffered) write path. In the event that
request is block aligned and a cached copy of the buffer in the ARC,
then it will be discarded from the ARC forcing all further reads to
retrieve the data from disk.

For O_DIRECT reads:
The only alignment restrictions are PAGE_SIZE alignment. In the event
that the requested data is in buffered (in the ARC) it will just be
copied from the ARC into the user buffer.

For both O_DIRECT writes and reads the O_DIRECT flag will be ignored in
the event that file contents are mmap'ed. In this case, all requests
that are at least PAGE_SIZE aligned will just fall back to the buffered
paths. If the request however is not PAGE_SIZE aligned, EINVAL will
be returned as always regardless if the file's contents are mmap'ed.

Since O_DIRECT writes go through the normal ZIO pipeline, the
following operations are supported just as with normal buffered writes:
Checksum
Compression
Encryption
Erasure Coding
There is one caveat for the data integrity of O_DIRECT writes that is
distinct for each of the OS's supported by ZFS.
FreeBSD - FreeBSD is able to place user pages under write protection so
          any data in the user buffers and written directly down to the
	  VDEV disks is guaranteed to not change. There is no concern
	  with data integrity and O_DIRECT writes.
Linux - Linux is not able to place anonymous user pages under write
        protection. Because of this, if the user decides to manipulate
	the page contents while the write operation is occurring, data
	integrity can not be guaranteed. However, there is a module
	parameter `zfs_vdev_direct_write_verify` that controls the
	if a O_DIRECT writes that can occur to a top-level VDEV before
	a checksum verify is run before the contents of the I/O buffer
        are committed to disk. In the event of a checksum verification
	failure the write will return EIO. The number of O_DIRECT write
	checksum verification errors can be observed by doing
	`zpool status -d`, which will list all verification errors that
	have occurred on a top-level VDEV. Along with `zpool status`, a
	ZED event will be issues as `dio_verify` when a checksum
	verification error occurs.

ZVOLs and dedup is not currently supported with Direct I/O.

A new dataset property `direct` has been added with the following 3
allowable values:
disabled - Accepts O_DIRECT flag, but silently ignores it and treats
	   the request as a buffered IO request.
standard - Follows the alignment restrictions  outlined above for
	   write/read IO requests when the O_DIRECT flag is used.
always   - Treats every write/read IO request as though it passed
           O_DIRECT and will do O_DIRECT if the alignment restrictions
	   are met otherwise will redirect through the ARC. This
	   property will not allow a request to fail.

There is also a module parameter zfs_dio_enabled that can be used to
force all reads and writes through the ARC. By setting this module
parameter to 0, it mimics as if the  direct dataset property is set to
disabled.

Reviewed-by: Brian Behlendorf <behlendorf@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Atkinson <batkinson@lanl.gov>
Co-authored-by: Mark Maybee <mark.maybee@delphix.com>
Co-authored-by: Matt Macy <mmacy@FreeBSD.org>
Co-authored-by: Brian Behlendorf <behlendorf@llnl.gov>
Closes #10018

man/man4/zfs.4

@@ -291,6 +291,14 @@ Default dnode block size as a power of 2.
 .It Sy zfs_default_ibs Ns = Ns Sy 17 Po 128 KiB Pc Pq int
 Default dnode indirect block size as a power of 2.
 .
+.It Sy zfs_dio_enabled Ns = Ns Sy 0 Ns | Ns 1 Pq int
+Enable Direct I/O.
+If this setting is 0, then all I/O requests will be directed through the ARC
+acting as though the dataset property
+.Sy direct
+was set to
+.Sy disabled .
+.
 .It Sy zfs_history_output_max Ns = Ns Sy 1048576 Ns B Po 1 MiB Pc Pq u64
 When attempting to log an output nvlist of an ioctl in the on-disk history,
 the output will not be stored if it is larger than this size (in bytes).
@@ -416,6 +424,26 @@ May be increased up to
 .Sy ASHIFT_MAX Po 16 Pc ,
 but this may negatively impact pool space efficiency.
 .
+.It Sy zfs_vdev_direct_write_verify Ns = Ns Sy Linux 1 | FreeBSD 0 Pq uint
+If non-zero, then a Direct I/O write's checksum will be verified every
+time the write is issued and before it is commited to the block pointer.
+In the event the checksum is not valid then the I/O operation will return EIO.
+This module parameter can be used to detect if the
+contents of the users buffer have changed in the process of doing a Direct I/O
+write.
+It can also help to identify if reported checksum errors are tied to Direct I/O
+writes.
+Each verify error causes a
+.Sy dio_verify
+zevent.
+Direct Write I/O checkum verify errors can be seen with
+.Nm zpool Cm status Fl d .
+The default value for this is 1 on Linux, but is 0 for
+.Fx
+because user pages can be placed under write protection in
+.Fx
+before the Direct I/O write is issued.
+.
 .It Sy zfs_vdev_min_auto_ashift Ns = Ns Sy ASHIFT_MIN Po 9 Pc Pq uint
 Minimum ashift used when creating new top-level vdevs.
 .
@@ -1093,6 +1121,9 @@ This will smoothly handle between ten times and a tenth of this number.
 .Pp
 .Sy zfs_delay_scale No \(mu Sy zfs_dirty_data_max Em must No be smaller than Sy 2^64 .
 .
+.It Sy zfs_dio_write_verify_events_per_second Ns = Ns Sy 20 Ns /s Pq uint
+Rate limit Direct I/O write verify events to this many per second.
+.
 .It Sy zfs_disable_ivset_guid_check Ns = Ns Sy 0 Ns | Ns 1 Pq int
 Disables requirement for IVset GUIDs to be present and match when doing a raw
 receive of encrypted datasets.

man/man7/zfsprops.7

@@ -1039,6 +1039,48 @@ See the
 section of
 .Xr zfsconcepts 7 .
 .It Xo
+.Sy direct Ns = Ns Sy disabled Ns | Ns Sy standard Ns | Ns Sy always
+.Xc
+Controls the behavior of Direct I/O requests
+.Pq e.g. Dv O_DIRECT .
+The
+.Sy standard
+behavior for Direct I/O requests is to bypass the ARC when possible.
+These requests will not be cached and performance will be limited by the
+raw speed of the underlying disks
+.Pq Dv this is the default .
+.Sy always
+causes every properly aligned read or write to be treated as a direct request.
+.Sy disabled
+causes the O_DIRECT flag to be silently ignored and all direct requests will
+be handled by the ARC.
+This is the default behavior for OpenZFS 2.2 and prior releases.
+.Pp
+Bypassing the ARC requires that a direct request be correctly aligned.
+For write requests the starting offset and size of the request must be
+.Sy recordsize Ns
+-aligned, if not then the unaligned portion of the request will be silently
+redirected through the ARC.
+For read requests there is no
+.Sy recordsize
+alignment restriction on either the starting offset or size.
+All direct requests must use a page-aligned memory buffer and the request
+size must be a multiple of the page size or an error is returned.
+.Pp
+Concurrently mixing buffered and direct requests to overlapping regions of
+a file can decrease performance.
+However, the resulting file will always be coherent.
+For example, a direct read after a buffered write will return the data
+from the buffered write.
+Furthermore, if an application uses
+.Xr mmap 2
+based file access then in order to maintain coherency all direct requests
+are converted to buffered requests while the file is mapped.
+Currently Direct I/O is not supported with zvols.
+If dedup is enabled on a dataset, Direct I/O writes will not check for
+deduplication.
+Deduplication and Direct I/O writes are currently incompatible.
+.It Xo
 .Sy dnodesize Ns = Ns Sy legacy Ns | Ns Sy auto Ns | Ns Sy 1k Ns | Ns
 .Sy 2k Ns | Ns Sy 4k Ns | Ns Sy 8k Ns | Ns Sy 16k
 .Xc

man/man8/zpool-events.8

@@ -98,6 +98,17 @@ This can be an indicator of problems with the underlying storage device.
 The number of delay events is ratelimited by the
 .Sy zfs_slow_io_events_per_second
 module parameter.
+.It Sy dio_verify
+Issued when there was a checksum verify error after a Direct I/O write has been
+issued.
+This event can only take place if the module parameter
+.Sy zfs_vdev_direct_write_verify
+is not set to zero.
+See
+.Xr zfs 4
+for more details on the
+.Sy zfs_vdev_direct_write_verify
+module paramter.
 .It Sy config
 Issued every time a vdev change have been done to the pool.
 .It Sy zpool
@@ -408,8 +419,9 @@ ZIO_STAGE_VDEV_IO_DONE:0x00400000:RW--XT
 ZIO_STAGE_VDEV_IO_ASSESS:0x00800000:RW--XT
 
 ZIO_STAGE_CHECKSUM_VERIFY:0x01000000:R-----
+ZIO_STAGE_DIO_CHECKSUM_VERIFY:0x02000000:-W----
 
-ZIO_STAGE_DONE:0x02000000:RWFCXT
+ZIO_STAGE_DONE:0x04000000:RWFCXT
 .TE
 .
 .Sh I/O FLAGS

man/man8/zpool-status.8

@@ -36,7 +36,7 @@
 .Sh SYNOPSIS
 .Nm zpool
 .Cm status
-.Op Fl DegiLpPstvx
+.Op Fl dDegiLpPstvx
 .Op Fl T Sy u Ns | Ns Sy d
 .Op Fl c Op Ar SCRIPT1 Ns Oo , Ns Ar SCRIPT2 Oc Ns …
 .Oo Ar pool Oc Ns …
@@ -81,6 +81,15 @@ to display vdevs in flat hierarchy instead of nested vdev objects.
 Specify
 .Sy --json-pool-key-guid
 to set pool GUID as key for pool objects instead of pool names.
+.It Fl d
+Display the number of Direct I/O write checksum verify errors that have occured
+on a top-level VDEV.
+See
+.Sx zfs_vdev_direct_write_verify
+in
+.Xr zfs 4
+for details about the conditions that can cause Direct I/O write checksum
+verify failures to occur.
 .It Fl D
 Display a histogram of deduplication statistics, showing the allocated
 .Pq physically present on disk