[PATCH] libaio engine: fake work-around for sync issue

Needed to make the fsync() fallback work.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix libaio engine SYNC

io_prep_fsync() isn't supported at all in Linux currently, so
fall back to fsync() if we get an -EINVAL return value.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] CFLAGS: -fstack-protector doesn't work on older compilers

Just remove it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Net engine typo

READ -> DDIR_READ

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] SG IO engine: reduce allocations and memory leaks

We currently do 3 allocations and frees in the queued getevents
handler, we can easily reduce that to 0 by statically allocating
the buffers at init time.

The cleanup handler didn't free any of the currently allocated
structures, fix that up.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] SG IO engine: remember to set io_u->file in init

It's now mandatory, since it supports multiple files now.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] SG io engine: better handling of multiple files

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Network engine cleanups and comments

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Improve dependency checking on builds

From the similar fix for blktrace, from Vasily Tarasov <vtaras@openvz.org>.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Enable _FORTIFY_SOURCE and stack protector checking

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] We should check system() return value

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Remove SIGSEGV handler

We don't catch segfaults anymore, so remove that bit from the signal handler.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] aio example should use O_DIRECT IO

Otherwise it isn't async. Since fio defaults to buffered IO now, we need
to correct this example.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Don't create files for engines that don't need them

Introduce FIO_NULLIO as a way to control that from the io engine.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Misc fixes

- Change the hour/day eta display to be more clever.
- Style

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Network engine fixes

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Network io engine updates

- Use recv/send
- Set MSG_MORE on send if we are going to transmit more data
- Cleanups

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Document ->queue() in the skeleton engine

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Support residual io counts from io engines

We need this for requeuing support, the network engine makes this
pretty apparent (it's not unusual to see short tranfers there).

Basically we add an xfer_buf and xfer_buflen member to the io_u,
and these are the fields that the io engine MUST use. That allows
fio to increment and reset these appropriately, and simply requeue
the io_u for service of the next part of it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Support for > 1 network connection

Now the net ioengine also supports the nrfiles option to open
multiple connections.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Network engine updates

A few comments, pass down port short instead of buffer.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Simple support for networked IO

Adds a new ioengine, net. Use with ioengine=net, it supports only strict
reading or writing (no mixed reads/writes) to/from a single host.
The filename given must contain the host and port to connect to (or
listen from).

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Rename io engines

Typing fio-engine-foo is annoying, just rename them to their type. They
already reside in a private directory, no point in putting the fio-engine
prefix on the files.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix buffer alignment on non-4kb archs

Retrieve page size through sysconf.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add 'description' option

Can help provide a textual description of a job, if it's submitted
to someone for run/review.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix sg ioengine

We need to do the ->init() call after having opened the file,
otherwise sg doesn't know what file type it is dealing with.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix double free in sg engine error path

We need to clear td->io_ops->data if we free the structure,
otherwise the ->cleanup() handler will try to free it again.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix typo

WRITE -> DDIR_WRITE

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Only generate global bandwidth log if write_bw_log set

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add support for writing aggregate bandwidth log

This is the same that is shown while fio is running, it can be useful
for graphing purposes as well.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Unlink write file if !overwrite and it exists

Repeatability across runs, unlink the file so that the job
runs like it wasn't there to begin with.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Don't make file unlink the default

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] The last io depth entry is actually >= 64, not > 32

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Include io depth distribution statistics

Calculate the distribution of io depths for the job run.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Remove the time init check

We cannot control constructor ordering easily, hence the check
code was also in fio_gettime().

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Shorten 'commitrate' to 'CR'

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fixup rate usec cycle calculation for small values

If someone set rate to 2KiB/sec or something low in that range,
we would integer divide by zero. Do calculation in reads per msec
instead of full seconds and fix the zero division.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Improve rate functionality

usec_sleep() now takes into account the granularity of nanosleep().
Also stop using the remainder returned by nanosleep(), it doesn't
appear to always be accurate. Instead just check the elapsed time
manually.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Properly shutdown in case of serious errors

Introduce fio_abort to do a forced shutdown of everything for things
like segfaults.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Update HOWTO

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Cleanup some output code

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Merge branch 'master' of ssh://router/data/git/fio

[PATCH] Update irange documentation

Also allow ':' as the range delimeter, and '/' as the
direction delimeter.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Help: thinktime is in microseconds

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Abort early for bad option given to --cmdhelp

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Really fixup disk util init

Include disk io ticks, if the job itself creates the file. Don't
include it, if we create the file before running the job.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Revert "[PATCH] Move disk utilization setup to thread itself"

This reverts commit d52fb0c44567cf4fdc2a2c307c32672cc5a57170.

[PATCH] Move disk utilization setup to thread itself

We don't want it to include potential file creation before the job runs.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Warn if both option variable offset and callback given

If the callback is present, it is preferred to the offsets. So
giving offsets with a callback specified doesn't make sense.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Pretty up option looping

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Change output dev= to stdev=

It's really standard deviation, so make it a little easier to
understand.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Change O_DIRECT vs buffered setup

Change the default from O_DIRECT IO to normal buffered IO. That makes
more sense, as O_DIRECT is a special case and should be manually
enabled as such.

Do this by adding a option negate switch, so we don't need two sets
of parameters to control these options.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Change timeout to runtime and rearrange option entries

Runtime is a more descriptive name - the documentation is updated
as well, however the parsing has added an alias functionality to
support older names as well. So 'timeout' still works.

Also rearrange option entries, now they matter because the --cmdhelp
option will print them in order.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Thread option didn't work

We stored to the wrong location, woops.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Implement a better mean and standard deviation calculation

Based on Knuth's, volume 2.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] strtoul() -> strtol() conversion

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Merge branch 'master' of ssh://git.kernel.dk/data/git/fio

[PATCH] Fix standard deviation calculation

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Be strict in checking option ranges

Fail if given value falls outside of range instead of adjusting
it automatically.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Option minval should be signed

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Introduce bool option type

We can automatically flag those with min/max values.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Show allowed option values for string matching

Helpful to avoid looking in the documentation.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Pretty up command help output

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Reversed order of thinktime_blocks modulo operation

We wish to find the remainder of block count vs thinktime_blocks,
not the other way around. This makes thinktime_blocks work correctly.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fixup ratecycle option

It set .name twice, the last one was supposed to be .help instead.
Also kill options->opt_set, it's not used.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add default option values

It's handy for the help text, and it also cleans up init.c by removing
a bunch of default option defines.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Move timeout structure where it is used

Also makes sure that it's always initialized to 0's.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Complain if bad option given to 'cmdhelp'

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add per-option help

It's hard to remember all the options that fio supports, so add a
'cmdhelp' option that either prints all possible options (if argument
is 'all'), or specific option help (if argument is the command).

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fio version 1.11

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix busy-looping with aio engine and depth > 1

It's ok for io_getevents() to return 0, if min_nr is 0. Check
first if io_getevents() returns >= min_nr, if so then just break
and return.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add extra disk util name check

If it's the same device name, don't print utilization more than once.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Debug printf

Kill it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Unlink and directory fix

Make 'unlink' a bool and fix bug in directory setup.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Non-overwrite random write fix

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Make note of mailing list address

Yay, there's a list.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Make sure the fio_options[] members are always zero filled

Make sure the remaining members are cleared, to avoid confusion.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add thinktime_blocks

Right now 'thinktime' waits after every IO. This may not be what you
need, sometimes you want to wait only after 'x' number of ios. This
options allows you to set the number of ios to complete, before waiting.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix deviation overflow

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix division by zero in deviance calculation

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Split CFLAGS into OPTFLAGS + CFLAGS

Makes it easier to control rpm generation.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Don't create libdir

Since we link in io engines, we don't need to create the libdir anymore.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Don't dump optarg on unknown option

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Version 1.10

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Allow mem=mmap to also use a file backing

We have the stuff in place for huge page backed memory, so it's little
extra code to support io buffers inside mmap file backed memory.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Warning on hugepage-size setting in HOWTO

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Add hugepage-size option

We default to 4MiB, but export the setting so it can be changed.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Document how to setup/use huge pages

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Remove debug fprintfs

Oops, leftover from the mmaphuge commit.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Support for mmap of hugetlb files as memory backing

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix character + O_DIRECT

Just clear ->odirect if it's a character device, it doesn't apply.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix hugetlb problems

Alignment was bad, and we need to check the shmhuge string before shm,
otherwise it'll match the latter.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Must use ->real_file_size

->file_size is the size of the file from offset, while ->real_file_size
is the total size of the file. So when comparing EOF, we should look
at ->real_file_size against the offset.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Support for hugetlb backed shared memory

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Only unlink for created files

If the file was already there, don't unlink it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Make unlink and overwrite the default

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix FreeBSD/Solaris linkage

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Fix warnings from icc

icc spewed a bunch of warnings on building fio, but it did actually build
and work. Some of them are real bugs, most are just "helpful" warnings.

icc doesn't like pointer arithmetic, however these are not fixed up. It
works as-is, just ignore those class of warnings.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] rusage stat

Sum up total over all loops, not for each one.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

[PATCH] Improve time accounting for a job

We need to account loop time in usecs, or the null io engine can
get the timing wrong. For the system usage stats, don't use summed
up loop counts, but calculate end-start time instead.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>