Rather than delete percentage complete, show it. Because we were hiding
percentage complete we were not aware of inaccuracies in the percentage
calculation that are introduced when you toggle blanking and
verification on various methods.
I could not find the original specification of this standard, although
many proprietary implementations exist (or claim to exist). Also there
are many 3rd-party descriptions of the method, which often differ in one
or another detail, but after some reading I could conclude that the most
likely definition should be as follows:
1. Fill the device with zeroes, do not verify.
2. Fill the device with ones, do not verify.
3. Fill the device with a PRNG stream, do verify.
In other words, this is a simplified version of nwipe's "dodshort".
Both display_help() and nwipe_options_parse() are consistent about the
--noblank option, in that it exists and has no short form. Make the man
page and nwipe_gui_noblank() indicate that correctly. Mention that "ops2"
never blanks the device regardless of the option.
Change code so that there is only one place where the log is flushed on
exit from the ncurses gui. Prior to this patch the flush was done twice
which resulted in duplicated lines and missing lines. The
log_current_element variable was being initialised after the first flush.
This caused subsequent writes to nwipe to write log_lines pointers passed
the end of the pointer array causing memory corruption and resulting in the
occassional segfault on attempts to print the log array.
During testing a single drive two cores consistently run at close to 100%,
one core at 100% and the other varying between 80-100%. This shouldn't be
the case. Although there would be two threads running, i.e. the gui status
thread and the disk drives wipe thread I would have expected the amount of
CPU required to calculate the on screen stats to be insignificant as they
are only updated once a second.
To reduce the gui stats thread to near zero% while still maintaining the
stats update at one second I inserted an nanosleep interval of 100ms. This
now means that only the one core now shows at 80-100% while all other cores
are close to zero.
Update the footers for the commands available in each
screen. Also added the possibility to cancel from the
rounds screen. It was possible to cancel with backspace
from the other screens (method prng etc) but this was not
displayed in the footers. Added the `Ctrl-C` command
to all the footers since this is a global command.
Added the `Left` command to the rounds screen, since it
was implemented but not documented.
Also made sure the Ctrl-C was placed at the end on every
footer.
Symptoms: If you control-C to exit nwipe at the end of a successful
wipe you would get a segmentation fault. If nwipe log data was being
sent to stdout rather than to a log file, then you would not see the
log in the terminal windows after nwipe had exited. This patch fixes
those problems by creating a tri-state wipe flag for each wipe
thread. The main thread after having launched the wipe threads
will wait for ALL wipe flags to report that the wipe routine has
completed and the thread has exited. Only at that time does the
main() routine then proceed with joining the threads and waiting
for the join to confirm the thread has indeed exited. This join
is important as the thread will not actually exit until the OS
has flushed the disk buffers. Currently nwipe does not use O_SYNC
where data is sent straight to disk. Therefore it's important
to wait for data to be flushed before exiting nwipe. Part of this
patch is introducing a "please wait .. disks are being flushed"
prior to exiting nwipe otherwise it might look like nwipe had
hung. A disk flush in terms of how long it takes, can be from
instantly to 30 seconds or more, depending on how much system
memory you have, when the last sync occurred and if you have
changed the sync option from it's default. Something else to
note is that when all wipes have finished nwipe displays "Enter
To Exit" on the status line at the bottom of the screen. I
believe typing 'enter' rather than control-c did not produce a
segmentation fault. Irrespective, both methods of exiting nwipe
have been tested and confirmed to now work correctly. Tested
overnight with two drives using multiple wipe methods,
verification and blanking.
Symptoms: If you control-C to exit nwipe at the end of a successful
wipe you would get a segmentation fault. If nwipe log data was being
sent to stdout rather than to a log file, then you would not see the
log in the terminal windows after nwipe had exited. This patch fixes
those problems by creating a tri-state wipe flag for each wipe
thread. The main thread after having launched the wipe threads
will wait for ALL wipe flags to report that the wipe routine has
completed and the thread has exited. Only at that time does the
main() routine then proceed with joining the threads and waiting
for the join to confirm the thread has indeed exited. This join
is important as the thread will not actually exit until the OS
has flushed the disk buffers. Currently nwipe does not use O_SYNC
where data is sent straight to disk. Therefore it's important
to wait for data to be flushed before exiting nwipe. Part of this
patch is introducing a "please wait .. disks are being flushed"
prior to exiting nwipe otherwise it might look like nwipe had
hung. A disk flush in terms of how long it takes, can be from
instantly to 30 seconds or more, depending on how much system
memory you have, when the last sync occurred and if you have
changed the sync option from it's default. Something else to
note is that when all wipes have finished nwipe displays "Enter
To Exit" on the status line at the bottom of the screen. I
believe typing 'enter' rather than control-c did not produce a
segmentation fault. Irrespective, both methods of exiting nwipe
have been tested and confirmed to now work correctly. Tested
overnight with two drives using multiple wipe methods,
verification and blanking.
