Replace getopt() and usleep() dependencies with portable C++ parsing and sleeps. Refuse to execute the GPIO loopback diagnostic off Linux/Raspberry Pi hardware, and omit the hardware-only target from non-Linux Meson builds.
READ TOC receives its allocation length from the initiator’s CDB and accepts values up to 65535 bytes. The controller’s transfer buffer is normally only 4096 bytes, so using that value directly could overrun the buffer and report more data than can be transferred.
Clamp the response length to the available buffer size and clear only that bounded range before constructing the response. Also bound the start-track scan before dereferencing the track vector, allowing a lead-out (0xAA) request to safely produce its descriptor when no matching track exists.
Add regression tests for the maximum allocation length and for READ TOC lead-out handling.
Resolve symlinked default image folders with canonical() so relative link targets are evaluated from their link location. Replace the tautological default-folder assertion with coverage for missing folders, relative paths, and relative symlink targets.
Reject MODE SELECT Format Device pages whose length differs from the 24-byte page returned by MODE SENSE. Use a complete valid page in the truncation regression and cover both invalid page lengths and transfer lengths exceeding the buffer.
Forward command-line arguments to GoogleTest and mark the Meson target as gtest protocol so filtering, seeds, and individual results are available. Harden the service tests against unavailable socket capabilities and environment-dependent privileged ports, and skip the Linux-only interface assertion elsewhere.
Validate that the declared MODE SELECT parameter list fits in the transfer buffer and contains a complete command header. Validate the block-descriptor list against the remaining parameter-list length before skipping it.
Reject truncated mode-page headers and page bodies before reading them. Report ILLEGAL REQUEST / PARAMETER LIST LENGTH ERROR for truncated headers, block descriptors, and mode pages, as required by SCSI-2 8.2.8; retain INVALID FIELD IN PARAMETER LIST for complete but invalid fields.
Update the hard-disk and magneto-optical fixtures to send complete Format Device pages with their actual MODE SELECT parameter-list lengths. Add regression coverage for truncated headers, descriptors, page headers, and page bodies.
Introduce a server-rendered Go control panel for managing PiSCSI devices, disk images, saved configurations, archives, uploads, system settings, embedded themes and translations, PWA assets, a mock daemon, installation and systemd support, generated protobuf bindings, tests, and CI builds for supported Raspberry Pi targets.
This is built for feature parity with the Python Web UI, with a few notable differences: piscsi-web runs as a non-privileged user, there is no user authentication, the upload form uses a plain stream with no Javascript, the ability to set 'pretty hostname' is removed, and reserved ID memos are removed too.
Update piscsi core to use /var/lib/piscsi/images as its default and resolve relative image folders from the server working directory. Allow valid folders outside the user's home directory, and align the tests, protocol comments, service example, and man pages with the new behavior.
Ship a script and makefile target for migrating user data from old to new file system locations.
I'm currently debugging why NetBSD cannot access a simulated
DEC RRD42 drive. The log shows this command is received:
Mar 6 09:31:55 piscsi PISCSI[479]: [2026-03-06 09:31:55.710] [trace] (ID 5) - Received unknown command: $0Q
This is actually 0x51 (= ASCII 'Q')
I'm using `scsidump` to dump old SCSI drives and from time to time, I run into
errors. With this patch, I can resume a transfer (and will only lose 64 KB.)
This reverts a previous change where configurations for the Eclipse IDE specifically was introduced.
I think it's more important to have an IDE agnostic formatting framework, which clang-format provides
Example command for formatting all C++ source code, as run from within the cpp dir:
find . -iname '*.h' -o -iname '*.cpp' | xargs clang-format -i
This will reduce the recurring maintenance overhead of keeping the same information up to date in multiple places
Only the most common use cases are covered in the usage help text now
In order for the Eject command to properly release the image and reset the status,
we need to override the Eject function with the appropriate logic
The other removable disk drive devices inherit the Disk class,
and hence do not need this
Mode Sense (6) supports a 32-bit number for the maximum block count, where Mode Sense (10) supports a 64-bit number for the count. The Seagate SCSI Commands Reference Manual from 2016, has this to say about a "Short LBA mode parameter block descriptor" as returned by Mode Sense (6):
> On a MODE SENSE command, if the number of logical blocks on the
> medium exceeds the maximum value that is able to be specified in
> the NUMBER OF LOGICAL BLOCKS field, the device server shall return
> a value of FFFFFFFh.
Similarly, the Read Capacity (10) command should return FFFFFFFFh if the capacity of the drive exceeds 2^32 sectors, and that a driver should then know to use Read Capacity (16) if it supports that command.
There may be an unexpected side-effect if presenting a drive of more than 2^32 sectors to a legacy host that does not support devices that large. The Read Capacity commands specify the value returned as the last addressable sector on the device. This means that typically, an application which uses the value returned by that command is going to add 1 to it to get the actual number of blocks on a given device. If the program is not aware of Read Capacity (16), and is not using greater than 32-bit math, then it might report to the user that the total number of sectors on the drive as 0. I don't view this as a huge problem, however. In that case, the legacy driver wouldn't correctly support the capacity of the drive anyway, so providing that driver with a device that is 2 ^ 32 sectors or larger wouldn't make sense from the user perspective.
The scsictl command parser looks for single letter shorthands for each device type;
this adds 't' for SCTP and changes SCLP to 'p'
Now you can do '-t tape' and '-t printer' for an easy to remember command
Also moves the SCTP default product/vendor into the device factory,
updates the unit tests, and an overhauls the man pages
For the Status, DataIn, and DataOut phases, piscsi has a 50 microsecond
sleep implemented, which has proven a stabilizing factor for the bus
on most SCSI initiators
However, Akai S1100 and other samplers of a similar vintage need a longer
sleep time for reliable operation, so a -s option for piscsi has been
introduced to set the value to an artbitrary number at runtime
The X68000 host bridge and associated host file system module
has not been functional for many years, as reported by multiple users
Removing the code will reduce complexity for users -
the SCBR device has caused some misunderstandings in the past -
while eliminating code that has been flagged as insecure by SonarQube
* Fix VCD timestamps.
VCD simulation timestamps are integers.
eb71c31cf1 changed them to doubles which
causes the files to be unreadable.
* Fix VCD byte values
no spaces allowed between b and the bits. Value will be left-extended
with zeros.
* Revert "Don't ResizeCache on sector change if no filename is defined (#1438)"
This reverts commit dd9a3296d4.
* Revert "Add ModeSense page 0x25 (DEC special function control page) (#1412)"
This reverts commit 1121b8d9d6.
* Revert "DiskCache needs a size"
This reverts commit 7cc8df271c.
* Revert "Honor sector size change via ModeSelect6 in scsicd (#1406)"
This reverts commit b7f65d33e2.
* Revert "Multiple fixes for ModeSelect (#1405)"
This reverts commit ad5eae93e7.
* Add tests for issue#1426
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* add test without medium
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Don't resize DiskCache without a file
DiskCache computes its size from a file.
Skip ResizeCache Without file (i.e. 'empty' CD)
Fixes#1426
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Add ModeSense page 0x25 (DECSpecialFunctionControlPage)
VAXServer 3100 (CPU KA41-E) console firmware issues ModeSense(6)
page 0x25 when probing for disks. A disk won't be recognized if
this returns an error.
The DEC SCSI specification[1], section 8.5 documents this page.
[1] https://manx-docs.org/collections/antonio/dec/dec-scsi.pdfFixes#1410
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Make ModeSelect() non-const
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Implement ModeSelect for scsicd
and honor sector size setting.
DEC's VMS can't handle 2k sector sizes and uses ModeSelect6 to set the
sector size to 512 bytes.
Fixes#1397
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Test sector size setting via ModeSelect in SCSICD
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Re-calculate total blocks when sector size changes
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Reset CD data tracks after sector size change
The track calculation is based on sector size and must be reset after
change of sector size.
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* resize cache after change of sector size
The disk cache is based on sector size and must be resized when the
sector size changes.
Disk::ResizeCache needs a `raw` parameter, make this value accessible
from the current cache.
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Make GetRawMode const
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
---------
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Allow 'empty' ModeSelect6
tl;dr
Treat a computed length of 0 as `has_valid_page_code`.
Details:
The SRM console (aka 'BIOS') of DEC Alpha sends an empty
ModeSelect6 with the following data:
~~~
ModeSelect6, CDB $151000000c00
~~~
That makes 12 byte(s) as follows
~~~
0 1 2 3 4 5 6 7 8 9 10 11
00 00 00 08 00 00 00 00 00 00 02 00
~~~
decoding it (accoring to [1], Section 8.3.3, Table 94) gives us
Mode Data Length 0
Medium Type 0
Device-specific 0
Block desc len 8
Density Code 0
Number of blks 0
Reserved 0
Block length 512
`scsi_command_util::ModeSelect` computes
~~~
offset = 4 + buf[3];
~~~
giving 12 and
~~~
length -= offset;
~~~
giving 0.
Thus it never enters the `while` loop and `has_valid_page_code` stays
`false`, raising an error.
[1] [Small Computer System Interface - 2 rev 10L.pdf](https://dn790004.ca.archive.org/0/items/SCSISpecificationDocumentsSCSIDocuments/Small%20Computer%20System%20Interface%20-%202%20rev%2010L.pdf)
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Allow ModeSelect with page code 1
OpenVMS Alpha (the operating system, not the SRM BIOS) uses
ModeSelect6 with a page code of 1.
The semantics are unknown, just accepting it works for me.
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Fix page length computation in ModeSelect
tl;dr
The 'skip to next ModeSelect page' computation was off-by-one, either not
taking the page code itself into account or missing the fact that the
page length is given as `n - 1`.
Fix:
Add 1 to the computed length.
Details:
OpenVMS Alpha sends a ModeSelect6 as follows
~~~
command:
ModeSelect6, CDB $151000001900
payload:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
00 00 00 08 00 00 00 00 00 00 02 00 01 0a 24 00 00 00 00 00 00 00 00 00 00
~~~
This translates to (accoring to [1], Section 8.3.3)
~~~
Mode Data Length 0
Medium Type 0
Device-specific 0
Block desc len 8
~~~
with the following offset / length computation _before_ the `while` loop
~~~
offset = 12
length = 13
~~~
The first payload section is
~~~
4 5 6 7 8 9 10 11
00 00 00 00 00 00 02 00
~~~
translating to
~~~
Density Code 0
Number of blks 0
Reserved 0
Block length 0x200 512
~~~
Then follows a pagecode 1 as
~~~
12 13 14 15 16 17 18 19 20 21 22 23 24
01 0a 24 00 00 00 00 00 00 00 00 00 00
~~~
translating to
~~~~
Page code 1
Page length -1 10
Mode parameters 24 00 00 00 00 00 00 00 00 00 00
~~~
computing (inside the `while` loop, as `// Advance to the next page`)
~~~
size = 10 + 2 = 12
~~~
followed by new `offset` and `length` values
~~~
offset = 25
length = 1
~~~
So it stays in the `while` loop (and has a larger-than-buffer `offset`
value)
Signed-off-by: Klaus Kämpf <kkaempf@gmail.com>
* Fix missing logging for a LUN when the LUN is explicitly specified
* Do not suppress controller messages when LUN is specified
* Fix misleading logging for DaynaPort
Daniel Markstedt
Eric Helgeson
Jan-Benedict Glaw
Chris Hooper
BogDan Vatra
Rick Altherr
Daniel Markstedt
Klaus Kämpf
Uwe Seimet
Klaus Kämpf
Uwe Seimet