Like all lab assignments in this class, you are prohibited from copying any content from the Internet including (discord or other group messaging apps) or discussing,sharing ideas, code, configuration, text, or anything else or getting help from anyone in or outside of the class. Consulting online sources is acceptable, but under no cir- cumstances should anything be copied. Failure to abide by this requirement will result in penalty as described in our course syllabus.

Note: Before proceeding to lab 3, ensure that you have completed the read functionality from lab 2.  You will receive credit for completing the last two test cases for read in lab 3 as well.

Your internship is going great.  You have gained experience with C programming, you have experienced your first segmentation faults, and you’ve come out on top.  You are brimming with confidence and ready to handle your next challenge.

Implementing mdadm_write

Your next job is to implement write functionality formdadm and then thoroughly test your implementation. Specifically, you will implement the following function:

int  mdadm_write(uint32_t  addr,  uint32_t  len,  const  uint8_t  *buf)

As you can tell, it has an interface that is similar to that of themdadm_read function, which you have already implemented. Specifically, it writes len bytes from the user-supplied buf buffer to your storage system, start- ing at address addr. You may notice that the buf parameter now has a const specifier. We put the const there to emphasize that it is an in parameter; that is, mdadm_write should only read from this parameter and not modify it. It is a good practice to specify const specifier for your in parameters that are arrays or structs.

Similar to mdadm_read, writing to an out-of-bound linear address should fail.  A read larger than 1,024 bytes should fail; in other words, len can be 1,024 at most. There are a few more restrictions that you will find out as you try to pass the tests.

Once you implement the above function, you have the basic functionality of your storage system in place. We have expanded the tester to include new tests for the write operations, in addition to existing read operations. You should try to pass these write tests first.

Testing using trace replay

As we discussed before, your mdadm implementation is a layer right above JBOD, and the purpose of mdadm is to unify multiple small disks under a unified storage system with a single address space.  An application built on top of mdadm will issue a mdadm_mount and then a series of mdadm_write and mdadm_read commands to implement the required functionality, and eventually, it will issue mdadm_unmount command. Those read/write commands can be issued at arbitrary addresses with arbitrary payloads and our small number of tests may have missed corner cases that may arise in practice.

Therefore, in addition to the unit tests, we have introduces trace files, which contain the list of commands that a system built on top of your mdadm implementation can issue. We have also added to the tester a function-

ality to replay the trace files. Now the tester has two modes of operation. If you run it without any arguments, it will run the unit tests:

$   ./tester

/ ********************************************************************/ Please  ensure  there  were  no  warnings  when  you  compiled  your  program

If  you  have  any  make  errors  or  warnings  then  you  will  receive  a  straight  0 .

Please  ensure  there  are  sufficent  comments  in  your  code

Do  not  foget  to  add  your  name  and  date  in  comments  on  top  of  your  mdadm.c / ********************************************************************/

Testing  your  code  now:running  test_mount_unmount:  passed

Total  score:  10/10

If you run it with -w  pathname arguments, it expects the pathname to point to a trace file that contains the list of commands. In your repository, there are three trace files under the traces directory: simple-input, linear-input, random-input.  Let’s look at the contents of one of them using the head command, which shows the first 10 lines of its argument:

$  head  traces/simple-input

MOUNT

WRITE  0  256  0

READ  1006848  256  0

WRITE  1006848  256  93

WRITE  1007104  256  94

WRITE  1007360  256  95

READ  559872  256  0

WRITE  559872  256  139

READ  827904  256  0

WRITE  827904  256  162

The first command mounts the storage system.  The second command is a write command, and the argu- ments are similar to the actual mdadm_write function arguments; that is, write at address 0, 256 bytes of bytes with contents of 0.  The third command reads 256 bytes from address  1006848 (the third argument to READ is ignored). And so on.

You can replay them on your implementation using the tester as follows:

$  ./tester  -w  traces/simple-input

If one of the commands fails, for example because the address is out of bounds, then the tester aborts with an error message saying on which line the error happened.  If the tester can successfully replay the trace until the end, it takes the cryptographic checksum of every block of every disk and prints them out on the screen, as above. Now you can use this information to tell if the final state of your disks is consistent with the final state of the reference implementation, if the above trace was replayed on a reference implementation.  You can do that by comparing your output to that of the reference implementation. The files that contain the corresponding cryptographic checksums from reference implementation are also under traces directory and they end with - expected-output. For example, here’s how you can test ifyour implementation’s trace output matches with that of reference implementation’s output for the simple-input trace:

$  ./tester  -w  traces/simple-input  >my-output

$  diff  -u  my-output  traces/simple-expected-output

The first line replays the trace file and redirects the output to my-output file in the current directory, while the second line runs the diff tool, which compares to files contents – when the files are identical, no output is displayed, which means your implementation’s final state after the commands in the trace file matches the reference implementation’s state. No output for diff is considered passed. Just running a trace file with tester is not considered pass.

If there is a difference in the diff command output, then there is a bug in your implementation; you can see which blocks contents differ and that may help you with debugging.For example you might see something like:

-SIG(disk,block)  5  239   :  0x8d  0x95  0xbf  0x5c  0xae  0xec  0xe0  0x0d  0x6a  0xa6 +SIG(disk,block)  5  239   :  0x42  0xae  0xb0  0xe5  0x68  0x15  0x81  0x7e  0x20  0x15

This means that signature of disk 5 block 239 does not match with the reference implementation.  Your code did not write the right content on disk 5, block 239. 0x8... part is the signature of the content of the block and does not tell you the exact content of the block.

Based on diff’s output you know what disk and block numbers were not written correctly.  This likely means there was an issue in your logic or you missed implementing a particular edge case.

Debugging tips:

•  Without gdb: Based on diff’s output you know what disk and block numbers were not written correctly. This likely means there was an issue in your logic or you missed implementing a particular edge case. You can find out the exact address that is causing the issue by strategically using printf statements to printout a) The addrandlen you are dealing with (ideally at being of madm write function), b) the length you are using to copy into buffer every time you use memcpy or before you write using jbod write operation c) print out disk number and block number before your write to ensure you are writing at the right disk and right block.

•  With gdb: You can run the tester filesingdb. You can do it with the following commands:

gdb   ./tester

(gdb)  set  args  -w  traces/simple-input

(gdb)  r