In this project, we'll finish the implementation of a web server in C.
What you need to write:
-
HTTP request parser
-
HTTP response builder
-
LRU cache
- Doubly linked list (some functionality provided)
- Use existing hashtable functionality (below)
-
Your code will interface with the existing code. Understanding the existing code is an expected part of this challenge.
What's already there:
- Low-level networking code
- Code that determines a MIME type from a file extension
- File-reading code
- A hashtable implementation
- A linked list implementation (used solely by the hashable--you don't need it)
- All system calls, including
send()
andrecv()
A web server is a piece of software that accepts HTTP requests (e.g. GET requests for HTML pages), and returns responses (e.g. HTML pages). Other common uses are GET requests for getting data from RESTful API endpoints, images within web pages, and POST requests to upload data to the server (e.g. a form submission or file upload).
We will write a simple web server that returns files and some specialized data on a certain endpoint.
-
http://localhost:3490/d20
should return a random number between 1 and 20 inclusive astext/plain
data. -
Any other URL should map to the
serverroot
directory and files that lie within. For example:http://localhost:3490/index.html
should serve the file
./serverroot/index.html
Examine the skeleton source code in server.c
and cache.c
for which pieces
you'll need to implement.
IMPORTANT Spend some time inventorying the code to see what is where. Write down notes. Write an outline. Note which functions call which other functions. Time spent up front doing this will reduce overall time spent down the road.
The existing code is all one big hint on how to attack the problem.
For the portions that are already written, study the moderately-well-commented code to see how it works.
There is a Makefile
provided. On the command line, type make
to build the
server.
Type ./server
to run the server.
Read through all the main and stretch goals before writing any code to get an overall view, then come back to goal #1 and dig in.
-
Examine
handle_http_request()
in the fileserver.c
.You'll want to parse the first line of the HTTP request header to see if this is a
GET
orPOST
request, and to see what the path is. You'll use this information to decide which handler function to call.The variable
request
inhandle_http_request()
holds the entire HTTP request once therecv()
call returns.Read the three components from the first line of the HTTP header. Hint:
sscanf()
.Right after that, call the appropriate handler based on the request type (
GET
,POST
) and the path (/d20
or other file path.) You can start by just checking for/d20
and then add arbitrary files later.Hint:
strcmp()
for matching the request method and path. Another hint:strcmp()
returns0
if the strings are the same!Note: you can't
switch()
on strings in C since it will compare the string pointer values instead of the string contents. You have to use anif
-else
block withstrcmp()
to get the job done.If you can't find an appropriate handler, call
resp_404()
instead to give them a "404 Not Found" response. -
Implement the
get_d20()
handler. This will callsend_response()
.See above at the beginning of the assignment for what
get_d20()
should pass tosend_response()
.If you need a hint as to what the
send_response()
call should look like, check out the usage of it inresp_404()
, just above there.Note that unlike the other responses that send back file contents, the
d20
endpoint will simply compute a random number and send it back. It does not read the number from a file.The
fd
variable that is passed widely around to all the functions holds a file descriptor. It's just a number use to represent an open communications path. Usually they point to regular files on disk, but in this case it points to an open socket network connection. All of the code to create and usefd
has been written already, but we still need to pass it around to the points it is used. -
Implement
send_response()
.This needs to build a complete HTTP response with the given parameters. It should write the response to the string in the
response
variable.The total length of the header and body should be stored in the
response_length
variable so that thesend()
call knows how many bytes to send out over the wire.See the HTTP section above for an example of an HTTP response and use that to build your own.
Hint:
sprintf()
for creating the HTTP response.strlen()
for computing content length.sprintf()
also returns the total number of bytes in the result string, which might be helpful.The HTTP
Content-Length
header only includes the length of the body, not the header. But theresponse_length
variable used bysend()
is the total length of both header and body. -
Implement arbitrary file serving.
Any other URL should map to the
serverroot
directory and files that lie within. For example:http://localhost:3490/index.html
serves file./serverroot/index.html
.http://localhost:3490/foo/bar/baz.html
serves file./serverroot/foo/bar/baz.html
.You might make use of the functionality in
file.c
to make this happen.You also need to set the
Content-Type
header depending on what data is in the file.mime.c
has useful functionality for this.
Implement an LRU cache. This will be used to cache files in RAM so you don't have to load them through the OS.
When a file is requested, the cache should be checked to see if it is there. If so, the file is served from the cache. If not, the file is loaded from disk, served, and saved to the cache.
The cache has a maximum number of entries. If it has more entries than the max, the least-recently used entries are discarded.
The cache consists of a doubly-linked list and a hash table.
The hashtable code is already written and can be found in hashtable.c
.
-
Implement
cache_put()
incache.c
.Algorithm:
- Allocate a new cache entry with the passed parameters.
- Insert the entry at the head of the doubly-linked list.
- Store the entry in the hashtable as well, indexed by the entry's
path
. - Increment the current size of the cache.
- If the cache size is greater than the max size:
- Remove the entry from the hashtable, using the entry's
path
and thehashtable_delete
function. - Remove the cache entry at the tail of the linked list.
- Free the cache entry.
- Ensure the size counter for the number of entries in the cache is correct.
- Remove the entry from the hashtable, using the entry's
-
Implement
cache_get()
incache.c
.Algorithm:
- Attempt to find the cache entry pointer by
path
in the hash table. - If not found, return
NULL
. - Move the cache entry to the head of the doubly-linked list.
- Return the cache entry pointer.
- Attempt to find the cache entry pointer by
-
Add caching functionality to
server.c
.When a file is requested, first check to see if the path to the file is in the cache (use the file path as the key).
If it's there, serve it back.
If it's not there:
- Load the file from disk (see
file.c
) - Store it in the cache
- Serve it
- Load the file from disk (see
There's a set of unit tests included to ensure that your cache implementation is functioning correctly. From the src
directory, run make tests
in order to run the unit tests against your implementation.
-
Implement
find_start_of_body()
to locate the start of the HTTP request body (just after the header). -
Implement the
post_save()
handler. Modify the main loop to pass the body into it. Have this handler write the file to disk. Hint:open()
,write()
,close()
.fopen()
,fwrite()
, andfclose()
variants can also be used, but the former three functions will be slightly more straightforward to use in this case.The response from
post_save()
should be of typeapplication/json
and should be{"status":"ok"}
.
We know that if the user hits http://localhost:3490/index.html
it should
return the file at ./serverroot/index.html
.
Make it so that if the user hits http://localhost:3490/
(which is endpoint
/
, on disk ./serverroot/
), if no file is found there, try adding an
index.html
to the end of the path and trying again.
So http://localhost:3490/
would first try:
./serverroot/
fail to find a file there, then try:
./serverroot/index.html
and succeed.
It doesn't make sense to cache things forever--what if the file changes on disk?
Add a created_at
timestamp to cache entries.
If an item is found in the cache, check to see if it is more than 1 minute old. If it is, delete it from the cache, then load the new one from disk as if it weren't found.
You'll have to add a cache_delete
function to your cache code that does the work of actually removing entries that are too old from the cache.
Difficulty: Pretty Dang Tough
Research the pthreads library.
When a new connection comes in, launch a thread to handle it.
Be sure to lock the cache when a thread accesses it so the threads don't step on each other's toes and corrupt the cache.
Also have thread cleanup handlers to handle threads that have died.