Cyrus IMAP Server: Hacking¶

This file aims to be a guide to Cyrus coding style/conventions/useful utilities for people trying to approach the code in a sane way. It's not well organized right now but hopefully that will improve with time ;)

Warning

This document is woefully out of date. While some parts of it are still accurate, it has not been reviewed in quite a while. If you're looking at the code and it doesn't seem to match this, don't assume the code is wrong. Talk to the developers! We'll revise this document, but don't hold your breath right now. (This comment written August 5, 2021.)

Memory Allocation¶

All cyrus memory allocation should be done through the libcyrus functions. These are all written to correctly call fatal() in the event of an out-of-memory condition.
In addition to xmalloc and xrealloc, we provide replacements for strdup, strndup, and a malloc that will guarantee zeroed block of memory (xzmalloc).
If you are going to need to do a large number of small allocations, and then free them all at once, you should look at the memory pool routines, which are much faster, but will leak memory until you free the entire pool at once.

strlcpy vs strncpy vs memcpy¶

use strlcpy when you know the size of the buffer, e.g.:

char buf[50];
strlcpy(buf, src, sizeof(buf));

use memcpy to truncate a string into a buffer you know is large enough. Note that when you do this the resulting buffer will NOT BE NULL TERMINATED:

memcpy(buf, src, 4);
buf[5] = '\0'

you should try to avoid strncpy, since it is much slower than memcpy (it zero-fills the rest of the buffer) and isn't as safe as strlcpy.
Use of the functions in this way will reduce the confusion involved in their various behaviors. In other words, this avoids things that look like:

strncpy(buf, src, sizeof(buf)-1);

map_refresh and map_free¶

In many cases, it is far more effective to read a file via the operating system's mmap facility than it is to via the traditional read() and lseek system calls. To this end, Cyrus provides an operating system independent wrapper around the mmap() services (or lack thereof) of the operating system.
Cyrus currently only supports read-only memory maps, all writes back to a file need to be done via the more traditional facilities. This is to enable very low-performance support for operating systems which do not provide an mmap() facility via a fake userspace mmap.
To create a map, simply call map_refresh on the map (details are in lib/map.h). To free it, call map_free on the same map.
Despite the fact that the maps are read-only, it is often useful to open the file descriptors O_RDWR, especially if the file descriptors could possibly be used for writing elsewhere in the code. Some operating systems REQUIRE file descriptors that are mmap()-ed to be opened O_RDWR, so just do it.

Network Functions¶

Cyrus abstracts socket stream access to a concept we refer to as "prot streams" Prot Streams take care of all of the necessary SASL and TLS/SSL encryption that may need to happen before data goes out/comes in from the network. The API is documented in lib/prot.h

(todo) Authorization Modules¶

General Hints¶

Some general hints that all made it into my 11/15 16:47 commit that I think may be generally useful to people hacking on the cyrus source:

Command line apps should link cli_fatal.o so they all fatal() in the same way, unless there is a really good reason they need to do something unique.
If you call cyrus_init() you must call cyrus_done() before you exit.
No one should ever call DB->init() or DB->done() cyrusdb functions except for in libcyrus_init().
I've been trying to keep #include statements for libcyrus and libimap alphabetical, and below any system includes, but this is merely my personal style
Don't exit at the bottom of main with exit(x) use return instead.
For all the command line utilities that need to be sure that they are running as the cyrus user, it should be the first thing they do, and they should exit with an appropriate fatal() call.
All services should have a shut_down call. It should be the ONLY way of exiting the application. fatal() should always make an attempt to call shut_down() if it can (though it should have a recursive fatal() trap just in case). Similarly, command line utilities probably don't need a shut_down().

Coding Standards¶

These are the generally agreed upon coding standards as thrashed out on the cyrus-devel list in June 2010.

Spacing is 4 characters with soft tabs at 8 - mixed tabs and spaces This corresponds to the vi settings sw=4 sts=4 ts=8 etc.
Group the '*' character with the variable not the type, i.e.

char *foo;  /* correct */
char* foo;  /* WRONG */

The keywords 'if', 'for', and 'while' take a space after the keyword. The parentheses around the following expression are not closely connected to the expression without any spaces. The ';' inside the 'for' expression have a space after them and not before. For example:

if (condition)              /* correct */
if(condition)               /* WRONG */
if( condition )             /* WRONG */

for (i = 0; i < x; i++)     /* correct */
for(i = 0; i < x; i++)      /* WRONG */
for (i = 0 ;i < x ;i++)     /* WRONG */
for(i = 0 ;i < x ;i++)      /* WRONG */

while (foo)                 /* correct */
while(foo)                  /* WRONG */
while( foo )                /* WRONG */

Use spaces around the double-character logical operator '||' but don't use spaces around single-character bitwise operator '|'.

int flags = FOO|BAR;        /* correct */
int flags = FOO | BAR;      /* WRONG */
if (itchy || scratchy)      /* correct */
if (itchy||scratchy)        /* WRONG */

Function definitions are followed by a brace on a line by itself, all other braces are inline. Return types are inline with function definition. Old K&R style function definitions are not allowed.

void thing(int val)         /* correct */
{                           /* correct */
    ...body...
}                           /* correct */

void                        /* WRONG */
thing(int val)              /* WRONG */
{                           /* WRONG */
    ...body...
}                           /* WRONG */

void                        /* WRONG */
thing(                      /* WRONG */
    int val)                /* WRONG */
{                           /* WRONG */
    ...body...
}                           /* WRONG */

void                        /* WRONG */
antique(val)                /* WRONG */
    int val;                /* WRONG */
{                           /* WRONG */
    ...body...
}                           /* WRONG */

void thing(int val) {       /* WRONG */
    ...body...
}                           /* WRONG */

void noargs(void)           /* correct */
{                           /* correct */
    ...body...
}                           /* correct */

void noargs()               /* WRONG */
{                           /* WRONG */
    ...body...
}                           /* WRONG */

Long argument lists should be split across multiple lines, with the second and subsequent lines indented so that they line up with the start of the first line of arguments.

void toomanyargs(int arg1, const char *arg2,        /* correct */
                 struct whatever *arg3, int arg4)   /* correct */
{                                                   /* correct */
    ...body...
}                                                   /* correct */

void toomanyargs(int arg1, const char *arg2,        /* WRONG */
struct whatever *arg3, int arg4)                    /* WRONG */
{                                                   /* WRONG */
    ...body...
}                                                   /* WRONG */

void toomanyargs(                                   /* WRONG */
    int arg1,                                       /* WRONG */
    const char *arg2,                               /* WRONG */
    struct whatever *arg3,                          /* WRONG */
    int arg4)                                       /* WRONG */
{                                                   /* WRONG */
    ...body...
}                                                   /* WRONG */

Within a function, braces are used in old-fashioned K&R style. Specifically:
- open braces are placed at the end of the line containing the statement (such as an 'if') to which they belong, after a single space.
- closing braces are placed on a line by themselves, aligned with the start of the statement to which their matching open brace belongs.
- this applies even when the closing brace is followed by an 'else' keyword.

Yes, it's ugly and hard to read, but you get used to it and most of the code is currently like that. Deal with it.

while (cond) {              /* correct */
    ...body...              /* correct */
}                           /* correct */

while (cond){               /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */

while (cond) { ...body... } /* WRONG */

while (cond) {              /* WRONG */
    ...body... }            /* WRONG */

while (cond)                /* WRONG */
{                           /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */

while (cond)                /* WRONG */
  {                         /* WRONG */
    ...body...              /* WRONG */
  }                         /* WRONG */

if (cond) {                 /* correct */
    ...body...              /* correct */
}                           /* correct */
else if (othercond) {       /* correct */
    ...body...              /* correct */
}                           /* correct */
else {                      /* correct */
    ...body...              /* correct */
}                           /* correct */

if (cond) {                 /* WRONG */
    ...body...              /* WRONG */
} else if (othercond) {     /* WRONG */
    ...body...              /* WRONG */
} else {                    /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */

if (cond)                   /* WRONG */
{                           /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */
else if (othercond)         /* WRONG */
{                           /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */
else                        /* WRONG */
{                           /* WRONG */
    ...body...              /* WRONG */
}                           /* WRONG */

The braces around a block used in an 'if'...'else if'...'else' may be omitted if the statement is very simple and clear, such as a single function call. This is a judgement call though, so play it safe and use braces.

if (cond)                   /* correct, maybe */
    function();             /* correct, maybe */
else                        /* correct, maybe */
    other_function();       /* correct, maybe */

The 'goto' keyword needs to be used very very sparingly and only with forethought. The only clearly good example is to goto a label at the end of a function to do cleanup under error conditions.

void foo(struct bar *b)     /* correct */
{                           /* correct */
    char *x = xmalloc(...); /* correct */

    if (b == NULL)          /* correct */
        goto error;         /* correct */

    if (b->quux != 42)      /* correct */
        goto error;         /* correct */

    ...do useful things...  /* correct */
error:                      /* correct */
    free(x);                /* correct */
}                           /* correct */

Very occasionally, it may be permissable to use 'goto' from within a complicated or multiply-nested loop, to the top of a loop, but only if using another control structure is less clear.

Generally, zero return is SUCCESS and integer return is an error code.
Use "const char *" where possible.

int is_tacky(const char *name)              /* correct */
{                                           /* correct */
    return !strcmp(name, "britney");        /* correct */
}                                           /* correct */

int is_tacky(char *name)                    /* WRONG */
{                                           /* WRONG */
    return !strcmp(name, "britney");        /* WRONG */
}                                           /* WRONG */

Use "struct buf" for variable length strings where possible.
RAII http://en.wikipedia.org/wiki/Resource_Acquisition_Is_Initialization In practice, this means each structure should have a single cleanup function which handles all possible states of the structure and is called whenever the structure needs cleaning up. Likewise, any resources allocated during a function should be cleaned up in the same function, in a single code block at the end of the function (see the comments on 'goto').
If you find yourself passing the same multiple parameters through many functions, create a struct and pass around a pointer to that instead.
DON'T EVER REUSE THE SAME VARIABLE FOR TWO DIFFERENT PURPOSES IN THE SAME FUNCTION. IN FACT, DON'T REUSE THE SAME VARIABLE _NAME_ FOR DIFFERENT PURPOSES. KTHXBYE. (note: this doesn't apply to 'i', 'n', etc which are used in multiple loops. It applies to using the same name for an absolute offset and a "within this mmap" offset though, and it also applies to using the same variable name for native order and network order numbers, which is where I've seen it a few times and been super frustrated!)
Write RFCs in comments capitalized with space after the RFC, like RFC 1234, not like rfc 1234 or RFC1234.