libratbox/src/linebuf.c

/*
 *  ircd-ratbox: A slightly useful ircd.
 *  linebuf.c: Maintains linebuffers.
 *
 *  Copyright (C) 2001-2002 Adrian Chadd <adrian@creative.net.au>
 *  Copyright (C) 2002 Hybrid Development Team
 *  Copyright (C) 2002-2005 ircd-ratbox development team
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301
 *  USA
 *
 *  $Id: linebuf.c 26092 2008-09-19 15:13:52Z androsyn $
 */

#include <libratbox_config.h>
#include <ratbox_lib.h>
#include <commio-int.h>

#ifndef NOBALLOC
static rb_bh *rb_linebuf_heap;
#endif

static int bufline_count = 0;

#ifndef LINEBUF_HEAP_SIZE
#define LINEBUF_HEAP_SIZE 2048
#endif

/*
 * rb_linebuf_init
 *
 * Initialise the linebuf mechanism
 */

void
rb_linebuf_init(size_t heap_size)
{
	rb_linebuf_heap = rb_bh_create(sizeof(buf_line_t), heap_size, "librb_linebuf_heap");
}

static buf_line_t *
rb_linebuf_allocate(void)
{
	buf_line_t *t;
	t = rb_bh_alloc(rb_linebuf_heap);
	return (t);

}

static void
rb_linebuf_free(buf_line_t * p)
{
	rb_bh_free(rb_linebuf_heap, p);
}

/*
 * rb_linebuf_new_line
 *
 * Create a new line, and link it to the given linebuf.
 * It will be initially empty.
 */
static buf_line_t *
rb_linebuf_new_line(buf_head_t * bufhead)
{
	buf_line_t *bufline;
	rb_dlink_node *node;

	bufline = rb_linebuf_allocate();
	if(bufline == NULL)
		return NULL;
	++bufline_count;


	node = rb_make_rb_dlink_node();

	/* Stick it at the end of the buf list */
	rb_dlinkAddTail(bufline, node, &bufhead->list);
	bufline->refcount++;

	/* And finally, update the allocated size */
	bufhead->alloclen++;
	bufhead->numlines++;

	return bufline;
}


/*
 * rb_linebuf_done_line
 *
 * We've finished with the given line, so deallocate it
 */
static void
rb_linebuf_done_line(buf_head_t * bufhead, buf_line_t * bufline, rb_dlink_node *node)
{
	/* Remove it from the linked list */
	rb_dlinkDestroy(node, &bufhead->list);

	/* Update the allocated size */
	bufhead->alloclen--;
	bufhead->len -= bufline->len;
	lrb_assert(bufhead->len >= 0);
	bufhead->numlines--;

	bufline->refcount--;
	lrb_assert(bufline->refcount >= 0);

	if(bufline->refcount == 0)
	{
		/* and finally, deallocate the buf */
		--bufline_count;
		lrb_assert(bufline_count >= 0);
		rb_linebuf_free(bufline);
	}
}


/*
 * skip to end of line or the crlfs, return the number of bytes ..
 */
static inline int
rb_linebuf_skip_crlf(char *ch, int len)
{
	int orig_len = len;

	/* First, skip until the first non-CRLF */
	for(; len; len--, ch++)
	{
		if(*ch == '\r')
			break;
		else if(*ch == '\n')
			break;
	}

	/* Then, skip until the last CRLF */
	for(; len; len--, ch++)
	{
		if((*ch != '\r') && (*ch != '\n'))
			break;
	}
	lrb_assert(orig_len > len);
	return (orig_len - len);
}


/*
 * rb_linebuf_newbuf
 *
 * Initialise the new buffer
 */
void
rb_linebuf_newbuf(buf_head_t * bufhead)
{
	/* not much to do right now :) */
	memset(bufhead, 0, sizeof(buf_head_t));
}

/*
 * rb_linebuf_donebuf
 *
 * Flush all the lines associated with this buffer
 */
void
rb_linebuf_donebuf(buf_head_t * bufhead)
{
	while(bufhead->list.head != NULL)
	{
		rb_linebuf_done_line(bufhead, (buf_line_t *) bufhead->list.head->data,
				     bufhead->list.head);
	}
}

/*
 * rb_linebuf_copy_line
 *
 * Okay..this functions comments made absolutely no sense.
 *
 * Basically what we do is this.  Find the first chunk of text
 * and then scan for a CRLF.  If we didn't find it, but we didn't
 * overflow our buffer..we wait for some more data.
 * If we found a CRLF, we replace them with a \0 character.
 * If we overflowed, we copy the most our buffer can handle, terminate
 * it with a \0 and return.
 *
 * The return value is the amount of data we consumed.  This could
 * be different than the size of the linebuffer, as when we discard
 * the overflow, we don't want to process it again.
 *
 * This still sucks in my opinion, but it seems to work.
 *
 * -Aaron
 */
static int
rb_linebuf_copy_line(buf_head_t * bufhead, buf_line_t * bufline, char *data, int len)
{
	int cpylen = 0;		/* how many bytes we've copied */
	char *ch = data;	/* Pointer to where we are in the read data */
	char *bufch = bufline->buf + bufline->len;
	int clen = 0;		/* how many bytes we've processed,
				   and don't ever want to see again.. */

	/* If its full or terminated, ignore it */

	bufline->raw = 0;
	lrb_assert(bufline->len < BUF_DATA_SIZE);
	if(bufline->terminated == 1)
		return 0;

	clen = cpylen = rb_linebuf_skip_crlf(ch, len);
	if(clen == -1)
		return -1;

	/* This is the ~overflow case..This doesn't happen often.. */
	if(cpylen > (BUF_DATA_SIZE - bufline->len - 1))
	{
		memcpy(bufch, ch, (BUF_DATA_SIZE - bufline->len - 1));
		bufline->buf[BUF_DATA_SIZE - 1] = '\0';
		bufch = bufline->buf + BUF_DATA_SIZE - 2;
		while(cpylen && (*bufch == '\r' || *bufch == '\n'))
		{
			*bufch = '\0';
			cpylen--;
			bufch--;
		}
		bufline->terminated = 1;
		bufline->len = BUF_DATA_SIZE - 1;
		bufhead->len += BUF_DATA_SIZE - 1;
		return clen;
	}

	memcpy(bufch, ch, cpylen);
	bufch += cpylen;
	*bufch = '\0';
	bufch--;

	if(*bufch != '\r' && *bufch != '\n')
	{
		/* No linefeed, bail for the next time */
		bufhead->len += cpylen;
		bufline->len += cpylen;
		bufline->terminated = 0;
		return clen;
	}

	/* Yank the CRLF off this, replace with a \0 */
	while(cpylen && (*bufch == '\r' || *bufch == '\n'))
	{
		*bufch = '\0';
		cpylen--;
		bufch--;
	}

	bufline->terminated = 1;
	bufhead->len += cpylen;
	bufline->len += cpylen;
	return clen;
}

/*
 * rb_linebuf_copy_raw
 *
 * Copy as much data as possible directly into a linebuf,
 * splitting at \r\n, but without altering any data.
 *
 */
static int
rb_linebuf_copy_raw(buf_head_t * bufhead, buf_line_t * bufline, char *data, int len)
{
	int cpylen = 0;		/* how many bytes we've copied */
	char *ch = data;	/* Pointer to where we are in the read data */
	char *bufch = bufline->buf + bufline->len;
	int clen = 0;		/* how many bytes we've processed,
				   and don't ever want to see again.. */

	/* If its full or terminated, ignore it */

	bufline->raw = 1;
	lrb_assert(bufline->len < BUF_DATA_SIZE);
	if(bufline->terminated == 1)
		return 0;

	clen = cpylen = rb_linebuf_skip_crlf(ch, len);
	if(clen == -1)
		return -1;

	/* This is the overflow case..This doesn't happen often.. */
	if(cpylen > (BUF_DATA_SIZE - bufline->len - 1))
	{
		clen = BUF_DATA_SIZE - bufline->len - 1;
		memcpy(bufch, ch, clen);
		bufline->buf[BUF_DATA_SIZE - 1] = '\0';
		bufch = bufline->buf + BUF_DATA_SIZE - 2;
		bufline->terminated = 1;
		bufline->len = BUF_DATA_SIZE - 1;
		bufhead->len += BUF_DATA_SIZE - 1;
		return clen;
	}

	memcpy(bufch, ch, cpylen);
	bufch += cpylen;
	*bufch = '\0';
	bufch--;

	if(*bufch != '\r' && *bufch != '\n')
	{
		/* No linefeed, bail for the next time */
		bufhead->len += cpylen;
		bufline->len += cpylen;
		bufline->terminated = 0;
		return clen;
	}

	bufline->terminated = 1;
	bufhead->len += cpylen;
	bufline->len += cpylen;
	return clen;
}


/*
 * rb_linebuf_parse
 *
 * Take a given buffer and break out as many buffers as we can.
 * If we find a CRLF, we terminate that buffer and create a new one.
 * If we don't find a CRLF whilst parsing a buffer, we don't mark it
 * 'finished', so the next loop through we can continue appending ..
 *
 * A few notes here, which you'll need to understand before continuing.
 *
 * - right now I'm only dealing with single sized buffers. Later on,
 *   I might consider chaining buffers together to get longer "lines"
 *   but seriously, I don't see the advantage right now.
 *
 * - This *is* designed to turn into a reference-counter-protected setup
 *   to dodge copious copies.
 */
int
rb_linebuf_parse(buf_head_t * bufhead, char *data, int len, int raw)
{
	buf_line_t *bufline;
	int cpylen;
	int linecnt = 0;

	/* First, if we have a partial buffer, try to squeze data into it */
	if(bufhead->list.tail != NULL)
	{
		/* Check we're doing the partial buffer thing */
		bufline = bufhead->list.tail->data;
		/* just try, the worst it could do is *reject* us .. */
		if(!raw)
			cpylen = rb_linebuf_copy_line(bufhead, bufline, data, len);
		else
			cpylen = rb_linebuf_copy_raw(bufhead, bufline, data, len);

		if(cpylen == -1)
			return -1;

		linecnt++;
		/* If we've copied the same as what we've got, quit now */
		if(cpylen == len)
			return linecnt;	/* all the data done so soon? */

		/* Skip the data and update len .. */
		len -= cpylen;
		lrb_assert(len >= 0);
		data += cpylen;
	}

	/* Next, the loop */
	while(len > 0)
	{
		/* We obviously need a new buffer, so .. */
		bufline = rb_linebuf_new_line(bufhead);

		/* And parse */
		if(!raw)
			cpylen = rb_linebuf_copy_line(bufhead, bufline, data, len);
		else
			cpylen = rb_linebuf_copy_raw(bufhead, bufline, data, len);

		if(cpylen == -1)
			return -1;

		len -= cpylen;
		lrb_assert(len >= 0);
		data += cpylen;
		linecnt++;
	}
	return linecnt;
}


/*
 * rb_linebuf_get
 *
 * get the next buffer from our line. For the time being it will copy
 * data into the given buffer and free the underlying linebuf.
 */
int
rb_linebuf_get(buf_head_t * bufhead, char *buf, int buflen, int partial, int raw)
{
	buf_line_t *bufline;
	int cpylen;
	char *start, *ch;

	/* make sure we have a line */
	if(bufhead->list.head == NULL)
		return 0;	/* Obviously not.. hrm. */

	bufline = bufhead->list.head->data;

	/* make sure that the buffer was actually *terminated */
	if(!(partial || bufline->terminated))
		return 0;	/* Wait for more data! */

	if(buflen < bufline->len)
		cpylen = buflen - 1;
	else
		cpylen = bufline->len;

	/* Copy it */
	start = bufline->buf;

	/* if we left extraneous '\r\n' characters in the string,
	 * and we don't want to read the raw data, clean up the string.
	 */
	if(bufline->raw && !raw)
	{
		/* skip leading EOL characters */
		while(cpylen && (*start == '\r' || *start == '\n'))
		{
			start++;
			cpylen--;
		}
		/* skip trailing EOL characters */
		ch = &start[cpylen - 1];
		while(cpylen && (*ch == '\r' || *ch == '\n'))
		{
			ch--;
			cpylen--;
		}
	}

	memcpy(buf, start, cpylen);

	/* convert CR/LF to NULL */
	if(!raw)
		buf[cpylen] = '\0';

	lrb_assert(cpylen >= 0);

	/* Deallocate the line */
	rb_linebuf_done_line(bufhead, bufline, bufhead->list.head);

	/* return how much we copied */
	return cpylen;
}

/*
 * rb_linebuf_attach
 *
 * attach the lines in a buf_head_t to another buf_head_t
 * without copying the data (using refcounts).
 */
void
rb_linebuf_attach(buf_head_t * bufhead, buf_head_t * new)
{
	rb_dlink_node *ptr;
	buf_line_t *line;

	RB_DLINK_FOREACH(ptr, new->list.head)
	{
		line = ptr->data;
		rb_dlinkAddTailAlloc(line, &bufhead->list);

		/* Update the allocated size */
		bufhead->alloclen++;
		bufhead->len += line->len;
		bufhead->numlines++;

		line->refcount++;
	}
}


/*
 * rb_linebuf_putmsg
 *
 * Similar to rb_linebuf_put, but designed for use by send.c.
 *
 * prefixfmt is used as a format for the varargs, and is inserted first.
 * Then format/va_args is appended to the buffer.
 */
void
rb_linebuf_putmsg(buf_head_t * bufhead, const char *format, va_list * va_args,
		  const char *prefixfmt, ...)
{
	buf_line_t *bufline;
	int len = 0;
	va_list prefix_args;

	/* make sure the previous line is terminated */
#ifndef NDEBUG
	if(bufhead->list.tail)
	{
		bufline = bufhead->list.tail->data;
		lrb_assert(bufline->terminated);
	}
#endif
	/* Create a new line */
	bufline = rb_linebuf_new_line(bufhead);

	if(prefixfmt != NULL)
	{
		va_start(prefix_args, prefixfmt);
		len = rb_vsnprintf(bufline->buf, BUF_DATA_SIZE, prefixfmt, prefix_args);
		va_end(prefix_args);
	}

	if(va_args != NULL)
	{
		len += rb_vsnprintf((bufline->buf + len), (BUF_DATA_SIZE - len), format, *va_args);
	}

	bufline->terminated = 1;

	/* Truncate the data if required */
	if(rb_unlikely(len > 510))
	{
		len = 510;
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
	}
	else if(rb_unlikely(len == 0))
	{
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
		bufline->buf[len] = '\0';
	}
	else
	{
		/* Chop trailing CRLF's .. */
		while((bufline->buf[len] == '\r') || (bufline->buf[len] == '\n')
		      || (bufline->buf[len] == '\0'))
		{
			len--;
		}

		bufline->buf[++len] = '\r';
		bufline->buf[++len] = '\n';
		bufline->buf[++len] = '\0';
	}

	bufline->len = len;
	bufhead->len += len;
}

void
rb_linebuf_putbuf(buf_head_t * bufhead, const char *buffer)
{
	buf_line_t *bufline;
	int len = 0;

	/* make sure the previous line is terminated */
#ifndef NDEBUG
	if(bufhead->list.tail)
	{
		bufline = bufhead->list.tail->data;
		lrb_assert(bufline->terminated);
	}
#endif
	/* Create a new line */
	bufline = rb_linebuf_new_line(bufhead);

	if(rb_unlikely(buffer != NULL))
		len = rb_strlcpy(bufline->buf, buffer, BUF_DATA_SIZE);

	bufline->terminated = 1;

	/* Truncate the data if required */
	if(rb_unlikely(len > 510))
	{
		len = 510;
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
	}
	else if(rb_unlikely(len == 0))
	{
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
		bufline->buf[len] = '\0';
	}
	else
	{
		/* Chop trailing CRLF's .. */
		while((bufline->buf[len] == '\r') || (bufline->buf[len] == '\n')
		      || (bufline->buf[len] == '\0'))
		{
			len--;
		}

		bufline->buf[++len] = '\r';
		bufline->buf[++len] = '\n';
		bufline->buf[++len] = '\0';
	}

	bufline->len = len;
	bufhead->len += len;


}

void
rb_linebuf_put(buf_head_t * bufhead, const char *format, ...)
{
	buf_line_t *bufline;
	int len = 0;
	va_list args;

	/* make sure the previous line is terminated */
#ifndef NDEBUG
	if(bufhead->list.tail)
	{
		bufline = bufhead->list.tail->data;
		lrb_assert(bufline->terminated);
	}
#endif
	/* Create a new line */
	bufline = rb_linebuf_new_line(bufhead);

	if(rb_unlikely(format != NULL))
	{
		va_start(args, format);
		len = rb_vsnprintf(bufline->buf, BUF_DATA_SIZE, format, args);
		va_end(args);
	}

	bufline->terminated = 1;

	/* Truncate the data if required */
	if(rb_unlikely(len > 510))
	{
		len = 510;
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
	}
	else if(rb_unlikely(len == 0))
	{
		bufline->buf[len++] = '\r';
		bufline->buf[len++] = '\n';
		bufline->buf[len] = '\0';
	}
	else
	{
		/* Chop trailing CRLF's .. */
		while((bufline->buf[len] == '\r') || (bufline->buf[len] == '\n')
		      || (bufline->buf[len] == '\0'))
		{
			len--;
		}

		bufline->buf[++len] = '\r';
		bufline->buf[++len] = '\n';
		bufline->buf[++len] = '\0';
	}

	bufline->len = len;
	bufhead->len += len;
}


/*
 * rb_linebuf_flush
 *
 * Flush data to the buffer. It tries to write as much data as possible
 * to the given socket. Any return values are passed straight through.
 * If there is no data in the socket, EWOULDBLOCK is set as an errno
 * rather than returning 0 (which would map to an EOF..)
 *
 * Notes: XXX We *should* have a clue here when a non-full buffer is arrived.
 *        and tag it so that we don't re-schedule another write until
 *        we have a CRLF.
 */
int
rb_linebuf_flush(rb_fde_t *F, buf_head_t * bufhead)
{
	buf_line_t *bufline;
	int retval;

/*
 * autoconf checks for this..but really just want to use it if we have a
 * native version even if libircd provides a fake version...
 */
#ifdef HAVE_WRITEV
	if(!rb_fd_ssl(F))
	{
		rb_dlink_node *ptr;
		int x = 0, y;
		int xret;
		static struct rb_iovec vec[RB_UIO_MAXIOV];

		memset(vec, 0, sizeof(vec));
		/* Check we actually have a first buffer */
		if(bufhead->list.head == NULL)
		{
			/* nope, so we return none .. */
			errno = EWOULDBLOCK;
			return -1;
		}

		ptr = bufhead->list.head;

		bufline = ptr->data;
		if(!bufline->terminated)
		{
			errno = EWOULDBLOCK;
			return -1;

		}

		vec[x].iov_base = bufline->buf + bufhead->writeofs;
		vec[x++].iov_len = bufline->len - bufhead->writeofs;
		ptr = ptr->next;

		do
		{
			if(ptr == NULL)
				break;

			bufline = ptr->data;
			if(!bufline->terminated)
				break;

			vec[x].iov_base = bufline->buf;
			vec[x].iov_len = bufline->len;
			ptr = ptr->next;

		}
		while(++x < RB_UIO_MAXIOV);

		if(x == 0)
		{
			errno = EWOULDBLOCK;
			return -1;
		}

		xret = retval = rb_writev(F, vec, x);
		if(retval <= 0)
			return retval;

		ptr = bufhead->list.head;

		for(y = 0; y < x; y++)
		{
			bufline = ptr->data;

			if(xret >= bufline->len - bufhead->writeofs)
			{
				xret -= bufline->len - bufhead->writeofs;
				ptr = ptr->next;
				rb_linebuf_done_line(bufhead, bufline, bufhead->list.head);
				bufhead->writeofs = 0;
			}
			else
			{
				bufhead->writeofs += xret;
				break;
			}
		}

		return retval;
	}
#endif

	/* this is the non-writev case */

	/* Check we actually have a first buffer */
	if(bufhead->list.head == NULL)
	{
		/* nope, so we return none .. */
		errno = EWOULDBLOCK;
		return -1;
	}

	bufline = bufhead->list.head->data;

	/* And that its actually full .. */
	if(!bufline->terminated)
	{
		errno = EWOULDBLOCK;
		return -1;
	}

	/* Now, try writing data */
	retval = rb_write(F, bufline->buf + bufhead->writeofs, bufline->len - bufhead->writeofs);

	if(retval <= 0)
		return retval;

	/* we've got data, so update the write offset */
	bufhead->writeofs += retval;

	/* if we've written everything *and* the CRLF, deallocate and update
	   bufhead */
	if(bufhead->writeofs == bufline->len)
	{
		bufhead->writeofs = 0;
		lrb_assert(bufhead->len >= 0);
		rb_linebuf_done_line(bufhead, bufline, bufhead->list.head);
	}

	/* Return line length */
	return retval;
}


/*
 * count linebufs for stats z
 */

void
rb_count_rb_linebuf_memory(size_t *count, size_t *rb_linebuf_memory_used)
{
	rb_bh_usage(rb_linebuf_heap, count, NULL, rb_linebuf_memory_used, NULL);
}