queue.c

Go to the documentation of this file.

/*
 *   This program is 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 St, Fifth Floor, Boston, MA 02110-1301, USA
 */
 
/**
 * $Id: d341b2177ca1a54d87bd1d48a81ede4a2e8e7d38 $
 * @file lib/bio/queue.c
 * @brief Binary IO abstractions for queues of raw packets
 *
 * @copyright 2024 Network RADIUS SAS (legal@networkradius.com)
 */
 
#include <freeradius-devel/bio/bio_priv.h>
#include <freeradius-devel/bio/queue.h>
#include <freeradius-devel/bio/null.h>
#include <freeradius-devel/util/dlist.h>
 
typedef struct fr_bio_queue_list_s      fr_bio_queue_list_t;
typedef struct fr_bio_queue_s           fr_bio_queue_t;
 
/*
 *      Define type-safe wrappers for head and entry definitions.
 */
FR_DLIST_TYPES(fr_bio_queue_list)
 
/*
 *      For delayed writes.
 *
 *      @todo - we can remove the "cancelled" field by setting packet_ctx == my?
 */
struct fr_bio_queue_entry_s {
        void            *packet_ctx;
        void const      *buffer;
        size_t          size;
        size_t          already_written;
        bool            cancelled;
 
        fr_bio_queue_t *my;
 
        FR_DLIST_ENTRY(fr_bio_queue_list)       entry;          //!< List entry.
};
 
FR_DLIST_FUNCS(fr_bio_queue_list, fr_bio_queue_entry_t, entry)
 
typedef struct fr_bio_queue_s {
        FR_BIO_COMMON;
 
        size_t                          max_saved;
 
        fr_bio_queue_saved_t            saved;
        fr_bio_queue_callback_t         sent;
        fr_bio_queue_callback_t         cancel;
 
        FR_DLIST_HEAD(fr_bio_queue_list)        pending;
        FR_DLIST_HEAD(fr_bio_queue_list)        free;
 
        fr_bio_queue_entry_t            array[];
} fr_bio_queue_t;
 
static ssize_t fr_bio_queue_write_buffer(fr_bio_t *bio, void *packet_ctx, void const *buffer, size_t size);
 
/** Forcibly cancel all outstanding packets.
 *
 *  Even partially written ones.  This function is called from
 *  shutdown(), when the destructor is called, or on fatal read / write
 *  errors.
 */
static void fr_bio_queue_list_cancel(fr_bio_queue_t *my)
{
        fr_bio_queue_entry_t *item;
 
        my->bio.read = fr_bio_fail_read;
        my->bio.write = fr_bio_fail_write;
 
        if (!my->cancel) return;
 
        if (fr_bio_queue_list_num_elements(&my->pending) == 0) return;
 
        /*
         *      Cancel any remaining saved items.
         */
        while ((item = fr_bio_queue_list_pop_head(&my->pending)) != NULL) {
                my->cancel(&my->bio, item->packet_ctx, item->buffer, item->size);
                item->cancelled = true;
                fr_bio_queue_list_insert_head(&my->free, item);
        }
}
 
static int fr_bio_queue_destructor(fr_bio_queue_t *my)
{
        FR_BIO_DESTRUCTOR_COMMON;
 
        fr_assert(my->cancel);  /* otherwise it would be fr_bio_destructor */
 
        fr_bio_queue_list_cancel(my);
 
        return 0;
}
 
/** Push a packet onto a list.
 *
 */
static ssize_t fr_bio_queue_list_push(fr_bio_queue_t *my, void *packet_ctx, const void *buffer, size_t size, size_t already_written)
{
        fr_bio_queue_entry_t    *item;
 
        item = fr_bio_queue_list_pop_head(&my->free);
        if (!item) return fr_bio_error(IO_WOULD_BLOCK);
 
        /*
         *      If we're the first entry in the saved list, we can have a partially written packet.
         *
         *      Otherwise, we're a subsequent entry, and we cannot have any data which is partially written.
         */
        fr_assert((fr_bio_queue_list_num_elements(&my->pending) == 0) ||
                  (already_written == 0));
 
        item->packet_ctx = packet_ctx;
        item->buffer = buffer;
        item->size = size;
        item->already_written = already_written;
        item->cancelled = false;
 
        fr_bio_queue_list_insert_tail(&my->pending, item);
 
        if (my->saved) my->saved(&my->bio, packet_ctx, buffer, size, item);
 
        return size;
}
 
/** Write one packet to the next bio.
 *
 *  If it blocks, save the packet and return OK to the caller.
 */
static ssize_t fr_bio_queue_write_next(fr_bio_t *bio, void *packet_ctx, void const *buffer, size_t size)
{
        ssize_t rcode;
        fr_bio_queue_t *my = talloc_get_type_abort(bio, fr_bio_queue_t);
        fr_bio_t *next;
 
        /*
         *      We can't call the next bio if there's still cached data to flush.
         */
        fr_assert(fr_bio_queue_list_num_elements(&my->pending) == 0);
 
        next = fr_bio_next(&my->bio);
        fr_assert(next != NULL);
 
        /*
         *      Write the data out.  If we write all of it, we're done.
         */
        rcode = next->write(next, packet_ctx, buffer, size);
        if ((size_t) rcode == size) return rcode;
 
        if (rcode < 0) {
                /*
                 *      A non-blocking error: return it back up the chain.
                 */
                if (rcode != fr_bio_error(IO_WOULD_BLOCK)) return rcode;
 
                /*
                 *      All other errors are fatal.
                 */
                fr_bio_queue_list_cancel(my);
                return rcode;
        }
 
        /*
         *      We were flushing the next buffer, return any data which was written.
         */
        if (!buffer) return rcode;
 
        /*
         *      The next bio wrote a partial packet.  Save the entire packet, and swap the write function to
         *      save all future packets in the saved list.
         */
        bio->write = fr_bio_queue_write_buffer;
 
        fr_assert(fr_bio_queue_list_num_elements(&my->free) > 0);
 
        /*
         *      This can only error out if the free list has no more entries.
         */
        return fr_bio_queue_list_push(my, packet_ctx, buffer, size, (size_t) rcode);
}
 
/** Flush the packet list.
 *
 */
static ssize_t fr_bio_queue_write_flush(fr_bio_queue_t *my, size_t size)
{
        size_t written;
        fr_bio_t *next;
 
        if (fr_bio_queue_list_num_elements(&my->pending) == 0) {
                my->bio.write = fr_bio_queue_write_next;
                return 0;
        }
 
        next = fr_bio_next(&my->bio);
        fr_assert(next != NULL);
 
        /*
         *      Loop over the saved packets, flushing them to the next bio.
         */
        written = 0;
        while (written < size) {
                ssize_t rcode;
                fr_bio_queue_entry_t *item;
 
                /*
                 *      No more saved packets to write: stop.
                 */
                item = fr_bio_queue_list_head(&my->pending);
                if (!item) break;
 
                /*
                 *      A cancelled item must be partially written.  A cancelled item which has zero bytes
                 *      written should not be in the saved list.
                 */
                fr_assert(!item->cancelled || (item->already_written > 0));
 
                /*
                 *      Push out however much data we can to the next bio.
                 */
                rcode = next->write(next, item->packet_ctx, ((uint8_t const *) item->buffer) + item->already_written, item->size - item->already_written);
                if (rcode == 0) break;
 
                if (rcode < 0) {
                        if (rcode == fr_bio_error(IO_WOULD_BLOCK)) break;
 
                        return rcode;
                }
 
                /*
                 *      Update the written count.
                 */
                written += rcode;
                item->already_written += rcode;
 
                if (item->already_written < item->size) break;
 
                /*
                 *      We don't run "sent" callbacks for cancelled items.
                 */
                if (item->cancelled) {
                        if (my->cancel) my->cancel(&my->bio, item->packet_ctx, item->buffer, item->size);
                } else {
                        if (my->sent) my->sent(&my->bio, item->packet_ctx, item->buffer, item->size);
                }
 
                (void) fr_bio_queue_list_pop_head(&my->pending);
#ifndef NDEBUG
                item->buffer = NULL;
                item->packet_ctx = NULL;
                item->size = 0;
                item->already_written = 0;
#endif
                item->cancelled = true;
 
                fr_bio_queue_list_insert_head(&my->free, item);
        }
 
        /*
         *      If we've written all of the saved packets, go back to writing to the "next" bio.
         */
        if (fr_bio_queue_list_head(&my->pending)) my->bio.write = fr_bio_queue_write_next;
 
        return written;
}
 
/** Write to the packet list buffer.
 *
 *  The special buffer pointer of NULL means flush().  On flush, we call next->read(), and if that succeeds,
 *  go back to "pass through" mode for the buffers.
 */
static ssize_t fr_bio_queue_write_buffer(fr_bio_t *bio, void *packet_ctx, void const *buffer, size_t size)
{
        fr_bio_queue_t *my = talloc_get_type_abort(bio, fr_bio_queue_t);
 
        if (!buffer) return fr_bio_queue_write_flush(my, size);
 
        /*
         *      This can only error out if the free list has no more entries.
         */
        return fr_bio_queue_list_push(my, packet_ctx, buffer, size, 0);
}
 
/**  Read one packet from next bio.
 *  
 *  This function does NOT respect packet boundaries.  The caller should use other APIs to determine how big
 *  the "next" packet is.
 *
 *  The caller may buffer the output data itself, or it may use other APIs to do checking.
 *
 *  The main
 */
static ssize_t fr_bio_queue_read(fr_bio_t *bio, void *packet_ctx, void *buffer, size_t size)
{
        int rcode;
        fr_bio_queue_t *my = talloc_get_type_abort(bio, fr_bio_queue_t);
        fr_bio_t *next;
 
        next = fr_bio_next(&my->bio);
        fr_assert(next != NULL);
 
        rcode = next->read(next, packet_ctx, buffer, size);
        if (rcode >= 0) return rcode;
 
        /*
         *      We didn't read anything, return that.
         */
        if (rcode == fr_bio_error(IO_WOULD_BLOCK)) return rcode;
 
        /*
         *      Error reading, which means that we can't write to it, either.  We don't care if the error is
         *      EOF or anything else.  We just cancel the outstanding packets, and shut ourselves down.
         */
        fr_bio_queue_list_cancel(my);
        return rcode;
}
 
/** Shutdown
 *
 *  Cancel / close has to be called before re-init.
 */
static int fr_bio_queue_shutdown(fr_bio_t *bio)
{
        fr_bio_queue_t *my = talloc_get_type_abort(bio, fr_bio_queue_t);
 
        fr_bio_queue_list_cancel(my);
 
        return 0;
}
 
 
/** Allocate a packet-based bio.
 *
 *  This bio assumes that each call to fr_bio_write() is for one packet, and only one packet.  If the next bio
 *  returns a partial write, or WOULD BLOCK, then information about the packet is cached.  Subsequent writes
 *  will write the partial data first, and then continue with subsequent writes.
 *
 *  The caller is responsible for not freeing the packet ctx or the packet buffer until either the write has
 *  been performed, or the write has been cancelled.
 *
 *  The read() API makes no provisions for reading complete packets.  It simply returns whatever the next bio
 *  allows.  If instead there is a need to read only complete packets, then the next bio should be
 *  fr_bio_mem_alloc() with a fr_bio_mem_set_verify()
 *
 *  The read() API may return 0.  There may have been data read from an underlying FD, but that data did not
 *  make it through the filters of the "next" bios.  e.g. Any underlying FD should be put into a "wait for
 *  readable" state.
 *
 *  The write() API will return a full write, even if the next layer is blocked.  Any underlying FD
 *  should be put into a "wait for writeable" state.  The packet which was supposed to be written has been
 *  cached, and cannot be cancelled as it is partially written.  The caller should likely start using another
 *  bio for writes.  If the caller continues to use the bio, then any subsequent writes will *always* cache
 *  the packets. @todo - we need to mark up the bio as "blocked", and then have a write_blocked() API?  ugh.
 *  or just add `bool blocked` and `bool eof` to both read/write bios
 *
 *  Once the underlying FD has become writeable, the caller should call fr_bio_write(bio, NULL, NULL, SIZE_MAX);
 *  That will cause the pending packets to be flushed.
 *
 *  The write() API may return that it's written a full packet, in which case it's either completely written to
 *  the next bio, or to the pending queue.
 *
 *  The read / write APIs can return WOULD_BLOCK, in which case nothing was done.  Any underlying FD should be
 *  put into a "wait for writeable" state.  Other errors from bios "further down" the chain can also be
 *  returned.
 *
 *  @param ctx          the talloc ctx
 *  @param max_saved    Maximum number of packets to cache.  Must be 1..1^17
 *  @param saved        callback to run when a packet is saved in the pending queue
 *  @param sent         callback to run when a packet is sent.
 *  @param cancel       callback to run when a packet is cancelled.
 *  @param next         the next bio which will perform the underlying reads and writes.
 *      - NULL on error, memory allocation failed
 *      - !NULL the bio
 */
fr_bio_t *fr_bio_queue_alloc(TALLOC_CTX *ctx, size_t max_saved,
                              fr_bio_queue_saved_t saved,
                              fr_bio_queue_callback_t sent,
                              fr_bio_queue_callback_t cancel,
                              fr_bio_t *next)
{
        size_t i;
        fr_bio_queue_t *my;
 
        if (!max_saved) max_saved = 1;
        if (max_saved > (1 << 17)) max_saved = 1 << 17;
 
        my = (fr_bio_queue_t *) talloc_zero_array(ctx, uint8_t, sizeof(fr_bio_queue_t) +
                                                   sizeof(fr_bio_queue_entry_t) * max_saved);
        if (!my) return NULL;
 
        talloc_set_type(my, fr_bio_queue_t);
 
        my->max_saved = max_saved;
 
        fr_bio_queue_list_init(&my->pending);
        fr_bio_queue_list_init(&my->free);
 
        my->saved = saved;
        my->sent = sent;
        my->cancel = cancel;
 
        for (i = 0; i < max_saved; i++) {
                my->array[i].my = my;
                my->array[i].cancelled = true;
                fr_bio_queue_list_insert_tail(&my->free, &my->array[i]);
        }
 
        my->bio.read = fr_bio_queue_read;
        my->bio.write = fr_bio_queue_write_next;
        my->cb.shutdown = fr_bio_queue_shutdown;
 
        fr_bio_chain(&my->bio, next);
 
        if (my->cancel) {
                talloc_set_destructor(my, fr_bio_queue_destructor);
        } else {
                talloc_set_destructor((fr_bio_t *) my, fr_bio_destructor);
        }
 
        return (fr_bio_t *) my;
}
 
/** Cancel the write for a packet.
 *
 *  Cancel one a saved packets, and call the cancel() routine if it exists.
 *
 *  There is no way to cancel all packets.  The caller must find the lowest bio in the chain, and shutdown it.
 *  e.g. by closing the socket via fr_bio_fd_close().  That function will take care of walking back up the
 *  chain, and shutting down each bio.
 *
 *  @param      bio     the #fr_bio_queue_t
 *  @param      item    The context returned from #fr_bio_queue_saved_t
 *  @return
 *      - <0 no such packet was found in the list of saved packets, OR the packet cannot be cancelled.
 *      - 0 the packet was cancelled.
 */
int fr_bio_queue_cancel(fr_bio_t *bio, fr_bio_queue_entry_t *item)
{
        fr_bio_queue_t *my = talloc_get_type_abort(bio, fr_bio_queue_t);
 
        if (!(item >= &my->array[0]) && (item < &my->array[my->max_saved])) {
                return -1;
        }
 
        /*
         *      Already cancelled, that's a NOOP.
         */
        if (item->cancelled) return 0;
 
        /*
         *      If the item has been partially written, AND we have a working write function, see if we can
         *      cancel it.
         */
        if (item->already_written && (my->bio.write != fr_bio_null_write)) {
                ssize_t rcode;
                fr_bio_t *next;
 
                next = fr_bio_next(bio);
                fr_assert(next != NULL);
 
                /*
                 *      If the write fails or returns nothing, the item can't be cancelled.
                 */
                rcode = next->write(next, item->packet_ctx, ((uint8_t const *) item->buffer) + item->already_written, item->size - item->already_written);
                if (rcode <= 0) return -1;
 
                /*
                 *      If we haven't written the full item, it can't be cancelled.
                 */
                item->already_written += rcode;
                if (item->already_written < item->size) return -1;
 
                /*
                 *      Else the item has been fully written, it can be safely cancelled.
                 */
        }
 
        /*
         *      Remove it from the saved list, and run the cancellation callback.
         */
        (void) fr_bio_queue_list_remove(&my->pending, item);
        fr_bio_queue_list_insert_head(&my->free, item);
 
        if (my->cancel) my->cancel(bio, item->packet_ctx, item->buffer, item->size);
 
        return 0;
}