lighttpd1.4/src/mod_fastcgi.c

#include <sys/types.h>
#include <unistd.h>
#include <errno.h>
#include <fcntl.h>
#include <string.h>
#include <stdlib.h>
#include <ctype.h>
#include <assert.h>
#include <signal.h>

#include "buffer.h"
#include "server.h"
#include "keyvalue.h"
#include "log.h"

#include "http_chunk.h"
#include "fdevent.h"
#include "connections.h"
#include "response.h"
#include "joblist.h"

#include "plugin.h"

#include "inet_ntop_cache.h"
#include "stat_cache.h"
#include "status_counter.h"

#ifdef HAVE_FASTCGI_FASTCGI_H
#include <fastcgi/fastcgi.h>
#else
#ifdef HAVE_FASTCGI_H
#include <fastcgi.h>
#else
#include "fastcgi.h"
#endif
#endif /* HAVE_FASTCGI_FASTCGI_H */
#include <stdio.h>

#ifdef HAVE_SYS_FILIO_H
# include <sys/filio.h>
#endif

#include "sys-socket.h"

#ifdef HAVE_SYS_UIO_H
#include <sys/uio.h>
#endif
#ifdef HAVE_SYS_WAIT_H
#include <sys/wait.h>
#endif

#define FCGI_ENV_ADD_CHECK(ret, con) \
	if (ret == -1) { \
		con->http_status = 400; \
		con->file_finished = 1; \
		return -1; \
	};

/*
 *
 * TODO:
 *
 * - add timeout for a connect to a non-fastcgi process
 *   (use state_timestamp + state)
 *
 */

typedef struct fcgi_proc {
	size_t id; /* id will be between 1 and max_procs */
	buffer *unixsocket; /* config.socket + "-" + id */
	unsigned port;  /* config.port + pno */

	buffer *connection_name; /* either tcp:<host>:<port> or unix:<socket> for debugging purposes */

	pid_t pid;   /* PID of the spawned process (0 if not spawned locally) */


	size_t load; /* number of requests waiting on this process */

	time_t last_used; /* see idle_timeout */
	size_t requests;  /* see max_requests */
	struct fcgi_proc *prev, *next; /* see first */

	time_t disabled_until; /* this proc is disabled until, use something else until then */

	int is_local;

	enum {
		PROC_STATE_UNSET,    /* init-phase */
		PROC_STATE_RUNNING,  /* alive */
		PROC_STATE_OVERLOADED, /* listen-queue is full,
					  don't send anything to this proc for the next 2 seconds */
		PROC_STATE_DIED_WAIT_FOR_PID, /* */
		PROC_STATE_DIED,     /* marked as dead, should be restarted */
		PROC_STATE_KILLED    /* was killed as we don't have the load anymore */
	} state;
} fcgi_proc;

typedef struct {
	/* the key that is used to reference this value */
	buffer *id;

	/* list of processes handling this extension
	 * sorted by lowest load
	 *
	 * whenever a job is done move it up in the list
	 * until it is sorted, move it down as soon as the
	 * job is started
	 */
	fcgi_proc *first;
	fcgi_proc *unused_procs;

	/*
	 * spawn at least min_procs, at max_procs.
	 *
	 * as soon as the load of the first entry
	 * is max_load_per_proc we spawn a new one
	 * and add it to the first entry and give it
	 * the load
	 *
	 */

	unsigned short min_procs;
	unsigned short max_procs;
	size_t num_procs;    /* how many procs are started */
	size_t active_procs; /* how many of them are really running */

	unsigned short max_load_per_proc;

	/*
	 * kick the process from the list if it was not
	 * used for idle_timeout until min_procs is
	 * reached. this helps to get the processlist
	 * small again we had a small peak load.
	 *
	 */

	unsigned short idle_timeout;

	/*
	 * time after a disabled remote connection is tried to be re-enabled
	 *
	 *
	 */

	unsigned short disable_time;

	/*
	 * some fastcgi processes get a little bit larger
	 * than wanted. max_requests_per_proc kills a
	 * process after a number of handled requests.
	 *
	 */
	size_t max_requests_per_proc;


	/* config */

	/*
	 * host:port
	 *
	 * if host is one of the local IP adresses the
	 * whole connection is local
	 *
	 * if port is not 0, and host is not specified,
	 * "localhost" (INADDR_LOOPBACK) is assumed.
	 *
	 */
	buffer *host;
	unsigned short port;

	/*
	 * Unix Domain Socket
	 *
	 * instead of TCP/IP we can use Unix Domain Sockets
	 * - more secure (you have fileperms to play with)
	 * - more control (on locally)
	 * - more speed (no extra overhead)
	 */
	buffer *unixsocket;

	/* if socket is local we can start the fastcgi
	 * process ourself
	 *
	 * bin-path is the path to the binary
	 *
	 * check min_procs and max_procs for the number
	 * of process to start up
	 */
	buffer *bin_path;

	/* bin-path is set bin-environment is taken to
	 * create the environement before starting the
	 * FastCGI process
	 *
	 */
	array *bin_env;

	array *bin_env_copy;

	/*
	 * docroot-translation between URL->phys and the
	 * remote host
	 *
	 * reasons:
	 * - different dir-layout if remote
	 * - chroot if local
	 *
	 */
	buffer *docroot;

	/*
	 * fastcgi-mode:
	 * - responser
	 * - authorizer
	 *
	 */
	unsigned short mode;

	/*
	 * check_local tells you if the phys file is stat()ed
	 * or not. FastCGI doesn't care if the service is
	 * remote. If the web-server side doesn't contain
	 * the fastcgi-files we should not stat() for them
	 * and say '404 not found'.
	 */
	unsigned short check_local;

	/*
	 * append PATH_INFO to SCRIPT_FILENAME
	 *
	 * php needs this if cgi.fix_pathinfo is provided
	 *
	 */

	unsigned short break_scriptfilename_for_php;

	/*
	 * If the backend includes X-LIGHTTPD-send-file in the response
	 * we use the value as filename and ignore the content.
	 *
	 */
	unsigned short allow_xsendfile;

	ssize_t load; /* replace by host->load */

	size_t max_id; /* corresponds most of the time to
	num_procs.

	only if a process is killed max_id waits for the process itself
	to die and decrements it afterwards */

	buffer *strip_request_uri;

	unsigned short kill_signal; /* we need a setting for this as libfcgi
				       applications prefer SIGUSR1 while the
				       rest of the world would use SIGTERM
				       *sigh* */
} fcgi_extension_host;

/*
 * one extension can have multiple hosts assigned
 * one host can spawn additional processes on the same
 *   socket (if we control it)
 *
 * ext -> host -> procs
 *    1:n     1:n
 *
 * if the fastcgi process is remote that whole goes down
 * to
 *
 * ext -> host -> procs
 *    1:n     1:1
 *
 * in case of PHP and FCGI_CHILDREN we have again a procs
 * but we don't control it directly.
 *
 */

typedef struct {
	buffer *key; /* like .php */

	int note_is_sent;
	int last_used_ndx;

	fcgi_extension_host **hosts;

	size_t used;
	size_t size;
} fcgi_extension;

typedef struct {
	fcgi_extension **exts;

	size_t used;
	size_t size;
} fcgi_exts;


typedef struct {
	fcgi_exts *exts;

	array *ext_mapping;

	int debug;
} plugin_config;

typedef struct {
	size_t *ptr;
	size_t used;
	size_t size;
} buffer_uint;

typedef struct {
	char **ptr;

	size_t size;
	size_t used;
} char_array;

/* generic plugin data, shared between all connections */
typedef struct {
	PLUGIN_DATA;
	buffer_uint fcgi_request_id;

	buffer *fcgi_env;

	buffer *path;
	buffer *parse_response;

	buffer *statuskey;

	plugin_config **config_storage;

	plugin_config conf; /* this is only used as long as no handler_ctx is setup */
} plugin_data;

/* connection specific data */
typedef enum {
	FCGI_STATE_UNSET,
	FCGI_STATE_INIT,
	FCGI_STATE_CONNECT_DELAYED,
	FCGI_STATE_PREPARE_WRITE,
	FCGI_STATE_WRITE,
	FCGI_STATE_READ
} fcgi_connection_state_t;

typedef struct {
	fcgi_proc *proc;
	fcgi_extension_host *host;
	fcgi_extension *ext;

	fcgi_connection_state_t state;
	time_t   state_timestamp;

	int      reconnects; /* number of reconnect attempts */

	chunkqueue *rb; /* read queue */
	chunkqueue *wb; /* write queue */

	buffer   *response_header;

	size_t    request_id;
	int       fd;        /* fd to the fastcgi process */
	int       fde_ndx;   /* index into the fd-event buffer */

	pid_t     pid;
	int       got_proc;

	int       send_content_body;

	plugin_config conf;

	connection *remote_conn;  /* dumb pointer */
	plugin_data *plugin_data; /* dumb pointer */
} handler_ctx;


/* ok, we need a prototype */
static handler_t fcgi_handle_fdevent(void *s, void *ctx, int revents);

int fastcgi_status_copy_procname(buffer *b, fcgi_extension_host *host, fcgi_proc *proc) {
	buffer_copy_string(b, "fastcgi.backend.");
	buffer_append_string_buffer(b, host->id);
	if (proc) {
		buffer_append_string(b, ".");
		buffer_append_long(b, proc->id);
	}

	return 0;
}

int fastcgi_status_init(server *srv, buffer *b, fcgi_extension_host *host, fcgi_proc *proc) {
#define CLEAN(x) \
	fastcgi_status_copy_procname(b, host, proc); \
	buffer_append_string(b, x); \
	status_counter_set(srv, CONST_BUF_LEN(b), 0);

	CLEAN(".disabled");
	CLEAN(".died");
	CLEAN(".overloaded");
	CLEAN(".connected");
	CLEAN(".load");

#undef CLEAN

#define CLEAN(x) \
	fastcgi_status_copy_procname(b, host, NULL); \
	buffer_append_string(b, x); \
	status_counter_set(srv, CONST_BUF_LEN(b), 0);

	CLEAN(".load");

#undef CLEAN

	return 0;
}

static handler_ctx * handler_ctx_init() {
	handler_ctx * hctx;

	hctx = calloc(1, sizeof(*hctx));
	assert(hctx);

	hctx->fde_ndx = -1;

	hctx->response_header = buffer_init();

	hctx->request_id = 0;
	hctx->state = FCGI_STATE_INIT;
	hctx->proc = NULL;

	hctx->fd = -1;

	hctx->reconnects = 0;
	hctx->send_content_body = 1;

	hctx->rb = chunkqueue_init();
	hctx->wb = chunkqueue_init();

	return hctx;
}

static void handler_ctx_free(handler_ctx *hctx) {
	if (hctx->host) {
		hctx->host->load--;
		hctx->host = NULL;
	}

	buffer_free(hctx->response_header);

	chunkqueue_free(hctx->rb);
	chunkqueue_free(hctx->wb);

	free(hctx);
}

fcgi_proc *fastcgi_process_init() {
	fcgi_proc *f;

	f = calloc(1, sizeof(*f));
	f->unixsocket = buffer_init();
	f->connection_name = buffer_init();

	f->prev = NULL;
	f->next = NULL;

	return f;
}

void fastcgi_process_free(fcgi_proc *f) {
	if (!f) return;

	fastcgi_process_free(f->next);

	buffer_free(f->unixsocket);
	buffer_free(f->connection_name);

	free(f);
}

fcgi_extension_host *fastcgi_host_init() {
	fcgi_extension_host *f;

	f = calloc(1, sizeof(*f));

	f->id = buffer_init();
	f->host = buffer_init();
	f->unixsocket = buffer_init();
	f->docroot = buffer_init();
	f->bin_path = buffer_init();
	f->bin_env = array_init();
	f->bin_env_copy = array_init();
	f->strip_request_uri = buffer_init();

	return f;
}

void fastcgi_host_free(fcgi_extension_host *h) {
	if (!h) return;

	buffer_free(h->id);
	buffer_free(h->host);
	buffer_free(h->unixsocket);
	buffer_free(h->docroot);
	buffer_free(h->bin_path);
	buffer_free(h->strip_request_uri);
	array_free(h->bin_env);
	array_free(h->bin_env_copy);

	fastcgi_process_free(h->first);
	fastcgi_process_free(h->unused_procs);

	free(h);

}

fcgi_exts *fastcgi_extensions_init() {
	fcgi_exts *f;

	f = calloc(1, sizeof(*f));

	return f;
}

void fastcgi_extensions_free(fcgi_exts *f) {
	size_t i;

	if (!f) return;

	for (i = 0; i < f->used; i++) {
		fcgi_extension *fe;
		size_t j;

		fe = f->exts[i];

		for (j = 0; j < fe->used; j++) {
			fcgi_extension_host *h;

			h = fe->hosts[j];

			fastcgi_host_free(h);
		}

		buffer_free(fe->key);
		free(fe->hosts);

		free(fe);
	}

	free(f->exts);

	free(f);
}

int fastcgi_extension_insert(fcgi_exts *ext, buffer *key, fcgi_extension_host *fh) {
	fcgi_extension *fe;
	size_t i;

	/* there is something */

	for (i = 0; i < ext->used; i++) {
		if (buffer_is_equal(key, ext->exts[i]->key)) {
			break;
		}
	}

	if (i == ext->used) {
		/* filextension is new */
		fe = calloc(1, sizeof(*fe));
		assert(fe);
		fe->key = buffer_init();
		fe->last_used_ndx = -1;
		buffer_copy_string_buffer(fe->key, key);

		/* */

		if (ext->size == 0) {
			ext->size = 8;
			ext->exts = malloc(ext->size * sizeof(*(ext->exts)));
			assert(ext->exts);
		} else if (ext->used == ext->size) {
			ext->size += 8;
			ext->exts = realloc(ext->exts, ext->size * sizeof(*(ext->exts)));
			assert(ext->exts);
		}
		ext->exts[ext->used++] = fe;
	} else {
		fe = ext->exts[i];
	}

	if (fe->size == 0) {
		fe->size = 4;
		fe->hosts = malloc(fe->size * sizeof(*(fe->hosts)));
		assert(fe->hosts);
	} else if (fe->size == fe->used) {
		fe->size += 4;
		fe->hosts = realloc(fe->hosts, fe->size * sizeof(*(fe->hosts)));
		assert(fe->hosts);
	}

	fe->hosts[fe->used++] = fh;

	return 0;

}

INIT_FUNC(mod_fastcgi_init) {
	plugin_data *p;

	p = calloc(1, sizeof(*p));

	p->fcgi_env = buffer_init();

	p->path = buffer_init();
	p->parse_response = buffer_init();

	p->statuskey = buffer_init();

	return p;
}


FREE_FUNC(mod_fastcgi_free) {
	plugin_data *p = p_d;
	buffer_uint *r = &(p->fcgi_request_id);

	UNUSED(srv);

	if (r->ptr) free(r->ptr);

	buffer_free(p->fcgi_env);
	buffer_free(p->path);
	buffer_free(p->parse_response);
	buffer_free(p->statuskey);

	if (p->config_storage) {
		size_t i, j, n;
		for (i = 0; i < srv->config_context->used; i++) {
			plugin_config *s = p->config_storage[i];
			fcgi_exts *exts;

			if (!s) continue;

			exts = s->exts;

			for (j = 0; j < exts->used; j++) {
				fcgi_extension *ex;

				ex = exts->exts[j];

				for (n = 0; n < ex->used; n++) {
					fcgi_proc *proc;
					fcgi_extension_host *host;

					host = ex->hosts[n];

					for (proc = host->first; proc; proc = proc->next) {
						if (proc->pid != 0) {
							kill(proc->pid, host->kill_signal);
						}

						if (proc->is_local &&
						    !buffer_is_empty(proc->unixsocket)) {
							unlink(proc->unixsocket->ptr);
						}
					}

					for (proc = host->unused_procs; proc; proc = proc->next) {
						if (proc->pid != 0) {
							kill(proc->pid, host->kill_signal);
						}
						if (proc->is_local &&
						    !buffer_is_empty(proc->unixsocket)) {
							unlink(proc->unixsocket->ptr);
						}
					}
				}
			}

			fastcgi_extensions_free(s->exts);
			array_free(s->ext_mapping);

			free(s);
		}
		free(p->config_storage);
	}

	free(p);

	return HANDLER_GO_ON;
}

static int env_add(char_array *env, const char *key, size_t key_len, const char *val, size_t val_len) {
	char *dst;

	if (!key || !val) return -1;

	dst = malloc(key_len + val_len + 3);
	memcpy(dst, key, key_len);
	dst[key_len] = '=';
	/* add the \0 from the value */
	memcpy(dst + key_len + 1, val, val_len + 1);

	if (env->size == 0) {
		env->size = 16;
		env->ptr = malloc(env->size * sizeof(*env->ptr));
	} else if (env->size == env->used + 1) {
		env->size += 16;
		env->ptr = realloc(env->ptr, env->size * sizeof(*env->ptr));
	}

	env->ptr[env->used++] = dst;

	return 0;
}

static int parse_binpath(char_array *env, buffer *b) {
	char *start;
	size_t i;
	/* search for spaces */

	start = b->ptr;
	for (i = 0; i < b->used - 1; i++) {
		switch(b->ptr[i]) {
		case ' ':
		case '\t':
			/* a WS, stop here and copy the argument */

			if (env->size == 0) {
				env->size = 16;
				env->ptr = malloc(env->size * sizeof(*env->ptr));
			} else if (env->size == env->used) {
				env->size += 16;
				env->ptr = realloc(env->ptr, env->size * sizeof(*env->ptr));
			}