3 Utility functions to read/write blobs of data from a file descriptor
4 and handle the case where we might need multiple read/writes to get all the
7 Copyright (C) Andrew Tridgell 2006
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3 of the License, or
12 (at your option) any later version.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program; if not, see <http://www.gnu.org/licenses/>.
24 #include "lib/tdb/include/tdb.h"
25 #include "lib/events/events.h"
26 #include "lib/util/dlinklist.h"
27 #include "system/network.h"
28 #include "system/filesys.h"
29 #include "../include/ctdb_private.h"
30 #include "../include/ctdb.h"
32 /* structures for packet queueing - see common/ctdb_io.c */
38 struct ctdb_queue_pkt {
39 struct ctdb_queue_pkt *next, *prev;
46 struct ctdb_context *ctdb;
47 struct ctdb_partial partial; /* partial input packet */
48 struct ctdb_queue_pkt *out_queue, *out_queue_tail;
49 uint32_t out_queue_length;
54 ctdb_queue_cb_fn_t callback;
60 int ctdb_queue_length(struct ctdb_queue *queue)
62 return queue->out_queue_length;
66 called when an incoming connection is readable
68 static void queue_io_read(struct ctdb_queue *queue)
72 uint8_t *data, *data_base;
74 if (ioctl(queue->fd, FIONREAD, &num_ready) != 0) {
78 /* the descriptor has been closed */
83 queue->partial.data = talloc_realloc_size(queue, queue->partial.data,
84 num_ready + queue->partial.length);
86 if (queue->partial.data == NULL) {
87 DEBUG(DEBUG_ERR,("read error alloc failed for %u\n",
88 num_ready + queue->partial.length));
92 nread = read(queue->fd, queue->partial.data + queue->partial.length, num_ready);
94 DEBUG(DEBUG_ERR,("read error nread=%d\n", (int)nread));
99 data = queue->partial.data;
100 nread += queue->partial.length;
102 queue->partial.data = NULL;
103 queue->partial.length = 0;
105 if (nread >= 4 && *(uint32_t *)data == nread) {
106 /* it is the responsibility of the incoming packet
107 function to free 'data' */
108 queue->callback(data, nread, queue->private_data);
114 while (nread >= 4 && *(uint32_t *)data <= nread) {
115 /* we have at least one packet */
118 bool destroyed = false;
120 len = *(uint32_t *)data;
122 /* bad packet! treat as EOF */
123 DEBUG(DEBUG_CRIT,("Invalid packet of length 0\n"));
126 d2 = talloc_memdup(queue, data, len);
128 DEBUG(DEBUG_ERR,("read error memdup failed for %u\n", len));
133 queue->destroyed = &destroyed;
134 queue->callback(d2, len, queue->private_data);
135 /* If callback freed us, don't do anything else. */
139 queue->destroyed = NULL;
146 /* we have only part of a packet */
147 if (data_base == data) {
148 queue->partial.data = data;
149 queue->partial.length = nread;
151 queue->partial.data = talloc_memdup(queue, data, nread);
152 if (queue->partial.data == NULL) {
153 DEBUG(DEBUG_ERR,("read error memdup partial failed for %u\n",
157 queue->partial.length = nread;
158 talloc_free(data_base);
163 talloc_free(data_base);
167 queue->callback(NULL, 0, queue->private_data);
171 /* used when an event triggers a dead queue */
172 static void queue_dead(struct event_context *ev, struct timed_event *te,
173 struct timeval t, void *private_data)
175 struct ctdb_queue *queue = talloc_get_type(private_data, struct ctdb_queue);
176 queue->callback(NULL, 0, queue->private_data);
181 called when an incoming connection is writeable
183 static void queue_io_write(struct ctdb_queue *queue)
185 while (queue->out_queue) {
186 struct ctdb_queue_pkt *pkt = queue->out_queue;
188 if (queue->ctdb->flags & CTDB_FLAG_TORTURE) {
189 n = write(queue->fd, pkt->data, 1);
191 n = write(queue->fd, pkt->data, pkt->length);
194 if (n == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
195 if (pkt->length != pkt->full_length) {
196 /* partial packet sent - we have to drop it */
197 DLIST_REMOVE(queue->out_queue, pkt);
198 queue->out_queue_length--;
201 talloc_free(queue->fde);
204 event_add_timed(queue->ctdb->ev, queue, timeval_zero(),
210 if (n != pkt->length) {
216 DLIST_REMOVE(queue->out_queue, pkt);
217 queue->out_queue_length--;
221 EVENT_FD_NOT_WRITEABLE(queue->fde);
225 called when an incoming connection is readable or writeable
227 static void queue_io_handler(struct event_context *ev, struct fd_event *fde,
228 uint16_t flags, void *private_data)
230 struct ctdb_queue *queue = talloc_get_type(private_data, struct ctdb_queue);
232 if (flags & EVENT_FD_READ) {
233 queue_io_read(queue);
235 queue_io_write(queue);
241 queue a packet for sending
243 int ctdb_queue_send(struct ctdb_queue *queue, uint8_t *data, uint32_t length)
245 struct ctdb_queue_pkt *pkt;
246 uint32_t length2, full_length;
248 if (queue->alignment) {
249 /* enforce the length and alignment rules from the tcp packet allocator */
250 length2 = (length+(queue->alignment-1)) & ~(queue->alignment-1);
251 *(uint32_t *)data = length2;
256 if (length2 != length) {
257 memset(data+length, 0, length2-length);
260 full_length = length2;
262 /* if the queue is empty then try an immediate write, avoiding
263 queue overhead. This relies on non-blocking sockets */
264 if (queue->out_queue == NULL && queue->fd != -1 &&
265 !(queue->ctdb->flags & CTDB_FLAG_TORTURE)) {
266 ssize_t n = write(queue->fd, data, length2);
267 if (n == -1 && errno != EAGAIN && errno != EWOULDBLOCK) {
268 talloc_free(queue->fde);
271 event_add_timed(queue->ctdb->ev, queue, timeval_zero(),
273 /* yes, we report success, as the dead node is
274 handled via a separate event */
281 if (length2 == 0) return 0;
284 pkt = talloc(queue, struct ctdb_queue_pkt);
285 CTDB_NO_MEMORY(queue->ctdb, pkt);
287 pkt->data = talloc_memdup(pkt, data, length2);
288 CTDB_NO_MEMORY(queue->ctdb, pkt->data);
290 pkt->length = length2;
291 pkt->full_length = full_length;
293 if (queue->out_queue == NULL && queue->fd != -1) {
294 EVENT_FD_WRITEABLE(queue->fde);
297 DLIST_ADD_END(queue->out_queue, pkt, NULL);
299 queue->out_queue_length++;
301 if (queue->ctdb->tunable.verbose_memory_names != 0) {
302 struct ctdb_req_header *hdr = (struct ctdb_req_header *)pkt->data;
303 switch (hdr->operation) {
304 case CTDB_REQ_CONTROL: {
305 struct ctdb_req_control *c = (struct ctdb_req_control *)hdr;
306 talloc_set_name(pkt, "ctdb_queue_pkt: control opcode=%u srvid=%llu datalen=%u",
307 (unsigned)c->opcode, (unsigned long long)c->srvid, (unsigned)c->datalen);
310 case CTDB_REQ_MESSAGE: {
311 struct ctdb_req_message *m = (struct ctdb_req_message *)hdr;
312 talloc_set_name(pkt, "ctdb_queue_pkt: message srvid=%llu datalen=%u",
313 (unsigned long long)m->srvid, (unsigned)m->datalen);
317 talloc_set_name(pkt, "ctdb_queue_pkt: operation=%u length=%u src=%u dest=%u",
318 (unsigned)hdr->operation, (unsigned)hdr->length,
319 (unsigned)hdr->srcnode, (unsigned)hdr->destnode);
329 setup the fd used by the queue
331 int ctdb_queue_set_fd(struct ctdb_queue *queue, int fd)
334 talloc_free(queue->fde);
338 queue->fde = event_add_fd(queue->ctdb->ev, queue, fd, EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
339 queue_io_handler, queue);
340 if (queue->fde == NULL) {
344 if (queue->out_queue) {
345 EVENT_FD_WRITEABLE(queue->fde);
352 /* If someone sets up this pointer, they want to know if the queue is freed */
353 static int queue_destructor(struct ctdb_queue *queue)
355 if (queue->destroyed != NULL)
356 *queue->destroyed = true;
361 setup a packet queue on a socket
363 struct ctdb_queue *ctdb_queue_setup(struct ctdb_context *ctdb,
364 TALLOC_CTX *mem_ctx, int fd, int alignment,
366 ctdb_queue_cb_fn_t callback,
369 struct ctdb_queue *queue;
371 queue = talloc_zero(mem_ctx, struct ctdb_queue);
372 CTDB_NO_MEMORY_NULL(ctdb, queue);
376 queue->alignment = alignment;
377 queue->private_data = private_data;
378 queue->callback = callback;
380 if (ctdb_queue_set_fd(queue, fd) != 0) {
385 talloc_set_destructor(queue, queue_destructor);