• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/9p/trans_fd.c
4  *
5  * Fd transport layer.  Includes deprecated socket layer.
6  *
7  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
8  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
9  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
10  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
11  */
12 
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/in.h>
16 #include <linux/module.h>
17 #include <linux/net.h>
18 #include <linux/ipv6.h>
19 #include <linux/kthread.h>
20 #include <linux/errno.h>
21 #include <linux/kernel.h>
22 #include <linux/un.h>
23 #include <linux/uaccess.h>
24 #include <linux/inet.h>
25 #include <linux/idr.h>
26 #include <linux/file.h>
27 #include <linux/parser.h>
28 #include <linux/slab.h>
29 #include <linux/seq_file.h>
30 #include <net/9p/9p.h>
31 #include <net/9p/client.h>
32 #include <net/9p/transport.h>
33 
34 #include <linux/syscalls.h> /* killme */
35 
36 #define P9_PORT 564
37 #define MAX_SOCK_BUF (64*1024)
38 #define MAXPOLLWADDR	2
39 
40 static struct p9_trans_module p9_tcp_trans;
41 static struct p9_trans_module p9_fd_trans;
42 
43 /**
44  * struct p9_fd_opts - per-transport options
45  * @rfd: file descriptor for reading (trans=fd)
46  * @wfd: file descriptor for writing (trans=fd)
47  * @port: port to connect to (trans=tcp)
48  *
49  */
50 
51 struct p9_fd_opts {
52 	int rfd;
53 	int wfd;
54 	u16 port;
55 	bool privport;
56 };
57 
58 /*
59   * Option Parsing (code inspired by NFS code)
60   *  - a little lazy - parse all fd-transport options
61   */
62 
63 enum {
64 	/* Options that take integer arguments */
65 	Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
66 	/* Options that take no arguments */
67 	Opt_privport,
68 };
69 
70 static const match_table_t tokens = {
71 	{Opt_port, "port=%u"},
72 	{Opt_rfdno, "rfdno=%u"},
73 	{Opt_wfdno, "wfdno=%u"},
74 	{Opt_privport, "privport"},
75 	{Opt_err, NULL},
76 };
77 
78 enum {
79 	Rworksched = 1,		/* read work scheduled or running */
80 	Rpending = 2,		/* can read */
81 	Wworksched = 4,		/* write work scheduled or running */
82 	Wpending = 8,		/* can write */
83 };
84 
85 struct p9_poll_wait {
86 	struct p9_conn *conn;
87 	wait_queue_entry_t wait;
88 	wait_queue_head_t *wait_addr;
89 };
90 
91 /**
92  * struct p9_conn - fd mux connection state information
93  * @mux_list: list link for mux to manage multiple connections (?)
94  * @client: reference to client instance for this connection
95  * @err: error state
96  * @req_list: accounting for requests which have been sent
97  * @unsent_req_list: accounting for requests that haven't been sent
98  * @req: current request being processed (if any)
99  * @tmp_buf: temporary buffer to read in header
100  * @rc: temporary fcall for reading current frame
101  * @wpos: write position for current frame
102  * @wsize: amount of data to write for current frame
103  * @wbuf: current write buffer
104  * @poll_pending_link: pending links to be polled per conn
105  * @poll_wait: array of wait_q's for various worker threads
106  * @pt: poll state
107  * @rq: current read work
108  * @wq: current write work
109  * @wsched: ????
110  *
111  */
112 
113 struct p9_conn {
114 	struct list_head mux_list;
115 	struct p9_client *client;
116 	int err;
117 	struct list_head req_list;
118 	struct list_head unsent_req_list;
119 	struct p9_req_t *rreq;
120 	struct p9_req_t *wreq;
121 	char tmp_buf[P9_HDRSZ];
122 	struct p9_fcall rc;
123 	int wpos;
124 	int wsize;
125 	char *wbuf;
126 	struct list_head poll_pending_link;
127 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
128 	poll_table pt;
129 	struct work_struct rq;
130 	struct work_struct wq;
131 	unsigned long wsched;
132 };
133 
134 /**
135  * struct p9_trans_fd - transport state
136  * @rd: reference to file to read from
137  * @wr: reference of file to write to
138  * @conn: connection state reference
139  *
140  */
141 
142 struct p9_trans_fd {
143 	struct file *rd;
144 	struct file *wr;
145 	struct p9_conn conn;
146 };
147 
148 static void p9_poll_workfn(struct work_struct *work);
149 
150 static DEFINE_SPINLOCK(p9_poll_lock);
151 static LIST_HEAD(p9_poll_pending_list);
152 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
153 
154 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
155 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
156 
p9_mux_poll_stop(struct p9_conn * m)157 static void p9_mux_poll_stop(struct p9_conn *m)
158 {
159 	unsigned long flags;
160 	int i;
161 
162 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
163 		struct p9_poll_wait *pwait = &m->poll_wait[i];
164 
165 		if (pwait->wait_addr) {
166 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
167 			pwait->wait_addr = NULL;
168 		}
169 	}
170 
171 	spin_lock_irqsave(&p9_poll_lock, flags);
172 	list_del_init(&m->poll_pending_link);
173 	spin_unlock_irqrestore(&p9_poll_lock, flags);
174 
175 	flush_work(&p9_poll_work);
176 }
177 
178 /**
179  * p9_conn_cancel - cancel all pending requests with error
180  * @m: mux data
181  * @err: error code
182  *
183  */
184 
p9_conn_cancel(struct p9_conn * m,int err)185 static void p9_conn_cancel(struct p9_conn *m, int err)
186 {
187 	struct p9_req_t *req, *rtmp;
188 	LIST_HEAD(cancel_list);
189 
190 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
191 
192 	spin_lock(&m->client->lock);
193 
194 	if (m->err) {
195 		spin_unlock(&m->client->lock);
196 		return;
197 	}
198 
199 	m->err = err;
200 
201 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
202 		list_move(&req->req_list, &cancel_list);
203 		req->status = REQ_STATUS_ERROR;
204 	}
205 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
206 		list_move(&req->req_list, &cancel_list);
207 		req->status = REQ_STATUS_ERROR;
208 	}
209 
210 	spin_unlock(&m->client->lock);
211 
212 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
213 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
214 		list_del(&req->req_list);
215 		if (!req->t_err)
216 			req->t_err = err;
217 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
218 	}
219 }
220 
221 static __poll_t
p9_fd_poll(struct p9_client * client,struct poll_table_struct * pt,int * err)222 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
223 {
224 	__poll_t ret;
225 	struct p9_trans_fd *ts = NULL;
226 
227 	if (client && client->status == Connected)
228 		ts = client->trans;
229 
230 	if (!ts) {
231 		if (err)
232 			*err = -EREMOTEIO;
233 		return EPOLLERR;
234 	}
235 
236 	ret = vfs_poll(ts->rd, pt);
237 	if (ts->rd != ts->wr)
238 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
239 	return ret;
240 }
241 
242 /**
243  * p9_fd_read- read from a fd
244  * @client: client instance
245  * @v: buffer to receive data into
246  * @len: size of receive buffer
247  *
248  */
249 
p9_fd_read(struct p9_client * client,void * v,int len)250 static int p9_fd_read(struct p9_client *client, void *v, int len)
251 {
252 	int ret;
253 	struct p9_trans_fd *ts = NULL;
254 	loff_t pos;
255 
256 	if (client && client->status != Disconnected)
257 		ts = client->trans;
258 
259 	if (!ts)
260 		return -EREMOTEIO;
261 
262 	if (!(ts->rd->f_flags & O_NONBLOCK))
263 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
264 
265 	pos = ts->rd->f_pos;
266 	ret = kernel_read(ts->rd, v, len, &pos);
267 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
268 		client->status = Disconnected;
269 	return ret;
270 }
271 
272 /**
273  * p9_read_work - called when there is some data to be read from a transport
274  * @work: container of work to be done
275  *
276  */
277 
p9_read_work(struct work_struct * work)278 static void p9_read_work(struct work_struct *work)
279 {
280 	__poll_t n;
281 	int err;
282 	struct p9_conn *m;
283 
284 	m = container_of(work, struct p9_conn, rq);
285 
286 	if (m->err < 0)
287 		return;
288 
289 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
290 
291 	if (!m->rc.sdata) {
292 		m->rc.sdata = m->tmp_buf;
293 		m->rc.offset = 0;
294 		m->rc.capacity = P9_HDRSZ; /* start by reading header */
295 	}
296 
297 	clear_bit(Rpending, &m->wsched);
298 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
299 		 m, m->rc.offset, m->rc.capacity,
300 		 m->rc.capacity - m->rc.offset);
301 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
302 			 m->rc.capacity - m->rc.offset);
303 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
304 	if (err == -EAGAIN)
305 		goto end_clear;
306 
307 	if (err <= 0)
308 		goto error;
309 
310 	m->rc.offset += err;
311 
312 	/* header read in */
313 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
314 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
315 
316 		/* Header size */
317 		m->rc.size = P9_HDRSZ;
318 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
319 		if (err) {
320 			p9_debug(P9_DEBUG_ERROR,
321 				 "error parsing header: %d\n", err);
322 			goto error;
323 		}
324 
325 		if (m->rc.size >= m->client->msize) {
326 			p9_debug(P9_DEBUG_ERROR,
327 				 "requested packet size too big: %d\n",
328 				 m->rc.size);
329 			err = -EIO;
330 			goto error;
331 		}
332 
333 		p9_debug(P9_DEBUG_TRANS,
334 			 "mux %p pkt: size: %d bytes tag: %d\n",
335 			 m, m->rc.size, m->rc.tag);
336 
337 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
338 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
339 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
340 				 m->rc.tag);
341 			err = -EIO;
342 			goto error;
343 		}
344 
345 		if (!m->rreq->rc.sdata) {
346 			p9_debug(P9_DEBUG_ERROR,
347 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
348 				 m->rc.tag, m->rreq);
349 			m->rreq = NULL;
350 			err = -EIO;
351 			goto error;
352 		}
353 		m->rc.sdata = m->rreq->rc.sdata;
354 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
355 		m->rc.capacity = m->rc.size;
356 	}
357 
358 	/* packet is read in
359 	 * not an else because some packets (like clunk) have no payload
360 	 */
361 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
362 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
363 		m->rreq->rc.size = m->rc.offset;
364 		spin_lock(&m->client->lock);
365 		if (m->rreq->status == REQ_STATUS_SENT) {
366 			list_del(&m->rreq->req_list);
367 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
368 		} else if (m->rreq->status == REQ_STATUS_FLSHD) {
369 			/* Ignore replies associated with a cancelled request. */
370 			p9_debug(P9_DEBUG_TRANS,
371 				 "Ignore replies associated with a cancelled request\n");
372 		} else {
373 			spin_unlock(&m->client->lock);
374 			p9_debug(P9_DEBUG_ERROR,
375 				 "Request tag %d errored out while we were reading the reply\n",
376 				 m->rc.tag);
377 			err = -EIO;
378 			goto error;
379 		}
380 		spin_unlock(&m->client->lock);
381 		m->rc.sdata = NULL;
382 		m->rc.offset = 0;
383 		m->rc.capacity = 0;
384 		p9_req_put(m->rreq);
385 		m->rreq = NULL;
386 	}
387 
388 end_clear:
389 	clear_bit(Rworksched, &m->wsched);
390 
391 	if (!list_empty(&m->req_list)) {
392 		if (test_and_clear_bit(Rpending, &m->wsched))
393 			n = EPOLLIN;
394 		else
395 			n = p9_fd_poll(m->client, NULL, NULL);
396 
397 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
398 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
399 			schedule_work(&m->rq);
400 		}
401 	}
402 
403 	return;
404 error:
405 	p9_conn_cancel(m, err);
406 	clear_bit(Rworksched, &m->wsched);
407 }
408 
409 /**
410  * p9_fd_write - write to a socket
411  * @client: client instance
412  * @v: buffer to send data from
413  * @len: size of send buffer
414  *
415  */
416 
p9_fd_write(struct p9_client * client,void * v,int len)417 static int p9_fd_write(struct p9_client *client, void *v, int len)
418 {
419 	ssize_t ret;
420 	struct p9_trans_fd *ts = NULL;
421 
422 	if (client && client->status != Disconnected)
423 		ts = client->trans;
424 
425 	if (!ts)
426 		return -EREMOTEIO;
427 
428 	if (!(ts->wr->f_flags & O_NONBLOCK))
429 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
430 
431 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
432 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
433 		client->status = Disconnected;
434 	return ret;
435 }
436 
437 /**
438  * p9_write_work - called when a transport can send some data
439  * @work: container for work to be done
440  *
441  */
442 
p9_write_work(struct work_struct * work)443 static void p9_write_work(struct work_struct *work)
444 {
445 	__poll_t n;
446 	int err;
447 	struct p9_conn *m;
448 	struct p9_req_t *req;
449 
450 	m = container_of(work, struct p9_conn, wq);
451 
452 	if (m->err < 0) {
453 		clear_bit(Wworksched, &m->wsched);
454 		return;
455 	}
456 
457 	if (!m->wsize) {
458 		spin_lock(&m->client->lock);
459 		if (list_empty(&m->unsent_req_list)) {
460 			clear_bit(Wworksched, &m->wsched);
461 			spin_unlock(&m->client->lock);
462 			return;
463 		}
464 
465 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
466 			       req_list);
467 		req->status = REQ_STATUS_SENT;
468 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
469 		list_move_tail(&req->req_list, &m->req_list);
470 
471 		m->wbuf = req->tc.sdata;
472 		m->wsize = req->tc.size;
473 		m->wpos = 0;
474 		p9_req_get(req);
475 		m->wreq = req;
476 		spin_unlock(&m->client->lock);
477 	}
478 
479 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
480 		 m, m->wpos, m->wsize);
481 	clear_bit(Wpending, &m->wsched);
482 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
483 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
484 	if (err == -EAGAIN)
485 		goto end_clear;
486 
487 
488 	if (err < 0)
489 		goto error;
490 	else if (err == 0) {
491 		err = -EREMOTEIO;
492 		goto error;
493 	}
494 
495 	m->wpos += err;
496 	if (m->wpos == m->wsize) {
497 		m->wpos = m->wsize = 0;
498 		p9_req_put(m->wreq);
499 		m->wreq = NULL;
500 	}
501 
502 end_clear:
503 	clear_bit(Wworksched, &m->wsched);
504 
505 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
506 		if (test_and_clear_bit(Wpending, &m->wsched))
507 			n = EPOLLOUT;
508 		else
509 			n = p9_fd_poll(m->client, NULL, NULL);
510 
511 		if ((n & EPOLLOUT) &&
512 		   !test_and_set_bit(Wworksched, &m->wsched)) {
513 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
514 			schedule_work(&m->wq);
515 		}
516 	}
517 
518 	return;
519 
520 error:
521 	p9_conn_cancel(m, err);
522 	clear_bit(Wworksched, &m->wsched);
523 }
524 
p9_pollwake(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)525 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
526 {
527 	struct p9_poll_wait *pwait =
528 		container_of(wait, struct p9_poll_wait, wait);
529 	struct p9_conn *m = pwait->conn;
530 	unsigned long flags;
531 
532 	spin_lock_irqsave(&p9_poll_lock, flags);
533 	if (list_empty(&m->poll_pending_link))
534 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
535 	spin_unlock_irqrestore(&p9_poll_lock, flags);
536 
537 	schedule_work(&p9_poll_work);
538 	return 1;
539 }
540 
541 /**
542  * p9_pollwait - add poll task to the wait queue
543  * @filp: file pointer being polled
544  * @wait_address: wait_q to block on
545  * @p: poll state
546  *
547  * called by files poll operation to add v9fs-poll task to files wait queue
548  */
549 
550 static void
p9_pollwait(struct file * filp,wait_queue_head_t * wait_address,poll_table * p)551 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
552 {
553 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
554 	struct p9_poll_wait *pwait = NULL;
555 	int i;
556 
557 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
558 		if (m->poll_wait[i].wait_addr == NULL) {
559 			pwait = &m->poll_wait[i];
560 			break;
561 		}
562 	}
563 
564 	if (!pwait) {
565 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
566 		return;
567 	}
568 
569 	pwait->conn = m;
570 	pwait->wait_addr = wait_address;
571 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
572 	add_wait_queue(wait_address, &pwait->wait);
573 }
574 
575 /**
576  * p9_conn_create - initialize the per-session mux data
577  * @client: client instance
578  *
579  * Note: Creates the polling task if this is the first session.
580  */
581 
p9_conn_create(struct p9_client * client)582 static void p9_conn_create(struct p9_client *client)
583 {
584 	__poll_t n;
585 	struct p9_trans_fd *ts = client->trans;
586 	struct p9_conn *m = &ts->conn;
587 
588 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
589 
590 	INIT_LIST_HEAD(&m->mux_list);
591 	m->client = client;
592 
593 	INIT_LIST_HEAD(&m->req_list);
594 	INIT_LIST_HEAD(&m->unsent_req_list);
595 	INIT_WORK(&m->rq, p9_read_work);
596 	INIT_WORK(&m->wq, p9_write_work);
597 	INIT_LIST_HEAD(&m->poll_pending_link);
598 	init_poll_funcptr(&m->pt, p9_pollwait);
599 
600 	n = p9_fd_poll(client, &m->pt, NULL);
601 	if (n & EPOLLIN) {
602 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
603 		set_bit(Rpending, &m->wsched);
604 	}
605 
606 	if (n & EPOLLOUT) {
607 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
608 		set_bit(Wpending, &m->wsched);
609 	}
610 }
611 
612 /**
613  * p9_poll_mux - polls a mux and schedules read or write works if necessary
614  * @m: connection to poll
615  *
616  */
617 
p9_poll_mux(struct p9_conn * m)618 static void p9_poll_mux(struct p9_conn *m)
619 {
620 	__poll_t n;
621 	int err = -ECONNRESET;
622 
623 	if (m->err < 0)
624 		return;
625 
626 	n = p9_fd_poll(m->client, NULL, &err);
627 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
628 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
629 		p9_conn_cancel(m, err);
630 	}
631 
632 	if (n & EPOLLIN) {
633 		set_bit(Rpending, &m->wsched);
634 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
635 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
636 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
637 			schedule_work(&m->rq);
638 		}
639 	}
640 
641 	if (n & EPOLLOUT) {
642 		set_bit(Wpending, &m->wsched);
643 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
644 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
645 		    !test_and_set_bit(Wworksched, &m->wsched)) {
646 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
647 			schedule_work(&m->wq);
648 		}
649 	}
650 }
651 
652 /**
653  * p9_fd_request - send 9P request
654  * The function can sleep until the request is scheduled for sending.
655  * The function can be interrupted. Return from the function is not
656  * a guarantee that the request is sent successfully.
657  *
658  * @client: client instance
659  * @req: request to be sent
660  *
661  */
662 
p9_fd_request(struct p9_client * client,struct p9_req_t * req)663 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
664 {
665 	__poll_t n;
666 	struct p9_trans_fd *ts = client->trans;
667 	struct p9_conn *m = &ts->conn;
668 
669 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
670 		 m, current, &req->tc, req->tc.id);
671 	if (m->err < 0)
672 		return m->err;
673 
674 	spin_lock(&client->lock);
675 	req->status = REQ_STATUS_UNSENT;
676 	list_add_tail(&req->req_list, &m->unsent_req_list);
677 	spin_unlock(&client->lock);
678 
679 	if (test_and_clear_bit(Wpending, &m->wsched))
680 		n = EPOLLOUT;
681 	else
682 		n = p9_fd_poll(m->client, NULL, NULL);
683 
684 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
685 		schedule_work(&m->wq);
686 
687 	return 0;
688 }
689 
p9_fd_cancel(struct p9_client * client,struct p9_req_t * req)690 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
691 {
692 	int ret = 1;
693 
694 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
695 
696 	spin_lock(&client->lock);
697 
698 	if (req->status == REQ_STATUS_UNSENT) {
699 		list_del(&req->req_list);
700 		req->status = REQ_STATUS_FLSHD;
701 		p9_req_put(req);
702 		ret = 0;
703 	}
704 	spin_unlock(&client->lock);
705 
706 	return ret;
707 }
708 
p9_fd_cancelled(struct p9_client * client,struct p9_req_t * req)709 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
710 {
711 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
712 
713 	spin_lock(&client->lock);
714 	/* Ignore cancelled request if message has been received
715 	 * before lock.
716 	 */
717 	if (req->status == REQ_STATUS_RCVD) {
718 		spin_unlock(&client->lock);
719 		return 0;
720 	}
721 
722 	/* we haven't received a response for oldreq,
723 	 * remove it from the list.
724 	 */
725 	list_del(&req->req_list);
726 	req->status = REQ_STATUS_FLSHD;
727 	spin_unlock(&client->lock);
728 	p9_req_put(req);
729 
730 	return 0;
731 }
732 
p9_fd_show_options(struct seq_file * m,struct p9_client * clnt)733 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
734 {
735 	if (clnt->trans_mod == &p9_tcp_trans) {
736 		if (clnt->trans_opts.tcp.port != P9_PORT)
737 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
738 	} else if (clnt->trans_mod == &p9_fd_trans) {
739 		if (clnt->trans_opts.fd.rfd != ~0)
740 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
741 		if (clnt->trans_opts.fd.wfd != ~0)
742 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
743 	}
744 	return 0;
745 }
746 
747 /**
748  * parse_opts - parse mount options into p9_fd_opts structure
749  * @params: options string passed from mount
750  * @opts: fd transport-specific structure to parse options into
751  *
752  * Returns 0 upon success, -ERRNO upon failure
753  */
754 
parse_opts(char * params,struct p9_fd_opts * opts)755 static int parse_opts(char *params, struct p9_fd_opts *opts)
756 {
757 	char *p;
758 	substring_t args[MAX_OPT_ARGS];
759 	int option;
760 	char *options, *tmp_options;
761 
762 	opts->port = P9_PORT;
763 	opts->rfd = ~0;
764 	opts->wfd = ~0;
765 	opts->privport = false;
766 
767 	if (!params)
768 		return 0;
769 
770 	tmp_options = kstrdup(params, GFP_KERNEL);
771 	if (!tmp_options) {
772 		p9_debug(P9_DEBUG_ERROR,
773 			 "failed to allocate copy of option string\n");
774 		return -ENOMEM;
775 	}
776 	options = tmp_options;
777 
778 	while ((p = strsep(&options, ",")) != NULL) {
779 		int token;
780 		int r;
781 		if (!*p)
782 			continue;
783 		token = match_token(p, tokens, args);
784 		if ((token != Opt_err) && (token != Opt_privport)) {
785 			r = match_int(&args[0], &option);
786 			if (r < 0) {
787 				p9_debug(P9_DEBUG_ERROR,
788 					 "integer field, but no integer?\n");
789 				continue;
790 			}
791 		}
792 		switch (token) {
793 		case Opt_port:
794 			opts->port = option;
795 			break;
796 		case Opt_rfdno:
797 			opts->rfd = option;
798 			break;
799 		case Opt_wfdno:
800 			opts->wfd = option;
801 			break;
802 		case Opt_privport:
803 			opts->privport = true;
804 			break;
805 		default:
806 			continue;
807 		}
808 	}
809 
810 	kfree(tmp_options);
811 	return 0;
812 }
813 
p9_fd_open(struct p9_client * client,int rfd,int wfd)814 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
815 {
816 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
817 					   GFP_KERNEL);
818 	if (!ts)
819 		return -ENOMEM;
820 
821 	ts->rd = fget(rfd);
822 	if (!ts->rd)
823 		goto out_free_ts;
824 	if (!(ts->rd->f_mode & FMODE_READ))
825 		goto out_put_rd;
826 	/* prevent workers from hanging on IO when fd is a pipe */
827 	ts->rd->f_flags |= O_NONBLOCK;
828 	ts->wr = fget(wfd);
829 	if (!ts->wr)
830 		goto out_put_rd;
831 	if (!(ts->wr->f_mode & FMODE_WRITE))
832 		goto out_put_wr;
833 	ts->wr->f_flags |= O_NONBLOCK;
834 
835 	client->trans = ts;
836 	client->status = Connected;
837 
838 	return 0;
839 
840 out_put_wr:
841 	fput(ts->wr);
842 out_put_rd:
843 	fput(ts->rd);
844 out_free_ts:
845 	kfree(ts);
846 	return -EIO;
847 }
848 
p9_socket_open(struct p9_client * client,struct socket * csocket)849 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
850 {
851 	struct p9_trans_fd *p;
852 	struct file *file;
853 
854 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
855 	if (!p) {
856 		sock_release(csocket);
857 		return -ENOMEM;
858 	}
859 
860 	csocket->sk->sk_allocation = GFP_NOIO;
861 	file = sock_alloc_file(csocket, 0, NULL);
862 	if (IS_ERR(file)) {
863 		pr_err("%s (%d): failed to map fd\n",
864 		       __func__, task_pid_nr(current));
865 		kfree(p);
866 		return PTR_ERR(file);
867 	}
868 
869 	get_file(file);
870 	p->wr = p->rd = file;
871 	client->trans = p;
872 	client->status = Connected;
873 
874 	p->rd->f_flags |= O_NONBLOCK;
875 
876 	p9_conn_create(client);
877 	return 0;
878 }
879 
880 /**
881  * p9_mux_destroy - cancels all pending requests of mux
882  * @m: mux to destroy
883  *
884  */
885 
p9_conn_destroy(struct p9_conn * m)886 static void p9_conn_destroy(struct p9_conn *m)
887 {
888 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
889 		 m, m->mux_list.prev, m->mux_list.next);
890 
891 	p9_mux_poll_stop(m);
892 	cancel_work_sync(&m->rq);
893 	if (m->rreq) {
894 		p9_req_put(m->rreq);
895 		m->rreq = NULL;
896 	}
897 	cancel_work_sync(&m->wq);
898 	if (m->wreq) {
899 		p9_req_put(m->wreq);
900 		m->wreq = NULL;
901 	}
902 
903 	p9_conn_cancel(m, -ECONNRESET);
904 
905 	m->client = NULL;
906 }
907 
908 /**
909  * p9_fd_close - shutdown file descriptor transport
910  * @client: client instance
911  *
912  */
913 
p9_fd_close(struct p9_client * client)914 static void p9_fd_close(struct p9_client *client)
915 {
916 	struct p9_trans_fd *ts;
917 
918 	if (!client)
919 		return;
920 
921 	ts = client->trans;
922 	if (!ts)
923 		return;
924 
925 	client->status = Disconnected;
926 
927 	p9_conn_destroy(&ts->conn);
928 
929 	if (ts->rd)
930 		fput(ts->rd);
931 	if (ts->wr)
932 		fput(ts->wr);
933 
934 	kfree(ts);
935 }
936 
937 /*
938  * stolen from NFS - maybe should be made a generic function?
939  */
valid_ipaddr4(const char * buf)940 static inline int valid_ipaddr4(const char *buf)
941 {
942 	int rc, count, in[4];
943 
944 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
945 	if (rc != 4)
946 		return -EINVAL;
947 	for (count = 0; count < 4; count++) {
948 		if (in[count] > 255)
949 			return -EINVAL;
950 	}
951 	return 0;
952 }
953 
p9_bind_privport(struct socket * sock)954 static int p9_bind_privport(struct socket *sock)
955 {
956 	struct sockaddr_in cl;
957 	int port, err = -EINVAL;
958 
959 	memset(&cl, 0, sizeof(cl));
960 	cl.sin_family = AF_INET;
961 	cl.sin_addr.s_addr = INADDR_ANY;
962 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
963 		cl.sin_port = htons((ushort)port);
964 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
965 		if (err != -EADDRINUSE)
966 			break;
967 	}
968 	return err;
969 }
970 
971 
972 static int
p9_fd_create_tcp(struct p9_client * client,const char * addr,char * args)973 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
974 {
975 	int err;
976 	struct socket *csocket;
977 	struct sockaddr_in sin_server;
978 	struct p9_fd_opts opts;
979 
980 	err = parse_opts(args, &opts);
981 	if (err < 0)
982 		return err;
983 
984 	if (addr == NULL || valid_ipaddr4(addr) < 0)
985 		return -EINVAL;
986 
987 	csocket = NULL;
988 
989 	client->trans_opts.tcp.port = opts.port;
990 	client->trans_opts.tcp.privport = opts.privport;
991 	sin_server.sin_family = AF_INET;
992 	sin_server.sin_addr.s_addr = in_aton(addr);
993 	sin_server.sin_port = htons(opts.port);
994 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
995 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
996 	if (err) {
997 		pr_err("%s (%d): problem creating socket\n",
998 		       __func__, task_pid_nr(current));
999 		return err;
1000 	}
1001 
1002 	if (opts.privport) {
1003 		err = p9_bind_privport(csocket);
1004 		if (err < 0) {
1005 			pr_err("%s (%d): problem binding to privport\n",
1006 			       __func__, task_pid_nr(current));
1007 			sock_release(csocket);
1008 			return err;
1009 		}
1010 	}
1011 
1012 	err = csocket->ops->connect(csocket,
1013 				    (struct sockaddr *)&sin_server,
1014 				    sizeof(struct sockaddr_in), 0);
1015 	if (err < 0) {
1016 		pr_err("%s (%d): problem connecting socket to %s\n",
1017 		       __func__, task_pid_nr(current), addr);
1018 		sock_release(csocket);
1019 		return err;
1020 	}
1021 
1022 	return p9_socket_open(client, csocket);
1023 }
1024 
1025 static int
p9_fd_create_unix(struct p9_client * client,const char * addr,char * args)1026 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1027 {
1028 	int err;
1029 	struct socket *csocket;
1030 	struct sockaddr_un sun_server;
1031 
1032 	csocket = NULL;
1033 
1034 	if (!addr || !strlen(addr))
1035 		return -EINVAL;
1036 
1037 	if (strlen(addr) >= UNIX_PATH_MAX) {
1038 		pr_err("%s (%d): address too long: %s\n",
1039 		       __func__, task_pid_nr(current), addr);
1040 		return -ENAMETOOLONG;
1041 	}
1042 
1043 	sun_server.sun_family = PF_UNIX;
1044 	strcpy(sun_server.sun_path, addr);
1045 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1046 			    SOCK_STREAM, 0, &csocket, 1);
1047 	if (err < 0) {
1048 		pr_err("%s (%d): problem creating socket\n",
1049 		       __func__, task_pid_nr(current));
1050 
1051 		return err;
1052 	}
1053 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1054 			sizeof(struct sockaddr_un) - 1, 0);
1055 	if (err < 0) {
1056 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1057 		       __func__, task_pid_nr(current), addr, err);
1058 		sock_release(csocket);
1059 		return err;
1060 	}
1061 
1062 	return p9_socket_open(client, csocket);
1063 }
1064 
1065 static int
p9_fd_create(struct p9_client * client,const char * addr,char * args)1066 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1067 {
1068 	int err;
1069 	struct p9_fd_opts opts;
1070 
1071 	parse_opts(args, &opts);
1072 	client->trans_opts.fd.rfd = opts.rfd;
1073 	client->trans_opts.fd.wfd = opts.wfd;
1074 
1075 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1076 		pr_err("Insufficient options for proto=fd\n");
1077 		return -ENOPROTOOPT;
1078 	}
1079 
1080 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1081 	if (err < 0)
1082 		return err;
1083 
1084 	p9_conn_create(client);
1085 
1086 	return 0;
1087 }
1088 
1089 static struct p9_trans_module p9_tcp_trans = {
1090 	.name = "tcp",
1091 	.maxsize = MAX_SOCK_BUF,
1092 	.def = 0,
1093 	.create = p9_fd_create_tcp,
1094 	.close = p9_fd_close,
1095 	.request = p9_fd_request,
1096 	.cancel = p9_fd_cancel,
1097 	.cancelled = p9_fd_cancelled,
1098 	.show_options = p9_fd_show_options,
1099 	.owner = THIS_MODULE,
1100 };
1101 
1102 static struct p9_trans_module p9_unix_trans = {
1103 	.name = "unix",
1104 	.maxsize = MAX_SOCK_BUF,
1105 	.def = 0,
1106 	.create = p9_fd_create_unix,
1107 	.close = p9_fd_close,
1108 	.request = p9_fd_request,
1109 	.cancel = p9_fd_cancel,
1110 	.cancelled = p9_fd_cancelled,
1111 	.show_options = p9_fd_show_options,
1112 	.owner = THIS_MODULE,
1113 };
1114 
1115 static struct p9_trans_module p9_fd_trans = {
1116 	.name = "fd",
1117 	.maxsize = MAX_SOCK_BUF,
1118 	.def = 0,
1119 	.create = p9_fd_create,
1120 	.close = p9_fd_close,
1121 	.request = p9_fd_request,
1122 	.cancel = p9_fd_cancel,
1123 	.cancelled = p9_fd_cancelled,
1124 	.show_options = p9_fd_show_options,
1125 	.owner = THIS_MODULE,
1126 };
1127 
1128 /**
1129  * p9_poll_workfn - poll worker thread
1130  * @work: work queue
1131  *
1132  * polls all v9fs transports for new events and queues the appropriate
1133  * work to the work queue
1134  *
1135  */
1136 
p9_poll_workfn(struct work_struct * work)1137 static void p9_poll_workfn(struct work_struct *work)
1138 {
1139 	unsigned long flags;
1140 
1141 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1142 
1143 	spin_lock_irqsave(&p9_poll_lock, flags);
1144 	while (!list_empty(&p9_poll_pending_list)) {
1145 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1146 							struct p9_conn,
1147 							poll_pending_link);
1148 		list_del_init(&conn->poll_pending_link);
1149 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1150 
1151 		p9_poll_mux(conn);
1152 
1153 		spin_lock_irqsave(&p9_poll_lock, flags);
1154 	}
1155 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1156 
1157 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1158 }
1159 
p9_trans_fd_init(void)1160 int p9_trans_fd_init(void)
1161 {
1162 	v9fs_register_trans(&p9_tcp_trans);
1163 	v9fs_register_trans(&p9_unix_trans);
1164 	v9fs_register_trans(&p9_fd_trans);
1165 
1166 	return 0;
1167 }
1168 
p9_trans_fd_exit(void)1169 void p9_trans_fd_exit(void)
1170 {
1171 	flush_work(&p9_poll_work);
1172 	v9fs_unregister_trans(&p9_tcp_trans);
1173 	v9fs_unregister_trans(&p9_unix_trans);
1174 	v9fs_unregister_trans(&p9_fd_trans);
1175 }
1176