• 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 (1024*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  * @privport: port is privileged
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_lock: lock protecting req_list and requests statuses
97  * @req_list: accounting for requests which have been sent
98  * @unsent_req_list: accounting for requests that haven't been sent
99  * @rreq: read request
100  * @wreq: write request
101  * @req: current request being processed (if any)
102  * @tmp_buf: temporary buffer to read in header
103  * @rc: temporary fcall for reading current frame
104  * @wpos: write position for current frame
105  * @wsize: amount of data to write for current frame
106  * @wbuf: current write buffer
107  * @poll_pending_link: pending links to be polled per conn
108  * @poll_wait: array of wait_q's for various worker threads
109  * @pt: poll state
110  * @rq: current read work
111  * @wq: current write work
112  * @wsched: ????
113  *
114  */
115 
116 struct p9_conn {
117 	struct list_head mux_list;
118 	struct p9_client *client;
119 	int err;
120 	spinlock_t req_lock;
121 	struct list_head req_list;
122 	struct list_head unsent_req_list;
123 	struct p9_req_t *rreq;
124 	struct p9_req_t *wreq;
125 	char tmp_buf[P9_HDRSZ];
126 	struct p9_fcall rc;
127 	int wpos;
128 	int wsize;
129 	char *wbuf;
130 	struct list_head poll_pending_link;
131 	struct p9_poll_wait poll_wait[MAXPOLLWADDR];
132 	poll_table pt;
133 	struct work_struct rq;
134 	struct work_struct wq;
135 	unsigned long wsched;
136 };
137 
138 /**
139  * struct p9_trans_fd - transport state
140  * @rd: reference to file to read from
141  * @wr: reference of file to write to
142  * @conn: connection state reference
143  *
144  */
145 
146 struct p9_trans_fd {
147 	struct file *rd;
148 	struct file *wr;
149 	struct p9_conn conn;
150 };
151 
152 static void p9_poll_workfn(struct work_struct *work);
153 
154 static DEFINE_SPINLOCK(p9_poll_lock);
155 static LIST_HEAD(p9_poll_pending_list);
156 static DECLARE_WORK(p9_poll_work, p9_poll_workfn);
157 
158 static unsigned int p9_ipport_resv_min = P9_DEF_MIN_RESVPORT;
159 static unsigned int p9_ipport_resv_max = P9_DEF_MAX_RESVPORT;
160 
p9_mux_poll_stop(struct p9_conn * m)161 static void p9_mux_poll_stop(struct p9_conn *m)
162 {
163 	unsigned long flags;
164 	int i;
165 
166 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
167 		struct p9_poll_wait *pwait = &m->poll_wait[i];
168 
169 		if (pwait->wait_addr) {
170 			remove_wait_queue(pwait->wait_addr, &pwait->wait);
171 			pwait->wait_addr = NULL;
172 		}
173 	}
174 
175 	spin_lock_irqsave(&p9_poll_lock, flags);
176 	list_del_init(&m->poll_pending_link);
177 	spin_unlock_irqrestore(&p9_poll_lock, flags);
178 
179 	flush_work(&p9_poll_work);
180 }
181 
182 /**
183  * p9_conn_cancel - cancel all pending requests with error
184  * @m: mux data
185  * @err: error code
186  *
187  */
188 
p9_conn_cancel(struct p9_conn * m,int err)189 static void p9_conn_cancel(struct p9_conn *m, int err)
190 {
191 	struct p9_req_t *req, *rtmp;
192 	LIST_HEAD(cancel_list);
193 
194 	p9_debug(P9_DEBUG_ERROR, "mux %p err %d\n", m, err);
195 
196 	spin_lock(&m->req_lock);
197 
198 	if (m->err) {
199 		spin_unlock(&m->req_lock);
200 		return;
201 	}
202 
203 	m->err = err;
204 
205 	list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
206 		list_move(&req->req_list, &cancel_list);
207 		req->status = REQ_STATUS_ERROR;
208 	}
209 	list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
210 		list_move(&req->req_list, &cancel_list);
211 		req->status = REQ_STATUS_ERROR;
212 	}
213 
214 	spin_unlock(&m->req_lock);
215 
216 	list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
217 		p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
218 		list_del(&req->req_list);
219 		if (!req->t_err)
220 			req->t_err = err;
221 		p9_client_cb(m->client, req, REQ_STATUS_ERROR);
222 	}
223 }
224 
225 static __poll_t
p9_fd_poll(struct p9_client * client,struct poll_table_struct * pt,int * err)226 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
227 {
228 	__poll_t ret;
229 	struct p9_trans_fd *ts = NULL;
230 
231 	if (client && client->status == Connected)
232 		ts = client->trans;
233 
234 	if (!ts) {
235 		if (err)
236 			*err = -EREMOTEIO;
237 		return EPOLLERR;
238 	}
239 
240 	ret = vfs_poll(ts->rd, pt);
241 	if (ts->rd != ts->wr)
242 		ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
243 	return ret;
244 }
245 
246 /**
247  * p9_fd_read- read from a fd
248  * @client: client instance
249  * @v: buffer to receive data into
250  * @len: size of receive buffer
251  *
252  */
253 
p9_fd_read(struct p9_client * client,void * v,int len)254 static int p9_fd_read(struct p9_client *client, void *v, int len)
255 {
256 	int ret;
257 	struct p9_trans_fd *ts = NULL;
258 	loff_t pos;
259 
260 	if (client && client->status != Disconnected)
261 		ts = client->trans;
262 
263 	if (!ts)
264 		return -EREMOTEIO;
265 
266 	if (!(ts->rd->f_flags & O_NONBLOCK))
267 		p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
268 
269 	pos = ts->rd->f_pos;
270 	ret = kernel_read(ts->rd, v, len, &pos);
271 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
272 		client->status = Disconnected;
273 	return ret;
274 }
275 
276 /**
277  * p9_read_work - called when there is some data to be read from a transport
278  * @work: container of work to be done
279  *
280  */
281 
p9_read_work(struct work_struct * work)282 static void p9_read_work(struct work_struct *work)
283 {
284 	__poll_t n;
285 	int err;
286 	struct p9_conn *m;
287 
288 	m = container_of(work, struct p9_conn, rq);
289 
290 	if (m->err < 0)
291 		return;
292 
293 	p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
294 
295 	if (!m->rc.sdata) {
296 		m->rc.sdata = m->tmp_buf;
297 		m->rc.offset = 0;
298 		m->rc.capacity = P9_HDRSZ; /* start by reading header */
299 	}
300 
301 	clear_bit(Rpending, &m->wsched);
302 	p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
303 		 m, m->rc.offset, m->rc.capacity,
304 		 m->rc.capacity - m->rc.offset);
305 	err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
306 			 m->rc.capacity - m->rc.offset);
307 	p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
308 	if (err == -EAGAIN)
309 		goto end_clear;
310 
311 	if (err <= 0)
312 		goto error;
313 
314 	m->rc.offset += err;
315 
316 	/* header read in */
317 	if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
318 		p9_debug(P9_DEBUG_TRANS, "got new header\n");
319 
320 		/* Header size */
321 		m->rc.size = P9_HDRSZ;
322 		err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
323 		if (err) {
324 			p9_debug(P9_DEBUG_ERROR,
325 				 "error parsing header: %d\n", err);
326 			goto error;
327 		}
328 
329 		if (m->rc.size >= m->client->msize) {
330 			p9_debug(P9_DEBUG_ERROR,
331 				 "requested packet size too big: %d\n",
332 				 m->rc.size);
333 			err = -EIO;
334 			goto error;
335 		}
336 
337 		p9_debug(P9_DEBUG_TRANS,
338 			 "mux %p pkt: size: %d bytes tag: %d\n",
339 			 m, m->rc.size, m->rc.tag);
340 
341 		m->rreq = p9_tag_lookup(m->client, m->rc.tag);
342 		if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
343 			p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
344 				 m->rc.tag);
345 			err = -EIO;
346 			goto error;
347 		}
348 
349 		if (!m->rreq->rc.sdata) {
350 			p9_debug(P9_DEBUG_ERROR,
351 				 "No recv fcall for tag %d (req %p), disconnecting!\n",
352 				 m->rc.tag, m->rreq);
353 			p9_req_put(m->client, m->rreq);
354 			m->rreq = NULL;
355 			err = -EIO;
356 			goto error;
357 		}
358 		m->rc.sdata = m->rreq->rc.sdata;
359 		memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
360 		m->rc.capacity = m->rc.size;
361 	}
362 
363 	/* packet is read in
364 	 * not an else because some packets (like clunk) have no payload
365 	 */
366 	if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
367 		p9_debug(P9_DEBUG_TRANS, "got new packet\n");
368 		m->rreq->rc.size = m->rc.offset;
369 		spin_lock(&m->req_lock);
370 		if (m->rreq->status == REQ_STATUS_SENT) {
371 			list_del(&m->rreq->req_list);
372 			p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
373 		} else if (m->rreq->status == REQ_STATUS_FLSHD) {
374 			/* Ignore replies associated with a cancelled request. */
375 			p9_debug(P9_DEBUG_TRANS,
376 				 "Ignore replies associated with a cancelled request\n");
377 		} else {
378 			spin_unlock(&m->req_lock);
379 			p9_debug(P9_DEBUG_ERROR,
380 				 "Request tag %d errored out while we were reading the reply\n",
381 				 m->rc.tag);
382 			err = -EIO;
383 			goto error;
384 		}
385 		spin_unlock(&m->req_lock);
386 		m->rc.sdata = NULL;
387 		m->rc.offset = 0;
388 		m->rc.capacity = 0;
389 		p9_req_put(m->client, m->rreq);
390 		m->rreq = NULL;
391 	}
392 
393 end_clear:
394 	clear_bit(Rworksched, &m->wsched);
395 
396 	if (!list_empty(&m->req_list)) {
397 		if (test_and_clear_bit(Rpending, &m->wsched))
398 			n = EPOLLIN;
399 		else
400 			n = p9_fd_poll(m->client, NULL, NULL);
401 
402 		if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
403 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
404 			schedule_work(&m->rq);
405 		}
406 	}
407 
408 	return;
409 error:
410 	p9_conn_cancel(m, err);
411 	clear_bit(Rworksched, &m->wsched);
412 }
413 
414 /**
415  * p9_fd_write - write to a socket
416  * @client: client instance
417  * @v: buffer to send data from
418  * @len: size of send buffer
419  *
420  */
421 
p9_fd_write(struct p9_client * client,void * v,int len)422 static int p9_fd_write(struct p9_client *client, void *v, int len)
423 {
424 	ssize_t ret;
425 	struct p9_trans_fd *ts = NULL;
426 
427 	if (client && client->status != Disconnected)
428 		ts = client->trans;
429 
430 	if (!ts)
431 		return -EREMOTEIO;
432 
433 	if (!(ts->wr->f_flags & O_NONBLOCK))
434 		p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
435 
436 	ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
437 	if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
438 		client->status = Disconnected;
439 	return ret;
440 }
441 
442 /**
443  * p9_write_work - called when a transport can send some data
444  * @work: container for work to be done
445  *
446  */
447 
p9_write_work(struct work_struct * work)448 static void p9_write_work(struct work_struct *work)
449 {
450 	__poll_t n;
451 	int err;
452 	struct p9_conn *m;
453 	struct p9_req_t *req;
454 
455 	m = container_of(work, struct p9_conn, wq);
456 
457 	if (m->err < 0) {
458 		clear_bit(Wworksched, &m->wsched);
459 		return;
460 	}
461 
462 	if (!m->wsize) {
463 		spin_lock(&m->req_lock);
464 		if (list_empty(&m->unsent_req_list)) {
465 			clear_bit(Wworksched, &m->wsched);
466 			spin_unlock(&m->req_lock);
467 			return;
468 		}
469 
470 		req = list_entry(m->unsent_req_list.next, struct p9_req_t,
471 			       req_list);
472 		req->status = REQ_STATUS_SENT;
473 		p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
474 		list_move_tail(&req->req_list, &m->req_list);
475 
476 		m->wbuf = req->tc.sdata;
477 		m->wsize = req->tc.size;
478 		m->wpos = 0;
479 		p9_req_get(req);
480 		m->wreq = req;
481 		spin_unlock(&m->req_lock);
482 	}
483 
484 	p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
485 		 m, m->wpos, m->wsize);
486 	clear_bit(Wpending, &m->wsched);
487 	err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
488 	p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
489 	if (err == -EAGAIN)
490 		goto end_clear;
491 
492 
493 	if (err < 0)
494 		goto error;
495 	else if (err == 0) {
496 		err = -EREMOTEIO;
497 		goto error;
498 	}
499 
500 	m->wpos += err;
501 	if (m->wpos == m->wsize) {
502 		m->wpos = m->wsize = 0;
503 		p9_req_put(m->client, m->wreq);
504 		m->wreq = NULL;
505 	}
506 
507 end_clear:
508 	clear_bit(Wworksched, &m->wsched);
509 
510 	if (m->wsize || !list_empty(&m->unsent_req_list)) {
511 		if (test_and_clear_bit(Wpending, &m->wsched))
512 			n = EPOLLOUT;
513 		else
514 			n = p9_fd_poll(m->client, NULL, NULL);
515 
516 		if ((n & EPOLLOUT) &&
517 		   !test_and_set_bit(Wworksched, &m->wsched)) {
518 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
519 			schedule_work(&m->wq);
520 		}
521 	}
522 
523 	return;
524 
525 error:
526 	p9_conn_cancel(m, err);
527 	clear_bit(Wworksched, &m->wsched);
528 }
529 
p9_pollwake(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)530 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
531 {
532 	struct p9_poll_wait *pwait =
533 		container_of(wait, struct p9_poll_wait, wait);
534 	struct p9_conn *m = pwait->conn;
535 	unsigned long flags;
536 
537 	spin_lock_irqsave(&p9_poll_lock, flags);
538 	if (list_empty(&m->poll_pending_link))
539 		list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
540 	spin_unlock_irqrestore(&p9_poll_lock, flags);
541 
542 	schedule_work(&p9_poll_work);
543 	return 1;
544 }
545 
546 /**
547  * p9_pollwait - add poll task to the wait queue
548  * @filp: file pointer being polled
549  * @wait_address: wait_q to block on
550  * @p: poll state
551  *
552  * called by files poll operation to add v9fs-poll task to files wait queue
553  */
554 
555 static void
p9_pollwait(struct file * filp,wait_queue_head_t * wait_address,poll_table * p)556 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
557 {
558 	struct p9_conn *m = container_of(p, struct p9_conn, pt);
559 	struct p9_poll_wait *pwait = NULL;
560 	int i;
561 
562 	for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
563 		if (m->poll_wait[i].wait_addr == NULL) {
564 			pwait = &m->poll_wait[i];
565 			break;
566 		}
567 	}
568 
569 	if (!pwait) {
570 		p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
571 		return;
572 	}
573 
574 	pwait->conn = m;
575 	pwait->wait_addr = wait_address;
576 	init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
577 	add_wait_queue(wait_address, &pwait->wait);
578 }
579 
580 /**
581  * p9_conn_create - initialize the per-session mux data
582  * @client: client instance
583  *
584  * Note: Creates the polling task if this is the first session.
585  */
586 
p9_conn_create(struct p9_client * client)587 static void p9_conn_create(struct p9_client *client)
588 {
589 	__poll_t n;
590 	struct p9_trans_fd *ts = client->trans;
591 	struct p9_conn *m = &ts->conn;
592 
593 	p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
594 
595 	INIT_LIST_HEAD(&m->mux_list);
596 	m->client = client;
597 
598 	spin_lock_init(&m->req_lock);
599 	INIT_LIST_HEAD(&m->req_list);
600 	INIT_LIST_HEAD(&m->unsent_req_list);
601 	INIT_WORK(&m->rq, p9_read_work);
602 	INIT_WORK(&m->wq, p9_write_work);
603 	INIT_LIST_HEAD(&m->poll_pending_link);
604 	init_poll_funcptr(&m->pt, p9_pollwait);
605 
606 	n = p9_fd_poll(client, &m->pt, NULL);
607 	if (n & EPOLLIN) {
608 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
609 		set_bit(Rpending, &m->wsched);
610 	}
611 
612 	if (n & EPOLLOUT) {
613 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
614 		set_bit(Wpending, &m->wsched);
615 	}
616 }
617 
618 /**
619  * p9_poll_mux - polls a mux and schedules read or write works if necessary
620  * @m: connection to poll
621  *
622  */
623 
p9_poll_mux(struct p9_conn * m)624 static void p9_poll_mux(struct p9_conn *m)
625 {
626 	__poll_t n;
627 	int err = -ECONNRESET;
628 
629 	if (m->err < 0)
630 		return;
631 
632 	n = p9_fd_poll(m->client, NULL, &err);
633 	if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
634 		p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
635 		p9_conn_cancel(m, err);
636 	}
637 
638 	if (n & EPOLLIN) {
639 		set_bit(Rpending, &m->wsched);
640 		p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
641 		if (!test_and_set_bit(Rworksched, &m->wsched)) {
642 			p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
643 			schedule_work(&m->rq);
644 		}
645 	}
646 
647 	if (n & EPOLLOUT) {
648 		set_bit(Wpending, &m->wsched);
649 		p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
650 		if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
651 		    !test_and_set_bit(Wworksched, &m->wsched)) {
652 			p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
653 			schedule_work(&m->wq);
654 		}
655 	}
656 }
657 
658 /**
659  * p9_fd_request - send 9P request
660  * The function can sleep until the request is scheduled for sending.
661  * The function can be interrupted. Return from the function is not
662  * a guarantee that the request is sent successfully.
663  *
664  * @client: client instance
665  * @req: request to be sent
666  *
667  */
668 
p9_fd_request(struct p9_client * client,struct p9_req_t * req)669 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
670 {
671 	__poll_t n;
672 	struct p9_trans_fd *ts = client->trans;
673 	struct p9_conn *m = &ts->conn;
674 
675 	p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
676 		 m, current, &req->tc, req->tc.id);
677 	if (m->err < 0)
678 		return m->err;
679 
680 	spin_lock(&m->req_lock);
681 	req->status = REQ_STATUS_UNSENT;
682 	list_add_tail(&req->req_list, &m->unsent_req_list);
683 	spin_unlock(&m->req_lock);
684 
685 	if (test_and_clear_bit(Wpending, &m->wsched))
686 		n = EPOLLOUT;
687 	else
688 		n = p9_fd_poll(m->client, NULL, NULL);
689 
690 	if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
691 		schedule_work(&m->wq);
692 
693 	return 0;
694 }
695 
p9_fd_cancel(struct p9_client * client,struct p9_req_t * req)696 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
697 {
698 	struct p9_trans_fd *ts = client->trans;
699 	struct p9_conn *m = &ts->conn;
700 	int ret = 1;
701 
702 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
703 
704 	spin_lock(&m->req_lock);
705 
706 	if (req->status == REQ_STATUS_UNSENT) {
707 		list_del(&req->req_list);
708 		req->status = REQ_STATUS_FLSHD;
709 		p9_req_put(client, req);
710 		ret = 0;
711 	}
712 	spin_unlock(&m->req_lock);
713 
714 	return ret;
715 }
716 
p9_fd_cancelled(struct p9_client * client,struct p9_req_t * req)717 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
718 {
719 	struct p9_trans_fd *ts = client->trans;
720 	struct p9_conn *m = &ts->conn;
721 
722 	p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
723 
724 	spin_lock(&m->req_lock);
725 	/* Ignore cancelled request if message has been received
726 	 * before lock.
727 	 */
728 	if (req->status == REQ_STATUS_RCVD) {
729 		spin_unlock(&m->req_lock);
730 		return 0;
731 	}
732 
733 	/* we haven't received a response for oldreq,
734 	 * remove it from the list.
735 	 */
736 	list_del(&req->req_list);
737 	req->status = REQ_STATUS_FLSHD;
738 	spin_unlock(&m->req_lock);
739 
740 	p9_req_put(client, req);
741 
742 	return 0;
743 }
744 
p9_fd_show_options(struct seq_file * m,struct p9_client * clnt)745 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
746 {
747 	if (clnt->trans_mod == &p9_tcp_trans) {
748 		if (clnt->trans_opts.tcp.port != P9_PORT)
749 			seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
750 	} else if (clnt->trans_mod == &p9_fd_trans) {
751 		if (clnt->trans_opts.fd.rfd != ~0)
752 			seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
753 		if (clnt->trans_opts.fd.wfd != ~0)
754 			seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
755 	}
756 	return 0;
757 }
758 
759 /**
760  * parse_opts - parse mount options into p9_fd_opts structure
761  * @params: options string passed from mount
762  * @opts: fd transport-specific structure to parse options into
763  *
764  * Returns 0 upon success, -ERRNO upon failure
765  */
766 
parse_opts(char * params,struct p9_fd_opts * opts)767 static int parse_opts(char *params, struct p9_fd_opts *opts)
768 {
769 	char *p;
770 	substring_t args[MAX_OPT_ARGS];
771 	int option;
772 	char *options, *tmp_options;
773 
774 	opts->port = P9_PORT;
775 	opts->rfd = ~0;
776 	opts->wfd = ~0;
777 	opts->privport = false;
778 
779 	if (!params)
780 		return 0;
781 
782 	tmp_options = kstrdup(params, GFP_KERNEL);
783 	if (!tmp_options) {
784 		p9_debug(P9_DEBUG_ERROR,
785 			 "failed to allocate copy of option string\n");
786 		return -ENOMEM;
787 	}
788 	options = tmp_options;
789 
790 	while ((p = strsep(&options, ",")) != NULL) {
791 		int token;
792 		int r;
793 		if (!*p)
794 			continue;
795 		token = match_token(p, tokens, args);
796 		if ((token != Opt_err) && (token != Opt_privport)) {
797 			r = match_int(&args[0], &option);
798 			if (r < 0) {
799 				p9_debug(P9_DEBUG_ERROR,
800 					 "integer field, but no integer?\n");
801 				continue;
802 			}
803 		}
804 		switch (token) {
805 		case Opt_port:
806 			opts->port = option;
807 			break;
808 		case Opt_rfdno:
809 			opts->rfd = option;
810 			break;
811 		case Opt_wfdno:
812 			opts->wfd = option;
813 			break;
814 		case Opt_privport:
815 			opts->privport = true;
816 			break;
817 		default:
818 			continue;
819 		}
820 	}
821 
822 	kfree(tmp_options);
823 	return 0;
824 }
825 
p9_fd_open(struct p9_client * client,int rfd,int wfd)826 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
827 {
828 	struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
829 					   GFP_KERNEL);
830 	if (!ts)
831 		return -ENOMEM;
832 
833 	ts->rd = fget(rfd);
834 	if (!ts->rd)
835 		goto out_free_ts;
836 	if (!(ts->rd->f_mode & FMODE_READ))
837 		goto out_put_rd;
838 	/* Prevent workers from hanging on IO when fd is a pipe.
839 	 * It's technically possible for userspace or concurrent mounts to
840 	 * modify this flag concurrently, which will likely result in a
841 	 * broken filesystem. However, just having bad flags here should
842 	 * not crash the kernel or cause any other sort of bug, so mark this
843 	 * particular data race as intentional so that tooling (like KCSAN)
844 	 * can allow it and detect further problems.
845 	 */
846 	data_race(ts->rd->f_flags |= O_NONBLOCK);
847 	ts->wr = fget(wfd);
848 	if (!ts->wr)
849 		goto out_put_rd;
850 	if (!(ts->wr->f_mode & FMODE_WRITE))
851 		goto out_put_wr;
852 	data_race(ts->wr->f_flags |= O_NONBLOCK);
853 
854 	client->trans = ts;
855 	client->status = Connected;
856 
857 	return 0;
858 
859 out_put_wr:
860 	fput(ts->wr);
861 out_put_rd:
862 	fput(ts->rd);
863 out_free_ts:
864 	kfree(ts);
865 	return -EIO;
866 }
867 
p9_socket_open(struct p9_client * client,struct socket * csocket)868 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
869 {
870 	struct p9_trans_fd *p;
871 	struct file *file;
872 
873 	p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
874 	if (!p) {
875 		sock_release(csocket);
876 		return -ENOMEM;
877 	}
878 
879 	csocket->sk->sk_allocation = GFP_NOIO;
880 	file = sock_alloc_file(csocket, 0, NULL);
881 	if (IS_ERR(file)) {
882 		pr_err("%s (%d): failed to map fd\n",
883 		       __func__, task_pid_nr(current));
884 		kfree(p);
885 		return PTR_ERR(file);
886 	}
887 
888 	get_file(file);
889 	p->wr = p->rd = file;
890 	client->trans = p;
891 	client->status = Connected;
892 
893 	p->rd->f_flags |= O_NONBLOCK;
894 
895 	p9_conn_create(client);
896 	return 0;
897 }
898 
899 /**
900  * p9_conn_destroy - cancels all pending requests of mux
901  * @m: mux to destroy
902  *
903  */
904 
p9_conn_destroy(struct p9_conn * m)905 static void p9_conn_destroy(struct p9_conn *m)
906 {
907 	p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
908 		 m, m->mux_list.prev, m->mux_list.next);
909 
910 	p9_mux_poll_stop(m);
911 	cancel_work_sync(&m->rq);
912 	if (m->rreq) {
913 		p9_req_put(m->client, m->rreq);
914 		m->rreq = NULL;
915 	}
916 	cancel_work_sync(&m->wq);
917 	if (m->wreq) {
918 		p9_req_put(m->client, m->wreq);
919 		m->wreq = NULL;
920 	}
921 
922 	p9_conn_cancel(m, -ECONNRESET);
923 
924 	m->client = NULL;
925 }
926 
927 /**
928  * p9_fd_close - shutdown file descriptor transport
929  * @client: client instance
930  *
931  */
932 
p9_fd_close(struct p9_client * client)933 static void p9_fd_close(struct p9_client *client)
934 {
935 	struct p9_trans_fd *ts;
936 
937 	if (!client)
938 		return;
939 
940 	ts = client->trans;
941 	if (!ts)
942 		return;
943 
944 	client->status = Disconnected;
945 
946 	p9_conn_destroy(&ts->conn);
947 
948 	if (ts->rd)
949 		fput(ts->rd);
950 	if (ts->wr)
951 		fput(ts->wr);
952 
953 	kfree(ts);
954 }
955 
956 /*
957  * stolen from NFS - maybe should be made a generic function?
958  */
valid_ipaddr4(const char * buf)959 static inline int valid_ipaddr4(const char *buf)
960 {
961 	int rc, count, in[4];
962 
963 	rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
964 	if (rc != 4)
965 		return -EINVAL;
966 	for (count = 0; count < 4; count++) {
967 		if (in[count] > 255)
968 			return -EINVAL;
969 	}
970 	return 0;
971 }
972 
p9_bind_privport(struct socket * sock)973 static int p9_bind_privport(struct socket *sock)
974 {
975 	struct sockaddr_in cl;
976 	int port, err = -EINVAL;
977 
978 	memset(&cl, 0, sizeof(cl));
979 	cl.sin_family = AF_INET;
980 	cl.sin_addr.s_addr = htonl(INADDR_ANY);
981 	for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
982 		cl.sin_port = htons((ushort)port);
983 		err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
984 		if (err != -EADDRINUSE)
985 			break;
986 	}
987 	return err;
988 }
989 
990 
991 static int
p9_fd_create_tcp(struct p9_client * client,const char * addr,char * args)992 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
993 {
994 	int err;
995 	struct socket *csocket;
996 	struct sockaddr_in sin_server;
997 	struct p9_fd_opts opts;
998 
999 	err = parse_opts(args, &opts);
1000 	if (err < 0)
1001 		return err;
1002 
1003 	if (addr == NULL || valid_ipaddr4(addr) < 0)
1004 		return -EINVAL;
1005 
1006 	csocket = NULL;
1007 
1008 	client->trans_opts.tcp.port = opts.port;
1009 	client->trans_opts.tcp.privport = opts.privport;
1010 	sin_server.sin_family = AF_INET;
1011 	sin_server.sin_addr.s_addr = in_aton(addr);
1012 	sin_server.sin_port = htons(opts.port);
1013 	err = __sock_create(current->nsproxy->net_ns, PF_INET,
1014 			    SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
1015 	if (err) {
1016 		pr_err("%s (%d): problem creating socket\n",
1017 		       __func__, task_pid_nr(current));
1018 		return err;
1019 	}
1020 
1021 	if (opts.privport) {
1022 		err = p9_bind_privport(csocket);
1023 		if (err < 0) {
1024 			pr_err("%s (%d): problem binding to privport\n",
1025 			       __func__, task_pid_nr(current));
1026 			sock_release(csocket);
1027 			return err;
1028 		}
1029 	}
1030 
1031 	err = csocket->ops->connect(csocket,
1032 				    (struct sockaddr *)&sin_server,
1033 				    sizeof(struct sockaddr_in), 0);
1034 	if (err < 0) {
1035 		pr_err("%s (%d): problem connecting socket to %s\n",
1036 		       __func__, task_pid_nr(current), addr);
1037 		sock_release(csocket);
1038 		return err;
1039 	}
1040 
1041 	return p9_socket_open(client, csocket);
1042 }
1043 
1044 static int
p9_fd_create_unix(struct p9_client * client,const char * addr,char * args)1045 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1046 {
1047 	int err;
1048 	struct socket *csocket;
1049 	struct sockaddr_un sun_server;
1050 
1051 	csocket = NULL;
1052 
1053 	if (!addr || !strlen(addr))
1054 		return -EINVAL;
1055 
1056 	if (strlen(addr) >= UNIX_PATH_MAX) {
1057 		pr_err("%s (%d): address too long: %s\n",
1058 		       __func__, task_pid_nr(current), addr);
1059 		return -ENAMETOOLONG;
1060 	}
1061 
1062 	sun_server.sun_family = PF_UNIX;
1063 	strcpy(sun_server.sun_path, addr);
1064 	err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1065 			    SOCK_STREAM, 0, &csocket, 1);
1066 	if (err < 0) {
1067 		pr_err("%s (%d): problem creating socket\n",
1068 		       __func__, task_pid_nr(current));
1069 
1070 		return err;
1071 	}
1072 	err = csocket->ops->connect(csocket, (struct sockaddr *)&sun_server,
1073 			sizeof(struct sockaddr_un) - 1, 0);
1074 	if (err < 0) {
1075 		pr_err("%s (%d): problem connecting socket: %s: %d\n",
1076 		       __func__, task_pid_nr(current), addr, err);
1077 		sock_release(csocket);
1078 		return err;
1079 	}
1080 
1081 	return p9_socket_open(client, csocket);
1082 }
1083 
1084 static int
p9_fd_create(struct p9_client * client,const char * addr,char * args)1085 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1086 {
1087 	int err;
1088 	struct p9_fd_opts opts;
1089 
1090 	parse_opts(args, &opts);
1091 	client->trans_opts.fd.rfd = opts.rfd;
1092 	client->trans_opts.fd.wfd = opts.wfd;
1093 
1094 	if (opts.rfd == ~0 || opts.wfd == ~0) {
1095 		pr_err("Insufficient options for proto=fd\n");
1096 		return -ENOPROTOOPT;
1097 	}
1098 
1099 	err = p9_fd_open(client, opts.rfd, opts.wfd);
1100 	if (err < 0)
1101 		return err;
1102 
1103 	p9_conn_create(client);
1104 
1105 	return 0;
1106 }
1107 
1108 static struct p9_trans_module p9_tcp_trans = {
1109 	.name = "tcp",
1110 	.maxsize = MAX_SOCK_BUF,
1111 	.def = 0,
1112 	.create = p9_fd_create_tcp,
1113 	.close = p9_fd_close,
1114 	.request = p9_fd_request,
1115 	.cancel = p9_fd_cancel,
1116 	.cancelled = p9_fd_cancelled,
1117 	.show_options = p9_fd_show_options,
1118 	.owner = THIS_MODULE,
1119 };
1120 
1121 static struct p9_trans_module p9_unix_trans = {
1122 	.name = "unix",
1123 	.maxsize = MAX_SOCK_BUF,
1124 	.def = 0,
1125 	.create = p9_fd_create_unix,
1126 	.close = p9_fd_close,
1127 	.request = p9_fd_request,
1128 	.cancel = p9_fd_cancel,
1129 	.cancelled = p9_fd_cancelled,
1130 	.show_options = p9_fd_show_options,
1131 	.owner = THIS_MODULE,
1132 };
1133 
1134 static struct p9_trans_module p9_fd_trans = {
1135 	.name = "fd",
1136 	.maxsize = MAX_SOCK_BUF,
1137 	.def = 0,
1138 	.create = p9_fd_create,
1139 	.close = p9_fd_close,
1140 	.request = p9_fd_request,
1141 	.cancel = p9_fd_cancel,
1142 	.cancelled = p9_fd_cancelled,
1143 	.show_options = p9_fd_show_options,
1144 	.owner = THIS_MODULE,
1145 };
1146 
1147 /**
1148  * p9_poll_workfn - poll worker thread
1149  * @work: work queue
1150  *
1151  * polls all v9fs transports for new events and queues the appropriate
1152  * work to the work queue
1153  *
1154  */
1155 
p9_poll_workfn(struct work_struct * work)1156 static void p9_poll_workfn(struct work_struct *work)
1157 {
1158 	unsigned long flags;
1159 
1160 	p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1161 
1162 	spin_lock_irqsave(&p9_poll_lock, flags);
1163 	while (!list_empty(&p9_poll_pending_list)) {
1164 		struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1165 							struct p9_conn,
1166 							poll_pending_link);
1167 		list_del_init(&conn->poll_pending_link);
1168 		spin_unlock_irqrestore(&p9_poll_lock, flags);
1169 
1170 		p9_poll_mux(conn);
1171 
1172 		spin_lock_irqsave(&p9_poll_lock, flags);
1173 	}
1174 	spin_unlock_irqrestore(&p9_poll_lock, flags);
1175 
1176 	p9_debug(P9_DEBUG_TRANS, "finish\n");
1177 }
1178 
p9_trans_fd_init(void)1179 int p9_trans_fd_init(void)
1180 {
1181 	v9fs_register_trans(&p9_tcp_trans);
1182 	v9fs_register_trans(&p9_unix_trans);
1183 	v9fs_register_trans(&p9_fd_trans);
1184 
1185 	return 0;
1186 }
1187 
p9_trans_fd_exit(void)1188 void p9_trans_fd_exit(void)
1189 {
1190 	flush_work(&p9_poll_work);
1191 	v9fs_unregister_trans(&p9_tcp_trans);
1192 	v9fs_unregister_trans(&p9_unix_trans);
1193 	v9fs_unregister_trans(&p9_fd_trans);
1194 }
1195