1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Fd transport layer. Includes deprecated socket layer.
4 *
5 * Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
6 * Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
7 * Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
8 * Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
9 */
10
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13 #include <linux/in.h>
14 #include <linux/module.h>
15 #include <linux/net.h>
16 #include <linux/ipv6.h>
17 #include <linux/kthread.h>
18 #include <linux/errno.h>
19 #include <linux/kernel.h>
20 #include <linux/un.h>
21 #include <linux/uaccess.h>
22 #include <linux/inet.h>
23 #include <linux/file.h>
24 #include <linux/parser.h>
25 #include <linux/slab.h>
26 #include <linux/seq_file.h>
27 #include <net/9p/9p.h>
28 #include <net/9p/client.h>
29 #include <net/9p/transport.h>
30
31 #include <linux/syscalls.h> /* killme */
32
33 #define P9_PORT 564
34 #define MAX_SOCK_BUF (1024*1024)
35 #define MAXPOLLWADDR 2
36
37 static struct p9_trans_module p9_tcp_trans;
38 static struct p9_trans_module p9_fd_trans;
39
40 /**
41 * struct p9_fd_opts - per-transport options
42 * @rfd: file descriptor for reading (trans=fd)
43 * @wfd: file descriptor for writing (trans=fd)
44 * @port: port to connect to (trans=tcp)
45 * @privport: port is privileged
46 */
47
48 struct p9_fd_opts {
49 int rfd;
50 int wfd;
51 u16 port;
52 bool privport;
53 };
54
55 /*
56 * Option Parsing (code inspired by NFS code)
57 * - a little lazy - parse all fd-transport options
58 */
59
60 enum {
61 /* Options that take integer arguments */
62 Opt_port, Opt_rfdno, Opt_wfdno, Opt_err,
63 /* Options that take no arguments */
64 Opt_privport,
65 };
66
67 static const match_table_t tokens = {
68 {Opt_port, "port=%u"},
69 {Opt_rfdno, "rfdno=%u"},
70 {Opt_wfdno, "wfdno=%u"},
71 {Opt_privport, "privport"},
72 {Opt_err, NULL},
73 };
74
75 enum {
76 Rworksched = 1, /* read work scheduled or running */
77 Rpending = 2, /* can read */
78 Wworksched = 4, /* write work scheduled or running */
79 Wpending = 8, /* can write */
80 };
81
82 struct p9_poll_wait {
83 struct p9_conn *conn;
84 wait_queue_entry_t wait;
85 wait_queue_head_t *wait_addr;
86 };
87
88 /**
89 * struct p9_conn - fd mux connection state information
90 * @mux_list: list link for mux to manage multiple connections (?)
91 * @client: reference to client instance for this connection
92 * @err: error state
93 * @req_lock: lock protecting req_list and requests statuses
94 * @req_list: accounting for requests which have been sent
95 * @unsent_req_list: accounting for requests that haven't been sent
96 * @rreq: read request
97 * @wreq: write request
98 * @tmp_buf: temporary buffer to read in header
99 * @rc: temporary fcall for reading current frame
100 * @wpos: write position for current frame
101 * @wsize: amount of data to write for current frame
102 * @wbuf: current write buffer
103 * @poll_pending_link: pending links to be polled per conn
104 * @poll_wait: array of wait_q's for various worker threads
105 * @pt: poll state
106 * @rq: current read work
107 * @wq: current write work
108 * @wsched: ????
109 *
110 */
111
112 struct p9_conn {
113 struct list_head mux_list;
114 struct p9_client *client;
115 int err;
116 spinlock_t req_lock;
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->req_lock);
193
194 if (READ_ONCE(m->err)) {
195 spin_unlock(&m->req_lock);
196 return;
197 }
198
199 WRITE_ONCE(m->err, err);
200 ASSERT_EXCLUSIVE_WRITER(m->err);
201
202 list_for_each_entry_safe(req, rtmp, &m->req_list, req_list) {
203 list_move(&req->req_list, &cancel_list);
204 WRITE_ONCE(req->status, REQ_STATUS_ERROR);
205 }
206 list_for_each_entry_safe(req, rtmp, &m->unsent_req_list, req_list) {
207 list_move(&req->req_list, &cancel_list);
208 WRITE_ONCE(req->status, REQ_STATUS_ERROR);
209 }
210
211 spin_unlock(&m->req_lock);
212
213 list_for_each_entry_safe(req, rtmp, &cancel_list, req_list) {
214 p9_debug(P9_DEBUG_ERROR, "call back req %p\n", req);
215 list_del(&req->req_list);
216 if (!req->t_err)
217 req->t_err = err;
218 p9_client_cb(m->client, req, REQ_STATUS_ERROR);
219 }
220 }
221
222 static __poll_t
p9_fd_poll(struct p9_client * client,struct poll_table_struct * pt,int * err)223 p9_fd_poll(struct p9_client *client, struct poll_table_struct *pt, int *err)
224 {
225 __poll_t ret;
226 struct p9_trans_fd *ts = NULL;
227
228 if (client && client->status == Connected)
229 ts = client->trans;
230
231 if (!ts) {
232 if (err)
233 *err = -EREMOTEIO;
234 return EPOLLERR;
235 }
236
237 ret = vfs_poll(ts->rd, pt);
238 if (ts->rd != ts->wr)
239 ret = (ret & ~EPOLLOUT) | (vfs_poll(ts->wr, pt) & ~EPOLLIN);
240 return ret;
241 }
242
243 /**
244 * p9_fd_read- read from a fd
245 * @client: client instance
246 * @v: buffer to receive data into
247 * @len: size of receive buffer
248 *
249 */
250
p9_fd_read(struct p9_client * client,void * v,int len)251 static int p9_fd_read(struct p9_client *client, void *v, int len)
252 {
253 int ret;
254 struct p9_trans_fd *ts = NULL;
255 loff_t pos;
256
257 if (client && client->status != Disconnected)
258 ts = client->trans;
259
260 if (!ts)
261 return -EREMOTEIO;
262
263 if (!(ts->rd->f_flags & O_NONBLOCK))
264 p9_debug(P9_DEBUG_ERROR, "blocking read ...\n");
265
266 pos = ts->rd->f_pos;
267 ret = kernel_read(ts->rd, v, len, &pos);
268 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
269 client->status = Disconnected;
270 return ret;
271 }
272
273 /**
274 * p9_read_work - called when there is some data to be read from a transport
275 * @work: container of work to be done
276 *
277 */
278
p9_read_work(struct work_struct * work)279 static void p9_read_work(struct work_struct *work)
280 {
281 __poll_t n;
282 int err;
283 struct p9_conn *m;
284
285 m = container_of(work, struct p9_conn, rq);
286
287 if (READ_ONCE(m->err) < 0)
288 return;
289
290 p9_debug(P9_DEBUG_TRANS, "start mux %p pos %zd\n", m, m->rc.offset);
291
292 if (!m->rc.sdata) {
293 m->rc.sdata = m->tmp_buf;
294 m->rc.offset = 0;
295 m->rc.capacity = P9_HDRSZ; /* start by reading header */
296 }
297
298 clear_bit(Rpending, &m->wsched);
299 p9_debug(P9_DEBUG_TRANS, "read mux %p pos %zd size: %zd = %zd\n",
300 m, m->rc.offset, m->rc.capacity,
301 m->rc.capacity - m->rc.offset);
302 err = p9_fd_read(m->client, m->rc.sdata + m->rc.offset,
303 m->rc.capacity - m->rc.offset);
304 p9_debug(P9_DEBUG_TRANS, "mux %p got %d bytes\n", m, err);
305 if (err == -EAGAIN)
306 goto end_clear;
307
308 if (err <= 0)
309 goto error;
310
311 m->rc.offset += err;
312
313 /* header read in */
314 if ((!m->rreq) && (m->rc.offset == m->rc.capacity)) {
315 p9_debug(P9_DEBUG_TRANS, "got new header\n");
316
317 /* Header size */
318 m->rc.size = P9_HDRSZ;
319 err = p9_parse_header(&m->rc, &m->rc.size, NULL, NULL, 0);
320 if (err) {
321 p9_debug(P9_DEBUG_ERROR,
322 "error parsing header: %d\n", err);
323 goto error;
324 }
325
326 p9_debug(P9_DEBUG_TRANS,
327 "mux %p pkt: size: %d bytes tag: %d\n",
328 m, m->rc.size, m->rc.tag);
329
330 m->rreq = p9_tag_lookup(m->client, m->rc.tag);
331 if (!m->rreq || (m->rreq->status != REQ_STATUS_SENT)) {
332 p9_debug(P9_DEBUG_ERROR, "Unexpected packet tag %d\n",
333 m->rc.tag);
334 err = -EIO;
335 goto error;
336 }
337
338 if (m->rc.size > m->rreq->rc.capacity) {
339 p9_debug(P9_DEBUG_ERROR,
340 "requested packet size too big: %d for tag %d with capacity %zd\n",
341 m->rc.size, m->rc.tag, m->rreq->rc.capacity);
342 err = -EIO;
343 goto error;
344 }
345
346 if (!m->rreq->rc.sdata) {
347 p9_debug(P9_DEBUG_ERROR,
348 "No recv fcall for tag %d (req %p), disconnecting!\n",
349 m->rc.tag, m->rreq);
350 p9_req_put(m->client, m->rreq);
351 m->rreq = NULL;
352 err = -EIO;
353 goto error;
354 }
355 m->rc.sdata = m->rreq->rc.sdata;
356 memcpy(m->rc.sdata, m->tmp_buf, m->rc.capacity);
357 m->rc.capacity = m->rc.size;
358 }
359
360 /* packet is read in
361 * not an else because some packets (like clunk) have no payload
362 */
363 if ((m->rreq) && (m->rc.offset == m->rc.capacity)) {
364 p9_debug(P9_DEBUG_TRANS, "got new packet\n");
365 m->rreq->rc.size = m->rc.offset;
366 spin_lock(&m->req_lock);
367 if (m->rreq->status == REQ_STATUS_SENT) {
368 list_del(&m->rreq->req_list);
369 p9_client_cb(m->client, m->rreq, REQ_STATUS_RCVD);
370 } else if (m->rreq->status == REQ_STATUS_FLSHD) {
371 /* Ignore replies associated with a cancelled request. */
372 p9_debug(P9_DEBUG_TRANS,
373 "Ignore replies associated with a cancelled request\n");
374 } else {
375 spin_unlock(&m->req_lock);
376 p9_debug(P9_DEBUG_ERROR,
377 "Request tag %d errored out while we were reading the reply\n",
378 m->rc.tag);
379 err = -EIO;
380 goto error;
381 }
382 spin_unlock(&m->req_lock);
383 m->rc.sdata = NULL;
384 m->rc.offset = 0;
385 m->rc.capacity = 0;
386 p9_req_put(m->client, m->rreq);
387 m->rreq = NULL;
388 }
389
390 end_clear:
391 clear_bit(Rworksched, &m->wsched);
392
393 if (!list_empty(&m->req_list)) {
394 if (test_and_clear_bit(Rpending, &m->wsched))
395 n = EPOLLIN;
396 else
397 n = p9_fd_poll(m->client, NULL, NULL);
398
399 if ((n & EPOLLIN) && !test_and_set_bit(Rworksched, &m->wsched)) {
400 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
401 schedule_work(&m->rq);
402 }
403 }
404
405 return;
406 error:
407 p9_conn_cancel(m, err);
408 clear_bit(Rworksched, &m->wsched);
409 }
410
411 /**
412 * p9_fd_write - write to a socket
413 * @client: client instance
414 * @v: buffer to send data from
415 * @len: size of send buffer
416 *
417 */
418
p9_fd_write(struct p9_client * client,void * v,int len)419 static int p9_fd_write(struct p9_client *client, void *v, int len)
420 {
421 ssize_t ret;
422 struct p9_trans_fd *ts = NULL;
423
424 if (client && client->status != Disconnected)
425 ts = client->trans;
426
427 if (!ts)
428 return -EREMOTEIO;
429
430 if (!(ts->wr->f_flags & O_NONBLOCK))
431 p9_debug(P9_DEBUG_ERROR, "blocking write ...\n");
432
433 ret = kernel_write(ts->wr, v, len, &ts->wr->f_pos);
434 if (ret <= 0 && ret != -ERESTARTSYS && ret != -EAGAIN)
435 client->status = Disconnected;
436 return ret;
437 }
438
439 /**
440 * p9_write_work - called when a transport can send some data
441 * @work: container for work to be done
442 *
443 */
444
p9_write_work(struct work_struct * work)445 static void p9_write_work(struct work_struct *work)
446 {
447 __poll_t n;
448 int err;
449 struct p9_conn *m;
450 struct p9_req_t *req;
451
452 m = container_of(work, struct p9_conn, wq);
453
454 if (READ_ONCE(m->err) < 0) {
455 clear_bit(Wworksched, &m->wsched);
456 return;
457 }
458
459 if (!m->wsize) {
460 spin_lock(&m->req_lock);
461 if (list_empty(&m->unsent_req_list)) {
462 clear_bit(Wworksched, &m->wsched);
463 spin_unlock(&m->req_lock);
464 return;
465 }
466
467 req = list_entry(m->unsent_req_list.next, struct p9_req_t,
468 req_list);
469 WRITE_ONCE(req->status, REQ_STATUS_SENT);
470 p9_debug(P9_DEBUG_TRANS, "move req %p\n", req);
471 list_move_tail(&req->req_list, &m->req_list);
472
473 m->wbuf = req->tc.sdata;
474 m->wsize = req->tc.size;
475 m->wpos = 0;
476 p9_req_get(req);
477 m->wreq = req;
478 spin_unlock(&m->req_lock);
479 }
480
481 p9_debug(P9_DEBUG_TRANS, "mux %p pos %d size %d\n",
482 m, m->wpos, m->wsize);
483 clear_bit(Wpending, &m->wsched);
484 err = p9_fd_write(m->client, m->wbuf + m->wpos, m->wsize - m->wpos);
485 p9_debug(P9_DEBUG_TRANS, "mux %p sent %d bytes\n", m, err);
486 if (err == -EAGAIN)
487 goto end_clear;
488
489
490 if (err < 0)
491 goto error;
492 else if (err == 0) {
493 err = -EREMOTEIO;
494 goto error;
495 }
496
497 m->wpos += err;
498 if (m->wpos == m->wsize) {
499 m->wpos = m->wsize = 0;
500 p9_req_put(m->client, m->wreq);
501 m->wreq = NULL;
502 }
503
504 end_clear:
505 clear_bit(Wworksched, &m->wsched);
506
507 if (m->wsize || !list_empty(&m->unsent_req_list)) {
508 if (test_and_clear_bit(Wpending, &m->wsched))
509 n = EPOLLOUT;
510 else
511 n = p9_fd_poll(m->client, NULL, NULL);
512
513 if ((n & EPOLLOUT) &&
514 !test_and_set_bit(Wworksched, &m->wsched)) {
515 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
516 schedule_work(&m->wq);
517 }
518 }
519
520 return;
521
522 error:
523 p9_conn_cancel(m, err);
524 clear_bit(Wworksched, &m->wsched);
525 }
526
p9_pollwake(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)527 static int p9_pollwake(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
528 {
529 struct p9_poll_wait *pwait =
530 container_of(wait, struct p9_poll_wait, wait);
531 struct p9_conn *m = pwait->conn;
532 unsigned long flags;
533
534 spin_lock_irqsave(&p9_poll_lock, flags);
535 if (list_empty(&m->poll_pending_link))
536 list_add_tail(&m->poll_pending_link, &p9_poll_pending_list);
537 spin_unlock_irqrestore(&p9_poll_lock, flags);
538
539 schedule_work(&p9_poll_work);
540 return 1;
541 }
542
543 /**
544 * p9_pollwait - add poll task to the wait queue
545 * @filp: file pointer being polled
546 * @wait_address: wait_q to block on
547 * @p: poll state
548 *
549 * called by files poll operation to add v9fs-poll task to files wait queue
550 */
551
552 static void
p9_pollwait(struct file * filp,wait_queue_head_t * wait_address,poll_table * p)553 p9_pollwait(struct file *filp, wait_queue_head_t *wait_address, poll_table *p)
554 {
555 struct p9_conn *m = container_of(p, struct p9_conn, pt);
556 struct p9_poll_wait *pwait = NULL;
557 int i;
558
559 for (i = 0; i < ARRAY_SIZE(m->poll_wait); i++) {
560 if (m->poll_wait[i].wait_addr == NULL) {
561 pwait = &m->poll_wait[i];
562 break;
563 }
564 }
565
566 if (!pwait) {
567 p9_debug(P9_DEBUG_ERROR, "not enough wait_address slots\n");
568 return;
569 }
570
571 pwait->conn = m;
572 pwait->wait_addr = wait_address;
573 init_waitqueue_func_entry(&pwait->wait, p9_pollwake);
574 add_wait_queue(wait_address, &pwait->wait);
575 }
576
577 /**
578 * p9_conn_create - initialize the per-session mux data
579 * @client: client instance
580 *
581 * Note: Creates the polling task if this is the first session.
582 */
583
p9_conn_create(struct p9_client * client)584 static void p9_conn_create(struct p9_client *client)
585 {
586 __poll_t n;
587 struct p9_trans_fd *ts = client->trans;
588 struct p9_conn *m = &ts->conn;
589
590 p9_debug(P9_DEBUG_TRANS, "client %p msize %d\n", client, client->msize);
591
592 INIT_LIST_HEAD(&m->mux_list);
593 m->client = client;
594
595 spin_lock_init(&m->req_lock);
596 INIT_LIST_HEAD(&m->req_list);
597 INIT_LIST_HEAD(&m->unsent_req_list);
598 INIT_WORK(&m->rq, p9_read_work);
599 INIT_WORK(&m->wq, p9_write_work);
600 INIT_LIST_HEAD(&m->poll_pending_link);
601 init_poll_funcptr(&m->pt, p9_pollwait);
602
603 n = p9_fd_poll(client, &m->pt, NULL);
604 if (n & EPOLLIN) {
605 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
606 set_bit(Rpending, &m->wsched);
607 }
608
609 if (n & EPOLLOUT) {
610 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
611 set_bit(Wpending, &m->wsched);
612 }
613 }
614
615 /**
616 * p9_poll_mux - polls a mux and schedules read or write works if necessary
617 * @m: connection to poll
618 *
619 */
620
p9_poll_mux(struct p9_conn * m)621 static void p9_poll_mux(struct p9_conn *m)
622 {
623 __poll_t n;
624 int err = -ECONNRESET;
625
626 if (READ_ONCE(m->err) < 0)
627 return;
628
629 n = p9_fd_poll(m->client, NULL, &err);
630 if (n & (EPOLLERR | EPOLLHUP | EPOLLNVAL)) {
631 p9_debug(P9_DEBUG_TRANS, "error mux %p err %d\n", m, n);
632 p9_conn_cancel(m, err);
633 }
634
635 if (n & EPOLLIN) {
636 set_bit(Rpending, &m->wsched);
637 p9_debug(P9_DEBUG_TRANS, "mux %p can read\n", m);
638 if (!test_and_set_bit(Rworksched, &m->wsched)) {
639 p9_debug(P9_DEBUG_TRANS, "sched read work %p\n", m);
640 schedule_work(&m->rq);
641 }
642 }
643
644 if (n & EPOLLOUT) {
645 set_bit(Wpending, &m->wsched);
646 p9_debug(P9_DEBUG_TRANS, "mux %p can write\n", m);
647 if ((m->wsize || !list_empty(&m->unsent_req_list)) &&
648 !test_and_set_bit(Wworksched, &m->wsched)) {
649 p9_debug(P9_DEBUG_TRANS, "sched write work %p\n", m);
650 schedule_work(&m->wq);
651 }
652 }
653 }
654
655 /**
656 * p9_fd_request - send 9P request
657 * The function can sleep until the request is scheduled for sending.
658 * The function can be interrupted. Return from the function is not
659 * a guarantee that the request is sent successfully.
660 *
661 * @client: client instance
662 * @req: request to be sent
663 *
664 */
665
p9_fd_request(struct p9_client * client,struct p9_req_t * req)666 static int p9_fd_request(struct p9_client *client, struct p9_req_t *req)
667 {
668 __poll_t n;
669 int err;
670 struct p9_trans_fd *ts = client->trans;
671 struct p9_conn *m = &ts->conn;
672
673 p9_debug(P9_DEBUG_TRANS, "mux %p task %p tcall %p id %d\n",
674 m, current, &req->tc, req->tc.id);
675
676 spin_lock(&m->req_lock);
677
678 err = READ_ONCE(m->err);
679 if (err < 0) {
680 spin_unlock(&m->req_lock);
681 return err;
682 }
683
684 WRITE_ONCE(req->status, REQ_STATUS_UNSENT);
685 list_add_tail(&req->req_list, &m->unsent_req_list);
686 spin_unlock(&m->req_lock);
687
688 if (test_and_clear_bit(Wpending, &m->wsched))
689 n = EPOLLOUT;
690 else
691 n = p9_fd_poll(m->client, NULL, NULL);
692
693 if (n & EPOLLOUT && !test_and_set_bit(Wworksched, &m->wsched))
694 schedule_work(&m->wq);
695
696 return 0;
697 }
698
p9_fd_cancel(struct p9_client * client,struct p9_req_t * req)699 static int p9_fd_cancel(struct p9_client *client, struct p9_req_t *req)
700 {
701 struct p9_trans_fd *ts = client->trans;
702 struct p9_conn *m = &ts->conn;
703 int ret = 1;
704
705 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
706
707 spin_lock(&m->req_lock);
708
709 if (req->status == REQ_STATUS_UNSENT) {
710 list_del(&req->req_list);
711 WRITE_ONCE(req->status, REQ_STATUS_FLSHD);
712 p9_req_put(client, req);
713 ret = 0;
714 }
715 spin_unlock(&m->req_lock);
716
717 return ret;
718 }
719
p9_fd_cancelled(struct p9_client * client,struct p9_req_t * req)720 static int p9_fd_cancelled(struct p9_client *client, struct p9_req_t *req)
721 {
722 struct p9_trans_fd *ts = client->trans;
723 struct p9_conn *m = &ts->conn;
724
725 p9_debug(P9_DEBUG_TRANS, "client %p req %p\n", client, req);
726
727 spin_lock(&m->req_lock);
728 /* Ignore cancelled request if status changed since the request was
729 * processed in p9_client_flush()
730 */
731 if (req->status != REQ_STATUS_SENT) {
732 spin_unlock(&m->req_lock);
733 return 0;
734 }
735
736 /* we haven't received a response for oldreq,
737 * remove it from the list.
738 */
739 list_del(&req->req_list);
740 WRITE_ONCE(req->status, REQ_STATUS_FLSHD);
741 spin_unlock(&m->req_lock);
742
743 p9_req_put(client, req);
744
745 return 0;
746 }
747
p9_fd_show_options(struct seq_file * m,struct p9_client * clnt)748 static int p9_fd_show_options(struct seq_file *m, struct p9_client *clnt)
749 {
750 if (clnt->trans_mod == &p9_tcp_trans) {
751 if (clnt->trans_opts.tcp.port != P9_PORT)
752 seq_printf(m, ",port=%u", clnt->trans_opts.tcp.port);
753 } else if (clnt->trans_mod == &p9_fd_trans) {
754 if (clnt->trans_opts.fd.rfd != ~0)
755 seq_printf(m, ",rfd=%u", clnt->trans_opts.fd.rfd);
756 if (clnt->trans_opts.fd.wfd != ~0)
757 seq_printf(m, ",wfd=%u", clnt->trans_opts.fd.wfd);
758 }
759 return 0;
760 }
761
762 /**
763 * parse_opts - parse mount options into p9_fd_opts structure
764 * @params: options string passed from mount
765 * @opts: fd transport-specific structure to parse options into
766 *
767 * Returns 0 upon success, -ERRNO upon failure
768 */
769
parse_opts(char * params,struct p9_fd_opts * opts)770 static int parse_opts(char *params, struct p9_fd_opts *opts)
771 {
772 char *p;
773 substring_t args[MAX_OPT_ARGS];
774 int option;
775 char *options, *tmp_options;
776
777 opts->port = P9_PORT;
778 opts->rfd = ~0;
779 opts->wfd = ~0;
780 opts->privport = false;
781
782 if (!params)
783 return 0;
784
785 tmp_options = kstrdup(params, GFP_KERNEL);
786 if (!tmp_options) {
787 p9_debug(P9_DEBUG_ERROR,
788 "failed to allocate copy of option string\n");
789 return -ENOMEM;
790 }
791 options = tmp_options;
792
793 while ((p = strsep(&options, ",")) != NULL) {
794 int token;
795 int r;
796 if (!*p)
797 continue;
798 token = match_token(p, tokens, args);
799 if ((token != Opt_err) && (token != Opt_privport)) {
800 r = match_int(&args[0], &option);
801 if (r < 0) {
802 p9_debug(P9_DEBUG_ERROR,
803 "integer field, but no integer?\n");
804 continue;
805 }
806 }
807 switch (token) {
808 case Opt_port:
809 opts->port = option;
810 break;
811 case Opt_rfdno:
812 opts->rfd = option;
813 break;
814 case Opt_wfdno:
815 opts->wfd = option;
816 break;
817 case Opt_privport:
818 opts->privport = true;
819 break;
820 default:
821 continue;
822 }
823 }
824
825 kfree(tmp_options);
826 return 0;
827 }
828
p9_fd_open(struct p9_client * client,int rfd,int wfd)829 static int p9_fd_open(struct p9_client *client, int rfd, int wfd)
830 {
831 struct p9_trans_fd *ts = kzalloc(sizeof(struct p9_trans_fd),
832 GFP_KERNEL);
833 if (!ts)
834 return -ENOMEM;
835
836 ts->rd = fget(rfd);
837 if (!ts->rd)
838 goto out_free_ts;
839 if (!(ts->rd->f_mode & FMODE_READ))
840 goto out_put_rd;
841 /* Prevent workers from hanging on IO when fd is a pipe.
842 * It's technically possible for userspace or concurrent mounts to
843 * modify this flag concurrently, which will likely result in a
844 * broken filesystem. However, just having bad flags here should
845 * not crash the kernel or cause any other sort of bug, so mark this
846 * particular data race as intentional so that tooling (like KCSAN)
847 * can allow it and detect further problems.
848 */
849 data_race(ts->rd->f_flags |= O_NONBLOCK);
850 ts->wr = fget(wfd);
851 if (!ts->wr)
852 goto out_put_rd;
853 if (!(ts->wr->f_mode & FMODE_WRITE))
854 goto out_put_wr;
855 data_race(ts->wr->f_flags |= O_NONBLOCK);
856
857 client->trans = ts;
858 client->status = Connected;
859
860 return 0;
861
862 out_put_wr:
863 fput(ts->wr);
864 out_put_rd:
865 fput(ts->rd);
866 out_free_ts:
867 kfree(ts);
868 return -EIO;
869 }
870
p9_socket_open(struct p9_client * client,struct socket * csocket)871 static int p9_socket_open(struct p9_client *client, struct socket *csocket)
872 {
873 struct p9_trans_fd *p;
874 struct file *file;
875
876 p = kzalloc(sizeof(struct p9_trans_fd), GFP_KERNEL);
877 if (!p) {
878 sock_release(csocket);
879 return -ENOMEM;
880 }
881
882 csocket->sk->sk_allocation = GFP_NOIO;
883 csocket->sk->sk_use_task_frag = false;
884 file = sock_alloc_file(csocket, 0, NULL);
885 if (IS_ERR(file)) {
886 pr_err("%s (%d): failed to map fd\n",
887 __func__, task_pid_nr(current));
888 kfree(p);
889 return PTR_ERR(file);
890 }
891
892 get_file(file);
893 p->wr = p->rd = file;
894 client->trans = p;
895 client->status = Connected;
896
897 p->rd->f_flags |= O_NONBLOCK;
898
899 p9_conn_create(client);
900 return 0;
901 }
902
903 /**
904 * p9_conn_destroy - cancels all pending requests of mux
905 * @m: mux to destroy
906 *
907 */
908
p9_conn_destroy(struct p9_conn * m)909 static void p9_conn_destroy(struct p9_conn *m)
910 {
911 p9_debug(P9_DEBUG_TRANS, "mux %p prev %p next %p\n",
912 m, m->mux_list.prev, m->mux_list.next);
913
914 p9_mux_poll_stop(m);
915 cancel_work_sync(&m->rq);
916 if (m->rreq) {
917 p9_req_put(m->client, m->rreq);
918 m->rreq = NULL;
919 }
920 cancel_work_sync(&m->wq);
921 if (m->wreq) {
922 p9_req_put(m->client, m->wreq);
923 m->wreq = NULL;
924 }
925
926 p9_conn_cancel(m, -ECONNRESET);
927
928 m->client = NULL;
929 }
930
931 /**
932 * p9_fd_close - shutdown file descriptor transport
933 * @client: client instance
934 *
935 */
936
p9_fd_close(struct p9_client * client)937 static void p9_fd_close(struct p9_client *client)
938 {
939 struct p9_trans_fd *ts;
940
941 if (!client)
942 return;
943
944 ts = client->trans;
945 if (!ts)
946 return;
947
948 client->status = Disconnected;
949
950 p9_conn_destroy(&ts->conn);
951
952 if (ts->rd)
953 fput(ts->rd);
954 if (ts->wr)
955 fput(ts->wr);
956
957 kfree(ts);
958 }
959
960 /*
961 * stolen from NFS - maybe should be made a generic function?
962 */
valid_ipaddr4(const char * buf)963 static inline int valid_ipaddr4(const char *buf)
964 {
965 int rc, count, in[4];
966
967 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
968 if (rc != 4)
969 return -EINVAL;
970 for (count = 0; count < 4; count++) {
971 if (in[count] > 255)
972 return -EINVAL;
973 }
974 return 0;
975 }
976
p9_bind_privport(struct socket * sock)977 static int p9_bind_privport(struct socket *sock)
978 {
979 struct sockaddr_in cl;
980 int port, err = -EINVAL;
981
982 memset(&cl, 0, sizeof(cl));
983 cl.sin_family = AF_INET;
984 cl.sin_addr.s_addr = htonl(INADDR_ANY);
985 for (port = p9_ipport_resv_max; port >= p9_ipport_resv_min; port--) {
986 cl.sin_port = htons((ushort)port);
987 err = kernel_bind(sock, (struct sockaddr *)&cl, sizeof(cl));
988 if (err != -EADDRINUSE)
989 break;
990 }
991 return err;
992 }
993
994
995 static int
p9_fd_create_tcp(struct p9_client * client,const char * addr,char * args)996 p9_fd_create_tcp(struct p9_client *client, const char *addr, char *args)
997 {
998 int err;
999 struct socket *csocket;
1000 struct sockaddr_in sin_server;
1001 struct p9_fd_opts opts;
1002
1003 err = parse_opts(args, &opts);
1004 if (err < 0)
1005 return err;
1006
1007 if (addr == NULL || valid_ipaddr4(addr) < 0)
1008 return -EINVAL;
1009
1010 csocket = NULL;
1011
1012 client->trans_opts.tcp.port = opts.port;
1013 client->trans_opts.tcp.privport = opts.privport;
1014 sin_server.sin_family = AF_INET;
1015 sin_server.sin_addr.s_addr = in_aton(addr);
1016 sin_server.sin_port = htons(opts.port);
1017 err = __sock_create(current->nsproxy->net_ns, PF_INET,
1018 SOCK_STREAM, IPPROTO_TCP, &csocket, 1);
1019 if (err) {
1020 pr_err("%s (%d): problem creating socket\n",
1021 __func__, task_pid_nr(current));
1022 return err;
1023 }
1024
1025 if (opts.privport) {
1026 err = p9_bind_privport(csocket);
1027 if (err < 0) {
1028 pr_err("%s (%d): problem binding to privport\n",
1029 __func__, task_pid_nr(current));
1030 sock_release(csocket);
1031 return err;
1032 }
1033 }
1034
1035 err = READ_ONCE(csocket->ops)->connect(csocket,
1036 (struct sockaddr *)&sin_server,
1037 sizeof(struct sockaddr_in), 0);
1038 if (err < 0) {
1039 pr_err("%s (%d): problem connecting socket to %s\n",
1040 __func__, task_pid_nr(current), addr);
1041 sock_release(csocket);
1042 return err;
1043 }
1044
1045 return p9_socket_open(client, csocket);
1046 }
1047
1048 static int
p9_fd_create_unix(struct p9_client * client,const char * addr,char * args)1049 p9_fd_create_unix(struct p9_client *client, const char *addr, char *args)
1050 {
1051 int err;
1052 struct socket *csocket;
1053 struct sockaddr_un sun_server;
1054
1055 csocket = NULL;
1056
1057 if (!addr || !strlen(addr))
1058 return -EINVAL;
1059
1060 if (strlen(addr) >= UNIX_PATH_MAX) {
1061 pr_err("%s (%d): address too long: %s\n",
1062 __func__, task_pid_nr(current), addr);
1063 return -ENAMETOOLONG;
1064 }
1065
1066 sun_server.sun_family = PF_UNIX;
1067 strcpy(sun_server.sun_path, addr);
1068 err = __sock_create(current->nsproxy->net_ns, PF_UNIX,
1069 SOCK_STREAM, 0, &csocket, 1);
1070 if (err < 0) {
1071 pr_err("%s (%d): problem creating socket\n",
1072 __func__, task_pid_nr(current));
1073
1074 return err;
1075 }
1076 err = READ_ONCE(csocket->ops)->connect(csocket, (struct sockaddr *)&sun_server,
1077 sizeof(struct sockaddr_un) - 1, 0);
1078 if (err < 0) {
1079 pr_err("%s (%d): problem connecting socket: %s: %d\n",
1080 __func__, task_pid_nr(current), addr, err);
1081 sock_release(csocket);
1082 return err;
1083 }
1084
1085 return p9_socket_open(client, csocket);
1086 }
1087
1088 static int
p9_fd_create(struct p9_client * client,const char * addr,char * args)1089 p9_fd_create(struct p9_client *client, const char *addr, char *args)
1090 {
1091 int err;
1092 struct p9_fd_opts opts;
1093
1094 err = parse_opts(args, &opts);
1095 if (err < 0)
1096 return err;
1097 client->trans_opts.fd.rfd = opts.rfd;
1098 client->trans_opts.fd.wfd = opts.wfd;
1099
1100 if (opts.rfd == ~0 || opts.wfd == ~0) {
1101 pr_err("Insufficient options for proto=fd\n");
1102 return -ENOPROTOOPT;
1103 }
1104
1105 err = p9_fd_open(client, opts.rfd, opts.wfd);
1106 if (err < 0)
1107 return err;
1108
1109 p9_conn_create(client);
1110
1111 return 0;
1112 }
1113
1114 static struct p9_trans_module p9_tcp_trans = {
1115 .name = "tcp",
1116 .maxsize = MAX_SOCK_BUF,
1117 .pooled_rbuffers = false,
1118 .def = 0,
1119 .create = p9_fd_create_tcp,
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 MODULE_ALIAS_9P("tcp");
1128
1129 static struct p9_trans_module p9_unix_trans = {
1130 .name = "unix",
1131 .maxsize = MAX_SOCK_BUF,
1132 .def = 0,
1133 .create = p9_fd_create_unix,
1134 .close = p9_fd_close,
1135 .request = p9_fd_request,
1136 .cancel = p9_fd_cancel,
1137 .cancelled = p9_fd_cancelled,
1138 .show_options = p9_fd_show_options,
1139 .owner = THIS_MODULE,
1140 };
1141 MODULE_ALIAS_9P("unix");
1142
1143 static struct p9_trans_module p9_fd_trans = {
1144 .name = "fd",
1145 .maxsize = MAX_SOCK_BUF,
1146 .def = 0,
1147 .create = p9_fd_create,
1148 .close = p9_fd_close,
1149 .request = p9_fd_request,
1150 .cancel = p9_fd_cancel,
1151 .cancelled = p9_fd_cancelled,
1152 .show_options = p9_fd_show_options,
1153 .owner = THIS_MODULE,
1154 };
1155 MODULE_ALIAS_9P("fd");
1156
1157 /**
1158 * p9_poll_workfn - poll worker thread
1159 * @work: work queue
1160 *
1161 * polls all v9fs transports for new events and queues the appropriate
1162 * work to the work queue
1163 *
1164 */
1165
p9_poll_workfn(struct work_struct * work)1166 static void p9_poll_workfn(struct work_struct *work)
1167 {
1168 unsigned long flags;
1169
1170 p9_debug(P9_DEBUG_TRANS, "start %p\n", current);
1171
1172 spin_lock_irqsave(&p9_poll_lock, flags);
1173 while (!list_empty(&p9_poll_pending_list)) {
1174 struct p9_conn *conn = list_first_entry(&p9_poll_pending_list,
1175 struct p9_conn,
1176 poll_pending_link);
1177 list_del_init(&conn->poll_pending_link);
1178 spin_unlock_irqrestore(&p9_poll_lock, flags);
1179
1180 p9_poll_mux(conn);
1181
1182 spin_lock_irqsave(&p9_poll_lock, flags);
1183 }
1184 spin_unlock_irqrestore(&p9_poll_lock, flags);
1185
1186 p9_debug(P9_DEBUG_TRANS, "finish\n");
1187 }
1188
p9_trans_fd_init(void)1189 static int __init p9_trans_fd_init(void)
1190 {
1191 v9fs_register_trans(&p9_tcp_trans);
1192 v9fs_register_trans(&p9_unix_trans);
1193 v9fs_register_trans(&p9_fd_trans);
1194
1195 return 0;
1196 }
1197
p9_trans_fd_exit(void)1198 static void __exit p9_trans_fd_exit(void)
1199 {
1200 flush_work(&p9_poll_work);
1201 v9fs_unregister_trans(&p9_tcp_trans);
1202 v9fs_unregister_trans(&p9_unix_trans);
1203 v9fs_unregister_trans(&p9_fd_trans);
1204 }
1205
1206 module_init(p9_trans_fd_init);
1207 module_exit(p9_trans_fd_exit);
1208
1209 MODULE_AUTHOR("Eric Van Hensbergen <ericvh@gmail.com>");
1210 MODULE_DESCRIPTION("Filedescriptor Transport for 9P");
1211 MODULE_LICENSE("GPL");
1212