[go: nahoru, domu]

1/*
2 * The USB Monitor, inspired by Dave Harding's USBMon.
3 *
4 * This is a binary format reader.
5 *
6 * Copyright (C) 2006 Paolo Abeni (paolo.abeni@email.it)
7 * Copyright (C) 2006,2007 Pete Zaitcev (zaitcev@redhat.com)
8 */
9
10#include <linux/kernel.h>
11#include <linux/types.h>
12#include <linux/fs.h>
13#include <linux/cdev.h>
14#include <linux/export.h>
15#include <linux/usb.h>
16#include <linux/poll.h>
17#include <linux/compat.h>
18#include <linux/mm.h>
19#include <linux/scatterlist.h>
20#include <linux/slab.h>
21
22#include <asm/uaccess.h>
23
24#include "usb_mon.h"
25
26/*
27 * Defined by USB 2.0 clause 9.3, table 9.2.
28 */
29#define SETUP_LEN  8
30
31/* ioctl macros */
32#define MON_IOC_MAGIC 0x92
33
34#define MON_IOCQ_URB_LEN _IO(MON_IOC_MAGIC, 1)
35/* #2 used to be MON_IOCX_URB, removed before it got into Linus tree */
36#define MON_IOCG_STATS _IOR(MON_IOC_MAGIC, 3, struct mon_bin_stats)
37#define MON_IOCT_RING_SIZE _IO(MON_IOC_MAGIC, 4)
38#define MON_IOCQ_RING_SIZE _IO(MON_IOC_MAGIC, 5)
39#define MON_IOCX_GET   _IOW(MON_IOC_MAGIC, 6, struct mon_bin_get)
40#define MON_IOCX_MFETCH _IOWR(MON_IOC_MAGIC, 7, struct mon_bin_mfetch)
41#define MON_IOCH_MFLUSH _IO(MON_IOC_MAGIC, 8)
42/* #9 was MON_IOCT_SETAPI */
43#define MON_IOCX_GETX   _IOW(MON_IOC_MAGIC, 10, struct mon_bin_get)
44
45#ifdef CONFIG_COMPAT
46#define MON_IOCX_GET32 _IOW(MON_IOC_MAGIC, 6, struct mon_bin_get32)
47#define MON_IOCX_MFETCH32 _IOWR(MON_IOC_MAGIC, 7, struct mon_bin_mfetch32)
48#define MON_IOCX_GETX32   _IOW(MON_IOC_MAGIC, 10, struct mon_bin_get32)
49#endif
50
51/*
52 * Some architectures have enormous basic pages (16KB for ia64, 64KB for ppc).
53 * But it's all right. Just use a simple way to make sure the chunk is never
54 * smaller than a page.
55 *
56 * N.B. An application does not know our chunk size.
57 *
58 * Woops, get_zeroed_page() returns a single page. I guess we're stuck with
59 * page-sized chunks for the time being.
60 */
61#define CHUNK_SIZE   PAGE_SIZE
62#define CHUNK_ALIGN(x)   (((x)+CHUNK_SIZE-1) & ~(CHUNK_SIZE-1))
63
64/*
65 * The magic limit was calculated so that it allows the monitoring
66 * application to pick data once in two ticks. This way, another application,
67 * which presumably drives the bus, gets to hog CPU, yet we collect our data.
68 * If HZ is 100, a 480 mbit/s bus drives 614 KB every jiffy. USB has an
69 * enormous overhead built into the bus protocol, so we need about 1000 KB.
70 *
71 * This is still too much for most cases, where we just snoop a few
72 * descriptor fetches for enumeration. So, the default is a "reasonable"
73 * amount for systems with HZ=250 and incomplete bus saturation.
74 *
75 * XXX What about multi-megabyte URBs which take minutes to transfer?
76 */
77#define BUFF_MAX  CHUNK_ALIGN(1200*1024)
78#define BUFF_DFL   CHUNK_ALIGN(300*1024)
79#define BUFF_MIN     CHUNK_ALIGN(8*1024)
80
81/*
82 * The per-event API header (2 per URB).
83 *
84 * This structure is seen in userland as defined by the documentation.
85 */
86struct mon_bin_hdr {
87	u64 id;			/* URB ID - from submission to callback */
88	unsigned char type;	/* Same as in text API; extensible. */
89	unsigned char xfer_type;	/* ISO, Intr, Control, Bulk */
90	unsigned char epnum;	/* Endpoint number and transfer direction */
91	unsigned char devnum;	/* Device address */
92	unsigned short busnum;	/* Bus number */
93	char flag_setup;
94	char flag_data;
95	s64 ts_sec;		/* gettimeofday */
96	s32 ts_usec;		/* gettimeofday */
97	int status;
98	unsigned int len_urb;	/* Length of data (submitted or actual) */
99	unsigned int len_cap;	/* Delivered length */
100	union {
101		unsigned char setup[SETUP_LEN];	/* Only for Control S-type */
102		struct iso_rec {
103			int error_count;
104			int numdesc;
105		} iso;
106	} s;
107	int interval;
108	int start_frame;
109	unsigned int xfer_flags;
110	unsigned int ndesc;	/* Actual number of ISO descriptors */
111};
112
113/*
114 * ISO vector, packed into the head of data stream.
115 * This has to take 16 bytes to make sure that the end of buffer
116 * wrap is not happening in the middle of a descriptor.
117 */
118struct mon_bin_isodesc {
119	int          iso_status;
120	unsigned int iso_off;
121	unsigned int iso_len;
122	u32 _pad;
123};
124
125/* per file statistic */
126struct mon_bin_stats {
127	u32 queued;
128	u32 dropped;
129};
130
131struct mon_bin_get {
132	struct mon_bin_hdr __user *hdr;	/* Can be 48 bytes or 64. */
133	void __user *data;
134	size_t alloc;		/* Length of data (can be zero) */
135};
136
137struct mon_bin_mfetch {
138	u32 __user *offvec;	/* Vector of events fetched */
139	u32 nfetch;		/* Number of events to fetch (out: fetched) */
140	u32 nflush;		/* Number of events to flush */
141};
142
143#ifdef CONFIG_COMPAT
144struct mon_bin_get32 {
145	u32 hdr32;
146	u32 data32;
147	u32 alloc32;
148};
149
150struct mon_bin_mfetch32 {
151        u32 offvec32;
152        u32 nfetch32;
153        u32 nflush32;
154};
155#endif
156
157/* Having these two values same prevents wrapping of the mon_bin_hdr */
158#define PKT_ALIGN   64
159#define PKT_SIZE    64
160
161#define PKT_SZ_API0 48	/* API 0 (2.6.20) size */
162#define PKT_SZ_API1 64	/* API 1 size: extra fields */
163
164#define ISODESC_MAX   128	/* Same number as usbfs allows, 2048 bytes. */
165
166/* max number of USB bus supported */
167#define MON_BIN_MAX_MINOR 128
168
169/*
170 * The buffer: map of used pages.
171 */
172struct mon_pgmap {
173	struct page *pg;
174	unsigned char *ptr;	/* XXX just use page_to_virt everywhere? */
175};
176
177/*
178 * This gets associated with an open file struct.
179 */
180struct mon_reader_bin {
181	/* The buffer: one per open. */
182	spinlock_t b_lock;		/* Protect b_cnt, b_in */
183	unsigned int b_size;		/* Current size of the buffer - bytes */
184	unsigned int b_cnt;		/* Bytes used */
185	unsigned int b_in, b_out;	/* Offsets into buffer - bytes */
186	unsigned int b_read;		/* Amount of read data in curr. pkt. */
187	struct mon_pgmap *b_vec;	/* The map array */
188	wait_queue_head_t b_wait;	/* Wait for data here */
189
190	struct mutex fetch_lock;	/* Protect b_read, b_out */
191	int mmap_active;
192
193	/* A list of these is needed for "bus 0". Some time later. */
194	struct mon_reader r;
195
196	/* Stats */
197	unsigned int cnt_lost;
198};
199
200static inline struct mon_bin_hdr *MON_OFF2HDR(const struct mon_reader_bin *rp,
201    unsigned int offset)
202{
203	return (struct mon_bin_hdr *)
204	    (rp->b_vec[offset / CHUNK_SIZE].ptr + offset % CHUNK_SIZE);
205}
206
207#define MON_RING_EMPTY(rp)	((rp)->b_cnt == 0)
208
209static unsigned char xfer_to_pipe[4] = {
210	PIPE_CONTROL, PIPE_ISOCHRONOUS, PIPE_BULK, PIPE_INTERRUPT
211};
212
213static struct class *mon_bin_class;
214static dev_t mon_bin_dev0;
215static struct cdev mon_bin_cdev;
216
217static void mon_buff_area_fill(const struct mon_reader_bin *rp,
218    unsigned int offset, unsigned int size);
219static int mon_bin_wait_event(struct file *file, struct mon_reader_bin *rp);
220static int mon_alloc_buff(struct mon_pgmap *map, int npages);
221static void mon_free_buff(struct mon_pgmap *map, int npages);
222
223/*
224 * This is a "chunked memcpy". It does not manipulate any counters.
225 */
226static unsigned int mon_copy_to_buff(const struct mon_reader_bin *this,
227    unsigned int off, const unsigned char *from, unsigned int length)
228{
229	unsigned int step_len;
230	unsigned char *buf;
231	unsigned int in_page;
232
233	while (length) {
234		/*
235		 * Determine step_len.
236		 */
237		step_len = length;
238		in_page = CHUNK_SIZE - (off & (CHUNK_SIZE-1));
239		if (in_page < step_len)
240			step_len = in_page;
241
242		/*
243		 * Copy data and advance pointers.
244		 */
245		buf = this->b_vec[off / CHUNK_SIZE].ptr + off % CHUNK_SIZE;
246		memcpy(buf, from, step_len);
247		if ((off += step_len) >= this->b_size) off = 0;
248		from += step_len;
249		length -= step_len;
250	}
251	return off;
252}
253
254/*
255 * This is a little worse than the above because it's "chunked copy_to_user".
256 * The return value is an error code, not an offset.
257 */
258static int copy_from_buf(const struct mon_reader_bin *this, unsigned int off,
259    char __user *to, int length)
260{
261	unsigned int step_len;
262	unsigned char *buf;
263	unsigned int in_page;
264
265	while (length) {
266		/*
267		 * Determine step_len.
268		 */
269		step_len = length;
270		in_page = CHUNK_SIZE - (off & (CHUNK_SIZE-1));
271		if (in_page < step_len)
272			step_len = in_page;
273
274		/*
275		 * Copy data and advance pointers.
276		 */
277		buf = this->b_vec[off / CHUNK_SIZE].ptr + off % CHUNK_SIZE;
278		if (copy_to_user(to, buf, step_len))
279			return -EINVAL;
280		if ((off += step_len) >= this->b_size) off = 0;
281		to += step_len;
282		length -= step_len;
283	}
284	return 0;
285}
286
287/*
288 * Allocate an (aligned) area in the buffer.
289 * This is called under b_lock.
290 * Returns ~0 on failure.
291 */
292static unsigned int mon_buff_area_alloc(struct mon_reader_bin *rp,
293    unsigned int size)
294{
295	unsigned int offset;
296
297	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
298	if (rp->b_cnt + size > rp->b_size)
299		return ~0;
300	offset = rp->b_in;
301	rp->b_cnt += size;
302	if ((rp->b_in += size) >= rp->b_size)
303		rp->b_in -= rp->b_size;
304	return offset;
305}
306
307/*
308 * This is the same thing as mon_buff_area_alloc, only it does not allow
309 * buffers to wrap. This is needed by applications which pass references
310 * into mmap-ed buffers up their stacks (libpcap can do that).
311 *
312 * Currently, we always have the header stuck with the data, although
313 * it is not strictly speaking necessary.
314 *
315 * When a buffer would wrap, we place a filler packet to mark the space.
316 */
317static unsigned int mon_buff_area_alloc_contiguous(struct mon_reader_bin *rp,
318    unsigned int size)
319{
320	unsigned int offset;
321	unsigned int fill_size;
322
323	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
324	if (rp->b_cnt + size > rp->b_size)
325		return ~0;
326	if (rp->b_in + size > rp->b_size) {
327		/*
328		 * This would wrap. Find if we still have space after
329		 * skipping to the end of the buffer. If we do, place
330		 * a filler packet and allocate a new packet.
331		 */
332		fill_size = rp->b_size - rp->b_in;
333		if (rp->b_cnt + size + fill_size > rp->b_size)
334			return ~0;
335		mon_buff_area_fill(rp, rp->b_in, fill_size);
336
337		offset = 0;
338		rp->b_in = size;
339		rp->b_cnt += size + fill_size;
340	} else if (rp->b_in + size == rp->b_size) {
341		offset = rp->b_in;
342		rp->b_in = 0;
343		rp->b_cnt += size;
344	} else {
345		offset = rp->b_in;
346		rp->b_in += size;
347		rp->b_cnt += size;
348	}
349	return offset;
350}
351
352/*
353 * Return a few (kilo-)bytes to the head of the buffer.
354 * This is used if a data fetch fails.
355 */
356static void mon_buff_area_shrink(struct mon_reader_bin *rp, unsigned int size)
357{
358
359	/* size &= ~(PKT_ALIGN-1);  -- we're called with aligned size */
360	rp->b_cnt -= size;
361	if (rp->b_in < size)
362		rp->b_in += rp->b_size;
363	rp->b_in -= size;
364}
365
366/*
367 * This has to be called under both b_lock and fetch_lock, because
368 * it accesses both b_cnt and b_out.
369 */
370static void mon_buff_area_free(struct mon_reader_bin *rp, unsigned int size)
371{
372
373	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
374	rp->b_cnt -= size;
375	if ((rp->b_out += size) >= rp->b_size)
376		rp->b_out -= rp->b_size;
377}
378
379static void mon_buff_area_fill(const struct mon_reader_bin *rp,
380    unsigned int offset, unsigned int size)
381{
382	struct mon_bin_hdr *ep;
383
384	ep = MON_OFF2HDR(rp, offset);
385	memset(ep, 0, PKT_SIZE);
386	ep->type = '@';
387	ep->len_cap = size - PKT_SIZE;
388}
389
390static inline char mon_bin_get_setup(unsigned char *setupb,
391    const struct urb *urb, char ev_type)
392{
393
394	if (urb->setup_packet == NULL)
395		return 'Z';
396	memcpy(setupb, urb->setup_packet, SETUP_LEN);
397	return 0;
398}
399
400static unsigned int mon_bin_get_data(const struct mon_reader_bin *rp,
401    unsigned int offset, struct urb *urb, unsigned int length,
402    char *flag)
403{
404	int i;
405	struct scatterlist *sg;
406	unsigned int this_len;
407
408	*flag = 0;
409	if (urb->num_sgs == 0) {
410		if (urb->transfer_buffer == NULL) {
411			*flag = 'Z';
412			return length;
413		}
414		mon_copy_to_buff(rp, offset, urb->transfer_buffer, length);
415		length = 0;
416
417	} else {
418		/* If IOMMU coalescing occurred, we cannot trust sg_page */
419		if (urb->transfer_flags & URB_DMA_SG_COMBINED) {
420			*flag = 'D';
421			return length;
422		}
423
424		/* Copy up to the first non-addressable segment */
425		for_each_sg(urb->sg, sg, urb->num_sgs, i) {
426			if (length == 0 || PageHighMem(sg_page(sg)))
427				break;
428			this_len = min_t(unsigned int, sg->length, length);
429			offset = mon_copy_to_buff(rp, offset, sg_virt(sg),
430					this_len);
431			length -= this_len;
432		}
433		if (i == 0)
434			*flag = 'D';
435	}
436
437	return length;
438}
439
440/*
441 * This is the look-ahead pass in case of 'C Zi', when actual_length cannot
442 * be used to determine the length of the whole contiguous buffer.
443 */
444static unsigned int mon_bin_collate_isodesc(const struct mon_reader_bin *rp,
445    struct urb *urb, unsigned int ndesc)
446{
447	struct usb_iso_packet_descriptor *fp;
448	unsigned int length;
449
450	length = 0;
451	fp = urb->iso_frame_desc;
452	while (ndesc-- != 0) {
453		if (fp->actual_length != 0) {
454			if (fp->offset + fp->actual_length > length)
455				length = fp->offset + fp->actual_length;
456		}
457		fp++;
458	}
459	return length;
460}
461
462static void mon_bin_get_isodesc(const struct mon_reader_bin *rp,
463    unsigned int offset, struct urb *urb, char ev_type, unsigned int ndesc)
464{
465	struct mon_bin_isodesc *dp;
466	struct usb_iso_packet_descriptor *fp;
467
468	fp = urb->iso_frame_desc;
469	while (ndesc-- != 0) {
470		dp = (struct mon_bin_isodesc *)
471		    (rp->b_vec[offset / CHUNK_SIZE].ptr + offset % CHUNK_SIZE);
472		dp->iso_status = fp->status;
473		dp->iso_off = fp->offset;
474		dp->iso_len = (ev_type == 'S') ? fp->length : fp->actual_length;
475		dp->_pad = 0;
476		if ((offset += sizeof(struct mon_bin_isodesc)) >= rp->b_size)
477			offset = 0;
478		fp++;
479	}
480}
481
482static void mon_bin_event(struct mon_reader_bin *rp, struct urb *urb,
483    char ev_type, int status)
484{
485	const struct usb_endpoint_descriptor *epd = &urb->ep->desc;
486	struct timeval ts;
487	unsigned long flags;
488	unsigned int urb_length;
489	unsigned int offset;
490	unsigned int length;
491	unsigned int delta;
492	unsigned int ndesc, lendesc;
493	unsigned char dir;
494	struct mon_bin_hdr *ep;
495	char data_tag = 0;
496
497	do_gettimeofday(&ts);
498
499	spin_lock_irqsave(&rp->b_lock, flags);
500
501	/*
502	 * Find the maximum allowable length, then allocate space.
503	 */
504	urb_length = (ev_type == 'S') ?
505	    urb->transfer_buffer_length : urb->actual_length;
506	length = urb_length;
507
508	if (usb_endpoint_xfer_isoc(epd)) {
509		if (urb->number_of_packets < 0) {
510			ndesc = 0;
511		} else if (urb->number_of_packets >= ISODESC_MAX) {
512			ndesc = ISODESC_MAX;
513		} else {
514			ndesc = urb->number_of_packets;
515		}
516		if (ev_type == 'C' && usb_urb_dir_in(urb))
517			length = mon_bin_collate_isodesc(rp, urb, ndesc);
518	} else {
519		ndesc = 0;
520	}
521	lendesc = ndesc*sizeof(struct mon_bin_isodesc);
522
523	/* not an issue unless there's a subtle bug in a HCD somewhere */
524	if (length >= urb->transfer_buffer_length)
525		length = urb->transfer_buffer_length;
526
527	if (length >= rp->b_size/5)
528		length = rp->b_size/5;
529
530	if (usb_urb_dir_in(urb)) {
531		if (ev_type == 'S') {
532			length = 0;
533			data_tag = '<';
534		}
535		/* Cannot rely on endpoint number in case of control ep.0 */
536		dir = USB_DIR_IN;
537	} else {
538		if (ev_type == 'C') {
539			length = 0;
540			data_tag = '>';
541		}
542		dir = 0;
543	}
544
545	if (rp->mmap_active) {
546		offset = mon_buff_area_alloc_contiguous(rp,
547						 length + PKT_SIZE + lendesc);
548	} else {
549		offset = mon_buff_area_alloc(rp, length + PKT_SIZE + lendesc);
550	}
551	if (offset == ~0) {
552		rp->cnt_lost++;
553		spin_unlock_irqrestore(&rp->b_lock, flags);
554		return;
555	}
556
557	ep = MON_OFF2HDR(rp, offset);
558	if ((offset += PKT_SIZE) >= rp->b_size) offset = 0;
559
560	/*
561	 * Fill the allocated area.
562	 */
563	memset(ep, 0, PKT_SIZE);
564	ep->type = ev_type;
565	ep->xfer_type = xfer_to_pipe[usb_endpoint_type(epd)];
566	ep->epnum = dir | usb_endpoint_num(epd);
567	ep->devnum = urb->dev->devnum;
568	ep->busnum = urb->dev->bus->busnum;
569	ep->id = (unsigned long) urb;
570	ep->ts_sec = ts.tv_sec;
571	ep->ts_usec = ts.tv_usec;
572	ep->status = status;
573	ep->len_urb = urb_length;
574	ep->len_cap = length + lendesc;
575	ep->xfer_flags = urb->transfer_flags;
576
577	if (usb_endpoint_xfer_int(epd)) {
578		ep->interval = urb->interval;
579	} else if (usb_endpoint_xfer_isoc(epd)) {
580		ep->interval = urb->interval;
581		ep->start_frame = urb->start_frame;
582		ep->s.iso.error_count = urb->error_count;
583		ep->s.iso.numdesc = urb->number_of_packets;
584	}
585
586	if (usb_endpoint_xfer_control(epd) && ev_type == 'S') {
587		ep->flag_setup = mon_bin_get_setup(ep->s.setup, urb, ev_type);
588	} else {
589		ep->flag_setup = '-';
590	}
591
592	if (ndesc != 0) {
593		ep->ndesc = ndesc;
594		mon_bin_get_isodesc(rp, offset, urb, ev_type, ndesc);
595		if ((offset += lendesc) >= rp->b_size)
596			offset -= rp->b_size;
597	}
598
599	if (length != 0) {
600		length = mon_bin_get_data(rp, offset, urb, length,
601				&ep->flag_data);
602		if (length > 0) {
603			delta = (ep->len_cap + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
604			ep->len_cap -= length;
605			delta -= (ep->len_cap + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
606			mon_buff_area_shrink(rp, delta);
607		}
608	} else {
609		ep->flag_data = data_tag;
610	}
611
612	spin_unlock_irqrestore(&rp->b_lock, flags);
613
614	wake_up(&rp->b_wait);
615}
616
617static void mon_bin_submit(void *data, struct urb *urb)
618{
619	struct mon_reader_bin *rp = data;
620	mon_bin_event(rp, urb, 'S', -EINPROGRESS);
621}
622
623static void mon_bin_complete(void *data, struct urb *urb, int status)
624{
625	struct mon_reader_bin *rp = data;
626	mon_bin_event(rp, urb, 'C', status);
627}
628
629static void mon_bin_error(void *data, struct urb *urb, int error)
630{
631	struct mon_reader_bin *rp = data;
632	struct timeval ts;
633	unsigned long flags;
634	unsigned int offset;
635	struct mon_bin_hdr *ep;
636
637	do_gettimeofday(&ts);
638
639	spin_lock_irqsave(&rp->b_lock, flags);
640
641	offset = mon_buff_area_alloc(rp, PKT_SIZE);
642	if (offset == ~0) {
643		/* Not incrementing cnt_lost. Just because. */
644		spin_unlock_irqrestore(&rp->b_lock, flags);
645		return;
646	}
647
648	ep = MON_OFF2HDR(rp, offset);
649
650	memset(ep, 0, PKT_SIZE);
651	ep->type = 'E';
652	ep->xfer_type = xfer_to_pipe[usb_endpoint_type(&urb->ep->desc)];
653	ep->epnum = usb_urb_dir_in(urb) ? USB_DIR_IN : 0;
654	ep->epnum |= usb_endpoint_num(&urb->ep->desc);
655	ep->devnum = urb->dev->devnum;
656	ep->busnum = urb->dev->bus->busnum;
657	ep->id = (unsigned long) urb;
658	ep->ts_sec = ts.tv_sec;
659	ep->ts_usec = ts.tv_usec;
660	ep->status = error;
661
662	ep->flag_setup = '-';
663	ep->flag_data = 'E';
664
665	spin_unlock_irqrestore(&rp->b_lock, flags);
666
667	wake_up(&rp->b_wait);
668}
669
670static int mon_bin_open(struct inode *inode, struct file *file)
671{
672	struct mon_bus *mbus;
673	struct mon_reader_bin *rp;
674	size_t size;
675	int rc;
676
677	mutex_lock(&mon_lock);
678	if ((mbus = mon_bus_lookup(iminor(inode))) == NULL) {
679		mutex_unlock(&mon_lock);
680		return -ENODEV;
681	}
682	if (mbus != &mon_bus0 && mbus->u_bus == NULL) {
683		printk(KERN_ERR TAG ": consistency error on open\n");
684		mutex_unlock(&mon_lock);
685		return -ENODEV;
686	}
687
688	rp = kzalloc(sizeof(struct mon_reader_bin), GFP_KERNEL);
689	if (rp == NULL) {
690		rc = -ENOMEM;
691		goto err_alloc;
692	}
693	spin_lock_init(&rp->b_lock);
694	init_waitqueue_head(&rp->b_wait);
695	mutex_init(&rp->fetch_lock);
696	rp->b_size = BUFF_DFL;
697
698	size = sizeof(struct mon_pgmap) * (rp->b_size/CHUNK_SIZE);
699	if ((rp->b_vec = kzalloc(size, GFP_KERNEL)) == NULL) {
700		rc = -ENOMEM;
701		goto err_allocvec;
702	}
703
704	if ((rc = mon_alloc_buff(rp->b_vec, rp->b_size/CHUNK_SIZE)) < 0)
705		goto err_allocbuff;
706
707	rp->r.m_bus = mbus;
708	rp->r.r_data = rp;
709	rp->r.rnf_submit = mon_bin_submit;
710	rp->r.rnf_error = mon_bin_error;
711	rp->r.rnf_complete = mon_bin_complete;
712
713	mon_reader_add(mbus, &rp->r);
714
715	file->private_data = rp;
716	mutex_unlock(&mon_lock);
717	return 0;
718
719err_allocbuff:
720	kfree(rp->b_vec);
721err_allocvec:
722	kfree(rp);
723err_alloc:
724	mutex_unlock(&mon_lock);
725	return rc;
726}
727
728/*
729 * Extract an event from buffer and copy it to user space.
730 * Wait if there is no event ready.
731 * Returns zero or error.
732 */
733static int mon_bin_get_event(struct file *file, struct mon_reader_bin *rp,
734    struct mon_bin_hdr __user *hdr, unsigned int hdrbytes,
735    void __user *data, unsigned int nbytes)
736{
737	unsigned long flags;
738	struct mon_bin_hdr *ep;
739	size_t step_len;
740	unsigned int offset;
741	int rc;
742
743	mutex_lock(&rp->fetch_lock);
744
745	if ((rc = mon_bin_wait_event(file, rp)) < 0) {
746		mutex_unlock(&rp->fetch_lock);
747		return rc;
748	}
749
750	ep = MON_OFF2HDR(rp, rp->b_out);
751
752	if (copy_to_user(hdr, ep, hdrbytes)) {
753		mutex_unlock(&rp->fetch_lock);
754		return -EFAULT;
755	}
756
757	step_len = min(ep->len_cap, nbytes);
758	if ((offset = rp->b_out + PKT_SIZE) >= rp->b_size) offset = 0;
759
760	if (copy_from_buf(rp, offset, data, step_len)) {
761		mutex_unlock(&rp->fetch_lock);
762		return -EFAULT;
763	}
764
765	spin_lock_irqsave(&rp->b_lock, flags);
766	mon_buff_area_free(rp, PKT_SIZE + ep->len_cap);
767	spin_unlock_irqrestore(&rp->b_lock, flags);
768	rp->b_read = 0;
769
770	mutex_unlock(&rp->fetch_lock);
771	return 0;
772}
773
774static int mon_bin_release(struct inode *inode, struct file *file)
775{
776	struct mon_reader_bin *rp = file->private_data;
777	struct mon_bus* mbus = rp->r.m_bus;
778
779	mutex_lock(&mon_lock);
780
781	if (mbus->nreaders <= 0) {
782		printk(KERN_ERR TAG ": consistency error on close\n");
783		mutex_unlock(&mon_lock);
784		return 0;
785	}
786	mon_reader_del(mbus, &rp->r);
787
788	mon_free_buff(rp->b_vec, rp->b_size/CHUNK_SIZE);
789	kfree(rp->b_vec);
790	kfree(rp);
791
792	mutex_unlock(&mon_lock);
793	return 0;
794}
795
796static ssize_t mon_bin_read(struct file *file, char __user *buf,
797    size_t nbytes, loff_t *ppos)
798{
799	struct mon_reader_bin *rp = file->private_data;
800	unsigned int hdrbytes = PKT_SZ_API0;
801	unsigned long flags;
802	struct mon_bin_hdr *ep;
803	unsigned int offset;
804	size_t step_len;
805	char *ptr;
806	ssize_t done = 0;
807	int rc;
808
809	mutex_lock(&rp->fetch_lock);
810
811	if ((rc = mon_bin_wait_event(file, rp)) < 0) {
812		mutex_unlock(&rp->fetch_lock);
813		return rc;
814	}
815
816	ep = MON_OFF2HDR(rp, rp->b_out);
817
818	if (rp->b_read < hdrbytes) {
819		step_len = min(nbytes, (size_t)(hdrbytes - rp->b_read));
820		ptr = ((char *)ep) + rp->b_read;
821		if (step_len && copy_to_user(buf, ptr, step_len)) {
822			mutex_unlock(&rp->fetch_lock);
823			return -EFAULT;
824		}
825		nbytes -= step_len;
826		buf += step_len;
827		rp->b_read += step_len;
828		done += step_len;
829	}
830
831	if (rp->b_read >= hdrbytes) {
832		step_len = ep->len_cap;
833		step_len -= rp->b_read - hdrbytes;
834		if (step_len > nbytes)
835			step_len = nbytes;
836		offset = rp->b_out + PKT_SIZE;
837		offset += rp->b_read - hdrbytes;
838		if (offset >= rp->b_size)
839			offset -= rp->b_size;
840		if (copy_from_buf(rp, offset, buf, step_len)) {
841			mutex_unlock(&rp->fetch_lock);
842			return -EFAULT;
843		}
844		nbytes -= step_len;
845		buf += step_len;
846		rp->b_read += step_len;
847		done += step_len;
848	}
849
850	/*
851	 * Check if whole packet was read, and if so, jump to the next one.
852	 */
853	if (rp->b_read >= hdrbytes + ep->len_cap) {
854		spin_lock_irqsave(&rp->b_lock, flags);
855		mon_buff_area_free(rp, PKT_SIZE + ep->len_cap);
856		spin_unlock_irqrestore(&rp->b_lock, flags);
857		rp->b_read = 0;
858	}
859
860	mutex_unlock(&rp->fetch_lock);
861	return done;
862}
863
864/*
865 * Remove at most nevents from chunked buffer.
866 * Returns the number of removed events.
867 */
868static int mon_bin_flush(struct mon_reader_bin *rp, unsigned nevents)
869{
870	unsigned long flags;
871	struct mon_bin_hdr *ep;
872	int i;
873
874	mutex_lock(&rp->fetch_lock);
875	spin_lock_irqsave(&rp->b_lock, flags);
876	for (i = 0; i < nevents; ++i) {
877		if (MON_RING_EMPTY(rp))
878			break;
879
880		ep = MON_OFF2HDR(rp, rp->b_out);
881		mon_buff_area_free(rp, PKT_SIZE + ep->len_cap);
882	}
883	spin_unlock_irqrestore(&rp->b_lock, flags);
884	rp->b_read = 0;
885	mutex_unlock(&rp->fetch_lock);
886	return i;
887}
888
889/*
890 * Fetch at most max event offsets into the buffer and put them into vec.
891 * The events are usually freed later with mon_bin_flush.
892 * Return the effective number of events fetched.
893 */
894static int mon_bin_fetch(struct file *file, struct mon_reader_bin *rp,
895    u32 __user *vec, unsigned int max)
896{
897	unsigned int cur_out;
898	unsigned int bytes, avail;
899	unsigned int size;
900	unsigned int nevents;
901	struct mon_bin_hdr *ep;
902	unsigned long flags;
903	int rc;
904
905	mutex_lock(&rp->fetch_lock);
906
907	if ((rc = mon_bin_wait_event(file, rp)) < 0) {
908		mutex_unlock(&rp->fetch_lock);
909		return rc;
910	}
911
912	spin_lock_irqsave(&rp->b_lock, flags);
913	avail = rp->b_cnt;
914	spin_unlock_irqrestore(&rp->b_lock, flags);
915
916	cur_out = rp->b_out;
917	nevents = 0;
918	bytes = 0;
919	while (bytes < avail) {
920		if (nevents >= max)
921			break;
922
923		ep = MON_OFF2HDR(rp, cur_out);
924		if (put_user(cur_out, &vec[nevents])) {
925			mutex_unlock(&rp->fetch_lock);
926			return -EFAULT;
927		}
928
929		nevents++;
930		size = ep->len_cap + PKT_SIZE;
931		size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
932		if ((cur_out += size) >= rp->b_size)
933			cur_out -= rp->b_size;
934		bytes += size;
935	}
936
937	mutex_unlock(&rp->fetch_lock);
938	return nevents;
939}
940
941/*
942 * Count events. This is almost the same as the above mon_bin_fetch,
943 * only we do not store offsets into user vector, and we have no limit.
944 */
945static int mon_bin_queued(struct mon_reader_bin *rp)
946{
947	unsigned int cur_out;
948	unsigned int bytes, avail;
949	unsigned int size;
950	unsigned int nevents;
951	struct mon_bin_hdr *ep;
952	unsigned long flags;
953
954	mutex_lock(&rp->fetch_lock);
955
956	spin_lock_irqsave(&rp->b_lock, flags);
957	avail = rp->b_cnt;
958	spin_unlock_irqrestore(&rp->b_lock, flags);
959
960	cur_out = rp->b_out;
961	nevents = 0;
962	bytes = 0;
963	while (bytes < avail) {
964		ep = MON_OFF2HDR(rp, cur_out);
965
966		nevents++;
967		size = ep->len_cap + PKT_SIZE;
968		size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
969		if ((cur_out += size) >= rp->b_size)
970			cur_out -= rp->b_size;
971		bytes += size;
972	}
973
974	mutex_unlock(&rp->fetch_lock);
975	return nevents;
976}
977
978/*
979 */
980static long mon_bin_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
981{
982	struct mon_reader_bin *rp = file->private_data;
983	// struct mon_bus* mbus = rp->r.m_bus;
984	int ret = 0;
985	struct mon_bin_hdr *ep;
986	unsigned long flags;
987
988	switch (cmd) {
989
990	case MON_IOCQ_URB_LEN:
991		/*
992		 * N.B. This only returns the size of data, without the header.
993		 */
994		spin_lock_irqsave(&rp->b_lock, flags);
995		if (!MON_RING_EMPTY(rp)) {
996			ep = MON_OFF2HDR(rp, rp->b_out);
997			ret = ep->len_cap;
998		}
999		spin_unlock_irqrestore(&rp->b_lock, flags);
1000		break;
1001
1002	case MON_IOCQ_RING_SIZE:
1003		ret = rp->b_size;
1004		break;
1005
1006	case MON_IOCT_RING_SIZE:
1007		/*
1008		 * Changing the buffer size will flush it's contents; the new
1009		 * buffer is allocated before releasing the old one to be sure
1010		 * the device will stay functional also in case of memory
1011		 * pressure.
1012		 */
1013		{
1014		int size;
1015		struct mon_pgmap *vec;
1016
1017		if (arg < BUFF_MIN || arg > BUFF_MAX)
1018			return -EINVAL;
1019
1020		size = CHUNK_ALIGN(arg);
1021		if ((vec = kzalloc(sizeof(struct mon_pgmap) * (size/CHUNK_SIZE),
1022		    GFP_KERNEL)) == NULL) {
1023			ret = -ENOMEM;
1024			break;
1025		}
1026
1027		ret = mon_alloc_buff(vec, size/CHUNK_SIZE);
1028		if (ret < 0) {
1029			kfree(vec);
1030			break;
1031		}
1032
1033		mutex_lock(&rp->fetch_lock);
1034		spin_lock_irqsave(&rp->b_lock, flags);
1035		mon_free_buff(rp->b_vec, rp->b_size/CHUNK_SIZE);
1036		kfree(rp->b_vec);
1037		rp->b_vec  = vec;
1038		rp->b_size = size;
1039		rp->b_read = rp->b_in = rp->b_out = rp->b_cnt = 0;
1040		rp->cnt_lost = 0;
1041		spin_unlock_irqrestore(&rp->b_lock, flags);
1042		mutex_unlock(&rp->fetch_lock);
1043		}
1044		break;
1045
1046	case MON_IOCH_MFLUSH:
1047		ret = mon_bin_flush(rp, arg);
1048		break;
1049
1050	case MON_IOCX_GET:
1051	case MON_IOCX_GETX:
1052		{
1053		struct mon_bin_get getb;
1054
1055		if (copy_from_user(&getb, (void __user *)arg,
1056					    sizeof(struct mon_bin_get)))
1057			return -EFAULT;
1058
1059		if (getb.alloc > 0x10000000)	/* Want to cast to u32 */
1060			return -EINVAL;
1061		ret = mon_bin_get_event(file, rp, getb.hdr,
1062		    (cmd == MON_IOCX_GET)? PKT_SZ_API0: PKT_SZ_API1,
1063		    getb.data, (unsigned int)getb.alloc);
1064		}
1065		break;
1066
1067	case MON_IOCX_MFETCH:
1068		{
1069		struct mon_bin_mfetch mfetch;
1070		struct mon_bin_mfetch __user *uptr;
1071
1072		uptr = (struct mon_bin_mfetch __user *)arg;
1073
1074		if (copy_from_user(&mfetch, uptr, sizeof(mfetch)))
1075			return -EFAULT;
1076
1077		if (mfetch.nflush) {
1078			ret = mon_bin_flush(rp, mfetch.nflush);
1079			if (ret < 0)
1080				return ret;
1081			if (put_user(ret, &uptr->nflush))
1082				return -EFAULT;
1083		}
1084		ret = mon_bin_fetch(file, rp, mfetch.offvec, mfetch.nfetch);
1085		if (ret < 0)
1086			return ret;
1087		if (put_user(ret, &uptr->nfetch))
1088			return -EFAULT;
1089		ret = 0;
1090		}
1091		break;
1092
1093	case MON_IOCG_STATS: {
1094		struct mon_bin_stats __user *sp;
1095		unsigned int nevents;
1096		unsigned int ndropped;
1097
1098		spin_lock_irqsave(&rp->b_lock, flags);
1099		ndropped = rp->cnt_lost;
1100		rp->cnt_lost = 0;
1101		spin_unlock_irqrestore(&rp->b_lock, flags);
1102		nevents = mon_bin_queued(rp);
1103
1104		sp = (struct mon_bin_stats __user *)arg;
1105		if (put_user(ndropped, &sp->dropped))
1106			return -EFAULT;
1107		if (put_user(nevents, &sp->queued))
1108			return -EFAULT;
1109
1110		}
1111		break;
1112
1113	default:
1114		return -ENOTTY;
1115	}
1116
1117	return ret;
1118}
1119
1120#ifdef CONFIG_COMPAT
1121static long mon_bin_compat_ioctl(struct file *file,
1122    unsigned int cmd, unsigned long arg)
1123{
1124	struct mon_reader_bin *rp = file->private_data;
1125	int ret;
1126
1127	switch (cmd) {
1128
1129	case MON_IOCX_GET32:
1130	case MON_IOCX_GETX32:
1131		{
1132		struct mon_bin_get32 getb;
1133
1134		if (copy_from_user(&getb, (void __user *)arg,
1135					    sizeof(struct mon_bin_get32)))
1136			return -EFAULT;
1137
1138		ret = mon_bin_get_event(file, rp, compat_ptr(getb.hdr32),
1139		    (cmd == MON_IOCX_GET32)? PKT_SZ_API0: PKT_SZ_API1,
1140		    compat_ptr(getb.data32), getb.alloc32);
1141		if (ret < 0)
1142			return ret;
1143		}
1144		return 0;
1145
1146	case MON_IOCX_MFETCH32:
1147		{
1148		struct mon_bin_mfetch32 mfetch;
1149		struct mon_bin_mfetch32 __user *uptr;
1150
1151		uptr = (struct mon_bin_mfetch32 __user *) compat_ptr(arg);
1152
1153		if (copy_from_user(&mfetch, uptr, sizeof(mfetch)))
1154			return -EFAULT;
1155
1156		if (mfetch.nflush32) {
1157			ret = mon_bin_flush(rp, mfetch.nflush32);
1158			if (ret < 0)
1159				return ret;
1160			if (put_user(ret, &uptr->nflush32))
1161				return -EFAULT;
1162		}
1163		ret = mon_bin_fetch(file, rp, compat_ptr(mfetch.offvec32),
1164		    mfetch.nfetch32);
1165		if (ret < 0)
1166			return ret;
1167		if (put_user(ret, &uptr->nfetch32))
1168			return -EFAULT;
1169		}
1170		return 0;
1171
1172	case MON_IOCG_STATS:
1173		return mon_bin_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
1174
1175	case MON_IOCQ_URB_LEN:
1176	case MON_IOCQ_RING_SIZE:
1177	case MON_IOCT_RING_SIZE:
1178	case MON_IOCH_MFLUSH:
1179		return mon_bin_ioctl(file, cmd, arg);
1180
1181	default:
1182		;
1183	}
1184	return -ENOTTY;
1185}
1186#endif /* CONFIG_COMPAT */
1187
1188static unsigned int
1189mon_bin_poll(struct file *file, struct poll_table_struct *wait)
1190{
1191	struct mon_reader_bin *rp = file->private_data;
1192	unsigned int mask = 0;
1193	unsigned long flags;
1194
1195	if (file->f_mode & FMODE_READ)
1196		poll_wait(file, &rp->b_wait, wait);
1197
1198	spin_lock_irqsave(&rp->b_lock, flags);
1199	if (!MON_RING_EMPTY(rp))
1200		mask |= POLLIN | POLLRDNORM;    /* readable */
1201	spin_unlock_irqrestore(&rp->b_lock, flags);
1202	return mask;
1203}
1204
1205/*
1206 * open and close: just keep track of how many times the device is
1207 * mapped, to use the proper memory allocation function.
1208 */
1209static void mon_bin_vma_open(struct vm_area_struct *vma)
1210{
1211	struct mon_reader_bin *rp = vma->vm_private_data;
1212	rp->mmap_active++;
1213}
1214
1215static void mon_bin_vma_close(struct vm_area_struct *vma)
1216{
1217	struct mon_reader_bin *rp = vma->vm_private_data;
1218	rp->mmap_active--;
1219}
1220
1221/*
1222 * Map ring pages to user space.
1223 */
1224static int mon_bin_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1225{
1226	struct mon_reader_bin *rp = vma->vm_private_data;
1227	unsigned long offset, chunk_idx;
1228	struct page *pageptr;
1229
1230	offset = vmf->pgoff << PAGE_SHIFT;
1231	if (offset >= rp->b_size)
1232		return VM_FAULT_SIGBUS;
1233	chunk_idx = offset / CHUNK_SIZE;
1234	pageptr = rp->b_vec[chunk_idx].pg;
1235	get_page(pageptr);
1236	vmf->page = pageptr;
1237	return 0;
1238}
1239
1240static const struct vm_operations_struct mon_bin_vm_ops = {
1241	.open =     mon_bin_vma_open,
1242	.close =    mon_bin_vma_close,
1243	.fault =    mon_bin_vma_fault,
1244};
1245
1246static int mon_bin_mmap(struct file *filp, struct vm_area_struct *vma)
1247{
1248	/* don't do anything here: "fault" will set up page table entries */
1249	vma->vm_ops = &mon_bin_vm_ops;
1250	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1251	vma->vm_private_data = filp->private_data;
1252	mon_bin_vma_open(vma);
1253	return 0;
1254}
1255
1256static const struct file_operations mon_fops_binary = {
1257	.owner =	THIS_MODULE,
1258	.open =		mon_bin_open,
1259	.llseek =	no_llseek,
1260	.read =		mon_bin_read,
1261	/* .write =	mon_text_write, */
1262	.poll =		mon_bin_poll,
1263	.unlocked_ioctl = mon_bin_ioctl,
1264#ifdef CONFIG_COMPAT
1265	.compat_ioctl =	mon_bin_compat_ioctl,
1266#endif
1267	.release =	mon_bin_release,
1268	.mmap =		mon_bin_mmap,
1269};
1270
1271static int mon_bin_wait_event(struct file *file, struct mon_reader_bin *rp)
1272{
1273	DECLARE_WAITQUEUE(waita, current);
1274	unsigned long flags;
1275
1276	add_wait_queue(&rp->b_wait, &waita);
1277	set_current_state(TASK_INTERRUPTIBLE);
1278
1279	spin_lock_irqsave(&rp->b_lock, flags);
1280	while (MON_RING_EMPTY(rp)) {
1281		spin_unlock_irqrestore(&rp->b_lock, flags);
1282
1283		if (file->f_flags & O_NONBLOCK) {
1284			set_current_state(TASK_RUNNING);
1285			remove_wait_queue(&rp->b_wait, &waita);
1286			return -EWOULDBLOCK; /* Same as EAGAIN in Linux */
1287		}
1288		schedule();
1289		if (signal_pending(current)) {
1290			remove_wait_queue(&rp->b_wait, &waita);
1291			return -EINTR;
1292		}
1293		set_current_state(TASK_INTERRUPTIBLE);
1294
1295		spin_lock_irqsave(&rp->b_lock, flags);
1296	}
1297	spin_unlock_irqrestore(&rp->b_lock, flags);
1298
1299	set_current_state(TASK_RUNNING);
1300	remove_wait_queue(&rp->b_wait, &waita);
1301	return 0;
1302}
1303
1304static int mon_alloc_buff(struct mon_pgmap *map, int npages)
1305{
1306	int n;
1307	unsigned long vaddr;
1308
1309	for (n = 0; n < npages; n++) {
1310		vaddr = get_zeroed_page(GFP_KERNEL);
1311		if (vaddr == 0) {
1312			while (n-- != 0)
1313				free_page((unsigned long) map[n].ptr);
1314			return -ENOMEM;
1315		}
1316		map[n].ptr = (unsigned char *) vaddr;
1317		map[n].pg = virt_to_page((void *) vaddr);
1318	}
1319	return 0;
1320}
1321
1322static void mon_free_buff(struct mon_pgmap *map, int npages)
1323{
1324	int n;
1325
1326	for (n = 0; n < npages; n++)
1327		free_page((unsigned long) map[n].ptr);
1328}
1329
1330int mon_bin_add(struct mon_bus *mbus, const struct usb_bus *ubus)
1331{
1332	struct device *dev;
1333	unsigned minor = ubus? ubus->busnum: 0;
1334
1335	if (minor >= MON_BIN_MAX_MINOR)
1336		return 0;
1337
1338	dev = device_create(mon_bin_class, ubus ? ubus->controller : NULL,
1339			    MKDEV(MAJOR(mon_bin_dev0), minor), NULL,
1340			    "usbmon%d", minor);
1341	if (IS_ERR(dev))
1342		return 0;
1343
1344	mbus->classdev = dev;
1345	return 1;
1346}
1347
1348void mon_bin_del(struct mon_bus *mbus)
1349{
1350	device_destroy(mon_bin_class, mbus->classdev->devt);
1351}
1352
1353int __init mon_bin_init(void)
1354{
1355	int rc;
1356
1357	mon_bin_class = class_create(THIS_MODULE, "usbmon");
1358	if (IS_ERR(mon_bin_class)) {
1359		rc = PTR_ERR(mon_bin_class);
1360		goto err_class;
1361	}
1362
1363	rc = alloc_chrdev_region(&mon_bin_dev0, 0, MON_BIN_MAX_MINOR, "usbmon");
1364	if (rc < 0)
1365		goto err_dev;
1366
1367	cdev_init(&mon_bin_cdev, &mon_fops_binary);
1368	mon_bin_cdev.owner = THIS_MODULE;
1369
1370	rc = cdev_add(&mon_bin_cdev, mon_bin_dev0, MON_BIN_MAX_MINOR);
1371	if (rc < 0)
1372		goto err_add;
1373
1374	return 0;
1375
1376err_add:
1377	unregister_chrdev_region(mon_bin_dev0, MON_BIN_MAX_MINOR);
1378err_dev:
1379	class_destroy(mon_bin_class);
1380err_class:
1381	return rc;
1382}
1383
1384void mon_bin_exit(void)
1385{
1386	cdev_del(&mon_bin_cdev);
1387	unregister_chrdev_region(mon_bin_dev0, MON_BIN_MAX_MINOR);
1388	class_destroy(mon_bin_class);
1389}
1390