mon_bin.c

linux 内核源代码

/*
 * The USB Monitor, inspired by Dave Harding's USBMon.
 *
 * This is a binary format reader.
 *
 * Copyright (C) 2006 Paolo Abeni ([e-mail])
 * Copyright (C) 2006,2007 Pete Zaitcev ([e-mail])
 */

#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/fs.h>
#include <linux/cdev.h>
#include <linux/usb.h>
#include <linux/poll.h>
#include <linux/compat.h>
#include <linux/mm.h>

#include <asm/uaccess.h>

#include "usb_mon.h"

/*
 * Defined by USB 2.0 clause 9.3, table 9.2.
 */
#define SETUP_LEN  8

/* ioctl macros */
#define MON_IOC_MAGIC 0x92

#define MON_IOCQ_URB_LEN _IO(MON_IOC_MAGIC, 1)
/* #2 used to be MON_IOCX_URB, removed before it got into Linus tree */
#define MON_IOCG_STATS _IOR(MON_IOC_MAGIC, 3, struct mon_bin_stats)
#define MON_IOCT_RING_SIZE _IO(MON_IOC_MAGIC, 4)
#define MON_IOCQ_RING_SIZE _IO(MON_IOC_MAGIC, 5)
#define MON_IOCX_GET   _IOW(MON_IOC_MAGIC, 6, struct mon_bin_get)
#define MON_IOCX_MFETCH _IOWR(MON_IOC_MAGIC, 7, struct mon_bin_mfetch)
#define MON_IOCH_MFLUSH _IO(MON_IOC_MAGIC, 8)
#ifdef CONFIG_COMPAT
#define MON_IOCX_GET32 _IOW(MON_IOC_MAGIC, 6, struct mon_bin_get32)
#define MON_IOCX_MFETCH32 _IOWR(MON_IOC_MAGIC, 7, struct mon_bin_mfetch32)
#endif

/*
 * Some architectures have enormous basic pages (16KB for ia64, 64KB for ppc).
 * But it's all right. Just use a simple way to make sure the chunk is never
 * smaller than a page.
 *
 * N.B. An application does not know our chunk size.
 *
 * Woops, get_zeroed_page() returns a single page. I guess we're stuck with
 * page-sized chunks for the time being.
 */
#define CHUNK_SIZE   PAGE_SIZE
#define CHUNK_ALIGN(x)   (((x)+CHUNK_SIZE-1) & ~(CHUNK_SIZE-1))

/*
 * The magic limit was calculated so that it allows the monitoring
 * application to pick data once in two ticks. This way, another application,
 * which presumably drives the bus, gets to hog CPU, yet we collect our data.
 * If HZ is 100, a 480 mbit/s bus drives 614 KB every jiffy. USB has an
 * enormous overhead built into the bus protocol, so we need about 1000 KB.
 *
 * This is still too much for most cases, where we just snoop a few
 * descriptor fetches for enumeration. So, the default is a "reasonable"
 * amount for systems with HZ=250 and incomplete bus saturation.
 *
 * XXX What about multi-megabyte URBs which take minutes to transfer?
 */
#define BUFF_MAX  CHUNK_ALIGN(1200*1024)
#define BUFF_DFL   CHUNK_ALIGN(300*1024)
#define BUFF_MIN     CHUNK_ALIGN(8*1024)

/*
 * The per-event API header (2 per URB).
 *
 * This structure is seen in userland as defined by the documentation.
 */
struct mon_bin_hdr {
	u64 id;			/* URB ID - from submission to callback */
	unsigned char type;	/* Same as in text API; extensible. */
	unsigned char xfer_type;	/* ISO, Intr, Control, Bulk */
	unsigned char epnum;	/* Endpoint number and transfer direction */
	unsigned char devnum;	/* Device address */
	unsigned short busnum;	/* Bus number */
	char flag_setup;
	char flag_data;
	s64 ts_sec;		/* gettimeofday */
	s32 ts_usec;		/* gettimeofday */
	int status;
	unsigned int len_urb;	/* Length of data (submitted or actual) */
	unsigned int len_cap;	/* Delivered length */
	unsigned char setup[SETUP_LEN];	/* Only for Control S-type */
};

/* per file statistic */
struct mon_bin_stats {
	u32 queued;
	u32 dropped;
};

struct mon_bin_get {
	struct mon_bin_hdr __user *hdr;	/* Only 48 bytes, not 64. */
	void __user *data;
	size_t alloc;		/* Length of data (can be zero) */
};

struct mon_bin_mfetch {
	u32 __user *offvec;	/* Vector of events fetched */
	u32 nfetch;		/* Number of events to fetch (out: fetched) */
	u32 nflush;		/* Number of events to flush */
};

#ifdef CONFIG_COMPAT
struct mon_bin_get32 {
	u32 hdr32;
	u32 data32;
	u32 alloc32;
};

struct mon_bin_mfetch32 {
        u32 offvec32;
        u32 nfetch32;
        u32 nflush32;
};
#endif

/* Having these two values same prevents wrapping of the mon_bin_hdr */
#define PKT_ALIGN   64
#define PKT_SIZE    64

/* max number of USB bus supported */
#define MON_BIN_MAX_MINOR 128

/*
 * The buffer: map of used pages.
 */
struct mon_pgmap {
	struct page *pg;
	unsigned char *ptr;	/* XXX just use page_to_virt everywhere? */
};

/*
 * This gets associated with an open file struct.
 */
struct mon_reader_bin {
	/* The buffer: one per open. */
	spinlock_t b_lock;		/* Protect b_cnt, b_in */
	unsigned int b_size;		/* Current size of the buffer - bytes */
	unsigned int b_cnt;		/* Bytes used */
	unsigned int b_in, b_out;	/* Offsets into buffer - bytes */
	unsigned int b_read;		/* Amount of read data in curr. pkt. */
	struct mon_pgmap *b_vec;	/* The map array */
	wait_queue_head_t b_wait;	/* Wait for data here */

	struct mutex fetch_lock;	/* Protect b_read, b_out */
	int mmap_active;

	/* A list of these is needed for "bus 0". Some time later. */
	struct mon_reader r;

	/* Stats */
	unsigned int cnt_lost;
};

static inline struct mon_bin_hdr *MON_OFF2HDR(const struct mon_reader_bin *rp,
    unsigned int offset)
{
	return (struct mon_bin_hdr *)
	    (rp->b_vec[offset / CHUNK_SIZE].ptr + offset % CHUNK_SIZE);
}

#define MON_RING_EMPTY(rp)	((rp)->b_cnt == 0)

static unsigned char xfer_to_pipe[4] = {
	PIPE_CONTROL, PIPE_ISOCHRONOUS, PIPE_BULK, PIPE_INTERRUPT
};

static struct class *mon_bin_class;
static dev_t mon_bin_dev0;
static struct cdev mon_bin_cdev;

static void mon_buff_area_fill(const struct mon_reader_bin *rp,
    unsigned int offset, unsigned int size);
static int mon_bin_wait_event(struct file *file, struct mon_reader_bin *rp);
static int mon_alloc_buff(struct mon_pgmap *map, int npages);
static void mon_free_buff(struct mon_pgmap *map, int npages);

/*
 * This is a "chunked memcpy". It does not manipulate any counters.
 * But it returns the new offset for repeated application.
 */
unsigned int mon_copy_to_buff(const struct mon_reader_bin *this,
    unsigned int off, const unsigned char *from, unsigned int length)
{
	unsigned int step_len;
	unsigned char *buf;
	unsigned int in_page;

	while (length) {
		/*
		 * Determine step_len.
		 */
		step_len = length;
		in_page = CHUNK_SIZE - (off & (CHUNK_SIZE-1));
		if (in_page < step_len)
			step_len = in_page;

		/*
		 * Copy data and advance pointers.
		 */
		buf = this->b_vec[off / CHUNK_SIZE].ptr + off % CHUNK_SIZE;
		memcpy(buf, from, step_len);
		if ((off += step_len) >= this->b_size) off = 0;
		from += step_len;
		length -= step_len;
	}
	return off;
}

/*
 * This is a little worse than the above because it's "chunked copy_to_user".
 * The return value is an error code, not an offset.
 */
static int copy_from_buf(const struct mon_reader_bin *this, unsigned int off,
    char __user *to, int length)
{
	unsigned int step_len;
	unsigned char *buf;
	unsigned int in_page;

	while (length) {
		/*
		 * Determine step_len.
		 */
		step_len = length;
		in_page = CHUNK_SIZE - (off & (CHUNK_SIZE-1));
		if (in_page < step_len)
			step_len = in_page;

		/*
		 * Copy data and advance pointers.
		 */
		buf = this->b_vec[off / CHUNK_SIZE].ptr + off % CHUNK_SIZE;
		if (copy_to_user(to, buf, step_len))
			return -EINVAL;
		if ((off += step_len) >= this->b_size) off = 0;
		to += step_len;
		length -= step_len;
	}
	return 0;
}

/*
 * Allocate an (aligned) area in the buffer.
 * This is called under b_lock.
 * Returns ~0 on failure.
 */
static unsigned int mon_buff_area_alloc(struct mon_reader_bin *rp,
    unsigned int size)
{
	unsigned int offset;

	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
	if (rp->b_cnt + size > rp->b_size)
		return ~0;
	offset = rp->b_in;
	rp->b_cnt += size;
	if ((rp->b_in += size) >= rp->b_size)
		rp->b_in -= rp->b_size;
	return offset;
}

/*
 * This is the same thing as mon_buff_area_alloc, only it does not allow
 * buffers to wrap. This is needed by applications which pass references
 * into mmap-ed buffers up their stacks (libpcap can do that).
 *
 * Currently, we always have the header stuck with the data, although
 * it is not strictly speaking necessary.
 *
 * When a buffer would wrap, we place a filler packet to mark the space.
 */
static unsigned int mon_buff_area_alloc_contiguous(struct mon_reader_bin *rp,
    unsigned int size)
{
	unsigned int offset;
	unsigned int fill_size;

	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
	if (rp->b_cnt + size > rp->b_size)
		return ~0;
	if (rp->b_in + size > rp->b_size) {
		/*
		 * This would wrap. Find if we still have space after
		 * skipping to the end of the buffer. If we do, place
		 * a filler packet and allocate a new packet.
		 */
		fill_size = rp->b_size - rp->b_in;
		if (rp->b_cnt + size + fill_size > rp->b_size)
			return ~0;
		mon_buff_area_fill(rp, rp->b_in, fill_size);

		offset = 0;
		rp->b_in = size;
		rp->b_cnt += size + fill_size;
	} else if (rp->b_in + size == rp->b_size) {
		offset = rp->b_in;
		rp->b_in = 0;
		rp->b_cnt += size;
	} else {
		offset = rp->b_in;
		rp->b_in += size;
		rp->b_cnt += size;
	}
	return offset;
}

/*
 * Return a few (kilo-)bytes to the head of the buffer.
 * This is used if a DMA fetch fails.
 */
static void mon_buff_area_shrink(struct mon_reader_bin *rp, unsigned int size)
{

	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
	rp->b_cnt -= size;
	if (rp->b_in < size)
		rp->b_in += rp->b_size;
	rp->b_in -= size;
}

/*
 * This has to be called under both b_lock and fetch_lock, because
 * it accesses both b_cnt and b_out.
 */
static void mon_buff_area_free(struct mon_reader_bin *rp, unsigned int size)
{

	size = (size + PKT_ALIGN-1) & ~(PKT_ALIGN-1);
	rp->b_cnt -= size;
	if ((rp->b_out += size) >= rp->b_size)
		rp->b_out -= rp->b_size;
}

static void mon_buff_area_fill(const struct mon_reader_bin *rp,
    unsigned int offset, unsigned int size)
{
	struct mon_bin_hdr *ep;

	ep = MON_OFF2HDR(rp, offset);
	memset(ep, 0, PKT_SIZE);
	ep->type = '@';
	ep->len_cap = size - PKT_SIZE;
}

static inline char mon_bin_get_setup(unsigned char *setupb,
    const struct urb *urb, char ev_type)
{

	if (!usb_endpoint_xfer_control(&urb->ep->desc) || ev_type != 'S')
		return '-';

	if (urb->setup_packet == NULL)
		return 'Z';

	memcpy(setupb, urb->setup_packet, SETUP_LEN);
	return 0;
}

static char mon_bin_get_data(const struct mon_reader_bin *rp,
    unsigned int offset, struct urb *urb, unsigned int length)
{

	if (urb->dev->bus->uses_dma &&
	    (urb->transfer_flags & URB_NO_TRANSFER_DMA_MAP)) {
		mon_dmapeek_vec(rp, offset, urb->transfer_dma, length);
		return 0;
	}

	if (urb->transfer_buffer == NULL)
		return 'Z';

	mon_copy_to_buff(rp, offset, urb->transfer_buffer, length);
	return 0;
}

static void mon_bin_event(struct mon_reader_bin *rp, struct urb *urb,
    char ev_type, int status)
{
	const struct usb_endpoint_descriptor *epd = &urb->ep->desc;
	unsigned long flags;
	struct timeval ts;
	unsigned int urb_length;
	unsigned int offset;
	unsigned int length;
	unsigned char dir;
	struct mon_bin_hdr *ep;
	char data_tag = 0;

	do_gettimeofday(&ts);

	spin_lock_irqsave(&rp->b_lock, flags);

	/*
	 * Find the maximum allowable length, then allocate space.
	 */
	urb_length = (ev_type == 'S') ?
	    urb->transfer_buffer_length : urb->actual_length;
	length = urb_length;

	if (length >= rp->b_size/5)
		length = rp->b_size/5;

	if (usb_urb_dir_in(urb)) {
		if (ev_type == 'S') {
			length = 0;
			data_tag = '<';
		}
		/* Cannot rely on endpoint number in case of control ep.0 */
		dir = USB_DIR_IN;
	} else {
		if (ev_type == 'C') {
			length = 0;
			data_tag = '>';
		}
		dir = 0;
	}

	if (rp->mmap_active)
		offset = mon_buff_area_alloc_contiguous(rp, length + PKT_SIZE);
	else
		offset = mon_buff_area_alloc(rp, length + PKT_SIZE);
	if (offset == ~0) {
		rp->cnt_lost++;
		spin_unlock_irqrestore(&rp->b_lock, flags);
		return;
	}

	ep = MON_OFF2HDR(rp, offset);
	if ((offset += PKT_SIZE) >= rp->b_size) offset = 0;

	/*
	 * Fill the allocated area.
	 */
	memset(ep, 0, PKT_SIZE);
	ep->type = ev_type;
	ep->xfer_type = xfer_to_pipe[usb_endpoint_type(epd)];
	ep->epnum = dir | usb_endpoint_num(epd);
	ep->devnum = urb->dev->devnum;
	ep->busnum = urb->dev->bus->busnum;
	ep->id = (unsigned long) urb;
	ep->ts_sec = ts.tv_sec;
	ep->ts_usec = ts.tv_usec;
	ep->status = status;
	ep->len_urb = urb_length;
	ep->len_cap = length;

	ep->flag_setup = mon_bin_get_setup(ep->setup, urb, ev_type);
	if (length != 0) {
		ep->flag_data = mon_bin_get_data(rp, offset, urb, length);
		if (ep->flag_data != 0) {	/* Yes, it's 0x00, not '0' */
			ep->len_cap = 0;
			mon_buff_area_shrink(rp, length);
		}
	} else {
		ep->flag_data = data_tag;
	}

	spin_unlock_irqrestore(&rp->b_lock, flags);

	wake_up(&rp->b_wait);
}

static void mon_bin_submit(void *data, struct urb *urb)
{
	struct mon_reader_bin *rp = data;
	mon_bin_event(rp, urb, 'S', -EINPROGRESS);
}

static void mon_bin_complete(void *data, struct urb *urb, int status)
{
	struct mon_reader_bin *rp = data;
	mon_bin_event(rp, urb, 'C', status);
}

static void mon_bin_error(void *data, struct urb *urb, int error)
{
	struct mon_reader_bin *rp = data;
	unsigned long flags;
	unsigned int offset;
	struct mon_bin_hdr *ep;

	spin_lock_irqsave(&rp->b_lock, flags);

	offset = mon_buff_area_alloc(rp, PKT_SIZE);
	if (offset == ~0) {
		/* Not incrementing cnt_lost. Just because. */
		spin_unlock_irqrestore(&rp->b_lock, flags);
		return;
	}

	ep = MON_OFF2HDR(rp, offset);

	memset(ep, 0, PKT_SIZE);
	ep->type = 'E';
	ep->xfer_type = xfer_to_pipe[usb_endpoint_type(&urb->ep->desc)];
	ep->epnum = usb_urb_dir_in(urb) ? USB_DIR_IN : 0;
	ep->epnum |= usb_endpoint_num(&urb->ep->desc);
	ep->devnum = urb->dev->devnum;
	ep->busnum = urb->dev->bus->busnum;
	ep->id = (unsigned long) urb;
	ep->status = error;

	ep->flag_setup = '-';
	ep->flag_data = 'E';

	spin_unlock_irqrestore(&rp->b_lock, flags);

	wake_up(&rp->b_wait);
}

static int mon_bin_open(struct inode *inode, struct file *file)
{
	struct mon_bus *mbus;
	struct mon_reader_bin *rp;
	size_t size;
	int rc;

	mutex_lock(&mon_lock);
	if ((mbus = mon_bus_lookup(iminor(inode))) == NULL) {
		mutex_unlock(&mon_lock);
		return -ENODEV;
	}
	if (mbus != &mon_bus0 && mbus->u_bus == NULL) {
		printk(KERN_ERR TAG ": consistency error on open\n");
		mutex_unlock(&mon_lock);
		return -ENODEV;
	}

	rp = kzalloc(sizeof(struct mon_reader_bin), GFP_KERNEL);
	if (rp == NULL) {
		rc = -ENOMEM;
		goto err_alloc;
	}
	spin_lock_init(&rp->b_lock);
	init_waitqueue_head(&rp->b_wait);
	mutex_init(&rp->fetch_lock);

	rp->b_size = BUFF_DFL;

	size = sizeof(struct mon_pgmap) * (rp->b_size/CHUNK_SIZE);
	if ((rp->b_vec = kzalloc(size, GFP_KERNEL)) == NULL) {
		rc = -ENOMEM;
		goto err_allocvec;
	}

	if ((rc = mon_alloc_buff(rp->b_vec, rp->b_size/CHUNK_SIZE)) < 0)
		goto err_allocbuff;

	rp->r.m_bus = mbus;
	rp->r.r_data = rp;
	rp->r.rnf_submit = mon_bin_submit;
	rp->r.rnf_error = mon_bin_error;
	rp->r.rnf_complete = mon_bin_complete;

	mon_reader_add(mbus, &rp->r);

	file->private_data = rp;
	mutex_unlock(&mon_lock);
	return 0;

err_allocbuff:
	kfree(rp->b_vec);
err_allocvec:
	kfree(rp);
err_alloc:
	mutex_unlock(&mon_lock);
	return rc;
}

/*
 * Extract an event from buffer and copy it to user space.
 * Wait if there is no event ready.
 * Returns zero or error.
 */
static int mon_bin_get_event(struct file *file, struct mon_reader_bin *rp,
    struct mon_bin_hdr __user *hdr, void __user *data, unsigned int nbytes)
{
	unsigned long flags;
	struct mon_bin_hdr *ep;
	size_t step_len;
	unsigned int offset;
	int rc;

	mutex_lock(&rp->fetch_lock);

	if ((rc = mon_bin_wait_event(file, rp)) < 0) {
		mutex_unlock(&rp->fetch_lock);
		return rc;
	}

	ep = MON_OFF2HDR(rp, rp->b_out);

	if (copy_to_user(hdr, ep, sizeof(struct mon_bin_hdr))) {
		mutex_unlock(&rp->fetch_lock);