2002-03-14 02:59:53

by Jean Tourrilhes

[permalink] [raw]
Subject: [PATCH 2.4.19-pre3] New wireless driver API part 1

Hi Marcelo,

Some Wireless LAN driver authors have requested me to backport
the new Wireless Extension API to 2.4.X so that they can hack their
drivers without having to deal with 2.5.X. This patch just do that, so
could you please add that to your tree ?

This patch is *very* safe, especially compared to other stuff
that made it's way in your kernel :
o was integrated in 2.5.3-pre2, straight backport (same patch)
o tested with various drivers in 2.5.X
o reviewed by Jeff Garzik ;-)
o doesn't change anything for driver still using the old API
(so current kernel/pcmcia drivers won't get broken).
o no part of the kernel using the new API, only outside drivers

For stability reason, I won't change any of the current driver
in the kernel, and those will continue to use the old API (unless you
or Jeff request me to update them).
Also, I may send you in the future the second part of the new
API (that got just integrated in 2.5.7-pre1), but I'm not in a
hurry... So, for now, first part only ;-)
Below, you will find the original blurb I sent to Linus...

Thanks in advance...

Jean

---------------------------------------------------------
Hi Linus,

I feel the first part my new driver API for Wireless
Extensions is ready to go in the kernel.

I don't want to bore you too much with details, but for
Wireless Extensions it replace the call to the device driver ioctl
function with individual call to handlers functions.
The advantage of the new API are :
o Handling of Extensions in driver broken in small contained functions
o Tighter checks of ioctl before calling the driver
o Flexible commit strategy (at least, the start of it)
o Backward compatibility (can be mixed with old API)
o Remove all ioctl assumptions in the driver
o Driver doesn't have to worry about memory and user-space issues
(copy_to/from_user handled by the API, not the driver)
The last point is important for the following reasons :
o You are now able to call the new driver API from any API you
want (including from within other parts of the kernel).
o Common mistakes are avoided (buffer overflow, user space copy
with irq disabled and so on).
More details available at :
http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Tools.html#newapi

Compare to my earlier patch (sent to the list in December), I
made the following changes :
o cleanup comments
o use "return X;" instead of "return(X);" (pointed out to me
by Jeff - thanks).

This has been tested on 2.5.3-pre2. Would you mind adding that
to your kernel ?
Regards,

Jean
---------------------------------------------------------

diff -u -p -r --new-file linux/include/linux-w12/netdevice.h linux/include/linux/netdevice.h
--- linux/include/linux-w12/netdevice.h Thu Nov 22 11:47:09 2001
+++ linux/include/linux/netdevice.h Thu Jan 17 12:00:39 2002
@@ -278,6 +278,10 @@ struct net_device
struct net_device_stats* (*get_stats)(struct net_device *dev);
struct iw_statistics* (*get_wireless_stats)(struct net_device *dev);

+ /* List of functions to handle Wireless Extensions (instead of ioctl).
+ * See <net/iw_handler.h> for details. Jean II */
+ struct iw_handler_def * wireless_handlers;
+
/*
* This marks the end of the "visible" part of the structure. All
* fields hereafter are internal to the system, and may change at
diff -u -p -r --new-file linux/include/linux-w12/wireless.h linux/include/linux/wireless.h
--- linux/include/linux-w12/wireless.h Thu Nov 22 11:47:12 2001
+++ linux/include/linux/wireless.h Thu Jan 17 12:04:08 2002
@@ -1,9 +1,10 @@
/*
* This file define a set of standard wireless extensions
*
- * Version : 12 5.10.01
+ * Version : 13 6.12.01
*
* Authors : Jean Tourrilhes - HPL - <[email protected]>
+ * Copyright (c) 1997-2001 Jean Tourrilhes, All Rights Reserved.
*/

#ifndef _LINUX_WIRELESS_H
@@ -11,6 +12,8 @@

/************************** DOCUMENTATION **************************/
/*
+ * Initial APIs (1996 -> onward) :
+ * -----------------------------
* Basically, the wireless extensions are for now a set of standard ioctl
* call + /proc/net/wireless
*
@@ -27,16 +30,27 @@
* We have the list of command plus a structure descibing the
* data exchanged...
* Note that to add these ioctl, I was obliged to modify :
- * net/core/dev.c (two place + add include)
- * net/ipv4/af_inet.c (one place + add include)
+ * # net/core/dev.c (two place + add include)
+ * # net/ipv4/af_inet.c (one place + add include)
*
* /proc/net/wireless is a copy of /proc/net/dev.
* We have a structure for data passed from the driver to /proc/net/wireless
* Too add this, I've modified :
- * net/core/dev.c (two other places)
- * include/linux/netdevice.h (one place)
- * include/linux/proc_fs.h (one place)
+ * # net/core/dev.c (two other places)
+ * # include/linux/netdevice.h (one place)
+ * # include/linux/proc_fs.h (one place)
+ *
+ * New driver API (2001 -> onward) :
+ * -------------------------------
+ * This file is only concerned with the user space API and common definitions.
+ * The new driver API is defined and documented in :
+ * # include/net/iw_handler.h
*
+ * Note as well that /proc/net/wireless implementation has now moved in :
+ * # include/linux/wireless.c
+ *
+ * Other comments :
+ * --------------
* Do not add here things that are redundant with other mechanisms
* (drivers init, ifconfig, /proc/net/dev, ...) and with are not
* wireless specific.
@@ -54,16 +68,14 @@
#include <linux/socket.h> /* for "struct sockaddr" et al */
#include <linux/if.h> /* for IFNAMSIZ and co... */

-/**************************** CONSTANTS ****************************/
-
-/* --------------------------- VERSION --------------------------- */
+/***************************** VERSION *****************************/
/*
* This constant is used to know the availability of the wireless
* extensions and to know which version of wireless extensions it is
* (there is some stuff that will be added in the future...)
* I just plan to increment with each new version.
*/
-#define WIRELESS_EXT 12
+#define WIRELESS_EXT 13

/*
* Changes :
@@ -123,12 +135,20 @@
* - Add DEV PRIVATE IOCTL to avoid collisions in SIOCDEVPRIVATE space
* - Add new statistics (frag, retry, beacon)
* - Add average quality (for user space calibration)
+ *
+ * V12 to V13
+ * ----------
+ * - Document creation of new driver API.
+ * - Extract union iwreq_data from struct iwreq (for new driver API).
+ * - Rename SIOCSIWNAME as SIOCSIWCOMMIT
*/

+/**************************** CONSTANTS ****************************/
+
/* -------------------------- IOCTL LIST -------------------------- */

/* Basic operations */
-#define SIOCSIWNAME 0x8B00 /* Unused */
+#define SIOCSIWCOMMIT 0x8B00 /* Commit pending changes to driver */
#define SIOCGIWNAME 0x8B01 /* get name == wireless protocol */
#define SIOCSIWNWID 0x8B02 /* set network id (the cell) */
#define SIOCGIWNWID 0x8B03 /* get network id */
@@ -414,13 +434,49 @@ struct iw_statistics

/* ------------------------ IOCTL REQUEST ------------------------ */
/*
+ * This structure defines the payload of an ioctl, and is used
+ * below.
+ *
+ * Note that this structure should fit on the memory footprint
+ * of iwreq (which is the same as ifreq), which mean a max size of
+ * 16 octets = 128 bits. Warning, pointers might be 64 bits wide...
+ * You should check this when increasing the structures defined
+ * above in this file...
+ */
+union iwreq_data
+{
+ /* Config - generic */
+ char name[IFNAMSIZ];
+ /* Name : used to verify the presence of wireless extensions.
+ * Name of the protocol/provider... */
+
+ struct iw_point essid; /* Extended network name */
+ struct iw_param nwid; /* network id (or domain - the cell) */
+ struct iw_freq freq; /* frequency or channel :
+ * 0-1000 = channel
+ * > 1000 = frequency in Hz */
+
+ struct iw_param sens; /* signal level threshold */
+ struct iw_param bitrate; /* default bit rate */
+ struct iw_param txpower; /* default transmit power */
+ struct iw_param rts; /* RTS threshold threshold */
+ struct iw_param frag; /* Fragmentation threshold */
+ __u32 mode; /* Operation mode */
+ struct iw_param retry; /* Retry limits & lifetime */
+
+ struct iw_point encoding; /* Encoding stuff : tokens */
+ struct iw_param power; /* PM duration/timeout */
+
+ struct sockaddr ap_addr; /* Access point address */
+
+ struct iw_point data; /* Other large parameters */
+};
+
+/*
* The structure to exchange data for ioctl.
* This structure is the same as 'struct ifreq', but (re)defined for
* convenience...
- *
- * Note that it should fit on the same memory footprint !
- * You should check this when increasing the above structures (16 octets)
- * 16 octets = 128 bits. Warning, pointers might be 64 bits wide...
+ * Do I need to remind you about structure size (32 octets) ?
*/
struct iwreq
{
@@ -429,35 +485,8 @@ struct iwreq
char ifrn_name[IFNAMSIZ]; /* if name, e.g. "eth0" */
} ifr_ifrn;

- /* Data part */
- union
- {
- /* Config - generic */
- char name[IFNAMSIZ];
- /* Name : used to verify the presence of wireless extensions.
- * Name of the protocol/provider... */
-
- struct iw_point essid; /* Extended network name */
- struct iw_param nwid; /* network id (or domain - the cell) */
- struct iw_freq freq; /* frequency or channel :
- * 0-1000 = channel
- * > 1000 = frequency in Hz */
-
- struct iw_param sens; /* signal level threshold */
- struct iw_param bitrate; /* default bit rate */
- struct iw_param txpower; /* default transmit power */
- struct iw_param rts; /* RTS threshold threshold */
- struct iw_param frag; /* Fragmentation threshold */
- __u32 mode; /* Operation mode */
- struct iw_param retry; /* Retry limits & lifetime */
-
- struct iw_point encoding; /* Encoding stuff : tokens */
- struct iw_param power; /* PM duration/timeout */
-
- struct sockaddr ap_addr; /* Access point address */
-
- struct iw_point data; /* Other large parameters */
- } u;
+ /* Data part (defined just above) */
+ union iwreq_data u;
};

/* -------------------------- IOCTL DATA -------------------------- */
diff -u -p -r --new-file linux/include/net-w12/iw_handler.h linux/include/net/iw_handler.h
--- linux/include/net-w12/iw_handler.h Wed Dec 31 16:00:00 1969
+++ linux/include/net/iw_handler.h Thu Jan 17 12:16:46 2002
@@ -0,0 +1,374 @@
+/*
+ * This file define the new driver API for Wireless Extensions
+ *
+ * Version : 2 6.12.01
+ *
+ * Authors : Jean Tourrilhes - HPL - <[email protected]>
+ * Copyright (c) 2001 Jean Tourrilhes, All Rights Reserved.
+ */
+
+#ifndef _IW_HANDLER_H
+#define _IW_HANDLER_H
+
+/************************** DOCUMENTATION **************************/
+/*
+ * Initial driver API (1996 -> onward) :
+ * -----------------------------------
+ * The initial API just sends the IOCTL request received from user space
+ * to the driver (via the driver ioctl handler). The driver has to
+ * handle all the rest...
+ *
+ * The initial API also defines a specific handler in struct net_device
+ * to handle wireless statistics.
+ *
+ * The initial APIs served us well and has proven a reasonably good design.
+ * However, there is a few shortcommings :
+ * o No events, everything is a request to the driver.
+ * o Large ioctl function in driver with gigantic switch statement
+ * (i.e. spaghetti code).
+ * o Driver has to mess up with copy_to/from_user, and in many cases
+ * does it unproperly. Common mistakes are :
+ * * buffer overflows (no checks or off by one checks)
+ * * call copy_to/from_user with irq disabled
+ * o The user space interface is tied to ioctl because of the use
+ * copy_to/from_user.
+ *
+ * New driver API (2001 -> onward) :
+ * -------------------------------
+ * The new driver API is just a bunch of standard functions (handlers),
+ * each handling a specific Wireless Extension. The driver just export
+ * the list of handler it supports, and those will be called apropriately.
+ *
+ * I tried to keep the main advantage of the previous API (simplicity,
+ * efficiency and light weight), and also I provide a good dose of backward
+ * compatibility (most structures are the same, driver can use both API
+ * simultaneously, ...).
+ * Hopefully, I've also addressed the shortcomming of the initial API.
+ *
+ * The advantage of the new API are :
+ * o Handling of Extensions in driver broken in small contained functions
+ * o Tighter checks of ioctl before calling the driver
+ * o Flexible commit strategy (at least, the start of it)
+ * o Backward compatibility (can be mixed with old API)
+ * o Driver doesn't have to worry about memory and user-space issues
+ * The last point is important for the following reasons :
+ * o You are now able to call the new driver API from any API you
+ * want (including from within other parts of the kernel).
+ * o Common mistakes are avoided (buffer overflow, user space copy
+ * with irq disabled and so on).
+ *
+ * The Drawback of the new API are :
+ * o bloat (especially kernel)
+ * o need to migrate existing drivers to new API
+ * My initial testing shows that the new API adds around 3kB to the kernel
+ * and save between 0 and 5kB from a typical driver.
+ * Also, as all structures and data types are unchanged, the migration is
+ * quite straightforward (but tedious).
+ *
+ * ---
+ *
+ * The new driver API is defined below in this file. User space should
+ * not be aware of what's happening down there...
+ *
+ * A new kernel wrapper is in charge of validating the IOCTLs and calling
+ * the appropriate driver handler. This is implemented in :
+ * # net/core/wireless.c
+ *
+ * The driver export the list of handlers in :
+ * # include/linux/netdevice.h (one place)
+ *
+ * The new driver API is available for WIRELESS_EXT >= 13.
+ * Good luck with migration to the new API ;-)
+ */
+
+/* ---------------------- THE IMPLEMENTATION ---------------------- */
+/*
+ * Some of the choice I've made are pretty controversials. Defining an
+ * API is very much weighting compromises. This goes into some of the
+ * details and the thinking behind the implementation.
+ *
+ * Implementation goals :
+ * --------------------
+ * The implementation goals were as follow :
+ * o Obvious : you should not need a PhD to understand what's happening,
+ * the benefit is easier maintainance.
+ * o Flexible : it should accomodate a wide variety of driver
+ * implementations and be as flexible as the old API.
+ * o Lean : it should be efficient memory wise to minimise the impact
+ * on kernel footprint.
+ * o Transparent to user space : the large number of user space
+ * applications that use Wireless Extensions should not need
+ * any modifications.
+ *
+ * Array of functions versus Struct of functions
+ * ---------------------------------------------
+ * 1) Having an array of functions allow the kernel code to access the
+ * handler in a single lookup, which is much more efficient (think hash
+ * table here).
+ * 2) The only drawback is that driver writer may put their handler in
+ * the wrong slot. This is trivial to test (I set the frequency, the
+ * bitrate changes). Once the handler is in the proper slot, it will be
+ * there forever, because the array is only extended at the end.
+ * 3) Backward/forward compatibility : adding new handler just require
+ * extending the array, so you can put newer driver in older kernel
+ * without having to patch the kernel code (and vice versa).
+ *
+ * All handler are of the same generic type
+ * ----------------------------------------
+ * That's a feature !!!
+ * 1) Having a generic handler allow to have generic code, which is more
+ * efficient. If each of the handler was individually typed I would need
+ * to add a big switch in the kernel (== more bloat). This solution is
+ * more scalable, adding new Wireless Extensions doesn't add new code.
+ * 2) You can use the same handler in different slots of the array. For
+ * hardware, it may be more efficient or logical to handle multiple
+ * Wireless Extensions with a single function, and the API allow you to
+ * do that. (An example would be a single record on the card to control
+ * both bitrate and frequency, the handler would read the old record,
+ * modify it according to info->cmd and rewrite it).
+ *
+ * Functions prototype uses union iwreq_data
+ * -----------------------------------------
+ * Some would have prefered functions defined this way :
+ * static int mydriver_ioctl_setrate(struct net_device *dev,
+ * long rate, int auto)
+ * 1) The kernel code doesn't "validate" the content of iwreq_data, and
+ * can't do it (different hardware may have different notion of what a
+ * valid frequency is), so we don't pretend that we do it.
+ * 2) The above form is not extendable. If I want to add a flag (for
+ * example to distinguish setting max rate and basic rate), I would
+ * break the prototype. Using iwreq_data is more flexible.
+ * 3) Also, the above form is not generic (see above).
+ * 4) I don't expect driver developper using the wrong field of the
+ * union (Doh !), so static typechecking doesn't add much value.
+ * 5) Lastly, you can skip the union by doing :
+ * static int mydriver_ioctl_setrate(struct net_device *dev,
+ * struct iw_request_info *info,
+ * struct iw_param *rrq,
+ * char *extra)
+ * And then adding the handler in the array like this :
+ * (iw_handler) mydriver_ioctl_setrate, // SIOCSIWRATE
+ *
+ * Using functions and not a registry
+ * ----------------------------------
+ * Another implementation option would have been for every instance to
+ * define a registry (a struct containing all the Wireless Extensions)
+ * and only have a function to commit the registry to the hardware.
+ * 1) This approach can be emulated by the current code, but not
+ * vice versa.
+ * 2) Some drivers don't keep any configuration in the driver, for them
+ * adding such a registry would be a significant bloat.
+ * 3) The code to translate from Wireless Extension to native format is
+ * needed anyway, so it would not reduce significantely the amount of code.
+ * 4) The current approach only selectively translate Wireless Extensions
+ * to native format and only selectively set, whereas the registry approach
+ * would require to translate all WE and set all parameters for any single
+ * change.
+ * 5) For many Wireless Extensions, the GET operation return the current
+ * dynamic value, not the value that was set.
+ *
+ * This header is <net/iw_handler.h>
+ * ---------------------------------
+ * 1) This header is kernel space only and should not be exported to
+ * user space. Headers in "include/linux/" are exported, headers in
+ * "include/net/" are not.
+ *
+ * Mixed 32/64 bit issues
+ * ----------------------
+ * The Wireless Extensions are designed to be 64 bit clean, by using only
+ * datatypes with explicit storage size.
+ * There are some issues related to kernel and user space using different
+ * memory model, and in particular 64bit kernel with 32bit user space.
+ * The problem is related to struct iw_point, that contains a pointer
+ * that *may* need to be translated.
+ * This is quite messy. The new API doesn't solve this problem (it can't),
+ * but is a step in the right direction :
+ * 1) Meta data about each ioctl is easily available, so we know what type
+ * of translation is needed.
+ * 2) The move of data between kernel and user space is only done in a single
+ * place in the kernel, so adding specific hooks in there is possible.
+ * 3) In the long term, it allows to move away from using ioctl as the
+ * user space API.
+ *
+ * So many comments and so few code
+ * --------------------------------
+ * That's a feature. Comments won't bloat the resulting kernel binary.
+ */
+
+/***************************** INCLUDES *****************************/
+
+#include <linux/wireless.h> /* IOCTL user space API */
+
+/***************************** VERSION *****************************/
+/*
+ * This constant is used to know which version of the driver API is
+ * available. Hopefully, this will be pretty stable and no changes
+ * will be needed...
+ * I just plan to increment with each new version.
+ */
+#define IW_HANDLER_VERSION 2
+
+/**************************** CONSTANTS ****************************/
+
+/* Special error message for the driver to indicate that we
+ * should do a commit after return from the iw_handler */
+#define EIWCOMMIT EINPROGRESS
+
+/* Flags available in struct iw_request_info */
+#define IW_REQUEST_FLAG_NONE 0x0000 /* No flag so far */
+
+/* Type of headers we know about (basically union iwreq_data) */
+#define IW_HEADER_TYPE_NULL 0 /* Not available */
+#define IW_HEADER_TYPE_CHAR 2 /* char [IFNAMSIZ] */
+#define IW_HEADER_TYPE_UINT 4 /* __u32 */
+#define IW_HEADER_TYPE_FREQ 5 /* struct iw_freq */
+#define IW_HEADER_TYPE_POINT 6 /* struct iw_point */
+#define IW_HEADER_TYPE_PARAM 7 /* struct iw_param */
+#define IW_HEADER_TYPE_ADDR 8 /* struct sockaddr */
+
+/* Handling flags */
+/* Most are not implemented. I just use them as a reminder of some
+ * cool features we might need one day ;-) */
+#define IW_DESCR_FLAG_NONE 0x0000 /* Obvious */
+/* Wrapper level flags */
+#define IW_DESCR_FLAG_DUMP 0x0001 /* Not part of the dump command */
+#define IW_DESCR_FLAG_EVENT 0x0002 /* Generate an event on SET */
+#define IW_DESCR_FLAG_RESTRICT 0x0004 /* GET request is ROOT only */
+/* Driver level flags */
+#define IW_DESCR_FLAG_WAIT 0x0100 /* Wait for driver event */
+
+/****************************** TYPES ******************************/
+
+/* ----------------------- WIRELESS HANDLER ----------------------- */
+/*
+ * A wireless handler is just a standard function, that looks like the
+ * ioctl handler.
+ * We also define there how a handler list look like... As the Wireless
+ * Extension space is quite dense, we use a simple array, which is faster
+ * (that's the perfect hash table ;-).
+ */
+
+/*
+ * Meta data about the request passed to the iw_handler.
+ * Most handlers can safely ignore what's in there.
+ * The 'cmd' field might come handy if you want to use the same handler
+ * for multiple command...
+ * This struct is also my long term insurance. I can add new fields here
+ * without breaking the prototype of iw_handler...
+ */
+struct iw_request_info
+{
+ __u16 cmd; /* Wireless Extension command */
+ __u16 flags; /* More to come ;-) */
+};
+
+/*
+ * This is how a function handling a Wireless Extension should look
+ * like (both get and set, standard and private).
+ */
+typedef int (*iw_handler)(struct net_device *dev, struct iw_request_info *info,
+ union iwreq_data *wrqu, char *extra);
+
+/*
+ * This define all the handler that the driver export.
+ * As you need only one per driver type, please use a static const
+ * shared by all driver instances... Same for the members...
+ * This will be linked from net_device in <linux/netdevice.h>
+ */
+struct iw_handler_def
+{
+ /* Number of handlers defined (more precisely, index of the
+ * last defined handler + 1) */
+ __u16 num_standard;
+ __u16 num_private;
+ /* Number of private arg description */
+ __u16 num_private_args;
+
+ /* Array of handlers for standard ioctls
+ * We will call dev->wireless_handlers->standard[ioctl - SIOCSIWNAME]
+ */
+ iw_handler * standard;
+
+ /* Array of handlers for private ioctls
+ * Will call dev->wireless_handlers->private[ioctl - SIOCIWFIRSTPRIV]
+ */
+ iw_handler * private;
+
+ /* Arguments of private handler. This one is just a list, so you
+ * can put it in any order you want and should not leave holes...
+ * We will automatically export that to user space... */
+ struct iw_priv_args * private_args;
+
+ /* In the long term, get_wireless_stats will move from
+ * 'struct net_device' to here, to minimise bloat. */
+};
+
+/* ----------------------- WIRELESS EVENTS ----------------------- */
+/*
+ * Currently we don't support events, so let's just plan for the
+ * future...
+ */
+
+/*
+ * A Wireless Event.
+ */
+// How do we define short header ? We don't want a flag on length.
+// Probably a flag on event ? Highest bit to zero...
+struct iw_event
+{
+ __u16 length; /* Lenght of this stuff */
+ __u16 event; /* Wireless IOCTL */
+ union iwreq_data header; /* IOCTL fixed payload */
+ char extra[0]; /* Optional IOCTL data */
+};
+
+/* ---------------------- IOCTL DESCRIPTION ---------------------- */
+/*
+ * One of the main goal of the new interface is to deal entirely with
+ * user space/kernel space memory move.
+ * For that, we need to know :
+ * o if iwreq is a pointer or contain the full data
+ * o what is the size of the data to copy
+ *
+ * For private IOCTLs, we use the same rules as used by iwpriv and
+ * defined in struct iw_priv_args.
+ *
+ * For standard IOCTLs, things are quite different and we need to
+ * use the stuctures below. Actually, this struct is also more
+ * efficient, but that's another story...
+ */
+
+/*
+ * Describe how a standard IOCTL looks like.
+ */
+struct iw_ioctl_description
+{
+ __u8 header_type; /* NULL, iw_point or other */
+ __u8 token_type; /* Future */
+ __u16 token_size; /* Granularity of payload */
+ __u16 min_tokens; /* Min acceptable token number */
+ __u16 max_tokens; /* Max acceptable token number */
+ __u32 flags; /* Special handling of the request */
+};
+
+/* Need to think of short header translation table. Later. */
+
+/**************************** PROTOTYPES ****************************/
+/*
+ * Functions part of the Wireless Extensions (defined in net/core/wireless.c).
+ * Those may be called only within the kernel.
+ */
+
+/* First : function strictly used inside the kernel */
+
+/* Handle /proc/net/wireless, called in net/code/dev.c */
+extern int dev_get_wireless_info(char * buffer, char **start, off_t offset,
+ int length);
+
+/* Handle IOCTLs, called in net/code/dev.c */
+extern int wireless_process_ioctl(struct ifreq *ifr, unsigned int cmd);
+
+/* Second : functions that may be called by driver modules */
+/* None yet */
+
+#endif /* _LINUX_WIRELESS_H */
diff -u -p -r --new-file linux/net/core-w12/Makefile linux/net/core/Makefile
--- linux/net/core-w12/Makefile Tue Oct 30 15:08:12 2001
+++ linux/net/core/Makefile Thu Jan 17 11:06:07 2002
@@ -26,5 +26,8 @@ obj-$(CONFIG_NET) += dev.o dev_mcast.o d
obj-$(CONFIG_NETFILTER) += netfilter.o
obj-$(CONFIG_NET_DIVERT) += dv.o
obj-$(CONFIG_NET_PROFILE) += profile.o
+obj-$(CONFIG_NET_RADIO) += wireless.o
+# Ugly. I wish all wireless drivers were moved in drivers/net/wireless
+obj-$(CONFIG_NET_PCMCIA_RADIO) += wireless.o

include $(TOPDIR)/Rules.make
diff -u -p -r --new-file linux/net/core-w12/dev.c linux/net/core/dev.c
--- linux/net/core-w12/dev.c Wed Nov 7 14:39:36 2001
+++ linux/net/core/dev.c Thu Jan 17 11:06:07 2002
@@ -102,6 +102,7 @@
#include <linux/module.h>
#if defined(CONFIG_NET_RADIO) || defined(CONFIG_NET_PCMCIA_RADIO)
#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
+#include <net/iw_handler.h>
#endif /* CONFIG_NET_RADIO || CONFIG_NET_PCMCIA_RADIO */
#ifdef CONFIG_PLIP
extern int plip_init(void);
@@ -1796,122 +1797,6 @@ static int dev_proc_stats(char *buffer,
#endif /* CONFIG_PROC_FS */


-#ifdef WIRELESS_EXT
-#ifdef CONFIG_PROC_FS
-
-/*
- * Print one entry of /proc/net/wireless
- * This is a clone of /proc/net/dev (just above)
- */
-static int sprintf_wireless_stats(char *buffer, struct net_device *dev)
-{
- /* Get stats from the driver */
- struct iw_statistics *stats = (dev->get_wireless_stats ?
- dev->get_wireless_stats(dev) :
- (struct iw_statistics *) NULL);
- int size;
-
- if (stats != (struct iw_statistics *) NULL) {
- size = sprintf(buffer,
- "%6s: %04x %3d%c %3d%c %3d%c %6d %6d %6d %6d %6d %6d\n",
- dev->name,
- stats->status,
- stats->qual.qual,
- stats->qual.updated & 1 ? '.' : ' ',
- stats->qual.level,
- stats->qual.updated & 2 ? '.' : ' ',
- stats->qual.noise,
- stats->qual.updated & 4 ? '.' : ' ',
- stats->discard.nwid,
- stats->discard.code,
- stats->discard.fragment,
- stats->discard.retries,
- stats->discard.misc,
- stats->miss.beacon);
- stats->qual.updated = 0;
- }
- else
- size = 0;
-
- return size;
-}
-
-/*
- * Print info for /proc/net/wireless (print all entries)
- * This is a clone of /proc/net/dev (just above)
- */
-static int dev_get_wireless_info(char * buffer, char **start, off_t offset,
- int length)
-{
- int len = 0;
- off_t begin = 0;
- off_t pos = 0;
- int size;
-
- struct net_device * dev;
-
- size = sprintf(buffer,
- "Inter-| sta-| Quality | Discarded packets | Missed\n"
- " face | tus | link level noise | nwid crypt frag retry misc | beacon\n"
- );
-
- pos += size;
- len += size;
-
- read_lock(&dev_base_lock);
- for (dev = dev_base; dev != NULL; dev = dev->next) {
- size = sprintf_wireless_stats(buffer + len, dev);
- len += size;
- pos = begin + len;
-
- if (pos < offset) {
- len = 0;
- begin = pos;
- }
- if (pos > offset + length)
- break;
- }
- read_unlock(&dev_base_lock);
-
- *start = buffer + (offset - begin); /* Start of wanted data */
- len -= (offset - begin); /* Start slop */
- if (len > length)
- len = length; /* Ending slop */
- if (len < 0)
- len = 0;
-
- return len;
-}
-#endif /* CONFIG_PROC_FS */
-
-/*
- * Allow programatic access to /proc/net/wireless even if /proc
- * doesn't exist... Also more efficient...
- */
-static inline int dev_iwstats(struct net_device *dev, struct ifreq *ifr)
-{
- /* Get stats from the driver */
- struct iw_statistics *stats = (dev->get_wireless_stats ?
- dev->get_wireless_stats(dev) :
- (struct iw_statistics *) NULL);
-
- if (stats != (struct iw_statistics *) NULL) {
- struct iwreq * wrq = (struct iwreq *)ifr;
-
- /* Copy statistics to the user buffer */
- if(copy_to_user(wrq->u.data.pointer, stats,
- sizeof(struct iw_statistics)))
- return -EFAULT;
-
- /* Check if we need to clear the update flag */
- if(wrq->u.data.flags != 0)
- stats->qual.updated = 0;
- return(0);
- } else
- return -EOPNOTSUPP;
-}
-#endif /* WIRELESS_EXT */
-
/**
* netdev_set_master - set up master/slave pair
* @slave: slave device
@@ -2209,11 +2094,6 @@ static int dev_ifsioc(struct ifreq *ifr,
notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
return 0;

-#ifdef WIRELESS_EXT
- case SIOCGIWSTATS:
- return dev_iwstats(dev, ifr);
-#endif /* WIRELESS_EXT */
-
/*
* Unknown or private ioctl
*/
@@ -2239,17 +2119,6 @@ static int dev_ifsioc(struct ifreq *ifr,
return -EOPNOTSUPP;
}

-#ifdef WIRELESS_EXT
- if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) {
- if (dev->do_ioctl) {
- if (!netif_device_present(dev))
- return -ENODEV;
- return dev->do_ioctl(dev, ifr, cmd);
- }
- return -EOPNOTSUPP;
- }
-#endif /* WIRELESS_EXT */
-
}
return -EINVAL;
}
@@ -2431,7 +2300,8 @@ int dev_ioctl(unsigned int cmd, void *ar
}
dev_load(ifr.ifr_name);
rtnl_lock();
- ret = dev_ifsioc(&ifr, cmd);
+ /* Follow me in net/core/wireless.c */
+ ret = wireless_process_ioctl(&ifr, cmd);
rtnl_unlock();
if (!ret && IW_IS_GET(cmd) &&
copy_to_user(arg, &ifr, sizeof(struct ifreq)))
@@ -2856,6 +2726,7 @@ int __init net_dev_init(void)
proc_net_create("dev", 0, dev_get_info);
create_proc_read_entry("net/softnet_stat", 0, 0, dev_proc_stats, NULL);
#ifdef WIRELESS_EXT
+ /* Available in net/core/wireless.c */
proc_net_create("wireless", 0, dev_get_wireless_info);
#endif /* WIRELESS_EXT */
#endif /* CONFIG_PROC_FS */
diff -u -p -r --new-file linux/net/core-w12/wireless.c linux/net/core/wireless.c
--- linux/net/core-w12/wireless.c Wed Dec 31 16:00:00 1969
+++ linux/net/core/wireless.c Mon Jan 21 11:13:23 2002
@@ -0,0 +1,733 @@
+/*
+ * This file implement the Wireless Extensions APIs.
+ *
+ * Authors : Jean Tourrilhes - HPL - <[email protected]>
+ * Copyright (c) 1997-2001 Jean Tourrilhes, All Rights Reserved.
+ *
+ * (As all part of the Linux kernel, this file is GPL)
+ */
+
+/************************** DOCUMENTATION **************************/
+/*
+ * API definition :
+ * --------------
+ * See <linux/wireless.h> for details of the APIs and the rest.
+ *
+ * History :
+ * -------
+ *
+ * v1 - 5.12.01 - Jean II
+ * o Created this file.
+ *
+ * v2 - 13.12.01 - Jean II
+ * o Move /proc/net/wireless stuff from net/core/dev.c to here
+ * o Make Wireless Extension IOCTLs go through here
+ * o Added iw_handler handling ;-)
+ * o Added standard ioctl description
+ * o Initial dumb commit strategy based on orinoco.c
+ */
+
+/***************************** INCLUDES *****************************/
+
+#include <asm/uaccess.h> /* copy_to_user() */
+#include <linux/config.h> /* Not needed ??? */
+#include <linux/types.h> /* off_t */
+#include <linux/netdevice.h> /* struct ifreq, dev_get_by_name() */
+
+#include <linux/wireless.h> /* Pretty obvious */
+#include <net/iw_handler.h> /* New driver API */
+
+/**************************** CONSTANTS ****************************/
+
+/* This will be turned on later on... */
+#undef WE_STRICT_WRITE /* Check write buffer size */
+
+/* Debuging stuff */
+#undef WE_IOCTL_DEBUG /* Debug IOCTL API */
+
+/************************* GLOBAL VARIABLES *************************/
+/*
+ * You should not use global variables, because or re-entrancy.
+ * On our case, it's only const, so it's OK...
+ */
+static const struct iw_ioctl_description standard_ioctl[] = {
+ /* SIOCSIWCOMMIT (internal) */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCGIWNAME */
+ { IW_HEADER_TYPE_CHAR, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWNWID */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, IW_DESCR_FLAG_EVENT},
+ /* SIOCGIWNWID */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWFREQ */
+ { IW_HEADER_TYPE_FREQ, 0, 0, 0, 0, IW_DESCR_FLAG_EVENT},
+ /* SIOCGIWFREQ */
+ { IW_HEADER_TYPE_FREQ, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWMODE */
+ { IW_HEADER_TYPE_UINT, 0, 0, 0, 0, IW_DESCR_FLAG_EVENT},
+ /* SIOCGIWMODE */
+ { IW_HEADER_TYPE_UINT, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWSENS */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWSENS */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWRANGE */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCGIWRANGE */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, sizeof(struct iw_range), IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWPRIV */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCGIWPRIV (handled directly by us) */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCSIWSTATS */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCGIWSTATS (handled directly by us) */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWSPY */
+ { IW_HEADER_TYPE_POINT, 0, sizeof(struct sockaddr), 0, IW_MAX_SPY, 0},
+ /* SIOCGIWSPY */
+ { IW_HEADER_TYPE_POINT, 0, (sizeof(struct sockaddr) + sizeof(struct iw_quality)), 0, IW_MAX_SPY, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCSIWAP */
+ { IW_HEADER_TYPE_ADDR, 0, 0, 0, 0, 0},
+ /* SIOCGIWAP */
+ { IW_HEADER_TYPE_ADDR, 0, 0, 0, 0, IW_DESCR_FLAG_DUMP},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCGIWAPLIST */
+ { IW_HEADER_TYPE_POINT, 0, (sizeof(struct sockaddr) + sizeof(struct iw_quality)), 0, IW_MAX_AP, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCSIWESSID */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, IW_ESSID_MAX_SIZE, IW_DESCR_FLAG_EVENT},
+ /* SIOCGIWESSID */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, IW_ESSID_MAX_SIZE, IW_DESCR_FLAG_DUMP},
+ /* SIOCSIWNICKN */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, IW_ESSID_MAX_SIZE, 0},
+ /* SIOCGIWNICKN */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, IW_ESSID_MAX_SIZE, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* -- hole -- */
+ { IW_HEADER_TYPE_NULL, 0, 0, 0, 0, 0},
+ /* SIOCSIWRATE */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWRATE */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWRTS */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWRTS */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWFRAG */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWFRAG */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWTXPOW */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWTXPOW */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWRETRY */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWRETRY */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCSIWENCODE */
+ { IW_HEADER_TYPE_POINT, 4, 1, 0, IW_ENCODING_TOKEN_MAX, IW_DESCR_FLAG_EVENT | IW_DESCR_FLAG_RESTRICT},
+ /* SIOCGIWENCODE */
+ { IW_HEADER_TYPE_POINT, 0, 1, 0, IW_ENCODING_TOKEN_MAX, IW_DESCR_FLAG_DUMP | IW_DESCR_FLAG_RESTRICT},
+ /* SIOCSIWPOWER */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+ /* SIOCGIWPOWER */
+ { IW_HEADER_TYPE_PARAM, 0, 0, 0, 0, 0},
+};
+
+/* Size (in bytes) of the various private data types */
+char priv_type_size[] = { 0, 1, 1, 0, 4, 4, 0, 0 };
+
+/************************ COMMON SUBROUTINES ************************/
+/*
+ * Stuff that may be used in various place or doesn't fit in one
+ * of the section below.
+ */
+
+/* ---------------------------------------------------------------- */
+/*
+ * Return the driver handler associated with a specific Wireless Extension.
+ * Called from various place, so make sure it remains efficient.
+ */
+static inline iw_handler get_handler(struct net_device *dev,
+ unsigned int cmd)
+{
+ unsigned int index; /* MUST be unsigned */
+
+ /* Check if we have some wireless handlers defined */
+ if(dev->wireless_handlers == NULL)
+ return NULL;
+
+ /* Try as a standard command */
+ index = cmd - SIOCIWFIRST;
+ if(index < dev->wireless_handlers->num_standard)
+ return dev->wireless_handlers->standard[index];
+
+ /* Try as a private command */
+ index = cmd - SIOCIWFIRSTPRIV;
+ if(index < dev->wireless_handlers->num_private)
+ return dev->wireless_handlers->private[index];
+
+ /* Not found */
+ return NULL;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Get statistics out of the driver
+ */
+static inline struct iw_statistics *get_wireless_stats(struct net_device *dev)
+{
+ return (dev->get_wireless_stats ?
+ dev->get_wireless_stats(dev) :
+ (struct iw_statistics *) NULL);
+ /* In the future, get_wireless_stats may move from 'struct net_device'
+ * to 'struct iw_handler_def', to de-bloat struct net_device.
+ * Definitely worse a thought... */
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Call the commit handler in the driver
+ * (if exist and if conditions are right)
+ *
+ * Note : our current commit strategy is currently pretty dumb,
+ * but we will be able to improve on that...
+ * The goal is to try to agreagate as many changes as possible
+ * before doing the commit. Drivers that will define a commit handler
+ * are usually those that need a reset after changing parameters, so
+ * we want to minimise the number of reset.
+ * A cool idea is to use a timer : at each "set" command, we re-set the
+ * timer, when the timer eventually fires, we call the driver.
+ * Hopefully, more on that later.
+ *
+ * Also, I'm waiting to see how many people will complain about the
+ * netif_running(dev) test. I'm open on that one...
+ * Hopefully, the driver will remember to do a commit in "open()" ;-)
+ */
+static inline int call_commit_handler(struct net_device * dev)
+{
+ if((netif_running(dev)) &&
+ (dev->wireless_handlers->standard[0] != NULL)) {
+ /* Call the commit handler on the driver */
+ return dev->wireless_handlers->standard[0](dev, NULL,
+ NULL, NULL);
+ } else
+ return 0; /* Command completed successfully */
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Number of private arguments
+ */
+static inline int get_priv_size(__u16 args)
+{
+ int num = args & IW_PRIV_SIZE_MASK;
+ int type = (args & IW_PRIV_TYPE_MASK) >> 12;
+
+ return num * priv_type_size[type];
+}
+
+
+/******************** /proc/net/wireless SUPPORT ********************/
+/*
+ * The /proc/net/wireless file is a human readable user-space interface
+ * exporting various wireless specific statistics from the wireless devices.
+ * This is the most popular part of the Wireless Extensions ;-)
+ *
+ * This interface is a pure clone of /proc/net/dev (in net/core/dev.c).
+ * The content of the file is basically the content of "struct iw_statistics".
+ */
+
+#ifdef CONFIG_PROC_FS
+
+/* ---------------------------------------------------------------- */
+/*
+ * Print one entry (line) of /proc/net/wireless
+ */
+static inline int sprintf_wireless_stats(char *buffer, struct net_device *dev)
+{
+ /* Get stats from the driver */
+ struct iw_statistics *stats;
+ int size;
+
+ stats = get_wireless_stats(dev);
+ if (stats != (struct iw_statistics *) NULL) {
+ size = sprintf(buffer,
+ "%6s: %04x %3d%c %3d%c %3d%c %6d %6d %6d %6d %6d %6d\n",
+ dev->name,
+ stats->status,
+ stats->qual.qual,
+ stats->qual.updated & 1 ? '.' : ' ',
+ stats->qual.level,
+ stats->qual.updated & 2 ? '.' : ' ',
+ stats->qual.noise,
+ stats->qual.updated & 4 ? '.' : ' ',
+ stats->discard.nwid,
+ stats->discard.code,
+ stats->discard.fragment,
+ stats->discard.retries,
+ stats->discard.misc,
+ stats->miss.beacon);
+ stats->qual.updated = 0;
+ }
+ else
+ size = 0;
+
+ return size;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Print info for /proc/net/wireless (print all entries)
+ */
+int dev_get_wireless_info(char * buffer, char **start, off_t offset,
+ int length)
+{
+ int len = 0;
+ off_t begin = 0;
+ off_t pos = 0;
+ int size;
+
+ struct net_device * dev;
+
+ size = sprintf(buffer,
+ "Inter-| sta-| Quality | Discarded packets | Missed\n"
+ " face | tus | link level noise | nwid crypt frag retry misc | beacon\n"
+ );
+
+ pos += size;
+ len += size;
+
+ read_lock(&dev_base_lock);
+ for (dev = dev_base; dev != NULL; dev = dev->next) {
+ size = sprintf_wireless_stats(buffer + len, dev);
+ len += size;
+ pos = begin + len;
+
+ if (pos < offset) {
+ len = 0;
+ begin = pos;
+ }
+ if (pos > offset + length)
+ break;
+ }
+ read_unlock(&dev_base_lock);
+
+ *start = buffer + (offset - begin); /* Start of wanted data */
+ len -= (offset - begin); /* Start slop */
+ if (len > length)
+ len = length; /* Ending slop */
+ if (len < 0)
+ len = 0;
+
+ return len;
+}
+#endif /* CONFIG_PROC_FS */
+
+/************************** IOCTL SUPPORT **************************/
+/*
+ * The original user space API to configure all those Wireless Extensions
+ * is through IOCTLs.
+ * In there, we check if we need to call the new driver API (iw_handler)
+ * or just call the driver ioctl handler.
+ */
+
+/* ---------------------------------------------------------------- */
+/*
+ * Allow programatic access to /proc/net/wireless even if /proc
+ * doesn't exist... Also more efficient...
+ */
+static inline int dev_iwstats(struct net_device *dev, struct ifreq *ifr)
+{
+ /* Get stats from the driver */
+ struct iw_statistics *stats;
+
+ stats = get_wireless_stats(dev);
+ if (stats != (struct iw_statistics *) NULL) {
+ struct iwreq * wrq = (struct iwreq *)ifr;
+
+ /* Copy statistics to the user buffer */
+ if(copy_to_user(wrq->u.data.pointer, stats,
+ sizeof(struct iw_statistics)))
+ return -EFAULT;
+
+ /* Check if we need to clear the update flag */
+ if(wrq->u.data.flags != 0)
+ stats->qual.updated = 0;
+ return 0;
+ } else
+ return -EOPNOTSUPP;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Export the driver private handler definition
+ * They will be picked up by tools like iwpriv...
+ */
+static inline int ioctl_export_private(struct net_device * dev,
+ struct ifreq * ifr)
+{
+ struct iwreq * iwr = (struct iwreq *) ifr;
+
+ /* Check if the driver has something to export */
+ if((dev->wireless_handlers->num_private_args == 0) ||
+ (dev->wireless_handlers->private_args == NULL))
+ return -EOPNOTSUPP;
+
+ /* Check NULL pointer */
+ if(iwr->u.data.pointer == NULL)
+ return -EFAULT;
+#ifdef WE_STRICT_WRITE
+ /* Check if there is enough buffer up there */
+ if(iwr->u.data.length < (SIOCIWLASTPRIV - SIOCIWFIRSTPRIV + 1))
+ return -E2BIG;
+#endif /* WE_STRICT_WRITE */
+
+ /* Set the number of available ioctls. */
+ iwr->u.data.length = dev->wireless_handlers->num_private_args;
+
+ /* Copy structure to the user buffer. */
+ if (copy_to_user(iwr->u.data.pointer,
+ dev->wireless_handlers->private_args,
+ sizeof(struct iw_priv_args) * iwr->u.data.length))
+ return -EFAULT;
+
+ return 0;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Wrapper to call a standard Wireless Extension handler.
+ * We do various checks and also take care of moving data between
+ * user space and kernel space.
+ */
+static inline int ioctl_standard_call(struct net_device * dev,
+ struct ifreq * ifr,
+ unsigned int cmd,
+ iw_handler handler)
+{
+ struct iwreq * iwr = (struct iwreq *) ifr;
+ const struct iw_ioctl_description * descr;
+ struct iw_request_info info;
+ int ret = -EINVAL;
+
+ /* Get the description of the IOCTL */
+ descr = &(standard_ioctl[cmd - SIOCIWFIRST]);
+
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "%s : Found standard handler for 0x%04X\n",
+ ifr->ifr_name, cmd);
+ printk(KERN_DEBUG "Header type : %d, token type : %d, token_size : %d, max_token : %d\n", descr->header_type, descr->token_type, descr->token_size, descr->max_tokens);
+#endif /* WE_IOCTL_DEBUG */
+
+ /* Prepare the call */
+ info.cmd = cmd;
+ info.flags = 0;
+
+ /* Check if we have a pointer to user space data or not */
+ if(descr->header_type != IW_HEADER_TYPE_POINT) {
+ /* No extra arguments. Trivial to handle */
+ ret = handler(dev, &info, &(iwr->u), NULL);
+ } else {
+ char * extra;
+ int err;
+
+ /* Check what user space is giving us */
+ if(IW_IS_SET(cmd)) {
+ /* Check NULL pointer */
+ if((iwr->u.data.pointer == NULL) &&
+ (iwr->u.data.length != 0))
+ return -EFAULT;
+ /* Check if number of token fits within bounds */
+ if(iwr->u.data.length > descr->max_tokens)
+ return -E2BIG;
+ if(iwr->u.data.length < descr->min_tokens)
+ return -EINVAL;
+ } else {
+ /* Check NULL pointer */
+ if(iwr->u.data.pointer == NULL)
+ return -EFAULT;
+#ifdef WE_STRICT_WRITE
+ /* Check if there is enough buffer up there */
+ if(iwr->u.data.length < descr->max_tokens)
+ return -E2BIG;
+#endif /* WE_STRICT_WRITE */
+ }
+
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Malloc %d bytes\n",
+ descr->max_tokens * descr->token_size);
+#endif /* WE_IOCTL_DEBUG */
+
+ /* Always allocate for max space. Easier, and won't last
+ * long... */
+ extra = kmalloc(descr->max_tokens * descr->token_size,
+ GFP_KERNEL);
+ if (extra == NULL) {
+ return -ENOMEM;
+ }
+
+ /* If it is a SET, get all the extra data in here */
+ if(IW_IS_SET(cmd) && (iwr->u.data.length != 0)) {
+ err = copy_from_user(extra, iwr->u.data.pointer,
+ iwr->u.data.length *
+ descr->token_size);
+ if (err) {
+ kfree(extra);
+ return -EFAULT;
+ }
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Got %d bytes\n",
+ iwr->u.data.length * descr->token_size);
+#endif /* WE_IOCTL_DEBUG */
+ }
+
+ /* Call the handler */
+ ret = handler(dev, &info, &(iwr->u), extra);
+
+ /* If we have something to return to the user */
+ if (!ret && IW_IS_GET(cmd)) {
+ err = copy_to_user(iwr->u.data.pointer, extra,
+ iwr->u.data.length *
+ descr->token_size);
+ if (err)
+ ret = -EFAULT;
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Wrote %d bytes\n",
+ iwr->u.data.length * descr->token_size);
+#endif /* WE_IOCTL_DEBUG */
+ }
+
+ /* Cleanup - I told you it wasn't that long ;-) */
+ kfree(extra);
+ }
+
+ /* Call commit handler if needed and defined */
+ if(ret == -EIWCOMMIT)
+ ret = call_commit_handler(dev);
+
+ /* Here, we will generate the appropriate event if needed */
+
+ return ret;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Wrapper to call a private Wireless Extension handler.
+ * We do various checks and also take care of moving data between
+ * user space and kernel space.
+ * It's not as nice and slimline as the standard wrapper. The cause
+ * is struct iw_priv_args, which was not really designed for the
+ * job we are going here.
+ *
+ * IMPORTANT : This function prevent to set and get data on the same
+ * IOCTL and enforce the SET/GET convention. Not doing it would be
+ * far too hairy...
+ * If you need to set and get data at the same time, please don't use
+ * a iw_handler but process it in your ioctl handler (i.e. use the
+ * old driver API).
+ */
+static inline int ioctl_private_call(struct net_device * dev,
+ struct ifreq * ifr,
+ unsigned int cmd,
+ iw_handler handler)
+{
+ struct iwreq * iwr = (struct iwreq *) ifr;
+ struct iw_priv_args * descr = NULL;
+ struct iw_request_info info;
+ int extra_size = 0;
+ int i;
+ int ret = -EINVAL;
+
+ /* Get the description of the IOCTL */
+ for(i = 0; i < dev->wireless_handlers->num_private_args; i++)
+ if(cmd == dev->wireless_handlers->private_args[i].cmd) {
+ descr = &(dev->wireless_handlers->private_args[i]);
+ break;
+ }
+
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "%s : Found private handler for 0x%04X\n",
+ ifr->ifr_name, cmd);
+ if(descr) {
+ printk(KERN_DEBUG "Name %s, set %X, get %X\n",
+ descr->name, descr->set_args, descr->get_args);
+ }
+#endif /* WE_IOCTL_DEBUG */
+
+ /* Compute the size of the set/get arguments */
+ if(descr != NULL) {
+ if(IW_IS_SET(cmd)) {
+ /* Size of set arguments */
+ extra_size = get_priv_size(descr->set_args);
+
+ /* Does it fits in iwr ? */
+ if((descr->set_args & IW_PRIV_SIZE_FIXED) &&
+ (extra_size < IFNAMSIZ))
+ extra_size = 0;
+ } else {
+ /* Size of set arguments */
+ extra_size = get_priv_size(descr->get_args);
+
+ /* Does it fits in iwr ? */
+ if((descr->get_args & IW_PRIV_SIZE_FIXED) &&
+ (extra_size < IFNAMSIZ))
+ extra_size = 0;
+ }
+ }
+
+ /* Prepare the call */
+ info.cmd = cmd;
+ info.flags = 0;
+
+ /* Check if we have a pointer to user space data or not. */
+ if(extra_size == 0) {
+ /* No extra arguments. Trivial to handle */
+ ret = handler(dev, &info, &(iwr->u), (char *) &(iwr->u));
+ } else {
+ char * extra;
+ int err;
+
+ /* Check what user space is giving us */
+ if(IW_IS_SET(cmd)) {
+ /* Check NULL pointer */
+ if((iwr->u.data.pointer == NULL) &&
+ (iwr->u.data.length != 0))
+ return -EFAULT;
+
+ /* Does it fits within bounds ? */
+ if(iwr->u.data.length > (descr->set_args &
+ IW_PRIV_SIZE_MASK))
+ return -E2BIG;
+ } else {
+ /* Check NULL pointer */
+ if(iwr->u.data.pointer == NULL)
+ return -EFAULT;
+ }
+
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Malloc %d bytes\n", extra_size);
+#endif /* WE_IOCTL_DEBUG */
+
+ /* Always allocate for max space. Easier, and won't last
+ * long... */
+ extra = kmalloc(extra_size, GFP_KERNEL);
+ if (extra == NULL) {
+ return -ENOMEM;
+ }
+
+ /* If it is a SET, get all the extra data in here */
+ if(IW_IS_SET(cmd) && (iwr->u.data.length != 0)) {
+ err = copy_from_user(extra, iwr->u.data.pointer,
+ extra_size);
+ if (err) {
+ kfree(extra);
+ return -EFAULT;
+ }
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Got %d elem\n", iwr->u.data.length);
+#endif /* WE_IOCTL_DEBUG */
+ }
+
+ /* Call the handler */
+ ret = handler(dev, &info, &(iwr->u), extra);
+
+ /* If we have something to return to the user */
+ if (!ret && IW_IS_GET(cmd)) {
+ err = copy_to_user(iwr->u.data.pointer, extra,
+ extra_size);
+ if (err)
+ ret = -EFAULT;
+#ifdef WE_IOCTL_DEBUG
+ printk(KERN_DEBUG "Wrote %d elem\n",
+ iwr->u.data.length);
+#endif /* WE_IOCTL_DEBUG */
+ }
+
+ /* Cleanup - I told you it wasn't that long ;-) */
+ kfree(extra);
+ }
+
+
+ /* Call commit handler if needed and defined */
+ if(ret == -EIWCOMMIT)
+ ret = call_commit_handler(dev);
+
+ return ret;
+}
+
+/* ---------------------------------------------------------------- */
+/*
+ * Main IOCTl dispatcher. Called from the main networking code
+ * (dev_ioctl() in net/core/dev.c).
+ * Check the type of IOCTL and call the appropriate wrapper...
+ */
+int wireless_process_ioctl(struct ifreq *ifr, unsigned int cmd)
+{
+ struct net_device *dev;
+ iw_handler handler;
+
+ /* Permissions are already checked in dev_ioctl() before calling us.
+ * The copy_to/from_user() of ifr is also dealt with in there */
+
+ /* Make sure the device exist */
+ if ((dev = __dev_get_by_name(ifr->ifr_name)) == NULL)
+ return -ENODEV;
+
+ /* A bunch of special cases, then the generic case...
+ * Note that 'cmd' is already filtered in dev_ioctl() with
+ * (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST) */
+ switch(cmd)
+ {
+ case SIOCGIWSTATS:
+ /* Get Wireless Stats */
+ return dev_iwstats(dev, ifr);
+
+ case SIOCGIWPRIV:
+ /* Check if we have some wireless handlers defined */
+ if(dev->wireless_handlers != NULL) {
+ /* We export to user space the definition of
+ * the private handler ourselves */
+ return ioctl_export_private(dev, ifr);
+ }
+ // ## Fall-through for old API ##
+ default:
+ /* Generic IOCTL */
+ /* Basic check */
+ if (!netif_device_present(dev))
+ return -ENODEV;
+ /* New driver API : try to find the handler */
+ handler = get_handler(dev, cmd);
+ if(handler != NULL) {
+ /* Standard and private are not the same */
+ if(cmd < SIOCIWFIRSTPRIV)
+ return ioctl_standard_call(dev,
+ ifr,
+ cmd,
+ handler);
+ else
+ return ioctl_private_call(dev,
+ ifr,
+ cmd,
+ handler);
+ }
+ /* Old driver API : call driver ioctl handler */
+ if (dev->do_ioctl) {
+ return dev->do_ioctl(dev, ifr, cmd);
+ }
+ return -EOPNOTSUPP;
+ }
+ /* Not reached */
+ return -EINVAL;
+}


2002-03-14 03:05:49

by Alan

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

> the new Wireless Extension API to 2.4.X so that they can hack their
> drivers without having to deal with 2.5.X. This patch just do that, so
> could you please add that to your tree ?

Would it be better to leave the patch where said driver people can get it,
and when they submit drivers needing it (if ever) then merge it ?

2002-03-14 03:09:49

by Jeff Garzik

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

Tangential question, what's up with the prism2 driver?

It seems like everybody I meet these days has a wireless card which uses
the prism2 driver from linux-wlan.org. And since I just got two of
these cards (D-Link DWL-650), I am strongly tempted to merge the driver
into the kernel.

How well does the prism2 driver work with the current wireless driver API?
Is there any particular reason why it is not in the kernel now?

Jeff




2002-03-14 03:12:18

by Jean Tourrilhes

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Thu, Mar 14, 2002 at 03:20:47AM +0000, Alan Cox wrote:
> > the new Wireless Extension API to 2.4.X so that they can hack their
> > drivers without having to deal with 2.5.X. This patch just do that, so
> > could you please add that to your tree ?
>
> Would it be better to leave the patch where said driver people can get it,
> and when they submit drivers needing it (if ever) then merge it ?

Maybe, maybe not.
1) Most Wireless LAN driver live outside the kernel. So, their
evolution is somewhat decoupled to the kernel, so the earlier the
patch goes it the better it is for those.
2) David Gibson, maintainer of the Orinoco driver, told me
that he would merge my new-API orinoco patches in his driver only when
the new API would be in 2.4.x (as you may have noticed, he hasn't
updated 2.5.X for a while). Chicken and Eggs.
3) Risk is minimal, bloat is minimal. What's the penality of
merging it now ?

I'm open to suggestions and will do what's best for
everybody. What other people think ?
But I feel now is a good time (as it seem that 2.4.19 will
have to go through some stabilisation process).

Thanks for the comment ;-)

Jean

2002-03-14 03:14:08

by Alan

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

> updated 2.5.X for a while). Chicken and Eggs.
> 3) Risk is minimal, bloat is minimal. What's the penality of
> merging it now ?

Sounds good reason to me

2002-03-14 03:16:08

by Jeff Garzik

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

Jean Tourrilhes wrote:

>1) Most Wireless LAN driver live outside the kernel. So, their
>evolution is somewhat decoupled to the kernel, so the earlier the
>patch goes it the better it is for those.
>
As you may have gathered from my last email, this is a bit annoying when
trying to find and stabalize a driver for a card you just got :)

> 2) David Gibson, maintainer of the Orinoco driver, told me
>that he would merge my new-API orinoco patches in his driver only when
>the new API would be in 2.4.x (as you may have noticed, he hasn't
>updated 2.5.X for a while). Chicken and Eggs.
>
Does that mean orinoco updates are coming for 2.5.x?

Any idea if the other drivers are getting updates too, like airo?

>I'm open to suggestions and will do what's best for
>everybody. What other people think ?
> But I feel now is a good time (as it seem that 2.4.19 will
>have to go through some stabilisation process).
>
My standard rule of thumb is turning out to be, deploy something
2.4.x-worthy in 2.5.x, wait a while, and then deploy it in 2.4.x. I
have no particular opinion about whether your patch should go in -right
now- but it seems like a fair patch for 2.4.x sooner or later.

Jeff





2002-03-14 03:19:49

by Jean Tourrilhes

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Wed, Mar 13, 2002 at 10:09:09PM -0500, Jeff Garzik wrote:
> Tangential question, what's up with the prism2 driver?

Which one ? ;-)
Orinoco can work sometime with those cards, David may have
some updates in the pipeline (especially for WEP). The Prism2 driver
from Jouni is actually pretty nice and dedicated for those cards.

> It seems like everybody I meet these days has a wireless card which uses
> the prism2 driver from linux-wlan.org. And since I just got two of
> these cards (D-Link DWL-650), I am strongly tempted to merge the driver
> into the kernel.

Before doing that, have a chat with Alan that had a close look
with it for a partial opinion. And you may want to go through the
source yourself.
My personal take is that linux-wlan-ng is nice, does a lot,
but is huge and complex, and it seems that Mark is not that interested
in merging it in the kernel. Keeping it as a separate module may make
most sense.
BTW, there was a thread about this on the linux-wlan mailing
lists.

> How well does the prism2 driver work with the current wireless driver API?

Very basic support (mostly read only). linux-wlan-ng has it's
own set of tools and MIB.

> Is there any particular reason why it is not in the kernel now?

Nobody ever pushed it in. I can't force people to add their
driver int he kernel (hint : compare the list on my web page with the
list in the kernel).

> Jeff

Have fun...

Jean

2002-03-14 03:26:19

by Jean Tourrilhes

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Wed, Mar 13, 2002 at 10:15:30PM -0500, Jeff Garzik wrote:
> Jean Tourrilhes wrote:
>
> >1) Most Wireless LAN driver live outside the kernel. So, their
> >evolution is somewhat decoupled to the kernel, so the earlier the
> >patch goes it the better it is for those.
> >
> As you may have gathered from my last email, this is a bit annoying when
> trying to find and stabalize a driver for a card you just got :)

Especially that there are so many driver to choose from, and
each is broken in some unique way.

> > 2) David Gibson, maintainer of the Orinoco driver, told me
> >that he would merge my new-API orinoco patches in his driver only when
> >the new API would be in 2.4.x (as you may have noticed, he hasn't
> >updated 2.5.X for a while). Chicken and Eggs.
> >
> Does that mean orinoco updates are coming for 2.5.x?

I asked him a couple of times, but so far nothing. He is
working hard on various issues in the driver.

> Any idea if the other drivers are getting updates too, like airo?

I think the CVS has a new version, including change to the
monitoring mode, but airo list is pretty quiet lately. airo.c is
pretty stable and bug free, so updates are not critical. I still need
to finish to convert it to the new API.

> My standard rule of thumb is turning out to be, deploy something
> 2.4.x-worthy in 2.5.x, wait a while, and then deploy it in 2.4.x. I
> have no particular opinion about whether your patch should go in -right
> now- but it seems like a fair patch for 2.4.x sooner or later.

Yes, you are right, it has been in 2.5.X for only two months,
which is quite short.

> Jeff

Have fun...

Jean

2002-03-14 10:46:25

by Bjorn Wesen

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Wed, 13 Mar 2002, Jeff Garzik wrote:
> Tangential question, what's up with the prism2 driver?
>
> It seems like everybody I meet these days has a wireless card which uses
> the prism2 driver from linux-wlan.org. And since I just got two of
> these cards (D-Link DWL-650), I am strongly tempted to merge the driver
> into the kernel.

Just a datapoint:

The orinico driver (already in the kernel) works fine with the DWL-650 card.
Tried it some days ago.. not a very big field trial but I inserted the card
and I got an eth0 from it and it worked, so thats the way I like it :)

On the other hand, the driver from linux-wlan (also for prism2-based cards)
did not compile because it required the external pcmcia-cs source-code etc..
it was not really "drop and compile". Anyway, I'm just happy one of them
worked :)

/Bjorn

2002-03-14 10:49:55

by Jeff Garzik

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

Bjorn Wesen wrote:

>Just a datapoint:
>
>The orinico driver (already in the kernel) works fine with the DWL-650 card.
>Tried it some days ago.. not a very big field trial but I inserted the card
>and I got an eth0 from it and it worked, so thats the way I like it :)
>


Not "just" a datapoint, a useful one. Thanks.

Jeff



2002-03-14 11:52:38

by Wichert Akkerman

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

In article <[email protected]>,
Bjorn Wesen <[email protected]> wrote:
>The orinico driver (already in the kernel) works fine with the DWL-650 card.
>Tried it some days ago.. not a very big field trial but I inserted the card
>and I got an eth0 from it and it worked, so thats the way I like it :)

Last time I tried the orinoco driver it failed to see my Lucent orinoco
card. The wlan-cs driver from the pcmcia sources works just fine though.

Wichert.


--
_________________________________________________________________
/ Nothing is fool-proof to a sufficiently talented fool \
| [email protected] http://www.liacs.nl/~wichert/ |
| 1024D/2FA3BC2D 576E 100B 518D 2F16 36B0 2805 3CB8 9250 2FA3 BC2D |

2002-03-15 03:37:20

by David Gibson

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Wed, Mar 13, 2002 at 10:15:30PM -0500, Jeff Garzik wrote:
> Jean Tourrilhes wrote:
>
> >1) Most Wireless LAN driver live outside the kernel. So, their
> >evolution is somewhat decoupled to the kernel, so the earlier the
> >patch goes it the better it is for those.
> >
> As you may have gathered from my last email, this is a bit annoying when
> trying to find and stabalize a driver for a card you just got :)

> > 2) David Gibson, maintainer of the Orinoco driver, told me
> >that he would merge my new-API orinoco patches in his driver only when
> >the new API would be in 2.4.x (as you may have noticed, he hasn't
> >updated 2.5.X for a while). Chicken and Eggs.
> >
> Does that mean orinoco updates are coming for 2.5.x?

Yes, I'll try to send some in. When 2.5 had just started I figured
Linus was busy enough with the bio stuff and the driver updates could
wait, since then I've been busy with other things so I haven't gotten
around to sending patches.

AFAIK none of the APIs which are relevant to me have changed so it
should just be a matter of copying the files across from 2.4.

--
David Gibson | For every complex problem there is a
[email protected] | solution which is simple, neat and
| wrong. -- H.L. Mencken
http://www.ozlabs.org/people/dgibson

2002-03-15 03:42:00

by Jean Tourrilhes

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Fri, Mar 15, 2002 at 01:43:00PM +1100, David Gibson wrote:
>
> Yes, I'll try to send some in. When 2.5 had just started I figured
> Linus was busy enough with the bio stuff and the driver updates could
> wait, since then I've been busy with other things so I haven't gotten
> around to sending patches.
>
> AFAIK none of the APIs which are relevant to me have changed so it
> should just be a matter of copying the files across from 2.4.

Hu ho... You are reading LKML. I'd better be careful about
what I'm saying ;-)
By the way, you are right, I've been using 9b in 2.5.6 without
any trouble. But, it seem that my patch will get in 2.4.X, so you will
have it both way.

By the way, what's the situation regarding our famous ALLOC
irq handling ? Any improvement ? I haven't tried v10, so I just wonder
if it's worth upgrading ;-)

Have fun...

Jean

2002-03-15 05:31:18

by David Gibson

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Thu, Mar 14, 2002 at 05:49:15AM -0500, Jeff Garzik wrote:
> Bjorn Wesen wrote:
>
> >Just a datapoint:
> >
> >The orinico driver (already in the kernel) works fine with the DWL-650
> >card. Tried it some days ago.. not a very big field trial but I inserted
> >the card and I got an eth0 from it and it worked, so thats the way I like
> >it :)
>
>
> Not "just" a datapoint, a useful one. Thanks.

Sadly not everybody is having as much luck. A lot of people are
reporting terribly throughput on Intersil cards like the DWL-650 and I
haven't managed to track the problem down yet.

--
David Gibson | For every complex problem there is a
[email protected] | solution which is simple, neat and
| wrong. -- H.L. Mencken
http://www.ozlabs.org/people/dgibson

2002-03-16 00:38:42

by Oliver Xymoron

[permalink] [raw]
Subject: Re: [PATCH 2.4.19-pre3] New wireless driver API part 1

On Fri, 15 Mar 2002, David Gibson wrote:

> On Thu, Mar 14, 2002 at 05:49:15AM -0500, Jeff Garzik wrote:
> > Bjorn Wesen wrote:
> >
> > >Just a datapoint:
> > >
> > >The orinico driver (already in the kernel) works fine with the DWL-650
> > >card. Tried it some days ago.. not a very big field trial but I inserted
> > >the card and I got an eth0 from it and it worked, so thats the way I like
> > >it :)
> >
> >
> > Not "just" a datapoint, a useful one. Thanks.
>
> Sadly not everybody is having as much luck. A lot of people are
> reporting terribly throughput on Intersil cards like the DWL-650 and I
> haven't managed to track the problem down yet.

Here's some further datapoints:

Vaio Z505SX, kernels through 2.4.16 - works great
IBM T22, 2.4.16+sound, 2.4.17+lowlatency-sound - works great
T22, 2.4.18+ and 2.5.5 - 2-3K/s

--
"Love the dolphins," she advised him. "Write by W.A.S.T.E.."

2002-03-29 01:09:32

by Jean Tourrilhes

[permalink] [raw]
Subject: [PATCH 2.4.19-pre4] New wireless driver API part 1

Hi again Marcelo,

Sorry to insist, but I need to get this patch in 2.4.X. I'm
now blocked on two wireless driver (orinoco.c and airo.c) because
their maintainers can't/won't use 2.5.X. And I hate to see all those
patches bitroting.

The deal : new driver API for Wireless Extensions (first
part). More details below.
Patch *attached*. I re-diffed against 2.4.19-pre4, and tested
with both 2.4.19-pre4 and 2.4.19-pre4-ac2 (for ac2, the patch applies
with fuzz).
Why would you accept the patch :
o More than 2 month already in 2.5.X
o Jeff says it's 2.4.X material :
http://marc.theaimsgroup.com/?l=linux-kernel&m=101607595127684&w=2
o Alan says "sounds good" :
http://marc.theaimsgroup.com/?l=linux-kernel&m=101607595327690&w=2

Tell me if there is any complications...

Have fun...

Jean

On Wed, Mar 13, 2002 at 06:59:15PM -0800, jt wrote:
> Hi Marcelo,
>
> Some Wireless LAN driver authors have requested me to backport
> the new Wireless Extension API to 2.4.X so that they can hack their
> drivers without having to deal with 2.5.X. This patch just do that, so
> could you please add that to your tree ?
>
> This patch is *very* safe, especially compared to other stuff
> that made it's way in your kernel :
> o was integrated in 2.5.3-pre2, straight backport (same patch)
> o tested with various drivers in 2.5.X
> o reviewed by Jeff Garzik ;-)
> o doesn't change anything for driver still using the old API
> (so current kernel/pcmcia drivers won't get broken).
> o no part of the kernel using the new API, only outside drivers
>
> For stability reason, I won't change any of the current driver
> in the kernel, and those will continue to use the old API (unless you
> or Jeff request me to update them).
> Also, I may send you in the future the second part of the new
> API (that got just integrated in 2.5.7-pre1), but I'm not in a
> hurry... So, for now, first part only ;-)
> Below, you will find the original blurb I sent to Linus...
>
> Thanks in advance...
>
> Jean
>
> ---------------------------------------------------------
> Hi Linus,
>
> I feel the first part my new driver API for Wireless
> Extensions is ready to go in the kernel.
>
> I don't want to bore you too much with details, but for
> Wireless Extensions it replace the call to the device driver ioctl
> function with individual call to handlers functions.
> The advantage of the new API are :
> o Handling of Extensions in driver broken in small contained functions
> o Tighter checks of ioctl before calling the driver
> o Flexible commit strategy (at least, the start of it)
> o Backward compatibility (can be mixed with old API)
> o Remove all ioctl assumptions in the driver
> o Driver doesn't have to worry about memory and user-space issues
> (copy_to/from_user handled by the API, not the driver)
> The last point is important for the following reasons :
> o You are now able to call the new driver API from any API you
> want (including from within other parts of the kernel).
> o Common mistakes are avoided (buffer overflow, user space copy
> with irq disabled and so on).
> More details available at :
> http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Tools.html#newapi
>
> Compare to my earlier patch (sent to the list in December), I
> made the following changes :
> o cleanup comments
> o use "return X;" instead of "return(X);" (pointed out to me
> by Jeff - thanks).
>
> This has been tested on 2.5.3-pre2. Would you mind adding that
> to your kernel ?
> Regards,
>
> Jean


Attachments:
(No filename) (3.63 kB)
iw_handlers.w13-6.diff (50.74 kB)
Download all attachments