Merge with git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git
[linux-2.6] / drivers / scsi / sg.c
1 /*
2  *  History:
3  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
4  *           to allow user process control of SCSI devices.
5  *  Development Sponsored by Killy Corp. NY NY
6  *
7  * Original driver (sg.c):
8  *        Copyright (C) 1992 Lawrence Foard
9  * Version 2 and 3 extensions to driver:
10  *        Copyright (C) 1998 - 2005 Douglas Gilbert
11  *
12  *  Modified  19-JAN-1998  Richard Gooch <rgooch@atnf.csiro.au>  Devfs support
13  *
14  * This program is free software; you can redistribute it and/or modify
15  * it under the terms of the GNU General Public License as published by
16  * the Free Software Foundation; either version 2, or (at your option)
17  * any later version.
18  *
19  */
20
21 static int sg_version_num = 30533;      /* 2 digits for each component */
22 #define SG_VERSION_STR "3.5.33"
23
24 /*
25  *  D. P. Gilbert (dgilbert@interlog.com, dougg@triode.net.au), notes:
26  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
27  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
28  *        (otherwise the macros compile to empty statements).
29  *
30  */
31 #include <linux/config.h>
32 #include <linux/module.h>
33
34 #include <linux/fs.h>
35 #include <linux/kernel.h>
36 #include <linux/sched.h>
37 #include <linux/string.h>
38 #include <linux/mm.h>
39 #include <linux/errno.h>
40 #include <linux/mtio.h>
41 #include <linux/ioctl.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/poll.h>
45 #include <linux/smp_lock.h>
46 #include <linux/moduleparam.h>
47 #include <linux/cdev.h>
48 #include <linux/seq_file.h>
49 #include <linux/blkdev.h>
50 #include <linux/delay.h>
51 #include <linux/scatterlist.h>
52
53 #include "scsi.h"
54 #include <scsi/scsi_dbg.h>
55 #include <scsi/scsi_host.h>
56 #include <scsi/scsi_driver.h>
57 #include <scsi/scsi_ioctl.h>
58 #include <scsi/sg.h>
59
60 #include "scsi_logging.h"
61
62 #ifdef CONFIG_SCSI_PROC_FS
63 #include <linux/proc_fs.h>
64 static char *sg_version_date = "20050908";
65
66 static int sg_proc_init(void);
67 static void sg_proc_cleanup(void);
68 #endif
69
70 #define SG_ALLOW_DIO_DEF 0
71 #define SG_ALLOW_DIO_CODE /* compile out by commenting this define */
72
73 #define SG_MAX_DEVS 32768
74
75 /*
76  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
77  * Then when using 32 bit integers x * m may overflow during the calculation.
78  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
79  * calculates the same, but prevents the overflow when both m and d
80  * are "small" numbers (like HZ and USER_HZ).
81  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
82  * in 32 bits.
83  */
84 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
85
86 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
87
88 int sg_big_buff = SG_DEF_RESERVED_SIZE;
89 /* N.B. This variable is readable and writeable via
90    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
91    of this size (or less if there is not enough memory) will be reserved
92    for use by this file descriptor. [Deprecated usage: this variable is also
93    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
94    the kernel (i.e. it is not a module).] */
95 static int def_reserved_size = -1;      /* picks up init parameter */
96 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
97
98 #define SG_SECTOR_SZ 512
99 #define SG_SECTOR_MSK (SG_SECTOR_SZ - 1)
100
101 #define SG_DEV_ARR_LUMP 32      /* amount to over allocate sg_dev_arr by */
102
103 static int sg_add(struct class_device *, struct class_interface *);
104 static void sg_remove(struct class_device *, struct class_interface *);
105
106 static DEFINE_RWLOCK(sg_dev_arr_lock);  /* Also used to lock
107                                                            file descriptor list for device */
108
109 static struct class_interface sg_interface = {
110         .add            = sg_add,
111         .remove         = sg_remove,
112 };
113
114 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
115         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
116         unsigned short sglist_len; /* size of malloc'd scatter-gather list ++ */
117         unsigned bufflen;       /* Size of (aggregate) data buffer */
118         unsigned b_malloc_len;  /* actual len malloc'ed in buffer */
119         struct scatterlist *buffer;/* scatter list */
120         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
121         unsigned char cmd_opcode; /* first byte of command */
122 } Sg_scatter_hold;
123
124 struct sg_device;               /* forward declarations */
125 struct sg_fd;
126
127 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
128         struct sg_request *nextrp;      /* NULL -> tail request (slist) */
129         struct sg_fd *parentfp; /* NULL -> not in use */
130         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
131         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
132         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
133         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
134         char orphan;            /* 1 -> drop on sight, 0 -> normal */
135         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
136         volatile char done;     /* 0->before bh, 1->before read, 2->read */
137 } Sg_request;
138
139 typedef struct sg_fd {          /* holds the state of a file descriptor */
140         struct sg_fd *nextfp;   /* NULL when last opened fd on this device */
141         struct sg_device *parentdp;     /* owning device */
142         wait_queue_head_t read_wait;    /* queue read until command done */
143         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
144         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
145         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
146         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
147         unsigned save_scat_len; /* original length of trunc. scat. element */
148         Sg_request *headrp;     /* head of request slist, NULL->empty */
149         struct fasync_struct *async_qp; /* used by asynchronous notification */
150         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
151         char low_dma;           /* as in parent but possibly overridden to 1 */
152         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
153         volatile char closed;   /* 1 -> fd closed but request(s) outstanding */
154         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
155         char next_cmd_len;      /* 0 -> automatic (def), >0 -> use on next write() */
156         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
157         char mmap_called;       /* 0 -> mmap() never called on this fd */
158 } Sg_fd;
159
160 typedef struct sg_device { /* holds the state of each scsi generic device */
161         struct scsi_device *device;
162         wait_queue_head_t o_excl_wait;  /* queue open() when O_EXCL in use */
163         int sg_tablesize;       /* adapter's max scatter-gather table size */
164         Sg_fd *headfp;          /* first open fd belonging to this device */
165         volatile char detached; /* 0->attached, 1->detached pending removal */
166         volatile char exclude;  /* opened for exclusive access */
167         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
168         struct gendisk *disk;
169         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
170 } Sg_device;
171
172 static int sg_fasync(int fd, struct file *filp, int mode);
173 /* tasklet or soft irq callback */
174 static void sg_cmd_done(void *data, char *sense, int result, int resid);
175 static int sg_start_req(Sg_request * srp);
176 static void sg_finish_rem_req(Sg_request * srp);
177 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
178 static int sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp,
179                          int tablesize);
180 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
181                            Sg_request * srp);
182 static ssize_t sg_new_write(Sg_fd * sfp, const char __user *buf, size_t count,
183                             int blocking, int read_only, Sg_request ** o_srp);
184 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
185                            unsigned char *cmnd, int timeout, int blocking);
186 static int sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
187                       int wr_xf, int *countp, unsigned char __user **up);
188 static int sg_write_xfer(Sg_request * srp);
189 static int sg_read_xfer(Sg_request * srp);
190 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
191 static void sg_remove_scat(Sg_scatter_hold * schp);
192 static void sg_build_reserve(Sg_fd * sfp, int req_size);
193 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
194 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
195 static struct page *sg_page_malloc(int rqSz, int lowDma, int *retSzp);
196 static void sg_page_free(struct page *page, int size);
197 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
198 static int sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
199 static void __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
200 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
201 static Sg_request *sg_add_request(Sg_fd * sfp);
202 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
203 static int sg_res_in_use(Sg_fd * sfp);
204 static int sg_allow_access(unsigned char opcode, char dev_type);
205 static int sg_build_direct(Sg_request * srp, Sg_fd * sfp, int dxfer_len);
206 static Sg_device *sg_get_dev(int dev);
207 #ifdef CONFIG_SCSI_PROC_FS
208 static int sg_last_dev(void);
209 #endif
210
211 static Sg_device **sg_dev_arr = NULL;
212 static int sg_dev_max;
213 static int sg_nr_dev;
214
215 #define SZ_SG_HEADER sizeof(struct sg_header)
216 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
217 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
218 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
219
220 static int
221 sg_open(struct inode *inode, struct file *filp)
222 {
223         int dev = iminor(inode);
224         int flags = filp->f_flags;
225         struct request_queue *q;
226         Sg_device *sdp;
227         Sg_fd *sfp;
228         int res;
229         int retval;
230
231         nonseekable_open(inode, filp);
232         SCSI_LOG_TIMEOUT(3, printk("sg_open: dev=%d, flags=0x%x\n", dev, flags));
233         sdp = sg_get_dev(dev);
234         if ((!sdp) || (!sdp->device))
235                 return -ENXIO;
236         if (sdp->detached)
237                 return -ENODEV;
238
239         /* This driver's module count bumped by fops_get in <linux/fs.h> */
240         /* Prevent the device driver from vanishing while we sleep */
241         retval = scsi_device_get(sdp->device);
242         if (retval)
243                 return retval;
244
245         if (!((flags & O_NONBLOCK) ||
246               scsi_block_when_processing_errors(sdp->device))) {
247                 retval = -ENXIO;
248                 /* we are in error recovery for this device */
249                 goto error_out;
250         }
251
252         if (flags & O_EXCL) {
253                 if (O_RDONLY == (flags & O_ACCMODE)) {
254                         retval = -EPERM; /* Can't lock it with read only access */
255                         goto error_out;
256                 }
257                 if (sdp->headfp && (flags & O_NONBLOCK)) {
258                         retval = -EBUSY;
259                         goto error_out;
260                 }
261                 res = 0;
262                 __wait_event_interruptible(sdp->o_excl_wait,
263                         ((sdp->headfp || sdp->exclude) ? 0 : (sdp->exclude = 1)), res);
264                 if (res) {
265                         retval = res;   /* -ERESTARTSYS because signal hit process */
266                         goto error_out;
267                 }
268         } else if (sdp->exclude) {      /* some other fd has an exclusive lock on dev */
269                 if (flags & O_NONBLOCK) {
270                         retval = -EBUSY;
271                         goto error_out;
272                 }
273                 res = 0;
274                 __wait_event_interruptible(sdp->o_excl_wait, (!sdp->exclude),
275                                            res);
276                 if (res) {
277                         retval = res;   /* -ERESTARTSYS because signal hit process */
278                         goto error_out;
279                 }
280         }
281         if (sdp->detached) {
282                 retval = -ENODEV;
283                 goto error_out;
284         }
285         if (!sdp->headfp) {     /* no existing opens on this device */
286                 sdp->sgdebug = 0;
287                 q = sdp->device->request_queue;
288                 sdp->sg_tablesize = min(q->max_hw_segments,
289                                         q->max_phys_segments);
290         }
291         if ((sfp = sg_add_sfp(sdp, dev)))
292                 filp->private_data = sfp;
293         else {
294                 if (flags & O_EXCL)
295                         sdp->exclude = 0;       /* undo if error */
296                 retval = -ENOMEM;
297                 goto error_out;
298         }
299         return 0;
300
301       error_out:
302         scsi_device_put(sdp->device);
303         return retval;
304 }
305
306 /* Following function was formerly called 'sg_close' */
307 static int
308 sg_release(struct inode *inode, struct file *filp)
309 {
310         Sg_device *sdp;
311         Sg_fd *sfp;
312
313         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
314                 return -ENXIO;
315         SCSI_LOG_TIMEOUT(3, printk("sg_release: %s\n", sdp->disk->disk_name));
316         sg_fasync(-1, filp, 0); /* remove filp from async notification list */
317         if (0 == sg_remove_sfp(sdp, sfp)) {     /* Returns 1 when sdp gone */
318                 if (!sdp->detached) {
319                         scsi_device_put(sdp->device);
320                 }
321                 sdp->exclude = 0;
322                 wake_up_interruptible(&sdp->o_excl_wait);
323         }
324         return 0;
325 }
326
327 static ssize_t
328 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
329 {
330         Sg_device *sdp;
331         Sg_fd *sfp;
332         Sg_request *srp;
333         int req_pack_id = -1;
334         sg_io_hdr_t *hp;
335         struct sg_header *old_hdr = NULL;
336         int retval = 0;
337
338         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
339                 return -ENXIO;
340         SCSI_LOG_TIMEOUT(3, printk("sg_read: %s, count=%d\n",
341                                    sdp->disk->disk_name, (int) count));
342
343         if (!access_ok(VERIFY_WRITE, buf, count))
344                 return -EFAULT;
345         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
346                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
347                 if (!old_hdr)
348                         return -ENOMEM;
349                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
350                         retval = -EFAULT;
351                         goto free_old_hdr;
352                 }
353                 if (old_hdr->reply_len < 0) {
354                         if (count >= SZ_SG_IO_HDR) {
355                                 sg_io_hdr_t *new_hdr;
356                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
357                                 if (!new_hdr) {
358                                         retval = -ENOMEM;
359                                         goto free_old_hdr;
360                                 }
361                                 retval =__copy_from_user
362                                     (new_hdr, buf, SZ_SG_IO_HDR);
363                                 req_pack_id = new_hdr->pack_id;
364                                 kfree(new_hdr);
365                                 if (retval) {
366                                         retval = -EFAULT;
367                                         goto free_old_hdr;
368                                 }
369                         }
370                 } else
371                         req_pack_id = old_hdr->pack_id;
372         }
373         srp = sg_get_rq_mark(sfp, req_pack_id);
374         if (!srp) {             /* now wait on packet to arrive */
375                 if (sdp->detached) {
376                         retval = -ENODEV;
377                         goto free_old_hdr;
378                 }
379                 if (filp->f_flags & O_NONBLOCK) {
380                         retval = -EAGAIN;
381                         goto free_old_hdr;
382                 }
383                 while (1) {
384                         retval = 0; /* following macro beats race condition */
385                         __wait_event_interruptible(sfp->read_wait,
386                                 (sdp->detached ||
387                                 (srp = sg_get_rq_mark(sfp, req_pack_id))), 
388                                 retval);
389                         if (sdp->detached) {
390                                 retval = -ENODEV;
391                                 goto free_old_hdr;
392                         }
393                         if (0 == retval)
394                                 break;
395
396                         /* -ERESTARTSYS as signal hit process */
397                         goto free_old_hdr;
398                 }
399         }
400         if (srp->header.interface_id != '\0') {
401                 retval = sg_new_read(sfp, buf, count, srp);
402                 goto free_old_hdr;
403         }
404
405         hp = &srp->header;
406         if (old_hdr == NULL) {
407                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
408                 if (! old_hdr) {
409                         retval = -ENOMEM;
410                         goto free_old_hdr;
411                 }
412         }
413         memset(old_hdr, 0, SZ_SG_HEADER);
414         old_hdr->reply_len = (int) hp->timeout;
415         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
416         old_hdr->pack_id = hp->pack_id;
417         old_hdr->twelve_byte =
418             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
419         old_hdr->target_status = hp->masked_status;
420         old_hdr->host_status = hp->host_status;
421         old_hdr->driver_status = hp->driver_status;
422         if ((CHECK_CONDITION & hp->masked_status) ||
423             (DRIVER_SENSE & hp->driver_status))
424                 memcpy(old_hdr->sense_buffer, srp->sense_b,
425                        sizeof (old_hdr->sense_buffer));
426         switch (hp->host_status) {
427         /* This setup of 'result' is for backward compatibility and is best
428            ignored by the user who should use target, host + driver status */
429         case DID_OK:
430         case DID_PASSTHROUGH:
431         case DID_SOFT_ERROR:
432                 old_hdr->result = 0;
433                 break;
434         case DID_NO_CONNECT:
435         case DID_BUS_BUSY:
436         case DID_TIME_OUT:
437                 old_hdr->result = EBUSY;
438                 break;
439         case DID_BAD_TARGET:
440         case DID_ABORT:
441         case DID_PARITY:
442         case DID_RESET:
443         case DID_BAD_INTR:
444                 old_hdr->result = EIO;
445                 break;
446         case DID_ERROR:
447                 old_hdr->result = (srp->sense_b[0] == 0 && 
448                                   hp->masked_status == GOOD) ? 0 : EIO;
449                 break;
450         default:
451                 old_hdr->result = EIO;
452                 break;
453         }
454
455         /* Now copy the result back to the user buffer.  */
456         if (count >= SZ_SG_HEADER) {
457                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
458                         retval = -EFAULT;
459                         goto free_old_hdr;
460                 }
461                 buf += SZ_SG_HEADER;
462                 if (count > old_hdr->reply_len)
463                         count = old_hdr->reply_len;
464                 if (count > SZ_SG_HEADER) {
465                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
466                                 retval = -EFAULT;
467                                 goto free_old_hdr;
468                         }
469                 }
470         } else
471                 count = (old_hdr->result == 0) ? 0 : -EIO;
472         sg_finish_rem_req(srp);
473         retval = count;
474 free_old_hdr:
475         kfree(old_hdr);
476         return retval;
477 }
478
479 static ssize_t
480 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
481 {
482         sg_io_hdr_t *hp = &srp->header;
483         int err = 0;
484         int len;
485
486         if (count < SZ_SG_IO_HDR) {
487                 err = -EINVAL;
488                 goto err_out;
489         }
490         hp->sb_len_wr = 0;
491         if ((hp->mx_sb_len > 0) && hp->sbp) {
492                 if ((CHECK_CONDITION & hp->masked_status) ||
493                     (DRIVER_SENSE & hp->driver_status)) {
494                         int sb_len = SCSI_SENSE_BUFFERSIZE;
495                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
496                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
497                         len = (len > sb_len) ? sb_len : len;
498                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
499                                 err = -EFAULT;
500                                 goto err_out;
501                         }
502                         hp->sb_len_wr = len;
503                 }
504         }
505         if (hp->masked_status || hp->host_status || hp->driver_status)
506                 hp->info |= SG_INFO_CHECK;
507         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
508                 err = -EFAULT;
509                 goto err_out;
510         }
511         err = sg_read_xfer(srp);
512       err_out:
513         sg_finish_rem_req(srp);
514         return (0 == err) ? count : err;
515 }
516
517 static ssize_t
518 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
519 {
520         int mxsize, cmd_size, k;
521         int input_size, blocking;
522         unsigned char opcode;
523         Sg_device *sdp;
524         Sg_fd *sfp;
525         Sg_request *srp;
526         struct sg_header old_hdr;
527         sg_io_hdr_t *hp;
528         unsigned char cmnd[MAX_COMMAND_SIZE];
529
530         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
531                 return -ENXIO;
532         SCSI_LOG_TIMEOUT(3, printk("sg_write: %s, count=%d\n",
533                                    sdp->disk->disk_name, (int) count));
534         if (sdp->detached)
535                 return -ENODEV;
536         if (!((filp->f_flags & O_NONBLOCK) ||
537               scsi_block_when_processing_errors(sdp->device)))
538                 return -ENXIO;
539
540         if (!access_ok(VERIFY_READ, buf, count))
541                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
542         if (count < SZ_SG_HEADER)
543                 return -EIO;
544         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
545                 return -EFAULT;
546         blocking = !(filp->f_flags & O_NONBLOCK);
547         if (old_hdr.reply_len < 0)
548                 return sg_new_write(sfp, buf, count, blocking, 0, NULL);
549         if (count < (SZ_SG_HEADER + 6))
550                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
551
552         if (!(srp = sg_add_request(sfp))) {
553                 SCSI_LOG_TIMEOUT(1, printk("sg_write: queue full\n"));
554                 return -EDOM;
555         }
556         buf += SZ_SG_HEADER;
557         __get_user(opcode, buf);
558         if (sfp->next_cmd_len > 0) {
559                 if (sfp->next_cmd_len > MAX_COMMAND_SIZE) {
560                         SCSI_LOG_TIMEOUT(1, printk("sg_write: command length too long\n"));
561                         sfp->next_cmd_len = 0;
562                         sg_remove_request(sfp, srp);
563                         return -EIO;
564                 }
565                 cmd_size = sfp->next_cmd_len;
566                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
567         } else {
568                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
569                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
570                         cmd_size = 12;
571         }
572         SCSI_LOG_TIMEOUT(4, printk(
573                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
574 /* Determine buffer size.  */
575         input_size = count - cmd_size;
576         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
577         mxsize -= SZ_SG_HEADER;
578         input_size -= SZ_SG_HEADER;
579         if (input_size < 0) {
580                 sg_remove_request(sfp, srp);
581                 return -EIO;    /* User did not pass enough bytes for this command. */
582         }
583         hp = &srp->header;
584         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
585         hp->cmd_len = (unsigned char) cmd_size;
586         hp->iovec_count = 0;
587         hp->mx_sb_len = 0;
588         if (input_size > 0)
589                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
590                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
591         else
592                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
593         hp->dxfer_len = mxsize;
594         hp->dxferp = (char __user *)buf + cmd_size;
595         hp->sbp = NULL;
596         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
597         hp->flags = input_size; /* structure abuse ... */
598         hp->pack_id = old_hdr.pack_id;
599         hp->usr_ptr = NULL;
600         if (__copy_from_user(cmnd, buf, cmd_size))
601                 return -EFAULT;
602         /*
603          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
604          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
605          * is a non-zero input_size, so emit a warning.
606          */
607         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV)
608                 if (printk_ratelimit())
609                         printk(KERN_WARNING
610                                "sg_write: data in/out %d/%d bytes for SCSI command 0x%x--"
611                                "guessing data in;\n" KERN_WARNING "   "
612                                "program %s not setting count and/or reply_len properly\n",
613                                old_hdr.reply_len - (int)SZ_SG_HEADER,
614                                input_size, (unsigned int) cmnd[0],
615                                current->comm);
616         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
617         return (k < 0) ? k : count;
618 }
619
620 static ssize_t
621 sg_new_write(Sg_fd * sfp, const char __user *buf, size_t count,
622              int blocking, int read_only, Sg_request ** o_srp)
623 {
624         int k;
625         Sg_request *srp;
626         sg_io_hdr_t *hp;
627         unsigned char cmnd[MAX_COMMAND_SIZE];
628         int timeout;
629         unsigned long ul_timeout;
630
631         if (count < SZ_SG_IO_HDR)
632                 return -EINVAL;
633         if (!access_ok(VERIFY_READ, buf, count))
634                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
635
636         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
637         if (!(srp = sg_add_request(sfp))) {
638                 SCSI_LOG_TIMEOUT(1, printk("sg_new_write: queue full\n"));
639                 return -EDOM;
640         }
641         hp = &srp->header;
642         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
643                 sg_remove_request(sfp, srp);
644                 return -EFAULT;
645         }
646         if (hp->interface_id != 'S') {
647                 sg_remove_request(sfp, srp);
648                 return -ENOSYS;
649         }
650         if (hp->flags & SG_FLAG_MMAP_IO) {
651                 if (hp->dxfer_len > sfp->reserve.bufflen) {
652                         sg_remove_request(sfp, srp);
653                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
654                 }
655                 if (hp->flags & SG_FLAG_DIRECT_IO) {
656                         sg_remove_request(sfp, srp);
657                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
658                 }
659                 if (sg_res_in_use(sfp)) {
660                         sg_remove_request(sfp, srp);
661                         return -EBUSY;  /* reserve buffer already being used */
662                 }
663         }
664         ul_timeout = msecs_to_jiffies(srp->header.timeout);
665         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
666         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
667                 sg_remove_request(sfp, srp);
668                 return -EMSGSIZE;
669         }
670         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
671                 sg_remove_request(sfp, srp);
672                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
673         }
674         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
675                 sg_remove_request(sfp, srp);
676                 return -EFAULT;
677         }
678         if (read_only &&
679             (!sg_allow_access(cmnd[0], sfp->parentdp->device->type))) {
680                 sg_remove_request(sfp, srp);
681                 return -EPERM;
682         }
683         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
684         if (k < 0)
685                 return k;
686         if (o_srp)
687                 *o_srp = srp;
688         return count;
689 }
690
691 static int
692 sg_common_write(Sg_fd * sfp, Sg_request * srp,
693                 unsigned char *cmnd, int timeout, int blocking)
694 {
695         int k, data_dir;
696         Sg_device *sdp = sfp->parentdp;
697         sg_io_hdr_t *hp = &srp->header;
698
699         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
700         hp->status = 0;
701         hp->masked_status = 0;
702         hp->msg_status = 0;
703         hp->info = 0;
704         hp->host_status = 0;
705         hp->driver_status = 0;
706         hp->resid = 0;
707         SCSI_LOG_TIMEOUT(4, printk("sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
708                           (int) cmnd[0], (int) hp->cmd_len));
709
710         if ((k = sg_start_req(srp))) {
711                 SCSI_LOG_TIMEOUT(1, printk("sg_write: start_req err=%d\n", k));
712                 sg_finish_rem_req(srp);
713                 return k;       /* probably out of space --> ENOMEM */
714         }
715         if ((k = sg_write_xfer(srp))) {
716                 SCSI_LOG_TIMEOUT(1, printk("sg_write: write_xfer, bad address\n"));
717                 sg_finish_rem_req(srp);
718                 return k;
719         }
720         if (sdp->detached) {
721                 sg_finish_rem_req(srp);
722                 return -ENODEV;
723         }
724
725         switch (hp->dxfer_direction) {
726         case SG_DXFER_TO_FROM_DEV:
727         case SG_DXFER_FROM_DEV:
728                 data_dir = DMA_FROM_DEVICE;
729                 break;
730         case SG_DXFER_TO_DEV:
731                 data_dir = DMA_TO_DEVICE;
732                 break;
733         case SG_DXFER_UNKNOWN:
734                 data_dir = DMA_BIDIRECTIONAL;
735                 break;
736         default:
737                 data_dir = DMA_NONE;
738                 break;
739         }
740         hp->duration = jiffies_to_msecs(jiffies);
741 /* Now send everything of to mid-level. The next time we hear about this
742    packet is when sg_cmd_done() is called (i.e. a callback). */
743         if (scsi_execute_async(sdp->device, cmnd, hp->cmd_len, data_dir, srp->data.buffer,
744                                 hp->dxfer_len, srp->data.k_use_sg, timeout,
745                                 SG_DEFAULT_RETRIES, srp, sg_cmd_done,
746                                 GFP_ATOMIC)) {
747                 SCSI_LOG_TIMEOUT(1, printk("sg_write: scsi_execute_async failed\n"));
748                 /*
749                  * most likely out of mem, but could also be a bad map
750                  */
751                 sg_finish_rem_req(srp);
752                 return -ENOMEM;
753         } else
754                 return 0;
755 }
756
757 static int
758 sg_srp_done(Sg_request *srp, Sg_fd *sfp)
759 {
760         unsigned long iflags;
761         int done;
762
763         read_lock_irqsave(&sfp->rq_list_lock, iflags);
764         done = srp->done;
765         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
766         return done;
767 }
768
769 static int
770 sg_ioctl(struct inode *inode, struct file *filp,
771          unsigned int cmd_in, unsigned long arg)
772 {
773         void __user *p = (void __user *)arg;
774         int __user *ip = p;
775         int result, val, read_only;
776         Sg_device *sdp;
777         Sg_fd *sfp;
778         Sg_request *srp;
779         unsigned long iflags;
780
781         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
782                 return -ENXIO;
783         SCSI_LOG_TIMEOUT(3, printk("sg_ioctl: %s, cmd=0x%x\n",
784                                    sdp->disk->disk_name, (int) cmd_in));
785         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
786
787         switch (cmd_in) {
788         case SG_IO:
789                 {
790                         int blocking = 1;       /* ignore O_NONBLOCK flag */
791
792                         if (sdp->detached)
793                                 return -ENODEV;
794                         if (!scsi_block_when_processing_errors(sdp->device))
795                                 return -ENXIO;
796                         if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
797                                 return -EFAULT;
798                         result =
799                             sg_new_write(sfp, p, SZ_SG_IO_HDR,
800                                          blocking, read_only, &srp);
801                         if (result < 0)
802                                 return result;
803                         srp->sg_io_owned = 1;
804                         while (1) {
805                                 result = 0;     /* following macro to beat race condition */
806                                 __wait_event_interruptible(sfp->read_wait,
807                                         (sdp->detached || sfp->closed || sg_srp_done(srp, sfp)),
808                                                            result);
809                                 if (sdp->detached)
810                                         return -ENODEV;
811                                 if (sfp->closed)
812                                         return 0;       /* request packet dropped already */
813                                 if (0 == result)
814                                         break;
815                                 srp->orphan = 1;
816                                 return result;  /* -ERESTARTSYS because signal hit process */
817                         }
818                         write_lock_irqsave(&sfp->rq_list_lock, iflags);
819                         srp->done = 2;
820                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
821                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
822                         return (result < 0) ? result : 0;
823                 }
824         case SG_SET_TIMEOUT:
825                 result = get_user(val, ip);
826                 if (result)
827                         return result;
828                 if (val < 0)
829                         return -EIO;
830                 if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
831                     val = MULDIV (INT_MAX, USER_HZ, HZ);
832                 sfp->timeout_user = val;
833                 sfp->timeout = MULDIV (val, HZ, USER_HZ);
834
835                 return 0;
836         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
837                                 /* strange ..., for backward compatibility */
838                 return sfp->timeout_user;
839         case SG_SET_FORCE_LOW_DMA:
840                 result = get_user(val, ip);
841                 if (result)
842                         return result;
843                 if (val) {
844                         sfp->low_dma = 1;
845                         if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
846                                 val = (int) sfp->reserve.bufflen;
847                                 sg_remove_scat(&sfp->reserve);
848                                 sg_build_reserve(sfp, val);
849                         }
850                 } else {
851                         if (sdp->detached)
852                                 return -ENODEV;
853                         sfp->low_dma = sdp->device->host->unchecked_isa_dma;
854                 }
855                 return 0;
856         case SG_GET_LOW_DMA:
857                 return put_user((int) sfp->low_dma, ip);
858         case SG_GET_SCSI_ID:
859                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
860                         return -EFAULT;
861                 else {
862                         sg_scsi_id_t __user *sg_idp = p;
863
864                         if (sdp->detached)
865                                 return -ENODEV;
866                         __put_user((int) sdp->device->host->host_no,
867                                    &sg_idp->host_no);
868                         __put_user((int) sdp->device->channel,
869                                    &sg_idp->channel);
870                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
871                         __put_user((int) sdp->device->lun, &sg_idp->lun);
872                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
873                         __put_user((short) sdp->device->host->cmd_per_lun,
874                                    &sg_idp->h_cmd_per_lun);
875                         __put_user((short) sdp->device->queue_depth,
876                                    &sg_idp->d_queue_depth);
877                         __put_user(0, &sg_idp->unused[0]);
878                         __put_user(0, &sg_idp->unused[1]);
879                         return 0;
880                 }
881         case SG_SET_FORCE_PACK_ID:
882                 result = get_user(val, ip);
883                 if (result)
884                         return result;
885                 sfp->force_packid = val ? 1 : 0;
886                 return 0;
887         case SG_GET_PACK_ID:
888                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
889                         return -EFAULT;
890                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
891                 for (srp = sfp->headrp; srp; srp = srp->nextrp) {
892                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
893                                 read_unlock_irqrestore(&sfp->rq_list_lock,
894                                                        iflags);
895                                 __put_user(srp->header.pack_id, ip);
896                                 return 0;
897                         }
898                 }
899                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
900                 __put_user(-1, ip);
901                 return 0;
902         case SG_GET_NUM_WAITING:
903                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
904                 for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
905                         if ((1 == srp->done) && (!srp->sg_io_owned))
906                                 ++val;
907                 }
908                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
909                 return put_user(val, ip);
910         case SG_GET_SG_TABLESIZE:
911                 return put_user(sdp->sg_tablesize, ip);
912         case SG_SET_RESERVED_SIZE:
913                 result = get_user(val, ip);
914                 if (result)
915                         return result;
916                 if (val < 0)
917                         return -EINVAL;
918                 if (val != sfp->reserve.bufflen) {
919                         if (sg_res_in_use(sfp) || sfp->mmap_called)
920                                 return -EBUSY;
921                         sg_remove_scat(&sfp->reserve);
922                         sg_build_reserve(sfp, val);
923                 }
924                 return 0;
925         case SG_GET_RESERVED_SIZE:
926                 val = (int) sfp->reserve.bufflen;
927                 return put_user(val, ip);
928         case SG_SET_COMMAND_Q:
929                 result = get_user(val, ip);
930                 if (result)
931                         return result;
932                 sfp->cmd_q = val ? 1 : 0;
933                 return 0;
934         case SG_GET_COMMAND_Q:
935                 return put_user((int) sfp->cmd_q, ip);
936         case SG_SET_KEEP_ORPHAN:
937                 result = get_user(val, ip);
938                 if (result)
939                         return result;
940                 sfp->keep_orphan = val;
941                 return 0;
942         case SG_GET_KEEP_ORPHAN:
943                 return put_user((int) sfp->keep_orphan, ip);
944         case SG_NEXT_CMD_LEN:
945                 result = get_user(val, ip);
946                 if (result)
947                         return result;
948                 sfp->next_cmd_len = (val > 0) ? val : 0;
949                 return 0;
950         case SG_GET_VERSION_NUM:
951                 return put_user(sg_version_num, ip);
952         case SG_GET_ACCESS_COUNT:
953                 /* faked - we don't have a real access count anymore */
954                 val = (sdp->device ? 1 : 0);
955                 return put_user(val, ip);
956         case SG_GET_REQUEST_TABLE:
957                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
958                         return -EFAULT;
959                 else {
960                         sg_req_info_t *rinfo;
961                         unsigned int ms;
962
963                         rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
964                                                                 GFP_KERNEL);
965                         if (!rinfo)
966                                 return -ENOMEM;
967                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
968                         for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
969                              ++val, srp = srp ? srp->nextrp : srp) {
970                                 memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
971                                 if (srp) {
972                                         rinfo[val].req_state = srp->done + 1;
973                                         rinfo[val].problem =
974                                             srp->header.masked_status & 
975                                             srp->header.host_status & 
976                                             srp->header.driver_status;
977                                         if (srp->done)
978                                                 rinfo[val].duration =
979                                                         srp->header.duration;
980                                         else {
981                                                 ms = jiffies_to_msecs(jiffies);
982                                                 rinfo[val].duration =
983                                                     (ms > srp->header.duration) ?
984                                                     (ms - srp->header.duration) : 0;
985                                         }
986                                         rinfo[val].orphan = srp->orphan;
987                                         rinfo[val].sg_io_owned =
988                                                         srp->sg_io_owned;
989                                         rinfo[val].pack_id =
990                                                         srp->header.pack_id;
991                                         rinfo[val].usr_ptr =
992                                                         srp->header.usr_ptr;
993                                 }
994                         }
995                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
996                         result = __copy_to_user(p, rinfo, 
997                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
998                         result = result ? -EFAULT : 0;
999                         kfree(rinfo);
1000                         return result;
1001                 }
1002         case SG_EMULATED_HOST:
1003                 if (sdp->detached)
1004                         return -ENODEV;
1005                 return put_user(sdp->device->host->hostt->emulated, ip);
1006         case SG_SCSI_RESET:
1007                 if (sdp->detached)
1008                         return -ENODEV;
1009                 if (filp->f_flags & O_NONBLOCK) {
1010                         if (scsi_host_in_recovery(sdp->device->host))
1011                                 return -EBUSY;
1012                 } else if (!scsi_block_when_processing_errors(sdp->device))
1013                         return -EBUSY;
1014                 result = get_user(val, ip);
1015                 if (result)
1016                         return result;
1017                 if (SG_SCSI_RESET_NOTHING == val)
1018                         return 0;
1019                 switch (val) {
1020                 case SG_SCSI_RESET_DEVICE:
1021                         val = SCSI_TRY_RESET_DEVICE;
1022                         break;
1023                 case SG_SCSI_RESET_BUS:
1024                         val = SCSI_TRY_RESET_BUS;
1025                         break;
1026                 case SG_SCSI_RESET_HOST:
1027                         val = SCSI_TRY_RESET_HOST;
1028                         break;
1029                 default:
1030                         return -EINVAL;
1031                 }
1032                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1033                         return -EACCES;
1034                 return (scsi_reset_provider(sdp->device, val) ==
1035                         SUCCESS) ? 0 : -EIO;
1036         case SCSI_IOCTL_SEND_COMMAND:
1037                 if (sdp->detached)
1038                         return -ENODEV;
1039                 if (read_only) {
1040                         unsigned char opcode = WRITE_6;
1041                         Scsi_Ioctl_Command __user *siocp = p;
1042
1043                         if (copy_from_user(&opcode, siocp->data, 1))
1044                                 return -EFAULT;
1045                         if (!sg_allow_access(opcode, sdp->device->type))
1046                                 return -EPERM;
1047                 }
1048                 return sg_scsi_ioctl(filp, sdp->device->request_queue, NULL, p);
1049         case SG_SET_DEBUG:
1050                 result = get_user(val, ip);
1051                 if (result)
1052                         return result;
1053                 sdp->sgdebug = (char) val;
1054                 return 0;
1055         case SCSI_IOCTL_GET_IDLUN:
1056         case SCSI_IOCTL_GET_BUS_NUMBER:
1057         case SCSI_IOCTL_PROBE_HOST:
1058         case SG_GET_TRANSFORM:
1059                 if (sdp->detached)
1060                         return -ENODEV;
1061                 return scsi_ioctl(sdp->device, cmd_in, p);
1062         default:
1063                 if (read_only)
1064                         return -EPERM;  /* don't know so take safe approach */
1065                 return scsi_ioctl(sdp->device, cmd_in, p);
1066         }
1067 }
1068
1069 #ifdef CONFIG_COMPAT
1070 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1071 {
1072         Sg_device *sdp;
1073         Sg_fd *sfp;
1074         struct scsi_device *sdev;
1075
1076         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1077                 return -ENXIO;
1078
1079         sdev = sdp->device;
1080         if (sdev->host->hostt->compat_ioctl) { 
1081                 int ret;
1082
1083                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1084
1085                 return ret;
1086         }
1087         
1088         return -ENOIOCTLCMD;
1089 }
1090 #endif
1091
1092 static unsigned int
1093 sg_poll(struct file *filp, poll_table * wait)
1094 {
1095         unsigned int res = 0;
1096         Sg_device *sdp;
1097         Sg_fd *sfp;
1098         Sg_request *srp;
1099         int count = 0;
1100         unsigned long iflags;
1101
1102         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp))
1103             || sfp->closed)
1104                 return POLLERR;
1105         poll_wait(filp, &sfp->read_wait, wait);
1106         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1107         for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1108                 /* if any read waiting, flag it */
1109                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1110                         res = POLLIN | POLLRDNORM;
1111                 ++count;
1112         }
1113         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1114
1115         if (sdp->detached)
1116                 res |= POLLHUP;
1117         else if (!sfp->cmd_q) {
1118                 if (0 == count)
1119                         res |= POLLOUT | POLLWRNORM;
1120         } else if (count < SG_MAX_QUEUE)
1121                 res |= POLLOUT | POLLWRNORM;
1122         SCSI_LOG_TIMEOUT(3, printk("sg_poll: %s, res=0x%x\n",
1123                                    sdp->disk->disk_name, (int) res));
1124         return res;
1125 }
1126
1127 static int
1128 sg_fasync(int fd, struct file *filp, int mode)
1129 {
1130         int retval;
1131         Sg_device *sdp;
1132         Sg_fd *sfp;
1133
1134         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1135                 return -ENXIO;
1136         SCSI_LOG_TIMEOUT(3, printk("sg_fasync: %s, mode=%d\n",
1137                                    sdp->disk->disk_name, mode));
1138
1139         retval = fasync_helper(fd, filp, mode, &sfp->async_qp);
1140         return (retval < 0) ? retval : 0;
1141 }
1142
1143 static struct page *
1144 sg_vma_nopage(struct vm_area_struct *vma, unsigned long addr, int *type)
1145 {
1146         Sg_fd *sfp;
1147         struct page *page = NOPAGE_SIGBUS;
1148         unsigned long offset, len, sa;
1149         Sg_scatter_hold *rsv_schp;
1150         struct scatterlist *sg;
1151         int k;
1152
1153         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1154                 return page;
1155         rsv_schp = &sfp->reserve;
1156         offset = addr - vma->vm_start;
1157         if (offset >= rsv_schp->bufflen)
1158                 return page;
1159         SCSI_LOG_TIMEOUT(3, printk("sg_vma_nopage: offset=%lu, scatg=%d\n",
1160                                    offset, rsv_schp->k_use_sg));
1161         sg = rsv_schp->buffer;
1162         sa = vma->vm_start;
1163         for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
1164              ++k, ++sg) {
1165                 len = vma->vm_end - sa;
1166                 len = (len < sg->length) ? len : sg->length;
1167                 if (offset < len) {
1168                         page = sg->page;
1169                         get_page(page); /* increment page count */
1170                         break;
1171                 }
1172                 sa += len;
1173                 offset -= len;
1174         }
1175
1176         if (type)
1177                 *type = VM_FAULT_MINOR;
1178         return page;
1179 }
1180
1181 static struct vm_operations_struct sg_mmap_vm_ops = {
1182         .nopage = sg_vma_nopage,
1183 };
1184
1185 static int
1186 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1187 {
1188         Sg_fd *sfp;
1189         unsigned long req_sz, len, sa;
1190         Sg_scatter_hold *rsv_schp;
1191         int k;
1192         struct scatterlist *sg;
1193
1194         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1195                 return -ENXIO;
1196         req_sz = vma->vm_end - vma->vm_start;
1197         SCSI_LOG_TIMEOUT(3, printk("sg_mmap starting, vm_start=%p, len=%d\n",
1198                                    (void *) vma->vm_start, (int) req_sz));
1199         if (vma->vm_pgoff)
1200                 return -EINVAL; /* want no offset */
1201         rsv_schp = &sfp->reserve;
1202         if (req_sz > rsv_schp->bufflen)
1203                 return -ENOMEM; /* cannot map more than reserved buffer */
1204
1205         sa = vma->vm_start;
1206         sg = rsv_schp->buffer;
1207         for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
1208              ++k, ++sg) {
1209                 len = vma->vm_end - sa;
1210                 len = (len < sg->length) ? len : sg->length;
1211                 sa += len;
1212         }
1213
1214         sfp->mmap_called = 1;
1215         vma->vm_flags |= VM_RESERVED;
1216         vma->vm_private_data = sfp;
1217         vma->vm_ops = &sg_mmap_vm_ops;
1218         return 0;
1219 }
1220
1221 /* This function is a "bottom half" handler that is called by the
1222  * mid level when a command is completed (or has failed). */
1223 static void
1224 sg_cmd_done(void *data, char *sense, int result, int resid)
1225 {
1226         Sg_request *srp = data;
1227         Sg_device *sdp = NULL;
1228         Sg_fd *sfp;
1229         unsigned long iflags;
1230         unsigned int ms;
1231
1232         if (NULL == srp) {
1233                 printk(KERN_ERR "sg_cmd_done: NULL request\n");
1234                 return;
1235         }
1236         sfp = srp->parentfp;
1237         if (sfp)
1238                 sdp = sfp->parentdp;
1239         if ((NULL == sdp) || sdp->detached) {
1240                 printk(KERN_INFO "sg_cmd_done: device detached\n");
1241                 return;
1242         }
1243
1244
1245         SCSI_LOG_TIMEOUT(4, printk("sg_cmd_done: %s, pack_id=%d, res=0x%x\n",
1246                 sdp->disk->disk_name, srp->header.pack_id, result));
1247         srp->header.resid = resid;
1248         ms = jiffies_to_msecs(jiffies);
1249         srp->header.duration = (ms > srp->header.duration) ?
1250                                 (ms - srp->header.duration) : 0;
1251         if (0 != result) {
1252                 struct scsi_sense_hdr sshdr;
1253
1254                 memcpy(srp->sense_b, sense, sizeof (srp->sense_b));
1255                 srp->header.status = 0xff & result;
1256                 srp->header.masked_status = status_byte(result);
1257                 srp->header.msg_status = msg_byte(result);
1258                 srp->header.host_status = host_byte(result);
1259                 srp->header.driver_status = driver_byte(result);
1260                 if ((sdp->sgdebug > 0) &&
1261                     ((CHECK_CONDITION == srp->header.masked_status) ||
1262                      (COMMAND_TERMINATED == srp->header.masked_status)))
1263                         __scsi_print_sense("sg_cmd_done", sense,
1264                                            SCSI_SENSE_BUFFERSIZE);
1265
1266                 /* Following if statement is a patch supplied by Eric Youngdale */
1267                 if (driver_byte(result) != 0
1268                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1269                     && !scsi_sense_is_deferred(&sshdr)
1270                     && sshdr.sense_key == UNIT_ATTENTION
1271                     && sdp->device->removable) {
1272                         /* Detected possible disc change. Set the bit - this */
1273                         /* may be used if there are filesystems using this device */
1274                         sdp->device->changed = 1;
1275                 }
1276         }
1277         /* Rely on write phase to clean out srp status values, so no "else" */
1278
1279         if (sfp->closed) {      /* whoops this fd already released, cleanup */
1280                 SCSI_LOG_TIMEOUT(1, printk("sg_cmd_done: already closed, freeing ...\n"));
1281                 sg_finish_rem_req(srp);
1282                 srp = NULL;
1283                 if (NULL == sfp->headrp) {
1284                         SCSI_LOG_TIMEOUT(1, printk("sg...bh: already closed, final cleanup\n"));
1285                         if (0 == sg_remove_sfp(sdp, sfp)) {     /* device still present */
1286                                 scsi_device_put(sdp->device);
1287                         }
1288                         sfp = NULL;
1289                 }
1290         } else if (srp && srp->orphan) {
1291                 if (sfp->keep_orphan)
1292                         srp->sg_io_owned = 0;
1293                 else {
1294                         sg_finish_rem_req(srp);
1295                         srp = NULL;
1296                 }
1297         }
1298         if (sfp && srp) {
1299                 /* Now wake up any sg_read() that is waiting for this packet. */
1300                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1301                 write_lock_irqsave(&sfp->rq_list_lock, iflags);
1302                 srp->done = 1;
1303                 wake_up_interruptible(&sfp->read_wait);
1304                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1305         }
1306 }
1307
1308 static struct file_operations sg_fops = {
1309         .owner = THIS_MODULE,
1310         .read = sg_read,
1311         .write = sg_write,
1312         .poll = sg_poll,
1313         .ioctl = sg_ioctl,
1314 #ifdef CONFIG_COMPAT
1315         .compat_ioctl = sg_compat_ioctl,
1316 #endif
1317         .open = sg_open,
1318         .mmap = sg_mmap,
1319         .release = sg_release,
1320         .fasync = sg_fasync,
1321 };
1322
1323 static struct class *sg_sysfs_class;
1324
1325 static int sg_sysfs_valid = 0;
1326
1327 static int sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1328 {
1329         struct request_queue *q = scsidp->request_queue;
1330         Sg_device *sdp;
1331         unsigned long iflags;
1332         void *old_sg_dev_arr = NULL;
1333         int k, error;
1334
1335         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1336         if (!sdp) {
1337                 printk(KERN_WARNING "kmalloc Sg_device failure\n");
1338                 return -ENOMEM;
1339         }
1340
1341         write_lock_irqsave(&sg_dev_arr_lock, iflags);
1342         if (unlikely(sg_nr_dev >= sg_dev_max)) {        /* try to resize */
1343                 Sg_device **tmp_da;
1344                 int tmp_dev_max = sg_nr_dev + SG_DEV_ARR_LUMP;
1345                 write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
1346
1347                 tmp_da = kzalloc(tmp_dev_max * sizeof(Sg_device *), GFP_KERNEL);
1348                 if (unlikely(!tmp_da))
1349                         goto expand_failed;
1350
1351                 write_lock_irqsave(&sg_dev_arr_lock, iflags);
1352                 memcpy(tmp_da, sg_dev_arr, sg_dev_max * sizeof(Sg_device *));
1353                 old_sg_dev_arr = sg_dev_arr;
1354                 sg_dev_arr = tmp_da;
1355                 sg_dev_max = tmp_dev_max;
1356         }
1357
1358         for (k = 0; k < sg_dev_max; k++)
1359                 if (!sg_dev_arr[k])
1360                         break;
1361         if (unlikely(k >= SG_MAX_DEVS))
1362                 goto overflow;
1363
1364         SCSI_LOG_TIMEOUT(3, printk("sg_alloc: dev=%d \n", k));
1365         sprintf(disk->disk_name, "sg%d", k);
1366         disk->first_minor = k;
1367         sdp->disk = disk;
1368         sdp->device = scsidp;
1369         init_waitqueue_head(&sdp->o_excl_wait);
1370         sdp->sg_tablesize = min(q->max_hw_segments, q->max_phys_segments);
1371
1372         sg_nr_dev++;
1373         sg_dev_arr[k] = sdp;
1374         write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
1375         error = k;
1376
1377  out:
1378         if (error < 0)
1379                 kfree(sdp);
1380         kfree(old_sg_dev_arr);
1381         return error;
1382
1383  expand_failed:
1384         printk(KERN_WARNING "sg_alloc: device array cannot be resized\n");
1385         error = -ENOMEM;
1386         goto out;
1387
1388  overflow:
1389         write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
1390         sdev_printk(KERN_WARNING, scsidp,
1391                     "Unable to attach sg device type=%d, minor "
1392                     "number exceeds %d\n", scsidp->type, SG_MAX_DEVS - 1);
1393         error = -ENODEV;
1394         goto out;
1395 }
1396
1397 static int
1398 sg_add(struct class_device *cl_dev, struct class_interface *cl_intf)
1399 {
1400         struct scsi_device *scsidp = to_scsi_device(cl_dev->dev);
1401         struct gendisk *disk;
1402         Sg_device *sdp = NULL;
1403         struct cdev * cdev = NULL;
1404         int error, k;
1405
1406         disk = alloc_disk(1);
1407         if (!disk) {
1408                 printk(KERN_WARNING "alloc_disk failed\n");
1409                 return -ENOMEM;
1410         }
1411         disk->major = SCSI_GENERIC_MAJOR;
1412
1413         error = -ENOMEM;
1414         cdev = cdev_alloc();
1415         if (!cdev) {
1416                 printk(KERN_WARNING "cdev_alloc failed\n");
1417                 goto out;
1418         }
1419         cdev->owner = THIS_MODULE;
1420         cdev->ops = &sg_fops;
1421
1422         error = sg_alloc(disk, scsidp);
1423         if (error < 0) {
1424                 printk(KERN_WARNING "sg_alloc failed\n");
1425                 goto out;
1426         }
1427         k = error;
1428         sdp = sg_dev_arr[k];
1429
1430         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, k), 1);
1431         if (error)
1432                 goto out;
1433
1434         sdp->cdev = cdev;
1435         if (sg_sysfs_valid) {
1436                 struct class_device * sg_class_member;
1437
1438                 sg_class_member = class_device_create(sg_sysfs_class, NULL,
1439                                 MKDEV(SCSI_GENERIC_MAJOR, k), 
1440                                 cl_dev->dev, "%s", 
1441                                 disk->disk_name);
1442                 if (IS_ERR(sg_class_member))
1443                         printk(KERN_WARNING "sg_add: "
1444                                 "class_device_create failed\n");
1445                 class_set_devdata(sg_class_member, sdp);
1446                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj, 
1447                                           &sg_class_member->kobj, "generic");
1448                 if (error)
1449                         printk(KERN_ERR "sg_add: unable to make symlink "
1450                                         "'generic' back to sg%d\n", k);
1451         } else
1452                 printk(KERN_WARNING "sg_add: sg_sys INvalid\n");
1453
1454         sdev_printk(KERN_NOTICE, scsidp,
1455                     "Attached scsi generic sg%d type %d\n", k,scsidp->type);
1456
1457         return 0;
1458
1459 out:
1460         put_disk(disk);
1461         if (cdev)
1462                 cdev_del(cdev);
1463         return error;
1464 }
1465
1466 static void
1467 sg_remove(struct class_device *cl_dev, struct class_interface *cl_intf)
1468 {
1469         struct scsi_device *scsidp = to_scsi_device(cl_dev->dev);
1470         Sg_device *sdp = NULL;
1471         unsigned long iflags;
1472         Sg_fd *sfp;
1473         Sg_fd *tsfp;
1474         Sg_request *srp;
1475         Sg_request *tsrp;
1476         int k, delay;
1477
1478         if (NULL == sg_dev_arr)
1479                 return;
1480         delay = 0;
1481         write_lock_irqsave(&sg_dev_arr_lock, iflags);
1482         for (k = 0; k < sg_dev_max; k++) {
1483                 sdp = sg_dev_arr[k];
1484                 if ((NULL == sdp) || (sdp->device != scsidp))
1485                         continue;       /* dirty but lowers nesting */
1486                 if (sdp->headfp) {
1487                         sdp->detached = 1;
1488                         for (sfp = sdp->headfp; sfp; sfp = tsfp) {
1489                                 tsfp = sfp->nextfp;
1490                                 for (srp = sfp->headrp; srp; srp = tsrp) {
1491                                         tsrp = srp->nextrp;
1492                                         if (sfp->closed || (0 == sg_srp_done(srp, sfp)))
1493                                                 sg_finish_rem_req(srp);
1494                                 }
1495                                 if (sfp->closed) {
1496                                         scsi_device_put(sdp->device);
1497                                         __sg_remove_sfp(sdp, sfp);
1498                                 } else {
1499                                         delay = 1;
1500                                         wake_up_interruptible(&sfp->read_wait);
1501                                         kill_fasync(&sfp->async_qp, SIGPOLL,
1502                                                     POLL_HUP);
1503                                 }
1504                         }
1505                         SCSI_LOG_TIMEOUT(3, printk("sg_detach: dev=%d, dirty\n", k));
1506                         if (NULL == sdp->headfp) {
1507                                 sg_dev_arr[k] = NULL;
1508                         }
1509                 } else {        /* nothing active, simple case */
1510                         SCSI_LOG_TIMEOUT(3, printk("sg_detach: dev=%d\n", k));
1511                         sg_dev_arr[k] = NULL;
1512                 }
1513                 sg_nr_dev--;
1514                 break;
1515         }
1516         write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
1517
1518         if (sdp) {
1519                 sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1520                 class_device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, k));
1521                 cdev_del(sdp->cdev);
1522                 sdp->cdev = NULL;
1523                 put_disk(sdp->disk);
1524                 sdp->disk = NULL;
1525                 if (NULL == sdp->headfp)
1526                         kfree((char *) sdp);
1527         }
1528
1529         if (delay)
1530                 msleep(10);     /* dirty detach so delay device destruction */
1531 }
1532
1533 /* Set 'perm' (4th argument) to 0 to disable module_param's definition
1534  * of sysfs parameters (which module_param doesn't yet support).
1535  * Sysfs parameters defined explicitly below.
1536  */
1537 module_param_named(def_reserved_size, def_reserved_size, int, S_IRUGO);
1538 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1539
1540 MODULE_AUTHOR("Douglas Gilbert");
1541 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1542 MODULE_LICENSE("GPL");
1543 MODULE_VERSION(SG_VERSION_STR);
1544 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1545
1546 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1547 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1548
1549 static int __init
1550 init_sg(void)
1551 {
1552         int rc;
1553
1554         if (def_reserved_size >= 0)
1555                 sg_big_buff = def_reserved_size;
1556
1557         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), 
1558                                     SG_MAX_DEVS, "sg");
1559         if (rc)
1560                 return rc;
1561         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1562         if ( IS_ERR(sg_sysfs_class) ) {
1563                 rc = PTR_ERR(sg_sysfs_class);
1564                 goto err_out;
1565         }
1566         sg_sysfs_valid = 1;
1567         rc = scsi_register_interface(&sg_interface);
1568         if (0 == rc) {
1569 #ifdef CONFIG_SCSI_PROC_FS
1570                 sg_proc_init();
1571 #endif                          /* CONFIG_SCSI_PROC_FS */
1572                 return 0;
1573         }
1574         class_destroy(sg_sysfs_class);
1575 err_out:
1576         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1577         return rc;
1578 }
1579
1580 static void __exit
1581 exit_sg(void)
1582 {
1583 #ifdef CONFIG_SCSI_PROC_FS
1584         sg_proc_cleanup();
1585 #endif                          /* CONFIG_SCSI_PROC_FS */
1586         scsi_unregister_interface(&sg_interface);
1587         class_destroy(sg_sysfs_class);
1588         sg_sysfs_valid = 0;
1589         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1590                                  SG_MAX_DEVS);
1591         kfree((char *)sg_dev_arr);
1592         sg_dev_arr = NULL;
1593         sg_dev_max = 0;
1594 }
1595
1596 static int
1597 sg_start_req(Sg_request * srp)
1598 {
1599         int res;
1600         Sg_fd *sfp = srp->parentfp;
1601         sg_io_hdr_t *hp = &srp->header;
1602         int dxfer_len = (int) hp->dxfer_len;
1603         int dxfer_dir = hp->dxfer_direction;
1604         Sg_scatter_hold *req_schp = &srp->data;
1605         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1606
1607         SCSI_LOG_TIMEOUT(4, printk("sg_start_req: dxfer_len=%d\n", dxfer_len));
1608         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1609                 return 0;
1610         if (sg_allow_dio && (hp->flags & SG_FLAG_DIRECT_IO) &&
1611             (dxfer_dir != SG_DXFER_UNKNOWN) && (0 == hp->iovec_count) &&
1612             (!sfp->parentdp->device->host->unchecked_isa_dma)) {
1613                 res = sg_build_direct(srp, sfp, dxfer_len);
1614                 if (res <= 0)   /* -ve -> error, 0 -> done, 1 -> try indirect */
1615                         return res;
1616         }
1617         if ((!sg_res_in_use(sfp)) && (dxfer_len <= rsv_schp->bufflen))
1618                 sg_link_reserve(sfp, srp, dxfer_len);
1619         else {
1620                 res = sg_build_indirect(req_schp, sfp, dxfer_len);
1621                 if (res) {
1622                         sg_remove_scat(req_schp);
1623                         return res;
1624                 }
1625         }
1626         return 0;
1627 }
1628
1629 static void
1630 sg_finish_rem_req(Sg_request * srp)
1631 {
1632         Sg_fd *sfp = srp->parentfp;
1633         Sg_scatter_hold *req_schp = &srp->data;
1634
1635         SCSI_LOG_TIMEOUT(4, printk("sg_finish_rem_req: res_used=%d\n", (int) srp->res_used));
1636         if (srp->res_used)
1637                 sg_unlink_reserve(sfp, srp);
1638         else
1639                 sg_remove_scat(req_schp);
1640         sg_remove_request(sfp, srp);
1641 }
1642
1643 static int
1644 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1645 {
1646         int sg_bufflen = tablesize * sizeof(struct scatterlist);
1647         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1648
1649         /*
1650          * TODO: test without low_dma, we should not need it since
1651          * the block layer will bounce the buffer for us
1652          *
1653          * XXX(hch): we shouldn't need GFP_DMA for the actual S/G list.
1654          */
1655         if (sfp->low_dma)
1656                  gfp_flags |= GFP_DMA;
1657         schp->buffer = kzalloc(sg_bufflen, gfp_flags);
1658         if (!schp->buffer)
1659                 return -ENOMEM;
1660         schp->sglist_len = sg_bufflen;
1661         return tablesize;       /* number of scat_gath elements allocated */
1662 }
1663
1664 #ifdef SG_ALLOW_DIO_CODE
1665 /* vvvvvvvv  following code borrowed from st driver's direct IO vvvvvvvvv */
1666         /* TODO: hopefully we can use the generic block layer code */
1667
1668 /* Pin down user pages and put them into a scatter gather list. Returns <= 0 if
1669    - mapping of all pages not successful
1670    (i.e., either completely successful or fails)
1671 */
1672 static int 
1673 st_map_user_pages(struct scatterlist *sgl, const unsigned int max_pages, 
1674                   unsigned long uaddr, size_t count, int rw)
1675 {
1676         unsigned long end = (uaddr + count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1677         unsigned long start = uaddr >> PAGE_SHIFT;
1678         const int nr_pages = end - start;
1679         int res, i, j;
1680         struct page **pages;
1681
1682         /* User attempted Overflow! */
1683         if ((uaddr + count) < uaddr)
1684                 return -EINVAL;
1685
1686         /* Too big */
1687         if (nr_pages > max_pages)
1688                 return -ENOMEM;
1689
1690         /* Hmm? */
1691         if (count == 0)
1692                 return 0;
1693
1694         if ((pages = kmalloc(max_pages * sizeof(*pages), GFP_ATOMIC)) == NULL)
1695                 return -ENOMEM;
1696
1697         /* Try to fault in all of the necessary pages */
1698         down_read(&current->mm->mmap_sem);
1699         /* rw==READ means read from drive, write into memory area */
1700         res = get_user_pages(
1701                 current,
1702                 current->mm,
1703                 uaddr,
1704                 nr_pages,
1705                 rw == READ,
1706                 0, /* don't force */
1707                 pages,
1708                 NULL);
1709         up_read(&current->mm->mmap_sem);
1710
1711         /* Errors and no page mapped should return here */
1712         if (res < nr_pages)
1713                 goto out_unmap;
1714
1715         for (i=0; i < nr_pages; i++) {
1716                 /* FIXME: flush superflous for rw==READ,
1717                  * probably wrong function for rw==WRITE
1718                  */
1719                 flush_dcache_page(pages[i]);
1720                 /* ?? Is locking needed? I don't think so */
1721                 /* if (TestSetPageLocked(pages[i]))
1722                    goto out_unlock; */
1723         }
1724
1725         sgl[0].page = pages[0];
1726         sgl[0].offset = uaddr & ~PAGE_MASK;
1727         if (nr_pages > 1) {
1728                 sgl[0].length = PAGE_SIZE - sgl[0].offset;
1729                 count -= sgl[0].length;
1730                 for (i=1; i < nr_pages ; i++) {
1731                         sgl[i].page = pages[i]; 
1732                         sgl[i].length = count < PAGE_SIZE ? count : PAGE_SIZE;
1733                         count -= PAGE_SIZE;
1734                 }
1735         }
1736         else {
1737                 sgl[0].length = count;
1738         }
1739
1740         kfree(pages);
1741         return nr_pages;
1742
1743  out_unmap:
1744         if (res > 0) {
1745                 for (j=0; j < res; j++)
1746                         page_cache_release(pages[j]);
1747                 res = 0;
1748         }
1749         kfree(pages);
1750         return res;
1751 }
1752
1753
1754 /* And unmap them... */
1755 static int 
1756 st_unmap_user_pages(struct scatterlist *sgl, const unsigned int nr_pages,
1757                     int dirtied)
1758 {
1759         int i;
1760
1761         for (i=0; i < nr_pages; i++) {
1762                 struct page *page = sgl[i].page;
1763
1764                 if (dirtied)
1765                         SetPageDirty(page);
1766                 /* unlock_page(page); */
1767                 /* FIXME: cache flush missing for rw==READ
1768                  * FIXME: call the correct reference counting function
1769                  */
1770                 page_cache_release(page);
1771         }
1772
1773         return 0;
1774 }
1775
1776 /* ^^^^^^^^  above code borrowed from st driver's direct IO ^^^^^^^^^ */
1777 #endif
1778
1779
1780 /* Returns: -ve -> error, 0 -> done, 1 -> try indirect */
1781 static int
1782 sg_build_direct(Sg_request * srp, Sg_fd * sfp, int dxfer_len)
1783 {
1784 #ifdef SG_ALLOW_DIO_CODE
1785         sg_io_hdr_t *hp = &srp->header;
1786         Sg_scatter_hold *schp = &srp->data;
1787         int sg_tablesize = sfp->parentdp->sg_tablesize;
1788         int mx_sc_elems, res;
1789         struct scsi_device *sdev = sfp->parentdp->device;
1790
1791         if (((unsigned long)hp->dxferp &
1792                         queue_dma_alignment(sdev->request_queue)) != 0)
1793                 return 1;
1794
1795         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1796         if (mx_sc_elems <= 0) {
1797                 return 1;
1798         }
1799         res = st_map_user_pages(schp->buffer, mx_sc_elems,
1800                                 (unsigned long)hp->dxferp, dxfer_len, 
1801                                 (SG_DXFER_TO_DEV == hp->dxfer_direction) ? 1 : 0);
1802         if (res <= 0) {
1803                 sg_remove_scat(schp);
1804                 return 1;
1805         }
1806         schp->k_use_sg = res;
1807         schp->dio_in_use = 1;
1808         hp->info |= SG_INFO_DIRECT_IO;
1809         return 0;
1810 #else
1811         return 1;
1812 #endif
1813 }
1814
1815 static int
1816 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1817 {
1818         struct scatterlist *sg;
1819         int ret_sz = 0, k, rem_sz, num, mx_sc_elems;
1820         int sg_tablesize = sfp->parentdp->sg_tablesize;
1821         int blk_size = buff_size;
1822         struct page *p = NULL;
1823
1824         if ((blk_size < 0) || (!sfp))
1825                 return -EFAULT;
1826         if (0 == blk_size)
1827                 ++blk_size;     /* don't know why */
1828 /* round request up to next highest SG_SECTOR_SZ byte boundary */
1829         blk_size = (blk_size + SG_SECTOR_MSK) & (~SG_SECTOR_MSK);
1830         SCSI_LOG_TIMEOUT(4, printk("sg_build_indirect: buff_size=%d, blk_size=%d\n",
1831                                    buff_size, blk_size));
1832
1833         /* N.B. ret_sz carried into this block ... */
1834         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1835         if (mx_sc_elems < 0)
1836                 return mx_sc_elems;     /* most likely -ENOMEM */
1837
1838         for (k = 0, sg = schp->buffer, rem_sz = blk_size;
1839              (rem_sz > 0) && (k < mx_sc_elems);
1840              ++k, rem_sz -= ret_sz, ++sg) {
1841                 
1842                 num = (rem_sz > SG_SCATTER_SZ) ? SG_SCATTER_SZ : rem_sz;
1843                 p = sg_page_malloc(num, sfp->low_dma, &ret_sz);
1844                 if (!p)
1845                         return -ENOMEM;
1846
1847                 sg->page = p;
1848                 sg->length = ret_sz;
1849
1850                 SCSI_LOG_TIMEOUT(5, printk("sg_build_build: k=%d, a=0x%p, len=%d\n",
1851                                   k, p, ret_sz));
1852         }               /* end of for loop */
1853
1854         schp->k_use_sg = k;
1855         SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k_use_sg=%d, rem_sz=%d\n", k, rem_sz));
1856
1857         schp->bufflen = blk_size;
1858         if (rem_sz > 0) /* must have failed */
1859                 return -ENOMEM;
1860
1861         return 0;
1862 }
1863
1864 static int
1865 sg_write_xfer(Sg_request * srp)
1866 {
1867         sg_io_hdr_t *hp = &srp->header;
1868         Sg_scatter_hold *schp = &srp->data;
1869         struct scatterlist *sg = schp->buffer;
1870         int num_xfer = 0;
1871         int j, k, onum, usglen, ksglen, res;
1872         int iovec_count = (int) hp->iovec_count;
1873         int dxfer_dir = hp->dxfer_direction;
1874         unsigned char *p;
1875         unsigned char __user *up;
1876         int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
1877
1878         if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_TO_DEV == dxfer_dir) ||
1879             (SG_DXFER_TO_FROM_DEV == dxfer_dir)) {
1880                 num_xfer = (int) (new_interface ? hp->dxfer_len : hp->flags);
1881                 if (schp->bufflen < num_xfer)
1882                         num_xfer = schp->bufflen;
1883         }
1884         if ((num_xfer <= 0) || (schp->dio_in_use) ||
1885             (new_interface
1886              && ((SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO) & hp->flags)))
1887                 return 0;
1888
1889         SCSI_LOG_TIMEOUT(4, printk("sg_write_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
1890                           num_xfer, iovec_count, schp->k_use_sg));
1891         if (iovec_count) {
1892                 onum = iovec_count;
1893                 if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
1894                         return -EFAULT;
1895         } else
1896                 onum = 1;
1897
1898         ksglen = sg->length;
1899         p = page_address(sg->page);
1900         for (j = 0, k = 0; j < onum; ++j) {
1901                 res = sg_u_iovec(hp, iovec_count, j, 1, &usglen, &up);
1902                 if (res)
1903                         return res;
1904
1905                 for (; p; ++sg, ksglen = sg->length,
1906                      p = page_address(sg->page)) {
1907                         if (usglen <= 0)
1908                                 break;
1909                         if (ksglen > usglen) {
1910                                 if (usglen >= num_xfer) {
1911                                         if (__copy_from_user(p, up, num_xfer))
1912                                                 return -EFAULT;
1913                                         return 0;
1914                                 }
1915                                 if (__copy_from_user(p, up, usglen))
1916                                         return -EFAULT;
1917                                 p += usglen;
1918                                 ksglen -= usglen;
1919                                 break;
1920                         } else {
1921                                 if (ksglen >= num_xfer) {
1922                                         if (__copy_from_user(p, up, num_xfer))
1923                                                 return -EFAULT;
1924                                         return 0;
1925                                 }
1926                                 if (__copy_from_user(p, up, ksglen))
1927                                         return -EFAULT;
1928                                 up += ksglen;
1929                                 usglen -= ksglen;
1930                         }
1931                         ++k;
1932                         if (k >= schp->k_use_sg)
1933                                 return 0;
1934                 }
1935         }
1936
1937         return 0;
1938 }
1939
1940 static int
1941 sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
1942            int wr_xf, int *countp, unsigned char __user **up)
1943 {
1944         int num_xfer = (int) hp->dxfer_len;
1945         unsigned char __user *p = hp->dxferp;
1946         int count;
1947
1948         if (0 == sg_num) {
1949                 if (wr_xf && ('\0' == hp->interface_id))
1950                         count = (int) hp->flags;        /* holds "old" input_size */
1951                 else
1952                         count = num_xfer;
1953         } else {
1954                 sg_iovec_t iovec;
1955                 if (__copy_from_user(&iovec, p + ind*SZ_SG_IOVEC, SZ_SG_IOVEC))
1956                         return -EFAULT;
1957                 p = iovec.iov_base;
1958                 count = (int) iovec.iov_len;
1959         }
1960         if (!access_ok(wr_xf ? VERIFY_READ : VERIFY_WRITE, p, count))
1961                 return -EFAULT;
1962         if (up)
1963                 *up = p;
1964         if (countp)
1965                 *countp = count;
1966         return 0;
1967 }
1968
1969 static void
1970 sg_remove_scat(Sg_scatter_hold * schp)
1971 {
1972         SCSI_LOG_TIMEOUT(4, printk("sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1973         if (schp->buffer && (schp->sglist_len > 0)) {
1974                 struct scatterlist *sg = schp->buffer;
1975
1976                 if (schp->dio_in_use) {
1977 #ifdef SG_ALLOW_DIO_CODE
1978                         st_unmap_user_pages(sg, schp->k_use_sg, TRUE);
1979 #endif
1980                 } else {
1981                         int k;
1982
1983                         for (k = 0; (k < schp->k_use_sg) && sg->page;
1984                              ++k, ++sg) {
1985                                 SCSI_LOG_TIMEOUT(5, printk(
1986                                     "sg_remove_scat: k=%d, a=0x%p, len=%d\n",
1987                                     k, sg->page, sg->length));
1988                                 sg_page_free(sg->page, sg->length);
1989                         }
1990                 }
1991                 kfree(schp->buffer);
1992         }
1993         memset(schp, 0, sizeof (*schp));
1994 }
1995
1996 static int
1997 sg_read_xfer(Sg_request * srp)
1998 {
1999         sg_io_hdr_t *hp = &srp->header;
2000         Sg_scatter_hold *schp = &srp->data;
2001         struct scatterlist *sg = schp->buffer;
2002         int num_xfer = 0;
2003         int j, k, onum, usglen, ksglen, res;
2004         int iovec_count = (int) hp->iovec_count;
2005         int dxfer_dir = hp->dxfer_direction;
2006         unsigned char *p;
2007         unsigned char __user *up;
2008         int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
2009
2010         if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_FROM_DEV == dxfer_dir)
2011             || (SG_DXFER_TO_FROM_DEV == dxfer_dir)) {
2012                 num_xfer = hp->dxfer_len;
2013                 if (schp->bufflen < num_xfer)
2014                         num_xfer = schp->bufflen;
2015         }
2016         if ((num_xfer <= 0) || (schp->dio_in_use) ||
2017             (new_interface
2018              && ((SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO) & hp->flags)))
2019                 return 0;
2020
2021         SCSI_LOG_TIMEOUT(4, printk("sg_read_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
2022                           num_xfer, iovec_count, schp->k_use_sg));
2023         if (iovec_count) {
2024                 onum = iovec_count;
2025                 if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
2026                         return -EFAULT;
2027         } else
2028                 onum = 1;
2029
2030         p = page_address(sg->page);
2031         ksglen = sg->length;
2032         for (j = 0, k = 0; j < onum; ++j) {
2033                 res = sg_u_iovec(hp, iovec_count, j, 0, &usglen, &up);
2034                 if (res)
2035                         return res;
2036
2037                 for (; p; ++sg, ksglen = sg->length,
2038                      p = page_address(sg->page)) {
2039                         if (usglen <= 0)
2040                                 break;
2041                         if (ksglen > usglen) {
2042                                 if (usglen >= num_xfer) {
2043                                         if (__copy_to_user(up, p, num_xfer))
2044                                                 return -EFAULT;
2045                                         return 0;
2046                                 }
2047                                 if (__copy_to_user(up, p, usglen))
2048                                         return -EFAULT;
2049                                 p += usglen;
2050                                 ksglen -= usglen;
2051                                 break;
2052                         } else {
2053                                 if (ksglen >= num_xfer) {
2054                                         if (__copy_to_user(up, p, num_xfer))
2055                                                 return -EFAULT;
2056                                         return 0;
2057                                 }
2058                                 if (__copy_to_user(up, p, ksglen))
2059                                         return -EFAULT;
2060                                 up += ksglen;
2061                                 usglen -= ksglen;
2062                         }
2063                         ++k;
2064                         if (k >= schp->k_use_sg)
2065                                 return 0;
2066                 }
2067         }
2068
2069         return 0;
2070 }
2071
2072 static int
2073 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2074 {
2075         Sg_scatter_hold *schp = &srp->data;
2076         struct scatterlist *sg = schp->buffer;
2077         int k, num;
2078
2079         SCSI_LOG_TIMEOUT(4, printk("sg_read_oxfer: num_read_xfer=%d\n",
2080                                    num_read_xfer));
2081         if ((!outp) || (num_read_xfer <= 0))
2082                 return 0;
2083
2084         for (k = 0; (k < schp->k_use_sg) && sg->page; ++k, ++sg) {
2085                 num = sg->length;
2086                 if (num > num_read_xfer) {
2087                         if (__copy_to_user(outp, page_address(sg->page),
2088                                            num_read_xfer))
2089                                 return -EFAULT;
2090                         break;
2091                 } else {
2092                         if (__copy_to_user(outp, page_address(sg->page),
2093                                            num))
2094                                 return -EFAULT;
2095                         num_read_xfer -= num;
2096                         if (num_read_xfer <= 0)
2097                                 break;
2098                         outp += num;
2099                 }
2100         }
2101
2102         return 0;
2103 }
2104
2105 static void
2106 sg_build_reserve(Sg_fd * sfp, int req_size)
2107 {
2108         Sg_scatter_hold *schp = &sfp->reserve;
2109
2110         SCSI_LOG_TIMEOUT(4, printk("sg_build_reserve: req_size=%d\n", req_size));
2111         do {
2112                 if (req_size < PAGE_SIZE)
2113                         req_size = PAGE_SIZE;
2114                 if (0 == sg_build_indirect(schp, sfp, req_size))
2115                         return;
2116                 else
2117                         sg_remove_scat(schp);
2118                 req_size >>= 1; /* divide by 2 */
2119         } while (req_size > (PAGE_SIZE / 2));
2120 }
2121
2122 static void
2123 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2124 {
2125         Sg_scatter_hold *req_schp = &srp->data;
2126         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2127         struct scatterlist *sg = rsv_schp->buffer;
2128         int k, num, rem;
2129
2130         srp->res_used = 1;
2131         SCSI_LOG_TIMEOUT(4, printk("sg_link_reserve: size=%d\n", size));
2132         rem = size;
2133
2134         for (k = 0; k < rsv_schp->k_use_sg; ++k, ++sg) {
2135                 num = sg->length;
2136                 if (rem <= num) {
2137                         sfp->save_scat_len = num;
2138                         sg->length = rem;
2139                         req_schp->k_use_sg = k + 1;
2140                         req_schp->sglist_len = rsv_schp->sglist_len;
2141                         req_schp->buffer = rsv_schp->buffer;
2142
2143                         req_schp->bufflen = size;
2144                         req_schp->b_malloc_len = rsv_schp->b_malloc_len;
2145                         break;
2146                 } else
2147                         rem -= num;
2148         }
2149
2150         if (k >= rsv_schp->k_use_sg)
2151                 SCSI_LOG_TIMEOUT(1, printk("sg_link_reserve: BAD size\n"));
2152 }
2153
2154 static void
2155 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2156 {
2157         Sg_scatter_hold *req_schp = &srp->data;
2158         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2159
2160         SCSI_LOG_TIMEOUT(4, printk("sg_unlink_reserve: req->k_use_sg=%d\n",
2161                                    (int) req_schp->k_use_sg));
2162         if ((rsv_schp->k_use_sg > 0) && (req_schp->k_use_sg > 0)) {
2163                 struct scatterlist *sg = rsv_schp->buffer;
2164
2165                 if (sfp->save_scat_len > 0)
2166                         (sg + (req_schp->k_use_sg - 1))->length =
2167                             (unsigned) sfp->save_scat_len;
2168                 else
2169                         SCSI_LOG_TIMEOUT(1, printk ("sg_unlink_reserve: BAD save_scat_len\n"));
2170         }
2171         req_schp->k_use_sg = 0;
2172         req_schp->bufflen = 0;
2173         req_schp->buffer = NULL;
2174         req_schp->sglist_len = 0;
2175         sfp->save_scat_len = 0;
2176         srp->res_used = 0;
2177 }
2178
2179 static Sg_request *
2180 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2181 {
2182         Sg_request *resp;
2183         unsigned long iflags;
2184
2185         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2186         for (resp = sfp->headrp; resp; resp = resp->nextrp) {
2187                 /* look for requests that are ready + not SG_IO owned */
2188                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2189                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2190                         resp->done = 2; /* guard against other readers */
2191                         break;
2192                 }
2193         }
2194         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2195         return resp;
2196 }
2197
2198 #ifdef CONFIG_SCSI_PROC_FS
2199 static Sg_request *
2200 sg_get_nth_request(Sg_fd * sfp, int nth)
2201 {
2202         Sg_request *resp;
2203         unsigned long iflags;
2204         int k;
2205
2206         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2207         for (k = 0, resp = sfp->headrp; resp && (k < nth);
2208              ++k, resp = resp->nextrp) ;
2209         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2210         return resp;
2211 }
2212 #endif
2213
2214 /* always adds to end of list */
2215 static Sg_request *
2216 sg_add_request(Sg_fd * sfp)
2217 {
2218         int k;
2219         unsigned long iflags;
2220         Sg_request *resp;
2221         Sg_request *rp = sfp->req_arr;
2222
2223         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2224         resp = sfp->headrp;
2225         if (!resp) {
2226                 memset(rp, 0, sizeof (Sg_request));
2227                 rp->parentfp = sfp;
2228                 resp = rp;
2229                 sfp->headrp = resp;
2230         } else {
2231                 if (0 == sfp->cmd_q)
2232                         resp = NULL;    /* command queuing disallowed */
2233                 else {
2234                         for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2235                                 if (!rp->parentfp)
2236                                         break;
2237                         }
2238                         if (k < SG_MAX_QUEUE) {
2239                                 memset(rp, 0, sizeof (Sg_request));
2240                                 rp->parentfp = sfp;
2241                                 while (resp->nextrp)
2242                                         resp = resp->nextrp;
2243                                 resp->nextrp = rp;
2244                                 resp = rp;
2245                         } else
2246                                 resp = NULL;
2247                 }
2248         }
2249         if (resp) {
2250                 resp->nextrp = NULL;
2251                 resp->header.duration = jiffies_to_msecs(jiffies);
2252         }
2253         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2254         return resp;
2255 }
2256
2257 /* Return of 1 for found; 0 for not found */
2258 static int
2259 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2260 {
2261         Sg_request *prev_rp;
2262         Sg_request *rp;
2263         unsigned long iflags;
2264         int res = 0;
2265
2266         if ((!sfp) || (!srp) || (!sfp->headrp))
2267                 return res;
2268         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2269         prev_rp = sfp->headrp;
2270         if (srp == prev_rp) {
2271                 sfp->headrp = prev_rp->nextrp;
2272                 prev_rp->parentfp = NULL;
2273                 res = 1;
2274         } else {
2275                 while ((rp = prev_rp->nextrp)) {
2276                         if (srp == rp) {
2277                                 prev_rp->nextrp = rp->nextrp;
2278                                 rp->parentfp = NULL;
2279                                 res = 1;
2280                                 break;
2281                         }
2282                         prev_rp = rp;
2283                 }
2284         }
2285         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2286         return res;
2287 }
2288
2289 #ifdef CONFIG_SCSI_PROC_FS
2290 static Sg_fd *
2291 sg_get_nth_sfp(Sg_device * sdp, int nth)
2292 {
2293         Sg_fd *resp;
2294         unsigned long iflags;
2295         int k;
2296
2297         read_lock_irqsave(&sg_dev_arr_lock, iflags);
2298         for (k = 0, resp = sdp->headfp; resp && (k < nth);
2299              ++k, resp = resp->nextfp) ;
2300         read_unlock_irqrestore(&sg_dev_arr_lock, iflags);
2301         return resp;
2302 }
2303 #endif
2304
2305 static Sg_fd *
2306 sg_add_sfp(Sg_device * sdp, int dev)
2307 {
2308         Sg_fd *sfp;
2309         unsigned long iflags;
2310
2311         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2312         if (!sfp)
2313                 return NULL;
2314
2315         init_waitqueue_head(&sfp->read_wait);
2316         rwlock_init(&sfp->rq_list_lock);
2317
2318         sfp->timeout = SG_DEFAULT_TIMEOUT;
2319         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2320         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2321         sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2322             sdp->device->host->unchecked_isa_dma : 1;
2323         sfp->cmd_q = SG_DEF_COMMAND_Q;
2324         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2325         sfp->parentdp = sdp;
2326         write_lock_irqsave(&sg_dev_arr_lock, iflags);
2327         if (!sdp->headfp)
2328                 sdp->headfp = sfp;
2329         else {                  /* add to tail of existing list */
2330                 Sg_fd *pfp = sdp->headfp;
2331                 while (pfp->nextfp)
2332                         pfp = pfp->nextfp;
2333                 pfp->nextfp = sfp;
2334         }
2335         write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
2336         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp: sfp=0x%p\n", sfp));
2337         sg_build_reserve(sfp, sg_big_buff);
2338         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp:   bufflen=%d, k_use_sg=%d\n",
2339                            sfp->reserve.bufflen, sfp->reserve.k_use_sg));
2340         return sfp;
2341 }
2342
2343 static void
2344 __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
2345 {
2346         Sg_fd *fp;
2347         Sg_fd *prev_fp;
2348
2349         prev_fp = sdp->headfp;
2350         if (sfp == prev_fp)
2351                 sdp->headfp = prev_fp->nextfp;
2352         else {
2353                 while ((fp = prev_fp->nextfp)) {
2354                         if (sfp == fp) {
2355                                 prev_fp->nextfp = fp->nextfp;
2356                                 break;
2357                         }
2358                         prev_fp = fp;
2359                 }
2360         }
2361         if (sfp->reserve.bufflen > 0) {
2362                 SCSI_LOG_TIMEOUT(6, 
2363                         printk("__sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2364                         (int) sfp->reserve.bufflen, (int) sfp->reserve.k_use_sg));
2365                 sg_remove_scat(&sfp->reserve);
2366         }
2367         sfp->parentdp = NULL;
2368         SCSI_LOG_TIMEOUT(6, printk("__sg_remove_sfp:    sfp=0x%p\n", sfp));
2369         kfree(sfp);
2370 }
2371
2372 /* Returns 0 in normal case, 1 when detached and sdp object removed */
2373 static int
2374 sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
2375 {
2376         Sg_request *srp;
2377         Sg_request *tsrp;
2378         int dirty = 0;
2379         int res = 0;
2380
2381         for (srp = sfp->headrp; srp; srp = tsrp) {
2382                 tsrp = srp->nextrp;
2383                 if (sg_srp_done(srp, sfp))
2384                         sg_finish_rem_req(srp);
2385                 else
2386                         ++dirty;
2387         }
2388         if (0 == dirty) {
2389                 unsigned long iflags;
2390
2391                 write_lock_irqsave(&sg_dev_arr_lock, iflags);
2392                 __sg_remove_sfp(sdp, sfp);
2393                 if (sdp->detached && (NULL == sdp->headfp)) {
2394                         int k, maxd;
2395
2396                         maxd = sg_dev_max;
2397                         for (k = 0; k < maxd; ++k) {
2398                                 if (sdp == sg_dev_arr[k])
2399                                         break;
2400                         }
2401                         if (k < maxd)
2402                                 sg_dev_arr[k] = NULL;
2403                         kfree((char *) sdp);
2404                         res = 1;
2405                 }
2406                 write_unlock_irqrestore(&sg_dev_arr_lock, iflags);
2407         } else {
2408                 /* MOD_INC's to inhibit unloading sg and associated adapter driver */
2409                 /* only bump the access_count if we actually succeeded in
2410                  * throwing another counter on the host module */
2411                 scsi_device_get(sdp->device);   /* XXX: retval ignored? */      
2412                 sfp->closed = 1;        /* flag dirty state on this fd */
2413                 SCSI_LOG_TIMEOUT(1, printk("sg_remove_sfp: worrisome, %d writes pending\n",
2414                                   dirty));
2415         }
2416         return res;
2417 }
2418
2419 static int
2420 sg_res_in_use(Sg_fd * sfp)
2421 {
2422         const Sg_request *srp;
2423         unsigned long iflags;
2424
2425         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2426         for (srp = sfp->headrp; srp; srp = srp->nextrp)
2427                 if (srp->res_used)
2428                         break;
2429         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2430         return srp ? 1 : 0;
2431 }
2432
2433 /* If retSzp==NULL want exact size or fail */
2434 static struct page *
2435 sg_page_malloc(int rqSz, int lowDma, int *retSzp)
2436 {
2437         struct page *resp = NULL;
2438         gfp_t page_mask;
2439         int order, a_size;
2440         int resSz = rqSz;
2441
2442         if (rqSz <= 0)
2443                 return resp;
2444
2445         if (lowDma)
2446                 page_mask = GFP_ATOMIC | GFP_DMA | __GFP_COMP | __GFP_NOWARN;
2447         else
2448                 page_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
2449
2450         for (order = 0, a_size = PAGE_SIZE; a_size < rqSz;
2451              order++, a_size <<= 1) ;
2452         resp = alloc_pages(page_mask, order);
2453         while ((!resp) && order && retSzp) {
2454                 --order;
2455                 a_size >>= 1;   /* divide by 2, until PAGE_SIZE */
2456                 resp =  alloc_pages(page_mask, order);  /* try half */
2457                 resSz = a_size;
2458         }
2459         if (resp) {
2460                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2461                         memset(page_address(resp), 0, resSz);
2462                 if (retSzp)
2463                         *retSzp = resSz;
2464         }
2465         return resp;
2466 }
2467
2468 static void
2469 sg_page_free(struct page *page, int size)
2470 {
2471         int order, a_size;
2472
2473         if (!page)
2474                 return;
2475         for (order = 0, a_size = PAGE_SIZE; a_size < size;
2476              order++, a_size <<= 1) ;
2477         __free_pages(page, order);
2478 }
2479
2480 #ifndef MAINTENANCE_IN_CMD
2481 #define MAINTENANCE_IN_CMD 0xa3
2482 #endif
2483
2484 static unsigned char allow_ops[] = { TEST_UNIT_READY, REQUEST_SENSE,
2485         INQUIRY, READ_CAPACITY, READ_BUFFER, READ_6, READ_10, READ_12,
2486         READ_16, MODE_SENSE, MODE_SENSE_10, LOG_SENSE, REPORT_LUNS,
2487         SERVICE_ACTION_IN, RECEIVE_DIAGNOSTIC, READ_LONG, MAINTENANCE_IN_CMD
2488 };
2489
2490 static int
2491 sg_allow_access(unsigned char opcode, char dev_type)
2492 {
2493         int k;
2494
2495         if (TYPE_SCANNER == dev_type)   /* TYPE_ROM maybe burner */
2496                 return 1;
2497         for (k = 0; k < sizeof (allow_ops); ++k) {
2498                 if (opcode == allow_ops[k])
2499                         return 1;
2500         }
2501         return 0;
2502 }
2503
2504 #ifdef CONFIG_SCSI_PROC_FS
2505 static int
2506 sg_last_dev(void)
2507 {
2508         int k;
2509         unsigned long iflags;
2510
2511         read_lock_irqsave(&sg_dev_arr_lock, iflags);
2512         for (k = sg_dev_max - 1; k >= 0; --k)
2513                 if (sg_dev_arr[k] && sg_dev_arr[k]->device)
2514                         break;
2515         read_unlock_irqrestore(&sg_dev_arr_lock, iflags);
2516         return k + 1;           /* origin 1 */
2517 }
2518 #endif
2519
2520 static Sg_device *
2521 sg_get_dev(int dev)
2522 {
2523         Sg_device *sdp = NULL;
2524         unsigned long iflags;
2525
2526         if (sg_dev_arr && (dev >= 0)) {
2527                 read_lock_irqsave(&sg_dev_arr_lock, iflags);
2528                 if (dev < sg_dev_max)
2529                         sdp = sg_dev_arr[dev];
2530                 read_unlock_irqrestore(&sg_dev_arr_lock, iflags);
2531         }
2532         return sdp;
2533 }
2534
2535 #ifdef CONFIG_SCSI_PROC_FS
2536
2537 static struct proc_dir_entry *sg_proc_sgp = NULL;
2538
2539 static char sg_proc_sg_dirname[] = "scsi/sg";
2540
2541 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2542
2543 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2544 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2545                                   size_t count, loff_t *off);
2546 static struct file_operations adio_fops = {
2547         /* .owner, .read and .llseek added in sg_proc_init() */
2548         .open = sg_proc_single_open_adio,
2549         .write = sg_proc_write_adio,
2550         .release = single_release,
2551 };
2552
2553 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2554 static ssize_t sg_proc_write_dressz(struct file *filp, 
2555                 const char __user *buffer, size_t count, loff_t *off);
2556 static struct file_operations dressz_fops = {
2557         .open = sg_proc_single_open_dressz,
2558         .write = sg_proc_write_dressz,
2559         .release = single_release,
2560 };
2561
2562 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2563 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2564 static struct file_operations version_fops = {
2565         .open = sg_proc_single_open_version,
2566         .release = single_release,
2567 };
2568
2569 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2570 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2571 static struct file_operations devhdr_fops = {
2572         .open = sg_proc_single_open_devhdr,
2573         .release = single_release,
2574 };
2575
2576 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2577 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2578 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2579 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2580 static void dev_seq_stop(struct seq_file *s, void *v);
2581 static struct file_operations dev_fops = {
2582         .open = sg_proc_open_dev,
2583         .release = seq_release,
2584 };
2585 static struct seq_operations dev_seq_ops = {
2586         .start = dev_seq_start,
2587         .next  = dev_seq_next,
2588         .stop  = dev_seq_stop,
2589         .show  = sg_proc_seq_show_dev,
2590 };
2591
2592 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2593 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2594 static struct file_operations devstrs_fops = {
2595         .open = sg_proc_open_devstrs,
2596         .release = seq_release,
2597 };
2598 static struct seq_operations devstrs_seq_ops = {
2599         .start = dev_seq_start,
2600         .next  = dev_seq_next,
2601         .stop  = dev_seq_stop,
2602         .show  = sg_proc_seq_show_devstrs,
2603 };
2604
2605 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2606 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2607 static struct file_operations debug_fops = {
2608         .open = sg_proc_open_debug,
2609         .release = seq_release,
2610 };
2611 static struct seq_operations debug_seq_ops = {
2612         .start = dev_seq_start,
2613         .next  = dev_seq_next,
2614         .stop  = dev_seq_stop,
2615         .show  = sg_proc_seq_show_debug,
2616 };
2617
2618
2619 struct sg_proc_leaf {
2620         const char * name;
2621         struct file_operations * fops;
2622 };
2623
2624 static struct sg_proc_leaf sg_proc_leaf_arr[] = {
2625         {"allow_dio", &adio_fops},
2626         {"debug", &debug_fops},
2627         {"def_reserved_size", &dressz_fops},
2628         {"device_hdr", &devhdr_fops},
2629         {"devices", &dev_fops},
2630         {"device_strs", &devstrs_fops},
2631         {"version", &version_fops}
2632 };
2633
2634 static int
2635 sg_proc_init(void)
2636 {
2637         int k, mask;
2638         int num_leaves =
2639             sizeof (sg_proc_leaf_arr) / sizeof (sg_proc_leaf_arr[0]);
2640         struct proc_dir_entry *pdep;
2641         struct sg_proc_leaf * leaf;
2642
2643         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2644         if (!sg_proc_sgp)
2645                 return 1;
2646         for (k = 0; k < num_leaves; ++k) {
2647                 leaf = &sg_proc_leaf_arr[k];
2648                 mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2649                 pdep = create_proc_entry(leaf->name, mask, sg_proc_sgp);
2650                 if (pdep) {
2651                         leaf->fops->owner = THIS_MODULE,
2652                         leaf->fops->read = seq_read,
2653                         leaf->fops->llseek = seq_lseek,
2654                         pdep->proc_fops = leaf->fops;
2655                 }
2656         }
2657         return 0;
2658 }
2659
2660 static void
2661 sg_proc_cleanup(void)
2662 {
2663         int k;
2664         int num_leaves =
2665             sizeof (sg_proc_leaf_arr) / sizeof (sg_proc_leaf_arr[0]);
2666
2667         if (!sg_proc_sgp)
2668                 return;
2669         for (k = 0; k < num_leaves; ++k)
2670                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2671         remove_proc_entry(sg_proc_sg_dirname, NULL);
2672 }
2673
2674
2675 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2676 {
2677         seq_printf(s, "%d\n", *((int *)s->private));
2678         return 0;
2679 }
2680
2681 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2682 {
2683         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2684 }
2685
2686 static ssize_t 
2687 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2688                    size_t count, loff_t *off)
2689 {
2690         int num;
2691         char buff[11];
2692
2693         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2694                 return -EACCES;
2695         num = (count < 10) ? count : 10;
2696         if (copy_from_user(buff, buffer, num))
2697                 return -EFAULT;
2698         buff[num] = '\0';
2699         sg_allow_dio = simple_strtoul(buff, NULL, 10) ? 1 : 0;
2700         return count;
2701 }
2702
2703 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2704 {
2705         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2706 }
2707
2708 static ssize_t 
2709 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2710                      size_t count, loff_t *off)
2711 {
2712         int num;
2713         unsigned long k = ULONG_MAX;
2714         char buff[11];
2715
2716         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2717                 return -EACCES;
2718         num = (count < 10) ? count : 10;
2719         if (copy_from_user(buff, buffer, num))
2720                 return -EFAULT;
2721         buff[num] = '\0';
2722         k = simple_strtoul(buff, NULL, 10);
2723         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2724                 sg_big_buff = k;
2725                 return count;
2726         }
2727         return -ERANGE;
2728 }
2729
2730 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2731 {
2732         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2733                    sg_version_date);
2734         return 0;
2735 }
2736
2737 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2738 {
2739         return single_open(file, sg_proc_seq_show_version, NULL);
2740 }
2741
2742 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2743 {
2744         seq_printf(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\t"
2745                    "online\n");
2746         return 0;
2747 }
2748
2749 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2750 {
2751         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2752 }
2753
2754 struct sg_proc_deviter {
2755         loff_t  index;
2756         size_t  max;
2757 };
2758
2759 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2760 {
2761         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2762
2763         s->private = it;
2764         if (! it)
2765                 return NULL;
2766
2767         if (NULL == sg_dev_arr)
2768                 return NULL;
2769         it->index = *pos;
2770         it->max = sg_last_dev();
2771         if (it->index >= it->max)
2772                 return NULL;
2773         return it;
2774 }
2775
2776 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2777 {
2778         struct sg_proc_deviter * it = s->private;
2779
2780         *pos = ++it->index;
2781         return (it->index < it->max) ? it : NULL;
2782 }
2783
2784 static void dev_seq_stop(struct seq_file *s, void *v)
2785 {
2786         kfree(s->private);
2787 }
2788
2789 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2790 {
2791         return seq_open(file, &dev_seq_ops);
2792 }
2793
2794 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2795 {
2796         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2797         Sg_device *sdp;
2798         struct scsi_device *scsidp;
2799
2800         sdp = it ? sg_get_dev(it->index) : NULL;
2801         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2802                 seq_printf(s, "%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\n",
2803                               scsidp->host->host_no, scsidp->channel,
2804                               scsidp->id, scsidp->lun, (int) scsidp->type,
2805                               1,
2806                               (int) scsidp->queue_depth,
2807                               (int) scsidp->device_busy,
2808                               (int) scsi_device_online(scsidp));
2809         else
2810                 seq_printf(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2811         return 0;
2812 }
2813
2814 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2815 {
2816         return seq_open(file, &devstrs_seq_ops);
2817 }
2818
2819 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2820 {
2821         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2822         Sg_device *sdp;
2823         struct scsi_device *scsidp;
2824
2825         sdp = it ? sg_get_dev(it->index) : NULL;
2826         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2827                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2828                            scsidp->vendor, scsidp->model, scsidp->rev);
2829         else
2830                 seq_printf(s, "<no active device>\n");
2831         return 0;
2832 }
2833
2834 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2835 {
2836         int k, m, new_interface, blen, usg;
2837         Sg_request *srp;
2838         Sg_fd *fp;
2839         const sg_io_hdr_t *hp;
2840         const char * cp;
2841         unsigned int ms;
2842
2843         for (k = 0; (fp = sg_get_nth_sfp(sdp, k)); ++k) {
2844                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2845                            "(res)sgat=%d low_dma=%d\n", k + 1,
2846                            jiffies_to_msecs(fp->timeout),
2847                            fp->reserve.bufflen,
2848                            (int) fp->reserve.k_use_sg,
2849                            (int) fp->low_dma);
2850                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=%d\n",
2851                            (int) fp->cmd_q, (int) fp->force_packid,
2852                            (int) fp->keep_orphan, (int) fp->closed);
2853                 for (m = 0; (srp = sg_get_nth_request(fp, m)); ++m) {
2854                         hp = &srp->header;
2855                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2856                         if (srp->res_used) {
2857                                 if (new_interface && 
2858                                     (SG_FLAG_MMAP_IO & hp->flags))
2859                                         cp = "     mmap>> ";
2860                                 else
2861                                         cp = "     rb>> ";
2862                         } else {
2863                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2864                                         cp = "     dio>> ";
2865                                 else
2866                                         cp = "     ";
2867                         }
2868                         seq_printf(s, cp);
2869                         blen = srp->data.bufflen;
2870                         usg = srp->data.k_use_sg;
2871                         seq_printf(s, srp->done ? 
2872                                    ((1 == srp->done) ?  "rcv:" : "fin:")
2873                                    : "act:");
2874                         seq_printf(s, " id=%d blen=%d",
2875                                    srp->header.pack_id, blen);
2876                         if (srp->done)
2877                                 seq_printf(s, " dur=%d", hp->duration);
2878                         else {
2879                                 ms = jiffies_to_msecs(jiffies);
2880                                 seq_printf(s, " t_o/elap=%d/%d",
2881                                         (new_interface ? hp->timeout :
2882                                                   jiffies_to_msecs(fp->timeout)),
2883                                         (ms > hp->duration ? ms - hp->duration : 0));
2884                         }
2885                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2886                                    (int) srp->data.cmd_opcode);
2887                 }
2888                 if (0 == m)
2889                         seq_printf(s, "     No requests active\n");
2890         }
2891 }
2892
2893 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2894 {
2895         return seq_open(file, &debug_seq_ops);
2896 }
2897
2898 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2899 {
2900         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2901         Sg_device *sdp;
2902
2903         if (it && (0 == it->index)) {
2904                 seq_printf(s, "dev_max(currently)=%d max_active_device=%d "
2905                            "(origin 1)\n", sg_dev_max, (int)it->max);
2906                 seq_printf(s, " def_reserved_size=%d\n", sg_big_buff);
2907         }
2908         sdp = it ? sg_get_dev(it->index) : NULL;
2909         if (sdp) {
2910                 struct scsi_device *scsidp = sdp->device;
2911
2912                 if (NULL == scsidp) {
2913                         seq_printf(s, "device %d detached ??\n", 
2914                                    (int)it->index);
2915                         return 0;
2916                 }
2917
2918                 if (sg_get_nth_sfp(sdp, 0)) {
2919                         seq_printf(s, " >>> device=%s ",
2920                                 sdp->disk->disk_name);
2921                         if (sdp->detached)
2922                                 seq_printf(s, "detached pending close ");
2923                         else
2924                                 seq_printf
2925                                     (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
2926                                      scsidp->host->host_no,
2927                                      scsidp->channel, scsidp->id,
2928                                      scsidp->lun,
2929                                      scsidp->host->hostt->emulated);
2930                         seq_printf(s, " sg_tablesize=%d excl=%d\n",
2931                                    sdp->sg_tablesize, sdp->exclude);
2932                 }
2933                 sg_proc_debug_helper(s, sdp);
2934         }
2935         return 0;
2936 }
2937
2938 #endif                          /* CONFIG_SCSI_PROC_FS */
2939
2940 module_init(init_sg);
2941 module_exit(exit_sg);