Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[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 = 30534;      /* 2 digits for each component */
22 #define SG_VERSION_STR "3.5.34"
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/module.h>
32
33 #include <linux/fs.h>
34 #include <linux/kernel.h>
35 #include <linux/sched.h>
36 #include <linux/string.h>
37 #include <linux/mm.h>
38 #include <linux/errno.h>
39 #include <linux/mtio.h>
40 #include <linux/ioctl.h>
41 #include <linux/fcntl.h>
42 #include <linux/init.h>
43 #include <linux/poll.h>
44 #include <linux/moduleparam.h>
45 #include <linux/cdev.h>
46 #include <linux/idr.h>
47 #include <linux/seq_file.h>
48 #include <linux/blkdev.h>
49 #include <linux/delay.h>
50 #include <linux/scatterlist.h>
51
52 #include "scsi.h"
53 #include <scsi/scsi_dbg.h>
54 #include <scsi/scsi_host.h>
55 #include <scsi/scsi_driver.h>
56 #include <scsi/scsi_ioctl.h>
57 #include <scsi/sg.h>
58
59 #include "scsi_logging.h"
60
61 #ifdef CONFIG_SCSI_PROC_FS
62 #include <linux/proc_fs.h>
63 static char *sg_version_date = "20061027";
64
65 static int sg_proc_init(void);
66 static void sg_proc_cleanup(void);
67 #endif
68
69 #define SG_ALLOW_DIO_DEF 0
70 #define SG_ALLOW_DIO_CODE /* compile out by commenting this define */
71
72 #define SG_MAX_DEVS 32768
73
74 /*
75  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
76  * Then when using 32 bit integers x * m may overflow during the calculation.
77  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
78  * calculates the same, but prevents the overflow when both m and d
79  * are "small" numbers (like HZ and USER_HZ).
80  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
81  * in 32 bits.
82  */
83 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
84
85 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
86
87 int sg_big_buff = SG_DEF_RESERVED_SIZE;
88 /* N.B. This variable is readable and writeable via
89    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
90    of this size (or less if there is not enough memory) will be reserved
91    for use by this file descriptor. [Deprecated usage: this variable is also
92    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
93    the kernel (i.e. it is not a module).] */
94 static int def_reserved_size = -1;      /* picks up init parameter */
95 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
96
97 static int scatter_elem_sz = SG_SCATTER_SZ;
98 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
99
100 #define SG_SECTOR_SZ 512
101 #define SG_SECTOR_MSK (SG_SECTOR_SZ - 1)
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_IDR(sg_index_idr);
107 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
108                                                            file descriptor list for device */
109
110 static struct class_interface sg_interface = {
111         .add            = sg_add,
112         .remove         = sg_remove,
113 };
114
115 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
116         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
117         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
118         unsigned bufflen;       /* Size of (aggregate) data buffer */
119         unsigned b_malloc_len;  /* actual len malloc'ed in buffer */
120         struct scatterlist *buffer;/* scatter list */
121         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
122         unsigned char cmd_opcode; /* first byte of command */
123 } Sg_scatter_hold;
124
125 struct sg_device;               /* forward declarations */
126 struct sg_fd;
127
128 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
129         struct sg_request *nextrp;      /* NULL -> tail request (slist) */
130         struct sg_fd *parentfp; /* NULL -> not in use */
131         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
132         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
133         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
134         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
135         char orphan;            /* 1 -> drop on sight, 0 -> normal */
136         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
137         volatile char done;     /* 0->before bh, 1->before read, 2->read */
138 } Sg_request;
139
140 typedef struct sg_fd {          /* holds the state of a file descriptor */
141         struct sg_fd *nextfp;   /* NULL when last opened fd on this device */
142         struct sg_device *parentdp;     /* owning device */
143         wait_queue_head_t read_wait;    /* queue read until command done */
144         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
145         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
146         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
147         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
148         unsigned save_scat_len; /* original length of trunc. scat. element */
149         Sg_request *headrp;     /* head of request slist, NULL->empty */
150         struct fasync_struct *async_qp; /* used by asynchronous notification */
151         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
152         char low_dma;           /* as in parent but possibly overridden to 1 */
153         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
154         volatile char closed;   /* 1 -> fd closed but request(s) outstanding */
155         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
156         char next_cmd_len;      /* 0 -> automatic (def), >0 -> use on next write() */
157         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
158         char mmap_called;       /* 0 -> mmap() never called on this fd */
159 } Sg_fd;
160
161 typedef struct sg_device { /* holds the state of each scsi generic device */
162         struct scsi_device *device;
163         wait_queue_head_t o_excl_wait;  /* queue open() when O_EXCL in use */
164         int sg_tablesize;       /* adapter's max scatter-gather table size */
165         u32 index;              /* device index number */
166         Sg_fd *headfp;          /* first open fd belonging to this device */
167         volatile char detached; /* 0->attached, 1->detached pending removal */
168         volatile char exclude;  /* opened for exclusive access */
169         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
170         struct gendisk *disk;
171         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
172 } Sg_device;
173
174 static int sg_fasync(int fd, struct file *filp, int mode);
175 /* tasklet or soft irq callback */
176 static void sg_cmd_done(void *data, char *sense, int result, int resid);
177 static int sg_start_req(Sg_request * srp);
178 static void sg_finish_rem_req(Sg_request * srp);
179 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
180 static int sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp,
181                          int tablesize);
182 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
183                            Sg_request * srp);
184 static ssize_t sg_new_write(Sg_fd * sfp, const char __user *buf, size_t count,
185                             int blocking, int read_only, Sg_request ** o_srp);
186 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
187                            unsigned char *cmnd, int timeout, int blocking);
188 static int sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
189                       int wr_xf, int *countp, unsigned char __user **up);
190 static int sg_write_xfer(Sg_request * srp);
191 static int sg_read_xfer(Sg_request * srp);
192 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
193 static void sg_remove_scat(Sg_scatter_hold * schp);
194 static void sg_build_reserve(Sg_fd * sfp, int req_size);
195 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
196 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
197 static struct page *sg_page_malloc(int rqSz, int lowDma, int *retSzp);
198 static void sg_page_free(struct page *page, int size);
199 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
200 static int sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
201 static void __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp);
202 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
203 static Sg_request *sg_add_request(Sg_fd * sfp);
204 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
205 static int sg_res_in_use(Sg_fd * sfp);
206 static int sg_allow_access(unsigned char opcode, char dev_type);
207 static int sg_build_direct(Sg_request * srp, Sg_fd * sfp, int dxfer_len);
208 static Sg_device *sg_get_dev(int dev);
209 #ifdef CONFIG_SCSI_PROC_FS
210 static int sg_last_dev(void);
211 #endif
212
213 #define SZ_SG_HEADER sizeof(struct sg_header)
214 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
215 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
216 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
217
218 static int
219 sg_open(struct inode *inode, struct file *filp)
220 {
221         int dev = iminor(inode);
222         int flags = filp->f_flags;
223         struct request_queue *q;
224         Sg_device *sdp;
225         Sg_fd *sfp;
226         int res;
227         int retval;
228
229         nonseekable_open(inode, filp);
230         SCSI_LOG_TIMEOUT(3, printk("sg_open: dev=%d, flags=0x%x\n", dev, flags));
231         sdp = sg_get_dev(dev);
232         if ((!sdp) || (!sdp->device))
233                 return -ENXIO;
234         if (sdp->detached)
235                 return -ENODEV;
236
237         /* This driver's module count bumped by fops_get in <linux/fs.h> */
238         /* Prevent the device driver from vanishing while we sleep */
239         retval = scsi_device_get(sdp->device);
240         if (retval)
241                 return retval;
242
243         if (!((flags & O_NONBLOCK) ||
244               scsi_block_when_processing_errors(sdp->device))) {
245                 retval = -ENXIO;
246                 /* we are in error recovery for this device */
247                 goto error_out;
248         }
249
250         if (flags & O_EXCL) {
251                 if (O_RDONLY == (flags & O_ACCMODE)) {
252                         retval = -EPERM; /* Can't lock it with read only access */
253                         goto error_out;
254                 }
255                 if (sdp->headfp && (flags & O_NONBLOCK)) {
256                         retval = -EBUSY;
257                         goto error_out;
258                 }
259                 res = 0;
260                 __wait_event_interruptible(sdp->o_excl_wait,
261                         ((sdp->headfp || sdp->exclude) ? 0 : (sdp->exclude = 1)), res);
262                 if (res) {
263                         retval = res;   /* -ERESTARTSYS because signal hit process */
264                         goto error_out;
265                 }
266         } else if (sdp->exclude) {      /* some other fd has an exclusive lock on dev */
267                 if (flags & O_NONBLOCK) {
268                         retval = -EBUSY;
269                         goto error_out;
270                 }
271                 res = 0;
272                 __wait_event_interruptible(sdp->o_excl_wait, (!sdp->exclude),
273                                            res);
274                 if (res) {
275                         retval = res;   /* -ERESTARTSYS because signal hit process */
276                         goto error_out;
277                 }
278         }
279         if (sdp->detached) {
280                 retval = -ENODEV;
281                 goto error_out;
282         }
283         if (!sdp->headfp) {     /* no existing opens on this device */
284                 sdp->sgdebug = 0;
285                 q = sdp->device->request_queue;
286                 sdp->sg_tablesize = min(q->max_hw_segments,
287                                         q->max_phys_segments);
288         }
289         if ((sfp = sg_add_sfp(sdp, dev)))
290                 filp->private_data = sfp;
291         else {
292                 if (flags & O_EXCL)
293                         sdp->exclude = 0;       /* undo if error */
294                 retval = -ENOMEM;
295                 goto error_out;
296         }
297         return 0;
298
299       error_out:
300         scsi_device_put(sdp->device);
301         return retval;
302 }
303
304 /* Following function was formerly called 'sg_close' */
305 static int
306 sg_release(struct inode *inode, struct file *filp)
307 {
308         Sg_device *sdp;
309         Sg_fd *sfp;
310
311         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
312                 return -ENXIO;
313         SCSI_LOG_TIMEOUT(3, printk("sg_release: %s\n", sdp->disk->disk_name));
314         sg_fasync(-1, filp, 0); /* remove filp from async notification list */
315         if (0 == sg_remove_sfp(sdp, sfp)) {     /* Returns 1 when sdp gone */
316                 if (!sdp->detached) {
317                         scsi_device_put(sdp->device);
318                 }
319                 sdp->exclude = 0;
320                 wake_up_interruptible(&sdp->o_excl_wait);
321         }
322         return 0;
323 }
324
325 static ssize_t
326 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
327 {
328         Sg_device *sdp;
329         Sg_fd *sfp;
330         Sg_request *srp;
331         int req_pack_id = -1;
332         sg_io_hdr_t *hp;
333         struct sg_header *old_hdr = NULL;
334         int retval = 0;
335
336         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
337                 return -ENXIO;
338         SCSI_LOG_TIMEOUT(3, printk("sg_read: %s, count=%d\n",
339                                    sdp->disk->disk_name, (int) count));
340
341         if (!access_ok(VERIFY_WRITE, buf, count))
342                 return -EFAULT;
343         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
344                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
345                 if (!old_hdr)
346                         return -ENOMEM;
347                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
348                         retval = -EFAULT;
349                         goto free_old_hdr;
350                 }
351                 if (old_hdr->reply_len < 0) {
352                         if (count >= SZ_SG_IO_HDR) {
353                                 sg_io_hdr_t *new_hdr;
354                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
355                                 if (!new_hdr) {
356                                         retval = -ENOMEM;
357                                         goto free_old_hdr;
358                                 }
359                                 retval =__copy_from_user
360                                     (new_hdr, buf, SZ_SG_IO_HDR);
361                                 req_pack_id = new_hdr->pack_id;
362                                 kfree(new_hdr);
363                                 if (retval) {
364                                         retval = -EFAULT;
365                                         goto free_old_hdr;
366                                 }
367                         }
368                 } else
369                         req_pack_id = old_hdr->pack_id;
370         }
371         srp = sg_get_rq_mark(sfp, req_pack_id);
372         if (!srp) {             /* now wait on packet to arrive */
373                 if (sdp->detached) {
374                         retval = -ENODEV;
375                         goto free_old_hdr;
376                 }
377                 if (filp->f_flags & O_NONBLOCK) {
378                         retval = -EAGAIN;
379                         goto free_old_hdr;
380                 }
381                 while (1) {
382                         retval = 0; /* following macro beats race condition */
383                         __wait_event_interruptible(sfp->read_wait,
384                                 (sdp->detached ||
385                                 (srp = sg_get_rq_mark(sfp, req_pack_id))), 
386                                 retval);
387                         if (sdp->detached) {
388                                 retval = -ENODEV;
389                                 goto free_old_hdr;
390                         }
391                         if (0 == retval)
392                                 break;
393
394                         /* -ERESTARTSYS as signal hit process */
395                         goto free_old_hdr;
396                 }
397         }
398         if (srp->header.interface_id != '\0') {
399                 retval = sg_new_read(sfp, buf, count, srp);
400                 goto free_old_hdr;
401         }
402
403         hp = &srp->header;
404         if (old_hdr == NULL) {
405                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
406                 if (! old_hdr) {
407                         retval = -ENOMEM;
408                         goto free_old_hdr;
409                 }
410         }
411         memset(old_hdr, 0, SZ_SG_HEADER);
412         old_hdr->reply_len = (int) hp->timeout;
413         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
414         old_hdr->pack_id = hp->pack_id;
415         old_hdr->twelve_byte =
416             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
417         old_hdr->target_status = hp->masked_status;
418         old_hdr->host_status = hp->host_status;
419         old_hdr->driver_status = hp->driver_status;
420         if ((CHECK_CONDITION & hp->masked_status) ||
421             (DRIVER_SENSE & hp->driver_status))
422                 memcpy(old_hdr->sense_buffer, srp->sense_b,
423                        sizeof (old_hdr->sense_buffer));
424         switch (hp->host_status) {
425         /* This setup of 'result' is for backward compatibility and is best
426            ignored by the user who should use target, host + driver status */
427         case DID_OK:
428         case DID_PASSTHROUGH:
429         case DID_SOFT_ERROR:
430                 old_hdr->result = 0;
431                 break;
432         case DID_NO_CONNECT:
433         case DID_BUS_BUSY:
434         case DID_TIME_OUT:
435                 old_hdr->result = EBUSY;
436                 break;
437         case DID_BAD_TARGET:
438         case DID_ABORT:
439         case DID_PARITY:
440         case DID_RESET:
441         case DID_BAD_INTR:
442                 old_hdr->result = EIO;
443                 break;
444         case DID_ERROR:
445                 old_hdr->result = (srp->sense_b[0] == 0 && 
446                                   hp->masked_status == GOOD) ? 0 : EIO;
447                 break;
448         default:
449                 old_hdr->result = EIO;
450                 break;
451         }
452
453         /* Now copy the result back to the user buffer.  */
454         if (count >= SZ_SG_HEADER) {
455                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
456                         retval = -EFAULT;
457                         goto free_old_hdr;
458                 }
459                 buf += SZ_SG_HEADER;
460                 if (count > old_hdr->reply_len)
461                         count = old_hdr->reply_len;
462                 if (count > SZ_SG_HEADER) {
463                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
464                                 retval = -EFAULT;
465                                 goto free_old_hdr;
466                         }
467                 }
468         } else
469                 count = (old_hdr->result == 0) ? 0 : -EIO;
470         sg_finish_rem_req(srp);
471         retval = count;
472 free_old_hdr:
473         kfree(old_hdr);
474         return retval;
475 }
476
477 static ssize_t
478 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
479 {
480         sg_io_hdr_t *hp = &srp->header;
481         int err = 0;
482         int len;
483
484         if (count < SZ_SG_IO_HDR) {
485                 err = -EINVAL;
486                 goto err_out;
487         }
488         hp->sb_len_wr = 0;
489         if ((hp->mx_sb_len > 0) && hp->sbp) {
490                 if ((CHECK_CONDITION & hp->masked_status) ||
491                     (DRIVER_SENSE & hp->driver_status)) {
492                         int sb_len = SCSI_SENSE_BUFFERSIZE;
493                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
494                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
495                         len = (len > sb_len) ? sb_len : len;
496                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
497                                 err = -EFAULT;
498                                 goto err_out;
499                         }
500                         hp->sb_len_wr = len;
501                 }
502         }
503         if (hp->masked_status || hp->host_status || hp->driver_status)
504                 hp->info |= SG_INFO_CHECK;
505         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
506                 err = -EFAULT;
507                 goto err_out;
508         }
509         err = sg_read_xfer(srp);
510       err_out:
511         sg_finish_rem_req(srp);
512         return (0 == err) ? count : err;
513 }
514
515 static ssize_t
516 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
517 {
518         int mxsize, cmd_size, k;
519         int input_size, blocking;
520         unsigned char opcode;
521         Sg_device *sdp;
522         Sg_fd *sfp;
523         Sg_request *srp;
524         struct sg_header old_hdr;
525         sg_io_hdr_t *hp;
526         unsigned char cmnd[MAX_COMMAND_SIZE];
527
528         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
529                 return -ENXIO;
530         SCSI_LOG_TIMEOUT(3, printk("sg_write: %s, count=%d\n",
531                                    sdp->disk->disk_name, (int) count));
532         if (sdp->detached)
533                 return -ENODEV;
534         if (!((filp->f_flags & O_NONBLOCK) ||
535               scsi_block_when_processing_errors(sdp->device)))
536                 return -ENXIO;
537
538         if (!access_ok(VERIFY_READ, buf, count))
539                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
540         if (count < SZ_SG_HEADER)
541                 return -EIO;
542         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
543                 return -EFAULT;
544         blocking = !(filp->f_flags & O_NONBLOCK);
545         if (old_hdr.reply_len < 0)
546                 return sg_new_write(sfp, buf, count, blocking, 0, NULL);
547         if (count < (SZ_SG_HEADER + 6))
548                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
549
550         if (!(srp = sg_add_request(sfp))) {
551                 SCSI_LOG_TIMEOUT(1, printk("sg_write: queue full\n"));
552                 return -EDOM;
553         }
554         buf += SZ_SG_HEADER;
555         __get_user(opcode, buf);
556         if (sfp->next_cmd_len > 0) {
557                 if (sfp->next_cmd_len > MAX_COMMAND_SIZE) {
558                         SCSI_LOG_TIMEOUT(1, printk("sg_write: command length too long\n"));
559                         sfp->next_cmd_len = 0;
560                         sg_remove_request(sfp, srp);
561                         return -EIO;
562                 }
563                 cmd_size = sfp->next_cmd_len;
564                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
565         } else {
566                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
567                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
568                         cmd_size = 12;
569         }
570         SCSI_LOG_TIMEOUT(4, printk(
571                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
572 /* Determine buffer size.  */
573         input_size = count - cmd_size;
574         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
575         mxsize -= SZ_SG_HEADER;
576         input_size -= SZ_SG_HEADER;
577         if (input_size < 0) {
578                 sg_remove_request(sfp, srp);
579                 return -EIO;    /* User did not pass enough bytes for this command. */
580         }
581         hp = &srp->header;
582         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
583         hp->cmd_len = (unsigned char) cmd_size;
584         hp->iovec_count = 0;
585         hp->mx_sb_len = 0;
586         if (input_size > 0)
587                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
588                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
589         else
590                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
591         hp->dxfer_len = mxsize;
592         hp->dxferp = (char __user *)buf + cmd_size;
593         hp->sbp = NULL;
594         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
595         hp->flags = input_size; /* structure abuse ... */
596         hp->pack_id = old_hdr.pack_id;
597         hp->usr_ptr = NULL;
598         if (__copy_from_user(cmnd, buf, cmd_size))
599                 return -EFAULT;
600         /*
601          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
602          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
603          * is a non-zero input_size, so emit a warning.
604          */
605         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV)
606                 if (printk_ratelimit())
607                         printk(KERN_WARNING
608                                "sg_write: data in/out %d/%d bytes for SCSI command 0x%x--"
609                                "guessing data in;\n" KERN_WARNING "   "
610                                "program %s not setting count and/or reply_len properly\n",
611                                old_hdr.reply_len - (int)SZ_SG_HEADER,
612                                input_size, (unsigned int) cmnd[0],
613                                current->comm);
614         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
615         return (k < 0) ? k : count;
616 }
617
618 static ssize_t
619 sg_new_write(Sg_fd * sfp, const char __user *buf, size_t count,
620              int blocking, int read_only, Sg_request ** o_srp)
621 {
622         int k;
623         Sg_request *srp;
624         sg_io_hdr_t *hp;
625         unsigned char cmnd[MAX_COMMAND_SIZE];
626         int timeout;
627         unsigned long ul_timeout;
628
629         if (count < SZ_SG_IO_HDR)
630                 return -EINVAL;
631         if (!access_ok(VERIFY_READ, buf, count))
632                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
633
634         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
635         if (!(srp = sg_add_request(sfp))) {
636                 SCSI_LOG_TIMEOUT(1, printk("sg_new_write: queue full\n"));
637                 return -EDOM;
638         }
639         hp = &srp->header;
640         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
641                 sg_remove_request(sfp, srp);
642                 return -EFAULT;
643         }
644         if (hp->interface_id != 'S') {
645                 sg_remove_request(sfp, srp);
646                 return -ENOSYS;
647         }
648         if (hp->flags & SG_FLAG_MMAP_IO) {
649                 if (hp->dxfer_len > sfp->reserve.bufflen) {
650                         sg_remove_request(sfp, srp);
651                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
652                 }
653                 if (hp->flags & SG_FLAG_DIRECT_IO) {
654                         sg_remove_request(sfp, srp);
655                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
656                 }
657                 if (sg_res_in_use(sfp)) {
658                         sg_remove_request(sfp, srp);
659                         return -EBUSY;  /* reserve buffer already being used */
660                 }
661         }
662         ul_timeout = msecs_to_jiffies(srp->header.timeout);
663         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
664         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
665                 sg_remove_request(sfp, srp);
666                 return -EMSGSIZE;
667         }
668         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
669                 sg_remove_request(sfp, srp);
670                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
671         }
672         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
673                 sg_remove_request(sfp, srp);
674                 return -EFAULT;
675         }
676         if (read_only &&
677             (!sg_allow_access(cmnd[0], sfp->parentdp->device->type))) {
678                 sg_remove_request(sfp, srp);
679                 return -EPERM;
680         }
681         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
682         if (k < 0)
683                 return k;
684         if (o_srp)
685                 *o_srp = srp;
686         return count;
687 }
688
689 static int
690 sg_common_write(Sg_fd * sfp, Sg_request * srp,
691                 unsigned char *cmnd, int timeout, int blocking)
692 {
693         int k, data_dir;
694         Sg_device *sdp = sfp->parentdp;
695         sg_io_hdr_t *hp = &srp->header;
696
697         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
698         hp->status = 0;
699         hp->masked_status = 0;
700         hp->msg_status = 0;
701         hp->info = 0;
702         hp->host_status = 0;
703         hp->driver_status = 0;
704         hp->resid = 0;
705         SCSI_LOG_TIMEOUT(4, printk("sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
706                           (int) cmnd[0], (int) hp->cmd_len));
707
708         if ((k = sg_start_req(srp))) {
709                 SCSI_LOG_TIMEOUT(1, printk("sg_common_write: start_req err=%d\n", k));
710                 sg_finish_rem_req(srp);
711                 return k;       /* probably out of space --> ENOMEM */
712         }
713         if ((k = sg_write_xfer(srp))) {
714                 SCSI_LOG_TIMEOUT(1, printk("sg_common_write: write_xfer, bad address\n"));
715                 sg_finish_rem_req(srp);
716                 return k;
717         }
718         if (sdp->detached) {
719                 sg_finish_rem_req(srp);
720                 return -ENODEV;
721         }
722
723         switch (hp->dxfer_direction) {
724         case SG_DXFER_TO_FROM_DEV:
725         case SG_DXFER_FROM_DEV:
726                 data_dir = DMA_FROM_DEVICE;
727                 break;
728         case SG_DXFER_TO_DEV:
729                 data_dir = DMA_TO_DEVICE;
730                 break;
731         case SG_DXFER_UNKNOWN:
732                 data_dir = DMA_BIDIRECTIONAL;
733                 break;
734         default:
735                 data_dir = DMA_NONE;
736                 break;
737         }
738         hp->duration = jiffies_to_msecs(jiffies);
739 /* Now send everything of to mid-level. The next time we hear about this
740    packet is when sg_cmd_done() is called (i.e. a callback). */
741         if (scsi_execute_async(sdp->device, cmnd, hp->cmd_len, data_dir, srp->data.buffer,
742                                 hp->dxfer_len, srp->data.k_use_sg, timeout,
743                                 SG_DEFAULT_RETRIES, srp, sg_cmd_done,
744                                 GFP_ATOMIC)) {
745                 SCSI_LOG_TIMEOUT(1, printk("sg_common_write: scsi_execute_async failed\n"));
746                 /*
747                  * most likely out of mem, but could also be a bad map
748                  */
749                 sg_finish_rem_req(srp);
750                 return -ENOMEM;
751         } else
752                 return 0;
753 }
754
755 static int
756 sg_srp_done(Sg_request *srp, Sg_fd *sfp)
757 {
758         unsigned long iflags;
759         int done;
760
761         read_lock_irqsave(&sfp->rq_list_lock, iflags);
762         done = srp->done;
763         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
764         return done;
765 }
766
767 static int
768 sg_ioctl(struct inode *inode, struct file *filp,
769          unsigned int cmd_in, unsigned long arg)
770 {
771         void __user *p = (void __user *)arg;
772         int __user *ip = p;
773         int result, val, read_only;
774         Sg_device *sdp;
775         Sg_fd *sfp;
776         Sg_request *srp;
777         unsigned long iflags;
778
779         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
780                 return -ENXIO;
781         SCSI_LOG_TIMEOUT(3, printk("sg_ioctl: %s, cmd=0x%x\n",
782                                    sdp->disk->disk_name, (int) cmd_in));
783         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
784
785         switch (cmd_in) {
786         case SG_IO:
787                 {
788                         int blocking = 1;       /* ignore O_NONBLOCK flag */
789
790                         if (sdp->detached)
791                                 return -ENODEV;
792                         if (!scsi_block_when_processing_errors(sdp->device))
793                                 return -ENXIO;
794                         if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
795                                 return -EFAULT;
796                         result =
797                             sg_new_write(sfp, p, SZ_SG_IO_HDR,
798                                          blocking, read_only, &srp);
799                         if (result < 0)
800                                 return result;
801                         srp->sg_io_owned = 1;
802                         while (1) {
803                                 result = 0;     /* following macro to beat race condition */
804                                 __wait_event_interruptible(sfp->read_wait,
805                                         (sdp->detached || sfp->closed || sg_srp_done(srp, sfp)),
806                                                            result);
807                                 if (sdp->detached)
808                                         return -ENODEV;
809                                 if (sfp->closed)
810                                         return 0;       /* request packet dropped already */
811                                 if (0 == result)
812                                         break;
813                                 srp->orphan = 1;
814                                 return result;  /* -ERESTARTSYS because signal hit process */
815                         }
816                         write_lock_irqsave(&sfp->rq_list_lock, iflags);
817                         srp->done = 2;
818                         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
819                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
820                         return (result < 0) ? result : 0;
821                 }
822         case SG_SET_TIMEOUT:
823                 result = get_user(val, ip);
824                 if (result)
825                         return result;
826                 if (val < 0)
827                         return -EIO;
828                 if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
829                     val = MULDIV (INT_MAX, USER_HZ, HZ);
830                 sfp->timeout_user = val;
831                 sfp->timeout = MULDIV (val, HZ, USER_HZ);
832
833                 return 0;
834         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
835                                 /* strange ..., for backward compatibility */
836                 return sfp->timeout_user;
837         case SG_SET_FORCE_LOW_DMA:
838                 result = get_user(val, ip);
839                 if (result)
840                         return result;
841                 if (val) {
842                         sfp->low_dma = 1;
843                         if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
844                                 val = (int) sfp->reserve.bufflen;
845                                 sg_remove_scat(&sfp->reserve);
846                                 sg_build_reserve(sfp, val);
847                         }
848                 } else {
849                         if (sdp->detached)
850                                 return -ENODEV;
851                         sfp->low_dma = sdp->device->host->unchecked_isa_dma;
852                 }
853                 return 0;
854         case SG_GET_LOW_DMA:
855                 return put_user((int) sfp->low_dma, ip);
856         case SG_GET_SCSI_ID:
857                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
858                         return -EFAULT;
859                 else {
860                         sg_scsi_id_t __user *sg_idp = p;
861
862                         if (sdp->detached)
863                                 return -ENODEV;
864                         __put_user((int) sdp->device->host->host_no,
865                                    &sg_idp->host_no);
866                         __put_user((int) sdp->device->channel,
867                                    &sg_idp->channel);
868                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
869                         __put_user((int) sdp->device->lun, &sg_idp->lun);
870                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
871                         __put_user((short) sdp->device->host->cmd_per_lun,
872                                    &sg_idp->h_cmd_per_lun);
873                         __put_user((short) sdp->device->queue_depth,
874                                    &sg_idp->d_queue_depth);
875                         __put_user(0, &sg_idp->unused[0]);
876                         __put_user(0, &sg_idp->unused[1]);
877                         return 0;
878                 }
879         case SG_SET_FORCE_PACK_ID:
880                 result = get_user(val, ip);
881                 if (result)
882                         return result;
883                 sfp->force_packid = val ? 1 : 0;
884                 return 0;
885         case SG_GET_PACK_ID:
886                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
887                         return -EFAULT;
888                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
889                 for (srp = sfp->headrp; srp; srp = srp->nextrp) {
890                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
891                                 read_unlock_irqrestore(&sfp->rq_list_lock,
892                                                        iflags);
893                                 __put_user(srp->header.pack_id, ip);
894                                 return 0;
895                         }
896                 }
897                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
898                 __put_user(-1, ip);
899                 return 0;
900         case SG_GET_NUM_WAITING:
901                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
902                 for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
903                         if ((1 == srp->done) && (!srp->sg_io_owned))
904                                 ++val;
905                 }
906                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
907                 return put_user(val, ip);
908         case SG_GET_SG_TABLESIZE:
909                 return put_user(sdp->sg_tablesize, ip);
910         case SG_SET_RESERVED_SIZE:
911                 result = get_user(val, ip);
912                 if (result)
913                         return result;
914                 if (val < 0)
915                         return -EINVAL;
916                 val = min_t(int, val,
917                                 sdp->device->request_queue->max_sectors * 512);
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 = min_t(int, sfp->reserve.bufflen,
927                                 sdp->device->request_queue->max_sectors * 512);
928                 return put_user(val, ip);
929         case SG_SET_COMMAND_Q:
930                 result = get_user(val, ip);
931                 if (result)
932                         return result;
933                 sfp->cmd_q = val ? 1 : 0;
934                 return 0;
935         case SG_GET_COMMAND_Q:
936                 return put_user((int) sfp->cmd_q, ip);
937         case SG_SET_KEEP_ORPHAN:
938                 result = get_user(val, ip);
939                 if (result)
940                         return result;
941                 sfp->keep_orphan = val;
942                 return 0;
943         case SG_GET_KEEP_ORPHAN:
944                 return put_user((int) sfp->keep_orphan, ip);
945         case SG_NEXT_CMD_LEN:
946                 result = get_user(val, ip);
947                 if (result)
948                         return result;
949                 sfp->next_cmd_len = (val > 0) ? val : 0;
950                 return 0;
951         case SG_GET_VERSION_NUM:
952                 return put_user(sg_version_num, ip);
953         case SG_GET_ACCESS_COUNT:
954                 /* faked - we don't have a real access count anymore */
955                 val = (sdp->device ? 1 : 0);
956                 return put_user(val, ip);
957         case SG_GET_REQUEST_TABLE:
958                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
959                         return -EFAULT;
960                 else {
961                         sg_req_info_t *rinfo;
962                         unsigned int ms;
963
964                         rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
965                                                                 GFP_KERNEL);
966                         if (!rinfo)
967                                 return -ENOMEM;
968                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
969                         for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
970                              ++val, srp = srp ? srp->nextrp : srp) {
971                                 memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
972                                 if (srp) {
973                                         rinfo[val].req_state = srp->done + 1;
974                                         rinfo[val].problem =
975                                             srp->header.masked_status & 
976                                             srp->header.host_status & 
977                                             srp->header.driver_status;
978                                         if (srp->done)
979                                                 rinfo[val].duration =
980                                                         srp->header.duration;
981                                         else {
982                                                 ms = jiffies_to_msecs(jiffies);
983                                                 rinfo[val].duration =
984                                                     (ms > srp->header.duration) ?
985                                                     (ms - srp->header.duration) : 0;
986                                         }
987                                         rinfo[val].orphan = srp->orphan;
988                                         rinfo[val].sg_io_owned =
989                                                         srp->sg_io_owned;
990                                         rinfo[val].pack_id =
991                                                         srp->header.pack_id;
992                                         rinfo[val].usr_ptr =
993                                                         srp->header.usr_ptr;
994                                 }
995                         }
996                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
997                         result = __copy_to_user(p, rinfo, 
998                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
999                         result = result ? -EFAULT : 0;
1000                         kfree(rinfo);
1001                         return result;
1002                 }
1003         case SG_EMULATED_HOST:
1004                 if (sdp->detached)
1005                         return -ENODEV;
1006                 return put_user(sdp->device->host->hostt->emulated, ip);
1007         case SG_SCSI_RESET:
1008                 if (sdp->detached)
1009                         return -ENODEV;
1010                 if (filp->f_flags & O_NONBLOCK) {
1011                         if (scsi_host_in_recovery(sdp->device->host))
1012                                 return -EBUSY;
1013                 } else if (!scsi_block_when_processing_errors(sdp->device))
1014                         return -EBUSY;
1015                 result = get_user(val, ip);
1016                 if (result)
1017                         return result;
1018                 if (SG_SCSI_RESET_NOTHING == val)
1019                         return 0;
1020                 switch (val) {
1021                 case SG_SCSI_RESET_DEVICE:
1022                         val = SCSI_TRY_RESET_DEVICE;
1023                         break;
1024                 case SG_SCSI_RESET_BUS:
1025                         val = SCSI_TRY_RESET_BUS;
1026                         break;
1027                 case SG_SCSI_RESET_HOST:
1028                         val = SCSI_TRY_RESET_HOST;
1029                         break;
1030                 default:
1031                         return -EINVAL;
1032                 }
1033                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1034                         return -EACCES;
1035                 return (scsi_reset_provider(sdp->device, val) ==
1036                         SUCCESS) ? 0 : -EIO;
1037         case SCSI_IOCTL_SEND_COMMAND:
1038                 if (sdp->detached)
1039                         return -ENODEV;
1040                 if (read_only) {
1041                         unsigned char opcode = WRITE_6;
1042                         Scsi_Ioctl_Command __user *siocp = p;
1043
1044                         if (copy_from_user(&opcode, siocp->data, 1))
1045                                 return -EFAULT;
1046                         if (!sg_allow_access(opcode, sdp->device->type))
1047                                 return -EPERM;
1048                 }
1049                 return sg_scsi_ioctl(filp, sdp->device->request_queue, NULL, p);
1050         case SG_SET_DEBUG:
1051                 result = get_user(val, ip);
1052                 if (result)
1053                         return result;
1054                 sdp->sgdebug = (char) val;
1055                 return 0;
1056         case SCSI_IOCTL_GET_IDLUN:
1057         case SCSI_IOCTL_GET_BUS_NUMBER:
1058         case SCSI_IOCTL_PROBE_HOST:
1059         case SG_GET_TRANSFORM:
1060                 if (sdp->detached)
1061                         return -ENODEV;
1062                 return scsi_ioctl(sdp->device, cmd_in, p);
1063         case BLKSECTGET:
1064                 return put_user(sdp->device->request_queue->max_sectors * 512,
1065                                 ip);
1066         default:
1067                 if (read_only)
1068                         return -EPERM;  /* don't know so take safe approach */
1069                 return scsi_ioctl(sdp->device, cmd_in, p);
1070         }
1071 }
1072
1073 #ifdef CONFIG_COMPAT
1074 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1075 {
1076         Sg_device *sdp;
1077         Sg_fd *sfp;
1078         struct scsi_device *sdev;
1079
1080         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1081                 return -ENXIO;
1082
1083         sdev = sdp->device;
1084         if (sdev->host->hostt->compat_ioctl) { 
1085                 int ret;
1086
1087                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1088
1089                 return ret;
1090         }
1091         
1092         return -ENOIOCTLCMD;
1093 }
1094 #endif
1095
1096 static unsigned int
1097 sg_poll(struct file *filp, poll_table * wait)
1098 {
1099         unsigned int res = 0;
1100         Sg_device *sdp;
1101         Sg_fd *sfp;
1102         Sg_request *srp;
1103         int count = 0;
1104         unsigned long iflags;
1105
1106         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp))
1107             || sfp->closed)
1108                 return POLLERR;
1109         poll_wait(filp, &sfp->read_wait, wait);
1110         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1111         for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1112                 /* if any read waiting, flag it */
1113                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1114                         res = POLLIN | POLLRDNORM;
1115                 ++count;
1116         }
1117         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1118
1119         if (sdp->detached)
1120                 res |= POLLHUP;
1121         else if (!sfp->cmd_q) {
1122                 if (0 == count)
1123                         res |= POLLOUT | POLLWRNORM;
1124         } else if (count < SG_MAX_QUEUE)
1125                 res |= POLLOUT | POLLWRNORM;
1126         SCSI_LOG_TIMEOUT(3, printk("sg_poll: %s, res=0x%x\n",
1127                                    sdp->disk->disk_name, (int) res));
1128         return res;
1129 }
1130
1131 static int
1132 sg_fasync(int fd, struct file *filp, int mode)
1133 {
1134         int retval;
1135         Sg_device *sdp;
1136         Sg_fd *sfp;
1137
1138         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1139                 return -ENXIO;
1140         SCSI_LOG_TIMEOUT(3, printk("sg_fasync: %s, mode=%d\n",
1141                                    sdp->disk->disk_name, mode));
1142
1143         retval = fasync_helper(fd, filp, mode, &sfp->async_qp);
1144         return (retval < 0) ? retval : 0;
1145 }
1146
1147 static struct page *
1148 sg_vma_nopage(struct vm_area_struct *vma, unsigned long addr, int *type)
1149 {
1150         Sg_fd *sfp;
1151         struct page *page = NOPAGE_SIGBUS;
1152         unsigned long offset, len, sa;
1153         Sg_scatter_hold *rsv_schp;
1154         struct scatterlist *sg;
1155         int k;
1156
1157         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1158                 return page;
1159         rsv_schp = &sfp->reserve;
1160         offset = addr - vma->vm_start;
1161         if (offset >= rsv_schp->bufflen)
1162                 return page;
1163         SCSI_LOG_TIMEOUT(3, printk("sg_vma_nopage: offset=%lu, scatg=%d\n",
1164                                    offset, rsv_schp->k_use_sg));
1165         sg = rsv_schp->buffer;
1166         sa = vma->vm_start;
1167         for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
1168              ++k, sg = sg_next(sg)) {
1169                 len = vma->vm_end - sa;
1170                 len = (len < sg->length) ? len : sg->length;
1171                 if (offset < len) {
1172                         page = virt_to_page(page_address(sg_page(sg)) + offset);
1173                         get_page(page); /* increment page count */
1174                         break;
1175                 }
1176                 sa += len;
1177                 offset -= len;
1178         }
1179
1180         if (type)
1181                 *type = VM_FAULT_MINOR;
1182         return page;
1183 }
1184
1185 static struct vm_operations_struct sg_mmap_vm_ops = {
1186         .nopage = sg_vma_nopage,
1187 };
1188
1189 static int
1190 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1191 {
1192         Sg_fd *sfp;
1193         unsigned long req_sz, len, sa;
1194         Sg_scatter_hold *rsv_schp;
1195         int k;
1196         struct scatterlist *sg;
1197
1198         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1199                 return -ENXIO;
1200         req_sz = vma->vm_end - vma->vm_start;
1201         SCSI_LOG_TIMEOUT(3, printk("sg_mmap starting, vm_start=%p, len=%d\n",
1202                                    (void *) vma->vm_start, (int) req_sz));
1203         if (vma->vm_pgoff)
1204                 return -EINVAL; /* want no offset */
1205         rsv_schp = &sfp->reserve;
1206         if (req_sz > rsv_schp->bufflen)
1207                 return -ENOMEM; /* cannot map more than reserved buffer */
1208
1209         sa = vma->vm_start;
1210         sg = rsv_schp->buffer;
1211         for (k = 0; (k < rsv_schp->k_use_sg) && (sa < vma->vm_end);
1212              ++k, sg = sg_next(sg)) {
1213                 len = vma->vm_end - sa;
1214                 len = (len < sg->length) ? len : sg->length;
1215                 sa += len;
1216         }
1217
1218         sfp->mmap_called = 1;
1219         vma->vm_flags |= VM_RESERVED;
1220         vma->vm_private_data = sfp;
1221         vma->vm_ops = &sg_mmap_vm_ops;
1222         return 0;
1223 }
1224
1225 /* This function is a "bottom half" handler that is called by the
1226  * mid level when a command is completed (or has failed). */
1227 static void
1228 sg_cmd_done(void *data, char *sense, int result, int resid)
1229 {
1230         Sg_request *srp = data;
1231         Sg_device *sdp = NULL;
1232         Sg_fd *sfp;
1233         unsigned long iflags;
1234         unsigned int ms;
1235
1236         if (NULL == srp) {
1237                 printk(KERN_ERR "sg_cmd_done: NULL request\n");
1238                 return;
1239         }
1240         sfp = srp->parentfp;
1241         if (sfp)
1242                 sdp = sfp->parentdp;
1243         if ((NULL == sdp) || sdp->detached) {
1244                 printk(KERN_INFO "sg_cmd_done: device detached\n");
1245                 return;
1246         }
1247
1248
1249         SCSI_LOG_TIMEOUT(4, printk("sg_cmd_done: %s, pack_id=%d, res=0x%x\n",
1250                 sdp->disk->disk_name, srp->header.pack_id, result));
1251         srp->header.resid = resid;
1252         ms = jiffies_to_msecs(jiffies);
1253         srp->header.duration = (ms > srp->header.duration) ?
1254                                 (ms - srp->header.duration) : 0;
1255         if (0 != result) {
1256                 struct scsi_sense_hdr sshdr;
1257
1258                 memcpy(srp->sense_b, sense, sizeof (srp->sense_b));
1259                 srp->header.status = 0xff & result;
1260                 srp->header.masked_status = status_byte(result);
1261                 srp->header.msg_status = msg_byte(result);
1262                 srp->header.host_status = host_byte(result);
1263                 srp->header.driver_status = driver_byte(result);
1264                 if ((sdp->sgdebug > 0) &&
1265                     ((CHECK_CONDITION == srp->header.masked_status) ||
1266                      (COMMAND_TERMINATED == srp->header.masked_status)))
1267                         __scsi_print_sense("sg_cmd_done", sense,
1268                                            SCSI_SENSE_BUFFERSIZE);
1269
1270                 /* Following if statement is a patch supplied by Eric Youngdale */
1271                 if (driver_byte(result) != 0
1272                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1273                     && !scsi_sense_is_deferred(&sshdr)
1274                     && sshdr.sense_key == UNIT_ATTENTION
1275                     && sdp->device->removable) {
1276                         /* Detected possible disc change. Set the bit - this */
1277                         /* may be used if there are filesystems using this device */
1278                         sdp->device->changed = 1;
1279                 }
1280         }
1281         /* Rely on write phase to clean out srp status values, so no "else" */
1282
1283         if (sfp->closed) {      /* whoops this fd already released, cleanup */
1284                 SCSI_LOG_TIMEOUT(1, printk("sg_cmd_done: already closed, freeing ...\n"));
1285                 sg_finish_rem_req(srp);
1286                 srp = NULL;
1287                 if (NULL == sfp->headrp) {
1288                         SCSI_LOG_TIMEOUT(1, printk("sg_cmd_done: already closed, final cleanup\n"));
1289                         if (0 == sg_remove_sfp(sdp, sfp)) {     /* device still present */
1290                                 scsi_device_put(sdp->device);
1291                         }
1292                         sfp = NULL;
1293                 }
1294         } else if (srp && srp->orphan) {
1295                 if (sfp->keep_orphan)
1296                         srp->sg_io_owned = 0;
1297                 else {
1298                         sg_finish_rem_req(srp);
1299                         srp = NULL;
1300                 }
1301         }
1302         if (sfp && srp) {
1303                 /* Now wake up any sg_read() that is waiting for this packet. */
1304                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1305                 write_lock_irqsave(&sfp->rq_list_lock, iflags);
1306                 srp->done = 1;
1307                 wake_up_interruptible(&sfp->read_wait);
1308                 write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1309         }
1310 }
1311
1312 static struct file_operations sg_fops = {
1313         .owner = THIS_MODULE,
1314         .read = sg_read,
1315         .write = sg_write,
1316         .poll = sg_poll,
1317         .ioctl = sg_ioctl,
1318 #ifdef CONFIG_COMPAT
1319         .compat_ioctl = sg_compat_ioctl,
1320 #endif
1321         .open = sg_open,
1322         .mmap = sg_mmap,
1323         .release = sg_release,
1324         .fasync = sg_fasync,
1325 };
1326
1327 static struct class *sg_sysfs_class;
1328
1329 static int sg_sysfs_valid = 0;
1330
1331 static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1332 {
1333         struct request_queue *q = scsidp->request_queue;
1334         Sg_device *sdp;
1335         unsigned long iflags;
1336         int error;
1337         u32 k;
1338
1339         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1340         if (!sdp) {
1341                 printk(KERN_WARNING "kmalloc Sg_device failure\n");
1342                 return ERR_PTR(-ENOMEM);
1343         }
1344         error = -ENOMEM;
1345         if (!idr_pre_get(&sg_index_idr, GFP_KERNEL)) {
1346                 printk(KERN_WARNING "idr expansion Sg_device failure\n");
1347                 goto out;
1348         }
1349
1350         write_lock_irqsave(&sg_index_lock, iflags);
1351         error = idr_get_new(&sg_index_idr, sdp, &k);
1352         write_unlock_irqrestore(&sg_index_lock, iflags);
1353
1354         if (error) {
1355                 printk(KERN_WARNING "idr allocation Sg_device failure: %d\n",
1356                        error);
1357                 goto out;
1358         }
1359
1360         if (unlikely(k >= SG_MAX_DEVS))
1361                 goto overflow;
1362
1363         SCSI_LOG_TIMEOUT(3, printk("sg_alloc: dev=%d \n", k));
1364         sprintf(disk->disk_name, "sg%d", k);
1365         disk->first_minor = k;
1366         sdp->disk = disk;
1367         sdp->device = scsidp;
1368         init_waitqueue_head(&sdp->o_excl_wait);
1369         sdp->sg_tablesize = min(q->max_hw_segments, q->max_phys_segments);
1370         sdp->index = k;
1371
1372         error = 0;
1373  out:
1374         if (error) {
1375                 kfree(sdp);
1376                 return ERR_PTR(error);
1377         }
1378         return sdp;
1379
1380  overflow:
1381         sdev_printk(KERN_WARNING, scsidp,
1382                     "Unable to attach sg device type=%d, minor "
1383                     "number exceeds %d\n", scsidp->type, SG_MAX_DEVS - 1);
1384         error = -ENODEV;
1385         goto out;
1386 }
1387
1388 static int
1389 sg_add(struct class_device *cl_dev, struct class_interface *cl_intf)
1390 {
1391         struct scsi_device *scsidp = to_scsi_device(cl_dev->dev);
1392         struct gendisk *disk;
1393         Sg_device *sdp = NULL;
1394         struct cdev * cdev = NULL;
1395         int error;
1396         unsigned long iflags;
1397
1398         disk = alloc_disk(1);
1399         if (!disk) {
1400                 printk(KERN_WARNING "alloc_disk failed\n");
1401                 return -ENOMEM;
1402         }
1403         disk->major = SCSI_GENERIC_MAJOR;
1404
1405         error = -ENOMEM;
1406         cdev = cdev_alloc();
1407         if (!cdev) {
1408                 printk(KERN_WARNING "cdev_alloc failed\n");
1409                 goto out;
1410         }
1411         cdev->owner = THIS_MODULE;
1412         cdev->ops = &sg_fops;
1413
1414         sdp = sg_alloc(disk, scsidp);
1415         if (IS_ERR(sdp)) {
1416                 printk(KERN_WARNING "sg_alloc failed\n");
1417                 error = PTR_ERR(sdp);
1418                 goto out;
1419         }
1420
1421         class_set_devdata(cl_dev, sdp);
1422         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1423         if (error)
1424                 goto cdev_add_err;
1425
1426         sdp->cdev = cdev;
1427         if (sg_sysfs_valid) {
1428                 struct class_device * sg_class_member;
1429
1430                 sg_class_member = class_device_create(sg_sysfs_class, NULL,
1431                                 MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1432                                 cl_dev->dev, "%s",
1433                                 disk->disk_name);
1434                 if (IS_ERR(sg_class_member))
1435                         printk(KERN_WARNING "sg_add: "
1436                                 "class_device_create failed\n");
1437                 class_set_devdata(sg_class_member, sdp);
1438                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj, 
1439                                           &sg_class_member->kobj, "generic");
1440                 if (error)
1441                         printk(KERN_ERR "sg_add: unable to make symlink "
1442                                         "'generic' back to sg%d\n", sdp->index);
1443         } else
1444                 printk(KERN_WARNING "sg_add: sg_sys Invalid\n");
1445
1446         sdev_printk(KERN_NOTICE, scsidp,
1447                     "Attached scsi generic sg%d type %d\n", sdp->index,
1448                     scsidp->type);
1449
1450         return 0;
1451
1452 cdev_add_err:
1453         write_lock_irqsave(&sg_index_lock, iflags);
1454         idr_remove(&sg_index_idr, sdp->index);
1455         write_unlock_irqrestore(&sg_index_lock, iflags);
1456         kfree(sdp);
1457
1458 out:
1459         put_disk(disk);
1460         if (cdev)
1461                 cdev_del(cdev);
1462         return error;
1463 }
1464
1465 static void
1466 sg_remove(struct class_device *cl_dev, struct class_interface *cl_intf)
1467 {
1468         struct scsi_device *scsidp = to_scsi_device(cl_dev->dev);
1469         Sg_device *sdp = class_get_devdata(cl_dev);
1470         unsigned long iflags;
1471         Sg_fd *sfp;
1472         Sg_fd *tsfp;
1473         Sg_request *srp;
1474         Sg_request *tsrp;
1475         int delay;
1476
1477         if (!sdp)
1478                 return;
1479
1480         delay = 0;
1481         write_lock_irqsave(&sg_index_lock, iflags);
1482         if (sdp->headfp) {
1483                 sdp->detached = 1;
1484                 for (sfp = sdp->headfp; sfp; sfp = tsfp) {
1485                         tsfp = sfp->nextfp;
1486                         for (srp = sfp->headrp; srp; srp = tsrp) {
1487                                 tsrp = srp->nextrp;
1488                                 if (sfp->closed || (0 == sg_srp_done(srp, sfp)))
1489                                         sg_finish_rem_req(srp);
1490                         }
1491                         if (sfp->closed) {
1492                                 scsi_device_put(sdp->device);
1493                                 __sg_remove_sfp(sdp, sfp);
1494                         } else {
1495                                 delay = 1;
1496                                 wake_up_interruptible(&sfp->read_wait);
1497                                 kill_fasync(&sfp->async_qp, SIGPOLL,
1498                                             POLL_HUP);
1499                         }
1500                 }
1501                 SCSI_LOG_TIMEOUT(3, printk("sg_remove: dev=%d, dirty\n", sdp->index));
1502                 if (NULL == sdp->headfp) {
1503                         idr_remove(&sg_index_idr, sdp->index);
1504                 }
1505         } else {        /* nothing active, simple case */
1506                 SCSI_LOG_TIMEOUT(3, printk("sg_remove: dev=%d\n", sdp->index));
1507                 idr_remove(&sg_index_idr, sdp->index);
1508         }
1509         write_unlock_irqrestore(&sg_index_lock, iflags);
1510
1511         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1512         class_device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1513         cdev_del(sdp->cdev);
1514         sdp->cdev = NULL;
1515         put_disk(sdp->disk);
1516         sdp->disk = NULL;
1517         if (NULL == sdp->headfp)
1518                 kfree(sdp);
1519
1520         if (delay)
1521                 msleep(10);     /* dirty detach so delay device destruction */
1522 }
1523
1524 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1525 module_param_named(def_reserved_size, def_reserved_size, int,
1526                    S_IRUGO | S_IWUSR);
1527 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1528
1529 MODULE_AUTHOR("Douglas Gilbert");
1530 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1531 MODULE_LICENSE("GPL");
1532 MODULE_VERSION(SG_VERSION_STR);
1533 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1534
1535 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1536                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1537 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1538 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1539
1540 static int __init
1541 init_sg(void)
1542 {
1543         int rc;
1544
1545         if (scatter_elem_sz < PAGE_SIZE) {
1546                 scatter_elem_sz = PAGE_SIZE;
1547                 scatter_elem_sz_prev = scatter_elem_sz;
1548         }
1549         if (def_reserved_size >= 0)
1550                 sg_big_buff = def_reserved_size;
1551         else
1552                 def_reserved_size = sg_big_buff;
1553
1554         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), 
1555                                     SG_MAX_DEVS, "sg");
1556         if (rc)
1557                 return rc;
1558         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1559         if ( IS_ERR(sg_sysfs_class) ) {
1560                 rc = PTR_ERR(sg_sysfs_class);
1561                 goto err_out;
1562         }
1563         sg_sysfs_valid = 1;
1564         rc = scsi_register_interface(&sg_interface);
1565         if (0 == rc) {
1566 #ifdef CONFIG_SCSI_PROC_FS
1567                 sg_proc_init();
1568 #endif                          /* CONFIG_SCSI_PROC_FS */
1569                 return 0;
1570         }
1571         class_destroy(sg_sysfs_class);
1572 err_out:
1573         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1574         return rc;
1575 }
1576
1577 static void __exit
1578 exit_sg(void)
1579 {
1580 #ifdef CONFIG_SCSI_PROC_FS
1581         sg_proc_cleanup();
1582 #endif                          /* CONFIG_SCSI_PROC_FS */
1583         scsi_unregister_interface(&sg_interface);
1584         class_destroy(sg_sysfs_class);
1585         sg_sysfs_valid = 0;
1586         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1587                                  SG_MAX_DEVS);
1588         idr_destroy(&sg_index_idr);
1589 }
1590
1591 static int
1592 sg_start_req(Sg_request * srp)
1593 {
1594         int res;
1595         Sg_fd *sfp = srp->parentfp;
1596         sg_io_hdr_t *hp = &srp->header;
1597         int dxfer_len = (int) hp->dxfer_len;
1598         int dxfer_dir = hp->dxfer_direction;
1599         Sg_scatter_hold *req_schp = &srp->data;
1600         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1601
1602         SCSI_LOG_TIMEOUT(4, printk("sg_start_req: dxfer_len=%d\n", dxfer_len));
1603         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1604                 return 0;
1605         if (sg_allow_dio && (hp->flags & SG_FLAG_DIRECT_IO) &&
1606             (dxfer_dir != SG_DXFER_UNKNOWN) && (0 == hp->iovec_count) &&
1607             (!sfp->parentdp->device->host->unchecked_isa_dma)) {
1608                 res = sg_build_direct(srp, sfp, dxfer_len);
1609                 if (res <= 0)   /* -ve -> error, 0 -> done, 1 -> try indirect */
1610                         return res;
1611         }
1612         if ((!sg_res_in_use(sfp)) && (dxfer_len <= rsv_schp->bufflen))
1613                 sg_link_reserve(sfp, srp, dxfer_len);
1614         else {
1615                 res = sg_build_indirect(req_schp, sfp, dxfer_len);
1616                 if (res) {
1617                         sg_remove_scat(req_schp);
1618                         return res;
1619                 }
1620         }
1621         return 0;
1622 }
1623
1624 static void
1625 sg_finish_rem_req(Sg_request * srp)
1626 {
1627         Sg_fd *sfp = srp->parentfp;
1628         Sg_scatter_hold *req_schp = &srp->data;
1629
1630         SCSI_LOG_TIMEOUT(4, printk("sg_finish_rem_req: res_used=%d\n", (int) srp->res_used));
1631         if (srp->res_used)
1632                 sg_unlink_reserve(sfp, srp);
1633         else
1634                 sg_remove_scat(req_schp);
1635         sg_remove_request(sfp, srp);
1636 }
1637
1638 static int
1639 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1640 {
1641         int sg_bufflen = tablesize * sizeof(struct scatterlist);
1642         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1643
1644         /*
1645          * TODO: test without low_dma, we should not need it since
1646          * the block layer will bounce the buffer for us
1647          *
1648          * XXX(hch): we shouldn't need GFP_DMA for the actual S/G list.
1649          */
1650         if (sfp->low_dma)
1651                  gfp_flags |= GFP_DMA;
1652         schp->buffer = kzalloc(sg_bufflen, gfp_flags);
1653         if (!schp->buffer)
1654                 return -ENOMEM;
1655         sg_init_table(schp->buffer, tablesize);
1656         schp->sglist_len = sg_bufflen;
1657         return tablesize;       /* number of scat_gath elements allocated */
1658 }
1659
1660 #ifdef SG_ALLOW_DIO_CODE
1661 /* vvvvvvvv  following code borrowed from st driver's direct IO vvvvvvvvv */
1662         /* TODO: hopefully we can use the generic block layer code */
1663
1664 /* Pin down user pages and put them into a scatter gather list. Returns <= 0 if
1665    - mapping of all pages not successful
1666    (i.e., either completely successful or fails)
1667 */
1668 static int 
1669 st_map_user_pages(struct scatterlist *sgl, const unsigned int max_pages, 
1670                   unsigned long uaddr, size_t count, int rw)
1671 {
1672         unsigned long end = (uaddr + count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1673         unsigned long start = uaddr >> PAGE_SHIFT;
1674         const int nr_pages = end - start;
1675         int res, i, j;
1676         struct page **pages;
1677
1678         /* User attempted Overflow! */
1679         if ((uaddr + count) < uaddr)
1680                 return -EINVAL;
1681
1682         /* Too big */
1683         if (nr_pages > max_pages)
1684                 return -ENOMEM;
1685
1686         /* Hmm? */
1687         if (count == 0)
1688                 return 0;
1689
1690         if ((pages = kmalloc(max_pages * sizeof(*pages), GFP_ATOMIC)) == NULL)
1691                 return -ENOMEM;
1692
1693         /* Try to fault in all of the necessary pages */
1694         down_read(&current->mm->mmap_sem);
1695         /* rw==READ means read from drive, write into memory area */
1696         res = get_user_pages(
1697                 current,
1698                 current->mm,
1699                 uaddr,
1700                 nr_pages,
1701                 rw == READ,
1702                 0, /* don't force */
1703                 pages,
1704                 NULL);
1705         up_read(&current->mm->mmap_sem);
1706
1707         /* Errors and no page mapped should return here */
1708         if (res < nr_pages)
1709                 goto out_unmap;
1710
1711         for (i=0; i < nr_pages; i++) {
1712                 /* FIXME: flush superflous for rw==READ,
1713                  * probably wrong function for rw==WRITE
1714                  */
1715                 flush_dcache_page(pages[i]);
1716                 /* ?? Is locking needed? I don't think so */
1717                 /* if (TestSetPageLocked(pages[i]))
1718                    goto out_unlock; */
1719         }
1720
1721         sg_set_page(sgl, pages[0], 0, uaddr & ~PAGE_MASK);
1722         if (nr_pages > 1) {
1723                 sgl[0].length = PAGE_SIZE - sgl[0].offset;
1724                 count -= sgl[0].length;
1725                 for (i=1; i < nr_pages ; i++)
1726                         sg_set_page(&sgl[i], pages[i], count < PAGE_SIZE ? count : PAGE_SIZE, 0);
1727         }
1728         else {
1729                 sgl[0].length = count;
1730         }
1731
1732         kfree(pages);
1733         return nr_pages;
1734
1735  out_unmap:
1736         if (res > 0) {
1737                 for (j=0; j < res; j++)
1738                         page_cache_release(pages[j]);
1739                 res = 0;
1740         }
1741         kfree(pages);
1742         return res;
1743 }
1744
1745
1746 /* And unmap them... */
1747 static int 
1748 st_unmap_user_pages(struct scatterlist *sgl, const unsigned int nr_pages,
1749                     int dirtied)
1750 {
1751         int i;
1752
1753         for (i=0; i < nr_pages; i++) {
1754                 struct page *page = sg_page(&sgl[i]);
1755
1756                 if (dirtied)
1757                         SetPageDirty(page);
1758                 /* unlock_page(page); */
1759                 /* FIXME: cache flush missing for rw==READ
1760                  * FIXME: call the correct reference counting function
1761                  */
1762                 page_cache_release(page);
1763         }
1764
1765         return 0;
1766 }
1767
1768 /* ^^^^^^^^  above code borrowed from st driver's direct IO ^^^^^^^^^ */
1769 #endif
1770
1771
1772 /* Returns: -ve -> error, 0 -> done, 1 -> try indirect */
1773 static int
1774 sg_build_direct(Sg_request * srp, Sg_fd * sfp, int dxfer_len)
1775 {
1776 #ifdef SG_ALLOW_DIO_CODE
1777         sg_io_hdr_t *hp = &srp->header;
1778         Sg_scatter_hold *schp = &srp->data;
1779         int sg_tablesize = sfp->parentdp->sg_tablesize;
1780         int mx_sc_elems, res;
1781         struct scsi_device *sdev = sfp->parentdp->device;
1782
1783         if (((unsigned long)hp->dxferp &
1784                         queue_dma_alignment(sdev->request_queue)) != 0)
1785                 return 1;
1786
1787         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1788         if (mx_sc_elems <= 0) {
1789                 return 1;
1790         }
1791         res = st_map_user_pages(schp->buffer, mx_sc_elems,
1792                                 (unsigned long)hp->dxferp, dxfer_len, 
1793                                 (SG_DXFER_TO_DEV == hp->dxfer_direction) ? 1 : 0);
1794         if (res <= 0) {
1795                 sg_remove_scat(schp);
1796                 return 1;
1797         }
1798         schp->k_use_sg = res;
1799         schp->dio_in_use = 1;
1800         hp->info |= SG_INFO_DIRECT_IO;
1801         return 0;
1802 #else
1803         return 1;
1804 #endif
1805 }
1806
1807 static int
1808 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1809 {
1810         struct scatterlist *sg;
1811         int ret_sz = 0, k, rem_sz, num, mx_sc_elems;
1812         int sg_tablesize = sfp->parentdp->sg_tablesize;
1813         int blk_size = buff_size;
1814         struct page *p = NULL;
1815
1816         if (blk_size < 0)
1817                 return -EFAULT;
1818         if (0 == blk_size)
1819                 ++blk_size;     /* don't know why */
1820 /* round request up to next highest SG_SECTOR_SZ byte boundary */
1821         blk_size = (blk_size + SG_SECTOR_MSK) & (~SG_SECTOR_MSK);
1822         SCSI_LOG_TIMEOUT(4, printk("sg_build_indirect: buff_size=%d, blk_size=%d\n",
1823                                    buff_size, blk_size));
1824
1825         /* N.B. ret_sz carried into this block ... */
1826         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1827         if (mx_sc_elems < 0)
1828                 return mx_sc_elems;     /* most likely -ENOMEM */
1829
1830         num = scatter_elem_sz;
1831         if (unlikely(num != scatter_elem_sz_prev)) {
1832                 if (num < PAGE_SIZE) {
1833                         scatter_elem_sz = PAGE_SIZE;
1834                         scatter_elem_sz_prev = PAGE_SIZE;
1835                 } else
1836                         scatter_elem_sz_prev = num;
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 = sg_next(sg)) {
1841                 
1842                 num = (rem_sz > scatter_elem_sz_prev) ?
1843                       scatter_elem_sz_prev : rem_sz;
1844                 p = sg_page_malloc(num, sfp->low_dma, &ret_sz);
1845                 if (!p)
1846                         return -ENOMEM;
1847
1848                 if (num == scatter_elem_sz_prev) {
1849                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1850                                 scatter_elem_sz = ret_sz;
1851                                 scatter_elem_sz_prev = ret_sz;
1852                         }
1853                 }
1854                 sg_set_page(sg, p, (ret_sz > num) ? num : ret_sz, 0);
1855
1856                 SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k=%d, num=%d, "
1857                                  "ret_sz=%d\n", k, num, ret_sz));
1858         }               /* end of for loop */
1859
1860         schp->k_use_sg = k;
1861         SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k_use_sg=%d, "
1862                          "rem_sz=%d\n", k, rem_sz));
1863
1864         schp->bufflen = blk_size;
1865         if (rem_sz > 0) /* must have failed */
1866                 return -ENOMEM;
1867
1868         return 0;
1869 }
1870
1871 static int
1872 sg_write_xfer(Sg_request * srp)
1873 {
1874         sg_io_hdr_t *hp = &srp->header;
1875         Sg_scatter_hold *schp = &srp->data;
1876         struct scatterlist *sg = schp->buffer;
1877         int num_xfer = 0;
1878         int j, k, onum, usglen, ksglen, res;
1879         int iovec_count = (int) hp->iovec_count;
1880         int dxfer_dir = hp->dxfer_direction;
1881         unsigned char *p;
1882         unsigned char __user *up;
1883         int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
1884
1885         if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_TO_DEV == dxfer_dir) ||
1886             (SG_DXFER_TO_FROM_DEV == dxfer_dir)) {
1887                 num_xfer = (int) (new_interface ? hp->dxfer_len : hp->flags);
1888                 if (schp->bufflen < num_xfer)
1889                         num_xfer = schp->bufflen;
1890         }
1891         if ((num_xfer <= 0) || (schp->dio_in_use) ||
1892             (new_interface
1893              && ((SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO) & hp->flags)))
1894                 return 0;
1895
1896         SCSI_LOG_TIMEOUT(4, printk("sg_write_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
1897                           num_xfer, iovec_count, schp->k_use_sg));
1898         if (iovec_count) {
1899                 onum = iovec_count;
1900                 if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
1901                         return -EFAULT;
1902         } else
1903                 onum = 1;
1904
1905         ksglen = sg->length;
1906         p = page_address(sg_page(sg));
1907         for (j = 0, k = 0; j < onum; ++j) {
1908                 res = sg_u_iovec(hp, iovec_count, j, 1, &usglen, &up);
1909                 if (res)
1910                         return res;
1911
1912                 for (; p; sg = sg_next(sg), ksglen = sg->length,
1913                      p = page_address(sg_page(sg))) {
1914                         if (usglen <= 0)
1915                                 break;
1916                         if (ksglen > usglen) {
1917                                 if (usglen >= num_xfer) {
1918                                         if (__copy_from_user(p, up, num_xfer))
1919                                                 return -EFAULT;
1920                                         return 0;
1921                                 }
1922                                 if (__copy_from_user(p, up, usglen))
1923                                         return -EFAULT;
1924                                 p += usglen;
1925                                 ksglen -= usglen;
1926                                 break;
1927                         } else {
1928                                 if (ksglen >= num_xfer) {
1929                                         if (__copy_from_user(p, up, num_xfer))
1930                                                 return -EFAULT;
1931                                         return 0;
1932                                 }
1933                                 if (__copy_from_user(p, up, ksglen))
1934                                         return -EFAULT;
1935                                 up += ksglen;
1936                                 usglen -= ksglen;
1937                         }
1938                         ++k;
1939                         if (k >= schp->k_use_sg)
1940                                 return 0;
1941                 }
1942         }
1943
1944         return 0;
1945 }
1946
1947 static int
1948 sg_u_iovec(sg_io_hdr_t * hp, int sg_num, int ind,
1949            int wr_xf, int *countp, unsigned char __user **up)
1950 {
1951         int num_xfer = (int) hp->dxfer_len;
1952         unsigned char __user *p = hp->dxferp;
1953         int count;
1954
1955         if (0 == sg_num) {
1956                 if (wr_xf && ('\0' == hp->interface_id))
1957                         count = (int) hp->flags;        /* holds "old" input_size */
1958                 else
1959                         count = num_xfer;
1960         } else {
1961                 sg_iovec_t iovec;
1962                 if (__copy_from_user(&iovec, p + ind*SZ_SG_IOVEC, SZ_SG_IOVEC))
1963                         return -EFAULT;
1964                 p = iovec.iov_base;
1965                 count = (int) iovec.iov_len;
1966         }
1967         if (!access_ok(wr_xf ? VERIFY_READ : VERIFY_WRITE, p, count))
1968                 return -EFAULT;
1969         if (up)
1970                 *up = p;
1971         if (countp)
1972                 *countp = count;
1973         return 0;
1974 }
1975
1976 static void
1977 sg_remove_scat(Sg_scatter_hold * schp)
1978 {
1979         SCSI_LOG_TIMEOUT(4, printk("sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1980         if (schp->buffer && (schp->sglist_len > 0)) {
1981                 struct scatterlist *sg = schp->buffer;
1982
1983                 if (schp->dio_in_use) {
1984 #ifdef SG_ALLOW_DIO_CODE
1985                         st_unmap_user_pages(sg, schp->k_use_sg, TRUE);
1986 #endif
1987                 } else {
1988                         int k;
1989
1990                         for (k = 0; (k < schp->k_use_sg) && sg_page(sg);
1991                              ++k, sg = sg_next(sg)) {
1992                                 SCSI_LOG_TIMEOUT(5, printk(
1993                                     "sg_remove_scat: k=%d, pg=0x%p, len=%d\n",
1994                                     k, sg_page(sg), sg->length));
1995                                 sg_page_free(sg_page(sg), sg->length);
1996                         }
1997                 }
1998                 kfree(schp->buffer);
1999         }
2000         memset(schp, 0, sizeof (*schp));
2001 }
2002
2003 static int
2004 sg_read_xfer(Sg_request * srp)
2005 {
2006         sg_io_hdr_t *hp = &srp->header;
2007         Sg_scatter_hold *schp = &srp->data;
2008         struct scatterlist *sg = schp->buffer;
2009         int num_xfer = 0;
2010         int j, k, onum, usglen, ksglen, res;
2011         int iovec_count = (int) hp->iovec_count;
2012         int dxfer_dir = hp->dxfer_direction;
2013         unsigned char *p;
2014         unsigned char __user *up;
2015         int new_interface = ('\0' == hp->interface_id) ? 0 : 1;
2016
2017         if ((SG_DXFER_UNKNOWN == dxfer_dir) || (SG_DXFER_FROM_DEV == dxfer_dir)
2018             || (SG_DXFER_TO_FROM_DEV == dxfer_dir)) {
2019                 num_xfer = hp->dxfer_len;
2020                 if (schp->bufflen < num_xfer)
2021                         num_xfer = schp->bufflen;
2022         }
2023         if ((num_xfer <= 0) || (schp->dio_in_use) ||
2024             (new_interface
2025              && ((SG_FLAG_NO_DXFER | SG_FLAG_MMAP_IO) & hp->flags)))
2026                 return 0;
2027
2028         SCSI_LOG_TIMEOUT(4, printk("sg_read_xfer: num_xfer=%d, iovec_count=%d, k_use_sg=%d\n",
2029                           num_xfer, iovec_count, schp->k_use_sg));
2030         if (iovec_count) {
2031                 onum = iovec_count;
2032                 if (!access_ok(VERIFY_READ, hp->dxferp, SZ_SG_IOVEC * onum))
2033                         return -EFAULT;
2034         } else
2035                 onum = 1;
2036
2037         p = page_address(sg_page(sg));
2038         ksglen = sg->length;
2039         for (j = 0, k = 0; j < onum; ++j) {
2040                 res = sg_u_iovec(hp, iovec_count, j, 0, &usglen, &up);
2041                 if (res)
2042                         return res;
2043
2044                 for (; p; sg = sg_next(sg), ksglen = sg->length,
2045                      p = page_address(sg_page(sg))) {
2046                         if (usglen <= 0)
2047                                 break;
2048                         if (ksglen > usglen) {
2049                                 if (usglen >= num_xfer) {
2050                                         if (__copy_to_user(up, p, num_xfer))
2051                                                 return -EFAULT;
2052                                         return 0;
2053                                 }
2054                                 if (__copy_to_user(up, p, usglen))
2055                                         return -EFAULT;
2056                                 p += usglen;
2057                                 ksglen -= usglen;
2058                                 break;
2059                         } else {
2060                                 if (ksglen >= num_xfer) {
2061                                         if (__copy_to_user(up, p, num_xfer))
2062                                                 return -EFAULT;
2063                                         return 0;
2064                                 }
2065                                 if (__copy_to_user(up, p, ksglen))
2066                                         return -EFAULT;
2067                                 up += ksglen;
2068                                 usglen -= ksglen;
2069                         }
2070                         ++k;
2071                         if (k >= schp->k_use_sg)
2072                                 return 0;
2073                 }
2074         }
2075
2076         return 0;
2077 }
2078
2079 static int
2080 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
2081 {
2082         Sg_scatter_hold *schp = &srp->data;
2083         struct scatterlist *sg = schp->buffer;
2084         int k, num;
2085
2086         SCSI_LOG_TIMEOUT(4, printk("sg_read_oxfer: num_read_xfer=%d\n",
2087                                    num_read_xfer));
2088         if ((!outp) || (num_read_xfer <= 0))
2089                 return 0;
2090
2091         for (k = 0; (k < schp->k_use_sg) && sg_page(sg); ++k, sg = sg_next(sg)) {
2092                 num = sg->length;
2093                 if (num > num_read_xfer) {
2094                         if (__copy_to_user(outp, page_address(sg_page(sg)),
2095                                            num_read_xfer))
2096                                 return -EFAULT;
2097                         break;
2098                 } else {
2099                         if (__copy_to_user(outp, page_address(sg_page(sg)),
2100                                            num))
2101                                 return -EFAULT;
2102                         num_read_xfer -= num;
2103                         if (num_read_xfer <= 0)
2104                                 break;
2105                         outp += num;
2106                 }
2107         }
2108
2109         return 0;
2110 }
2111
2112 static void
2113 sg_build_reserve(Sg_fd * sfp, int req_size)
2114 {
2115         Sg_scatter_hold *schp = &sfp->reserve;
2116
2117         SCSI_LOG_TIMEOUT(4, printk("sg_build_reserve: req_size=%d\n", req_size));
2118         do {
2119                 if (req_size < PAGE_SIZE)
2120                         req_size = PAGE_SIZE;
2121                 if (0 == sg_build_indirect(schp, sfp, req_size))
2122                         return;
2123                 else
2124                         sg_remove_scat(schp);
2125                 req_size >>= 1; /* divide by 2 */
2126         } while (req_size > (PAGE_SIZE / 2));
2127 }
2128
2129 static void
2130 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2131 {
2132         Sg_scatter_hold *req_schp = &srp->data;
2133         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2134         struct scatterlist *sg = rsv_schp->buffer;
2135         int k, num, rem;
2136
2137         srp->res_used = 1;
2138         SCSI_LOG_TIMEOUT(4, printk("sg_link_reserve: size=%d\n", size));
2139         rem = size;
2140
2141         for (k = 0; k < rsv_schp->k_use_sg; ++k, sg = sg_next(sg)) {
2142                 num = sg->length;
2143                 if (rem <= num) {
2144                         sfp->save_scat_len = num;
2145                         sg->length = rem;
2146                         req_schp->k_use_sg = k + 1;
2147                         req_schp->sglist_len = rsv_schp->sglist_len;
2148                         req_schp->buffer = rsv_schp->buffer;
2149
2150                         req_schp->bufflen = size;
2151                         req_schp->b_malloc_len = rsv_schp->b_malloc_len;
2152                         break;
2153                 } else
2154                         rem -= num;
2155         }
2156
2157         if (k >= rsv_schp->k_use_sg)
2158                 SCSI_LOG_TIMEOUT(1, printk("sg_link_reserve: BAD size\n"));
2159 }
2160
2161 static void
2162 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2163 {
2164         Sg_scatter_hold *req_schp = &srp->data;
2165         Sg_scatter_hold *rsv_schp = &sfp->reserve;
2166
2167         SCSI_LOG_TIMEOUT(4, printk("sg_unlink_reserve: req->k_use_sg=%d\n",
2168                                    (int) req_schp->k_use_sg));
2169         if ((rsv_schp->k_use_sg > 0) && (req_schp->k_use_sg > 0)) {
2170                 struct scatterlist *sg = rsv_schp->buffer;
2171
2172                 if (sfp->save_scat_len > 0)
2173                         (sg + (req_schp->k_use_sg - 1))->length =
2174                             (unsigned) sfp->save_scat_len;
2175                 else
2176                         SCSI_LOG_TIMEOUT(1, printk ("sg_unlink_reserve: BAD save_scat_len\n"));
2177         }
2178         req_schp->k_use_sg = 0;
2179         req_schp->bufflen = 0;
2180         req_schp->buffer = NULL;
2181         req_schp->sglist_len = 0;
2182         sfp->save_scat_len = 0;
2183         srp->res_used = 0;
2184 }
2185
2186 static Sg_request *
2187 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2188 {
2189         Sg_request *resp;
2190         unsigned long iflags;
2191
2192         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2193         for (resp = sfp->headrp; resp; resp = resp->nextrp) {
2194                 /* look for requests that are ready + not SG_IO owned */
2195                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2196                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2197                         resp->done = 2; /* guard against other readers */
2198                         break;
2199                 }
2200         }
2201         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2202         return resp;
2203 }
2204
2205 #ifdef CONFIG_SCSI_PROC_FS
2206 static Sg_request *
2207 sg_get_nth_request(Sg_fd * sfp, int nth)
2208 {
2209         Sg_request *resp;
2210         unsigned long iflags;
2211         int k;
2212
2213         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2214         for (k = 0, resp = sfp->headrp; resp && (k < nth);
2215              ++k, resp = resp->nextrp) ;
2216         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2217         return resp;
2218 }
2219 #endif
2220
2221 /* always adds to end of list */
2222 static Sg_request *
2223 sg_add_request(Sg_fd * sfp)
2224 {
2225         int k;
2226         unsigned long iflags;
2227         Sg_request *resp;
2228         Sg_request *rp = sfp->req_arr;
2229
2230         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2231         resp = sfp->headrp;
2232         if (!resp) {
2233                 memset(rp, 0, sizeof (Sg_request));
2234                 rp->parentfp = sfp;
2235                 resp = rp;
2236                 sfp->headrp = resp;
2237         } else {
2238                 if (0 == sfp->cmd_q)
2239                         resp = NULL;    /* command queuing disallowed */
2240                 else {
2241                         for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2242                                 if (!rp->parentfp)
2243                                         break;
2244                         }
2245                         if (k < SG_MAX_QUEUE) {
2246                                 memset(rp, 0, sizeof (Sg_request));
2247                                 rp->parentfp = sfp;
2248                                 while (resp->nextrp)
2249                                         resp = resp->nextrp;
2250                                 resp->nextrp = rp;
2251                                 resp = rp;
2252                         } else
2253                                 resp = NULL;
2254                 }
2255         }
2256         if (resp) {
2257                 resp->nextrp = NULL;
2258                 resp->header.duration = jiffies_to_msecs(jiffies);
2259         }
2260         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2261         return resp;
2262 }
2263
2264 /* Return of 1 for found; 0 for not found */
2265 static int
2266 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2267 {
2268         Sg_request *prev_rp;
2269         Sg_request *rp;
2270         unsigned long iflags;
2271         int res = 0;
2272
2273         if ((!sfp) || (!srp) || (!sfp->headrp))
2274                 return res;
2275         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2276         prev_rp = sfp->headrp;
2277         if (srp == prev_rp) {
2278                 sfp->headrp = prev_rp->nextrp;
2279                 prev_rp->parentfp = NULL;
2280                 res = 1;
2281         } else {
2282                 while ((rp = prev_rp->nextrp)) {
2283                         if (srp == rp) {
2284                                 prev_rp->nextrp = rp->nextrp;
2285                                 rp->parentfp = NULL;
2286                                 res = 1;
2287                                 break;
2288                         }
2289                         prev_rp = rp;
2290                 }
2291         }
2292         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2293         return res;
2294 }
2295
2296 #ifdef CONFIG_SCSI_PROC_FS
2297 static Sg_fd *
2298 sg_get_nth_sfp(Sg_device * sdp, int nth)
2299 {
2300         Sg_fd *resp;
2301         unsigned long iflags;
2302         int k;
2303
2304         read_lock_irqsave(&sg_index_lock, iflags);
2305         for (k = 0, resp = sdp->headfp; resp && (k < nth);
2306              ++k, resp = resp->nextfp) ;
2307         read_unlock_irqrestore(&sg_index_lock, iflags);
2308         return resp;
2309 }
2310 #endif
2311
2312 static Sg_fd *
2313 sg_add_sfp(Sg_device * sdp, int dev)
2314 {
2315         Sg_fd *sfp;
2316         unsigned long iflags;
2317         int bufflen;
2318
2319         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2320         if (!sfp)
2321                 return NULL;
2322
2323         init_waitqueue_head(&sfp->read_wait);
2324         rwlock_init(&sfp->rq_list_lock);
2325
2326         sfp->timeout = SG_DEFAULT_TIMEOUT;
2327         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2328         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2329         sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2330             sdp->device->host->unchecked_isa_dma : 1;
2331         sfp->cmd_q = SG_DEF_COMMAND_Q;
2332         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2333         sfp->parentdp = sdp;
2334         write_lock_irqsave(&sg_index_lock, iflags);
2335         if (!sdp->headfp)
2336                 sdp->headfp = sfp;
2337         else {                  /* add to tail of existing list */
2338                 Sg_fd *pfp = sdp->headfp;
2339                 while (pfp->nextfp)
2340                         pfp = pfp->nextfp;
2341                 pfp->nextfp = sfp;
2342         }
2343         write_unlock_irqrestore(&sg_index_lock, iflags);
2344         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp: sfp=0x%p\n", sfp));
2345         if (unlikely(sg_big_buff != def_reserved_size))
2346                 sg_big_buff = def_reserved_size;
2347
2348         bufflen = min_t(int, sg_big_buff,
2349                         sdp->device->request_queue->max_sectors * 512);
2350         sg_build_reserve(sfp, bufflen);
2351         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp:   bufflen=%d, k_use_sg=%d\n",
2352                            sfp->reserve.bufflen, sfp->reserve.k_use_sg));
2353         return sfp;
2354 }
2355
2356 static void
2357 __sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
2358 {
2359         Sg_fd *fp;
2360         Sg_fd *prev_fp;
2361
2362         prev_fp = sdp->headfp;
2363         if (sfp == prev_fp)
2364                 sdp->headfp = prev_fp->nextfp;
2365         else {
2366                 while ((fp = prev_fp->nextfp)) {
2367                         if (sfp == fp) {
2368                                 prev_fp->nextfp = fp->nextfp;
2369                                 break;
2370                         }
2371                         prev_fp = fp;
2372                 }
2373         }
2374         if (sfp->reserve.bufflen > 0) {
2375                 SCSI_LOG_TIMEOUT(6, 
2376                         printk("__sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2377                         (int) sfp->reserve.bufflen, (int) sfp->reserve.k_use_sg));
2378                 sg_remove_scat(&sfp->reserve);
2379         }
2380         sfp->parentdp = NULL;
2381         SCSI_LOG_TIMEOUT(6, printk("__sg_remove_sfp:    sfp=0x%p\n", sfp));
2382         kfree(sfp);
2383 }
2384
2385 /* Returns 0 in normal case, 1 when detached and sdp object removed */
2386 static int
2387 sg_remove_sfp(Sg_device * sdp, Sg_fd * sfp)
2388 {
2389         Sg_request *srp;
2390         Sg_request *tsrp;
2391         int dirty = 0;
2392         int res = 0;
2393
2394         for (srp = sfp->headrp; srp; srp = tsrp) {
2395                 tsrp = srp->nextrp;
2396                 if (sg_srp_done(srp, sfp))
2397                         sg_finish_rem_req(srp);
2398                 else
2399                         ++dirty;
2400         }
2401         if (0 == dirty) {
2402                 unsigned long iflags;
2403
2404                 write_lock_irqsave(&sg_index_lock, iflags);
2405                 __sg_remove_sfp(sdp, sfp);
2406                 if (sdp->detached && (NULL == sdp->headfp)) {
2407                         idr_remove(&sg_index_idr, sdp->index);
2408                         kfree(sdp);
2409                         res = 1;
2410                 }
2411                 write_unlock_irqrestore(&sg_index_lock, iflags);
2412         } else {
2413                 /* MOD_INC's to inhibit unloading sg and associated adapter driver */
2414                 /* only bump the access_count if we actually succeeded in
2415                  * throwing another counter on the host module */
2416                 scsi_device_get(sdp->device);   /* XXX: retval ignored? */      
2417                 sfp->closed = 1;        /* flag dirty state on this fd */
2418                 SCSI_LOG_TIMEOUT(1, printk("sg_remove_sfp: worrisome, %d writes pending\n",
2419                                   dirty));
2420         }
2421         return res;
2422 }
2423
2424 static int
2425 sg_res_in_use(Sg_fd * sfp)
2426 {
2427         const Sg_request *srp;
2428         unsigned long iflags;
2429
2430         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2431         for (srp = sfp->headrp; srp; srp = srp->nextrp)
2432                 if (srp->res_used)
2433                         break;
2434         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2435         return srp ? 1 : 0;
2436 }
2437
2438 /* The size fetched (value output via retSzp) set when non-NULL return */
2439 static struct page *
2440 sg_page_malloc(int rqSz, int lowDma, int *retSzp)
2441 {
2442         struct page *resp = NULL;
2443         gfp_t page_mask;
2444         int order, a_size;
2445         int resSz;
2446
2447         if ((rqSz <= 0) || (NULL == retSzp))
2448                 return resp;
2449
2450         if (lowDma)
2451                 page_mask = GFP_ATOMIC | GFP_DMA | __GFP_COMP | __GFP_NOWARN;
2452         else
2453                 page_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
2454
2455         for (order = 0, a_size = PAGE_SIZE; a_size < rqSz;
2456              order++, a_size <<= 1) ;
2457         resSz = a_size;         /* rounded up if necessary */
2458         resp = alloc_pages(page_mask, order);
2459         while ((!resp) && order) {
2460                 --order;
2461                 a_size >>= 1;   /* divide by 2, until PAGE_SIZE */
2462                 resp =  alloc_pages(page_mask, order);  /* try half */
2463                 resSz = a_size;
2464         }
2465         if (resp) {
2466                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2467                         memset(page_address(resp), 0, resSz);
2468                 *retSzp = resSz;
2469         }
2470         return resp;
2471 }
2472
2473 static void
2474 sg_page_free(struct page *page, int size)
2475 {
2476         int order, a_size;
2477
2478         if (!page)
2479                 return;
2480         for (order = 0, a_size = PAGE_SIZE; a_size < size;
2481              order++, a_size <<= 1) ;
2482         __free_pages(page, order);
2483 }
2484
2485 #ifndef MAINTENANCE_IN_CMD
2486 #define MAINTENANCE_IN_CMD 0xa3
2487 #endif
2488
2489 static unsigned char allow_ops[] = { TEST_UNIT_READY, REQUEST_SENSE,
2490         INQUIRY, READ_CAPACITY, READ_BUFFER, READ_6, READ_10, READ_12,
2491         READ_16, MODE_SENSE, MODE_SENSE_10, LOG_SENSE, REPORT_LUNS,
2492         SERVICE_ACTION_IN, RECEIVE_DIAGNOSTIC, READ_LONG, MAINTENANCE_IN_CMD
2493 };
2494
2495 static int
2496 sg_allow_access(unsigned char opcode, char dev_type)
2497 {
2498         int k;
2499
2500         if (TYPE_SCANNER == dev_type)   /* TYPE_ROM maybe burner */
2501                 return 1;
2502         for (k = 0; k < sizeof (allow_ops); ++k) {
2503                 if (opcode == allow_ops[k])
2504                         return 1;
2505         }
2506         return 0;
2507 }
2508
2509 #ifdef CONFIG_SCSI_PROC_FS
2510 static int
2511 sg_idr_max_id(int id, void *p, void *data)
2512 {
2513         int *k = data;
2514
2515         if (*k < id)
2516                 *k = id;
2517
2518         return 0;
2519 }
2520
2521 static int
2522 sg_last_dev(void)
2523 {
2524         int k = 0;
2525         unsigned long iflags;
2526
2527         read_lock_irqsave(&sg_index_lock, iflags);
2528         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2529         read_unlock_irqrestore(&sg_index_lock, iflags);
2530         return k + 1;           /* origin 1 */
2531 }
2532 #endif
2533
2534 static Sg_device *
2535 sg_get_dev(int dev)
2536 {
2537         Sg_device *sdp;
2538         unsigned long iflags;
2539
2540         read_lock_irqsave(&sg_index_lock, iflags);
2541         sdp = idr_find(&sg_index_idr, dev);
2542         read_unlock_irqrestore(&sg_index_lock, iflags);
2543
2544         return sdp;
2545 }
2546
2547 #ifdef CONFIG_SCSI_PROC_FS
2548
2549 static struct proc_dir_entry *sg_proc_sgp = NULL;
2550
2551 static char sg_proc_sg_dirname[] = "scsi/sg";
2552
2553 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2554
2555 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2556 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2557                                   size_t count, loff_t *off);
2558 static struct file_operations adio_fops = {
2559         /* .owner, .read and .llseek added in sg_proc_init() */
2560         .open = sg_proc_single_open_adio,
2561         .write = sg_proc_write_adio,
2562         .release = single_release,
2563 };
2564
2565 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2566 static ssize_t sg_proc_write_dressz(struct file *filp, 
2567                 const char __user *buffer, size_t count, loff_t *off);
2568 static struct file_operations dressz_fops = {
2569         .open = sg_proc_single_open_dressz,
2570         .write = sg_proc_write_dressz,
2571         .release = single_release,
2572 };
2573
2574 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2575 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2576 static struct file_operations version_fops = {
2577         .open = sg_proc_single_open_version,
2578         .release = single_release,
2579 };
2580
2581 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2582 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2583 static struct file_operations devhdr_fops = {
2584         .open = sg_proc_single_open_devhdr,
2585         .release = single_release,
2586 };
2587
2588 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2589 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2590 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2591 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2592 static void dev_seq_stop(struct seq_file *s, void *v);
2593 static struct file_operations dev_fops = {
2594         .open = sg_proc_open_dev,
2595         .release = seq_release,
2596 };
2597 static struct seq_operations dev_seq_ops = {
2598         .start = dev_seq_start,
2599         .next  = dev_seq_next,
2600         .stop  = dev_seq_stop,
2601         .show  = sg_proc_seq_show_dev,
2602 };
2603
2604 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2605 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2606 static struct file_operations devstrs_fops = {
2607         .open = sg_proc_open_devstrs,
2608         .release = seq_release,
2609 };
2610 static struct seq_operations devstrs_seq_ops = {
2611         .start = dev_seq_start,
2612         .next  = dev_seq_next,
2613         .stop  = dev_seq_stop,
2614         .show  = sg_proc_seq_show_devstrs,
2615 };
2616
2617 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2618 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2619 static struct file_operations debug_fops = {
2620         .open = sg_proc_open_debug,
2621         .release = seq_release,
2622 };
2623 static struct seq_operations debug_seq_ops = {
2624         .start = dev_seq_start,
2625         .next  = dev_seq_next,
2626         .stop  = dev_seq_stop,
2627         .show  = sg_proc_seq_show_debug,
2628 };
2629
2630
2631 struct sg_proc_leaf {
2632         const char * name;
2633         struct file_operations * fops;
2634 };
2635
2636 static struct sg_proc_leaf sg_proc_leaf_arr[] = {
2637         {"allow_dio", &adio_fops},
2638         {"debug", &debug_fops},
2639         {"def_reserved_size", &dressz_fops},
2640         {"device_hdr", &devhdr_fops},
2641         {"devices", &dev_fops},
2642         {"device_strs", &devstrs_fops},
2643         {"version", &version_fops}
2644 };
2645
2646 static int
2647 sg_proc_init(void)
2648 {
2649         int k, mask;
2650         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2651         struct proc_dir_entry *pdep;
2652         struct sg_proc_leaf * leaf;
2653
2654         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2655         if (!sg_proc_sgp)
2656                 return 1;
2657         for (k = 0; k < num_leaves; ++k) {
2658                 leaf = &sg_proc_leaf_arr[k];
2659                 mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2660                 pdep = create_proc_entry(leaf->name, mask, sg_proc_sgp);
2661                 if (pdep) {
2662                         leaf->fops->owner = THIS_MODULE,
2663                         leaf->fops->read = seq_read,
2664                         leaf->fops->llseek = seq_lseek,
2665                         pdep->proc_fops = leaf->fops;
2666                 }
2667         }
2668         return 0;
2669 }
2670
2671 static void
2672 sg_proc_cleanup(void)
2673 {
2674         int k;
2675         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2676
2677         if (!sg_proc_sgp)
2678                 return;
2679         for (k = 0; k < num_leaves; ++k)
2680                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2681         remove_proc_entry(sg_proc_sg_dirname, NULL);
2682 }
2683
2684
2685 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2686 {
2687         seq_printf(s, "%d\n", *((int *)s->private));
2688         return 0;
2689 }
2690
2691 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2692 {
2693         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2694 }
2695
2696 static ssize_t 
2697 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2698                    size_t count, loff_t *off)
2699 {
2700         int num;
2701         char buff[11];
2702
2703         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2704                 return -EACCES;
2705         num = (count < 10) ? count : 10;
2706         if (copy_from_user(buff, buffer, num))
2707                 return -EFAULT;
2708         buff[num] = '\0';
2709         sg_allow_dio = simple_strtoul(buff, NULL, 10) ? 1 : 0;
2710         return count;
2711 }
2712
2713 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2714 {
2715         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2716 }
2717
2718 static ssize_t 
2719 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2720                      size_t count, loff_t *off)
2721 {
2722         int num;
2723         unsigned long k = ULONG_MAX;
2724         char buff[11];
2725
2726         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2727                 return -EACCES;
2728         num = (count < 10) ? count : 10;
2729         if (copy_from_user(buff, buffer, num))
2730                 return -EFAULT;
2731         buff[num] = '\0';
2732         k = simple_strtoul(buff, NULL, 10);
2733         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2734                 sg_big_buff = k;
2735                 return count;
2736         }
2737         return -ERANGE;
2738 }
2739
2740 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2741 {
2742         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2743                    sg_version_date);
2744         return 0;
2745 }
2746
2747 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2748 {
2749         return single_open(file, sg_proc_seq_show_version, NULL);
2750 }
2751
2752 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2753 {
2754         seq_printf(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\t"
2755                    "online\n");
2756         return 0;
2757 }
2758
2759 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2760 {
2761         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2762 }
2763
2764 struct sg_proc_deviter {
2765         loff_t  index;
2766         size_t  max;
2767 };
2768
2769 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2770 {
2771         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2772
2773         s->private = it;
2774         if (! it)
2775                 return NULL;
2776
2777         it->index = *pos;
2778         it->max = sg_last_dev();
2779         if (it->index >= it->max)
2780                 return NULL;
2781         return it;
2782 }
2783
2784 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2785 {
2786         struct sg_proc_deviter * it = s->private;
2787
2788         *pos = ++it->index;
2789         return (it->index < it->max) ? it : NULL;
2790 }
2791
2792 static void dev_seq_stop(struct seq_file *s, void *v)
2793 {
2794         kfree(s->private);
2795 }
2796
2797 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2798 {
2799         return seq_open(file, &dev_seq_ops);
2800 }
2801
2802 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2803 {
2804         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2805         Sg_device *sdp;
2806         struct scsi_device *scsidp;
2807
2808         sdp = it ? sg_get_dev(it->index) : NULL;
2809         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2810                 seq_printf(s, "%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\n",
2811                               scsidp->host->host_no, scsidp->channel,
2812                               scsidp->id, scsidp->lun, (int) scsidp->type,
2813                               1,
2814                               (int) scsidp->queue_depth,
2815                               (int) scsidp->device_busy,
2816                               (int) scsi_device_online(scsidp));
2817         else
2818                 seq_printf(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2819         return 0;
2820 }
2821
2822 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2823 {
2824         return seq_open(file, &devstrs_seq_ops);
2825 }
2826
2827 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2828 {
2829         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2830         Sg_device *sdp;
2831         struct scsi_device *scsidp;
2832
2833         sdp = it ? sg_get_dev(it->index) : NULL;
2834         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2835                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2836                            scsidp->vendor, scsidp->model, scsidp->rev);
2837         else
2838                 seq_printf(s, "<no active device>\n");
2839         return 0;
2840 }
2841
2842 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2843 {
2844         int k, m, new_interface, blen, usg;
2845         Sg_request *srp;
2846         Sg_fd *fp;
2847         const sg_io_hdr_t *hp;
2848         const char * cp;
2849         unsigned int ms;
2850
2851         for (k = 0; (fp = sg_get_nth_sfp(sdp, k)); ++k) {
2852                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2853                            "(res)sgat=%d low_dma=%d\n", k + 1,
2854                            jiffies_to_msecs(fp->timeout),
2855                            fp->reserve.bufflen,
2856                            (int) fp->reserve.k_use_sg,
2857                            (int) fp->low_dma);
2858                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=%d\n",
2859                            (int) fp->cmd_q, (int) fp->force_packid,
2860                            (int) fp->keep_orphan, (int) fp->closed);
2861                 for (m = 0; (srp = sg_get_nth_request(fp, m)); ++m) {
2862                         hp = &srp->header;
2863                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2864                         if (srp->res_used) {
2865                                 if (new_interface && 
2866                                     (SG_FLAG_MMAP_IO & hp->flags))
2867                                         cp = "     mmap>> ";
2868                                 else
2869                                         cp = "     rb>> ";
2870                         } else {
2871                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2872                                         cp = "     dio>> ";
2873                                 else
2874                                         cp = "     ";
2875                         }
2876                         seq_printf(s, cp);
2877                         blen = srp->data.bufflen;
2878                         usg = srp->data.k_use_sg;
2879                         seq_printf(s, srp->done ? 
2880                                    ((1 == srp->done) ?  "rcv:" : "fin:")
2881                                    : "act:");
2882                         seq_printf(s, " id=%d blen=%d",
2883                                    srp->header.pack_id, blen);
2884                         if (srp->done)
2885                                 seq_printf(s, " dur=%d", hp->duration);
2886                         else {
2887                                 ms = jiffies_to_msecs(jiffies);
2888                                 seq_printf(s, " t_o/elap=%d/%d",
2889                                         (new_interface ? hp->timeout :
2890                                                   jiffies_to_msecs(fp->timeout)),
2891                                         (ms > hp->duration ? ms - hp->duration : 0));
2892                         }
2893                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2894                                    (int) srp->data.cmd_opcode);
2895                 }
2896                 if (0 == m)
2897                         seq_printf(s, "     No requests active\n");
2898         }
2899 }
2900
2901 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2902 {
2903         return seq_open(file, &debug_seq_ops);
2904 }
2905
2906 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2907 {
2908         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2909         Sg_device *sdp;
2910
2911         if (it && (0 == it->index)) {
2912                 seq_printf(s, "max_active_device=%d(origin 1)\n",
2913                            (int)it->max);
2914                 seq_printf(s, " def_reserved_size=%d\n", sg_big_buff);
2915         }
2916         sdp = it ? sg_get_dev(it->index) : NULL;
2917         if (sdp) {
2918                 struct scsi_device *scsidp = sdp->device;
2919
2920                 if (NULL == scsidp) {
2921                         seq_printf(s, "device %d detached ??\n", 
2922                                    (int)it->index);
2923                         return 0;
2924                 }
2925
2926                 if (sg_get_nth_sfp(sdp, 0)) {
2927                         seq_printf(s, " >>> device=%s ",
2928                                 sdp->disk->disk_name);
2929                         if (sdp->detached)
2930                                 seq_printf(s, "detached pending close ");
2931                         else
2932                                 seq_printf
2933                                     (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
2934                                      scsidp->host->host_no,
2935                                      scsidp->channel, scsidp->id,
2936                                      scsidp->lun,
2937                                      scsidp->host->hostt->emulated);
2938                         seq_printf(s, " sg_tablesize=%d excl=%d\n",
2939                                    sdp->sg_tablesize, sdp->exclude);
2940                 }
2941                 sg_proc_debug_helper(s, sdp);
2942         }
2943         return 0;
2944 }
2945
2946 #endif                          /* CONFIG_SCSI_PROC_FS */
2947
2948 module_init(init_sg);
2949 module_exit(exit_sg);