scsi: convert to idr_alloc()
[~shefty/rdma-dev.git] / 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/slab.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/poll.h>
45 #include <linux/moduleparam.h>
46 #include <linux/cdev.h>
47 #include <linux/idr.h>
48 #include <linux/seq_file.h>
49 #include <linux/blkdev.h>
50 #include <linux/delay.h>
51 #include <linux/blktrace_api.h>
52 #include <linux/mutex.h>
53 #include <linux/ratelimit.h>
54
55 #include "scsi.h"
56 #include <scsi/scsi_dbg.h>
57 #include <scsi/scsi_host.h>
58 #include <scsi/scsi_driver.h>
59 #include <scsi/scsi_ioctl.h>
60 #include <scsi/sg.h>
61
62 #include "scsi_logging.h"
63
64 #ifdef CONFIG_SCSI_PROC_FS
65 #include <linux/proc_fs.h>
66 static char *sg_version_date = "20061027";
67
68 static int sg_proc_init(void);
69 static void sg_proc_cleanup(void);
70 #endif
71
72 #define SG_ALLOW_DIO_DEF 0
73
74 #define SG_MAX_DEVS 32768
75
76 /*
77  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
78  * Then when using 32 bit integers x * m may overflow during the calculation.
79  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
80  * calculates the same, but prevents the overflow when both m and d
81  * are "small" numbers (like HZ and USER_HZ).
82  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
83  * in 32 bits.
84  */
85 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
86
87 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
88
89 int sg_big_buff = SG_DEF_RESERVED_SIZE;
90 /* N.B. This variable is readable and writeable via
91    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
92    of this size (or less if there is not enough memory) will be reserved
93    for use by this file descriptor. [Deprecated usage: this variable is also
94    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
95    the kernel (i.e. it is not a module).] */
96 static int def_reserved_size = -1;      /* picks up init parameter */
97 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
98
99 static int scatter_elem_sz = SG_SCATTER_SZ;
100 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
101
102 #define SG_SECTOR_SZ 512
103
104 static int sg_add(struct device *, struct class_interface *);
105 static void sg_remove(struct device *, struct class_interface *);
106
107 static DEFINE_SPINLOCK(sg_open_exclusive_lock);
108
109 static DEFINE_IDR(sg_index_idr);
110 static DEFINE_RWLOCK(sg_index_lock);    /* Also used to lock
111                                                            file descriptor list for device */
112
113 static struct class_interface sg_interface = {
114         .add_dev        = sg_add,
115         .remove_dev     = sg_remove,
116 };
117
118 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
119         unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
120         unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
121         unsigned bufflen;       /* Size of (aggregate) data buffer */
122         struct page **pages;
123         int page_order;
124         char dio_in_use;        /* 0->indirect IO (or mmap), 1->dio */
125         unsigned char cmd_opcode; /* first byte of command */
126 } Sg_scatter_hold;
127
128 struct sg_device;               /* forward declarations */
129 struct sg_fd;
130
131 typedef struct sg_request {     /* SG_MAX_QUEUE requests outstanding per file */
132         struct sg_request *nextrp;      /* NULL -> tail request (slist) */
133         struct sg_fd *parentfp; /* NULL -> not in use */
134         Sg_scatter_hold data;   /* hold buffer, perhaps scatter list */
135         sg_io_hdr_t header;     /* scsi command+info, see <scsi/sg.h> */
136         unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
137         char res_used;          /* 1 -> using reserve buffer, 0 -> not ... */
138         char orphan;            /* 1 -> drop on sight, 0 -> normal */
139         char sg_io_owned;       /* 1 -> packet belongs to SG_IO */
140         /* done protected by rq_list_lock */
141         char done;              /* 0->before bh, 1->before read, 2->read */
142         struct request *rq;
143         struct bio *bio;
144         struct execute_work ew;
145 } Sg_request;
146
147 typedef struct sg_fd {          /* holds the state of a file descriptor */
148         /* sfd_siblings is protected by sg_index_lock */
149         struct list_head sfd_siblings;
150         struct sg_device *parentdp;     /* owning device */
151         wait_queue_head_t read_wait;    /* queue read until command done */
152         rwlock_t rq_list_lock;  /* protect access to list in req_arr */
153         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
154         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
155         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
156         unsigned save_scat_len; /* original length of trunc. scat. element */
157         Sg_request *headrp;     /* head of request slist, NULL->empty */
158         struct fasync_struct *async_qp; /* used by asynchronous notification */
159         Sg_request req_arr[SG_MAX_QUEUE];       /* used as singly-linked list */
160         char low_dma;           /* as in parent but possibly overridden to 1 */
161         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
162         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
163         char next_cmd_len;      /* 0 -> automatic (def), >0 -> use on next write() */
164         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
165         char mmap_called;       /* 0 -> mmap() never called on this fd */
166         struct kref f_ref;
167         struct execute_work ew;
168 } Sg_fd;
169
170 typedef struct sg_device { /* holds the state of each scsi generic device */
171         struct scsi_device *device;
172         wait_queue_head_t o_excl_wait;  /* queue open() when O_EXCL in use */
173         int sg_tablesize;       /* adapter's max scatter-gather table size */
174         u32 index;              /* device index number */
175         /* sfds is protected by sg_index_lock */
176         struct list_head sfds;
177         volatile char detached; /* 0->attached, 1->detached pending removal */
178         /* exclude protected by sg_open_exclusive_lock */
179         char exclude;           /* opened for exclusive access */
180         char sgdebug;           /* 0->off, 1->sense, 9->dump dev, 10-> all devs */
181         struct gendisk *disk;
182         struct cdev * cdev;     /* char_dev [sysfs: /sys/cdev/major/sg<n>] */
183         struct kref d_ref;
184 } Sg_device;
185
186 /* tasklet or soft irq callback */
187 static void sg_rq_end_io(struct request *rq, int uptodate);
188 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
189 static int sg_finish_rem_req(Sg_request * srp);
190 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
191 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
192                            Sg_request * srp);
193 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
194                         const char __user *buf, size_t count, int blocking,
195                         int read_only, int sg_io_owned, Sg_request **o_srp);
196 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
197                            unsigned char *cmnd, int timeout, int blocking);
198 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
199 static void sg_remove_scat(Sg_scatter_hold * schp);
200 static void sg_build_reserve(Sg_fd * sfp, int req_size);
201 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
202 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
203 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
204 static void sg_remove_sfp(struct kref *);
205 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
206 static Sg_request *sg_add_request(Sg_fd * sfp);
207 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
208 static int sg_res_in_use(Sg_fd * sfp);
209 static Sg_device *sg_get_dev(int dev);
210 static void sg_put_dev(Sg_device *sdp);
211
212 #define SZ_SG_HEADER sizeof(struct sg_header)
213 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
214 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
215 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
216
217 static int sg_allow_access(struct file *filp, unsigned char *cmd)
218 {
219         struct sg_fd *sfp = filp->private_data;
220
221         if (sfp->parentdp->device->type == TYPE_SCANNER)
222                 return 0;
223
224         return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
225 }
226
227 static int get_exclude(Sg_device *sdp)
228 {
229         unsigned long flags;
230         int ret;
231
232         spin_lock_irqsave(&sg_open_exclusive_lock, flags);
233         ret = sdp->exclude;
234         spin_unlock_irqrestore(&sg_open_exclusive_lock, flags);
235         return ret;
236 }
237
238 static int set_exclude(Sg_device *sdp, char val)
239 {
240         unsigned long flags;
241
242         spin_lock_irqsave(&sg_open_exclusive_lock, flags);
243         sdp->exclude = val;
244         spin_unlock_irqrestore(&sg_open_exclusive_lock, flags);
245         return val;
246 }
247
248 static int sfds_list_empty(Sg_device *sdp)
249 {
250         unsigned long flags;
251         int ret;
252
253         read_lock_irqsave(&sg_index_lock, flags);
254         ret = list_empty(&sdp->sfds);
255         read_unlock_irqrestore(&sg_index_lock, flags);
256         return ret;
257 }
258
259 static int
260 sg_open(struct inode *inode, struct file *filp)
261 {
262         int dev = iminor(inode);
263         int flags = filp->f_flags;
264         struct request_queue *q;
265         Sg_device *sdp;
266         Sg_fd *sfp;
267         int res;
268         int retval;
269
270         nonseekable_open(inode, filp);
271         SCSI_LOG_TIMEOUT(3, printk("sg_open: dev=%d, flags=0x%x\n", dev, flags));
272         sdp = sg_get_dev(dev);
273         if (IS_ERR(sdp)) {
274                 retval = PTR_ERR(sdp);
275                 sdp = NULL;
276                 goto sg_put;
277         }
278
279         /* This driver's module count bumped by fops_get in <linux/fs.h> */
280         /* Prevent the device driver from vanishing while we sleep */
281         retval = scsi_device_get(sdp->device);
282         if (retval)
283                 goto sg_put;
284
285         retval = scsi_autopm_get_device(sdp->device);
286         if (retval)
287                 goto sdp_put;
288
289         if (!((flags & O_NONBLOCK) ||
290               scsi_block_when_processing_errors(sdp->device))) {
291                 retval = -ENXIO;
292                 /* we are in error recovery for this device */
293                 goto error_out;
294         }
295
296         if (flags & O_EXCL) {
297                 if (O_RDONLY == (flags & O_ACCMODE)) {
298                         retval = -EPERM; /* Can't lock it with read only access */
299                         goto error_out;
300                 }
301                 if (!sfds_list_empty(sdp) && (flags & O_NONBLOCK)) {
302                         retval = -EBUSY;
303                         goto error_out;
304                 }
305                 res = wait_event_interruptible(sdp->o_excl_wait,
306                                            ((!sfds_list_empty(sdp) || get_exclude(sdp)) ? 0 : set_exclude(sdp, 1)));
307                 if (res) {
308                         retval = res;   /* -ERESTARTSYS because signal hit process */
309                         goto error_out;
310                 }
311         } else if (get_exclude(sdp)) {  /* some other fd has an exclusive lock on dev */
312                 if (flags & O_NONBLOCK) {
313                         retval = -EBUSY;
314                         goto error_out;
315                 }
316                 res = wait_event_interruptible(sdp->o_excl_wait, !get_exclude(sdp));
317                 if (res) {
318                         retval = res;   /* -ERESTARTSYS because signal hit process */
319                         goto error_out;
320                 }
321         }
322         if (sdp->detached) {
323                 retval = -ENODEV;
324                 goto error_out;
325         }
326         if (sfds_list_empty(sdp)) {     /* no existing opens on this device */
327                 sdp->sgdebug = 0;
328                 q = sdp->device->request_queue;
329                 sdp->sg_tablesize = queue_max_segments(q);
330         }
331         if ((sfp = sg_add_sfp(sdp, dev)))
332                 filp->private_data = sfp;
333         else {
334                 if (flags & O_EXCL) {
335                         set_exclude(sdp, 0);    /* undo if error */
336                         wake_up_interruptible(&sdp->o_excl_wait);
337                 }
338                 retval = -ENOMEM;
339                 goto error_out;
340         }
341         retval = 0;
342 error_out:
343         if (retval) {
344                 scsi_autopm_put_device(sdp->device);
345 sdp_put:
346                 scsi_device_put(sdp->device);
347         }
348 sg_put:
349         if (sdp)
350                 sg_put_dev(sdp);
351         return retval;
352 }
353
354 /* Following function was formerly called 'sg_close' */
355 static int
356 sg_release(struct inode *inode, struct file *filp)
357 {
358         Sg_device *sdp;
359         Sg_fd *sfp;
360
361         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
362                 return -ENXIO;
363         SCSI_LOG_TIMEOUT(3, printk("sg_release: %s\n", sdp->disk->disk_name));
364
365         set_exclude(sdp, 0);
366         wake_up_interruptible(&sdp->o_excl_wait);
367
368         scsi_autopm_put_device(sdp->device);
369         kref_put(&sfp->f_ref, sg_remove_sfp);
370         return 0;
371 }
372
373 static ssize_t
374 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
375 {
376         Sg_device *sdp;
377         Sg_fd *sfp;
378         Sg_request *srp;
379         int req_pack_id = -1;
380         sg_io_hdr_t *hp;
381         struct sg_header *old_hdr = NULL;
382         int retval = 0;
383
384         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
385                 return -ENXIO;
386         SCSI_LOG_TIMEOUT(3, printk("sg_read: %s, count=%d\n",
387                                    sdp->disk->disk_name, (int) count));
388
389         if (!access_ok(VERIFY_WRITE, buf, count))
390                 return -EFAULT;
391         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
392                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
393                 if (!old_hdr)
394                         return -ENOMEM;
395                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
396                         retval = -EFAULT;
397                         goto free_old_hdr;
398                 }
399                 if (old_hdr->reply_len < 0) {
400                         if (count >= SZ_SG_IO_HDR) {
401                                 sg_io_hdr_t *new_hdr;
402                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
403                                 if (!new_hdr) {
404                                         retval = -ENOMEM;
405                                         goto free_old_hdr;
406                                 }
407                                 retval =__copy_from_user
408                                     (new_hdr, buf, SZ_SG_IO_HDR);
409                                 req_pack_id = new_hdr->pack_id;
410                                 kfree(new_hdr);
411                                 if (retval) {
412                                         retval = -EFAULT;
413                                         goto free_old_hdr;
414                                 }
415                         }
416                 } else
417                         req_pack_id = old_hdr->pack_id;
418         }
419         srp = sg_get_rq_mark(sfp, req_pack_id);
420         if (!srp) {             /* now wait on packet to arrive */
421                 if (sdp->detached) {
422                         retval = -ENODEV;
423                         goto free_old_hdr;
424                 }
425                 if (filp->f_flags & O_NONBLOCK) {
426                         retval = -EAGAIN;
427                         goto free_old_hdr;
428                 }
429                 retval = wait_event_interruptible(sfp->read_wait,
430                         (sdp->detached ||
431                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
432                 if (sdp->detached) {
433                         retval = -ENODEV;
434                         goto free_old_hdr;
435                 }
436                 if (retval) {
437                         /* -ERESTARTSYS as signal hit process */
438                         goto free_old_hdr;
439                 }
440         }
441         if (srp->header.interface_id != '\0') {
442                 retval = sg_new_read(sfp, buf, count, srp);
443                 goto free_old_hdr;
444         }
445
446         hp = &srp->header;
447         if (old_hdr == NULL) {
448                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
449                 if (! old_hdr) {
450                         retval = -ENOMEM;
451                         goto free_old_hdr;
452                 }
453         }
454         memset(old_hdr, 0, SZ_SG_HEADER);
455         old_hdr->reply_len = (int) hp->timeout;
456         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
457         old_hdr->pack_id = hp->pack_id;
458         old_hdr->twelve_byte =
459             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
460         old_hdr->target_status = hp->masked_status;
461         old_hdr->host_status = hp->host_status;
462         old_hdr->driver_status = hp->driver_status;
463         if ((CHECK_CONDITION & hp->masked_status) ||
464             (DRIVER_SENSE & hp->driver_status))
465                 memcpy(old_hdr->sense_buffer, srp->sense_b,
466                        sizeof (old_hdr->sense_buffer));
467         switch (hp->host_status) {
468         /* This setup of 'result' is for backward compatibility and is best
469            ignored by the user who should use target, host + driver status */
470         case DID_OK:
471         case DID_PASSTHROUGH:
472         case DID_SOFT_ERROR:
473                 old_hdr->result = 0;
474                 break;
475         case DID_NO_CONNECT:
476         case DID_BUS_BUSY:
477         case DID_TIME_OUT:
478                 old_hdr->result = EBUSY;
479                 break;
480         case DID_BAD_TARGET:
481         case DID_ABORT:
482         case DID_PARITY:
483         case DID_RESET:
484         case DID_BAD_INTR:
485                 old_hdr->result = EIO;
486                 break;
487         case DID_ERROR:
488                 old_hdr->result = (srp->sense_b[0] == 0 && 
489                                   hp->masked_status == GOOD) ? 0 : EIO;
490                 break;
491         default:
492                 old_hdr->result = EIO;
493                 break;
494         }
495
496         /* Now copy the result back to the user buffer.  */
497         if (count >= SZ_SG_HEADER) {
498                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
499                         retval = -EFAULT;
500                         goto free_old_hdr;
501                 }
502                 buf += SZ_SG_HEADER;
503                 if (count > old_hdr->reply_len)
504                         count = old_hdr->reply_len;
505                 if (count > SZ_SG_HEADER) {
506                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
507                                 retval = -EFAULT;
508                                 goto free_old_hdr;
509                         }
510                 }
511         } else
512                 count = (old_hdr->result == 0) ? 0 : -EIO;
513         sg_finish_rem_req(srp);
514         retval = count;
515 free_old_hdr:
516         kfree(old_hdr);
517         return retval;
518 }
519
520 static ssize_t
521 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
522 {
523         sg_io_hdr_t *hp = &srp->header;
524         int err = 0;
525         int len;
526
527         if (count < SZ_SG_IO_HDR) {
528                 err = -EINVAL;
529                 goto err_out;
530         }
531         hp->sb_len_wr = 0;
532         if ((hp->mx_sb_len > 0) && hp->sbp) {
533                 if ((CHECK_CONDITION & hp->masked_status) ||
534                     (DRIVER_SENSE & hp->driver_status)) {
535                         int sb_len = SCSI_SENSE_BUFFERSIZE;
536                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
537                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
538                         len = (len > sb_len) ? sb_len : len;
539                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
540                                 err = -EFAULT;
541                                 goto err_out;
542                         }
543                         hp->sb_len_wr = len;
544                 }
545         }
546         if (hp->masked_status || hp->host_status || hp->driver_status)
547                 hp->info |= SG_INFO_CHECK;
548         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
549                 err = -EFAULT;
550                 goto err_out;
551         }
552 err_out:
553         err = sg_finish_rem_req(srp);
554         return (0 == err) ? count : err;
555 }
556
557 static ssize_t
558 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
559 {
560         int mxsize, cmd_size, k;
561         int input_size, blocking;
562         unsigned char opcode;
563         Sg_device *sdp;
564         Sg_fd *sfp;
565         Sg_request *srp;
566         struct sg_header old_hdr;
567         sg_io_hdr_t *hp;
568         unsigned char cmnd[MAX_COMMAND_SIZE];
569
570         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
571                 return -ENXIO;
572         SCSI_LOG_TIMEOUT(3, printk("sg_write: %s, count=%d\n",
573                                    sdp->disk->disk_name, (int) count));
574         if (sdp->detached)
575                 return -ENODEV;
576         if (!((filp->f_flags & O_NONBLOCK) ||
577               scsi_block_when_processing_errors(sdp->device)))
578                 return -ENXIO;
579
580         if (!access_ok(VERIFY_READ, buf, count))
581                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
582         if (count < SZ_SG_HEADER)
583                 return -EIO;
584         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
585                 return -EFAULT;
586         blocking = !(filp->f_flags & O_NONBLOCK);
587         if (old_hdr.reply_len < 0)
588                 return sg_new_write(sfp, filp, buf, count,
589                                     blocking, 0, 0, NULL);
590         if (count < (SZ_SG_HEADER + 6))
591                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
592
593         if (!(srp = sg_add_request(sfp))) {
594                 SCSI_LOG_TIMEOUT(1, printk("sg_write: queue full\n"));
595                 return -EDOM;
596         }
597         buf += SZ_SG_HEADER;
598         __get_user(opcode, buf);
599         if (sfp->next_cmd_len > 0) {
600                 if (sfp->next_cmd_len > MAX_COMMAND_SIZE) {
601                         SCSI_LOG_TIMEOUT(1, printk("sg_write: command length too long\n"));
602                         sfp->next_cmd_len = 0;
603                         sg_remove_request(sfp, srp);
604                         return -EIO;
605                 }
606                 cmd_size = sfp->next_cmd_len;
607                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
608         } else {
609                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
610                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
611                         cmd_size = 12;
612         }
613         SCSI_LOG_TIMEOUT(4, printk(
614                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
615 /* Determine buffer size.  */
616         input_size = count - cmd_size;
617         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
618         mxsize -= SZ_SG_HEADER;
619         input_size -= SZ_SG_HEADER;
620         if (input_size < 0) {
621                 sg_remove_request(sfp, srp);
622                 return -EIO;    /* User did not pass enough bytes for this command. */
623         }
624         hp = &srp->header;
625         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
626         hp->cmd_len = (unsigned char) cmd_size;
627         hp->iovec_count = 0;
628         hp->mx_sb_len = 0;
629         if (input_size > 0)
630                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
631                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
632         else
633                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
634         hp->dxfer_len = mxsize;
635         if (hp->dxfer_direction == SG_DXFER_TO_DEV)
636                 hp->dxferp = (char __user *)buf + cmd_size;
637         else
638                 hp->dxferp = NULL;
639         hp->sbp = NULL;
640         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
641         hp->flags = input_size; /* structure abuse ... */
642         hp->pack_id = old_hdr.pack_id;
643         hp->usr_ptr = NULL;
644         if (__copy_from_user(cmnd, buf, cmd_size))
645                 return -EFAULT;
646         /*
647          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
648          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
649          * is a non-zero input_size, so emit a warning.
650          */
651         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
652                 static char cmd[TASK_COMM_LEN];
653                 if (strcmp(current->comm, cmd)) {
654                         printk_ratelimited(KERN_WARNING
655                                            "sg_write: data in/out %d/%d bytes "
656                                            "for SCSI command 0x%x-- guessing "
657                                            "data in;\n   program %s not setting "
658                                            "count and/or reply_len properly\n",
659                                            old_hdr.reply_len - (int)SZ_SG_HEADER,
660                                            input_size, (unsigned int) cmnd[0],
661                                            current->comm);
662                         strcpy(cmd, current->comm);
663                 }
664         }
665         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
666         return (k < 0) ? k : count;
667 }
668
669 static ssize_t
670 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
671                  size_t count, int blocking, int read_only, int sg_io_owned,
672                  Sg_request **o_srp)
673 {
674         int k;
675         Sg_request *srp;
676         sg_io_hdr_t *hp;
677         unsigned char cmnd[MAX_COMMAND_SIZE];
678         int timeout;
679         unsigned long ul_timeout;
680
681         if (count < SZ_SG_IO_HDR)
682                 return -EINVAL;
683         if (!access_ok(VERIFY_READ, buf, count))
684                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
685
686         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
687         if (!(srp = sg_add_request(sfp))) {
688                 SCSI_LOG_TIMEOUT(1, printk("sg_new_write: queue full\n"));
689                 return -EDOM;
690         }
691         srp->sg_io_owned = sg_io_owned;
692         hp = &srp->header;
693         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
694                 sg_remove_request(sfp, srp);
695                 return -EFAULT;
696         }
697         if (hp->interface_id != 'S') {
698                 sg_remove_request(sfp, srp);
699                 return -ENOSYS;
700         }
701         if (hp->flags & SG_FLAG_MMAP_IO) {
702                 if (hp->dxfer_len > sfp->reserve.bufflen) {
703                         sg_remove_request(sfp, srp);
704                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
705                 }
706                 if (hp->flags & SG_FLAG_DIRECT_IO) {
707                         sg_remove_request(sfp, srp);
708                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
709                 }
710                 if (sg_res_in_use(sfp)) {
711                         sg_remove_request(sfp, srp);
712                         return -EBUSY;  /* reserve buffer already being used */
713                 }
714         }
715         ul_timeout = msecs_to_jiffies(srp->header.timeout);
716         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
717         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
718                 sg_remove_request(sfp, srp);
719                 return -EMSGSIZE;
720         }
721         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
722                 sg_remove_request(sfp, srp);
723                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
724         }
725         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
726                 sg_remove_request(sfp, srp);
727                 return -EFAULT;
728         }
729         if (read_only && sg_allow_access(file, cmnd)) {
730                 sg_remove_request(sfp, srp);
731                 return -EPERM;
732         }
733         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
734         if (k < 0)
735                 return k;
736         if (o_srp)
737                 *o_srp = srp;
738         return count;
739 }
740
741 static int
742 sg_common_write(Sg_fd * sfp, Sg_request * srp,
743                 unsigned char *cmnd, int timeout, int blocking)
744 {
745         int k, data_dir;
746         Sg_device *sdp = sfp->parentdp;
747         sg_io_hdr_t *hp = &srp->header;
748
749         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
750         hp->status = 0;
751         hp->masked_status = 0;
752         hp->msg_status = 0;
753         hp->info = 0;
754         hp->host_status = 0;
755         hp->driver_status = 0;
756         hp->resid = 0;
757         SCSI_LOG_TIMEOUT(4, printk("sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
758                           (int) cmnd[0], (int) hp->cmd_len));
759
760         k = sg_start_req(srp, cmnd);
761         if (k) {
762                 SCSI_LOG_TIMEOUT(1, printk("sg_common_write: start_req err=%d\n", k));
763                 sg_finish_rem_req(srp);
764                 return k;       /* probably out of space --> ENOMEM */
765         }
766         if (sdp->detached) {
767                 if (srp->bio)
768                         blk_end_request_all(srp->rq, -EIO);
769                 sg_finish_rem_req(srp);
770                 return -ENODEV;
771         }
772
773         switch (hp->dxfer_direction) {
774         case SG_DXFER_TO_FROM_DEV:
775         case SG_DXFER_FROM_DEV:
776                 data_dir = DMA_FROM_DEVICE;
777                 break;
778         case SG_DXFER_TO_DEV:
779                 data_dir = DMA_TO_DEVICE;
780                 break;
781         case SG_DXFER_UNKNOWN:
782                 data_dir = DMA_BIDIRECTIONAL;
783                 break;
784         default:
785                 data_dir = DMA_NONE;
786                 break;
787         }
788         hp->duration = jiffies_to_msecs(jiffies);
789
790         srp->rq->timeout = timeout;
791         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
792         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
793                               srp->rq, 1, sg_rq_end_io);
794         return 0;
795 }
796
797 static int srp_done(Sg_fd *sfp, Sg_request *srp)
798 {
799         unsigned long flags;
800         int ret;
801
802         read_lock_irqsave(&sfp->rq_list_lock, flags);
803         ret = srp->done;
804         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
805         return ret;
806 }
807
808 static long
809 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
810 {
811         void __user *p = (void __user *)arg;
812         int __user *ip = p;
813         int result, val, read_only;
814         Sg_device *sdp;
815         Sg_fd *sfp;
816         Sg_request *srp;
817         unsigned long iflags;
818
819         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
820                 return -ENXIO;
821
822         SCSI_LOG_TIMEOUT(3, printk("sg_ioctl: %s, cmd=0x%x\n",
823                                    sdp->disk->disk_name, (int) cmd_in));
824         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
825
826         switch (cmd_in) {
827         case SG_IO:
828                 if (sdp->detached)
829                         return -ENODEV;
830                 if (!scsi_block_when_processing_errors(sdp->device))
831                         return -ENXIO;
832                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
833                         return -EFAULT;
834                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
835                                  1, read_only, 1, &srp);
836                 if (result < 0)
837                         return result;
838                 result = wait_event_interruptible(sfp->read_wait,
839                         (srp_done(sfp, srp) || sdp->detached));
840                 if (sdp->detached)
841                         return -ENODEV;
842                 write_lock_irq(&sfp->rq_list_lock);
843                 if (srp->done) {
844                         srp->done = 2;
845                         write_unlock_irq(&sfp->rq_list_lock);
846                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
847                         return (result < 0) ? result : 0;
848                 }
849                 srp->orphan = 1;
850                 write_unlock_irq(&sfp->rq_list_lock);
851                 return result;  /* -ERESTARTSYS because signal hit process */
852         case SG_SET_TIMEOUT:
853                 result = get_user(val, ip);
854                 if (result)
855                         return result;
856                 if (val < 0)
857                         return -EIO;
858                 if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
859                     val = MULDIV (INT_MAX, USER_HZ, HZ);
860                 sfp->timeout_user = val;
861                 sfp->timeout = MULDIV (val, HZ, USER_HZ);
862
863                 return 0;
864         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
865                                 /* strange ..., for backward compatibility */
866                 return sfp->timeout_user;
867         case SG_SET_FORCE_LOW_DMA:
868                 result = get_user(val, ip);
869                 if (result)
870                         return result;
871                 if (val) {
872                         sfp->low_dma = 1;
873                         if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
874                                 val = (int) sfp->reserve.bufflen;
875                                 sg_remove_scat(&sfp->reserve);
876                                 sg_build_reserve(sfp, val);
877                         }
878                 } else {
879                         if (sdp->detached)
880                                 return -ENODEV;
881                         sfp->low_dma = sdp->device->host->unchecked_isa_dma;
882                 }
883                 return 0;
884         case SG_GET_LOW_DMA:
885                 return put_user((int) sfp->low_dma, ip);
886         case SG_GET_SCSI_ID:
887                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
888                         return -EFAULT;
889                 else {
890                         sg_scsi_id_t __user *sg_idp = p;
891
892                         if (sdp->detached)
893                                 return -ENODEV;
894                         __put_user((int) sdp->device->host->host_no,
895                                    &sg_idp->host_no);
896                         __put_user((int) sdp->device->channel,
897                                    &sg_idp->channel);
898                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
899                         __put_user((int) sdp->device->lun, &sg_idp->lun);
900                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
901                         __put_user((short) sdp->device->host->cmd_per_lun,
902                                    &sg_idp->h_cmd_per_lun);
903                         __put_user((short) sdp->device->queue_depth,
904                                    &sg_idp->d_queue_depth);
905                         __put_user(0, &sg_idp->unused[0]);
906                         __put_user(0, &sg_idp->unused[1]);
907                         return 0;
908                 }
909         case SG_SET_FORCE_PACK_ID:
910                 result = get_user(val, ip);
911                 if (result)
912                         return result;
913                 sfp->force_packid = val ? 1 : 0;
914                 return 0;
915         case SG_GET_PACK_ID:
916                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
917                         return -EFAULT;
918                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
919                 for (srp = sfp->headrp; srp; srp = srp->nextrp) {
920                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
921                                 read_unlock_irqrestore(&sfp->rq_list_lock,
922                                                        iflags);
923                                 __put_user(srp->header.pack_id, ip);
924                                 return 0;
925                         }
926                 }
927                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
928                 __put_user(-1, ip);
929                 return 0;
930         case SG_GET_NUM_WAITING:
931                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
932                 for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
933                         if ((1 == srp->done) && (!srp->sg_io_owned))
934                                 ++val;
935                 }
936                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
937                 return put_user(val, ip);
938         case SG_GET_SG_TABLESIZE:
939                 return put_user(sdp->sg_tablesize, ip);
940         case SG_SET_RESERVED_SIZE:
941                 result = get_user(val, ip);
942                 if (result)
943                         return result;
944                 if (val < 0)
945                         return -EINVAL;
946                 val = min_t(int, val,
947                             queue_max_sectors(sdp->device->request_queue) * 512);
948                 if (val != sfp->reserve.bufflen) {
949                         if (sg_res_in_use(sfp) || sfp->mmap_called)
950                                 return -EBUSY;
951                         sg_remove_scat(&sfp->reserve);
952                         sg_build_reserve(sfp, val);
953                 }
954                 return 0;
955         case SG_GET_RESERVED_SIZE:
956                 val = min_t(int, sfp->reserve.bufflen,
957                             queue_max_sectors(sdp->device->request_queue) * 512);
958                 return put_user(val, ip);
959         case SG_SET_COMMAND_Q:
960                 result = get_user(val, ip);
961                 if (result)
962                         return result;
963                 sfp->cmd_q = val ? 1 : 0;
964                 return 0;
965         case SG_GET_COMMAND_Q:
966                 return put_user((int) sfp->cmd_q, ip);
967         case SG_SET_KEEP_ORPHAN:
968                 result = get_user(val, ip);
969                 if (result)
970                         return result;
971                 sfp->keep_orphan = val;
972                 return 0;
973         case SG_GET_KEEP_ORPHAN:
974                 return put_user((int) sfp->keep_orphan, ip);
975         case SG_NEXT_CMD_LEN:
976                 result = get_user(val, ip);
977                 if (result)
978                         return result;
979                 sfp->next_cmd_len = (val > 0) ? val : 0;
980                 return 0;
981         case SG_GET_VERSION_NUM:
982                 return put_user(sg_version_num, ip);
983         case SG_GET_ACCESS_COUNT:
984                 /* faked - we don't have a real access count anymore */
985                 val = (sdp->device ? 1 : 0);
986                 return put_user(val, ip);
987         case SG_GET_REQUEST_TABLE:
988                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
989                         return -EFAULT;
990                 else {
991                         sg_req_info_t *rinfo;
992                         unsigned int ms;
993
994                         rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
995                                                                 GFP_KERNEL);
996                         if (!rinfo)
997                                 return -ENOMEM;
998                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
999                         for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
1000                              ++val, srp = srp ? srp->nextrp : srp) {
1001                                 memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
1002                                 if (srp) {
1003                                         rinfo[val].req_state = srp->done + 1;
1004                                         rinfo[val].problem =
1005                                             srp->header.masked_status & 
1006                                             srp->header.host_status & 
1007                                             srp->header.driver_status;
1008                                         if (srp->done)
1009                                                 rinfo[val].duration =
1010                                                         srp->header.duration;
1011                                         else {
1012                                                 ms = jiffies_to_msecs(jiffies);
1013                                                 rinfo[val].duration =
1014                                                     (ms > srp->header.duration) ?
1015                                                     (ms - srp->header.duration) : 0;
1016                                         }
1017                                         rinfo[val].orphan = srp->orphan;
1018                                         rinfo[val].sg_io_owned =
1019                                                         srp->sg_io_owned;
1020                                         rinfo[val].pack_id =
1021                                                         srp->header.pack_id;
1022                                         rinfo[val].usr_ptr =
1023                                                         srp->header.usr_ptr;
1024                                 }
1025                         }
1026                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1027                         result = __copy_to_user(p, rinfo, 
1028                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1029                         result = result ? -EFAULT : 0;
1030                         kfree(rinfo);
1031                         return result;
1032                 }
1033         case SG_EMULATED_HOST:
1034                 if (sdp->detached)
1035                         return -ENODEV;
1036                 return put_user(sdp->device->host->hostt->emulated, ip);
1037         case SG_SCSI_RESET:
1038                 if (sdp->detached)
1039                         return -ENODEV;
1040                 if (filp->f_flags & O_NONBLOCK) {
1041                         if (scsi_host_in_recovery(sdp->device->host))
1042                                 return -EBUSY;
1043                 } else if (!scsi_block_when_processing_errors(sdp->device))
1044                         return -EBUSY;
1045                 result = get_user(val, ip);
1046                 if (result)
1047                         return result;
1048                 if (SG_SCSI_RESET_NOTHING == val)
1049                         return 0;
1050                 switch (val) {
1051                 case SG_SCSI_RESET_DEVICE:
1052                         val = SCSI_TRY_RESET_DEVICE;
1053                         break;
1054                 case SG_SCSI_RESET_TARGET:
1055                         val = SCSI_TRY_RESET_TARGET;
1056                         break;
1057                 case SG_SCSI_RESET_BUS:
1058                         val = SCSI_TRY_RESET_BUS;
1059                         break;
1060                 case SG_SCSI_RESET_HOST:
1061                         val = SCSI_TRY_RESET_HOST;
1062                         break;
1063                 default:
1064                         return -EINVAL;
1065                 }
1066                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1067                         return -EACCES;
1068                 return (scsi_reset_provider(sdp->device, val) ==
1069                         SUCCESS) ? 0 : -EIO;
1070         case SCSI_IOCTL_SEND_COMMAND:
1071                 if (sdp->detached)
1072                         return -ENODEV;
1073                 if (read_only) {
1074                         unsigned char opcode = WRITE_6;
1075                         Scsi_Ioctl_Command __user *siocp = p;
1076
1077                         if (copy_from_user(&opcode, siocp->data, 1))
1078                                 return -EFAULT;
1079                         if (sg_allow_access(filp, &opcode))
1080                                 return -EPERM;
1081                 }
1082                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1083         case SG_SET_DEBUG:
1084                 result = get_user(val, ip);
1085                 if (result)
1086                         return result;
1087                 sdp->sgdebug = (char) val;
1088                 return 0;
1089         case SCSI_IOCTL_GET_IDLUN:
1090         case SCSI_IOCTL_GET_BUS_NUMBER:
1091         case SCSI_IOCTL_PROBE_HOST:
1092         case SG_GET_TRANSFORM:
1093                 if (sdp->detached)
1094                         return -ENODEV;
1095                 return scsi_ioctl(sdp->device, cmd_in, p);
1096         case BLKSECTGET:
1097                 return put_user(queue_max_sectors(sdp->device->request_queue) * 512,
1098                                 ip);
1099         case BLKTRACESETUP:
1100                 return blk_trace_setup(sdp->device->request_queue,
1101                                        sdp->disk->disk_name,
1102                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1103                                        NULL,
1104                                        (char *)arg);
1105         case BLKTRACESTART:
1106                 return blk_trace_startstop(sdp->device->request_queue, 1);
1107         case BLKTRACESTOP:
1108                 return blk_trace_startstop(sdp->device->request_queue, 0);
1109         case BLKTRACETEARDOWN:
1110                 return blk_trace_remove(sdp->device->request_queue);
1111         default:
1112                 if (read_only)
1113                         return -EPERM;  /* don't know so take safe approach */
1114                 return scsi_ioctl(sdp->device, cmd_in, p);
1115         }
1116 }
1117
1118 #ifdef CONFIG_COMPAT
1119 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1120 {
1121         Sg_device *sdp;
1122         Sg_fd *sfp;
1123         struct scsi_device *sdev;
1124
1125         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1126                 return -ENXIO;
1127
1128         sdev = sdp->device;
1129         if (sdev->host->hostt->compat_ioctl) { 
1130                 int ret;
1131
1132                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1133
1134                 return ret;
1135         }
1136         
1137         return -ENOIOCTLCMD;
1138 }
1139 #endif
1140
1141 static unsigned int
1142 sg_poll(struct file *filp, poll_table * wait)
1143 {
1144         unsigned int res = 0;
1145         Sg_device *sdp;
1146         Sg_fd *sfp;
1147         Sg_request *srp;
1148         int count = 0;
1149         unsigned long iflags;
1150
1151         sfp = filp->private_data;
1152         if (!sfp)
1153                 return POLLERR;
1154         sdp = sfp->parentdp;
1155         if (!sdp)
1156                 return POLLERR;
1157         poll_wait(filp, &sfp->read_wait, wait);
1158         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1159         for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1160                 /* if any read waiting, flag it */
1161                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1162                         res = POLLIN | POLLRDNORM;
1163                 ++count;
1164         }
1165         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1166
1167         if (sdp->detached)
1168                 res |= POLLHUP;
1169         else if (!sfp->cmd_q) {
1170                 if (0 == count)
1171                         res |= POLLOUT | POLLWRNORM;
1172         } else if (count < SG_MAX_QUEUE)
1173                 res |= POLLOUT | POLLWRNORM;
1174         SCSI_LOG_TIMEOUT(3, printk("sg_poll: %s, res=0x%x\n",
1175                                    sdp->disk->disk_name, (int) res));
1176         return res;
1177 }
1178
1179 static int
1180 sg_fasync(int fd, struct file *filp, int mode)
1181 {
1182         Sg_device *sdp;
1183         Sg_fd *sfp;
1184
1185         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1186                 return -ENXIO;
1187         SCSI_LOG_TIMEOUT(3, printk("sg_fasync: %s, mode=%d\n",
1188                                    sdp->disk->disk_name, mode));
1189
1190         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1191 }
1192
1193 static int
1194 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1195 {
1196         Sg_fd *sfp;
1197         unsigned long offset, len, sa;
1198         Sg_scatter_hold *rsv_schp;
1199         int k, length;
1200
1201         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1202                 return VM_FAULT_SIGBUS;
1203         rsv_schp = &sfp->reserve;
1204         offset = vmf->pgoff << PAGE_SHIFT;
1205         if (offset >= rsv_schp->bufflen)
1206                 return VM_FAULT_SIGBUS;
1207         SCSI_LOG_TIMEOUT(3, printk("sg_vma_fault: offset=%lu, scatg=%d\n",
1208                                    offset, rsv_schp->k_use_sg));
1209         sa = vma->vm_start;
1210         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1211         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1212                 len = vma->vm_end - sa;
1213                 len = (len < length) ? len : length;
1214                 if (offset < len) {
1215                         struct page *page = nth_page(rsv_schp->pages[k],
1216                                                      offset >> PAGE_SHIFT);
1217                         get_page(page); /* increment page count */
1218                         vmf->page = page;
1219                         return 0; /* success */
1220                 }
1221                 sa += len;
1222                 offset -= len;
1223         }
1224
1225         return VM_FAULT_SIGBUS;
1226 }
1227
1228 static const struct vm_operations_struct sg_mmap_vm_ops = {
1229         .fault = sg_vma_fault,
1230 };
1231
1232 static int
1233 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1234 {
1235         Sg_fd *sfp;
1236         unsigned long req_sz, len, sa;
1237         Sg_scatter_hold *rsv_schp;
1238         int k, length;
1239
1240         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1241                 return -ENXIO;
1242         req_sz = vma->vm_end - vma->vm_start;
1243         SCSI_LOG_TIMEOUT(3, printk("sg_mmap starting, vm_start=%p, len=%d\n",
1244                                    (void *) vma->vm_start, (int) req_sz));
1245         if (vma->vm_pgoff)
1246                 return -EINVAL; /* want no offset */
1247         rsv_schp = &sfp->reserve;
1248         if (req_sz > rsv_schp->bufflen)
1249                 return -ENOMEM; /* cannot map more than reserved buffer */
1250
1251         sa = vma->vm_start;
1252         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1253         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1254                 len = vma->vm_end - sa;
1255                 len = (len < length) ? len : length;
1256                 sa += len;
1257         }
1258
1259         sfp->mmap_called = 1;
1260         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1261         vma->vm_private_data = sfp;
1262         vma->vm_ops = &sg_mmap_vm_ops;
1263         return 0;
1264 }
1265
1266 static void sg_rq_end_io_usercontext(struct work_struct *work)
1267 {
1268         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1269         struct sg_fd *sfp = srp->parentfp;
1270
1271         sg_finish_rem_req(srp);
1272         kref_put(&sfp->f_ref, sg_remove_sfp);
1273 }
1274
1275 /*
1276  * This function is a "bottom half" handler that is called by the mid
1277  * level when a command is completed (or has failed).
1278  */
1279 static void sg_rq_end_io(struct request *rq, int uptodate)
1280 {
1281         struct sg_request *srp = rq->end_io_data;
1282         Sg_device *sdp;
1283         Sg_fd *sfp;
1284         unsigned long iflags;
1285         unsigned int ms;
1286         char *sense;
1287         int result, resid, done = 1;
1288
1289         if (WARN_ON(srp->done != 0))
1290                 return;
1291
1292         sfp = srp->parentfp;
1293         if (WARN_ON(sfp == NULL))
1294                 return;
1295
1296         sdp = sfp->parentdp;
1297         if (unlikely(sdp->detached))
1298                 printk(KERN_INFO "sg_rq_end_io: device detached\n");
1299
1300         sense = rq->sense;
1301         result = rq->errors;
1302         resid = rq->resid_len;
1303
1304         SCSI_LOG_TIMEOUT(4, printk("sg_cmd_done: %s, pack_id=%d, res=0x%x\n",
1305                 sdp->disk->disk_name, srp->header.pack_id, result));
1306         srp->header.resid = resid;
1307         ms = jiffies_to_msecs(jiffies);
1308         srp->header.duration = (ms > srp->header.duration) ?
1309                                 (ms - srp->header.duration) : 0;
1310         if (0 != result) {
1311                 struct scsi_sense_hdr sshdr;
1312
1313                 srp->header.status = 0xff & result;
1314                 srp->header.masked_status = status_byte(result);
1315                 srp->header.msg_status = msg_byte(result);
1316                 srp->header.host_status = host_byte(result);
1317                 srp->header.driver_status = driver_byte(result);
1318                 if ((sdp->sgdebug > 0) &&
1319                     ((CHECK_CONDITION == srp->header.masked_status) ||
1320                      (COMMAND_TERMINATED == srp->header.masked_status)))
1321                         __scsi_print_sense("sg_cmd_done", sense,
1322                                            SCSI_SENSE_BUFFERSIZE);
1323
1324                 /* Following if statement is a patch supplied by Eric Youngdale */
1325                 if (driver_byte(result) != 0
1326                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1327                     && !scsi_sense_is_deferred(&sshdr)
1328                     && sshdr.sense_key == UNIT_ATTENTION
1329                     && sdp->device->removable) {
1330                         /* Detected possible disc change. Set the bit - this */
1331                         /* may be used if there are filesystems using this device */
1332                         sdp->device->changed = 1;
1333                 }
1334         }
1335         /* Rely on write phase to clean out srp status values, so no "else" */
1336
1337         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1338         if (unlikely(srp->orphan)) {
1339                 if (sfp->keep_orphan)
1340                         srp->sg_io_owned = 0;
1341                 else
1342                         done = 0;
1343         }
1344         srp->done = done;
1345         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1346
1347         if (likely(done)) {
1348                 /* Now wake up any sg_read() that is waiting for this
1349                  * packet.
1350                  */
1351                 wake_up_interruptible(&sfp->read_wait);
1352                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1353                 kref_put(&sfp->f_ref, sg_remove_sfp);
1354         } else {
1355                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1356                 schedule_work(&srp->ew.work);
1357         }
1358 }
1359
1360 static const struct file_operations sg_fops = {
1361         .owner = THIS_MODULE,
1362         .read = sg_read,
1363         .write = sg_write,
1364         .poll = sg_poll,
1365         .unlocked_ioctl = sg_ioctl,
1366 #ifdef CONFIG_COMPAT
1367         .compat_ioctl = sg_compat_ioctl,
1368 #endif
1369         .open = sg_open,
1370         .mmap = sg_mmap,
1371         .release = sg_release,
1372         .fasync = sg_fasync,
1373         .llseek = no_llseek,
1374 };
1375
1376 static struct class *sg_sysfs_class;
1377
1378 static int sg_sysfs_valid = 0;
1379
1380 static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1381 {
1382         struct request_queue *q = scsidp->request_queue;
1383         Sg_device *sdp;
1384         unsigned long iflags;
1385         int error;
1386         u32 k;
1387
1388         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1389         if (!sdp) {
1390                 printk(KERN_WARNING "kmalloc Sg_device failure\n");
1391                 return ERR_PTR(-ENOMEM);
1392         }
1393
1394         idr_preload(GFP_KERNEL);
1395         write_lock_irqsave(&sg_index_lock, iflags);
1396
1397         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1398         if (error < 0) {
1399                 if (error == -ENOSPC) {
1400                         sdev_printk(KERN_WARNING, scsidp,
1401                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1402                                     scsidp->type, SG_MAX_DEVS - 1);
1403                         error = -ENODEV;
1404                 } else {
1405                         printk(KERN_WARNING
1406                                "idr allocation Sg_device failure: %d\n", error);
1407                 }
1408                 goto out_unlock;
1409         }
1410         k = error;
1411
1412         SCSI_LOG_TIMEOUT(3, printk("sg_alloc: dev=%d \n", k));
1413         sprintf(disk->disk_name, "sg%d", k);
1414         disk->first_minor = k;
1415         sdp->disk = disk;
1416         sdp->device = scsidp;
1417         INIT_LIST_HEAD(&sdp->sfds);
1418         init_waitqueue_head(&sdp->o_excl_wait);
1419         sdp->sg_tablesize = queue_max_segments(q);
1420         sdp->index = k;
1421         kref_init(&sdp->d_ref);
1422         error = 0;
1423
1424 out_unlock:
1425         write_unlock_irqrestore(&sg_index_lock, iflags);
1426         idr_preload_end();
1427
1428         if (error) {
1429                 kfree(sdp);
1430                 return ERR_PTR(error);
1431         }
1432         return sdp;
1433 }
1434
1435 static int
1436 sg_add(struct device *cl_dev, struct class_interface *cl_intf)
1437 {
1438         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1439         struct gendisk *disk;
1440         Sg_device *sdp = NULL;
1441         struct cdev * cdev = NULL;
1442         int error;
1443         unsigned long iflags;
1444
1445         disk = alloc_disk(1);
1446         if (!disk) {
1447                 printk(KERN_WARNING "alloc_disk failed\n");
1448                 return -ENOMEM;
1449         }
1450         disk->major = SCSI_GENERIC_MAJOR;
1451
1452         error = -ENOMEM;
1453         cdev = cdev_alloc();
1454         if (!cdev) {
1455                 printk(KERN_WARNING "cdev_alloc failed\n");
1456                 goto out;
1457         }
1458         cdev->owner = THIS_MODULE;
1459         cdev->ops = &sg_fops;
1460
1461         sdp = sg_alloc(disk, scsidp);
1462         if (IS_ERR(sdp)) {
1463                 printk(KERN_WARNING "sg_alloc failed\n");
1464                 error = PTR_ERR(sdp);
1465                 goto out;
1466         }
1467
1468         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1469         if (error)
1470                 goto cdev_add_err;
1471
1472         sdp->cdev = cdev;
1473         if (sg_sysfs_valid) {
1474                 struct device *sg_class_member;
1475
1476                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1477                                                 MKDEV(SCSI_GENERIC_MAJOR,
1478                                                       sdp->index),
1479                                                 sdp, "%s", disk->disk_name);
1480                 if (IS_ERR(sg_class_member)) {
1481                         printk(KERN_ERR "sg_add: "
1482                                "device_create failed\n");
1483                         error = PTR_ERR(sg_class_member);
1484                         goto cdev_add_err;
1485                 }
1486                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1487                                           &sg_class_member->kobj, "generic");
1488                 if (error)
1489                         printk(KERN_ERR "sg_add: unable to make symlink "
1490                                         "'generic' back to sg%d\n", sdp->index);
1491         } else
1492                 printk(KERN_WARNING "sg_add: sg_sys Invalid\n");
1493
1494         sdev_printk(KERN_NOTICE, scsidp,
1495                     "Attached scsi generic sg%d type %d\n", sdp->index,
1496                     scsidp->type);
1497
1498         dev_set_drvdata(cl_dev, sdp);
1499
1500         return 0;
1501
1502 cdev_add_err:
1503         write_lock_irqsave(&sg_index_lock, iflags);
1504         idr_remove(&sg_index_idr, sdp->index);
1505         write_unlock_irqrestore(&sg_index_lock, iflags);
1506         kfree(sdp);
1507
1508 out:
1509         put_disk(disk);
1510         if (cdev)
1511                 cdev_del(cdev);
1512         return error;
1513 }
1514
1515 static void sg_device_destroy(struct kref *kref)
1516 {
1517         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1518         unsigned long flags;
1519
1520         /* CAUTION!  Note that the device can still be found via idr_find()
1521          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1522          * any other cleanup.
1523          */
1524
1525         write_lock_irqsave(&sg_index_lock, flags);
1526         idr_remove(&sg_index_idr, sdp->index);
1527         write_unlock_irqrestore(&sg_index_lock, flags);
1528
1529         SCSI_LOG_TIMEOUT(3,
1530                 printk("sg_device_destroy: %s\n",
1531                         sdp->disk->disk_name));
1532
1533         put_disk(sdp->disk);
1534         kfree(sdp);
1535 }
1536
1537 static void sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
1538 {
1539         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1540         Sg_device *sdp = dev_get_drvdata(cl_dev);
1541         unsigned long iflags;
1542         Sg_fd *sfp;
1543
1544         if (!sdp || sdp->detached)
1545                 return;
1546
1547         SCSI_LOG_TIMEOUT(3, printk("sg_remove: %s\n", sdp->disk->disk_name));
1548
1549         /* Need a write lock to set sdp->detached. */
1550         write_lock_irqsave(&sg_index_lock, iflags);
1551         sdp->detached = 1;
1552         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1553                 wake_up_interruptible(&sfp->read_wait);
1554                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1555         }
1556         write_unlock_irqrestore(&sg_index_lock, iflags);
1557
1558         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1559         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1560         cdev_del(sdp->cdev);
1561         sdp->cdev = NULL;
1562
1563         sg_put_dev(sdp);
1564 }
1565
1566 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1567 module_param_named(def_reserved_size, def_reserved_size, int,
1568                    S_IRUGO | S_IWUSR);
1569 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1570
1571 MODULE_AUTHOR("Douglas Gilbert");
1572 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1573 MODULE_LICENSE("GPL");
1574 MODULE_VERSION(SG_VERSION_STR);
1575 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1576
1577 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1578                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1579 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1580 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1581
1582 static int __init
1583 init_sg(void)
1584 {
1585         int rc;
1586
1587         if (scatter_elem_sz < PAGE_SIZE) {
1588                 scatter_elem_sz = PAGE_SIZE;
1589                 scatter_elem_sz_prev = scatter_elem_sz;
1590         }
1591         if (def_reserved_size >= 0)
1592                 sg_big_buff = def_reserved_size;
1593         else
1594                 def_reserved_size = sg_big_buff;
1595
1596         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), 
1597                                     SG_MAX_DEVS, "sg");
1598         if (rc)
1599                 return rc;
1600         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1601         if ( IS_ERR(sg_sysfs_class) ) {
1602                 rc = PTR_ERR(sg_sysfs_class);
1603                 goto err_out;
1604         }
1605         sg_sysfs_valid = 1;
1606         rc = scsi_register_interface(&sg_interface);
1607         if (0 == rc) {
1608 #ifdef CONFIG_SCSI_PROC_FS
1609                 sg_proc_init();
1610 #endif                          /* CONFIG_SCSI_PROC_FS */
1611                 return 0;
1612         }
1613         class_destroy(sg_sysfs_class);
1614 err_out:
1615         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1616         return rc;
1617 }
1618
1619 static void __exit
1620 exit_sg(void)
1621 {
1622 #ifdef CONFIG_SCSI_PROC_FS
1623         sg_proc_cleanup();
1624 #endif                          /* CONFIG_SCSI_PROC_FS */
1625         scsi_unregister_interface(&sg_interface);
1626         class_destroy(sg_sysfs_class);
1627         sg_sysfs_valid = 0;
1628         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1629                                  SG_MAX_DEVS);
1630         idr_destroy(&sg_index_idr);
1631 }
1632
1633 static int sg_start_req(Sg_request *srp, unsigned char *cmd)
1634 {
1635         int res;
1636         struct request *rq;
1637         Sg_fd *sfp = srp->parentfp;
1638         sg_io_hdr_t *hp = &srp->header;
1639         int dxfer_len = (int) hp->dxfer_len;
1640         int dxfer_dir = hp->dxfer_direction;
1641         unsigned int iov_count = hp->iovec_count;
1642         Sg_scatter_hold *req_schp = &srp->data;
1643         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1644         struct request_queue *q = sfp->parentdp->device->request_queue;
1645         struct rq_map_data *md, map_data;
1646         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1647
1648         SCSI_LOG_TIMEOUT(4, printk(KERN_INFO "sg_start_req: dxfer_len=%d\n",
1649                                    dxfer_len));
1650
1651         rq = blk_get_request(q, rw, GFP_ATOMIC);
1652         if (!rq)
1653                 return -ENOMEM;
1654
1655         memcpy(rq->cmd, cmd, hp->cmd_len);
1656
1657         rq->cmd_len = hp->cmd_len;
1658         rq->cmd_type = REQ_TYPE_BLOCK_PC;
1659
1660         srp->rq = rq;
1661         rq->end_io_data = srp;
1662         rq->sense = srp->sense_b;
1663         rq->retries = SG_DEFAULT_RETRIES;
1664
1665         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1666                 return 0;
1667
1668         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1669             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1670             !sfp->parentdp->device->host->unchecked_isa_dma &&
1671             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1672                 md = NULL;
1673         else
1674                 md = &map_data;
1675
1676         if (md) {
1677                 if (!sg_res_in_use(sfp) && dxfer_len <= rsv_schp->bufflen)
1678                         sg_link_reserve(sfp, srp, dxfer_len);
1679                 else {
1680                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1681                         if (res)
1682                                 return res;
1683                 }
1684
1685                 md->pages = req_schp->pages;
1686                 md->page_order = req_schp->page_order;
1687                 md->nr_entries = req_schp->k_use_sg;
1688                 md->offset = 0;
1689                 md->null_mapped = hp->dxferp ? 0 : 1;
1690                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1691                         md->from_user = 1;
1692                 else
1693                         md->from_user = 0;
1694         }
1695
1696         if (iov_count) {
1697                 int len, size = sizeof(struct sg_iovec) * iov_count;
1698                 struct iovec *iov;
1699
1700                 iov = memdup_user(hp->dxferp, size);
1701                 if (IS_ERR(iov))
1702                         return PTR_ERR(iov);
1703
1704                 len = iov_length(iov, iov_count);
1705                 if (hp->dxfer_len < len) {
1706                         iov_count = iov_shorten(iov, iov_count, hp->dxfer_len);
1707                         len = hp->dxfer_len;
1708                 }
1709
1710                 res = blk_rq_map_user_iov(q, rq, md, (struct sg_iovec *)iov,
1711                                           iov_count,
1712                                           len, GFP_ATOMIC);
1713                 kfree(iov);
1714         } else
1715                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1716                                       hp->dxfer_len, GFP_ATOMIC);
1717
1718         if (!res) {
1719                 srp->bio = rq->bio;
1720
1721                 if (!md) {
1722                         req_schp->dio_in_use = 1;
1723                         hp->info |= SG_INFO_DIRECT_IO;
1724                 }
1725         }
1726         return res;
1727 }
1728
1729 static int sg_finish_rem_req(Sg_request * srp)
1730 {
1731         int ret = 0;
1732
1733         Sg_fd *sfp = srp->parentfp;
1734         Sg_scatter_hold *req_schp = &srp->data;
1735
1736         SCSI_LOG_TIMEOUT(4, printk("sg_finish_rem_req: res_used=%d\n", (int) srp->res_used));
1737         if (srp->rq) {
1738                 if (srp->bio)
1739                         ret = blk_rq_unmap_user(srp->bio);
1740
1741                 blk_put_request(srp->rq);
1742         }
1743
1744         if (srp->res_used)
1745                 sg_unlink_reserve(sfp, srp);
1746         else
1747                 sg_remove_scat(req_schp);
1748
1749         sg_remove_request(sfp, srp);
1750
1751         return ret;
1752 }
1753
1754 static int
1755 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1756 {
1757         int sg_bufflen = tablesize * sizeof(struct page *);
1758         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1759
1760         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1761         if (!schp->pages)
1762                 return -ENOMEM;
1763         schp->sglist_len = sg_bufflen;
1764         return tablesize;       /* number of scat_gath elements allocated */
1765 }
1766
1767 static int
1768 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1769 {
1770         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1771         int sg_tablesize = sfp->parentdp->sg_tablesize;
1772         int blk_size = buff_size, order;
1773         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1774
1775         if (blk_size < 0)
1776                 return -EFAULT;
1777         if (0 == blk_size)
1778                 ++blk_size;     /* don't know why */
1779         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1780         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1781         SCSI_LOG_TIMEOUT(4, printk("sg_build_indirect: buff_size=%d, blk_size=%d\n",
1782                                    buff_size, blk_size));
1783
1784         /* N.B. ret_sz carried into this block ... */
1785         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1786         if (mx_sc_elems < 0)
1787                 return mx_sc_elems;     /* most likely -ENOMEM */
1788
1789         num = scatter_elem_sz;
1790         if (unlikely(num != scatter_elem_sz_prev)) {
1791                 if (num < PAGE_SIZE) {
1792                         scatter_elem_sz = PAGE_SIZE;
1793                         scatter_elem_sz_prev = PAGE_SIZE;
1794                 } else
1795                         scatter_elem_sz_prev = num;
1796         }
1797
1798         if (sfp->low_dma)
1799                 gfp_mask |= GFP_DMA;
1800
1801         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1802                 gfp_mask |= __GFP_ZERO;
1803
1804         order = get_order(num);
1805 retry:
1806         ret_sz = 1 << (PAGE_SHIFT + order);
1807
1808         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1809              k++, rem_sz -= ret_sz) {
1810
1811                 num = (rem_sz > scatter_elem_sz_prev) ?
1812                         scatter_elem_sz_prev : rem_sz;
1813
1814                 schp->pages[k] = alloc_pages(gfp_mask, order);
1815                 if (!schp->pages[k])
1816                         goto out;
1817
1818                 if (num == scatter_elem_sz_prev) {
1819                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1820                                 scatter_elem_sz = ret_sz;
1821                                 scatter_elem_sz_prev = ret_sz;
1822                         }
1823                 }
1824
1825                 SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k=%d, num=%d, "
1826                                  "ret_sz=%d\n", k, num, ret_sz));
1827         }               /* end of for loop */
1828
1829         schp->page_order = order;
1830         schp->k_use_sg = k;
1831         SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k_use_sg=%d, "
1832                          "rem_sz=%d\n", k, rem_sz));
1833
1834         schp->bufflen = blk_size;
1835         if (rem_sz > 0) /* must have failed */
1836                 return -ENOMEM;
1837         return 0;
1838 out:
1839         for (i = 0; i < k; i++)
1840                 __free_pages(schp->pages[i], order);
1841
1842         if (--order >= 0)
1843                 goto retry;
1844
1845         return -ENOMEM;
1846 }
1847
1848 static void
1849 sg_remove_scat(Sg_scatter_hold * schp)
1850 {
1851         SCSI_LOG_TIMEOUT(4, printk("sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1852         if (schp->pages && schp->sglist_len > 0) {
1853                 if (!schp->dio_in_use) {
1854                         int k;
1855
1856                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1857                                 SCSI_LOG_TIMEOUT(5, printk(
1858                                     "sg_remove_scat: k=%d, pg=0x%p\n",
1859                                     k, schp->pages[k]));
1860                                 __free_pages(schp->pages[k], schp->page_order);
1861                         }
1862
1863                         kfree(schp->pages);
1864                 }
1865         }
1866         memset(schp, 0, sizeof (*schp));
1867 }
1868
1869 static int
1870 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1871 {
1872         Sg_scatter_hold *schp = &srp->data;
1873         int k, num;
1874
1875         SCSI_LOG_TIMEOUT(4, printk("sg_read_oxfer: num_read_xfer=%d\n",
1876                                    num_read_xfer));
1877         if ((!outp) || (num_read_xfer <= 0))
1878                 return 0;
1879
1880         num = 1 << (PAGE_SHIFT + schp->page_order);
1881         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1882                 if (num > num_read_xfer) {
1883                         if (__copy_to_user(outp, page_address(schp->pages[k]),
1884                                            num_read_xfer))
1885                                 return -EFAULT;
1886                         break;
1887                 } else {
1888                         if (__copy_to_user(outp, page_address(schp->pages[k]),
1889                                            num))
1890                                 return -EFAULT;
1891                         num_read_xfer -= num;
1892                         if (num_read_xfer <= 0)
1893                                 break;
1894                         outp += num;
1895                 }
1896         }
1897
1898         return 0;
1899 }
1900
1901 static void
1902 sg_build_reserve(Sg_fd * sfp, int req_size)
1903 {
1904         Sg_scatter_hold *schp = &sfp->reserve;
1905
1906         SCSI_LOG_TIMEOUT(4, printk("sg_build_reserve: req_size=%d\n", req_size));
1907         do {
1908                 if (req_size < PAGE_SIZE)
1909                         req_size = PAGE_SIZE;
1910                 if (0 == sg_build_indirect(schp, sfp, req_size))
1911                         return;
1912                 else
1913                         sg_remove_scat(schp);
1914                 req_size >>= 1; /* divide by 2 */
1915         } while (req_size > (PAGE_SIZE / 2));
1916 }
1917
1918 static void
1919 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
1920 {
1921         Sg_scatter_hold *req_schp = &srp->data;
1922         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1923         int k, num, rem;
1924
1925         srp->res_used = 1;
1926         SCSI_LOG_TIMEOUT(4, printk("sg_link_reserve: size=%d\n", size));
1927         rem = size;
1928
1929         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1930         for (k = 0; k < rsv_schp->k_use_sg; k++) {
1931                 if (rem <= num) {
1932                         req_schp->k_use_sg = k + 1;
1933                         req_schp->sglist_len = rsv_schp->sglist_len;
1934                         req_schp->pages = rsv_schp->pages;
1935
1936                         req_schp->bufflen = size;
1937                         req_schp->page_order = rsv_schp->page_order;
1938                         break;
1939                 } else
1940                         rem -= num;
1941         }
1942
1943         if (k >= rsv_schp->k_use_sg)
1944                 SCSI_LOG_TIMEOUT(1, printk("sg_link_reserve: BAD size\n"));
1945 }
1946
1947 static void
1948 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
1949 {
1950         Sg_scatter_hold *req_schp = &srp->data;
1951
1952         SCSI_LOG_TIMEOUT(4, printk("sg_unlink_reserve: req->k_use_sg=%d\n",
1953                                    (int) req_schp->k_use_sg));
1954         req_schp->k_use_sg = 0;
1955         req_schp->bufflen = 0;
1956         req_schp->pages = NULL;
1957         req_schp->page_order = 0;
1958         req_schp->sglist_len = 0;
1959         sfp->save_scat_len = 0;
1960         srp->res_used = 0;
1961 }
1962
1963 static Sg_request *
1964 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
1965 {
1966         Sg_request *resp;
1967         unsigned long iflags;
1968
1969         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1970         for (resp = sfp->headrp; resp; resp = resp->nextrp) {
1971                 /* look for requests that are ready + not SG_IO owned */
1972                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
1973                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
1974                         resp->done = 2; /* guard against other readers */
1975                         break;
1976                 }
1977         }
1978         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1979         return resp;
1980 }
1981
1982 /* always adds to end of list */
1983 static Sg_request *
1984 sg_add_request(Sg_fd * sfp)
1985 {
1986         int k;
1987         unsigned long iflags;
1988         Sg_request *resp;
1989         Sg_request *rp = sfp->req_arr;
1990
1991         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1992         resp = sfp->headrp;
1993         if (!resp) {
1994                 memset(rp, 0, sizeof (Sg_request));
1995                 rp->parentfp = sfp;
1996                 resp = rp;
1997                 sfp->headrp = resp;
1998         } else {
1999                 if (0 == sfp->cmd_q)
2000                         resp = NULL;    /* command queuing disallowed */
2001                 else {
2002                         for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2003                                 if (!rp->parentfp)
2004                                         break;
2005                         }
2006                         if (k < SG_MAX_QUEUE) {
2007                                 memset(rp, 0, sizeof (Sg_request));
2008                                 rp->parentfp = sfp;
2009                                 while (resp->nextrp)
2010                                         resp = resp->nextrp;
2011                                 resp->nextrp = rp;
2012                                 resp = rp;
2013                         } else
2014                                 resp = NULL;
2015                 }
2016         }
2017         if (resp) {
2018                 resp->nextrp = NULL;
2019                 resp->header.duration = jiffies_to_msecs(jiffies);
2020         }
2021         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2022         return resp;
2023 }
2024
2025 /* Return of 1 for found; 0 for not found */
2026 static int
2027 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2028 {
2029         Sg_request *prev_rp;
2030         Sg_request *rp;
2031         unsigned long iflags;
2032         int res = 0;
2033
2034         if ((!sfp) || (!srp) || (!sfp->headrp))
2035                 return res;
2036         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2037         prev_rp = sfp->headrp;
2038         if (srp == prev_rp) {
2039                 sfp->headrp = prev_rp->nextrp;
2040                 prev_rp->parentfp = NULL;
2041                 res = 1;
2042         } else {
2043                 while ((rp = prev_rp->nextrp)) {
2044                         if (srp == rp) {
2045                                 prev_rp->nextrp = rp->nextrp;
2046                                 rp->parentfp = NULL;
2047                                 res = 1;
2048                                 break;
2049                         }
2050                         prev_rp = rp;
2051                 }
2052         }
2053         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2054         return res;
2055 }
2056
2057 static Sg_fd *
2058 sg_add_sfp(Sg_device * sdp, int dev)
2059 {
2060         Sg_fd *sfp;
2061         unsigned long iflags;
2062         int bufflen;
2063
2064         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2065         if (!sfp)
2066                 return NULL;
2067
2068         init_waitqueue_head(&sfp->read_wait);
2069         rwlock_init(&sfp->rq_list_lock);
2070
2071         kref_init(&sfp->f_ref);
2072         sfp->timeout = SG_DEFAULT_TIMEOUT;
2073         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2074         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2075         sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2076             sdp->device->host->unchecked_isa_dma : 1;
2077         sfp->cmd_q = SG_DEF_COMMAND_Q;
2078         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2079         sfp->parentdp = sdp;
2080         write_lock_irqsave(&sg_index_lock, iflags);
2081         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2082         write_unlock_irqrestore(&sg_index_lock, iflags);
2083         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp: sfp=0x%p\n", sfp));
2084         if (unlikely(sg_big_buff != def_reserved_size))
2085                 sg_big_buff = def_reserved_size;
2086
2087         bufflen = min_t(int, sg_big_buff,
2088                         queue_max_sectors(sdp->device->request_queue) * 512);
2089         sg_build_reserve(sfp, bufflen);
2090         SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp:   bufflen=%d, k_use_sg=%d\n",
2091                            sfp->reserve.bufflen, sfp->reserve.k_use_sg));
2092
2093         kref_get(&sdp->d_ref);
2094         __module_get(THIS_MODULE);
2095         return sfp;
2096 }
2097
2098 static void sg_remove_sfp_usercontext(struct work_struct *work)
2099 {
2100         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2101         struct sg_device *sdp = sfp->parentdp;
2102
2103         /* Cleanup any responses which were never read(). */
2104         while (sfp->headrp)
2105                 sg_finish_rem_req(sfp->headrp);
2106
2107         if (sfp->reserve.bufflen > 0) {
2108                 SCSI_LOG_TIMEOUT(6,
2109                         printk("sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2110                                 (int) sfp->reserve.bufflen,
2111                                 (int) sfp->reserve.k_use_sg));
2112                 sg_remove_scat(&sfp->reserve);
2113         }
2114
2115         SCSI_LOG_TIMEOUT(6,
2116                 printk("sg_remove_sfp: %s, sfp=0x%p\n",
2117                         sdp->disk->disk_name,
2118                         sfp));
2119         kfree(sfp);
2120
2121         scsi_device_put(sdp->device);
2122         sg_put_dev(sdp);
2123         module_put(THIS_MODULE);
2124 }
2125
2126 static void sg_remove_sfp(struct kref *kref)
2127 {
2128         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2129         struct sg_device *sdp = sfp->parentdp;
2130         unsigned long iflags;
2131
2132         write_lock_irqsave(&sg_index_lock, iflags);
2133         list_del(&sfp->sfd_siblings);
2134         write_unlock_irqrestore(&sg_index_lock, iflags);
2135         wake_up_interruptible(&sdp->o_excl_wait);
2136
2137         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2138         schedule_work(&sfp->ew.work);
2139 }
2140
2141 static int
2142 sg_res_in_use(Sg_fd * sfp)
2143 {
2144         const Sg_request *srp;
2145         unsigned long iflags;
2146
2147         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2148         for (srp = sfp->headrp; srp; srp = srp->nextrp)
2149                 if (srp->res_used)
2150                         break;
2151         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2152         return srp ? 1 : 0;
2153 }
2154
2155 #ifdef CONFIG_SCSI_PROC_FS
2156 static int
2157 sg_idr_max_id(int id, void *p, void *data)
2158 {
2159         int *k = data;
2160
2161         if (*k < id)
2162                 *k = id;
2163
2164         return 0;
2165 }
2166
2167 static int
2168 sg_last_dev(void)
2169 {
2170         int k = -1;
2171         unsigned long iflags;
2172
2173         read_lock_irqsave(&sg_index_lock, iflags);
2174         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2175         read_unlock_irqrestore(&sg_index_lock, iflags);
2176         return k + 1;           /* origin 1 */
2177 }
2178 #endif
2179
2180 /* must be called with sg_index_lock held */
2181 static Sg_device *sg_lookup_dev(int dev)
2182 {
2183         return idr_find(&sg_index_idr, dev);
2184 }
2185
2186 static Sg_device *sg_get_dev(int dev)
2187 {
2188         struct sg_device *sdp;
2189         unsigned long flags;
2190
2191         read_lock_irqsave(&sg_index_lock, flags);
2192         sdp = sg_lookup_dev(dev);
2193         if (!sdp)
2194                 sdp = ERR_PTR(-ENXIO);
2195         else if (sdp->detached) {
2196                 /* If sdp->detached, then the refcount may already be 0, in
2197                  * which case it would be a bug to do kref_get().
2198                  */
2199                 sdp = ERR_PTR(-ENODEV);
2200         } else
2201                 kref_get(&sdp->d_ref);
2202         read_unlock_irqrestore(&sg_index_lock, flags);
2203
2204         return sdp;
2205 }
2206
2207 static void sg_put_dev(struct sg_device *sdp)
2208 {
2209         kref_put(&sdp->d_ref, sg_device_destroy);
2210 }
2211
2212 #ifdef CONFIG_SCSI_PROC_FS
2213
2214 static struct proc_dir_entry *sg_proc_sgp = NULL;
2215
2216 static char sg_proc_sg_dirname[] = "scsi/sg";
2217
2218 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2219
2220 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2221 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2222                                   size_t count, loff_t *off);
2223 static const struct file_operations adio_fops = {
2224         .owner = THIS_MODULE,
2225         .open = sg_proc_single_open_adio,
2226         .read = seq_read,
2227         .llseek = seq_lseek,
2228         .write = sg_proc_write_adio,
2229         .release = single_release,
2230 };
2231
2232 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2233 static ssize_t sg_proc_write_dressz(struct file *filp, 
2234                 const char __user *buffer, size_t count, loff_t *off);
2235 static const struct file_operations dressz_fops = {
2236         .owner = THIS_MODULE,
2237         .open = sg_proc_single_open_dressz,
2238         .read = seq_read,
2239         .llseek = seq_lseek,
2240         .write = sg_proc_write_dressz,
2241         .release = single_release,
2242 };
2243
2244 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2245 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2246 static const struct file_operations version_fops = {
2247         .owner = THIS_MODULE,
2248         .open = sg_proc_single_open_version,
2249         .read = seq_read,
2250         .llseek = seq_lseek,
2251         .release = single_release,
2252 };
2253
2254 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2255 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2256 static const struct file_operations devhdr_fops = {
2257         .owner = THIS_MODULE,
2258         .open = sg_proc_single_open_devhdr,
2259         .read = seq_read,
2260         .llseek = seq_lseek,
2261         .release = single_release,
2262 };
2263
2264 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2265 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2266 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2267 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2268 static void dev_seq_stop(struct seq_file *s, void *v);
2269 static const struct file_operations dev_fops = {
2270         .owner = THIS_MODULE,
2271         .open = sg_proc_open_dev,
2272         .read = seq_read,
2273         .llseek = seq_lseek,
2274         .release = seq_release,
2275 };
2276 static const struct seq_operations dev_seq_ops = {
2277         .start = dev_seq_start,
2278         .next  = dev_seq_next,
2279         .stop  = dev_seq_stop,
2280         .show  = sg_proc_seq_show_dev,
2281 };
2282
2283 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2284 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2285 static const struct file_operations devstrs_fops = {
2286         .owner = THIS_MODULE,
2287         .open = sg_proc_open_devstrs,
2288         .read = seq_read,
2289         .llseek = seq_lseek,
2290         .release = seq_release,
2291 };
2292 static const struct seq_operations devstrs_seq_ops = {
2293         .start = dev_seq_start,
2294         .next  = dev_seq_next,
2295         .stop  = dev_seq_stop,
2296         .show  = sg_proc_seq_show_devstrs,
2297 };
2298
2299 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2300 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2301 static const struct file_operations debug_fops = {
2302         .owner = THIS_MODULE,
2303         .open = sg_proc_open_debug,
2304         .read = seq_read,
2305         .llseek = seq_lseek,
2306         .release = seq_release,
2307 };
2308 static const struct seq_operations debug_seq_ops = {
2309         .start = dev_seq_start,
2310         .next  = dev_seq_next,
2311         .stop  = dev_seq_stop,
2312         .show  = sg_proc_seq_show_debug,
2313 };
2314
2315
2316 struct sg_proc_leaf {
2317         const char * name;
2318         const struct file_operations * fops;
2319 };
2320
2321 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2322         {"allow_dio", &adio_fops},
2323         {"debug", &debug_fops},
2324         {"def_reserved_size", &dressz_fops},
2325         {"device_hdr", &devhdr_fops},
2326         {"devices", &dev_fops},
2327         {"device_strs", &devstrs_fops},
2328         {"version", &version_fops}
2329 };
2330
2331 static int
2332 sg_proc_init(void)
2333 {
2334         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2335         int k;
2336
2337         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2338         if (!sg_proc_sgp)
2339                 return 1;
2340         for (k = 0; k < num_leaves; ++k) {
2341                 const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2342                 umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2343                 proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2344         }
2345         return 0;
2346 }
2347
2348 static void
2349 sg_proc_cleanup(void)
2350 {
2351         int k;
2352         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2353
2354         if (!sg_proc_sgp)
2355                 return;
2356         for (k = 0; k < num_leaves; ++k)
2357                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2358         remove_proc_entry(sg_proc_sg_dirname, NULL);
2359 }
2360
2361
2362 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2363 {
2364         seq_printf(s, "%d\n", *((int *)s->private));
2365         return 0;
2366 }
2367
2368 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2369 {
2370         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2371 }
2372
2373 static ssize_t 
2374 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2375                    size_t count, loff_t *off)
2376 {
2377         int err;
2378         unsigned long num;
2379
2380         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2381                 return -EACCES;
2382         err = kstrtoul_from_user(buffer, count, 0, &num);
2383         if (err)
2384                 return err;
2385         sg_allow_dio = num ? 1 : 0;
2386         return count;
2387 }
2388
2389 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2390 {
2391         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2392 }
2393
2394 static ssize_t 
2395 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2396                      size_t count, loff_t *off)
2397 {
2398         int err;
2399         unsigned long k = ULONG_MAX;
2400
2401         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2402                 return -EACCES;
2403
2404         err = kstrtoul_from_user(buffer, count, 0, &k);
2405         if (err)
2406                 return err;
2407         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2408                 sg_big_buff = k;
2409                 return count;
2410         }
2411         return -ERANGE;
2412 }
2413
2414 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2415 {
2416         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2417                    sg_version_date);
2418         return 0;
2419 }
2420
2421 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2422 {
2423         return single_open(file, sg_proc_seq_show_version, NULL);
2424 }
2425
2426 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2427 {
2428         seq_printf(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\t"
2429                    "online\n");
2430         return 0;
2431 }
2432
2433 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2434 {
2435         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2436 }
2437
2438 struct sg_proc_deviter {
2439         loff_t  index;
2440         size_t  max;
2441 };
2442
2443 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2444 {
2445         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2446
2447         s->private = it;
2448         if (! it)
2449                 return NULL;
2450
2451         it->index = *pos;
2452         it->max = sg_last_dev();
2453         if (it->index >= it->max)
2454                 return NULL;
2455         return it;
2456 }
2457
2458 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2459 {
2460         struct sg_proc_deviter * it = s->private;
2461
2462         *pos = ++it->index;
2463         return (it->index < it->max) ? it : NULL;
2464 }
2465
2466 static void dev_seq_stop(struct seq_file *s, void *v)
2467 {
2468         kfree(s->private);
2469 }
2470
2471 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2472 {
2473         return seq_open(file, &dev_seq_ops);
2474 }
2475
2476 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2477 {
2478         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2479         Sg_device *sdp;
2480         struct scsi_device *scsidp;
2481         unsigned long iflags;
2482
2483         read_lock_irqsave(&sg_index_lock, iflags);
2484         sdp = it ? sg_lookup_dev(it->index) : NULL;
2485         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2486                 seq_printf(s, "%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\n",
2487                               scsidp->host->host_no, scsidp->channel,
2488                               scsidp->id, scsidp->lun, (int) scsidp->type,
2489                               1,
2490                               (int) scsidp->queue_depth,
2491                               (int) scsidp->device_busy,
2492                               (int) scsi_device_online(scsidp));
2493         else
2494                 seq_printf(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2495         read_unlock_irqrestore(&sg_index_lock, iflags);
2496         return 0;
2497 }
2498
2499 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2500 {
2501         return seq_open(file, &devstrs_seq_ops);
2502 }
2503
2504 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2505 {
2506         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2507         Sg_device *sdp;
2508         struct scsi_device *scsidp;
2509         unsigned long iflags;
2510
2511         read_lock_irqsave(&sg_index_lock, iflags);
2512         sdp = it ? sg_lookup_dev(it->index) : NULL;
2513         if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2514                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2515                            scsidp->vendor, scsidp->model, scsidp->rev);
2516         else
2517                 seq_printf(s, "<no active device>\n");
2518         read_unlock_irqrestore(&sg_index_lock, iflags);
2519         return 0;
2520 }
2521
2522 /* must be called while holding sg_index_lock */
2523 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2524 {
2525         int k, m, new_interface, blen, usg;
2526         Sg_request *srp;
2527         Sg_fd *fp;
2528         const sg_io_hdr_t *hp;
2529         const char * cp;
2530         unsigned int ms;
2531
2532         k = 0;
2533         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2534                 k++;
2535                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2536                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2537                            "(res)sgat=%d low_dma=%d\n", k,
2538                            jiffies_to_msecs(fp->timeout),
2539                            fp->reserve.bufflen,
2540                            (int) fp->reserve.k_use_sg,
2541                            (int) fp->low_dma);
2542                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2543                            (int) fp->cmd_q, (int) fp->force_packid,
2544                            (int) fp->keep_orphan);
2545                 for (m = 0, srp = fp->headrp;
2546                                 srp != NULL;
2547                                 ++m, srp = srp->nextrp) {
2548                         hp = &srp->header;
2549                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2550                         if (srp->res_used) {
2551                                 if (new_interface && 
2552                                     (SG_FLAG_MMAP_IO & hp->flags))
2553                                         cp = "     mmap>> ";
2554                                 else
2555                                         cp = "     rb>> ";
2556                         } else {
2557                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2558                                         cp = "     dio>> ";
2559                                 else
2560                                         cp = "     ";
2561                         }
2562                         seq_printf(s, cp);
2563                         blen = srp->data.bufflen;
2564                         usg = srp->data.k_use_sg;
2565                         seq_printf(s, srp->done ? 
2566                                    ((1 == srp->done) ?  "rcv:" : "fin:")
2567                                    : "act:");
2568                         seq_printf(s, " id=%d blen=%d",
2569                                    srp->header.pack_id, blen);
2570                         if (srp->done)
2571                                 seq_printf(s, " dur=%d", hp->duration);
2572                         else {
2573                                 ms = jiffies_to_msecs(jiffies);
2574                                 seq_printf(s, " t_o/elap=%d/%d",
2575                                         (new_interface ? hp->timeout :
2576                                                   jiffies_to_msecs(fp->timeout)),
2577                                         (ms > hp->duration ? ms - hp->duration : 0));
2578                         }
2579                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2580                                    (int) srp->data.cmd_opcode);
2581                 }
2582                 if (0 == m)
2583                         seq_printf(s, "     No requests active\n");
2584                 read_unlock(&fp->rq_list_lock);
2585         }
2586 }
2587
2588 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2589 {
2590         return seq_open(file, &debug_seq_ops);
2591 }
2592
2593 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2594 {
2595         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2596         Sg_device *sdp;
2597         unsigned long iflags;
2598
2599         if (it && (0 == it->index)) {
2600                 seq_printf(s, "max_active_device=%d(origin 1)\n",
2601                            (int)it->max);
2602                 seq_printf(s, " def_reserved_size=%d\n", sg_big_buff);
2603         }
2604
2605         read_lock_irqsave(&sg_index_lock, iflags);
2606         sdp = it ? sg_lookup_dev(it->index) : NULL;
2607         if (sdp && !list_empty(&sdp->sfds)) {
2608                 struct scsi_device *scsidp = sdp->device;
2609
2610                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2611                 if (sdp->detached)
2612                         seq_printf(s, "detached pending close ");
2613                 else
2614                         seq_printf
2615                             (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
2616                              scsidp->host->host_no,
2617                              scsidp->channel, scsidp->id,
2618                              scsidp->lun,
2619                              scsidp->host->hostt->emulated);
2620                 seq_printf(s, " sg_tablesize=%d excl=%d\n",
2621                            sdp->sg_tablesize, get_exclude(sdp));
2622                 sg_proc_debug_helper(s, sdp);
2623         }
2624         read_unlock_irqrestore(&sg_index_lock, iflags);
2625         return 0;
2626 }
2627
2628 #endif                          /* CONFIG_SCSI_PROC_FS */
2629
2630 module_init(init_sg);
2631 module_exit(exit_sg);