xref: /openbmc/linux/drivers/scsi/sg.c (revision f7777dcc)
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/aio.h>
39 #include <linux/errno.h>
40 #include <linux/mtio.h>
41 #include <linux/ioctl.h>
42 #include <linux/slab.h>
43 #include <linux/fcntl.h>
44 #include <linux/init.h>
45 #include <linux/poll.h>
46 #include <linux/moduleparam.h>
47 #include <linux/cdev.h>
48 #include <linux/idr.h>
49 #include <linux/seq_file.h>
50 #include <linux/blkdev.h>
51 #include <linux/delay.h>
52 #include <linux/blktrace_api.h>
53 #include <linux/mutex.h>
54 #include <linux/ratelimit.h>
55 
56 #include "scsi.h"
57 #include <scsi/scsi_dbg.h>
58 #include <scsi/scsi_host.h>
59 #include <scsi/scsi_driver.h>
60 #include <scsi/scsi_ioctl.h>
61 #include <scsi/sg.h>
62 
63 #include "scsi_logging.h"
64 
65 #ifdef CONFIG_SCSI_PROC_FS
66 #include <linux/proc_fs.h>
67 static char *sg_version_date = "20061027";
68 
69 static int sg_proc_init(void);
70 static void sg_proc_cleanup(void);
71 #endif
72 
73 #define SG_ALLOW_DIO_DEF 0
74 
75 #define SG_MAX_DEVS 32768
76 
77 /*
78  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
79  * Then when using 32 bit integers x * m may overflow during the calculation.
80  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
81  * calculates the same, but prevents the overflow when both m and d
82  * are "small" numbers (like HZ and USER_HZ).
83  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
84  * in 32 bits.
85  */
86 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
87 
88 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
89 
90 int sg_big_buff = SG_DEF_RESERVED_SIZE;
91 /* N.B. This variable is readable and writeable via
92    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
93    of this size (or less if there is not enough memory) will be reserved
94    for use by this file descriptor. [Deprecated usage: this variable is also
95    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
96    the kernel (i.e. it is not a module).] */
97 static int def_reserved_size = -1;	/* picks up init parameter */
98 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
99 
100 static int scatter_elem_sz = SG_SCATTER_SZ;
101 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
102 
103 #define SG_SECTOR_SZ 512
104 
105 static int sg_add(struct device *, struct class_interface *);
106 static void sg_remove(struct device *, struct class_interface *);
107 
108 static DEFINE_IDR(sg_index_idr);
109 static DEFINE_RWLOCK(sg_index_lock);
110 
111 static struct class_interface sg_interface = {
112 	.add_dev	= sg_add,
113 	.remove_dev	= sg_remove,
114 };
115 
116 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
117 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
118 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
119 	unsigned bufflen;	/* Size of (aggregate) data buffer */
120 	struct page **pages;
121 	int page_order;
122 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
123 	unsigned char cmd_opcode; /* first byte of command */
124 } Sg_scatter_hold;
125 
126 struct sg_device;		/* forward declarations */
127 struct sg_fd;
128 
129 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
130 	struct sg_request *nextrp;	/* NULL -> tail request (slist) */
131 	struct sg_fd *parentfp;	/* NULL -> not in use */
132 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
133 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
134 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
135 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
136 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
137 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
138 	/* done protected by rq_list_lock */
139 	char done;		/* 0->before bh, 1->before read, 2->read */
140 	struct request *rq;
141 	struct bio *bio;
142 	struct execute_work ew;
143 } Sg_request;
144 
145 typedef struct sg_fd {		/* holds the state of a file descriptor */
146 	struct list_head sfd_siblings; /* protected by sfd_lock of device */
147 	struct sg_device *parentdp;	/* owning device */
148 	wait_queue_head_t read_wait;	/* queue read until command done */
149 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
150 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
151 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
152 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
153 	unsigned save_scat_len;	/* original length of trunc. scat. element */
154 	Sg_request *headrp;	/* head of request slist, NULL->empty */
155 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
156 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
157 	char low_dma;		/* as in parent but possibly overridden to 1 */
158 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
159 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
160 	char next_cmd_len;	/* 0 -> automatic (def), >0 -> use on next write() */
161 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
162 	char mmap_called;	/* 0 -> mmap() never called on this fd */
163 	struct kref f_ref;
164 	struct execute_work ew;
165 } Sg_fd;
166 
167 typedef struct sg_device { /* holds the state of each scsi generic device */
168 	struct scsi_device *device;
169 	int sg_tablesize;	/* adapter's max scatter-gather table size */
170 	u32 index;		/* device index number */
171 	spinlock_t sfd_lock;	/* protect file descriptor list for device */
172 	struct list_head sfds;
173 	struct rw_semaphore o_sem;	/* exclude open should hold this rwsem */
174 	volatile char detached;	/* 0->attached, 1->detached pending removal */
175 	char exclude;		/* opened for exclusive access */
176 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
177 	struct gendisk *disk;
178 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
179 	struct kref d_ref;
180 } Sg_device;
181 
182 /* tasklet or soft irq callback */
183 static void sg_rq_end_io(struct request *rq, int uptodate);
184 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
185 static int sg_finish_rem_req(Sg_request * srp);
186 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
187 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
188 			   Sg_request * srp);
189 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
190 			const char __user *buf, size_t count, int blocking,
191 			int read_only, int sg_io_owned, Sg_request **o_srp);
192 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
193 			   unsigned char *cmnd, int timeout, int blocking);
194 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
195 static void sg_remove_scat(Sg_scatter_hold * schp);
196 static void sg_build_reserve(Sg_fd * sfp, int req_size);
197 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
198 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
199 static Sg_fd *sg_add_sfp(Sg_device * sdp, int dev);
200 static void sg_remove_sfp(struct kref *);
201 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
202 static Sg_request *sg_add_request(Sg_fd * sfp);
203 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
204 static int sg_res_in_use(Sg_fd * sfp);
205 static Sg_device *sg_get_dev(int dev);
206 static void sg_put_dev(Sg_device *sdp);
207 
208 #define SZ_SG_HEADER sizeof(struct sg_header)
209 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
210 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
211 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
212 
213 static int sg_allow_access(struct file *filp, unsigned char *cmd)
214 {
215 	struct sg_fd *sfp = filp->private_data;
216 
217 	if (sfp->parentdp->device->type == TYPE_SCANNER)
218 		return 0;
219 
220 	return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
221 }
222 
223 static int sfds_list_empty(Sg_device *sdp)
224 {
225 	unsigned long flags;
226 	int ret;
227 
228 	spin_lock_irqsave(&sdp->sfd_lock, flags);
229 	ret = list_empty(&sdp->sfds);
230 	spin_unlock_irqrestore(&sdp->sfd_lock, flags);
231 	return ret;
232 }
233 
234 static int
235 sg_open(struct inode *inode, struct file *filp)
236 {
237 	int dev = iminor(inode);
238 	int flags = filp->f_flags;
239 	struct request_queue *q;
240 	Sg_device *sdp;
241 	Sg_fd *sfp;
242 	int retval;
243 
244 	nonseekable_open(inode, filp);
245 	SCSI_LOG_TIMEOUT(3, printk("sg_open: dev=%d, flags=0x%x\n", dev, flags));
246 	sdp = sg_get_dev(dev);
247 	if (IS_ERR(sdp)) {
248 		retval = PTR_ERR(sdp);
249 		sdp = NULL;
250 		goto sg_put;
251 	}
252 
253 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
254 	/* Prevent the device driver from vanishing while we sleep */
255 	retval = scsi_device_get(sdp->device);
256 	if (retval)
257 		goto sg_put;
258 
259 	retval = scsi_autopm_get_device(sdp->device);
260 	if (retval)
261 		goto sdp_put;
262 
263 	if (!((flags & O_NONBLOCK) ||
264 	      scsi_block_when_processing_errors(sdp->device))) {
265 		retval = -ENXIO;
266 		/* we are in error recovery for this device */
267 		goto error_out;
268 	}
269 
270 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE))) {
271 		retval = -EPERM; /* Can't lock it with read only access */
272 		goto error_out;
273 	}
274 	if (flags & O_NONBLOCK) {
275 		if (flags & O_EXCL) {
276 			if (!down_write_trylock(&sdp->o_sem)) {
277 				retval = -EBUSY;
278 				goto error_out;
279 			}
280 		} else {
281 			if (!down_read_trylock(&sdp->o_sem)) {
282 				retval = -EBUSY;
283 				goto error_out;
284 			}
285 		}
286 	} else {
287 		if (flags & O_EXCL)
288 			down_write(&sdp->o_sem);
289 		else
290 			down_read(&sdp->o_sem);
291 	}
292 	/* Since write lock is held, no need to check sfd_list */
293 	if (flags & O_EXCL)
294 		sdp->exclude = 1;	/* used by release lock */
295 
296 	if (sfds_list_empty(sdp)) {	/* no existing opens on this device */
297 		sdp->sgdebug = 0;
298 		q = sdp->device->request_queue;
299 		sdp->sg_tablesize = queue_max_segments(q);
300 	}
301 	sfp = sg_add_sfp(sdp, dev);
302 	if (!IS_ERR(sfp))
303 		filp->private_data = sfp;
304 		/* retval is already provably zero at this point because of the
305 		 * check after retval = scsi_autopm_get_device(sdp->device))
306 		 */
307 	else {
308 		retval = PTR_ERR(sfp);
309 
310 		if (flags & O_EXCL) {
311 			sdp->exclude = 0;	/* undo if error */
312 			up_write(&sdp->o_sem);
313 		} else
314 			up_read(&sdp->o_sem);
315 error_out:
316 		scsi_autopm_put_device(sdp->device);
317 sdp_put:
318 		scsi_device_put(sdp->device);
319 	}
320 sg_put:
321 	if (sdp)
322 		sg_put_dev(sdp);
323 	return retval;
324 }
325 
326 /* Following function was formerly called 'sg_close' */
327 static int
328 sg_release(struct inode *inode, struct file *filp)
329 {
330 	Sg_device *sdp;
331 	Sg_fd *sfp;
332 	int excl;
333 
334 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
335 		return -ENXIO;
336 	SCSI_LOG_TIMEOUT(3, printk("sg_release: %s\n", sdp->disk->disk_name));
337 
338 	excl = sdp->exclude;
339 	sdp->exclude = 0;
340 	if (excl)
341 		up_write(&sdp->o_sem);
342 	else
343 		up_read(&sdp->o_sem);
344 
345 	scsi_autopm_put_device(sdp->device);
346 	kref_put(&sfp->f_ref, sg_remove_sfp);
347 	return 0;
348 }
349 
350 static ssize_t
351 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
352 {
353 	Sg_device *sdp;
354 	Sg_fd *sfp;
355 	Sg_request *srp;
356 	int req_pack_id = -1;
357 	sg_io_hdr_t *hp;
358 	struct sg_header *old_hdr = NULL;
359 	int retval = 0;
360 
361 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
362 		return -ENXIO;
363 	SCSI_LOG_TIMEOUT(3, printk("sg_read: %s, count=%d\n",
364 				   sdp->disk->disk_name, (int) count));
365 
366 	if (!access_ok(VERIFY_WRITE, buf, count))
367 		return -EFAULT;
368 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
369 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
370 		if (!old_hdr)
371 			return -ENOMEM;
372 		if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
373 			retval = -EFAULT;
374 			goto free_old_hdr;
375 		}
376 		if (old_hdr->reply_len < 0) {
377 			if (count >= SZ_SG_IO_HDR) {
378 				sg_io_hdr_t *new_hdr;
379 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
380 				if (!new_hdr) {
381 					retval = -ENOMEM;
382 					goto free_old_hdr;
383 				}
384 				retval =__copy_from_user
385 				    (new_hdr, buf, SZ_SG_IO_HDR);
386 				req_pack_id = new_hdr->pack_id;
387 				kfree(new_hdr);
388 				if (retval) {
389 					retval = -EFAULT;
390 					goto free_old_hdr;
391 				}
392 			}
393 		} else
394 			req_pack_id = old_hdr->pack_id;
395 	}
396 	srp = sg_get_rq_mark(sfp, req_pack_id);
397 	if (!srp) {		/* now wait on packet to arrive */
398 		if (sdp->detached) {
399 			retval = -ENODEV;
400 			goto free_old_hdr;
401 		}
402 		if (filp->f_flags & O_NONBLOCK) {
403 			retval = -EAGAIN;
404 			goto free_old_hdr;
405 		}
406 		retval = wait_event_interruptible(sfp->read_wait,
407 			(sdp->detached ||
408 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
409 		if (sdp->detached) {
410 			retval = -ENODEV;
411 			goto free_old_hdr;
412 		}
413 		if (retval) {
414 			/* -ERESTARTSYS as signal hit process */
415 			goto free_old_hdr;
416 		}
417 	}
418 	if (srp->header.interface_id != '\0') {
419 		retval = sg_new_read(sfp, buf, count, srp);
420 		goto free_old_hdr;
421 	}
422 
423 	hp = &srp->header;
424 	if (old_hdr == NULL) {
425 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
426 		if (! old_hdr) {
427 			retval = -ENOMEM;
428 			goto free_old_hdr;
429 		}
430 	}
431 	memset(old_hdr, 0, SZ_SG_HEADER);
432 	old_hdr->reply_len = (int) hp->timeout;
433 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
434 	old_hdr->pack_id = hp->pack_id;
435 	old_hdr->twelve_byte =
436 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
437 	old_hdr->target_status = hp->masked_status;
438 	old_hdr->host_status = hp->host_status;
439 	old_hdr->driver_status = hp->driver_status;
440 	if ((CHECK_CONDITION & hp->masked_status) ||
441 	    (DRIVER_SENSE & hp->driver_status))
442 		memcpy(old_hdr->sense_buffer, srp->sense_b,
443 		       sizeof (old_hdr->sense_buffer));
444 	switch (hp->host_status) {
445 	/* This setup of 'result' is for backward compatibility and is best
446 	   ignored by the user who should use target, host + driver status */
447 	case DID_OK:
448 	case DID_PASSTHROUGH:
449 	case DID_SOFT_ERROR:
450 		old_hdr->result = 0;
451 		break;
452 	case DID_NO_CONNECT:
453 	case DID_BUS_BUSY:
454 	case DID_TIME_OUT:
455 		old_hdr->result = EBUSY;
456 		break;
457 	case DID_BAD_TARGET:
458 	case DID_ABORT:
459 	case DID_PARITY:
460 	case DID_RESET:
461 	case DID_BAD_INTR:
462 		old_hdr->result = EIO;
463 		break;
464 	case DID_ERROR:
465 		old_hdr->result = (srp->sense_b[0] == 0 &&
466 				  hp->masked_status == GOOD) ? 0 : EIO;
467 		break;
468 	default:
469 		old_hdr->result = EIO;
470 		break;
471 	}
472 
473 	/* Now copy the result back to the user buffer.  */
474 	if (count >= SZ_SG_HEADER) {
475 		if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
476 			retval = -EFAULT;
477 			goto free_old_hdr;
478 		}
479 		buf += SZ_SG_HEADER;
480 		if (count > old_hdr->reply_len)
481 			count = old_hdr->reply_len;
482 		if (count > SZ_SG_HEADER) {
483 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
484 				retval = -EFAULT;
485 				goto free_old_hdr;
486 			}
487 		}
488 	} else
489 		count = (old_hdr->result == 0) ? 0 : -EIO;
490 	sg_finish_rem_req(srp);
491 	retval = count;
492 free_old_hdr:
493 	kfree(old_hdr);
494 	return retval;
495 }
496 
497 static ssize_t
498 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
499 {
500 	sg_io_hdr_t *hp = &srp->header;
501 	int err = 0;
502 	int len;
503 
504 	if (count < SZ_SG_IO_HDR) {
505 		err = -EINVAL;
506 		goto err_out;
507 	}
508 	hp->sb_len_wr = 0;
509 	if ((hp->mx_sb_len > 0) && hp->sbp) {
510 		if ((CHECK_CONDITION & hp->masked_status) ||
511 		    (DRIVER_SENSE & hp->driver_status)) {
512 			int sb_len = SCSI_SENSE_BUFFERSIZE;
513 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
514 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
515 			len = (len > sb_len) ? sb_len : len;
516 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
517 				err = -EFAULT;
518 				goto err_out;
519 			}
520 			hp->sb_len_wr = len;
521 		}
522 	}
523 	if (hp->masked_status || hp->host_status || hp->driver_status)
524 		hp->info |= SG_INFO_CHECK;
525 	if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
526 		err = -EFAULT;
527 		goto err_out;
528 	}
529 err_out:
530 	err = sg_finish_rem_req(srp);
531 	return (0 == err) ? count : err;
532 }
533 
534 static ssize_t
535 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
536 {
537 	int mxsize, cmd_size, k;
538 	int input_size, blocking;
539 	unsigned char opcode;
540 	Sg_device *sdp;
541 	Sg_fd *sfp;
542 	Sg_request *srp;
543 	struct sg_header old_hdr;
544 	sg_io_hdr_t *hp;
545 	unsigned char cmnd[MAX_COMMAND_SIZE];
546 
547 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
548 		return -ENXIO;
549 	SCSI_LOG_TIMEOUT(3, printk("sg_write: %s, count=%d\n",
550 				   sdp->disk->disk_name, (int) count));
551 	if (sdp->detached)
552 		return -ENODEV;
553 	if (!((filp->f_flags & O_NONBLOCK) ||
554 	      scsi_block_when_processing_errors(sdp->device)))
555 		return -ENXIO;
556 
557 	if (!access_ok(VERIFY_READ, buf, count))
558 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
559 	if (count < SZ_SG_HEADER)
560 		return -EIO;
561 	if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
562 		return -EFAULT;
563 	blocking = !(filp->f_flags & O_NONBLOCK);
564 	if (old_hdr.reply_len < 0)
565 		return sg_new_write(sfp, filp, buf, count,
566 				    blocking, 0, 0, NULL);
567 	if (count < (SZ_SG_HEADER + 6))
568 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
569 
570 	if (!(srp = sg_add_request(sfp))) {
571 		SCSI_LOG_TIMEOUT(1, printk("sg_write: queue full\n"));
572 		return -EDOM;
573 	}
574 	buf += SZ_SG_HEADER;
575 	__get_user(opcode, buf);
576 	if (sfp->next_cmd_len > 0) {
577 		if (sfp->next_cmd_len > MAX_COMMAND_SIZE) {
578 			SCSI_LOG_TIMEOUT(1, printk("sg_write: command length too long\n"));
579 			sfp->next_cmd_len = 0;
580 			sg_remove_request(sfp, srp);
581 			return -EIO;
582 		}
583 		cmd_size = sfp->next_cmd_len;
584 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
585 	} else {
586 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
587 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
588 			cmd_size = 12;
589 	}
590 	SCSI_LOG_TIMEOUT(4, printk(
591 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
592 /* Determine buffer size.  */
593 	input_size = count - cmd_size;
594 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
595 	mxsize -= SZ_SG_HEADER;
596 	input_size -= SZ_SG_HEADER;
597 	if (input_size < 0) {
598 		sg_remove_request(sfp, srp);
599 		return -EIO;	/* User did not pass enough bytes for this command. */
600 	}
601 	hp = &srp->header;
602 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
603 	hp->cmd_len = (unsigned char) cmd_size;
604 	hp->iovec_count = 0;
605 	hp->mx_sb_len = 0;
606 	if (input_size > 0)
607 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
608 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
609 	else
610 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
611 	hp->dxfer_len = mxsize;
612 	if (hp->dxfer_direction == SG_DXFER_TO_DEV)
613 		hp->dxferp = (char __user *)buf + cmd_size;
614 	else
615 		hp->dxferp = NULL;
616 	hp->sbp = NULL;
617 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
618 	hp->flags = input_size;	/* structure abuse ... */
619 	hp->pack_id = old_hdr.pack_id;
620 	hp->usr_ptr = NULL;
621 	if (__copy_from_user(cmnd, buf, cmd_size))
622 		return -EFAULT;
623 	/*
624 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
625 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
626 	 * is a non-zero input_size, so emit a warning.
627 	 */
628 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
629 		static char cmd[TASK_COMM_LEN];
630 		if (strcmp(current->comm, cmd)) {
631 			printk_ratelimited(KERN_WARNING
632 					   "sg_write: data in/out %d/%d bytes "
633 					   "for SCSI command 0x%x-- guessing "
634 					   "data in;\n   program %s not setting "
635 					   "count and/or reply_len properly\n",
636 					   old_hdr.reply_len - (int)SZ_SG_HEADER,
637 					   input_size, (unsigned int) cmnd[0],
638 					   current->comm);
639 			strcpy(cmd, current->comm);
640 		}
641 	}
642 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
643 	return (k < 0) ? k : count;
644 }
645 
646 static ssize_t
647 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
648 		 size_t count, int blocking, int read_only, int sg_io_owned,
649 		 Sg_request **o_srp)
650 {
651 	int k;
652 	Sg_request *srp;
653 	sg_io_hdr_t *hp;
654 	unsigned char cmnd[MAX_COMMAND_SIZE];
655 	int timeout;
656 	unsigned long ul_timeout;
657 
658 	if (count < SZ_SG_IO_HDR)
659 		return -EINVAL;
660 	if (!access_ok(VERIFY_READ, buf, count))
661 		return -EFAULT; /* protects following copy_from_user()s + get_user()s */
662 
663 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
664 	if (!(srp = sg_add_request(sfp))) {
665 		SCSI_LOG_TIMEOUT(1, printk("sg_new_write: queue full\n"));
666 		return -EDOM;
667 	}
668 	srp->sg_io_owned = sg_io_owned;
669 	hp = &srp->header;
670 	if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
671 		sg_remove_request(sfp, srp);
672 		return -EFAULT;
673 	}
674 	if (hp->interface_id != 'S') {
675 		sg_remove_request(sfp, srp);
676 		return -ENOSYS;
677 	}
678 	if (hp->flags & SG_FLAG_MMAP_IO) {
679 		if (hp->dxfer_len > sfp->reserve.bufflen) {
680 			sg_remove_request(sfp, srp);
681 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
682 		}
683 		if (hp->flags & SG_FLAG_DIRECT_IO) {
684 			sg_remove_request(sfp, srp);
685 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
686 		}
687 		if (sg_res_in_use(sfp)) {
688 			sg_remove_request(sfp, srp);
689 			return -EBUSY;	/* reserve buffer already being used */
690 		}
691 	}
692 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
693 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
694 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
695 		sg_remove_request(sfp, srp);
696 		return -EMSGSIZE;
697 	}
698 	if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
699 		sg_remove_request(sfp, srp);
700 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
701 	}
702 	if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
703 		sg_remove_request(sfp, srp);
704 		return -EFAULT;
705 	}
706 	if (read_only && sg_allow_access(file, cmnd)) {
707 		sg_remove_request(sfp, srp);
708 		return -EPERM;
709 	}
710 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
711 	if (k < 0)
712 		return k;
713 	if (o_srp)
714 		*o_srp = srp;
715 	return count;
716 }
717 
718 static int
719 sg_common_write(Sg_fd * sfp, Sg_request * srp,
720 		unsigned char *cmnd, int timeout, int blocking)
721 {
722 	int k, data_dir;
723 	Sg_device *sdp = sfp->parentdp;
724 	sg_io_hdr_t *hp = &srp->header;
725 
726 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
727 	hp->status = 0;
728 	hp->masked_status = 0;
729 	hp->msg_status = 0;
730 	hp->info = 0;
731 	hp->host_status = 0;
732 	hp->driver_status = 0;
733 	hp->resid = 0;
734 	SCSI_LOG_TIMEOUT(4, printk("sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
735 			  (int) cmnd[0], (int) hp->cmd_len));
736 
737 	k = sg_start_req(srp, cmnd);
738 	if (k) {
739 		SCSI_LOG_TIMEOUT(1, printk("sg_common_write: start_req err=%d\n", k));
740 		sg_finish_rem_req(srp);
741 		return k;	/* probably out of space --> ENOMEM */
742 	}
743 	if (sdp->detached) {
744 		if (srp->bio)
745 			blk_end_request_all(srp->rq, -EIO);
746 		sg_finish_rem_req(srp);
747 		return -ENODEV;
748 	}
749 
750 	switch (hp->dxfer_direction) {
751 	case SG_DXFER_TO_FROM_DEV:
752 	case SG_DXFER_FROM_DEV:
753 		data_dir = DMA_FROM_DEVICE;
754 		break;
755 	case SG_DXFER_TO_DEV:
756 		data_dir = DMA_TO_DEVICE;
757 		break;
758 	case SG_DXFER_UNKNOWN:
759 		data_dir = DMA_BIDIRECTIONAL;
760 		break;
761 	default:
762 		data_dir = DMA_NONE;
763 		break;
764 	}
765 	hp->duration = jiffies_to_msecs(jiffies);
766 
767 	srp->rq->timeout = timeout;
768 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
769 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
770 			      srp->rq, 1, sg_rq_end_io);
771 	return 0;
772 }
773 
774 static int srp_done(Sg_fd *sfp, Sg_request *srp)
775 {
776 	unsigned long flags;
777 	int ret;
778 
779 	read_lock_irqsave(&sfp->rq_list_lock, flags);
780 	ret = srp->done;
781 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
782 	return ret;
783 }
784 
785 static long
786 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
787 {
788 	void __user *p = (void __user *)arg;
789 	int __user *ip = p;
790 	int result, val, read_only;
791 	Sg_device *sdp;
792 	Sg_fd *sfp;
793 	Sg_request *srp;
794 	unsigned long iflags;
795 
796 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
797 		return -ENXIO;
798 
799 	SCSI_LOG_TIMEOUT(3, printk("sg_ioctl: %s, cmd=0x%x\n",
800 				   sdp->disk->disk_name, (int) cmd_in));
801 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
802 
803 	switch (cmd_in) {
804 	case SG_IO:
805 		if (sdp->detached)
806 			return -ENODEV;
807 		if (!scsi_block_when_processing_errors(sdp->device))
808 			return -ENXIO;
809 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
810 			return -EFAULT;
811 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
812 				 1, read_only, 1, &srp);
813 		if (result < 0)
814 			return result;
815 		result = wait_event_interruptible(sfp->read_wait,
816 			(srp_done(sfp, srp) || sdp->detached));
817 		if (sdp->detached)
818 			return -ENODEV;
819 		write_lock_irq(&sfp->rq_list_lock);
820 		if (srp->done) {
821 			srp->done = 2;
822 			write_unlock_irq(&sfp->rq_list_lock);
823 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
824 			return (result < 0) ? result : 0;
825 		}
826 		srp->orphan = 1;
827 		write_unlock_irq(&sfp->rq_list_lock);
828 		return result;	/* -ERESTARTSYS because signal hit process */
829 	case SG_SET_TIMEOUT:
830 		result = get_user(val, ip);
831 		if (result)
832 			return result;
833 		if (val < 0)
834 			return -EIO;
835 		if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
836 		    val = MULDIV (INT_MAX, USER_HZ, HZ);
837 		sfp->timeout_user = val;
838 		sfp->timeout = MULDIV (val, HZ, USER_HZ);
839 
840 		return 0;
841 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
842 				/* strange ..., for backward compatibility */
843 		return sfp->timeout_user;
844 	case SG_SET_FORCE_LOW_DMA:
845 		result = get_user(val, ip);
846 		if (result)
847 			return result;
848 		if (val) {
849 			sfp->low_dma = 1;
850 			if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
851 				val = (int) sfp->reserve.bufflen;
852 				sg_remove_scat(&sfp->reserve);
853 				sg_build_reserve(sfp, val);
854 			}
855 		} else {
856 			if (sdp->detached)
857 				return -ENODEV;
858 			sfp->low_dma = sdp->device->host->unchecked_isa_dma;
859 		}
860 		return 0;
861 	case SG_GET_LOW_DMA:
862 		return put_user((int) sfp->low_dma, ip);
863 	case SG_GET_SCSI_ID:
864 		if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
865 			return -EFAULT;
866 		else {
867 			sg_scsi_id_t __user *sg_idp = p;
868 
869 			if (sdp->detached)
870 				return -ENODEV;
871 			__put_user((int) sdp->device->host->host_no,
872 				   &sg_idp->host_no);
873 			__put_user((int) sdp->device->channel,
874 				   &sg_idp->channel);
875 			__put_user((int) sdp->device->id, &sg_idp->scsi_id);
876 			__put_user((int) sdp->device->lun, &sg_idp->lun);
877 			__put_user((int) sdp->device->type, &sg_idp->scsi_type);
878 			__put_user((short) sdp->device->host->cmd_per_lun,
879 				   &sg_idp->h_cmd_per_lun);
880 			__put_user((short) sdp->device->queue_depth,
881 				   &sg_idp->d_queue_depth);
882 			__put_user(0, &sg_idp->unused[0]);
883 			__put_user(0, &sg_idp->unused[1]);
884 			return 0;
885 		}
886 	case SG_SET_FORCE_PACK_ID:
887 		result = get_user(val, ip);
888 		if (result)
889 			return result;
890 		sfp->force_packid = val ? 1 : 0;
891 		return 0;
892 	case SG_GET_PACK_ID:
893 		if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
894 			return -EFAULT;
895 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
896 		for (srp = sfp->headrp; srp; srp = srp->nextrp) {
897 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
898 				read_unlock_irqrestore(&sfp->rq_list_lock,
899 						       iflags);
900 				__put_user(srp->header.pack_id, ip);
901 				return 0;
902 			}
903 		}
904 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
905 		__put_user(-1, ip);
906 		return 0;
907 	case SG_GET_NUM_WAITING:
908 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
909 		for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
910 			if ((1 == srp->done) && (!srp->sg_io_owned))
911 				++val;
912 		}
913 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
914 		return put_user(val, ip);
915 	case SG_GET_SG_TABLESIZE:
916 		return put_user(sdp->sg_tablesize, ip);
917 	case SG_SET_RESERVED_SIZE:
918 		result = get_user(val, ip);
919 		if (result)
920 			return result;
921                 if (val < 0)
922                         return -EINVAL;
923 		val = min_t(int, val,
924 			    queue_max_sectors(sdp->device->request_queue) * 512);
925 		if (val != sfp->reserve.bufflen) {
926 			if (sg_res_in_use(sfp) || sfp->mmap_called)
927 				return -EBUSY;
928 			sg_remove_scat(&sfp->reserve);
929 			sg_build_reserve(sfp, val);
930 		}
931 		return 0;
932 	case SG_GET_RESERVED_SIZE:
933 		val = min_t(int, sfp->reserve.bufflen,
934 			    queue_max_sectors(sdp->device->request_queue) * 512);
935 		return put_user(val, ip);
936 	case SG_SET_COMMAND_Q:
937 		result = get_user(val, ip);
938 		if (result)
939 			return result;
940 		sfp->cmd_q = val ? 1 : 0;
941 		return 0;
942 	case SG_GET_COMMAND_Q:
943 		return put_user((int) sfp->cmd_q, ip);
944 	case SG_SET_KEEP_ORPHAN:
945 		result = get_user(val, ip);
946 		if (result)
947 			return result;
948 		sfp->keep_orphan = val;
949 		return 0;
950 	case SG_GET_KEEP_ORPHAN:
951 		return put_user((int) sfp->keep_orphan, ip);
952 	case SG_NEXT_CMD_LEN:
953 		result = get_user(val, ip);
954 		if (result)
955 			return result;
956 		sfp->next_cmd_len = (val > 0) ? val : 0;
957 		return 0;
958 	case SG_GET_VERSION_NUM:
959 		return put_user(sg_version_num, ip);
960 	case SG_GET_ACCESS_COUNT:
961 		/* faked - we don't have a real access count anymore */
962 		val = (sdp->device ? 1 : 0);
963 		return put_user(val, ip);
964 	case SG_GET_REQUEST_TABLE:
965 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
966 			return -EFAULT;
967 		else {
968 			sg_req_info_t *rinfo;
969 			unsigned int ms;
970 
971 			rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
972 								GFP_KERNEL);
973 			if (!rinfo)
974 				return -ENOMEM;
975 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
976 			for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
977 			     ++val, srp = srp ? srp->nextrp : srp) {
978 				memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
979 				if (srp) {
980 					rinfo[val].req_state = srp->done + 1;
981 					rinfo[val].problem =
982 					    srp->header.masked_status &
983 					    srp->header.host_status &
984 					    srp->header.driver_status;
985 					if (srp->done)
986 						rinfo[val].duration =
987 							srp->header.duration;
988 					else {
989 						ms = jiffies_to_msecs(jiffies);
990 						rinfo[val].duration =
991 						    (ms > srp->header.duration) ?
992 						    (ms - srp->header.duration) : 0;
993 					}
994 					rinfo[val].orphan = srp->orphan;
995 					rinfo[val].sg_io_owned =
996 							srp->sg_io_owned;
997 					rinfo[val].pack_id =
998 							srp->header.pack_id;
999 					rinfo[val].usr_ptr =
1000 							srp->header.usr_ptr;
1001 				}
1002 			}
1003 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1004 			result = __copy_to_user(p, rinfo,
1005 						SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1006 			result = result ? -EFAULT : 0;
1007 			kfree(rinfo);
1008 			return result;
1009 		}
1010 	case SG_EMULATED_HOST:
1011 		if (sdp->detached)
1012 			return -ENODEV;
1013 		return put_user(sdp->device->host->hostt->emulated, ip);
1014 	case SG_SCSI_RESET:
1015 		if (sdp->detached)
1016 			return -ENODEV;
1017 		if (filp->f_flags & O_NONBLOCK) {
1018 			if (scsi_host_in_recovery(sdp->device->host))
1019 				return -EBUSY;
1020 		} else if (!scsi_block_when_processing_errors(sdp->device))
1021 			return -EBUSY;
1022 		result = get_user(val, ip);
1023 		if (result)
1024 			return result;
1025 		if (SG_SCSI_RESET_NOTHING == val)
1026 			return 0;
1027 		switch (val) {
1028 		case SG_SCSI_RESET_DEVICE:
1029 			val = SCSI_TRY_RESET_DEVICE;
1030 			break;
1031 		case SG_SCSI_RESET_TARGET:
1032 			val = SCSI_TRY_RESET_TARGET;
1033 			break;
1034 		case SG_SCSI_RESET_BUS:
1035 			val = SCSI_TRY_RESET_BUS;
1036 			break;
1037 		case SG_SCSI_RESET_HOST:
1038 			val = SCSI_TRY_RESET_HOST;
1039 			break;
1040 		default:
1041 			return -EINVAL;
1042 		}
1043 		if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1044 			return -EACCES;
1045 		return (scsi_reset_provider(sdp->device, val) ==
1046 			SUCCESS) ? 0 : -EIO;
1047 	case SCSI_IOCTL_SEND_COMMAND:
1048 		if (sdp->detached)
1049 			return -ENODEV;
1050 		if (read_only) {
1051 			unsigned char opcode = WRITE_6;
1052 			Scsi_Ioctl_Command __user *siocp = p;
1053 
1054 			if (copy_from_user(&opcode, siocp->data, 1))
1055 				return -EFAULT;
1056 			if (sg_allow_access(filp, &opcode))
1057 				return -EPERM;
1058 		}
1059 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1060 	case SG_SET_DEBUG:
1061 		result = get_user(val, ip);
1062 		if (result)
1063 			return result;
1064 		sdp->sgdebug = (char) val;
1065 		return 0;
1066 	case SCSI_IOCTL_GET_IDLUN:
1067 	case SCSI_IOCTL_GET_BUS_NUMBER:
1068 	case SCSI_IOCTL_PROBE_HOST:
1069 	case SG_GET_TRANSFORM:
1070 		if (sdp->detached)
1071 			return -ENODEV;
1072 		return scsi_ioctl(sdp->device, cmd_in, p);
1073 	case BLKSECTGET:
1074 		return put_user(queue_max_sectors(sdp->device->request_queue) * 512,
1075 				ip);
1076 	case BLKTRACESETUP:
1077 		return blk_trace_setup(sdp->device->request_queue,
1078 				       sdp->disk->disk_name,
1079 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1080 				       NULL,
1081 				       (char *)arg);
1082 	case BLKTRACESTART:
1083 		return blk_trace_startstop(sdp->device->request_queue, 1);
1084 	case BLKTRACESTOP:
1085 		return blk_trace_startstop(sdp->device->request_queue, 0);
1086 	case BLKTRACETEARDOWN:
1087 		return blk_trace_remove(sdp->device->request_queue);
1088 	default:
1089 		if (read_only)
1090 			return -EPERM;	/* don't know so take safe approach */
1091 		return scsi_ioctl(sdp->device, cmd_in, p);
1092 	}
1093 }
1094 
1095 #ifdef CONFIG_COMPAT
1096 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1097 {
1098 	Sg_device *sdp;
1099 	Sg_fd *sfp;
1100 	struct scsi_device *sdev;
1101 
1102 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1103 		return -ENXIO;
1104 
1105 	sdev = sdp->device;
1106 	if (sdev->host->hostt->compat_ioctl) {
1107 		int ret;
1108 
1109 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1110 
1111 		return ret;
1112 	}
1113 
1114 	return -ENOIOCTLCMD;
1115 }
1116 #endif
1117 
1118 static unsigned int
1119 sg_poll(struct file *filp, poll_table * wait)
1120 {
1121 	unsigned int res = 0;
1122 	Sg_device *sdp;
1123 	Sg_fd *sfp;
1124 	Sg_request *srp;
1125 	int count = 0;
1126 	unsigned long iflags;
1127 
1128 	sfp = filp->private_data;
1129 	if (!sfp)
1130 		return POLLERR;
1131 	sdp = sfp->parentdp;
1132 	if (!sdp)
1133 		return POLLERR;
1134 	poll_wait(filp, &sfp->read_wait, wait);
1135 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1136 	for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1137 		/* if any read waiting, flag it */
1138 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1139 			res = POLLIN | POLLRDNORM;
1140 		++count;
1141 	}
1142 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1143 
1144 	if (sdp->detached)
1145 		res |= POLLHUP;
1146 	else if (!sfp->cmd_q) {
1147 		if (0 == count)
1148 			res |= POLLOUT | POLLWRNORM;
1149 	} else if (count < SG_MAX_QUEUE)
1150 		res |= POLLOUT | POLLWRNORM;
1151 	SCSI_LOG_TIMEOUT(3, printk("sg_poll: %s, res=0x%x\n",
1152 				   sdp->disk->disk_name, (int) res));
1153 	return res;
1154 }
1155 
1156 static int
1157 sg_fasync(int fd, struct file *filp, int mode)
1158 {
1159 	Sg_device *sdp;
1160 	Sg_fd *sfp;
1161 
1162 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1163 		return -ENXIO;
1164 	SCSI_LOG_TIMEOUT(3, printk("sg_fasync: %s, mode=%d\n",
1165 				   sdp->disk->disk_name, mode));
1166 
1167 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1168 }
1169 
1170 static int
1171 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1172 {
1173 	Sg_fd *sfp;
1174 	unsigned long offset, len, sa;
1175 	Sg_scatter_hold *rsv_schp;
1176 	int k, length;
1177 
1178 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1179 		return VM_FAULT_SIGBUS;
1180 	rsv_schp = &sfp->reserve;
1181 	offset = vmf->pgoff << PAGE_SHIFT;
1182 	if (offset >= rsv_schp->bufflen)
1183 		return VM_FAULT_SIGBUS;
1184 	SCSI_LOG_TIMEOUT(3, printk("sg_vma_fault: offset=%lu, scatg=%d\n",
1185 				   offset, rsv_schp->k_use_sg));
1186 	sa = vma->vm_start;
1187 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1188 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1189 		len = vma->vm_end - sa;
1190 		len = (len < length) ? len : length;
1191 		if (offset < len) {
1192 			struct page *page = nth_page(rsv_schp->pages[k],
1193 						     offset >> PAGE_SHIFT);
1194 			get_page(page);	/* increment page count */
1195 			vmf->page = page;
1196 			return 0; /* success */
1197 		}
1198 		sa += len;
1199 		offset -= len;
1200 	}
1201 
1202 	return VM_FAULT_SIGBUS;
1203 }
1204 
1205 static const struct vm_operations_struct sg_mmap_vm_ops = {
1206 	.fault = sg_vma_fault,
1207 };
1208 
1209 static int
1210 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1211 {
1212 	Sg_fd *sfp;
1213 	unsigned long req_sz, len, sa;
1214 	Sg_scatter_hold *rsv_schp;
1215 	int k, length;
1216 
1217 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1218 		return -ENXIO;
1219 	req_sz = vma->vm_end - vma->vm_start;
1220 	SCSI_LOG_TIMEOUT(3, printk("sg_mmap starting, vm_start=%p, len=%d\n",
1221 				   (void *) vma->vm_start, (int) req_sz));
1222 	if (vma->vm_pgoff)
1223 		return -EINVAL;	/* want no offset */
1224 	rsv_schp = &sfp->reserve;
1225 	if (req_sz > rsv_schp->bufflen)
1226 		return -ENOMEM;	/* cannot map more than reserved buffer */
1227 
1228 	sa = vma->vm_start;
1229 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1230 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1231 		len = vma->vm_end - sa;
1232 		len = (len < length) ? len : length;
1233 		sa += len;
1234 	}
1235 
1236 	sfp->mmap_called = 1;
1237 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1238 	vma->vm_private_data = sfp;
1239 	vma->vm_ops = &sg_mmap_vm_ops;
1240 	return 0;
1241 }
1242 
1243 static void sg_rq_end_io_usercontext(struct work_struct *work)
1244 {
1245 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1246 	struct sg_fd *sfp = srp->parentfp;
1247 
1248 	sg_finish_rem_req(srp);
1249 	kref_put(&sfp->f_ref, sg_remove_sfp);
1250 }
1251 
1252 /*
1253  * This function is a "bottom half" handler that is called by the mid
1254  * level when a command is completed (or has failed).
1255  */
1256 static void sg_rq_end_io(struct request *rq, int uptodate)
1257 {
1258 	struct sg_request *srp = rq->end_io_data;
1259 	Sg_device *sdp;
1260 	Sg_fd *sfp;
1261 	unsigned long iflags;
1262 	unsigned int ms;
1263 	char *sense;
1264 	int result, resid, done = 1;
1265 
1266 	if (WARN_ON(srp->done != 0))
1267 		return;
1268 
1269 	sfp = srp->parentfp;
1270 	if (WARN_ON(sfp == NULL))
1271 		return;
1272 
1273 	sdp = sfp->parentdp;
1274 	if (unlikely(sdp->detached))
1275 		printk(KERN_INFO "sg_rq_end_io: device detached\n");
1276 
1277 	sense = rq->sense;
1278 	result = rq->errors;
1279 	resid = rq->resid_len;
1280 
1281 	SCSI_LOG_TIMEOUT(4, printk("sg_cmd_done: %s, pack_id=%d, res=0x%x\n",
1282 		sdp->disk->disk_name, srp->header.pack_id, result));
1283 	srp->header.resid = resid;
1284 	ms = jiffies_to_msecs(jiffies);
1285 	srp->header.duration = (ms > srp->header.duration) ?
1286 				(ms - srp->header.duration) : 0;
1287 	if (0 != result) {
1288 		struct scsi_sense_hdr sshdr;
1289 
1290 		srp->header.status = 0xff & result;
1291 		srp->header.masked_status = status_byte(result);
1292 		srp->header.msg_status = msg_byte(result);
1293 		srp->header.host_status = host_byte(result);
1294 		srp->header.driver_status = driver_byte(result);
1295 		if ((sdp->sgdebug > 0) &&
1296 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1297 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1298 			__scsi_print_sense("sg_cmd_done", sense,
1299 					   SCSI_SENSE_BUFFERSIZE);
1300 
1301 		/* Following if statement is a patch supplied by Eric Youngdale */
1302 		if (driver_byte(result) != 0
1303 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1304 		    && !scsi_sense_is_deferred(&sshdr)
1305 		    && sshdr.sense_key == UNIT_ATTENTION
1306 		    && sdp->device->removable) {
1307 			/* Detected possible disc change. Set the bit - this */
1308 			/* may be used if there are filesystems using this device */
1309 			sdp->device->changed = 1;
1310 		}
1311 	}
1312 	/* Rely on write phase to clean out srp status values, so no "else" */
1313 
1314 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1315 	if (unlikely(srp->orphan)) {
1316 		if (sfp->keep_orphan)
1317 			srp->sg_io_owned = 0;
1318 		else
1319 			done = 0;
1320 	}
1321 	srp->done = done;
1322 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1323 
1324 	if (likely(done)) {
1325 		/* Now wake up any sg_read() that is waiting for this
1326 		 * packet.
1327 		 */
1328 		wake_up_interruptible(&sfp->read_wait);
1329 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1330 		kref_put(&sfp->f_ref, sg_remove_sfp);
1331 	} else {
1332 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1333 		schedule_work(&srp->ew.work);
1334 	}
1335 }
1336 
1337 static const struct file_operations sg_fops = {
1338 	.owner = THIS_MODULE,
1339 	.read = sg_read,
1340 	.write = sg_write,
1341 	.poll = sg_poll,
1342 	.unlocked_ioctl = sg_ioctl,
1343 #ifdef CONFIG_COMPAT
1344 	.compat_ioctl = sg_compat_ioctl,
1345 #endif
1346 	.open = sg_open,
1347 	.mmap = sg_mmap,
1348 	.release = sg_release,
1349 	.fasync = sg_fasync,
1350 	.llseek = no_llseek,
1351 };
1352 
1353 static struct class *sg_sysfs_class;
1354 
1355 static int sg_sysfs_valid = 0;
1356 
1357 static Sg_device *sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1358 {
1359 	struct request_queue *q = scsidp->request_queue;
1360 	Sg_device *sdp;
1361 	unsigned long iflags;
1362 	int error;
1363 	u32 k;
1364 
1365 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1366 	if (!sdp) {
1367 		printk(KERN_WARNING "kmalloc Sg_device failure\n");
1368 		return ERR_PTR(-ENOMEM);
1369 	}
1370 
1371 	idr_preload(GFP_KERNEL);
1372 	write_lock_irqsave(&sg_index_lock, iflags);
1373 
1374 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1375 	if (error < 0) {
1376 		if (error == -ENOSPC) {
1377 			sdev_printk(KERN_WARNING, scsidp,
1378 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1379 				    scsidp->type, SG_MAX_DEVS - 1);
1380 			error = -ENODEV;
1381 		} else {
1382 			printk(KERN_WARNING
1383 			       "idr allocation Sg_device failure: %d\n", error);
1384 		}
1385 		goto out_unlock;
1386 	}
1387 	k = error;
1388 
1389 	SCSI_LOG_TIMEOUT(3, printk("sg_alloc: dev=%d \n", k));
1390 	sprintf(disk->disk_name, "sg%d", k);
1391 	disk->first_minor = k;
1392 	sdp->disk = disk;
1393 	sdp->device = scsidp;
1394 	spin_lock_init(&sdp->sfd_lock);
1395 	INIT_LIST_HEAD(&sdp->sfds);
1396 	init_rwsem(&sdp->o_sem);
1397 	sdp->sg_tablesize = queue_max_segments(q);
1398 	sdp->index = k;
1399 	kref_init(&sdp->d_ref);
1400 	error = 0;
1401 
1402 out_unlock:
1403 	write_unlock_irqrestore(&sg_index_lock, iflags);
1404 	idr_preload_end();
1405 
1406 	if (error) {
1407 		kfree(sdp);
1408 		return ERR_PTR(error);
1409 	}
1410 	return sdp;
1411 }
1412 
1413 static int
1414 sg_add(struct device *cl_dev, struct class_interface *cl_intf)
1415 {
1416 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1417 	struct gendisk *disk;
1418 	Sg_device *sdp = NULL;
1419 	struct cdev * cdev = NULL;
1420 	int error;
1421 	unsigned long iflags;
1422 
1423 	disk = alloc_disk(1);
1424 	if (!disk) {
1425 		printk(KERN_WARNING "alloc_disk failed\n");
1426 		return -ENOMEM;
1427 	}
1428 	disk->major = SCSI_GENERIC_MAJOR;
1429 
1430 	error = -ENOMEM;
1431 	cdev = cdev_alloc();
1432 	if (!cdev) {
1433 		printk(KERN_WARNING "cdev_alloc failed\n");
1434 		goto out;
1435 	}
1436 	cdev->owner = THIS_MODULE;
1437 	cdev->ops = &sg_fops;
1438 
1439 	sdp = sg_alloc(disk, scsidp);
1440 	if (IS_ERR(sdp)) {
1441 		printk(KERN_WARNING "sg_alloc failed\n");
1442 		error = PTR_ERR(sdp);
1443 		goto out;
1444 	}
1445 
1446 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1447 	if (error)
1448 		goto cdev_add_err;
1449 
1450 	sdp->cdev = cdev;
1451 	if (sg_sysfs_valid) {
1452 		struct device *sg_class_member;
1453 
1454 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1455 						MKDEV(SCSI_GENERIC_MAJOR,
1456 						      sdp->index),
1457 						sdp, "%s", disk->disk_name);
1458 		if (IS_ERR(sg_class_member)) {
1459 			printk(KERN_ERR "sg_add: "
1460 			       "device_create failed\n");
1461 			error = PTR_ERR(sg_class_member);
1462 			goto cdev_add_err;
1463 		}
1464 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1465 					  &sg_class_member->kobj, "generic");
1466 		if (error)
1467 			printk(KERN_ERR "sg_add: unable to make symlink "
1468 					"'generic' back to sg%d\n", sdp->index);
1469 	} else
1470 		printk(KERN_WARNING "sg_add: sg_sys Invalid\n");
1471 
1472 	sdev_printk(KERN_NOTICE, scsidp,
1473 		    "Attached scsi generic sg%d type %d\n", sdp->index,
1474 		    scsidp->type);
1475 
1476 	dev_set_drvdata(cl_dev, sdp);
1477 
1478 	return 0;
1479 
1480 cdev_add_err:
1481 	write_lock_irqsave(&sg_index_lock, iflags);
1482 	idr_remove(&sg_index_idr, sdp->index);
1483 	write_unlock_irqrestore(&sg_index_lock, iflags);
1484 	kfree(sdp);
1485 
1486 out:
1487 	put_disk(disk);
1488 	if (cdev)
1489 		cdev_del(cdev);
1490 	return error;
1491 }
1492 
1493 static void sg_device_destroy(struct kref *kref)
1494 {
1495 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1496 	unsigned long flags;
1497 
1498 	/* CAUTION!  Note that the device can still be found via idr_find()
1499 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1500 	 * any other cleanup.
1501 	 */
1502 
1503 	write_lock_irqsave(&sg_index_lock, flags);
1504 	idr_remove(&sg_index_idr, sdp->index);
1505 	write_unlock_irqrestore(&sg_index_lock, flags);
1506 
1507 	SCSI_LOG_TIMEOUT(3,
1508 		printk("sg_device_destroy: %s\n",
1509 			sdp->disk->disk_name));
1510 
1511 	put_disk(sdp->disk);
1512 	kfree(sdp);
1513 }
1514 
1515 static void sg_remove(struct device *cl_dev, struct class_interface *cl_intf)
1516 {
1517 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1518 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1519 	unsigned long iflags;
1520 	Sg_fd *sfp;
1521 
1522 	if (!sdp || sdp->detached)
1523 		return;
1524 
1525 	SCSI_LOG_TIMEOUT(3, printk("sg_remove: %s\n", sdp->disk->disk_name));
1526 
1527 	/* Need a write lock to set sdp->detached. */
1528 	write_lock_irqsave(&sg_index_lock, iflags);
1529 	spin_lock(&sdp->sfd_lock);
1530 	sdp->detached = 1;
1531 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1532 		wake_up_interruptible(&sfp->read_wait);
1533 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1534 	}
1535 	spin_unlock(&sdp->sfd_lock);
1536 	write_unlock_irqrestore(&sg_index_lock, iflags);
1537 
1538 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1539 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1540 	cdev_del(sdp->cdev);
1541 	sdp->cdev = NULL;
1542 
1543 	sg_put_dev(sdp);
1544 }
1545 
1546 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1547 module_param_named(def_reserved_size, def_reserved_size, int,
1548 		   S_IRUGO | S_IWUSR);
1549 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1550 
1551 MODULE_AUTHOR("Douglas Gilbert");
1552 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1553 MODULE_LICENSE("GPL");
1554 MODULE_VERSION(SG_VERSION_STR);
1555 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1556 
1557 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1558                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1559 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1560 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1561 
1562 static int __init
1563 init_sg(void)
1564 {
1565 	int rc;
1566 
1567 	if (scatter_elem_sz < PAGE_SIZE) {
1568 		scatter_elem_sz = PAGE_SIZE;
1569 		scatter_elem_sz_prev = scatter_elem_sz;
1570 	}
1571 	if (def_reserved_size >= 0)
1572 		sg_big_buff = def_reserved_size;
1573 	else
1574 		def_reserved_size = sg_big_buff;
1575 
1576 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1577 				    SG_MAX_DEVS, "sg");
1578 	if (rc)
1579 		return rc;
1580         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1581         if ( IS_ERR(sg_sysfs_class) ) {
1582 		rc = PTR_ERR(sg_sysfs_class);
1583 		goto err_out;
1584         }
1585 	sg_sysfs_valid = 1;
1586 	rc = scsi_register_interface(&sg_interface);
1587 	if (0 == rc) {
1588 #ifdef CONFIG_SCSI_PROC_FS
1589 		sg_proc_init();
1590 #endif				/* CONFIG_SCSI_PROC_FS */
1591 		return 0;
1592 	}
1593 	class_destroy(sg_sysfs_class);
1594 err_out:
1595 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1596 	return rc;
1597 }
1598 
1599 static void __exit
1600 exit_sg(void)
1601 {
1602 #ifdef CONFIG_SCSI_PROC_FS
1603 	sg_proc_cleanup();
1604 #endif				/* CONFIG_SCSI_PROC_FS */
1605 	scsi_unregister_interface(&sg_interface);
1606 	class_destroy(sg_sysfs_class);
1607 	sg_sysfs_valid = 0;
1608 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1609 				 SG_MAX_DEVS);
1610 	idr_destroy(&sg_index_idr);
1611 }
1612 
1613 static int sg_start_req(Sg_request *srp, unsigned char *cmd)
1614 {
1615 	int res;
1616 	struct request *rq;
1617 	Sg_fd *sfp = srp->parentfp;
1618 	sg_io_hdr_t *hp = &srp->header;
1619 	int dxfer_len = (int) hp->dxfer_len;
1620 	int dxfer_dir = hp->dxfer_direction;
1621 	unsigned int iov_count = hp->iovec_count;
1622 	Sg_scatter_hold *req_schp = &srp->data;
1623 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1624 	struct request_queue *q = sfp->parentdp->device->request_queue;
1625 	struct rq_map_data *md, map_data;
1626 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1627 
1628 	SCSI_LOG_TIMEOUT(4, printk(KERN_INFO "sg_start_req: dxfer_len=%d\n",
1629 				   dxfer_len));
1630 
1631 	rq = blk_get_request(q, rw, GFP_ATOMIC);
1632 	if (!rq)
1633 		return -ENOMEM;
1634 
1635 	memcpy(rq->cmd, cmd, hp->cmd_len);
1636 
1637 	rq->cmd_len = hp->cmd_len;
1638 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
1639 
1640 	srp->rq = rq;
1641 	rq->end_io_data = srp;
1642 	rq->sense = srp->sense_b;
1643 	rq->retries = SG_DEFAULT_RETRIES;
1644 
1645 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1646 		return 0;
1647 
1648 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1649 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1650 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1651 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1652 		md = NULL;
1653 	else
1654 		md = &map_data;
1655 
1656 	if (md) {
1657 		if (!sg_res_in_use(sfp) && dxfer_len <= rsv_schp->bufflen)
1658 			sg_link_reserve(sfp, srp, dxfer_len);
1659 		else {
1660 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1661 			if (res)
1662 				return res;
1663 		}
1664 
1665 		md->pages = req_schp->pages;
1666 		md->page_order = req_schp->page_order;
1667 		md->nr_entries = req_schp->k_use_sg;
1668 		md->offset = 0;
1669 		md->null_mapped = hp->dxferp ? 0 : 1;
1670 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1671 			md->from_user = 1;
1672 		else
1673 			md->from_user = 0;
1674 	}
1675 
1676 	if (iov_count) {
1677 		int len, size = sizeof(struct sg_iovec) * iov_count;
1678 		struct iovec *iov;
1679 
1680 		iov = memdup_user(hp->dxferp, size);
1681 		if (IS_ERR(iov))
1682 			return PTR_ERR(iov);
1683 
1684 		len = iov_length(iov, iov_count);
1685 		if (hp->dxfer_len < len) {
1686 			iov_count = iov_shorten(iov, iov_count, hp->dxfer_len);
1687 			len = hp->dxfer_len;
1688 		}
1689 
1690 		res = blk_rq_map_user_iov(q, rq, md, (struct sg_iovec *)iov,
1691 					  iov_count,
1692 					  len, GFP_ATOMIC);
1693 		kfree(iov);
1694 	} else
1695 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1696 				      hp->dxfer_len, GFP_ATOMIC);
1697 
1698 	if (!res) {
1699 		srp->bio = rq->bio;
1700 
1701 		if (!md) {
1702 			req_schp->dio_in_use = 1;
1703 			hp->info |= SG_INFO_DIRECT_IO;
1704 		}
1705 	}
1706 	return res;
1707 }
1708 
1709 static int sg_finish_rem_req(Sg_request * srp)
1710 {
1711 	int ret = 0;
1712 
1713 	Sg_fd *sfp = srp->parentfp;
1714 	Sg_scatter_hold *req_schp = &srp->data;
1715 
1716 	SCSI_LOG_TIMEOUT(4, printk("sg_finish_rem_req: res_used=%d\n", (int) srp->res_used));
1717 	if (srp->rq) {
1718 		if (srp->bio)
1719 			ret = blk_rq_unmap_user(srp->bio);
1720 
1721 		blk_put_request(srp->rq);
1722 	}
1723 
1724 	if (srp->res_used)
1725 		sg_unlink_reserve(sfp, srp);
1726 	else
1727 		sg_remove_scat(req_schp);
1728 
1729 	sg_remove_request(sfp, srp);
1730 
1731 	return ret;
1732 }
1733 
1734 static int
1735 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1736 {
1737 	int sg_bufflen = tablesize * sizeof(struct page *);
1738 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1739 
1740 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1741 	if (!schp->pages)
1742 		return -ENOMEM;
1743 	schp->sglist_len = sg_bufflen;
1744 	return tablesize;	/* number of scat_gath elements allocated */
1745 }
1746 
1747 static int
1748 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1749 {
1750 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1751 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1752 	int blk_size = buff_size, order;
1753 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1754 
1755 	if (blk_size < 0)
1756 		return -EFAULT;
1757 	if (0 == blk_size)
1758 		++blk_size;	/* don't know why */
1759 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1760 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1761 	SCSI_LOG_TIMEOUT(4, printk("sg_build_indirect: buff_size=%d, blk_size=%d\n",
1762 				   buff_size, blk_size));
1763 
1764 	/* N.B. ret_sz carried into this block ... */
1765 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1766 	if (mx_sc_elems < 0)
1767 		return mx_sc_elems;	/* most likely -ENOMEM */
1768 
1769 	num = scatter_elem_sz;
1770 	if (unlikely(num != scatter_elem_sz_prev)) {
1771 		if (num < PAGE_SIZE) {
1772 			scatter_elem_sz = PAGE_SIZE;
1773 			scatter_elem_sz_prev = PAGE_SIZE;
1774 		} else
1775 			scatter_elem_sz_prev = num;
1776 	}
1777 
1778 	if (sfp->low_dma)
1779 		gfp_mask |= GFP_DMA;
1780 
1781 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1782 		gfp_mask |= __GFP_ZERO;
1783 
1784 	order = get_order(num);
1785 retry:
1786 	ret_sz = 1 << (PAGE_SHIFT + order);
1787 
1788 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1789 	     k++, rem_sz -= ret_sz) {
1790 
1791 		num = (rem_sz > scatter_elem_sz_prev) ?
1792 			scatter_elem_sz_prev : rem_sz;
1793 
1794 		schp->pages[k] = alloc_pages(gfp_mask, order);
1795 		if (!schp->pages[k])
1796 			goto out;
1797 
1798 		if (num == scatter_elem_sz_prev) {
1799 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1800 				scatter_elem_sz = ret_sz;
1801 				scatter_elem_sz_prev = ret_sz;
1802 			}
1803 		}
1804 
1805 		SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k=%d, num=%d, "
1806 				 "ret_sz=%d\n", k, num, ret_sz));
1807 	}		/* end of for loop */
1808 
1809 	schp->page_order = order;
1810 	schp->k_use_sg = k;
1811 	SCSI_LOG_TIMEOUT(5, printk("sg_build_indirect: k_use_sg=%d, "
1812 			 "rem_sz=%d\n", k, rem_sz));
1813 
1814 	schp->bufflen = blk_size;
1815 	if (rem_sz > 0)	/* must have failed */
1816 		return -ENOMEM;
1817 	return 0;
1818 out:
1819 	for (i = 0; i < k; i++)
1820 		__free_pages(schp->pages[i], order);
1821 
1822 	if (--order >= 0)
1823 		goto retry;
1824 
1825 	return -ENOMEM;
1826 }
1827 
1828 static void
1829 sg_remove_scat(Sg_scatter_hold * schp)
1830 {
1831 	SCSI_LOG_TIMEOUT(4, printk("sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1832 	if (schp->pages && schp->sglist_len > 0) {
1833 		if (!schp->dio_in_use) {
1834 			int k;
1835 
1836 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1837 				SCSI_LOG_TIMEOUT(5, printk(
1838 				    "sg_remove_scat: k=%d, pg=0x%p\n",
1839 				    k, schp->pages[k]));
1840 				__free_pages(schp->pages[k], schp->page_order);
1841 			}
1842 
1843 			kfree(schp->pages);
1844 		}
1845 	}
1846 	memset(schp, 0, sizeof (*schp));
1847 }
1848 
1849 static int
1850 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1851 {
1852 	Sg_scatter_hold *schp = &srp->data;
1853 	int k, num;
1854 
1855 	SCSI_LOG_TIMEOUT(4, printk("sg_read_oxfer: num_read_xfer=%d\n",
1856 				   num_read_xfer));
1857 	if ((!outp) || (num_read_xfer <= 0))
1858 		return 0;
1859 
1860 	num = 1 << (PAGE_SHIFT + schp->page_order);
1861 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1862 		if (num > num_read_xfer) {
1863 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1864 					   num_read_xfer))
1865 				return -EFAULT;
1866 			break;
1867 		} else {
1868 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1869 					   num))
1870 				return -EFAULT;
1871 			num_read_xfer -= num;
1872 			if (num_read_xfer <= 0)
1873 				break;
1874 			outp += num;
1875 		}
1876 	}
1877 
1878 	return 0;
1879 }
1880 
1881 static void
1882 sg_build_reserve(Sg_fd * sfp, int req_size)
1883 {
1884 	Sg_scatter_hold *schp = &sfp->reserve;
1885 
1886 	SCSI_LOG_TIMEOUT(4, printk("sg_build_reserve: req_size=%d\n", req_size));
1887 	do {
1888 		if (req_size < PAGE_SIZE)
1889 			req_size = PAGE_SIZE;
1890 		if (0 == sg_build_indirect(schp, sfp, req_size))
1891 			return;
1892 		else
1893 			sg_remove_scat(schp);
1894 		req_size >>= 1;	/* divide by 2 */
1895 	} while (req_size > (PAGE_SIZE / 2));
1896 }
1897 
1898 static void
1899 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
1900 {
1901 	Sg_scatter_hold *req_schp = &srp->data;
1902 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1903 	int k, num, rem;
1904 
1905 	srp->res_used = 1;
1906 	SCSI_LOG_TIMEOUT(4, printk("sg_link_reserve: size=%d\n", size));
1907 	rem = size;
1908 
1909 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1910 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
1911 		if (rem <= num) {
1912 			req_schp->k_use_sg = k + 1;
1913 			req_schp->sglist_len = rsv_schp->sglist_len;
1914 			req_schp->pages = rsv_schp->pages;
1915 
1916 			req_schp->bufflen = size;
1917 			req_schp->page_order = rsv_schp->page_order;
1918 			break;
1919 		} else
1920 			rem -= num;
1921 	}
1922 
1923 	if (k >= rsv_schp->k_use_sg)
1924 		SCSI_LOG_TIMEOUT(1, printk("sg_link_reserve: BAD size\n"));
1925 }
1926 
1927 static void
1928 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
1929 {
1930 	Sg_scatter_hold *req_schp = &srp->data;
1931 
1932 	SCSI_LOG_TIMEOUT(4, printk("sg_unlink_reserve: req->k_use_sg=%d\n",
1933 				   (int) req_schp->k_use_sg));
1934 	req_schp->k_use_sg = 0;
1935 	req_schp->bufflen = 0;
1936 	req_schp->pages = NULL;
1937 	req_schp->page_order = 0;
1938 	req_schp->sglist_len = 0;
1939 	sfp->save_scat_len = 0;
1940 	srp->res_used = 0;
1941 }
1942 
1943 static Sg_request *
1944 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
1945 {
1946 	Sg_request *resp;
1947 	unsigned long iflags;
1948 
1949 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1950 	for (resp = sfp->headrp; resp; resp = resp->nextrp) {
1951 		/* look for requests that are ready + not SG_IO owned */
1952 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
1953 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
1954 			resp->done = 2;	/* guard against other readers */
1955 			break;
1956 		}
1957 	}
1958 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1959 	return resp;
1960 }
1961 
1962 /* always adds to end of list */
1963 static Sg_request *
1964 sg_add_request(Sg_fd * sfp)
1965 {
1966 	int k;
1967 	unsigned long iflags;
1968 	Sg_request *resp;
1969 	Sg_request *rp = sfp->req_arr;
1970 
1971 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1972 	resp = sfp->headrp;
1973 	if (!resp) {
1974 		memset(rp, 0, sizeof (Sg_request));
1975 		rp->parentfp = sfp;
1976 		resp = rp;
1977 		sfp->headrp = resp;
1978 	} else {
1979 		if (0 == sfp->cmd_q)
1980 			resp = NULL;	/* command queuing disallowed */
1981 		else {
1982 			for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
1983 				if (!rp->parentfp)
1984 					break;
1985 			}
1986 			if (k < SG_MAX_QUEUE) {
1987 				memset(rp, 0, sizeof (Sg_request));
1988 				rp->parentfp = sfp;
1989 				while (resp->nextrp)
1990 					resp = resp->nextrp;
1991 				resp->nextrp = rp;
1992 				resp = rp;
1993 			} else
1994 				resp = NULL;
1995 		}
1996 	}
1997 	if (resp) {
1998 		resp->nextrp = NULL;
1999 		resp->header.duration = jiffies_to_msecs(jiffies);
2000 	}
2001 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2002 	return resp;
2003 }
2004 
2005 /* Return of 1 for found; 0 for not found */
2006 static int
2007 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2008 {
2009 	Sg_request *prev_rp;
2010 	Sg_request *rp;
2011 	unsigned long iflags;
2012 	int res = 0;
2013 
2014 	if ((!sfp) || (!srp) || (!sfp->headrp))
2015 		return res;
2016 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2017 	prev_rp = sfp->headrp;
2018 	if (srp == prev_rp) {
2019 		sfp->headrp = prev_rp->nextrp;
2020 		prev_rp->parentfp = NULL;
2021 		res = 1;
2022 	} else {
2023 		while ((rp = prev_rp->nextrp)) {
2024 			if (srp == rp) {
2025 				prev_rp->nextrp = rp->nextrp;
2026 				rp->parentfp = NULL;
2027 				res = 1;
2028 				break;
2029 			}
2030 			prev_rp = rp;
2031 		}
2032 	}
2033 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2034 	return res;
2035 }
2036 
2037 static Sg_fd *
2038 sg_add_sfp(Sg_device * sdp, int dev)
2039 {
2040 	Sg_fd *sfp;
2041 	unsigned long iflags;
2042 	int bufflen;
2043 
2044 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2045 	if (!sfp)
2046 		return ERR_PTR(-ENOMEM);
2047 
2048 	init_waitqueue_head(&sfp->read_wait);
2049 	rwlock_init(&sfp->rq_list_lock);
2050 
2051 	kref_init(&sfp->f_ref);
2052 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2053 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2054 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2055 	sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2056 	    sdp->device->host->unchecked_isa_dma : 1;
2057 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2058 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2059 	sfp->parentdp = sdp;
2060 	spin_lock_irqsave(&sdp->sfd_lock, iflags);
2061 	if (sdp->detached) {
2062 		spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
2063 		return ERR_PTR(-ENODEV);
2064 	}
2065 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2066 	spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
2067 	SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp: sfp=0x%p\n", sfp));
2068 	if (unlikely(sg_big_buff != def_reserved_size))
2069 		sg_big_buff = def_reserved_size;
2070 
2071 	bufflen = min_t(int, sg_big_buff,
2072 			queue_max_sectors(sdp->device->request_queue) * 512);
2073 	sg_build_reserve(sfp, bufflen);
2074 	SCSI_LOG_TIMEOUT(3, printk("sg_add_sfp:   bufflen=%d, k_use_sg=%d\n",
2075 			   sfp->reserve.bufflen, sfp->reserve.k_use_sg));
2076 
2077 	kref_get(&sdp->d_ref);
2078 	__module_get(THIS_MODULE);
2079 	return sfp;
2080 }
2081 
2082 static void sg_remove_sfp_usercontext(struct work_struct *work)
2083 {
2084 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2085 	struct sg_device *sdp = sfp->parentdp;
2086 
2087 	/* Cleanup any responses which were never read(). */
2088 	while (sfp->headrp)
2089 		sg_finish_rem_req(sfp->headrp);
2090 
2091 	if (sfp->reserve.bufflen > 0) {
2092 		SCSI_LOG_TIMEOUT(6,
2093 			printk("sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2094 				(int) sfp->reserve.bufflen,
2095 				(int) sfp->reserve.k_use_sg));
2096 		sg_remove_scat(&sfp->reserve);
2097 	}
2098 
2099 	SCSI_LOG_TIMEOUT(6,
2100 		printk("sg_remove_sfp: %s, sfp=0x%p\n",
2101 			sdp->disk->disk_name,
2102 			sfp));
2103 	kfree(sfp);
2104 
2105 	scsi_device_put(sdp->device);
2106 	sg_put_dev(sdp);
2107 	module_put(THIS_MODULE);
2108 }
2109 
2110 static void sg_remove_sfp(struct kref *kref)
2111 {
2112 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2113 	struct sg_device *sdp = sfp->parentdp;
2114 	unsigned long iflags;
2115 
2116 	spin_lock_irqsave(&sdp->sfd_lock, iflags);
2117 	list_del(&sfp->sfd_siblings);
2118 	spin_unlock_irqrestore(&sdp->sfd_lock, iflags);
2119 
2120 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2121 	schedule_work(&sfp->ew.work);
2122 }
2123 
2124 static int
2125 sg_res_in_use(Sg_fd * sfp)
2126 {
2127 	const Sg_request *srp;
2128 	unsigned long iflags;
2129 
2130 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
2131 	for (srp = sfp->headrp; srp; srp = srp->nextrp)
2132 		if (srp->res_used)
2133 			break;
2134 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2135 	return srp ? 1 : 0;
2136 }
2137 
2138 #ifdef CONFIG_SCSI_PROC_FS
2139 static int
2140 sg_idr_max_id(int id, void *p, void *data)
2141 {
2142 	int *k = data;
2143 
2144 	if (*k < id)
2145 		*k = id;
2146 
2147 	return 0;
2148 }
2149 
2150 static int
2151 sg_last_dev(void)
2152 {
2153 	int k = -1;
2154 	unsigned long iflags;
2155 
2156 	read_lock_irqsave(&sg_index_lock, iflags);
2157 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2158 	read_unlock_irqrestore(&sg_index_lock, iflags);
2159 	return k + 1;		/* origin 1 */
2160 }
2161 #endif
2162 
2163 /* must be called with sg_index_lock held */
2164 static Sg_device *sg_lookup_dev(int dev)
2165 {
2166 	return idr_find(&sg_index_idr, dev);
2167 }
2168 
2169 static Sg_device *sg_get_dev(int dev)
2170 {
2171 	struct sg_device *sdp;
2172 	unsigned long flags;
2173 
2174 	read_lock_irqsave(&sg_index_lock, flags);
2175 	sdp = sg_lookup_dev(dev);
2176 	if (!sdp)
2177 		sdp = ERR_PTR(-ENXIO);
2178 	else if (sdp->detached) {
2179 		/* If sdp->detached, then the refcount may already be 0, in
2180 		 * which case it would be a bug to do kref_get().
2181 		 */
2182 		sdp = ERR_PTR(-ENODEV);
2183 	} else
2184 		kref_get(&sdp->d_ref);
2185 	read_unlock_irqrestore(&sg_index_lock, flags);
2186 
2187 	return sdp;
2188 }
2189 
2190 static void sg_put_dev(struct sg_device *sdp)
2191 {
2192 	kref_put(&sdp->d_ref, sg_device_destroy);
2193 }
2194 
2195 #ifdef CONFIG_SCSI_PROC_FS
2196 
2197 static struct proc_dir_entry *sg_proc_sgp = NULL;
2198 
2199 static char sg_proc_sg_dirname[] = "scsi/sg";
2200 
2201 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2202 
2203 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2204 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2205 			          size_t count, loff_t *off);
2206 static const struct file_operations adio_fops = {
2207 	.owner = THIS_MODULE,
2208 	.open = sg_proc_single_open_adio,
2209 	.read = seq_read,
2210 	.llseek = seq_lseek,
2211 	.write = sg_proc_write_adio,
2212 	.release = single_release,
2213 };
2214 
2215 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2216 static ssize_t sg_proc_write_dressz(struct file *filp,
2217 		const char __user *buffer, size_t count, loff_t *off);
2218 static const struct file_operations dressz_fops = {
2219 	.owner = THIS_MODULE,
2220 	.open = sg_proc_single_open_dressz,
2221 	.read = seq_read,
2222 	.llseek = seq_lseek,
2223 	.write = sg_proc_write_dressz,
2224 	.release = single_release,
2225 };
2226 
2227 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2228 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2229 static const struct file_operations version_fops = {
2230 	.owner = THIS_MODULE,
2231 	.open = sg_proc_single_open_version,
2232 	.read = seq_read,
2233 	.llseek = seq_lseek,
2234 	.release = single_release,
2235 };
2236 
2237 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2238 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2239 static const struct file_operations devhdr_fops = {
2240 	.owner = THIS_MODULE,
2241 	.open = sg_proc_single_open_devhdr,
2242 	.read = seq_read,
2243 	.llseek = seq_lseek,
2244 	.release = single_release,
2245 };
2246 
2247 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2248 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2249 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2250 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2251 static void dev_seq_stop(struct seq_file *s, void *v);
2252 static const struct file_operations dev_fops = {
2253 	.owner = THIS_MODULE,
2254 	.open = sg_proc_open_dev,
2255 	.read = seq_read,
2256 	.llseek = seq_lseek,
2257 	.release = seq_release,
2258 };
2259 static const struct seq_operations dev_seq_ops = {
2260 	.start = dev_seq_start,
2261 	.next  = dev_seq_next,
2262 	.stop  = dev_seq_stop,
2263 	.show  = sg_proc_seq_show_dev,
2264 };
2265 
2266 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2267 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2268 static const struct file_operations devstrs_fops = {
2269 	.owner = THIS_MODULE,
2270 	.open = sg_proc_open_devstrs,
2271 	.read = seq_read,
2272 	.llseek = seq_lseek,
2273 	.release = seq_release,
2274 };
2275 static const struct seq_operations devstrs_seq_ops = {
2276 	.start = dev_seq_start,
2277 	.next  = dev_seq_next,
2278 	.stop  = dev_seq_stop,
2279 	.show  = sg_proc_seq_show_devstrs,
2280 };
2281 
2282 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2283 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2284 static const struct file_operations debug_fops = {
2285 	.owner = THIS_MODULE,
2286 	.open = sg_proc_open_debug,
2287 	.read = seq_read,
2288 	.llseek = seq_lseek,
2289 	.release = seq_release,
2290 };
2291 static const struct seq_operations debug_seq_ops = {
2292 	.start = dev_seq_start,
2293 	.next  = dev_seq_next,
2294 	.stop  = dev_seq_stop,
2295 	.show  = sg_proc_seq_show_debug,
2296 };
2297 
2298 
2299 struct sg_proc_leaf {
2300 	const char * name;
2301 	const struct file_operations * fops;
2302 };
2303 
2304 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2305 	{"allow_dio", &adio_fops},
2306 	{"debug", &debug_fops},
2307 	{"def_reserved_size", &dressz_fops},
2308 	{"device_hdr", &devhdr_fops},
2309 	{"devices", &dev_fops},
2310 	{"device_strs", &devstrs_fops},
2311 	{"version", &version_fops}
2312 };
2313 
2314 static int
2315 sg_proc_init(void)
2316 {
2317 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2318 	int k;
2319 
2320 	sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2321 	if (!sg_proc_sgp)
2322 		return 1;
2323 	for (k = 0; k < num_leaves; ++k) {
2324 		const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2325 		umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2326 		proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2327 	}
2328 	return 0;
2329 }
2330 
2331 static void
2332 sg_proc_cleanup(void)
2333 {
2334 	int k;
2335 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2336 
2337 	if (!sg_proc_sgp)
2338 		return;
2339 	for (k = 0; k < num_leaves; ++k)
2340 		remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2341 	remove_proc_entry(sg_proc_sg_dirname, NULL);
2342 }
2343 
2344 
2345 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2346 {
2347 	seq_printf(s, "%d\n", *((int *)s->private));
2348 	return 0;
2349 }
2350 
2351 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2352 {
2353 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2354 }
2355 
2356 static ssize_t
2357 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2358 		   size_t count, loff_t *off)
2359 {
2360 	int err;
2361 	unsigned long num;
2362 
2363 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2364 		return -EACCES;
2365 	err = kstrtoul_from_user(buffer, count, 0, &num);
2366 	if (err)
2367 		return err;
2368 	sg_allow_dio = num ? 1 : 0;
2369 	return count;
2370 }
2371 
2372 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2373 {
2374 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2375 }
2376 
2377 static ssize_t
2378 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2379 		     size_t count, loff_t *off)
2380 {
2381 	int err;
2382 	unsigned long k = ULONG_MAX;
2383 
2384 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2385 		return -EACCES;
2386 
2387 	err = kstrtoul_from_user(buffer, count, 0, &k);
2388 	if (err)
2389 		return err;
2390 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2391 		sg_big_buff = k;
2392 		return count;
2393 	}
2394 	return -ERANGE;
2395 }
2396 
2397 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2398 {
2399 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2400 		   sg_version_date);
2401 	return 0;
2402 }
2403 
2404 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2405 {
2406 	return single_open(file, sg_proc_seq_show_version, NULL);
2407 }
2408 
2409 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2410 {
2411 	seq_printf(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\t"
2412 		   "online\n");
2413 	return 0;
2414 }
2415 
2416 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2417 {
2418 	return single_open(file, sg_proc_seq_show_devhdr, NULL);
2419 }
2420 
2421 struct sg_proc_deviter {
2422 	loff_t	index;
2423 	size_t	max;
2424 };
2425 
2426 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2427 {
2428 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2429 
2430 	s->private = it;
2431 	if (! it)
2432 		return NULL;
2433 
2434 	it->index = *pos;
2435 	it->max = sg_last_dev();
2436 	if (it->index >= it->max)
2437 		return NULL;
2438 	return it;
2439 }
2440 
2441 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2442 {
2443 	struct sg_proc_deviter * it = s->private;
2444 
2445 	*pos = ++it->index;
2446 	return (it->index < it->max) ? it : NULL;
2447 }
2448 
2449 static void dev_seq_stop(struct seq_file *s, void *v)
2450 {
2451 	kfree(s->private);
2452 }
2453 
2454 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2455 {
2456         return seq_open(file, &dev_seq_ops);
2457 }
2458 
2459 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2460 {
2461 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2462 	Sg_device *sdp;
2463 	struct scsi_device *scsidp;
2464 	unsigned long iflags;
2465 
2466 	read_lock_irqsave(&sg_index_lock, iflags);
2467 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2468 	if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2469 		seq_printf(s, "%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\t%d\n",
2470 			      scsidp->host->host_no, scsidp->channel,
2471 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2472 			      1,
2473 			      (int) scsidp->queue_depth,
2474 			      (int) scsidp->device_busy,
2475 			      (int) scsi_device_online(scsidp));
2476 	else
2477 		seq_printf(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2478 	read_unlock_irqrestore(&sg_index_lock, iflags);
2479 	return 0;
2480 }
2481 
2482 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2483 {
2484         return seq_open(file, &devstrs_seq_ops);
2485 }
2486 
2487 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2488 {
2489 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2490 	Sg_device *sdp;
2491 	struct scsi_device *scsidp;
2492 	unsigned long iflags;
2493 
2494 	read_lock_irqsave(&sg_index_lock, iflags);
2495 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2496 	if (sdp && (scsidp = sdp->device) && (!sdp->detached))
2497 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2498 			   scsidp->vendor, scsidp->model, scsidp->rev);
2499 	else
2500 		seq_printf(s, "<no active device>\n");
2501 	read_unlock_irqrestore(&sg_index_lock, iflags);
2502 	return 0;
2503 }
2504 
2505 /* must be called while holding sg_index_lock and sfd_lock */
2506 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2507 {
2508 	int k, m, new_interface, blen, usg;
2509 	Sg_request *srp;
2510 	Sg_fd *fp;
2511 	const sg_io_hdr_t *hp;
2512 	const char * cp;
2513 	unsigned int ms;
2514 
2515 	k = 0;
2516 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2517 		k++;
2518 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2519 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2520 			   "(res)sgat=%d low_dma=%d\n", k,
2521 			   jiffies_to_msecs(fp->timeout),
2522 			   fp->reserve.bufflen,
2523 			   (int) fp->reserve.k_use_sg,
2524 			   (int) fp->low_dma);
2525 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2526 			   (int) fp->cmd_q, (int) fp->force_packid,
2527 			   (int) fp->keep_orphan);
2528 		for (m = 0, srp = fp->headrp;
2529 				srp != NULL;
2530 				++m, srp = srp->nextrp) {
2531 			hp = &srp->header;
2532 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2533 			if (srp->res_used) {
2534 				if (new_interface &&
2535 				    (SG_FLAG_MMAP_IO & hp->flags))
2536 					cp = "     mmap>> ";
2537 				else
2538 					cp = "     rb>> ";
2539 			} else {
2540 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2541 					cp = "     dio>> ";
2542 				else
2543 					cp = "     ";
2544 			}
2545 			seq_printf(s, cp);
2546 			blen = srp->data.bufflen;
2547 			usg = srp->data.k_use_sg;
2548 			seq_printf(s, srp->done ?
2549 				   ((1 == srp->done) ?  "rcv:" : "fin:")
2550 				   : "act:");
2551 			seq_printf(s, " id=%d blen=%d",
2552 				   srp->header.pack_id, blen);
2553 			if (srp->done)
2554 				seq_printf(s, " dur=%d", hp->duration);
2555 			else {
2556 				ms = jiffies_to_msecs(jiffies);
2557 				seq_printf(s, " t_o/elap=%d/%d",
2558 					(new_interface ? hp->timeout :
2559 						  jiffies_to_msecs(fp->timeout)),
2560 					(ms > hp->duration ? ms - hp->duration : 0));
2561 			}
2562 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2563 				   (int) srp->data.cmd_opcode);
2564 		}
2565 		if (0 == m)
2566 			seq_printf(s, "     No requests active\n");
2567 		read_unlock(&fp->rq_list_lock);
2568 	}
2569 }
2570 
2571 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2572 {
2573         return seq_open(file, &debug_seq_ops);
2574 }
2575 
2576 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2577 {
2578 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2579 	Sg_device *sdp;
2580 	unsigned long iflags;
2581 
2582 	if (it && (0 == it->index)) {
2583 		seq_printf(s, "max_active_device=%d(origin 1)\n",
2584 			   (int)it->max);
2585 		seq_printf(s, " def_reserved_size=%d\n", sg_big_buff);
2586 	}
2587 
2588 	read_lock_irqsave(&sg_index_lock, iflags);
2589 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2590 	if (sdp) {
2591 		spin_lock(&sdp->sfd_lock);
2592 		if (!list_empty(&sdp->sfds)) {
2593 			struct scsi_device *scsidp = sdp->device;
2594 
2595 			seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2596 			if (sdp->detached)
2597 				seq_printf(s, "detached pending close ");
2598 			else
2599 				seq_printf
2600 				    (s, "scsi%d chan=%d id=%d lun=%d   em=%d",
2601 				     scsidp->host->host_no,
2602 				     scsidp->channel, scsidp->id,
2603 				     scsidp->lun,
2604 				     scsidp->host->hostt->emulated);
2605 			seq_printf(s, " sg_tablesize=%d excl=%d\n",
2606 				   sdp->sg_tablesize, sdp->exclude);
2607 			sg_proc_debug_helper(s, sdp);
2608 		}
2609 		spin_unlock(&sdp->sfd_lock);
2610 	}
2611 	read_unlock_irqrestore(&sg_index_lock, iflags);
2612 	return 0;
2613 }
2614 
2615 #endif				/* CONFIG_SCSI_PROC_FS */
2616 
2617 module_init(init_sg);
2618 module_exit(exit_sg);
2619