xref: /openbmc/linux/drivers/scsi/sg.c (revision 3805e6a1)
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 - 2014 Douglas Gilbert
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License as published by
14  * the Free Software Foundation; either version 2, or (at your option)
15  * any later version.
16  *
17  */
18 
19 static int sg_version_num = 30536;	/* 2 digits for each component */
20 #define SG_VERSION_STR "3.5.36"
21 
22 /*
23  *  D. P. Gilbert (dgilbert@interlog.com), notes:
24  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
25  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
26  *        (otherwise the macros compile to empty statements).
27  *
28  */
29 #include <linux/module.h>
30 
31 #include <linux/fs.h>
32 #include <linux/kernel.h>
33 #include <linux/sched.h>
34 #include <linux/string.h>
35 #include <linux/mm.h>
36 #include <linux/errno.h>
37 #include <linux/mtio.h>
38 #include <linux/ioctl.h>
39 #include <linux/slab.h>
40 #include <linux/fcntl.h>
41 #include <linux/init.h>
42 #include <linux/poll.h>
43 #include <linux/moduleparam.h>
44 #include <linux/cdev.h>
45 #include <linux/idr.h>
46 #include <linux/seq_file.h>
47 #include <linux/blkdev.h>
48 #include <linux/delay.h>
49 #include <linux/blktrace_api.h>
50 #include <linux/mutex.h>
51 #include <linux/atomic.h>
52 #include <linux/ratelimit.h>
53 #include <linux/uio.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 = "20140603";
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 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
77  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
78  * than 16 bytes are "variable length" whose length is a multiple of 4
79  */
80 #define SG_MAX_CDB_SIZE 252
81 
82 /*
83  * Suppose you want to calculate the formula muldiv(x,m,d)=int(x * m / d)
84  * Then when using 32 bit integers x * m may overflow during the calculation.
85  * Replacing muldiv(x) by muldiv(x)=((x % d) * m) / d + int(x / d) * m
86  * calculates the same, but prevents the overflow when both m and d
87  * are "small" numbers (like HZ and USER_HZ).
88  * Of course an overflow is inavoidable if the result of muldiv doesn't fit
89  * in 32 bits.
90  */
91 #define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
92 
93 #define SG_DEFAULT_TIMEOUT MULDIV(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
94 
95 int sg_big_buff = SG_DEF_RESERVED_SIZE;
96 /* N.B. This variable is readable and writeable via
97    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
98    of this size (or less if there is not enough memory) will be reserved
99    for use by this file descriptor. [Deprecated usage: this variable is also
100    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
101    the kernel (i.e. it is not a module).] */
102 static int def_reserved_size = -1;	/* picks up init parameter */
103 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
104 
105 static int scatter_elem_sz = SG_SCATTER_SZ;
106 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
107 
108 #define SG_SECTOR_SZ 512
109 
110 static int sg_add_device(struct device *, struct class_interface *);
111 static void sg_remove_device(struct device *, struct class_interface *);
112 
113 static DEFINE_IDR(sg_index_idr);
114 static DEFINE_RWLOCK(sg_index_lock);	/* Also used to lock
115 							   file descriptor list for device */
116 
117 static struct class_interface sg_interface = {
118 	.add_dev        = sg_add_device,
119 	.remove_dev     = sg_remove_device,
120 };
121 
122 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
123 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
124 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
125 	unsigned bufflen;	/* Size of (aggregate) data buffer */
126 	struct page **pages;
127 	int page_order;
128 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
129 	unsigned char cmd_opcode; /* first byte of command */
130 } Sg_scatter_hold;
131 
132 struct sg_device;		/* forward declarations */
133 struct sg_fd;
134 
135 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
136 	struct sg_request *nextrp;	/* NULL -> tail request (slist) */
137 	struct sg_fd *parentfp;	/* NULL -> not in use */
138 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
139 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
140 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
141 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
142 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
143 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
144 	/* done protected by rq_list_lock */
145 	char done;		/* 0->before bh, 1->before read, 2->read */
146 	struct request *rq;
147 	struct bio *bio;
148 	struct execute_work ew;
149 } Sg_request;
150 
151 typedef struct sg_fd {		/* holds the state of a file descriptor */
152 	struct list_head sfd_siblings;  /* protected by device's sfd_lock */
153 	struct sg_device *parentdp;	/* owning device */
154 	wait_queue_head_t read_wait;	/* queue read until command done */
155 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
156 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
157 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
158 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
159 	unsigned save_scat_len;	/* original length of trunc. scat. element */
160 	Sg_request *headrp;	/* head of request slist, NULL->empty */
161 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
162 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
163 	char low_dma;		/* as in parent but possibly overridden to 1 */
164 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
165 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
166 	unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
167 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
168 	char mmap_called;	/* 0 -> mmap() never called on this fd */
169 	struct kref f_ref;
170 	struct execute_work ew;
171 } Sg_fd;
172 
173 typedef struct sg_device { /* holds the state of each scsi generic device */
174 	struct scsi_device *device;
175 	wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
176 	struct mutex open_rel_lock;     /* held when in open() or release() */
177 	int sg_tablesize;	/* adapter's max scatter-gather table size */
178 	u32 index;		/* device index number */
179 	struct list_head sfds;
180 	rwlock_t sfd_lock;      /* protect access to sfd list */
181 	atomic_t detaching;     /* 0->device usable, 1->device detaching */
182 	bool exclude;		/* 1->open(O_EXCL) succeeded and is active */
183 	int open_cnt;		/* count of opens (perhaps < num(sfds) ) */
184 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
185 	struct gendisk *disk;
186 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
187 	struct kref d_ref;
188 } Sg_device;
189 
190 /* tasklet or soft irq callback */
191 static void sg_rq_end_io(struct request *rq, int uptodate);
192 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
193 static int sg_finish_rem_req(Sg_request * srp);
194 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
195 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
196 			   Sg_request * srp);
197 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
198 			const char __user *buf, size_t count, int blocking,
199 			int read_only, int sg_io_owned, Sg_request **o_srp);
200 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
201 			   unsigned char *cmnd, int timeout, int blocking);
202 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
203 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
204 static void sg_build_reserve(Sg_fd * sfp, int req_size);
205 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
206 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
207 static Sg_fd *sg_add_sfp(Sg_device * sdp);
208 static void sg_remove_sfp(struct kref *);
209 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
210 static Sg_request *sg_add_request(Sg_fd * sfp);
211 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
212 static int sg_res_in_use(Sg_fd * sfp);
213 static Sg_device *sg_get_dev(int dev);
214 static void sg_device_destroy(struct kref *kref);
215 
216 #define SZ_SG_HEADER sizeof(struct sg_header)
217 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
218 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
219 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
220 
221 #define sg_printk(prefix, sdp, fmt, a...) \
222 	sdev_prefix_printk(prefix, (sdp)->device,		\
223 			   (sdp)->disk->disk_name, fmt, ##a)
224 
225 static int sg_allow_access(struct file *filp, unsigned char *cmd)
226 {
227 	struct sg_fd *sfp = filp->private_data;
228 
229 	if (sfp->parentdp->device->type == TYPE_SCANNER)
230 		return 0;
231 
232 	return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
233 }
234 
235 static int
236 open_wait(Sg_device *sdp, int flags)
237 {
238 	int retval = 0;
239 
240 	if (flags & O_EXCL) {
241 		while (sdp->open_cnt > 0) {
242 			mutex_unlock(&sdp->open_rel_lock);
243 			retval = wait_event_interruptible(sdp->open_wait,
244 					(atomic_read(&sdp->detaching) ||
245 					 !sdp->open_cnt));
246 			mutex_lock(&sdp->open_rel_lock);
247 
248 			if (retval) /* -ERESTARTSYS */
249 				return retval;
250 			if (atomic_read(&sdp->detaching))
251 				return -ENODEV;
252 		}
253 	} else {
254 		while (sdp->exclude) {
255 			mutex_unlock(&sdp->open_rel_lock);
256 			retval = wait_event_interruptible(sdp->open_wait,
257 					(atomic_read(&sdp->detaching) ||
258 					 !sdp->exclude));
259 			mutex_lock(&sdp->open_rel_lock);
260 
261 			if (retval) /* -ERESTARTSYS */
262 				return retval;
263 			if (atomic_read(&sdp->detaching))
264 				return -ENODEV;
265 		}
266 	}
267 
268 	return retval;
269 }
270 
271 /* Returns 0 on success, else a negated errno value */
272 static int
273 sg_open(struct inode *inode, struct file *filp)
274 {
275 	int dev = iminor(inode);
276 	int flags = filp->f_flags;
277 	struct request_queue *q;
278 	Sg_device *sdp;
279 	Sg_fd *sfp;
280 	int retval;
281 
282 	nonseekable_open(inode, filp);
283 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
284 		return -EPERM; /* Can't lock it with read only access */
285 	sdp = sg_get_dev(dev);
286 	if (IS_ERR(sdp))
287 		return PTR_ERR(sdp);
288 
289 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
290 				      "sg_open: flags=0x%x\n", flags));
291 
292 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
293 	/* Prevent the device driver from vanishing while we sleep */
294 	retval = scsi_device_get(sdp->device);
295 	if (retval)
296 		goto sg_put;
297 
298 	retval = scsi_autopm_get_device(sdp->device);
299 	if (retval)
300 		goto sdp_put;
301 
302 	/* scsi_block_when_processing_errors() may block so bypass
303 	 * check if O_NONBLOCK. Permits SCSI commands to be issued
304 	 * during error recovery. Tread carefully. */
305 	if (!((flags & O_NONBLOCK) ||
306 	      scsi_block_when_processing_errors(sdp->device))) {
307 		retval = -ENXIO;
308 		/* we are in error recovery for this device */
309 		goto error_out;
310 	}
311 
312 	mutex_lock(&sdp->open_rel_lock);
313 	if (flags & O_NONBLOCK) {
314 		if (flags & O_EXCL) {
315 			if (sdp->open_cnt > 0) {
316 				retval = -EBUSY;
317 				goto error_mutex_locked;
318 			}
319 		} else {
320 			if (sdp->exclude) {
321 				retval = -EBUSY;
322 				goto error_mutex_locked;
323 			}
324 		}
325 	} else {
326 		retval = open_wait(sdp, flags);
327 		if (retval) /* -ERESTARTSYS or -ENODEV */
328 			goto error_mutex_locked;
329 	}
330 
331 	/* N.B. at this point we are holding the open_rel_lock */
332 	if (flags & O_EXCL)
333 		sdp->exclude = true;
334 
335 	if (sdp->open_cnt < 1) {  /* no existing opens */
336 		sdp->sgdebug = 0;
337 		q = sdp->device->request_queue;
338 		sdp->sg_tablesize = queue_max_segments(q);
339 	}
340 	sfp = sg_add_sfp(sdp);
341 	if (IS_ERR(sfp)) {
342 		retval = PTR_ERR(sfp);
343 		goto out_undo;
344 	}
345 
346 	filp->private_data = sfp;
347 	sdp->open_cnt++;
348 	mutex_unlock(&sdp->open_rel_lock);
349 
350 	retval = 0;
351 sg_put:
352 	kref_put(&sdp->d_ref, sg_device_destroy);
353 	return retval;
354 
355 out_undo:
356 	if (flags & O_EXCL) {
357 		sdp->exclude = false;   /* undo if error */
358 		wake_up_interruptible(&sdp->open_wait);
359 	}
360 error_mutex_locked:
361 	mutex_unlock(&sdp->open_rel_lock);
362 error_out:
363 	scsi_autopm_put_device(sdp->device);
364 sdp_put:
365 	scsi_device_put(sdp->device);
366 	goto sg_put;
367 }
368 
369 /* Release resources associated with a successful sg_open()
370  * Returns 0 on success, else a negated errno value */
371 static int
372 sg_release(struct inode *inode, struct file *filp)
373 {
374 	Sg_device *sdp;
375 	Sg_fd *sfp;
376 
377 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
378 		return -ENXIO;
379 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
380 
381 	mutex_lock(&sdp->open_rel_lock);
382 	scsi_autopm_put_device(sdp->device);
383 	kref_put(&sfp->f_ref, sg_remove_sfp);
384 	sdp->open_cnt--;
385 
386 	/* possibly many open()s waiting on exlude clearing, start many;
387 	 * only open(O_EXCL)s wait on 0==open_cnt so only start one */
388 	if (sdp->exclude) {
389 		sdp->exclude = false;
390 		wake_up_interruptible_all(&sdp->open_wait);
391 	} else if (0 == sdp->open_cnt) {
392 		wake_up_interruptible(&sdp->open_wait);
393 	}
394 	mutex_unlock(&sdp->open_rel_lock);
395 	return 0;
396 }
397 
398 static ssize_t
399 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
400 {
401 	Sg_device *sdp;
402 	Sg_fd *sfp;
403 	Sg_request *srp;
404 	int req_pack_id = -1;
405 	sg_io_hdr_t *hp;
406 	struct sg_header *old_hdr = NULL;
407 	int retval = 0;
408 
409 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
410 		return -ENXIO;
411 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
412 				      "sg_read: count=%d\n", (int) count));
413 
414 	if (!access_ok(VERIFY_WRITE, buf, count))
415 		return -EFAULT;
416 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
417 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
418 		if (!old_hdr)
419 			return -ENOMEM;
420 		if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
421 			retval = -EFAULT;
422 			goto free_old_hdr;
423 		}
424 		if (old_hdr->reply_len < 0) {
425 			if (count >= SZ_SG_IO_HDR) {
426 				sg_io_hdr_t *new_hdr;
427 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
428 				if (!new_hdr) {
429 					retval = -ENOMEM;
430 					goto free_old_hdr;
431 				}
432 				retval =__copy_from_user
433 				    (new_hdr, buf, SZ_SG_IO_HDR);
434 				req_pack_id = new_hdr->pack_id;
435 				kfree(new_hdr);
436 				if (retval) {
437 					retval = -EFAULT;
438 					goto free_old_hdr;
439 				}
440 			}
441 		} else
442 			req_pack_id = old_hdr->pack_id;
443 	}
444 	srp = sg_get_rq_mark(sfp, req_pack_id);
445 	if (!srp) {		/* now wait on packet to arrive */
446 		if (atomic_read(&sdp->detaching)) {
447 			retval = -ENODEV;
448 			goto free_old_hdr;
449 		}
450 		if (filp->f_flags & O_NONBLOCK) {
451 			retval = -EAGAIN;
452 			goto free_old_hdr;
453 		}
454 		retval = wait_event_interruptible(sfp->read_wait,
455 			(atomic_read(&sdp->detaching) ||
456 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
457 		if (atomic_read(&sdp->detaching)) {
458 			retval = -ENODEV;
459 			goto free_old_hdr;
460 		}
461 		if (retval) {
462 			/* -ERESTARTSYS as signal hit process */
463 			goto free_old_hdr;
464 		}
465 	}
466 	if (srp->header.interface_id != '\0') {
467 		retval = sg_new_read(sfp, buf, count, srp);
468 		goto free_old_hdr;
469 	}
470 
471 	hp = &srp->header;
472 	if (old_hdr == NULL) {
473 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
474 		if (! old_hdr) {
475 			retval = -ENOMEM;
476 			goto free_old_hdr;
477 		}
478 	}
479 	memset(old_hdr, 0, SZ_SG_HEADER);
480 	old_hdr->reply_len = (int) hp->timeout;
481 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
482 	old_hdr->pack_id = hp->pack_id;
483 	old_hdr->twelve_byte =
484 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
485 	old_hdr->target_status = hp->masked_status;
486 	old_hdr->host_status = hp->host_status;
487 	old_hdr->driver_status = hp->driver_status;
488 	if ((CHECK_CONDITION & hp->masked_status) ||
489 	    (DRIVER_SENSE & hp->driver_status))
490 		memcpy(old_hdr->sense_buffer, srp->sense_b,
491 		       sizeof (old_hdr->sense_buffer));
492 	switch (hp->host_status) {
493 	/* This setup of 'result' is for backward compatibility and is best
494 	   ignored by the user who should use target, host + driver status */
495 	case DID_OK:
496 	case DID_PASSTHROUGH:
497 	case DID_SOFT_ERROR:
498 		old_hdr->result = 0;
499 		break;
500 	case DID_NO_CONNECT:
501 	case DID_BUS_BUSY:
502 	case DID_TIME_OUT:
503 		old_hdr->result = EBUSY;
504 		break;
505 	case DID_BAD_TARGET:
506 	case DID_ABORT:
507 	case DID_PARITY:
508 	case DID_RESET:
509 	case DID_BAD_INTR:
510 		old_hdr->result = EIO;
511 		break;
512 	case DID_ERROR:
513 		old_hdr->result = (srp->sense_b[0] == 0 &&
514 				  hp->masked_status == GOOD) ? 0 : EIO;
515 		break;
516 	default:
517 		old_hdr->result = EIO;
518 		break;
519 	}
520 
521 	/* Now copy the result back to the user buffer.  */
522 	if (count >= SZ_SG_HEADER) {
523 		if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
524 			retval = -EFAULT;
525 			goto free_old_hdr;
526 		}
527 		buf += SZ_SG_HEADER;
528 		if (count > old_hdr->reply_len)
529 			count = old_hdr->reply_len;
530 		if (count > SZ_SG_HEADER) {
531 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
532 				retval = -EFAULT;
533 				goto free_old_hdr;
534 			}
535 		}
536 	} else
537 		count = (old_hdr->result == 0) ? 0 : -EIO;
538 	sg_finish_rem_req(srp);
539 	retval = count;
540 free_old_hdr:
541 	kfree(old_hdr);
542 	return retval;
543 }
544 
545 static ssize_t
546 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
547 {
548 	sg_io_hdr_t *hp = &srp->header;
549 	int err = 0, err2;
550 	int len;
551 
552 	if (count < SZ_SG_IO_HDR) {
553 		err = -EINVAL;
554 		goto err_out;
555 	}
556 	hp->sb_len_wr = 0;
557 	if ((hp->mx_sb_len > 0) && hp->sbp) {
558 		if ((CHECK_CONDITION & hp->masked_status) ||
559 		    (DRIVER_SENSE & hp->driver_status)) {
560 			int sb_len = SCSI_SENSE_BUFFERSIZE;
561 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
562 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
563 			len = (len > sb_len) ? sb_len : len;
564 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
565 				err = -EFAULT;
566 				goto err_out;
567 			}
568 			hp->sb_len_wr = len;
569 		}
570 	}
571 	if (hp->masked_status || hp->host_status || hp->driver_status)
572 		hp->info |= SG_INFO_CHECK;
573 	if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
574 		err = -EFAULT;
575 		goto err_out;
576 	}
577 err_out:
578 	err2 = sg_finish_rem_req(srp);
579 	return err ? : err2 ? : count;
580 }
581 
582 static ssize_t
583 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
584 {
585 	int mxsize, cmd_size, k;
586 	int input_size, blocking;
587 	unsigned char opcode;
588 	Sg_device *sdp;
589 	Sg_fd *sfp;
590 	Sg_request *srp;
591 	struct sg_header old_hdr;
592 	sg_io_hdr_t *hp;
593 	unsigned char cmnd[SG_MAX_CDB_SIZE];
594 
595 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
596 		return -ENXIO;
597 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
598 				      "sg_write: count=%d\n", (int) count));
599 	if (atomic_read(&sdp->detaching))
600 		return -ENODEV;
601 	if (!((filp->f_flags & O_NONBLOCK) ||
602 	      scsi_block_when_processing_errors(sdp->device)))
603 		return -ENXIO;
604 
605 	if (!access_ok(VERIFY_READ, buf, count))
606 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
607 	if (count < SZ_SG_HEADER)
608 		return -EIO;
609 	if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
610 		return -EFAULT;
611 	blocking = !(filp->f_flags & O_NONBLOCK);
612 	if (old_hdr.reply_len < 0)
613 		return sg_new_write(sfp, filp, buf, count,
614 				    blocking, 0, 0, NULL);
615 	if (count < (SZ_SG_HEADER + 6))
616 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
617 
618 	if (!(srp = sg_add_request(sfp))) {
619 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
620 					      "sg_write: queue full\n"));
621 		return -EDOM;
622 	}
623 	buf += SZ_SG_HEADER;
624 	__get_user(opcode, buf);
625 	if (sfp->next_cmd_len > 0) {
626 		cmd_size = sfp->next_cmd_len;
627 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
628 	} else {
629 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
630 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
631 			cmd_size = 12;
632 	}
633 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
634 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
635 /* Determine buffer size.  */
636 	input_size = count - cmd_size;
637 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
638 	mxsize -= SZ_SG_HEADER;
639 	input_size -= SZ_SG_HEADER;
640 	if (input_size < 0) {
641 		sg_remove_request(sfp, srp);
642 		return -EIO;	/* User did not pass enough bytes for this command. */
643 	}
644 	hp = &srp->header;
645 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
646 	hp->cmd_len = (unsigned char) cmd_size;
647 	hp->iovec_count = 0;
648 	hp->mx_sb_len = 0;
649 	if (input_size > 0)
650 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
651 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
652 	else
653 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
654 	hp->dxfer_len = mxsize;
655 	if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
656 	    (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
657 		hp->dxferp = (char __user *)buf + cmd_size;
658 	else
659 		hp->dxferp = NULL;
660 	hp->sbp = NULL;
661 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
662 	hp->flags = input_size;	/* structure abuse ... */
663 	hp->pack_id = old_hdr.pack_id;
664 	hp->usr_ptr = NULL;
665 	if (__copy_from_user(cmnd, buf, cmd_size))
666 		return -EFAULT;
667 	/*
668 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
669 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
670 	 * is a non-zero input_size, so emit a warning.
671 	 */
672 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
673 		static char cmd[TASK_COMM_LEN];
674 		if (strcmp(current->comm, cmd)) {
675 			printk_ratelimited(KERN_WARNING
676 					   "sg_write: data in/out %d/%d bytes "
677 					   "for SCSI command 0x%x-- guessing "
678 					   "data in;\n   program %s not setting "
679 					   "count and/or reply_len properly\n",
680 					   old_hdr.reply_len - (int)SZ_SG_HEADER,
681 					   input_size, (unsigned int) cmnd[0],
682 					   current->comm);
683 			strcpy(cmd, current->comm);
684 		}
685 	}
686 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
687 	return (k < 0) ? k : count;
688 }
689 
690 static ssize_t
691 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
692 		 size_t count, int blocking, int read_only, int sg_io_owned,
693 		 Sg_request **o_srp)
694 {
695 	int k;
696 	Sg_request *srp;
697 	sg_io_hdr_t *hp;
698 	unsigned char cmnd[SG_MAX_CDB_SIZE];
699 	int timeout;
700 	unsigned long ul_timeout;
701 
702 	if (count < SZ_SG_IO_HDR)
703 		return -EINVAL;
704 	if (!access_ok(VERIFY_READ, buf, count))
705 		return -EFAULT; /* protects following copy_from_user()s + get_user()s */
706 
707 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
708 	if (!(srp = sg_add_request(sfp))) {
709 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
710 					      "sg_new_write: queue full\n"));
711 		return -EDOM;
712 	}
713 	srp->sg_io_owned = sg_io_owned;
714 	hp = &srp->header;
715 	if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
716 		sg_remove_request(sfp, srp);
717 		return -EFAULT;
718 	}
719 	if (hp->interface_id != 'S') {
720 		sg_remove_request(sfp, srp);
721 		return -ENOSYS;
722 	}
723 	if (hp->flags & SG_FLAG_MMAP_IO) {
724 		if (hp->dxfer_len > sfp->reserve.bufflen) {
725 			sg_remove_request(sfp, srp);
726 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
727 		}
728 		if (hp->flags & SG_FLAG_DIRECT_IO) {
729 			sg_remove_request(sfp, srp);
730 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
731 		}
732 		if (sg_res_in_use(sfp)) {
733 			sg_remove_request(sfp, srp);
734 			return -EBUSY;	/* reserve buffer already being used */
735 		}
736 	}
737 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
738 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
739 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
740 		sg_remove_request(sfp, srp);
741 		return -EMSGSIZE;
742 	}
743 	if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
744 		sg_remove_request(sfp, srp);
745 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
746 	}
747 	if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
748 		sg_remove_request(sfp, srp);
749 		return -EFAULT;
750 	}
751 	if (read_only && sg_allow_access(file, cmnd)) {
752 		sg_remove_request(sfp, srp);
753 		return -EPERM;
754 	}
755 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
756 	if (k < 0)
757 		return k;
758 	if (o_srp)
759 		*o_srp = srp;
760 	return count;
761 }
762 
763 static int
764 sg_common_write(Sg_fd * sfp, Sg_request * srp,
765 		unsigned char *cmnd, int timeout, int blocking)
766 {
767 	int k, at_head;
768 	Sg_device *sdp = sfp->parentdp;
769 	sg_io_hdr_t *hp = &srp->header;
770 
771 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
772 	hp->status = 0;
773 	hp->masked_status = 0;
774 	hp->msg_status = 0;
775 	hp->info = 0;
776 	hp->host_status = 0;
777 	hp->driver_status = 0;
778 	hp->resid = 0;
779 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
780 			"sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
781 			(int) cmnd[0], (int) hp->cmd_len));
782 
783 	k = sg_start_req(srp, cmnd);
784 	if (k) {
785 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
786 			"sg_common_write: start_req err=%d\n", k));
787 		sg_finish_rem_req(srp);
788 		return k;	/* probably out of space --> ENOMEM */
789 	}
790 	if (atomic_read(&sdp->detaching)) {
791 		if (srp->bio) {
792 			if (srp->rq->cmd != srp->rq->__cmd)
793 				kfree(srp->rq->cmd);
794 
795 			blk_end_request_all(srp->rq, -EIO);
796 			srp->rq = NULL;
797 		}
798 
799 		sg_finish_rem_req(srp);
800 		return -ENODEV;
801 	}
802 
803 	hp->duration = jiffies_to_msecs(jiffies);
804 	if (hp->interface_id != '\0' &&	/* v3 (or later) interface */
805 	    (SG_FLAG_Q_AT_TAIL & hp->flags))
806 		at_head = 0;
807 	else
808 		at_head = 1;
809 
810 	srp->rq->timeout = timeout;
811 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
812 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
813 			      srp->rq, at_head, sg_rq_end_io);
814 	return 0;
815 }
816 
817 static int srp_done(Sg_fd *sfp, Sg_request *srp)
818 {
819 	unsigned long flags;
820 	int ret;
821 
822 	read_lock_irqsave(&sfp->rq_list_lock, flags);
823 	ret = srp->done;
824 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
825 	return ret;
826 }
827 
828 static int max_sectors_bytes(struct request_queue *q)
829 {
830 	unsigned int max_sectors = queue_max_sectors(q);
831 
832 	max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
833 
834 	return max_sectors << 9;
835 }
836 
837 static long
838 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
839 {
840 	void __user *p = (void __user *)arg;
841 	int __user *ip = p;
842 	int result, val, read_only;
843 	Sg_device *sdp;
844 	Sg_fd *sfp;
845 	Sg_request *srp;
846 	unsigned long iflags;
847 
848 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
849 		return -ENXIO;
850 
851 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
852 				   "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
853 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
854 
855 	switch (cmd_in) {
856 	case SG_IO:
857 		if (atomic_read(&sdp->detaching))
858 			return -ENODEV;
859 		if (!scsi_block_when_processing_errors(sdp->device))
860 			return -ENXIO;
861 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
862 			return -EFAULT;
863 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
864 				 1, read_only, 1, &srp);
865 		if (result < 0)
866 			return result;
867 		result = wait_event_interruptible(sfp->read_wait,
868 			(srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
869 		if (atomic_read(&sdp->detaching))
870 			return -ENODEV;
871 		write_lock_irq(&sfp->rq_list_lock);
872 		if (srp->done) {
873 			srp->done = 2;
874 			write_unlock_irq(&sfp->rq_list_lock);
875 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
876 			return (result < 0) ? result : 0;
877 		}
878 		srp->orphan = 1;
879 		write_unlock_irq(&sfp->rq_list_lock);
880 		return result;	/* -ERESTARTSYS because signal hit process */
881 	case SG_SET_TIMEOUT:
882 		result = get_user(val, ip);
883 		if (result)
884 			return result;
885 		if (val < 0)
886 			return -EIO;
887 		if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
888 		    val = MULDIV (INT_MAX, USER_HZ, HZ);
889 		sfp->timeout_user = val;
890 		sfp->timeout = MULDIV (val, HZ, USER_HZ);
891 
892 		return 0;
893 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
894 				/* strange ..., for backward compatibility */
895 		return sfp->timeout_user;
896 	case SG_SET_FORCE_LOW_DMA:
897 		result = get_user(val, ip);
898 		if (result)
899 			return result;
900 		if (val) {
901 			sfp->low_dma = 1;
902 			if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
903 				val = (int) sfp->reserve.bufflen;
904 				sg_remove_scat(sfp, &sfp->reserve);
905 				sg_build_reserve(sfp, val);
906 			}
907 		} else {
908 			if (atomic_read(&sdp->detaching))
909 				return -ENODEV;
910 			sfp->low_dma = sdp->device->host->unchecked_isa_dma;
911 		}
912 		return 0;
913 	case SG_GET_LOW_DMA:
914 		return put_user((int) sfp->low_dma, ip);
915 	case SG_GET_SCSI_ID:
916 		if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
917 			return -EFAULT;
918 		else {
919 			sg_scsi_id_t __user *sg_idp = p;
920 
921 			if (atomic_read(&sdp->detaching))
922 				return -ENODEV;
923 			__put_user((int) sdp->device->host->host_no,
924 				   &sg_idp->host_no);
925 			__put_user((int) sdp->device->channel,
926 				   &sg_idp->channel);
927 			__put_user((int) sdp->device->id, &sg_idp->scsi_id);
928 			__put_user((int) sdp->device->lun, &sg_idp->lun);
929 			__put_user((int) sdp->device->type, &sg_idp->scsi_type);
930 			__put_user((short) sdp->device->host->cmd_per_lun,
931 				   &sg_idp->h_cmd_per_lun);
932 			__put_user((short) sdp->device->queue_depth,
933 				   &sg_idp->d_queue_depth);
934 			__put_user(0, &sg_idp->unused[0]);
935 			__put_user(0, &sg_idp->unused[1]);
936 			return 0;
937 		}
938 	case SG_SET_FORCE_PACK_ID:
939 		result = get_user(val, ip);
940 		if (result)
941 			return result;
942 		sfp->force_packid = val ? 1 : 0;
943 		return 0;
944 	case SG_GET_PACK_ID:
945 		if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
946 			return -EFAULT;
947 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
948 		for (srp = sfp->headrp; srp; srp = srp->nextrp) {
949 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
950 				read_unlock_irqrestore(&sfp->rq_list_lock,
951 						       iflags);
952 				__put_user(srp->header.pack_id, ip);
953 				return 0;
954 			}
955 		}
956 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
957 		__put_user(-1, ip);
958 		return 0;
959 	case SG_GET_NUM_WAITING:
960 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
961 		for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
962 			if ((1 == srp->done) && (!srp->sg_io_owned))
963 				++val;
964 		}
965 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
966 		return put_user(val, ip);
967 	case SG_GET_SG_TABLESIZE:
968 		return put_user(sdp->sg_tablesize, ip);
969 	case SG_SET_RESERVED_SIZE:
970 		result = get_user(val, ip);
971 		if (result)
972 			return result;
973                 if (val < 0)
974                         return -EINVAL;
975 		val = min_t(int, val,
976 			    max_sectors_bytes(sdp->device->request_queue));
977 		if (val != sfp->reserve.bufflen) {
978 			if (sg_res_in_use(sfp) || sfp->mmap_called)
979 				return -EBUSY;
980 			sg_remove_scat(sfp, &sfp->reserve);
981 			sg_build_reserve(sfp, val);
982 		}
983 		return 0;
984 	case SG_GET_RESERVED_SIZE:
985 		val = min_t(int, sfp->reserve.bufflen,
986 			    max_sectors_bytes(sdp->device->request_queue));
987 		return put_user(val, ip);
988 	case SG_SET_COMMAND_Q:
989 		result = get_user(val, ip);
990 		if (result)
991 			return result;
992 		sfp->cmd_q = val ? 1 : 0;
993 		return 0;
994 	case SG_GET_COMMAND_Q:
995 		return put_user((int) sfp->cmd_q, ip);
996 	case SG_SET_KEEP_ORPHAN:
997 		result = get_user(val, ip);
998 		if (result)
999 			return result;
1000 		sfp->keep_orphan = val;
1001 		return 0;
1002 	case SG_GET_KEEP_ORPHAN:
1003 		return put_user((int) sfp->keep_orphan, ip);
1004 	case SG_NEXT_CMD_LEN:
1005 		result = get_user(val, ip);
1006 		if (result)
1007 			return result;
1008 		sfp->next_cmd_len = (val > 0) ? val : 0;
1009 		return 0;
1010 	case SG_GET_VERSION_NUM:
1011 		return put_user(sg_version_num, ip);
1012 	case SG_GET_ACCESS_COUNT:
1013 		/* faked - we don't have a real access count anymore */
1014 		val = (sdp->device ? 1 : 0);
1015 		return put_user(val, ip);
1016 	case SG_GET_REQUEST_TABLE:
1017 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1018 			return -EFAULT;
1019 		else {
1020 			sg_req_info_t *rinfo;
1021 			unsigned int ms;
1022 
1023 			rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1024 								GFP_KERNEL);
1025 			if (!rinfo)
1026 				return -ENOMEM;
1027 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
1028 			for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
1029 			     ++val, srp = srp ? srp->nextrp : srp) {
1030 				memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
1031 				if (srp) {
1032 					rinfo[val].req_state = srp->done + 1;
1033 					rinfo[val].problem =
1034 					    srp->header.masked_status &
1035 					    srp->header.host_status &
1036 					    srp->header.driver_status;
1037 					if (srp->done)
1038 						rinfo[val].duration =
1039 							srp->header.duration;
1040 					else {
1041 						ms = jiffies_to_msecs(jiffies);
1042 						rinfo[val].duration =
1043 						    (ms > srp->header.duration) ?
1044 						    (ms - srp->header.duration) : 0;
1045 					}
1046 					rinfo[val].orphan = srp->orphan;
1047 					rinfo[val].sg_io_owned =
1048 							srp->sg_io_owned;
1049 					rinfo[val].pack_id =
1050 							srp->header.pack_id;
1051 					rinfo[val].usr_ptr =
1052 							srp->header.usr_ptr;
1053 				}
1054 			}
1055 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1056 			result = __copy_to_user(p, rinfo,
1057 						SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1058 			result = result ? -EFAULT : 0;
1059 			kfree(rinfo);
1060 			return result;
1061 		}
1062 	case SG_EMULATED_HOST:
1063 		if (atomic_read(&sdp->detaching))
1064 			return -ENODEV;
1065 		return put_user(sdp->device->host->hostt->emulated, ip);
1066 	case SCSI_IOCTL_SEND_COMMAND:
1067 		if (atomic_read(&sdp->detaching))
1068 			return -ENODEV;
1069 		if (read_only) {
1070 			unsigned char opcode = WRITE_6;
1071 			Scsi_Ioctl_Command __user *siocp = p;
1072 
1073 			if (copy_from_user(&opcode, siocp->data, 1))
1074 				return -EFAULT;
1075 			if (sg_allow_access(filp, &opcode))
1076 				return -EPERM;
1077 		}
1078 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1079 	case SG_SET_DEBUG:
1080 		result = get_user(val, ip);
1081 		if (result)
1082 			return result;
1083 		sdp->sgdebug = (char) val;
1084 		return 0;
1085 	case BLKSECTGET:
1086 		return put_user(max_sectors_bytes(sdp->device->request_queue),
1087 				ip);
1088 	case BLKTRACESETUP:
1089 		return blk_trace_setup(sdp->device->request_queue,
1090 				       sdp->disk->disk_name,
1091 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1092 				       NULL,
1093 				       (char *)arg);
1094 	case BLKTRACESTART:
1095 		return blk_trace_startstop(sdp->device->request_queue, 1);
1096 	case BLKTRACESTOP:
1097 		return blk_trace_startstop(sdp->device->request_queue, 0);
1098 	case BLKTRACETEARDOWN:
1099 		return blk_trace_remove(sdp->device->request_queue);
1100 	case SCSI_IOCTL_GET_IDLUN:
1101 	case SCSI_IOCTL_GET_BUS_NUMBER:
1102 	case SCSI_IOCTL_PROBE_HOST:
1103 	case SG_GET_TRANSFORM:
1104 	case SG_SCSI_RESET:
1105 		if (atomic_read(&sdp->detaching))
1106 			return -ENODEV;
1107 		break;
1108 	default:
1109 		if (read_only)
1110 			return -EPERM;	/* don't know so take safe approach */
1111 		break;
1112 	}
1113 
1114 	result = scsi_ioctl_block_when_processing_errors(sdp->device,
1115 			cmd_in, filp->f_flags & O_NDELAY);
1116 	if (result)
1117 		return result;
1118 	return scsi_ioctl(sdp->device, cmd_in, p);
1119 }
1120 
1121 #ifdef CONFIG_COMPAT
1122 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1123 {
1124 	Sg_device *sdp;
1125 	Sg_fd *sfp;
1126 	struct scsi_device *sdev;
1127 
1128 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1129 		return -ENXIO;
1130 
1131 	sdev = sdp->device;
1132 	if (sdev->host->hostt->compat_ioctl) {
1133 		int ret;
1134 
1135 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1136 
1137 		return ret;
1138 	}
1139 
1140 	return -ENOIOCTLCMD;
1141 }
1142 #endif
1143 
1144 static unsigned int
1145 sg_poll(struct file *filp, poll_table * wait)
1146 {
1147 	unsigned int res = 0;
1148 	Sg_device *sdp;
1149 	Sg_fd *sfp;
1150 	Sg_request *srp;
1151 	int count = 0;
1152 	unsigned long iflags;
1153 
1154 	sfp = filp->private_data;
1155 	if (!sfp)
1156 		return POLLERR;
1157 	sdp = sfp->parentdp;
1158 	if (!sdp)
1159 		return POLLERR;
1160 	poll_wait(filp, &sfp->read_wait, wait);
1161 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1162 	for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1163 		/* if any read waiting, flag it */
1164 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1165 			res = POLLIN | POLLRDNORM;
1166 		++count;
1167 	}
1168 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1169 
1170 	if (atomic_read(&sdp->detaching))
1171 		res |= POLLHUP;
1172 	else if (!sfp->cmd_q) {
1173 		if (0 == count)
1174 			res |= POLLOUT | POLLWRNORM;
1175 	} else if (count < SG_MAX_QUEUE)
1176 		res |= POLLOUT | POLLWRNORM;
1177 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1178 				      "sg_poll: res=0x%x\n", (int) res));
1179 	return res;
1180 }
1181 
1182 static int
1183 sg_fasync(int fd, struct file *filp, int mode)
1184 {
1185 	Sg_device *sdp;
1186 	Sg_fd *sfp;
1187 
1188 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1189 		return -ENXIO;
1190 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1191 				      "sg_fasync: mode=%d\n", mode));
1192 
1193 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1194 }
1195 
1196 static int
1197 sg_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1198 {
1199 	Sg_fd *sfp;
1200 	unsigned long offset, len, sa;
1201 	Sg_scatter_hold *rsv_schp;
1202 	int k, length;
1203 
1204 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1205 		return VM_FAULT_SIGBUS;
1206 	rsv_schp = &sfp->reserve;
1207 	offset = vmf->pgoff << PAGE_SHIFT;
1208 	if (offset >= rsv_schp->bufflen)
1209 		return VM_FAULT_SIGBUS;
1210 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1211 				      "sg_vma_fault: offset=%lu, scatg=%d\n",
1212 				      offset, rsv_schp->k_use_sg));
1213 	sa = vma->vm_start;
1214 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1215 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1216 		len = vma->vm_end - sa;
1217 		len = (len < length) ? len : length;
1218 		if (offset < len) {
1219 			struct page *page = nth_page(rsv_schp->pages[k],
1220 						     offset >> PAGE_SHIFT);
1221 			get_page(page);	/* increment page count */
1222 			vmf->page = page;
1223 			return 0; /* success */
1224 		}
1225 		sa += len;
1226 		offset -= len;
1227 	}
1228 
1229 	return VM_FAULT_SIGBUS;
1230 }
1231 
1232 static const struct vm_operations_struct sg_mmap_vm_ops = {
1233 	.fault = sg_vma_fault,
1234 };
1235 
1236 static int
1237 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1238 {
1239 	Sg_fd *sfp;
1240 	unsigned long req_sz, len, sa;
1241 	Sg_scatter_hold *rsv_schp;
1242 	int k, length;
1243 
1244 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1245 		return -ENXIO;
1246 	req_sz = vma->vm_end - vma->vm_start;
1247 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1248 				      "sg_mmap starting, vm_start=%p, len=%d\n",
1249 				      (void *) vma->vm_start, (int) req_sz));
1250 	if (vma->vm_pgoff)
1251 		return -EINVAL;	/* want no offset */
1252 	rsv_schp = &sfp->reserve;
1253 	if (req_sz > rsv_schp->bufflen)
1254 		return -ENOMEM;	/* cannot map more than reserved buffer */
1255 
1256 	sa = vma->vm_start;
1257 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1258 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1259 		len = vma->vm_end - sa;
1260 		len = (len < length) ? len : length;
1261 		sa += len;
1262 	}
1263 
1264 	sfp->mmap_called = 1;
1265 	vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1266 	vma->vm_private_data = sfp;
1267 	vma->vm_ops = &sg_mmap_vm_ops;
1268 	return 0;
1269 }
1270 
1271 static void
1272 sg_rq_end_io_usercontext(struct work_struct *work)
1273 {
1274 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1275 	struct sg_fd *sfp = srp->parentfp;
1276 
1277 	sg_finish_rem_req(srp);
1278 	kref_put(&sfp->f_ref, sg_remove_sfp);
1279 }
1280 
1281 /*
1282  * This function is a "bottom half" handler that is called by the mid
1283  * level when a command is completed (or has failed).
1284  */
1285 static void
1286 sg_rq_end_io(struct request *rq, int uptodate)
1287 {
1288 	struct sg_request *srp = rq->end_io_data;
1289 	Sg_device *sdp;
1290 	Sg_fd *sfp;
1291 	unsigned long iflags;
1292 	unsigned int ms;
1293 	char *sense;
1294 	int result, resid, done = 1;
1295 
1296 	if (WARN_ON(srp->done != 0))
1297 		return;
1298 
1299 	sfp = srp->parentfp;
1300 	if (WARN_ON(sfp == NULL))
1301 		return;
1302 
1303 	sdp = sfp->parentdp;
1304 	if (unlikely(atomic_read(&sdp->detaching)))
1305 		pr_info("%s: device detaching\n", __func__);
1306 
1307 	sense = rq->sense;
1308 	result = rq->errors;
1309 	resid = rq->resid_len;
1310 
1311 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1312 				      "sg_cmd_done: pack_id=%d, res=0x%x\n",
1313 				      srp->header.pack_id, result));
1314 	srp->header.resid = resid;
1315 	ms = jiffies_to_msecs(jiffies);
1316 	srp->header.duration = (ms > srp->header.duration) ?
1317 				(ms - srp->header.duration) : 0;
1318 	if (0 != result) {
1319 		struct scsi_sense_hdr sshdr;
1320 
1321 		srp->header.status = 0xff & result;
1322 		srp->header.masked_status = status_byte(result);
1323 		srp->header.msg_status = msg_byte(result);
1324 		srp->header.host_status = host_byte(result);
1325 		srp->header.driver_status = driver_byte(result);
1326 		if ((sdp->sgdebug > 0) &&
1327 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1328 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1329 			__scsi_print_sense(sdp->device, __func__, sense,
1330 					   SCSI_SENSE_BUFFERSIZE);
1331 
1332 		/* Following if statement is a patch supplied by Eric Youngdale */
1333 		if (driver_byte(result) != 0
1334 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1335 		    && !scsi_sense_is_deferred(&sshdr)
1336 		    && sshdr.sense_key == UNIT_ATTENTION
1337 		    && sdp->device->removable) {
1338 			/* Detected possible disc change. Set the bit - this */
1339 			/* may be used if there are filesystems using this device */
1340 			sdp->device->changed = 1;
1341 		}
1342 	}
1343 	/* Rely on write phase to clean out srp status values, so no "else" */
1344 
1345 	/*
1346 	 * Free the request as soon as it is complete so that its resources
1347 	 * can be reused without waiting for userspace to read() the
1348 	 * result.  But keep the associated bio (if any) around until
1349 	 * blk_rq_unmap_user() can be called from user context.
1350 	 */
1351 	srp->rq = NULL;
1352 	if (rq->cmd != rq->__cmd)
1353 		kfree(rq->cmd);
1354 	__blk_put_request(rq->q, rq);
1355 
1356 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1357 	if (unlikely(srp->orphan)) {
1358 		if (sfp->keep_orphan)
1359 			srp->sg_io_owned = 0;
1360 		else
1361 			done = 0;
1362 	}
1363 	srp->done = done;
1364 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1365 
1366 	if (likely(done)) {
1367 		/* Now wake up any sg_read() that is waiting for this
1368 		 * packet.
1369 		 */
1370 		wake_up_interruptible(&sfp->read_wait);
1371 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1372 		kref_put(&sfp->f_ref, sg_remove_sfp);
1373 	} else {
1374 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1375 		schedule_work(&srp->ew.work);
1376 	}
1377 }
1378 
1379 static const struct file_operations sg_fops = {
1380 	.owner = THIS_MODULE,
1381 	.read = sg_read,
1382 	.write = sg_write,
1383 	.poll = sg_poll,
1384 	.unlocked_ioctl = sg_ioctl,
1385 #ifdef CONFIG_COMPAT
1386 	.compat_ioctl = sg_compat_ioctl,
1387 #endif
1388 	.open = sg_open,
1389 	.mmap = sg_mmap,
1390 	.release = sg_release,
1391 	.fasync = sg_fasync,
1392 	.llseek = no_llseek,
1393 };
1394 
1395 static struct class *sg_sysfs_class;
1396 
1397 static int sg_sysfs_valid = 0;
1398 
1399 static Sg_device *
1400 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1401 {
1402 	struct request_queue *q = scsidp->request_queue;
1403 	Sg_device *sdp;
1404 	unsigned long iflags;
1405 	int error;
1406 	u32 k;
1407 
1408 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1409 	if (!sdp) {
1410 		sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1411 			    "failure\n", __func__);
1412 		return ERR_PTR(-ENOMEM);
1413 	}
1414 
1415 	idr_preload(GFP_KERNEL);
1416 	write_lock_irqsave(&sg_index_lock, iflags);
1417 
1418 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1419 	if (error < 0) {
1420 		if (error == -ENOSPC) {
1421 			sdev_printk(KERN_WARNING, scsidp,
1422 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1423 				    scsidp->type, SG_MAX_DEVS - 1);
1424 			error = -ENODEV;
1425 		} else {
1426 			sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1427 				    "allocation Sg_device failure: %d\n",
1428 				    __func__, error);
1429 		}
1430 		goto out_unlock;
1431 	}
1432 	k = error;
1433 
1434 	SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1435 					"sg_alloc: dev=%d \n", k));
1436 	sprintf(disk->disk_name, "sg%d", k);
1437 	disk->first_minor = k;
1438 	sdp->disk = disk;
1439 	sdp->device = scsidp;
1440 	mutex_init(&sdp->open_rel_lock);
1441 	INIT_LIST_HEAD(&sdp->sfds);
1442 	init_waitqueue_head(&sdp->open_wait);
1443 	atomic_set(&sdp->detaching, 0);
1444 	rwlock_init(&sdp->sfd_lock);
1445 	sdp->sg_tablesize = queue_max_segments(q);
1446 	sdp->index = k;
1447 	kref_init(&sdp->d_ref);
1448 	error = 0;
1449 
1450 out_unlock:
1451 	write_unlock_irqrestore(&sg_index_lock, iflags);
1452 	idr_preload_end();
1453 
1454 	if (error) {
1455 		kfree(sdp);
1456 		return ERR_PTR(error);
1457 	}
1458 	return sdp;
1459 }
1460 
1461 static int
1462 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1463 {
1464 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1465 	struct gendisk *disk;
1466 	Sg_device *sdp = NULL;
1467 	struct cdev * cdev = NULL;
1468 	int error;
1469 	unsigned long iflags;
1470 
1471 	disk = alloc_disk(1);
1472 	if (!disk) {
1473 		pr_warn("%s: alloc_disk failed\n", __func__);
1474 		return -ENOMEM;
1475 	}
1476 	disk->major = SCSI_GENERIC_MAJOR;
1477 
1478 	error = -ENOMEM;
1479 	cdev = cdev_alloc();
1480 	if (!cdev) {
1481 		pr_warn("%s: cdev_alloc failed\n", __func__);
1482 		goto out;
1483 	}
1484 	cdev->owner = THIS_MODULE;
1485 	cdev->ops = &sg_fops;
1486 
1487 	sdp = sg_alloc(disk, scsidp);
1488 	if (IS_ERR(sdp)) {
1489 		pr_warn("%s: sg_alloc failed\n", __func__);
1490 		error = PTR_ERR(sdp);
1491 		goto out;
1492 	}
1493 
1494 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1495 	if (error)
1496 		goto cdev_add_err;
1497 
1498 	sdp->cdev = cdev;
1499 	if (sg_sysfs_valid) {
1500 		struct device *sg_class_member;
1501 
1502 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1503 						MKDEV(SCSI_GENERIC_MAJOR,
1504 						      sdp->index),
1505 						sdp, "%s", disk->disk_name);
1506 		if (IS_ERR(sg_class_member)) {
1507 			pr_err("%s: device_create failed\n", __func__);
1508 			error = PTR_ERR(sg_class_member);
1509 			goto cdev_add_err;
1510 		}
1511 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1512 					  &sg_class_member->kobj, "generic");
1513 		if (error)
1514 			pr_err("%s: unable to make symlink 'generic' back "
1515 			       "to sg%d\n", __func__, sdp->index);
1516 	} else
1517 		pr_warn("%s: sg_sys Invalid\n", __func__);
1518 
1519 	sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1520 		    "type %d\n", sdp->index, scsidp->type);
1521 
1522 	dev_set_drvdata(cl_dev, sdp);
1523 
1524 	return 0;
1525 
1526 cdev_add_err:
1527 	write_lock_irqsave(&sg_index_lock, iflags);
1528 	idr_remove(&sg_index_idr, sdp->index);
1529 	write_unlock_irqrestore(&sg_index_lock, iflags);
1530 	kfree(sdp);
1531 
1532 out:
1533 	put_disk(disk);
1534 	if (cdev)
1535 		cdev_del(cdev);
1536 	return error;
1537 }
1538 
1539 static void
1540 sg_device_destroy(struct kref *kref)
1541 {
1542 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1543 	unsigned long flags;
1544 
1545 	/* CAUTION!  Note that the device can still be found via idr_find()
1546 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1547 	 * any other cleanup.
1548 	 */
1549 
1550 	write_lock_irqsave(&sg_index_lock, flags);
1551 	idr_remove(&sg_index_idr, sdp->index);
1552 	write_unlock_irqrestore(&sg_index_lock, flags);
1553 
1554 	SCSI_LOG_TIMEOUT(3,
1555 		sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1556 
1557 	put_disk(sdp->disk);
1558 	kfree(sdp);
1559 }
1560 
1561 static void
1562 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1563 {
1564 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1565 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1566 	unsigned long iflags;
1567 	Sg_fd *sfp;
1568 	int val;
1569 
1570 	if (!sdp)
1571 		return;
1572 	/* want sdp->detaching non-zero as soon as possible */
1573 	val = atomic_inc_return(&sdp->detaching);
1574 	if (val > 1)
1575 		return; /* only want to do following once per device */
1576 
1577 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1578 				      "%s\n", __func__));
1579 
1580 	read_lock_irqsave(&sdp->sfd_lock, iflags);
1581 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1582 		wake_up_interruptible_all(&sfp->read_wait);
1583 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1584 	}
1585 	wake_up_interruptible_all(&sdp->open_wait);
1586 	read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1587 
1588 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1589 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1590 	cdev_del(sdp->cdev);
1591 	sdp->cdev = NULL;
1592 
1593 	kref_put(&sdp->d_ref, sg_device_destroy);
1594 }
1595 
1596 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1597 module_param_named(def_reserved_size, def_reserved_size, int,
1598 		   S_IRUGO | S_IWUSR);
1599 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1600 
1601 MODULE_AUTHOR("Douglas Gilbert");
1602 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1603 MODULE_LICENSE("GPL");
1604 MODULE_VERSION(SG_VERSION_STR);
1605 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1606 
1607 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1608                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1609 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1610 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1611 
1612 static int __init
1613 init_sg(void)
1614 {
1615 	int rc;
1616 
1617 	if (scatter_elem_sz < PAGE_SIZE) {
1618 		scatter_elem_sz = PAGE_SIZE;
1619 		scatter_elem_sz_prev = scatter_elem_sz;
1620 	}
1621 	if (def_reserved_size >= 0)
1622 		sg_big_buff = def_reserved_size;
1623 	else
1624 		def_reserved_size = sg_big_buff;
1625 
1626 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1627 				    SG_MAX_DEVS, "sg");
1628 	if (rc)
1629 		return rc;
1630         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1631         if ( IS_ERR(sg_sysfs_class) ) {
1632 		rc = PTR_ERR(sg_sysfs_class);
1633 		goto err_out;
1634         }
1635 	sg_sysfs_valid = 1;
1636 	rc = scsi_register_interface(&sg_interface);
1637 	if (0 == rc) {
1638 #ifdef CONFIG_SCSI_PROC_FS
1639 		sg_proc_init();
1640 #endif				/* CONFIG_SCSI_PROC_FS */
1641 		return 0;
1642 	}
1643 	class_destroy(sg_sysfs_class);
1644 err_out:
1645 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1646 	return rc;
1647 }
1648 
1649 static void __exit
1650 exit_sg(void)
1651 {
1652 #ifdef CONFIG_SCSI_PROC_FS
1653 	sg_proc_cleanup();
1654 #endif				/* CONFIG_SCSI_PROC_FS */
1655 	scsi_unregister_interface(&sg_interface);
1656 	class_destroy(sg_sysfs_class);
1657 	sg_sysfs_valid = 0;
1658 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1659 				 SG_MAX_DEVS);
1660 	idr_destroy(&sg_index_idr);
1661 }
1662 
1663 static int
1664 sg_start_req(Sg_request *srp, unsigned char *cmd)
1665 {
1666 	int res;
1667 	struct request *rq;
1668 	Sg_fd *sfp = srp->parentfp;
1669 	sg_io_hdr_t *hp = &srp->header;
1670 	int dxfer_len = (int) hp->dxfer_len;
1671 	int dxfer_dir = hp->dxfer_direction;
1672 	unsigned int iov_count = hp->iovec_count;
1673 	Sg_scatter_hold *req_schp = &srp->data;
1674 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1675 	struct request_queue *q = sfp->parentdp->device->request_queue;
1676 	struct rq_map_data *md, map_data;
1677 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1678 	unsigned char *long_cmdp = NULL;
1679 
1680 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1681 				      "sg_start_req: dxfer_len=%d\n",
1682 				      dxfer_len));
1683 
1684 	if (hp->cmd_len > BLK_MAX_CDB) {
1685 		long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1686 		if (!long_cmdp)
1687 			return -ENOMEM;
1688 	}
1689 
1690 	/*
1691 	 * NOTE
1692 	 *
1693 	 * With scsi-mq enabled, there are a fixed number of preallocated
1694 	 * requests equal in number to shost->can_queue.  If all of the
1695 	 * preallocated requests are already in use, then using GFP_ATOMIC with
1696 	 * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1697 	 * will cause blk_get_request() to sleep until an active command
1698 	 * completes, freeing up a request.  Neither option is ideal, but
1699 	 * GFP_KERNEL is the better choice to prevent userspace from getting an
1700 	 * unexpected EWOULDBLOCK.
1701 	 *
1702 	 * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1703 	 * does not sleep except under memory pressure.
1704 	 */
1705 	rq = blk_get_request(q, rw, GFP_KERNEL);
1706 	if (IS_ERR(rq)) {
1707 		kfree(long_cmdp);
1708 		return PTR_ERR(rq);
1709 	}
1710 
1711 	blk_rq_set_block_pc(rq);
1712 
1713 	if (hp->cmd_len > BLK_MAX_CDB)
1714 		rq->cmd = long_cmdp;
1715 	memcpy(rq->cmd, cmd, hp->cmd_len);
1716 	rq->cmd_len = hp->cmd_len;
1717 
1718 	srp->rq = rq;
1719 	rq->end_io_data = srp;
1720 	rq->sense = srp->sense_b;
1721 	rq->retries = SG_DEFAULT_RETRIES;
1722 
1723 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1724 		return 0;
1725 
1726 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1727 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1728 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1729 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1730 		md = NULL;
1731 	else
1732 		md = &map_data;
1733 
1734 	if (md) {
1735 		if (!sg_res_in_use(sfp) && dxfer_len <= rsv_schp->bufflen)
1736 			sg_link_reserve(sfp, srp, dxfer_len);
1737 		else {
1738 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1739 			if (res)
1740 				return res;
1741 		}
1742 
1743 		md->pages = req_schp->pages;
1744 		md->page_order = req_schp->page_order;
1745 		md->nr_entries = req_schp->k_use_sg;
1746 		md->offset = 0;
1747 		md->null_mapped = hp->dxferp ? 0 : 1;
1748 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1749 			md->from_user = 1;
1750 		else
1751 			md->from_user = 0;
1752 	}
1753 
1754 	if (iov_count) {
1755 		struct iovec *iov = NULL;
1756 		struct iov_iter i;
1757 
1758 		res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1759 		if (res < 0)
1760 			return res;
1761 
1762 		iov_iter_truncate(&i, hp->dxfer_len);
1763 
1764 		res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1765 		kfree(iov);
1766 	} else
1767 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1768 				      hp->dxfer_len, GFP_ATOMIC);
1769 
1770 	if (!res) {
1771 		srp->bio = rq->bio;
1772 
1773 		if (!md) {
1774 			req_schp->dio_in_use = 1;
1775 			hp->info |= SG_INFO_DIRECT_IO;
1776 		}
1777 	}
1778 	return res;
1779 }
1780 
1781 static int
1782 sg_finish_rem_req(Sg_request *srp)
1783 {
1784 	int ret = 0;
1785 
1786 	Sg_fd *sfp = srp->parentfp;
1787 	Sg_scatter_hold *req_schp = &srp->data;
1788 
1789 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1790 				      "sg_finish_rem_req: res_used=%d\n",
1791 				      (int) srp->res_used));
1792 	if (srp->bio)
1793 		ret = blk_rq_unmap_user(srp->bio);
1794 
1795 	if (srp->rq) {
1796 		if (srp->rq->cmd != srp->rq->__cmd)
1797 			kfree(srp->rq->cmd);
1798 		blk_put_request(srp->rq);
1799 	}
1800 
1801 	if (srp->res_used)
1802 		sg_unlink_reserve(sfp, srp);
1803 	else
1804 		sg_remove_scat(sfp, req_schp);
1805 
1806 	sg_remove_request(sfp, srp);
1807 
1808 	return ret;
1809 }
1810 
1811 static int
1812 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1813 {
1814 	int sg_bufflen = tablesize * sizeof(struct page *);
1815 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1816 
1817 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1818 	if (!schp->pages)
1819 		return -ENOMEM;
1820 	schp->sglist_len = sg_bufflen;
1821 	return tablesize;	/* number of scat_gath elements allocated */
1822 }
1823 
1824 static int
1825 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1826 {
1827 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1828 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1829 	int blk_size = buff_size, order;
1830 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1831 
1832 	if (blk_size < 0)
1833 		return -EFAULT;
1834 	if (0 == blk_size)
1835 		++blk_size;	/* don't know why */
1836 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1837 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1838 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1839 		"sg_build_indirect: buff_size=%d, blk_size=%d\n",
1840 		buff_size, blk_size));
1841 
1842 	/* N.B. ret_sz carried into this block ... */
1843 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1844 	if (mx_sc_elems < 0)
1845 		return mx_sc_elems;	/* most likely -ENOMEM */
1846 
1847 	num = scatter_elem_sz;
1848 	if (unlikely(num != scatter_elem_sz_prev)) {
1849 		if (num < PAGE_SIZE) {
1850 			scatter_elem_sz = PAGE_SIZE;
1851 			scatter_elem_sz_prev = PAGE_SIZE;
1852 		} else
1853 			scatter_elem_sz_prev = num;
1854 	}
1855 
1856 	if (sfp->low_dma)
1857 		gfp_mask |= GFP_DMA;
1858 
1859 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1860 		gfp_mask |= __GFP_ZERO;
1861 
1862 	order = get_order(num);
1863 retry:
1864 	ret_sz = 1 << (PAGE_SHIFT + order);
1865 
1866 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1867 	     k++, rem_sz -= ret_sz) {
1868 
1869 		num = (rem_sz > scatter_elem_sz_prev) ?
1870 			scatter_elem_sz_prev : rem_sz;
1871 
1872 		schp->pages[k] = alloc_pages(gfp_mask, order);
1873 		if (!schp->pages[k])
1874 			goto out;
1875 
1876 		if (num == scatter_elem_sz_prev) {
1877 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1878 				scatter_elem_sz = ret_sz;
1879 				scatter_elem_sz_prev = ret_sz;
1880 			}
1881 		}
1882 
1883 		SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1884 				 "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1885 				 k, num, ret_sz));
1886 	}		/* end of for loop */
1887 
1888 	schp->page_order = order;
1889 	schp->k_use_sg = k;
1890 	SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1891 			 "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1892 			 k, rem_sz));
1893 
1894 	schp->bufflen = blk_size;
1895 	if (rem_sz > 0)	/* must have failed */
1896 		return -ENOMEM;
1897 	return 0;
1898 out:
1899 	for (i = 0; i < k; i++)
1900 		__free_pages(schp->pages[i], order);
1901 
1902 	if (--order >= 0)
1903 		goto retry;
1904 
1905 	return -ENOMEM;
1906 }
1907 
1908 static void
1909 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1910 {
1911 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1912 			 "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1913 	if (schp->pages && schp->sglist_len > 0) {
1914 		if (!schp->dio_in_use) {
1915 			int k;
1916 
1917 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1918 				SCSI_LOG_TIMEOUT(5,
1919 					sg_printk(KERN_INFO, sfp->parentdp,
1920 					"sg_remove_scat: k=%d, pg=0x%p\n",
1921 					k, schp->pages[k]));
1922 				__free_pages(schp->pages[k], schp->page_order);
1923 			}
1924 
1925 			kfree(schp->pages);
1926 		}
1927 	}
1928 	memset(schp, 0, sizeof (*schp));
1929 }
1930 
1931 static int
1932 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1933 {
1934 	Sg_scatter_hold *schp = &srp->data;
1935 	int k, num;
1936 
1937 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1938 			 "sg_read_oxfer: num_read_xfer=%d\n",
1939 			 num_read_xfer));
1940 	if ((!outp) || (num_read_xfer <= 0))
1941 		return 0;
1942 
1943 	num = 1 << (PAGE_SHIFT + schp->page_order);
1944 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1945 		if (num > num_read_xfer) {
1946 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1947 					   num_read_xfer))
1948 				return -EFAULT;
1949 			break;
1950 		} else {
1951 			if (__copy_to_user(outp, page_address(schp->pages[k]),
1952 					   num))
1953 				return -EFAULT;
1954 			num_read_xfer -= num;
1955 			if (num_read_xfer <= 0)
1956 				break;
1957 			outp += num;
1958 		}
1959 	}
1960 
1961 	return 0;
1962 }
1963 
1964 static void
1965 sg_build_reserve(Sg_fd * sfp, int req_size)
1966 {
1967 	Sg_scatter_hold *schp = &sfp->reserve;
1968 
1969 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1970 			 "sg_build_reserve: req_size=%d\n", req_size));
1971 	do {
1972 		if (req_size < PAGE_SIZE)
1973 			req_size = PAGE_SIZE;
1974 		if (0 == sg_build_indirect(schp, sfp, req_size))
1975 			return;
1976 		else
1977 			sg_remove_scat(sfp, schp);
1978 		req_size >>= 1;	/* divide by 2 */
1979 	} while (req_size > (PAGE_SIZE / 2));
1980 }
1981 
1982 static void
1983 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
1984 {
1985 	Sg_scatter_hold *req_schp = &srp->data;
1986 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1987 	int k, num, rem;
1988 
1989 	srp->res_used = 1;
1990 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1991 			 "sg_link_reserve: size=%d\n", size));
1992 	rem = size;
1993 
1994 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1995 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
1996 		if (rem <= num) {
1997 			req_schp->k_use_sg = k + 1;
1998 			req_schp->sglist_len = rsv_schp->sglist_len;
1999 			req_schp->pages = rsv_schp->pages;
2000 
2001 			req_schp->bufflen = size;
2002 			req_schp->page_order = rsv_schp->page_order;
2003 			break;
2004 		} else
2005 			rem -= num;
2006 	}
2007 
2008 	if (k >= rsv_schp->k_use_sg)
2009 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2010 				 "sg_link_reserve: BAD size\n"));
2011 }
2012 
2013 static void
2014 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2015 {
2016 	Sg_scatter_hold *req_schp = &srp->data;
2017 
2018 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2019 				      "sg_unlink_reserve: req->k_use_sg=%d\n",
2020 				      (int) req_schp->k_use_sg));
2021 	req_schp->k_use_sg = 0;
2022 	req_schp->bufflen = 0;
2023 	req_schp->pages = NULL;
2024 	req_schp->page_order = 0;
2025 	req_schp->sglist_len = 0;
2026 	sfp->save_scat_len = 0;
2027 	srp->res_used = 0;
2028 }
2029 
2030 static Sg_request *
2031 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2032 {
2033 	Sg_request *resp;
2034 	unsigned long iflags;
2035 
2036 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2037 	for (resp = sfp->headrp; resp; resp = resp->nextrp) {
2038 		/* look for requests that are ready + not SG_IO owned */
2039 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
2040 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2041 			resp->done = 2;	/* guard against other readers */
2042 			break;
2043 		}
2044 	}
2045 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2046 	return resp;
2047 }
2048 
2049 /* always adds to end of list */
2050 static Sg_request *
2051 sg_add_request(Sg_fd * sfp)
2052 {
2053 	int k;
2054 	unsigned long iflags;
2055 	Sg_request *resp;
2056 	Sg_request *rp = sfp->req_arr;
2057 
2058 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2059 	resp = sfp->headrp;
2060 	if (!resp) {
2061 		memset(rp, 0, sizeof (Sg_request));
2062 		rp->parentfp = sfp;
2063 		resp = rp;
2064 		sfp->headrp = resp;
2065 	} else {
2066 		if (0 == sfp->cmd_q)
2067 			resp = NULL;	/* command queuing disallowed */
2068 		else {
2069 			for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2070 				if (!rp->parentfp)
2071 					break;
2072 			}
2073 			if (k < SG_MAX_QUEUE) {
2074 				memset(rp, 0, sizeof (Sg_request));
2075 				rp->parentfp = sfp;
2076 				while (resp->nextrp)
2077 					resp = resp->nextrp;
2078 				resp->nextrp = rp;
2079 				resp = rp;
2080 			} else
2081 				resp = NULL;
2082 		}
2083 	}
2084 	if (resp) {
2085 		resp->nextrp = NULL;
2086 		resp->header.duration = jiffies_to_msecs(jiffies);
2087 	}
2088 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2089 	return resp;
2090 }
2091 
2092 /* Return of 1 for found; 0 for not found */
2093 static int
2094 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2095 {
2096 	Sg_request *prev_rp;
2097 	Sg_request *rp;
2098 	unsigned long iflags;
2099 	int res = 0;
2100 
2101 	if ((!sfp) || (!srp) || (!sfp->headrp))
2102 		return res;
2103 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2104 	prev_rp = sfp->headrp;
2105 	if (srp == prev_rp) {
2106 		sfp->headrp = prev_rp->nextrp;
2107 		prev_rp->parentfp = NULL;
2108 		res = 1;
2109 	} else {
2110 		while ((rp = prev_rp->nextrp)) {
2111 			if (srp == rp) {
2112 				prev_rp->nextrp = rp->nextrp;
2113 				rp->parentfp = NULL;
2114 				res = 1;
2115 				break;
2116 			}
2117 			prev_rp = rp;
2118 		}
2119 	}
2120 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2121 	return res;
2122 }
2123 
2124 static Sg_fd *
2125 sg_add_sfp(Sg_device * sdp)
2126 {
2127 	Sg_fd *sfp;
2128 	unsigned long iflags;
2129 	int bufflen;
2130 
2131 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2132 	if (!sfp)
2133 		return ERR_PTR(-ENOMEM);
2134 
2135 	init_waitqueue_head(&sfp->read_wait);
2136 	rwlock_init(&sfp->rq_list_lock);
2137 
2138 	kref_init(&sfp->f_ref);
2139 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2140 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2141 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2142 	sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2143 	    sdp->device->host->unchecked_isa_dma : 1;
2144 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2145 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2146 	sfp->parentdp = sdp;
2147 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2148 	if (atomic_read(&sdp->detaching)) {
2149 		write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2150 		return ERR_PTR(-ENODEV);
2151 	}
2152 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2153 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2154 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2155 				      "sg_add_sfp: sfp=0x%p\n", sfp));
2156 	if (unlikely(sg_big_buff != def_reserved_size))
2157 		sg_big_buff = def_reserved_size;
2158 
2159 	bufflen = min_t(int, sg_big_buff,
2160 			max_sectors_bytes(sdp->device->request_queue));
2161 	sg_build_reserve(sfp, bufflen);
2162 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2163 				      "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2164 				      sfp->reserve.bufflen,
2165 				      sfp->reserve.k_use_sg));
2166 
2167 	kref_get(&sdp->d_ref);
2168 	__module_get(THIS_MODULE);
2169 	return sfp;
2170 }
2171 
2172 static void
2173 sg_remove_sfp_usercontext(struct work_struct *work)
2174 {
2175 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2176 	struct sg_device *sdp = sfp->parentdp;
2177 
2178 	/* Cleanup any responses which were never read(). */
2179 	while (sfp->headrp)
2180 		sg_finish_rem_req(sfp->headrp);
2181 
2182 	if (sfp->reserve.bufflen > 0) {
2183 		SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2184 				"sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2185 				(int) sfp->reserve.bufflen,
2186 				(int) sfp->reserve.k_use_sg));
2187 		sg_remove_scat(sfp, &sfp->reserve);
2188 	}
2189 
2190 	SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2191 			"sg_remove_sfp: sfp=0x%p\n", sfp));
2192 	kfree(sfp);
2193 
2194 	scsi_device_put(sdp->device);
2195 	kref_put(&sdp->d_ref, sg_device_destroy);
2196 	module_put(THIS_MODULE);
2197 }
2198 
2199 static void
2200 sg_remove_sfp(struct kref *kref)
2201 {
2202 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2203 	struct sg_device *sdp = sfp->parentdp;
2204 	unsigned long iflags;
2205 
2206 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2207 	list_del(&sfp->sfd_siblings);
2208 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2209 
2210 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2211 	schedule_work(&sfp->ew.work);
2212 }
2213 
2214 static int
2215 sg_res_in_use(Sg_fd * sfp)
2216 {
2217 	const Sg_request *srp;
2218 	unsigned long iflags;
2219 
2220 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
2221 	for (srp = sfp->headrp; srp; srp = srp->nextrp)
2222 		if (srp->res_used)
2223 			break;
2224 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2225 	return srp ? 1 : 0;
2226 }
2227 
2228 #ifdef CONFIG_SCSI_PROC_FS
2229 static int
2230 sg_idr_max_id(int id, void *p, void *data)
2231 {
2232 	int *k = data;
2233 
2234 	if (*k < id)
2235 		*k = id;
2236 
2237 	return 0;
2238 }
2239 
2240 static int
2241 sg_last_dev(void)
2242 {
2243 	int k = -1;
2244 	unsigned long iflags;
2245 
2246 	read_lock_irqsave(&sg_index_lock, iflags);
2247 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2248 	read_unlock_irqrestore(&sg_index_lock, iflags);
2249 	return k + 1;		/* origin 1 */
2250 }
2251 #endif
2252 
2253 /* must be called with sg_index_lock held */
2254 static Sg_device *sg_lookup_dev(int dev)
2255 {
2256 	return idr_find(&sg_index_idr, dev);
2257 }
2258 
2259 static Sg_device *
2260 sg_get_dev(int dev)
2261 {
2262 	struct sg_device *sdp;
2263 	unsigned long flags;
2264 
2265 	read_lock_irqsave(&sg_index_lock, flags);
2266 	sdp = sg_lookup_dev(dev);
2267 	if (!sdp)
2268 		sdp = ERR_PTR(-ENXIO);
2269 	else if (atomic_read(&sdp->detaching)) {
2270 		/* If sdp->detaching, then the refcount may already be 0, in
2271 		 * which case it would be a bug to do kref_get().
2272 		 */
2273 		sdp = ERR_PTR(-ENODEV);
2274 	} else
2275 		kref_get(&sdp->d_ref);
2276 	read_unlock_irqrestore(&sg_index_lock, flags);
2277 
2278 	return sdp;
2279 }
2280 
2281 #ifdef CONFIG_SCSI_PROC_FS
2282 
2283 static struct proc_dir_entry *sg_proc_sgp = NULL;
2284 
2285 static char sg_proc_sg_dirname[] = "scsi/sg";
2286 
2287 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2288 
2289 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2290 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2291 			          size_t count, loff_t *off);
2292 static const struct file_operations adio_fops = {
2293 	.owner = THIS_MODULE,
2294 	.open = sg_proc_single_open_adio,
2295 	.read = seq_read,
2296 	.llseek = seq_lseek,
2297 	.write = sg_proc_write_adio,
2298 	.release = single_release,
2299 };
2300 
2301 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2302 static ssize_t sg_proc_write_dressz(struct file *filp,
2303 		const char __user *buffer, size_t count, loff_t *off);
2304 static const struct file_operations dressz_fops = {
2305 	.owner = THIS_MODULE,
2306 	.open = sg_proc_single_open_dressz,
2307 	.read = seq_read,
2308 	.llseek = seq_lseek,
2309 	.write = sg_proc_write_dressz,
2310 	.release = single_release,
2311 };
2312 
2313 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2314 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2315 static const struct file_operations version_fops = {
2316 	.owner = THIS_MODULE,
2317 	.open = sg_proc_single_open_version,
2318 	.read = seq_read,
2319 	.llseek = seq_lseek,
2320 	.release = single_release,
2321 };
2322 
2323 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2324 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2325 static const struct file_operations devhdr_fops = {
2326 	.owner = THIS_MODULE,
2327 	.open = sg_proc_single_open_devhdr,
2328 	.read = seq_read,
2329 	.llseek = seq_lseek,
2330 	.release = single_release,
2331 };
2332 
2333 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2334 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2335 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2336 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2337 static void dev_seq_stop(struct seq_file *s, void *v);
2338 static const struct file_operations dev_fops = {
2339 	.owner = THIS_MODULE,
2340 	.open = sg_proc_open_dev,
2341 	.read = seq_read,
2342 	.llseek = seq_lseek,
2343 	.release = seq_release,
2344 };
2345 static const struct seq_operations dev_seq_ops = {
2346 	.start = dev_seq_start,
2347 	.next  = dev_seq_next,
2348 	.stop  = dev_seq_stop,
2349 	.show  = sg_proc_seq_show_dev,
2350 };
2351 
2352 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2353 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2354 static const struct file_operations devstrs_fops = {
2355 	.owner = THIS_MODULE,
2356 	.open = sg_proc_open_devstrs,
2357 	.read = seq_read,
2358 	.llseek = seq_lseek,
2359 	.release = seq_release,
2360 };
2361 static const struct seq_operations devstrs_seq_ops = {
2362 	.start = dev_seq_start,
2363 	.next  = dev_seq_next,
2364 	.stop  = dev_seq_stop,
2365 	.show  = sg_proc_seq_show_devstrs,
2366 };
2367 
2368 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2369 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2370 static const struct file_operations debug_fops = {
2371 	.owner = THIS_MODULE,
2372 	.open = sg_proc_open_debug,
2373 	.read = seq_read,
2374 	.llseek = seq_lseek,
2375 	.release = seq_release,
2376 };
2377 static const struct seq_operations debug_seq_ops = {
2378 	.start = dev_seq_start,
2379 	.next  = dev_seq_next,
2380 	.stop  = dev_seq_stop,
2381 	.show  = sg_proc_seq_show_debug,
2382 };
2383 
2384 
2385 struct sg_proc_leaf {
2386 	const char * name;
2387 	const struct file_operations * fops;
2388 };
2389 
2390 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2391 	{"allow_dio", &adio_fops},
2392 	{"debug", &debug_fops},
2393 	{"def_reserved_size", &dressz_fops},
2394 	{"device_hdr", &devhdr_fops},
2395 	{"devices", &dev_fops},
2396 	{"device_strs", &devstrs_fops},
2397 	{"version", &version_fops}
2398 };
2399 
2400 static int
2401 sg_proc_init(void)
2402 {
2403 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2404 	int k;
2405 
2406 	sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2407 	if (!sg_proc_sgp)
2408 		return 1;
2409 	for (k = 0; k < num_leaves; ++k) {
2410 		const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2411 		umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2412 		proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2413 	}
2414 	return 0;
2415 }
2416 
2417 static void
2418 sg_proc_cleanup(void)
2419 {
2420 	int k;
2421 	int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2422 
2423 	if (!sg_proc_sgp)
2424 		return;
2425 	for (k = 0; k < num_leaves; ++k)
2426 		remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2427 	remove_proc_entry(sg_proc_sg_dirname, NULL);
2428 }
2429 
2430 
2431 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2432 {
2433 	seq_printf(s, "%d\n", *((int *)s->private));
2434 	return 0;
2435 }
2436 
2437 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2438 {
2439 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2440 }
2441 
2442 static ssize_t
2443 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2444 		   size_t count, loff_t *off)
2445 {
2446 	int err;
2447 	unsigned long num;
2448 
2449 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2450 		return -EACCES;
2451 	err = kstrtoul_from_user(buffer, count, 0, &num);
2452 	if (err)
2453 		return err;
2454 	sg_allow_dio = num ? 1 : 0;
2455 	return count;
2456 }
2457 
2458 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2459 {
2460 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2461 }
2462 
2463 static ssize_t
2464 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2465 		     size_t count, loff_t *off)
2466 {
2467 	int err;
2468 	unsigned long k = ULONG_MAX;
2469 
2470 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2471 		return -EACCES;
2472 
2473 	err = kstrtoul_from_user(buffer, count, 0, &k);
2474 	if (err)
2475 		return err;
2476 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2477 		sg_big_buff = k;
2478 		return count;
2479 	}
2480 	return -ERANGE;
2481 }
2482 
2483 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2484 {
2485 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2486 		   sg_version_date);
2487 	return 0;
2488 }
2489 
2490 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2491 {
2492 	return single_open(file, sg_proc_seq_show_version, NULL);
2493 }
2494 
2495 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2496 {
2497 	seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2498 	return 0;
2499 }
2500 
2501 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2502 {
2503 	return single_open(file, sg_proc_seq_show_devhdr, NULL);
2504 }
2505 
2506 struct sg_proc_deviter {
2507 	loff_t	index;
2508 	size_t	max;
2509 };
2510 
2511 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2512 {
2513 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2514 
2515 	s->private = it;
2516 	if (! it)
2517 		return NULL;
2518 
2519 	it->index = *pos;
2520 	it->max = sg_last_dev();
2521 	if (it->index >= it->max)
2522 		return NULL;
2523 	return it;
2524 }
2525 
2526 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2527 {
2528 	struct sg_proc_deviter * it = s->private;
2529 
2530 	*pos = ++it->index;
2531 	return (it->index < it->max) ? it : NULL;
2532 }
2533 
2534 static void dev_seq_stop(struct seq_file *s, void *v)
2535 {
2536 	kfree(s->private);
2537 }
2538 
2539 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2540 {
2541         return seq_open(file, &dev_seq_ops);
2542 }
2543 
2544 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2545 {
2546 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2547 	Sg_device *sdp;
2548 	struct scsi_device *scsidp;
2549 	unsigned long iflags;
2550 
2551 	read_lock_irqsave(&sg_index_lock, iflags);
2552 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2553 	if ((NULL == sdp) || (NULL == sdp->device) ||
2554 	    (atomic_read(&sdp->detaching)))
2555 		seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2556 	else {
2557 		scsidp = sdp->device;
2558 		seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2559 			      scsidp->host->host_no, scsidp->channel,
2560 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2561 			      1,
2562 			      (int) scsidp->queue_depth,
2563 			      (int) atomic_read(&scsidp->device_busy),
2564 			      (int) scsi_device_online(scsidp));
2565 	}
2566 	read_unlock_irqrestore(&sg_index_lock, iflags);
2567 	return 0;
2568 }
2569 
2570 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2571 {
2572         return seq_open(file, &devstrs_seq_ops);
2573 }
2574 
2575 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2576 {
2577 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2578 	Sg_device *sdp;
2579 	struct scsi_device *scsidp;
2580 	unsigned long iflags;
2581 
2582 	read_lock_irqsave(&sg_index_lock, iflags);
2583 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2584 	scsidp = sdp ? sdp->device : NULL;
2585 	if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2586 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2587 			   scsidp->vendor, scsidp->model, scsidp->rev);
2588 	else
2589 		seq_puts(s, "<no active device>\n");
2590 	read_unlock_irqrestore(&sg_index_lock, iflags);
2591 	return 0;
2592 }
2593 
2594 /* must be called while holding sg_index_lock */
2595 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2596 {
2597 	int k, m, new_interface, blen, usg;
2598 	Sg_request *srp;
2599 	Sg_fd *fp;
2600 	const sg_io_hdr_t *hp;
2601 	const char * cp;
2602 	unsigned int ms;
2603 
2604 	k = 0;
2605 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2606 		k++;
2607 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2608 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2609 			   "(res)sgat=%d low_dma=%d\n", k,
2610 			   jiffies_to_msecs(fp->timeout),
2611 			   fp->reserve.bufflen,
2612 			   (int) fp->reserve.k_use_sg,
2613 			   (int) fp->low_dma);
2614 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2615 			   (int) fp->cmd_q, (int) fp->force_packid,
2616 			   (int) fp->keep_orphan);
2617 		for (m = 0, srp = fp->headrp;
2618 				srp != NULL;
2619 				++m, srp = srp->nextrp) {
2620 			hp = &srp->header;
2621 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2622 			if (srp->res_used) {
2623 				if (new_interface &&
2624 				    (SG_FLAG_MMAP_IO & hp->flags))
2625 					cp = "     mmap>> ";
2626 				else
2627 					cp = "     rb>> ";
2628 			} else {
2629 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2630 					cp = "     dio>> ";
2631 				else
2632 					cp = "     ";
2633 			}
2634 			seq_puts(s, cp);
2635 			blen = srp->data.bufflen;
2636 			usg = srp->data.k_use_sg;
2637 			seq_puts(s, srp->done ?
2638 				 ((1 == srp->done) ?  "rcv:" : "fin:")
2639 				  : "act:");
2640 			seq_printf(s, " id=%d blen=%d",
2641 				   srp->header.pack_id, blen);
2642 			if (srp->done)
2643 				seq_printf(s, " dur=%d", hp->duration);
2644 			else {
2645 				ms = jiffies_to_msecs(jiffies);
2646 				seq_printf(s, " t_o/elap=%d/%d",
2647 					(new_interface ? hp->timeout :
2648 						  jiffies_to_msecs(fp->timeout)),
2649 					(ms > hp->duration ? ms - hp->duration : 0));
2650 			}
2651 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2652 				   (int) srp->data.cmd_opcode);
2653 		}
2654 		if (0 == m)
2655 			seq_puts(s, "     No requests active\n");
2656 		read_unlock(&fp->rq_list_lock);
2657 	}
2658 }
2659 
2660 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2661 {
2662         return seq_open(file, &debug_seq_ops);
2663 }
2664 
2665 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2666 {
2667 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2668 	Sg_device *sdp;
2669 	unsigned long iflags;
2670 
2671 	if (it && (0 == it->index))
2672 		seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2673 			   (int)it->max, sg_big_buff);
2674 
2675 	read_lock_irqsave(&sg_index_lock, iflags);
2676 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2677 	if (NULL == sdp)
2678 		goto skip;
2679 	read_lock(&sdp->sfd_lock);
2680 	if (!list_empty(&sdp->sfds)) {
2681 		seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2682 		if (atomic_read(&sdp->detaching))
2683 			seq_puts(s, "detaching pending close ");
2684 		else if (sdp->device) {
2685 			struct scsi_device *scsidp = sdp->device;
2686 
2687 			seq_printf(s, "%d:%d:%d:%llu   em=%d",
2688 				   scsidp->host->host_no,
2689 				   scsidp->channel, scsidp->id,
2690 				   scsidp->lun,
2691 				   scsidp->host->hostt->emulated);
2692 		}
2693 		seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2694 			   sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2695 		sg_proc_debug_helper(s, sdp);
2696 	}
2697 	read_unlock(&sdp->sfd_lock);
2698 skip:
2699 	read_unlock_irqrestore(&sg_index_lock, iflags);
2700 	return 0;
2701 }
2702 
2703 #endif				/* CONFIG_SCSI_PROC_FS */
2704 
2705 module_init(init_sg);
2706 module_exit(exit_sg);
2707