xref: /openbmc/linux/drivers/scsi/sg.c (revision 8bd1369b)
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 #include <linux/cred.h> /* for sg_check_file_access() */
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 = "20140603";
68 
69 static int sg_proc_init(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 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
83 
84 int sg_big_buff = SG_DEF_RESERVED_SIZE;
85 /* N.B. This variable is readable and writeable via
86    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
87    of this size (or less if there is not enough memory) will be reserved
88    for use by this file descriptor. [Deprecated usage: this variable is also
89    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
90    the kernel (i.e. it is not a module).] */
91 static int def_reserved_size = -1;	/* picks up init parameter */
92 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
93 
94 static int scatter_elem_sz = SG_SCATTER_SZ;
95 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
96 
97 #define SG_SECTOR_SZ 512
98 
99 static int sg_add_device(struct device *, struct class_interface *);
100 static void sg_remove_device(struct device *, struct class_interface *);
101 
102 static DEFINE_IDR(sg_index_idr);
103 static DEFINE_RWLOCK(sg_index_lock);	/* Also used to lock
104 							   file descriptor list for device */
105 
106 static struct class_interface sg_interface = {
107 	.add_dev        = sg_add_device,
108 	.remove_dev     = sg_remove_device,
109 };
110 
111 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
112 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
113 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
114 	unsigned bufflen;	/* Size of (aggregate) data buffer */
115 	struct page **pages;
116 	int page_order;
117 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
118 	unsigned char cmd_opcode; /* first byte of command */
119 } Sg_scatter_hold;
120 
121 struct sg_device;		/* forward declarations */
122 struct sg_fd;
123 
124 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
125 	struct list_head entry;	/* list entry */
126 	struct sg_fd *parentfp;	/* NULL -> not in use */
127 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
128 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
129 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
130 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
131 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
132 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
133 	/* done protected by rq_list_lock */
134 	char done;		/* 0->before bh, 1->before read, 2->read */
135 	struct request *rq;
136 	struct bio *bio;
137 	struct execute_work ew;
138 } Sg_request;
139 
140 typedef struct sg_fd {		/* holds the state of a file descriptor */
141 	struct list_head sfd_siblings;  /* protected by device's sfd_lock */
142 	struct sg_device *parentdp;	/* owning device */
143 	wait_queue_head_t read_wait;	/* queue read until command done */
144 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
145 	struct mutex f_mutex;	/* protect against changes in this fd */
146 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
147 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
148 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
149 	struct list_head rq_list; /* head of request list */
150 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
151 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
152 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
153 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
154 	unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
155 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
156 	char mmap_called;	/* 0 -> mmap() never called on this fd */
157 	char res_in_use;	/* 1 -> 'reserve' array in use */
158 	struct kref f_ref;
159 	struct execute_work ew;
160 } Sg_fd;
161 
162 typedef struct sg_device { /* holds the state of each scsi generic device */
163 	struct scsi_device *device;
164 	wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
165 	struct mutex open_rel_lock;     /* held when in open() or release() */
166 	int sg_tablesize;	/* adapter's max scatter-gather table size */
167 	u32 index;		/* device index number */
168 	struct list_head sfds;
169 	rwlock_t sfd_lock;      /* protect access to sfd list */
170 	atomic_t detaching;     /* 0->device usable, 1->device detaching */
171 	bool exclude;		/* 1->open(O_EXCL) succeeded and is active */
172 	int open_cnt;		/* count of opens (perhaps < num(sfds) ) */
173 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
174 	struct gendisk *disk;
175 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
176 	struct kref d_ref;
177 } Sg_device;
178 
179 /* tasklet or soft irq callback */
180 static void sg_rq_end_io(struct request *rq, blk_status_t status);
181 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
182 static int sg_finish_rem_req(Sg_request * srp);
183 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
184 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
185 			   Sg_request * srp);
186 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
187 			const char __user *buf, size_t count, int blocking,
188 			int read_only, int sg_io_owned, Sg_request **o_srp);
189 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
190 			   unsigned char *cmnd, int timeout, int blocking);
191 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
192 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
193 static void sg_build_reserve(Sg_fd * sfp, int req_size);
194 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
195 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
196 static Sg_fd *sg_add_sfp(Sg_device * sdp);
197 static void sg_remove_sfp(struct kref *);
198 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
199 static Sg_request *sg_add_request(Sg_fd * sfp);
200 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
201 static Sg_device *sg_get_dev(int dev);
202 static void sg_device_destroy(struct kref *kref);
203 
204 #define SZ_SG_HEADER sizeof(struct sg_header)
205 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
206 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
207 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
208 
209 #define sg_printk(prefix, sdp, fmt, a...) \
210 	sdev_prefix_printk(prefix, (sdp)->device,		\
211 			   (sdp)->disk->disk_name, fmt, ##a)
212 
213 /*
214  * The SCSI interfaces that use read() and write() as an asynchronous variant of
215  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
216  * to trigger read() and write() calls from various contexts with elevated
217  * privileges. This can lead to kernel memory corruption (e.g. if these
218  * interfaces are called through splice()) and privilege escalation inside
219  * userspace (e.g. if a process with access to such a device passes a file
220  * descriptor to a SUID binary as stdin/stdout/stderr).
221  *
222  * This function provides protection for the legacy API by restricting the
223  * calling context.
224  */
225 static int sg_check_file_access(struct file *filp, const char *caller)
226 {
227 	if (filp->f_cred != current_real_cred()) {
228 		pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
229 			caller, task_tgid_vnr(current), current->comm);
230 		return -EPERM;
231 	}
232 	if (uaccess_kernel()) {
233 		pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
234 			caller, task_tgid_vnr(current), current->comm);
235 		return -EACCES;
236 	}
237 	return 0;
238 }
239 
240 static int sg_allow_access(struct file *filp, unsigned char *cmd)
241 {
242 	struct sg_fd *sfp = filp->private_data;
243 
244 	if (sfp->parentdp->device->type == TYPE_SCANNER)
245 		return 0;
246 
247 	return blk_verify_command(cmd, filp->f_mode);
248 }
249 
250 static int
251 open_wait(Sg_device *sdp, int flags)
252 {
253 	int retval = 0;
254 
255 	if (flags & O_EXCL) {
256 		while (sdp->open_cnt > 0) {
257 			mutex_unlock(&sdp->open_rel_lock);
258 			retval = wait_event_interruptible(sdp->open_wait,
259 					(atomic_read(&sdp->detaching) ||
260 					 !sdp->open_cnt));
261 			mutex_lock(&sdp->open_rel_lock);
262 
263 			if (retval) /* -ERESTARTSYS */
264 				return retval;
265 			if (atomic_read(&sdp->detaching))
266 				return -ENODEV;
267 		}
268 	} else {
269 		while (sdp->exclude) {
270 			mutex_unlock(&sdp->open_rel_lock);
271 			retval = wait_event_interruptible(sdp->open_wait,
272 					(atomic_read(&sdp->detaching) ||
273 					 !sdp->exclude));
274 			mutex_lock(&sdp->open_rel_lock);
275 
276 			if (retval) /* -ERESTARTSYS */
277 				return retval;
278 			if (atomic_read(&sdp->detaching))
279 				return -ENODEV;
280 		}
281 	}
282 
283 	return retval;
284 }
285 
286 /* Returns 0 on success, else a negated errno value */
287 static int
288 sg_open(struct inode *inode, struct file *filp)
289 {
290 	int dev = iminor(inode);
291 	int flags = filp->f_flags;
292 	struct request_queue *q;
293 	Sg_device *sdp;
294 	Sg_fd *sfp;
295 	int retval;
296 
297 	nonseekable_open(inode, filp);
298 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
299 		return -EPERM; /* Can't lock it with read only access */
300 	sdp = sg_get_dev(dev);
301 	if (IS_ERR(sdp))
302 		return PTR_ERR(sdp);
303 
304 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
305 				      "sg_open: flags=0x%x\n", flags));
306 
307 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
308 	/* Prevent the device driver from vanishing while we sleep */
309 	retval = scsi_device_get(sdp->device);
310 	if (retval)
311 		goto sg_put;
312 
313 	retval = scsi_autopm_get_device(sdp->device);
314 	if (retval)
315 		goto sdp_put;
316 
317 	/* scsi_block_when_processing_errors() may block so bypass
318 	 * check if O_NONBLOCK. Permits SCSI commands to be issued
319 	 * during error recovery. Tread carefully. */
320 	if (!((flags & O_NONBLOCK) ||
321 	      scsi_block_when_processing_errors(sdp->device))) {
322 		retval = -ENXIO;
323 		/* we are in error recovery for this device */
324 		goto error_out;
325 	}
326 
327 	mutex_lock(&sdp->open_rel_lock);
328 	if (flags & O_NONBLOCK) {
329 		if (flags & O_EXCL) {
330 			if (sdp->open_cnt > 0) {
331 				retval = -EBUSY;
332 				goto error_mutex_locked;
333 			}
334 		} else {
335 			if (sdp->exclude) {
336 				retval = -EBUSY;
337 				goto error_mutex_locked;
338 			}
339 		}
340 	} else {
341 		retval = open_wait(sdp, flags);
342 		if (retval) /* -ERESTARTSYS or -ENODEV */
343 			goto error_mutex_locked;
344 	}
345 
346 	/* N.B. at this point we are holding the open_rel_lock */
347 	if (flags & O_EXCL)
348 		sdp->exclude = true;
349 
350 	if (sdp->open_cnt < 1) {  /* no existing opens */
351 		sdp->sgdebug = 0;
352 		q = sdp->device->request_queue;
353 		sdp->sg_tablesize = queue_max_segments(q);
354 	}
355 	sfp = sg_add_sfp(sdp);
356 	if (IS_ERR(sfp)) {
357 		retval = PTR_ERR(sfp);
358 		goto out_undo;
359 	}
360 
361 	filp->private_data = sfp;
362 	sdp->open_cnt++;
363 	mutex_unlock(&sdp->open_rel_lock);
364 
365 	retval = 0;
366 sg_put:
367 	kref_put(&sdp->d_ref, sg_device_destroy);
368 	return retval;
369 
370 out_undo:
371 	if (flags & O_EXCL) {
372 		sdp->exclude = false;   /* undo if error */
373 		wake_up_interruptible(&sdp->open_wait);
374 	}
375 error_mutex_locked:
376 	mutex_unlock(&sdp->open_rel_lock);
377 error_out:
378 	scsi_autopm_put_device(sdp->device);
379 sdp_put:
380 	scsi_device_put(sdp->device);
381 	goto sg_put;
382 }
383 
384 /* Release resources associated with a successful sg_open()
385  * Returns 0 on success, else a negated errno value */
386 static int
387 sg_release(struct inode *inode, struct file *filp)
388 {
389 	Sg_device *sdp;
390 	Sg_fd *sfp;
391 
392 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
393 		return -ENXIO;
394 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
395 
396 	mutex_lock(&sdp->open_rel_lock);
397 	scsi_autopm_put_device(sdp->device);
398 	kref_put(&sfp->f_ref, sg_remove_sfp);
399 	sdp->open_cnt--;
400 
401 	/* possibly many open()s waiting on exlude clearing, start many;
402 	 * only open(O_EXCL)s wait on 0==open_cnt so only start one */
403 	if (sdp->exclude) {
404 		sdp->exclude = false;
405 		wake_up_interruptible_all(&sdp->open_wait);
406 	} else if (0 == sdp->open_cnt) {
407 		wake_up_interruptible(&sdp->open_wait);
408 	}
409 	mutex_unlock(&sdp->open_rel_lock);
410 	return 0;
411 }
412 
413 static ssize_t
414 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
415 {
416 	Sg_device *sdp;
417 	Sg_fd *sfp;
418 	Sg_request *srp;
419 	int req_pack_id = -1;
420 	sg_io_hdr_t *hp;
421 	struct sg_header *old_hdr = NULL;
422 	int retval = 0;
423 
424 	/*
425 	 * This could cause a response to be stranded. Close the associated
426 	 * file descriptor to free up any resources being held.
427 	 */
428 	retval = sg_check_file_access(filp, __func__);
429 	if (retval)
430 		return retval;
431 
432 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
433 		return -ENXIO;
434 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
435 				      "sg_read: count=%d\n", (int) count));
436 
437 	if (!access_ok(VERIFY_WRITE, buf, count))
438 		return -EFAULT;
439 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
440 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
441 		if (!old_hdr)
442 			return -ENOMEM;
443 		if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
444 			retval = -EFAULT;
445 			goto free_old_hdr;
446 		}
447 		if (old_hdr->reply_len < 0) {
448 			if (count >= SZ_SG_IO_HDR) {
449 				sg_io_hdr_t *new_hdr;
450 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
451 				if (!new_hdr) {
452 					retval = -ENOMEM;
453 					goto free_old_hdr;
454 				}
455 				retval =__copy_from_user
456 				    (new_hdr, buf, SZ_SG_IO_HDR);
457 				req_pack_id = new_hdr->pack_id;
458 				kfree(new_hdr);
459 				if (retval) {
460 					retval = -EFAULT;
461 					goto free_old_hdr;
462 				}
463 			}
464 		} else
465 			req_pack_id = old_hdr->pack_id;
466 	}
467 	srp = sg_get_rq_mark(sfp, req_pack_id);
468 	if (!srp) {		/* now wait on packet to arrive */
469 		if (atomic_read(&sdp->detaching)) {
470 			retval = -ENODEV;
471 			goto free_old_hdr;
472 		}
473 		if (filp->f_flags & O_NONBLOCK) {
474 			retval = -EAGAIN;
475 			goto free_old_hdr;
476 		}
477 		retval = wait_event_interruptible(sfp->read_wait,
478 			(atomic_read(&sdp->detaching) ||
479 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
480 		if (atomic_read(&sdp->detaching)) {
481 			retval = -ENODEV;
482 			goto free_old_hdr;
483 		}
484 		if (retval) {
485 			/* -ERESTARTSYS as signal hit process */
486 			goto free_old_hdr;
487 		}
488 	}
489 	if (srp->header.interface_id != '\0') {
490 		retval = sg_new_read(sfp, buf, count, srp);
491 		goto free_old_hdr;
492 	}
493 
494 	hp = &srp->header;
495 	if (old_hdr == NULL) {
496 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
497 		if (! old_hdr) {
498 			retval = -ENOMEM;
499 			goto free_old_hdr;
500 		}
501 	}
502 	memset(old_hdr, 0, SZ_SG_HEADER);
503 	old_hdr->reply_len = (int) hp->timeout;
504 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
505 	old_hdr->pack_id = hp->pack_id;
506 	old_hdr->twelve_byte =
507 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
508 	old_hdr->target_status = hp->masked_status;
509 	old_hdr->host_status = hp->host_status;
510 	old_hdr->driver_status = hp->driver_status;
511 	if ((CHECK_CONDITION & hp->masked_status) ||
512 	    (DRIVER_SENSE & hp->driver_status))
513 		memcpy(old_hdr->sense_buffer, srp->sense_b,
514 		       sizeof (old_hdr->sense_buffer));
515 	switch (hp->host_status) {
516 	/* This setup of 'result' is for backward compatibility and is best
517 	   ignored by the user who should use target, host + driver status */
518 	case DID_OK:
519 	case DID_PASSTHROUGH:
520 	case DID_SOFT_ERROR:
521 		old_hdr->result = 0;
522 		break;
523 	case DID_NO_CONNECT:
524 	case DID_BUS_BUSY:
525 	case DID_TIME_OUT:
526 		old_hdr->result = EBUSY;
527 		break;
528 	case DID_BAD_TARGET:
529 	case DID_ABORT:
530 	case DID_PARITY:
531 	case DID_RESET:
532 	case DID_BAD_INTR:
533 		old_hdr->result = EIO;
534 		break;
535 	case DID_ERROR:
536 		old_hdr->result = (srp->sense_b[0] == 0 &&
537 				  hp->masked_status == GOOD) ? 0 : EIO;
538 		break;
539 	default:
540 		old_hdr->result = EIO;
541 		break;
542 	}
543 
544 	/* Now copy the result back to the user buffer.  */
545 	if (count >= SZ_SG_HEADER) {
546 		if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
547 			retval = -EFAULT;
548 			goto free_old_hdr;
549 		}
550 		buf += SZ_SG_HEADER;
551 		if (count > old_hdr->reply_len)
552 			count = old_hdr->reply_len;
553 		if (count > SZ_SG_HEADER) {
554 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
555 				retval = -EFAULT;
556 				goto free_old_hdr;
557 			}
558 		}
559 	} else
560 		count = (old_hdr->result == 0) ? 0 : -EIO;
561 	sg_finish_rem_req(srp);
562 	sg_remove_request(sfp, srp);
563 	retval = count;
564 free_old_hdr:
565 	kfree(old_hdr);
566 	return retval;
567 }
568 
569 static ssize_t
570 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
571 {
572 	sg_io_hdr_t *hp = &srp->header;
573 	int err = 0, err2;
574 	int len;
575 
576 	if (count < SZ_SG_IO_HDR) {
577 		err = -EINVAL;
578 		goto err_out;
579 	}
580 	hp->sb_len_wr = 0;
581 	if ((hp->mx_sb_len > 0) && hp->sbp) {
582 		if ((CHECK_CONDITION & hp->masked_status) ||
583 		    (DRIVER_SENSE & hp->driver_status)) {
584 			int sb_len = SCSI_SENSE_BUFFERSIZE;
585 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
586 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
587 			len = (len > sb_len) ? sb_len : len;
588 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
589 				err = -EFAULT;
590 				goto err_out;
591 			}
592 			hp->sb_len_wr = len;
593 		}
594 	}
595 	if (hp->masked_status || hp->host_status || hp->driver_status)
596 		hp->info |= SG_INFO_CHECK;
597 	if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
598 		err = -EFAULT;
599 		goto err_out;
600 	}
601 err_out:
602 	err2 = sg_finish_rem_req(srp);
603 	sg_remove_request(sfp, srp);
604 	return err ? : err2 ? : count;
605 }
606 
607 static ssize_t
608 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
609 {
610 	int mxsize, cmd_size, k;
611 	int input_size, blocking;
612 	unsigned char opcode;
613 	Sg_device *sdp;
614 	Sg_fd *sfp;
615 	Sg_request *srp;
616 	struct sg_header old_hdr;
617 	sg_io_hdr_t *hp;
618 	unsigned char cmnd[SG_MAX_CDB_SIZE];
619 	int retval;
620 
621 	retval = sg_check_file_access(filp, __func__);
622 	if (retval)
623 		return retval;
624 
625 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
626 		return -ENXIO;
627 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
628 				      "sg_write: count=%d\n", (int) count));
629 	if (atomic_read(&sdp->detaching))
630 		return -ENODEV;
631 	if (!((filp->f_flags & O_NONBLOCK) ||
632 	      scsi_block_when_processing_errors(sdp->device)))
633 		return -ENXIO;
634 
635 	if (!access_ok(VERIFY_READ, buf, count))
636 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
637 	if (count < SZ_SG_HEADER)
638 		return -EIO;
639 	if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
640 		return -EFAULT;
641 	blocking = !(filp->f_flags & O_NONBLOCK);
642 	if (old_hdr.reply_len < 0)
643 		return sg_new_write(sfp, filp, buf, count,
644 				    blocking, 0, 0, NULL);
645 	if (count < (SZ_SG_HEADER + 6))
646 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
647 
648 	if (!(srp = sg_add_request(sfp))) {
649 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
650 					      "sg_write: queue full\n"));
651 		return -EDOM;
652 	}
653 	buf += SZ_SG_HEADER;
654 	__get_user(opcode, buf);
655 	mutex_lock(&sfp->f_mutex);
656 	if (sfp->next_cmd_len > 0) {
657 		cmd_size = sfp->next_cmd_len;
658 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
659 	} else {
660 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
661 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
662 			cmd_size = 12;
663 	}
664 	mutex_unlock(&sfp->f_mutex);
665 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
666 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
667 /* Determine buffer size.  */
668 	input_size = count - cmd_size;
669 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
670 	mxsize -= SZ_SG_HEADER;
671 	input_size -= SZ_SG_HEADER;
672 	if (input_size < 0) {
673 		sg_remove_request(sfp, srp);
674 		return -EIO;	/* User did not pass enough bytes for this command. */
675 	}
676 	hp = &srp->header;
677 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
678 	hp->cmd_len = (unsigned char) cmd_size;
679 	hp->iovec_count = 0;
680 	hp->mx_sb_len = 0;
681 	if (input_size > 0)
682 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
683 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
684 	else
685 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
686 	hp->dxfer_len = mxsize;
687 	if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
688 	    (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
689 		hp->dxferp = (char __user *)buf + cmd_size;
690 	else
691 		hp->dxferp = NULL;
692 	hp->sbp = NULL;
693 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
694 	hp->flags = input_size;	/* structure abuse ... */
695 	hp->pack_id = old_hdr.pack_id;
696 	hp->usr_ptr = NULL;
697 	if (__copy_from_user(cmnd, buf, cmd_size))
698 		return -EFAULT;
699 	/*
700 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
701 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
702 	 * is a non-zero input_size, so emit a warning.
703 	 */
704 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
705 		printk_ratelimited(KERN_WARNING
706 				   "sg_write: data in/out %d/%d bytes "
707 				   "for SCSI command 0x%x-- guessing "
708 				   "data in;\n   program %s not setting "
709 				   "count and/or reply_len properly\n",
710 				   old_hdr.reply_len - (int)SZ_SG_HEADER,
711 				   input_size, (unsigned int) cmnd[0],
712 				   current->comm);
713 	}
714 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
715 	return (k < 0) ? k : count;
716 }
717 
718 static ssize_t
719 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
720 		 size_t count, int blocking, int read_only, int sg_io_owned,
721 		 Sg_request **o_srp)
722 {
723 	int k;
724 	Sg_request *srp;
725 	sg_io_hdr_t *hp;
726 	unsigned char cmnd[SG_MAX_CDB_SIZE];
727 	int timeout;
728 	unsigned long ul_timeout;
729 
730 	if (count < SZ_SG_IO_HDR)
731 		return -EINVAL;
732 	if (!access_ok(VERIFY_READ, buf, count))
733 		return -EFAULT; /* protects following copy_from_user()s + get_user()s */
734 
735 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
736 	if (!(srp = sg_add_request(sfp))) {
737 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
738 					      "sg_new_write: queue full\n"));
739 		return -EDOM;
740 	}
741 	srp->sg_io_owned = sg_io_owned;
742 	hp = &srp->header;
743 	if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
744 		sg_remove_request(sfp, srp);
745 		return -EFAULT;
746 	}
747 	if (hp->interface_id != 'S') {
748 		sg_remove_request(sfp, srp);
749 		return -ENOSYS;
750 	}
751 	if (hp->flags & SG_FLAG_MMAP_IO) {
752 		if (hp->dxfer_len > sfp->reserve.bufflen) {
753 			sg_remove_request(sfp, srp);
754 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
755 		}
756 		if (hp->flags & SG_FLAG_DIRECT_IO) {
757 			sg_remove_request(sfp, srp);
758 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
759 		}
760 		if (sfp->res_in_use) {
761 			sg_remove_request(sfp, srp);
762 			return -EBUSY;	/* reserve buffer already being used */
763 		}
764 	}
765 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
766 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
767 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
768 		sg_remove_request(sfp, srp);
769 		return -EMSGSIZE;
770 	}
771 	if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
772 		sg_remove_request(sfp, srp);
773 		return -EFAULT;	/* protects following copy_from_user()s + get_user()s */
774 	}
775 	if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
776 		sg_remove_request(sfp, srp);
777 		return -EFAULT;
778 	}
779 	if (read_only && sg_allow_access(file, cmnd)) {
780 		sg_remove_request(sfp, srp);
781 		return -EPERM;
782 	}
783 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
784 	if (k < 0)
785 		return k;
786 	if (o_srp)
787 		*o_srp = srp;
788 	return count;
789 }
790 
791 static int
792 sg_common_write(Sg_fd * sfp, Sg_request * srp,
793 		unsigned char *cmnd, int timeout, int blocking)
794 {
795 	int k, at_head;
796 	Sg_device *sdp = sfp->parentdp;
797 	sg_io_hdr_t *hp = &srp->header;
798 
799 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
800 	hp->status = 0;
801 	hp->masked_status = 0;
802 	hp->msg_status = 0;
803 	hp->info = 0;
804 	hp->host_status = 0;
805 	hp->driver_status = 0;
806 	hp->resid = 0;
807 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
808 			"sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
809 			(int) cmnd[0], (int) hp->cmd_len));
810 
811 	if (hp->dxfer_len >= SZ_256M)
812 		return -EINVAL;
813 
814 	k = sg_start_req(srp, cmnd);
815 	if (k) {
816 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
817 			"sg_common_write: start_req err=%d\n", k));
818 		sg_finish_rem_req(srp);
819 		sg_remove_request(sfp, srp);
820 		return k;	/* probably out of space --> ENOMEM */
821 	}
822 	if (atomic_read(&sdp->detaching)) {
823 		if (srp->bio) {
824 			scsi_req_free_cmd(scsi_req(srp->rq));
825 			blk_end_request_all(srp->rq, BLK_STS_IOERR);
826 			srp->rq = NULL;
827 		}
828 
829 		sg_finish_rem_req(srp);
830 		sg_remove_request(sfp, srp);
831 		return -ENODEV;
832 	}
833 
834 	hp->duration = jiffies_to_msecs(jiffies);
835 	if (hp->interface_id != '\0' &&	/* v3 (or later) interface */
836 	    (SG_FLAG_Q_AT_TAIL & hp->flags))
837 		at_head = 0;
838 	else
839 		at_head = 1;
840 
841 	srp->rq->timeout = timeout;
842 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
843 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
844 			      srp->rq, at_head, sg_rq_end_io);
845 	return 0;
846 }
847 
848 static int srp_done(Sg_fd *sfp, Sg_request *srp)
849 {
850 	unsigned long flags;
851 	int ret;
852 
853 	read_lock_irqsave(&sfp->rq_list_lock, flags);
854 	ret = srp->done;
855 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
856 	return ret;
857 }
858 
859 static int max_sectors_bytes(struct request_queue *q)
860 {
861 	unsigned int max_sectors = queue_max_sectors(q);
862 
863 	max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
864 
865 	return max_sectors << 9;
866 }
867 
868 static void
869 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
870 {
871 	Sg_request *srp;
872 	int val;
873 	unsigned int ms;
874 
875 	val = 0;
876 	list_for_each_entry(srp, &sfp->rq_list, entry) {
877 		if (val >= SG_MAX_QUEUE)
878 			break;
879 		rinfo[val].req_state = srp->done + 1;
880 		rinfo[val].problem =
881 			srp->header.masked_status &
882 			srp->header.host_status &
883 			srp->header.driver_status;
884 		if (srp->done)
885 			rinfo[val].duration =
886 				srp->header.duration;
887 		else {
888 			ms = jiffies_to_msecs(jiffies);
889 			rinfo[val].duration =
890 				(ms > srp->header.duration) ?
891 				(ms - srp->header.duration) : 0;
892 		}
893 		rinfo[val].orphan = srp->orphan;
894 		rinfo[val].sg_io_owned = srp->sg_io_owned;
895 		rinfo[val].pack_id = srp->header.pack_id;
896 		rinfo[val].usr_ptr = srp->header.usr_ptr;
897 		val++;
898 	}
899 }
900 
901 static long
902 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
903 {
904 	void __user *p = (void __user *)arg;
905 	int __user *ip = p;
906 	int result, val, read_only;
907 	Sg_device *sdp;
908 	Sg_fd *sfp;
909 	Sg_request *srp;
910 	unsigned long iflags;
911 
912 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
913 		return -ENXIO;
914 
915 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
916 				   "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
917 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
918 
919 	switch (cmd_in) {
920 	case SG_IO:
921 		if (atomic_read(&sdp->detaching))
922 			return -ENODEV;
923 		if (!scsi_block_when_processing_errors(sdp->device))
924 			return -ENXIO;
925 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
926 			return -EFAULT;
927 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
928 				 1, read_only, 1, &srp);
929 		if (result < 0)
930 			return result;
931 		result = wait_event_interruptible(sfp->read_wait,
932 			(srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
933 		if (atomic_read(&sdp->detaching))
934 			return -ENODEV;
935 		write_lock_irq(&sfp->rq_list_lock);
936 		if (srp->done) {
937 			srp->done = 2;
938 			write_unlock_irq(&sfp->rq_list_lock);
939 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
940 			return (result < 0) ? result : 0;
941 		}
942 		srp->orphan = 1;
943 		write_unlock_irq(&sfp->rq_list_lock);
944 		return result;	/* -ERESTARTSYS because signal hit process */
945 	case SG_SET_TIMEOUT:
946 		result = get_user(val, ip);
947 		if (result)
948 			return result;
949 		if (val < 0)
950 			return -EIO;
951 		if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
952 			val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
953 				    INT_MAX);
954 		sfp->timeout_user = val;
955 		sfp->timeout = mult_frac(val, HZ, USER_HZ);
956 
957 		return 0;
958 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
959 				/* strange ..., for backward compatibility */
960 		return sfp->timeout_user;
961 	case SG_SET_FORCE_LOW_DMA:
962 		/*
963 		 * N.B. This ioctl never worked properly, but failed to
964 		 * return an error value. So returning '0' to keep compability
965 		 * with legacy applications.
966 		 */
967 		return 0;
968 	case SG_GET_LOW_DMA:
969 		return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
970 	case SG_GET_SCSI_ID:
971 		if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
972 			return -EFAULT;
973 		else {
974 			sg_scsi_id_t __user *sg_idp = p;
975 
976 			if (atomic_read(&sdp->detaching))
977 				return -ENODEV;
978 			__put_user((int) sdp->device->host->host_no,
979 				   &sg_idp->host_no);
980 			__put_user((int) sdp->device->channel,
981 				   &sg_idp->channel);
982 			__put_user((int) sdp->device->id, &sg_idp->scsi_id);
983 			__put_user((int) sdp->device->lun, &sg_idp->lun);
984 			__put_user((int) sdp->device->type, &sg_idp->scsi_type);
985 			__put_user((short) sdp->device->host->cmd_per_lun,
986 				   &sg_idp->h_cmd_per_lun);
987 			__put_user((short) sdp->device->queue_depth,
988 				   &sg_idp->d_queue_depth);
989 			__put_user(0, &sg_idp->unused[0]);
990 			__put_user(0, &sg_idp->unused[1]);
991 			return 0;
992 		}
993 	case SG_SET_FORCE_PACK_ID:
994 		result = get_user(val, ip);
995 		if (result)
996 			return result;
997 		sfp->force_packid = val ? 1 : 0;
998 		return 0;
999 	case SG_GET_PACK_ID:
1000 		if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
1001 			return -EFAULT;
1002 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
1003 		list_for_each_entry(srp, &sfp->rq_list, entry) {
1004 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
1005 				read_unlock_irqrestore(&sfp->rq_list_lock,
1006 						       iflags);
1007 				__put_user(srp->header.pack_id, ip);
1008 				return 0;
1009 			}
1010 		}
1011 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1012 		__put_user(-1, ip);
1013 		return 0;
1014 	case SG_GET_NUM_WAITING:
1015 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
1016 		val = 0;
1017 		list_for_each_entry(srp, &sfp->rq_list, entry) {
1018 			if ((1 == srp->done) && (!srp->sg_io_owned))
1019 				++val;
1020 		}
1021 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1022 		return put_user(val, ip);
1023 	case SG_GET_SG_TABLESIZE:
1024 		return put_user(sdp->sg_tablesize, ip);
1025 	case SG_SET_RESERVED_SIZE:
1026 		result = get_user(val, ip);
1027 		if (result)
1028 			return result;
1029                 if (val < 0)
1030                         return -EINVAL;
1031 		val = min_t(int, val,
1032 			    max_sectors_bytes(sdp->device->request_queue));
1033 		mutex_lock(&sfp->f_mutex);
1034 		if (val != sfp->reserve.bufflen) {
1035 			if (sfp->mmap_called ||
1036 			    sfp->res_in_use) {
1037 				mutex_unlock(&sfp->f_mutex);
1038 				return -EBUSY;
1039 			}
1040 
1041 			sg_remove_scat(sfp, &sfp->reserve);
1042 			sg_build_reserve(sfp, val);
1043 		}
1044 		mutex_unlock(&sfp->f_mutex);
1045 		return 0;
1046 	case SG_GET_RESERVED_SIZE:
1047 		val = min_t(int, sfp->reserve.bufflen,
1048 			    max_sectors_bytes(sdp->device->request_queue));
1049 		return put_user(val, ip);
1050 	case SG_SET_COMMAND_Q:
1051 		result = get_user(val, ip);
1052 		if (result)
1053 			return result;
1054 		sfp->cmd_q = val ? 1 : 0;
1055 		return 0;
1056 	case SG_GET_COMMAND_Q:
1057 		return put_user((int) sfp->cmd_q, ip);
1058 	case SG_SET_KEEP_ORPHAN:
1059 		result = get_user(val, ip);
1060 		if (result)
1061 			return result;
1062 		sfp->keep_orphan = val;
1063 		return 0;
1064 	case SG_GET_KEEP_ORPHAN:
1065 		return put_user((int) sfp->keep_orphan, ip);
1066 	case SG_NEXT_CMD_LEN:
1067 		result = get_user(val, ip);
1068 		if (result)
1069 			return result;
1070 		if (val > SG_MAX_CDB_SIZE)
1071 			return -ENOMEM;
1072 		sfp->next_cmd_len = (val > 0) ? val : 0;
1073 		return 0;
1074 	case SG_GET_VERSION_NUM:
1075 		return put_user(sg_version_num, ip);
1076 	case SG_GET_ACCESS_COUNT:
1077 		/* faked - we don't have a real access count anymore */
1078 		val = (sdp->device ? 1 : 0);
1079 		return put_user(val, ip);
1080 	case SG_GET_REQUEST_TABLE:
1081 		if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1082 			return -EFAULT;
1083 		else {
1084 			sg_req_info_t *rinfo;
1085 
1086 			rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1087 					GFP_KERNEL);
1088 			if (!rinfo)
1089 				return -ENOMEM;
1090 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
1091 			sg_fill_request_table(sfp, rinfo);
1092 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1093 			result = __copy_to_user(p, rinfo,
1094 						SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1095 			result = result ? -EFAULT : 0;
1096 			kfree(rinfo);
1097 			return result;
1098 		}
1099 	case SG_EMULATED_HOST:
1100 		if (atomic_read(&sdp->detaching))
1101 			return -ENODEV;
1102 		return put_user(sdp->device->host->hostt->emulated, ip);
1103 	case SCSI_IOCTL_SEND_COMMAND:
1104 		if (atomic_read(&sdp->detaching))
1105 			return -ENODEV;
1106 		if (read_only) {
1107 			unsigned char opcode = WRITE_6;
1108 			Scsi_Ioctl_Command __user *siocp = p;
1109 
1110 			if (copy_from_user(&opcode, siocp->data, 1))
1111 				return -EFAULT;
1112 			if (sg_allow_access(filp, &opcode))
1113 				return -EPERM;
1114 		}
1115 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1116 	case SG_SET_DEBUG:
1117 		result = get_user(val, ip);
1118 		if (result)
1119 			return result;
1120 		sdp->sgdebug = (char) val;
1121 		return 0;
1122 	case BLKSECTGET:
1123 		return put_user(max_sectors_bytes(sdp->device->request_queue),
1124 				ip);
1125 	case BLKTRACESETUP:
1126 		return blk_trace_setup(sdp->device->request_queue,
1127 				       sdp->disk->disk_name,
1128 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1129 				       NULL, p);
1130 	case BLKTRACESTART:
1131 		return blk_trace_startstop(sdp->device->request_queue, 1);
1132 	case BLKTRACESTOP:
1133 		return blk_trace_startstop(sdp->device->request_queue, 0);
1134 	case BLKTRACETEARDOWN:
1135 		return blk_trace_remove(sdp->device->request_queue);
1136 	case SCSI_IOCTL_GET_IDLUN:
1137 	case SCSI_IOCTL_GET_BUS_NUMBER:
1138 	case SCSI_IOCTL_PROBE_HOST:
1139 	case SG_GET_TRANSFORM:
1140 	case SG_SCSI_RESET:
1141 		if (atomic_read(&sdp->detaching))
1142 			return -ENODEV;
1143 		break;
1144 	default:
1145 		if (read_only)
1146 			return -EPERM;	/* don't know so take safe approach */
1147 		break;
1148 	}
1149 
1150 	result = scsi_ioctl_block_when_processing_errors(sdp->device,
1151 			cmd_in, filp->f_flags & O_NDELAY);
1152 	if (result)
1153 		return result;
1154 	return scsi_ioctl(sdp->device, cmd_in, p);
1155 }
1156 
1157 #ifdef CONFIG_COMPAT
1158 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1159 {
1160 	Sg_device *sdp;
1161 	Sg_fd *sfp;
1162 	struct scsi_device *sdev;
1163 
1164 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1165 		return -ENXIO;
1166 
1167 	sdev = sdp->device;
1168 	if (sdev->host->hostt->compat_ioctl) {
1169 		int ret;
1170 
1171 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1172 
1173 		return ret;
1174 	}
1175 
1176 	return -ENOIOCTLCMD;
1177 }
1178 #endif
1179 
1180 static __poll_t
1181 sg_poll(struct file *filp, poll_table * wait)
1182 {
1183 	__poll_t res = 0;
1184 	Sg_device *sdp;
1185 	Sg_fd *sfp;
1186 	Sg_request *srp;
1187 	int count = 0;
1188 	unsigned long iflags;
1189 
1190 	sfp = filp->private_data;
1191 	if (!sfp)
1192 		return EPOLLERR;
1193 	sdp = sfp->parentdp;
1194 	if (!sdp)
1195 		return EPOLLERR;
1196 	poll_wait(filp, &sfp->read_wait, wait);
1197 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1198 	list_for_each_entry(srp, &sfp->rq_list, entry) {
1199 		/* if any read waiting, flag it */
1200 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1201 			res = EPOLLIN | EPOLLRDNORM;
1202 		++count;
1203 	}
1204 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1205 
1206 	if (atomic_read(&sdp->detaching))
1207 		res |= EPOLLHUP;
1208 	else if (!sfp->cmd_q) {
1209 		if (0 == count)
1210 			res |= EPOLLOUT | EPOLLWRNORM;
1211 	} else if (count < SG_MAX_QUEUE)
1212 		res |= EPOLLOUT | EPOLLWRNORM;
1213 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1214 				      "sg_poll: res=0x%x\n", (__force u32) res));
1215 	return res;
1216 }
1217 
1218 static int
1219 sg_fasync(int fd, struct file *filp, int mode)
1220 {
1221 	Sg_device *sdp;
1222 	Sg_fd *sfp;
1223 
1224 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1225 		return -ENXIO;
1226 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1227 				      "sg_fasync: mode=%d\n", mode));
1228 
1229 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1230 }
1231 
1232 static vm_fault_t
1233 sg_vma_fault(struct vm_fault *vmf)
1234 {
1235 	struct vm_area_struct *vma = vmf->vma;
1236 	Sg_fd *sfp;
1237 	unsigned long offset, len, sa;
1238 	Sg_scatter_hold *rsv_schp;
1239 	int k, length;
1240 
1241 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1242 		return VM_FAULT_SIGBUS;
1243 	rsv_schp = &sfp->reserve;
1244 	offset = vmf->pgoff << PAGE_SHIFT;
1245 	if (offset >= rsv_schp->bufflen)
1246 		return VM_FAULT_SIGBUS;
1247 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1248 				      "sg_vma_fault: offset=%lu, scatg=%d\n",
1249 				      offset, rsv_schp->k_use_sg));
1250 	sa = vma->vm_start;
1251 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1252 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1253 		len = vma->vm_end - sa;
1254 		len = (len < length) ? len : length;
1255 		if (offset < len) {
1256 			struct page *page = nth_page(rsv_schp->pages[k],
1257 						     offset >> PAGE_SHIFT);
1258 			get_page(page);	/* increment page count */
1259 			vmf->page = page;
1260 			return 0; /* success */
1261 		}
1262 		sa += len;
1263 		offset -= len;
1264 	}
1265 
1266 	return VM_FAULT_SIGBUS;
1267 }
1268 
1269 static const struct vm_operations_struct sg_mmap_vm_ops = {
1270 	.fault = sg_vma_fault,
1271 };
1272 
1273 static int
1274 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1275 {
1276 	Sg_fd *sfp;
1277 	unsigned long req_sz, len, sa;
1278 	Sg_scatter_hold *rsv_schp;
1279 	int k, length;
1280 	int ret = 0;
1281 
1282 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1283 		return -ENXIO;
1284 	req_sz = vma->vm_end - vma->vm_start;
1285 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1286 				      "sg_mmap starting, vm_start=%p, len=%d\n",
1287 				      (void *) vma->vm_start, (int) req_sz));
1288 	if (vma->vm_pgoff)
1289 		return -EINVAL;	/* want no offset */
1290 	rsv_schp = &sfp->reserve;
1291 	mutex_lock(&sfp->f_mutex);
1292 	if (req_sz > rsv_schp->bufflen) {
1293 		ret = -ENOMEM;	/* cannot map more than reserved buffer */
1294 		goto out;
1295 	}
1296 
1297 	sa = vma->vm_start;
1298 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1299 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1300 		len = vma->vm_end - sa;
1301 		len = (len < length) ? len : length;
1302 		sa += len;
1303 	}
1304 
1305 	sfp->mmap_called = 1;
1306 	vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1307 	vma->vm_private_data = sfp;
1308 	vma->vm_ops = &sg_mmap_vm_ops;
1309 out:
1310 	mutex_unlock(&sfp->f_mutex);
1311 	return ret;
1312 }
1313 
1314 static void
1315 sg_rq_end_io_usercontext(struct work_struct *work)
1316 {
1317 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1318 	struct sg_fd *sfp = srp->parentfp;
1319 
1320 	sg_finish_rem_req(srp);
1321 	sg_remove_request(sfp, srp);
1322 	kref_put(&sfp->f_ref, sg_remove_sfp);
1323 }
1324 
1325 /*
1326  * This function is a "bottom half" handler that is called by the mid
1327  * level when a command is completed (or has failed).
1328  */
1329 static void
1330 sg_rq_end_io(struct request *rq, blk_status_t status)
1331 {
1332 	struct sg_request *srp = rq->end_io_data;
1333 	struct scsi_request *req = scsi_req(rq);
1334 	Sg_device *sdp;
1335 	Sg_fd *sfp;
1336 	unsigned long iflags;
1337 	unsigned int ms;
1338 	char *sense;
1339 	int result, resid, done = 1;
1340 
1341 	if (WARN_ON(srp->done != 0))
1342 		return;
1343 
1344 	sfp = srp->parentfp;
1345 	if (WARN_ON(sfp == NULL))
1346 		return;
1347 
1348 	sdp = sfp->parentdp;
1349 	if (unlikely(atomic_read(&sdp->detaching)))
1350 		pr_info("%s: device detaching\n", __func__);
1351 
1352 	sense = req->sense;
1353 	result = req->result;
1354 	resid = req->resid_len;
1355 
1356 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1357 				      "sg_cmd_done: pack_id=%d, res=0x%x\n",
1358 				      srp->header.pack_id, result));
1359 	srp->header.resid = resid;
1360 	ms = jiffies_to_msecs(jiffies);
1361 	srp->header.duration = (ms > srp->header.duration) ?
1362 				(ms - srp->header.duration) : 0;
1363 	if (0 != result) {
1364 		struct scsi_sense_hdr sshdr;
1365 
1366 		srp->header.status = 0xff & result;
1367 		srp->header.masked_status = status_byte(result);
1368 		srp->header.msg_status = msg_byte(result);
1369 		srp->header.host_status = host_byte(result);
1370 		srp->header.driver_status = driver_byte(result);
1371 		if ((sdp->sgdebug > 0) &&
1372 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1373 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1374 			__scsi_print_sense(sdp->device, __func__, sense,
1375 					   SCSI_SENSE_BUFFERSIZE);
1376 
1377 		/* Following if statement is a patch supplied by Eric Youngdale */
1378 		if (driver_byte(result) != 0
1379 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1380 		    && !scsi_sense_is_deferred(&sshdr)
1381 		    && sshdr.sense_key == UNIT_ATTENTION
1382 		    && sdp->device->removable) {
1383 			/* Detected possible disc change. Set the bit - this */
1384 			/* may be used if there are filesystems using this device */
1385 			sdp->device->changed = 1;
1386 		}
1387 	}
1388 
1389 	if (req->sense_len)
1390 		memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1391 
1392 	/* Rely on write phase to clean out srp status values, so no "else" */
1393 
1394 	/*
1395 	 * Free the request as soon as it is complete so that its resources
1396 	 * can be reused without waiting for userspace to read() the
1397 	 * result.  But keep the associated bio (if any) around until
1398 	 * blk_rq_unmap_user() can be called from user context.
1399 	 */
1400 	srp->rq = NULL;
1401 	scsi_req_free_cmd(scsi_req(rq));
1402 	__blk_put_request(rq->q, rq);
1403 
1404 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1405 	if (unlikely(srp->orphan)) {
1406 		if (sfp->keep_orphan)
1407 			srp->sg_io_owned = 0;
1408 		else
1409 			done = 0;
1410 	}
1411 	srp->done = done;
1412 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1413 
1414 	if (likely(done)) {
1415 		/* Now wake up any sg_read() that is waiting for this
1416 		 * packet.
1417 		 */
1418 		wake_up_interruptible(&sfp->read_wait);
1419 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1420 		kref_put(&sfp->f_ref, sg_remove_sfp);
1421 	} else {
1422 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1423 		schedule_work(&srp->ew.work);
1424 	}
1425 }
1426 
1427 static const struct file_operations sg_fops = {
1428 	.owner = THIS_MODULE,
1429 	.read = sg_read,
1430 	.write = sg_write,
1431 	.poll = sg_poll,
1432 	.unlocked_ioctl = sg_ioctl,
1433 #ifdef CONFIG_COMPAT
1434 	.compat_ioctl = sg_compat_ioctl,
1435 #endif
1436 	.open = sg_open,
1437 	.mmap = sg_mmap,
1438 	.release = sg_release,
1439 	.fasync = sg_fasync,
1440 	.llseek = no_llseek,
1441 };
1442 
1443 static struct class *sg_sysfs_class;
1444 
1445 static int sg_sysfs_valid = 0;
1446 
1447 static Sg_device *
1448 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1449 {
1450 	struct request_queue *q = scsidp->request_queue;
1451 	Sg_device *sdp;
1452 	unsigned long iflags;
1453 	int error;
1454 	u32 k;
1455 
1456 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1457 	if (!sdp) {
1458 		sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1459 			    "failure\n", __func__);
1460 		return ERR_PTR(-ENOMEM);
1461 	}
1462 
1463 	idr_preload(GFP_KERNEL);
1464 	write_lock_irqsave(&sg_index_lock, iflags);
1465 
1466 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1467 	if (error < 0) {
1468 		if (error == -ENOSPC) {
1469 			sdev_printk(KERN_WARNING, scsidp,
1470 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1471 				    scsidp->type, SG_MAX_DEVS - 1);
1472 			error = -ENODEV;
1473 		} else {
1474 			sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1475 				    "allocation Sg_device failure: %d\n",
1476 				    __func__, error);
1477 		}
1478 		goto out_unlock;
1479 	}
1480 	k = error;
1481 
1482 	SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1483 					"sg_alloc: dev=%d \n", k));
1484 	sprintf(disk->disk_name, "sg%d", k);
1485 	disk->first_minor = k;
1486 	sdp->disk = disk;
1487 	sdp->device = scsidp;
1488 	mutex_init(&sdp->open_rel_lock);
1489 	INIT_LIST_HEAD(&sdp->sfds);
1490 	init_waitqueue_head(&sdp->open_wait);
1491 	atomic_set(&sdp->detaching, 0);
1492 	rwlock_init(&sdp->sfd_lock);
1493 	sdp->sg_tablesize = queue_max_segments(q);
1494 	sdp->index = k;
1495 	kref_init(&sdp->d_ref);
1496 	error = 0;
1497 
1498 out_unlock:
1499 	write_unlock_irqrestore(&sg_index_lock, iflags);
1500 	idr_preload_end();
1501 
1502 	if (error) {
1503 		kfree(sdp);
1504 		return ERR_PTR(error);
1505 	}
1506 	return sdp;
1507 }
1508 
1509 static int
1510 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1511 {
1512 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1513 	struct gendisk *disk;
1514 	Sg_device *sdp = NULL;
1515 	struct cdev * cdev = NULL;
1516 	int error;
1517 	unsigned long iflags;
1518 
1519 	disk = alloc_disk(1);
1520 	if (!disk) {
1521 		pr_warn("%s: alloc_disk failed\n", __func__);
1522 		return -ENOMEM;
1523 	}
1524 	disk->major = SCSI_GENERIC_MAJOR;
1525 
1526 	error = -ENOMEM;
1527 	cdev = cdev_alloc();
1528 	if (!cdev) {
1529 		pr_warn("%s: cdev_alloc failed\n", __func__);
1530 		goto out;
1531 	}
1532 	cdev->owner = THIS_MODULE;
1533 	cdev->ops = &sg_fops;
1534 
1535 	sdp = sg_alloc(disk, scsidp);
1536 	if (IS_ERR(sdp)) {
1537 		pr_warn("%s: sg_alloc failed\n", __func__);
1538 		error = PTR_ERR(sdp);
1539 		goto out;
1540 	}
1541 
1542 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1543 	if (error)
1544 		goto cdev_add_err;
1545 
1546 	sdp->cdev = cdev;
1547 	if (sg_sysfs_valid) {
1548 		struct device *sg_class_member;
1549 
1550 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1551 						MKDEV(SCSI_GENERIC_MAJOR,
1552 						      sdp->index),
1553 						sdp, "%s", disk->disk_name);
1554 		if (IS_ERR(sg_class_member)) {
1555 			pr_err("%s: device_create failed\n", __func__);
1556 			error = PTR_ERR(sg_class_member);
1557 			goto cdev_add_err;
1558 		}
1559 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1560 					  &sg_class_member->kobj, "generic");
1561 		if (error)
1562 			pr_err("%s: unable to make symlink 'generic' back "
1563 			       "to sg%d\n", __func__, sdp->index);
1564 	} else
1565 		pr_warn("%s: sg_sys Invalid\n", __func__);
1566 
1567 	sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1568 		    "type %d\n", sdp->index, scsidp->type);
1569 
1570 	dev_set_drvdata(cl_dev, sdp);
1571 
1572 	return 0;
1573 
1574 cdev_add_err:
1575 	write_lock_irqsave(&sg_index_lock, iflags);
1576 	idr_remove(&sg_index_idr, sdp->index);
1577 	write_unlock_irqrestore(&sg_index_lock, iflags);
1578 	kfree(sdp);
1579 
1580 out:
1581 	put_disk(disk);
1582 	if (cdev)
1583 		cdev_del(cdev);
1584 	return error;
1585 }
1586 
1587 static void
1588 sg_device_destroy(struct kref *kref)
1589 {
1590 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1591 	unsigned long flags;
1592 
1593 	/* CAUTION!  Note that the device can still be found via idr_find()
1594 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1595 	 * any other cleanup.
1596 	 */
1597 
1598 	write_lock_irqsave(&sg_index_lock, flags);
1599 	idr_remove(&sg_index_idr, sdp->index);
1600 	write_unlock_irqrestore(&sg_index_lock, flags);
1601 
1602 	SCSI_LOG_TIMEOUT(3,
1603 		sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1604 
1605 	put_disk(sdp->disk);
1606 	kfree(sdp);
1607 }
1608 
1609 static void
1610 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1611 {
1612 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1613 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1614 	unsigned long iflags;
1615 	Sg_fd *sfp;
1616 	int val;
1617 
1618 	if (!sdp)
1619 		return;
1620 	/* want sdp->detaching non-zero as soon as possible */
1621 	val = atomic_inc_return(&sdp->detaching);
1622 	if (val > 1)
1623 		return; /* only want to do following once per device */
1624 
1625 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1626 				      "%s\n", __func__));
1627 
1628 	read_lock_irqsave(&sdp->sfd_lock, iflags);
1629 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1630 		wake_up_interruptible_all(&sfp->read_wait);
1631 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1632 	}
1633 	wake_up_interruptible_all(&sdp->open_wait);
1634 	read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1635 
1636 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1637 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1638 	cdev_del(sdp->cdev);
1639 	sdp->cdev = NULL;
1640 
1641 	kref_put(&sdp->d_ref, sg_device_destroy);
1642 }
1643 
1644 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1645 module_param_named(def_reserved_size, def_reserved_size, int,
1646 		   S_IRUGO | S_IWUSR);
1647 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1648 
1649 MODULE_AUTHOR("Douglas Gilbert");
1650 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1651 MODULE_LICENSE("GPL");
1652 MODULE_VERSION(SG_VERSION_STR);
1653 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1654 
1655 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1656                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1657 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1658 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1659 
1660 static int __init
1661 init_sg(void)
1662 {
1663 	int rc;
1664 
1665 	if (scatter_elem_sz < PAGE_SIZE) {
1666 		scatter_elem_sz = PAGE_SIZE;
1667 		scatter_elem_sz_prev = scatter_elem_sz;
1668 	}
1669 	if (def_reserved_size >= 0)
1670 		sg_big_buff = def_reserved_size;
1671 	else
1672 		def_reserved_size = sg_big_buff;
1673 
1674 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1675 				    SG_MAX_DEVS, "sg");
1676 	if (rc)
1677 		return rc;
1678         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1679         if ( IS_ERR(sg_sysfs_class) ) {
1680 		rc = PTR_ERR(sg_sysfs_class);
1681 		goto err_out;
1682         }
1683 	sg_sysfs_valid = 1;
1684 	rc = scsi_register_interface(&sg_interface);
1685 	if (0 == rc) {
1686 #ifdef CONFIG_SCSI_PROC_FS
1687 		sg_proc_init();
1688 #endif				/* CONFIG_SCSI_PROC_FS */
1689 		return 0;
1690 	}
1691 	class_destroy(sg_sysfs_class);
1692 err_out:
1693 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1694 	return rc;
1695 }
1696 
1697 static void __exit
1698 exit_sg(void)
1699 {
1700 #ifdef CONFIG_SCSI_PROC_FS
1701 	remove_proc_subtree("scsi/sg", NULL);
1702 #endif				/* CONFIG_SCSI_PROC_FS */
1703 	scsi_unregister_interface(&sg_interface);
1704 	class_destroy(sg_sysfs_class);
1705 	sg_sysfs_valid = 0;
1706 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1707 				 SG_MAX_DEVS);
1708 	idr_destroy(&sg_index_idr);
1709 }
1710 
1711 static int
1712 sg_start_req(Sg_request *srp, unsigned char *cmd)
1713 {
1714 	int res;
1715 	struct request *rq;
1716 	struct scsi_request *req;
1717 	Sg_fd *sfp = srp->parentfp;
1718 	sg_io_hdr_t *hp = &srp->header;
1719 	int dxfer_len = (int) hp->dxfer_len;
1720 	int dxfer_dir = hp->dxfer_direction;
1721 	unsigned int iov_count = hp->iovec_count;
1722 	Sg_scatter_hold *req_schp = &srp->data;
1723 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1724 	struct request_queue *q = sfp->parentdp->device->request_queue;
1725 	struct rq_map_data *md, map_data;
1726 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1727 	unsigned char *long_cmdp = NULL;
1728 
1729 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1730 				      "sg_start_req: dxfer_len=%d\n",
1731 				      dxfer_len));
1732 
1733 	if (hp->cmd_len > BLK_MAX_CDB) {
1734 		long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1735 		if (!long_cmdp)
1736 			return -ENOMEM;
1737 	}
1738 
1739 	/*
1740 	 * NOTE
1741 	 *
1742 	 * With scsi-mq enabled, there are a fixed number of preallocated
1743 	 * requests equal in number to shost->can_queue.  If all of the
1744 	 * preallocated requests are already in use, then using GFP_ATOMIC with
1745 	 * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1746 	 * will cause blk_get_request() to sleep until an active command
1747 	 * completes, freeing up a request.  Neither option is ideal, but
1748 	 * GFP_KERNEL is the better choice to prevent userspace from getting an
1749 	 * unexpected EWOULDBLOCK.
1750 	 *
1751 	 * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1752 	 * does not sleep except under memory pressure.
1753 	 */
1754 	rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1755 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
1756 	if (IS_ERR(rq)) {
1757 		kfree(long_cmdp);
1758 		return PTR_ERR(rq);
1759 	}
1760 	req = scsi_req(rq);
1761 
1762 	if (hp->cmd_len > BLK_MAX_CDB)
1763 		req->cmd = long_cmdp;
1764 	memcpy(req->cmd, cmd, hp->cmd_len);
1765 	req->cmd_len = hp->cmd_len;
1766 
1767 	srp->rq = rq;
1768 	rq->end_io_data = srp;
1769 	req->retries = SG_DEFAULT_RETRIES;
1770 
1771 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1772 		return 0;
1773 
1774 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1775 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1776 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1777 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1778 		md = NULL;
1779 	else
1780 		md = &map_data;
1781 
1782 	if (md) {
1783 		mutex_lock(&sfp->f_mutex);
1784 		if (dxfer_len <= rsv_schp->bufflen &&
1785 		    !sfp->res_in_use) {
1786 			sfp->res_in_use = 1;
1787 			sg_link_reserve(sfp, srp, dxfer_len);
1788 		} else if (hp->flags & SG_FLAG_MMAP_IO) {
1789 			res = -EBUSY; /* sfp->res_in_use == 1 */
1790 			if (dxfer_len > rsv_schp->bufflen)
1791 				res = -ENOMEM;
1792 			mutex_unlock(&sfp->f_mutex);
1793 			return res;
1794 		} else {
1795 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1796 			if (res) {
1797 				mutex_unlock(&sfp->f_mutex);
1798 				return res;
1799 			}
1800 		}
1801 		mutex_unlock(&sfp->f_mutex);
1802 
1803 		md->pages = req_schp->pages;
1804 		md->page_order = req_schp->page_order;
1805 		md->nr_entries = req_schp->k_use_sg;
1806 		md->offset = 0;
1807 		md->null_mapped = hp->dxferp ? 0 : 1;
1808 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1809 			md->from_user = 1;
1810 		else
1811 			md->from_user = 0;
1812 	}
1813 
1814 	if (iov_count) {
1815 		struct iovec *iov = NULL;
1816 		struct iov_iter i;
1817 
1818 		res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1819 		if (res < 0)
1820 			return res;
1821 
1822 		iov_iter_truncate(&i, hp->dxfer_len);
1823 		if (!iov_iter_count(&i)) {
1824 			kfree(iov);
1825 			return -EINVAL;
1826 		}
1827 
1828 		res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1829 		kfree(iov);
1830 	} else
1831 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1832 				      hp->dxfer_len, GFP_ATOMIC);
1833 
1834 	if (!res) {
1835 		srp->bio = rq->bio;
1836 
1837 		if (!md) {
1838 			req_schp->dio_in_use = 1;
1839 			hp->info |= SG_INFO_DIRECT_IO;
1840 		}
1841 	}
1842 	return res;
1843 }
1844 
1845 static int
1846 sg_finish_rem_req(Sg_request *srp)
1847 {
1848 	int ret = 0;
1849 
1850 	Sg_fd *sfp = srp->parentfp;
1851 	Sg_scatter_hold *req_schp = &srp->data;
1852 
1853 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1854 				      "sg_finish_rem_req: res_used=%d\n",
1855 				      (int) srp->res_used));
1856 	if (srp->bio)
1857 		ret = blk_rq_unmap_user(srp->bio);
1858 
1859 	if (srp->rq) {
1860 		scsi_req_free_cmd(scsi_req(srp->rq));
1861 		blk_put_request(srp->rq);
1862 	}
1863 
1864 	if (srp->res_used)
1865 		sg_unlink_reserve(sfp, srp);
1866 	else
1867 		sg_remove_scat(sfp, req_schp);
1868 
1869 	return ret;
1870 }
1871 
1872 static int
1873 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1874 {
1875 	int sg_bufflen = tablesize * sizeof(struct page *);
1876 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1877 
1878 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1879 	if (!schp->pages)
1880 		return -ENOMEM;
1881 	schp->sglist_len = sg_bufflen;
1882 	return tablesize;	/* number of scat_gath elements allocated */
1883 }
1884 
1885 static int
1886 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1887 {
1888 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1889 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1890 	int blk_size = buff_size, order;
1891 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1892 	struct sg_device *sdp = sfp->parentdp;
1893 
1894 	if (blk_size < 0)
1895 		return -EFAULT;
1896 	if (0 == blk_size)
1897 		++blk_size;	/* don't know why */
1898 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1899 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1900 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1901 		"sg_build_indirect: buff_size=%d, blk_size=%d\n",
1902 		buff_size, blk_size));
1903 
1904 	/* N.B. ret_sz carried into this block ... */
1905 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1906 	if (mx_sc_elems < 0)
1907 		return mx_sc_elems;	/* most likely -ENOMEM */
1908 
1909 	num = scatter_elem_sz;
1910 	if (unlikely(num != scatter_elem_sz_prev)) {
1911 		if (num < PAGE_SIZE) {
1912 			scatter_elem_sz = PAGE_SIZE;
1913 			scatter_elem_sz_prev = PAGE_SIZE;
1914 		} else
1915 			scatter_elem_sz_prev = num;
1916 	}
1917 
1918 	if (sdp->device->host->unchecked_isa_dma)
1919 		gfp_mask |= GFP_DMA;
1920 
1921 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1922 		gfp_mask |= __GFP_ZERO;
1923 
1924 	order = get_order(num);
1925 retry:
1926 	ret_sz = 1 << (PAGE_SHIFT + order);
1927 
1928 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1929 	     k++, rem_sz -= ret_sz) {
1930 
1931 		num = (rem_sz > scatter_elem_sz_prev) ?
1932 			scatter_elem_sz_prev : rem_sz;
1933 
1934 		schp->pages[k] = alloc_pages(gfp_mask | __GFP_ZERO, order);
1935 		if (!schp->pages[k])
1936 			goto out;
1937 
1938 		if (num == scatter_elem_sz_prev) {
1939 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1940 				scatter_elem_sz = ret_sz;
1941 				scatter_elem_sz_prev = ret_sz;
1942 			}
1943 		}
1944 
1945 		SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1946 				 "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1947 				 k, num, ret_sz));
1948 	}		/* end of for loop */
1949 
1950 	schp->page_order = order;
1951 	schp->k_use_sg = k;
1952 	SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1953 			 "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1954 			 k, rem_sz));
1955 
1956 	schp->bufflen = blk_size;
1957 	if (rem_sz > 0)	/* must have failed */
1958 		return -ENOMEM;
1959 	return 0;
1960 out:
1961 	for (i = 0; i < k; i++)
1962 		__free_pages(schp->pages[i], order);
1963 
1964 	if (--order >= 0)
1965 		goto retry;
1966 
1967 	return -ENOMEM;
1968 }
1969 
1970 static void
1971 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1972 {
1973 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1974 			 "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1975 	if (schp->pages && schp->sglist_len > 0) {
1976 		if (!schp->dio_in_use) {
1977 			int k;
1978 
1979 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1980 				SCSI_LOG_TIMEOUT(5,
1981 					sg_printk(KERN_INFO, sfp->parentdp,
1982 					"sg_remove_scat: k=%d, pg=0x%p\n",
1983 					k, schp->pages[k]));
1984 				__free_pages(schp->pages[k], schp->page_order);
1985 			}
1986 
1987 			kfree(schp->pages);
1988 		}
1989 	}
1990 	memset(schp, 0, sizeof (*schp));
1991 }
1992 
1993 static int
1994 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1995 {
1996 	Sg_scatter_hold *schp = &srp->data;
1997 	int k, num;
1998 
1999 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2000 			 "sg_read_oxfer: num_read_xfer=%d\n",
2001 			 num_read_xfer));
2002 	if ((!outp) || (num_read_xfer <= 0))
2003 		return 0;
2004 
2005 	num = 1 << (PAGE_SHIFT + schp->page_order);
2006 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2007 		if (num > num_read_xfer) {
2008 			if (__copy_to_user(outp, page_address(schp->pages[k]),
2009 					   num_read_xfer))
2010 				return -EFAULT;
2011 			break;
2012 		} else {
2013 			if (__copy_to_user(outp, page_address(schp->pages[k]),
2014 					   num))
2015 				return -EFAULT;
2016 			num_read_xfer -= num;
2017 			if (num_read_xfer <= 0)
2018 				break;
2019 			outp += num;
2020 		}
2021 	}
2022 
2023 	return 0;
2024 }
2025 
2026 static void
2027 sg_build_reserve(Sg_fd * sfp, int req_size)
2028 {
2029 	Sg_scatter_hold *schp = &sfp->reserve;
2030 
2031 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2032 			 "sg_build_reserve: req_size=%d\n", req_size));
2033 	do {
2034 		if (req_size < PAGE_SIZE)
2035 			req_size = PAGE_SIZE;
2036 		if (0 == sg_build_indirect(schp, sfp, req_size))
2037 			return;
2038 		else
2039 			sg_remove_scat(sfp, schp);
2040 		req_size >>= 1;	/* divide by 2 */
2041 	} while (req_size > (PAGE_SIZE / 2));
2042 }
2043 
2044 static void
2045 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2046 {
2047 	Sg_scatter_hold *req_schp = &srp->data;
2048 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
2049 	int k, num, rem;
2050 
2051 	srp->res_used = 1;
2052 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2053 			 "sg_link_reserve: size=%d\n", size));
2054 	rem = size;
2055 
2056 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2057 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
2058 		if (rem <= num) {
2059 			req_schp->k_use_sg = k + 1;
2060 			req_schp->sglist_len = rsv_schp->sglist_len;
2061 			req_schp->pages = rsv_schp->pages;
2062 
2063 			req_schp->bufflen = size;
2064 			req_schp->page_order = rsv_schp->page_order;
2065 			break;
2066 		} else
2067 			rem -= num;
2068 	}
2069 
2070 	if (k >= rsv_schp->k_use_sg)
2071 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2072 				 "sg_link_reserve: BAD size\n"));
2073 }
2074 
2075 static void
2076 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2077 {
2078 	Sg_scatter_hold *req_schp = &srp->data;
2079 
2080 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2081 				      "sg_unlink_reserve: req->k_use_sg=%d\n",
2082 				      (int) req_schp->k_use_sg));
2083 	req_schp->k_use_sg = 0;
2084 	req_schp->bufflen = 0;
2085 	req_schp->pages = NULL;
2086 	req_schp->page_order = 0;
2087 	req_schp->sglist_len = 0;
2088 	srp->res_used = 0;
2089 	/* Called without mutex lock to avoid deadlock */
2090 	sfp->res_in_use = 0;
2091 }
2092 
2093 static Sg_request *
2094 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2095 {
2096 	Sg_request *resp;
2097 	unsigned long iflags;
2098 
2099 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2100 	list_for_each_entry(resp, &sfp->rq_list, entry) {
2101 		/* look for requests that are ready + not SG_IO owned */
2102 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
2103 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2104 			resp->done = 2;	/* guard against other readers */
2105 			write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2106 			return resp;
2107 		}
2108 	}
2109 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2110 	return NULL;
2111 }
2112 
2113 /* always adds to end of list */
2114 static Sg_request *
2115 sg_add_request(Sg_fd * sfp)
2116 {
2117 	int k;
2118 	unsigned long iflags;
2119 	Sg_request *rp = sfp->req_arr;
2120 
2121 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2122 	if (!list_empty(&sfp->rq_list)) {
2123 		if (!sfp->cmd_q)
2124 			goto out_unlock;
2125 
2126 		for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2127 			if (!rp->parentfp)
2128 				break;
2129 		}
2130 		if (k >= SG_MAX_QUEUE)
2131 			goto out_unlock;
2132 	}
2133 	memset(rp, 0, sizeof (Sg_request));
2134 	rp->parentfp = sfp;
2135 	rp->header.duration = jiffies_to_msecs(jiffies);
2136 	list_add_tail(&rp->entry, &sfp->rq_list);
2137 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2138 	return rp;
2139 out_unlock:
2140 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2141 	return NULL;
2142 }
2143 
2144 /* Return of 1 for found; 0 for not found */
2145 static int
2146 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2147 {
2148 	unsigned long iflags;
2149 	int res = 0;
2150 
2151 	if (!sfp || !srp || list_empty(&sfp->rq_list))
2152 		return res;
2153 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2154 	if (!list_empty(&srp->entry)) {
2155 		list_del(&srp->entry);
2156 		srp->parentfp = NULL;
2157 		res = 1;
2158 	}
2159 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2160 	return res;
2161 }
2162 
2163 static Sg_fd *
2164 sg_add_sfp(Sg_device * sdp)
2165 {
2166 	Sg_fd *sfp;
2167 	unsigned long iflags;
2168 	int bufflen;
2169 
2170 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2171 	if (!sfp)
2172 		return ERR_PTR(-ENOMEM);
2173 
2174 	init_waitqueue_head(&sfp->read_wait);
2175 	rwlock_init(&sfp->rq_list_lock);
2176 	INIT_LIST_HEAD(&sfp->rq_list);
2177 	kref_init(&sfp->f_ref);
2178 	mutex_init(&sfp->f_mutex);
2179 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2180 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2181 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2182 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2183 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2184 	sfp->parentdp = sdp;
2185 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2186 	if (atomic_read(&sdp->detaching)) {
2187 		write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2188 		return ERR_PTR(-ENODEV);
2189 	}
2190 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2191 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2192 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2193 				      "sg_add_sfp: sfp=0x%p\n", sfp));
2194 	if (unlikely(sg_big_buff != def_reserved_size))
2195 		sg_big_buff = def_reserved_size;
2196 
2197 	bufflen = min_t(int, sg_big_buff,
2198 			max_sectors_bytes(sdp->device->request_queue));
2199 	sg_build_reserve(sfp, bufflen);
2200 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2201 				      "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2202 				      sfp->reserve.bufflen,
2203 				      sfp->reserve.k_use_sg));
2204 
2205 	kref_get(&sdp->d_ref);
2206 	__module_get(THIS_MODULE);
2207 	return sfp;
2208 }
2209 
2210 static void
2211 sg_remove_sfp_usercontext(struct work_struct *work)
2212 {
2213 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2214 	struct sg_device *sdp = sfp->parentdp;
2215 	Sg_request *srp;
2216 	unsigned long iflags;
2217 
2218 	/* Cleanup any responses which were never read(). */
2219 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2220 	while (!list_empty(&sfp->rq_list)) {
2221 		srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2222 		sg_finish_rem_req(srp);
2223 		list_del(&srp->entry);
2224 		srp->parentfp = NULL;
2225 	}
2226 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2227 
2228 	if (sfp->reserve.bufflen > 0) {
2229 		SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2230 				"sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2231 				(int) sfp->reserve.bufflen,
2232 				(int) sfp->reserve.k_use_sg));
2233 		sg_remove_scat(sfp, &sfp->reserve);
2234 	}
2235 
2236 	SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2237 			"sg_remove_sfp: sfp=0x%p\n", sfp));
2238 	kfree(sfp);
2239 
2240 	scsi_device_put(sdp->device);
2241 	kref_put(&sdp->d_ref, sg_device_destroy);
2242 	module_put(THIS_MODULE);
2243 }
2244 
2245 static void
2246 sg_remove_sfp(struct kref *kref)
2247 {
2248 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2249 	struct sg_device *sdp = sfp->parentdp;
2250 	unsigned long iflags;
2251 
2252 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2253 	list_del(&sfp->sfd_siblings);
2254 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2255 
2256 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2257 	schedule_work(&sfp->ew.work);
2258 }
2259 
2260 #ifdef CONFIG_SCSI_PROC_FS
2261 static int
2262 sg_idr_max_id(int id, void *p, void *data)
2263 {
2264 	int *k = data;
2265 
2266 	if (*k < id)
2267 		*k = id;
2268 
2269 	return 0;
2270 }
2271 
2272 static int
2273 sg_last_dev(void)
2274 {
2275 	int k = -1;
2276 	unsigned long iflags;
2277 
2278 	read_lock_irqsave(&sg_index_lock, iflags);
2279 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2280 	read_unlock_irqrestore(&sg_index_lock, iflags);
2281 	return k + 1;		/* origin 1 */
2282 }
2283 #endif
2284 
2285 /* must be called with sg_index_lock held */
2286 static Sg_device *sg_lookup_dev(int dev)
2287 {
2288 	return idr_find(&sg_index_idr, dev);
2289 }
2290 
2291 static Sg_device *
2292 sg_get_dev(int dev)
2293 {
2294 	struct sg_device *sdp;
2295 	unsigned long flags;
2296 
2297 	read_lock_irqsave(&sg_index_lock, flags);
2298 	sdp = sg_lookup_dev(dev);
2299 	if (!sdp)
2300 		sdp = ERR_PTR(-ENXIO);
2301 	else if (atomic_read(&sdp->detaching)) {
2302 		/* If sdp->detaching, then the refcount may already be 0, in
2303 		 * which case it would be a bug to do kref_get().
2304 		 */
2305 		sdp = ERR_PTR(-ENODEV);
2306 	} else
2307 		kref_get(&sdp->d_ref);
2308 	read_unlock_irqrestore(&sg_index_lock, flags);
2309 
2310 	return sdp;
2311 }
2312 
2313 #ifdef CONFIG_SCSI_PROC_FS
2314 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2315 
2316 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2317 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2318 			          size_t count, loff_t *off);
2319 static const struct file_operations adio_fops = {
2320 	.owner = THIS_MODULE,
2321 	.open = sg_proc_single_open_adio,
2322 	.read = seq_read,
2323 	.llseek = seq_lseek,
2324 	.write = sg_proc_write_adio,
2325 	.release = single_release,
2326 };
2327 
2328 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2329 static ssize_t sg_proc_write_dressz(struct file *filp,
2330 		const char __user *buffer, size_t count, loff_t *off);
2331 static const struct file_operations dressz_fops = {
2332 	.owner = THIS_MODULE,
2333 	.open = sg_proc_single_open_dressz,
2334 	.read = seq_read,
2335 	.llseek = seq_lseek,
2336 	.write = sg_proc_write_dressz,
2337 	.release = single_release,
2338 };
2339 
2340 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2341 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2342 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2343 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2344 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2345 static void dev_seq_stop(struct seq_file *s, void *v);
2346 static const struct seq_operations dev_seq_ops = {
2347 	.start = dev_seq_start,
2348 	.next  = dev_seq_next,
2349 	.stop  = dev_seq_stop,
2350 	.show  = sg_proc_seq_show_dev,
2351 };
2352 
2353 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2354 static const struct seq_operations devstrs_seq_ops = {
2355 	.start = dev_seq_start,
2356 	.next  = dev_seq_next,
2357 	.stop  = dev_seq_stop,
2358 	.show  = sg_proc_seq_show_devstrs,
2359 };
2360 
2361 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2362 static const struct seq_operations debug_seq_ops = {
2363 	.start = dev_seq_start,
2364 	.next  = dev_seq_next,
2365 	.stop  = dev_seq_stop,
2366 	.show  = sg_proc_seq_show_debug,
2367 };
2368 
2369 static int
2370 sg_proc_init(void)
2371 {
2372 	struct proc_dir_entry *p;
2373 
2374 	p = proc_mkdir("scsi/sg", NULL);
2375 	if (!p)
2376 		return 1;
2377 
2378 	proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_fops);
2379 	proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2380 	proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_fops);
2381 	proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2382 	proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2383 	proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2384 	proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2385 	return 0;
2386 }
2387 
2388 
2389 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2390 {
2391 	seq_printf(s, "%d\n", *((int *)s->private));
2392 	return 0;
2393 }
2394 
2395 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2396 {
2397 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2398 }
2399 
2400 static ssize_t
2401 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2402 		   size_t count, loff_t *off)
2403 {
2404 	int err;
2405 	unsigned long num;
2406 
2407 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2408 		return -EACCES;
2409 	err = kstrtoul_from_user(buffer, count, 0, &num);
2410 	if (err)
2411 		return err;
2412 	sg_allow_dio = num ? 1 : 0;
2413 	return count;
2414 }
2415 
2416 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2417 {
2418 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2419 }
2420 
2421 static ssize_t
2422 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2423 		     size_t count, loff_t *off)
2424 {
2425 	int err;
2426 	unsigned long k = ULONG_MAX;
2427 
2428 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2429 		return -EACCES;
2430 
2431 	err = kstrtoul_from_user(buffer, count, 0, &k);
2432 	if (err)
2433 		return err;
2434 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2435 		sg_big_buff = k;
2436 		return count;
2437 	}
2438 	return -ERANGE;
2439 }
2440 
2441 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2442 {
2443 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2444 		   sg_version_date);
2445 	return 0;
2446 }
2447 
2448 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2449 {
2450 	seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2451 	return 0;
2452 }
2453 
2454 struct sg_proc_deviter {
2455 	loff_t	index;
2456 	size_t	max;
2457 };
2458 
2459 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2460 {
2461 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2462 
2463 	s->private = it;
2464 	if (! it)
2465 		return NULL;
2466 
2467 	it->index = *pos;
2468 	it->max = sg_last_dev();
2469 	if (it->index >= it->max)
2470 		return NULL;
2471 	return it;
2472 }
2473 
2474 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2475 {
2476 	struct sg_proc_deviter * it = s->private;
2477 
2478 	*pos = ++it->index;
2479 	return (it->index < it->max) ? it : NULL;
2480 }
2481 
2482 static void dev_seq_stop(struct seq_file *s, void *v)
2483 {
2484 	kfree(s->private);
2485 }
2486 
2487 static int sg_proc_seq_show_dev(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 ((NULL == sdp) || (NULL == sdp->device) ||
2497 	    (atomic_read(&sdp->detaching)))
2498 		seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2499 	else {
2500 		scsidp = sdp->device;
2501 		seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2502 			      scsidp->host->host_no, scsidp->channel,
2503 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2504 			      1,
2505 			      (int) scsidp->queue_depth,
2506 			      (int) atomic_read(&scsidp->device_busy),
2507 			      (int) scsi_device_online(scsidp));
2508 	}
2509 	read_unlock_irqrestore(&sg_index_lock, iflags);
2510 	return 0;
2511 }
2512 
2513 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2514 {
2515 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2516 	Sg_device *sdp;
2517 	struct scsi_device *scsidp;
2518 	unsigned long iflags;
2519 
2520 	read_lock_irqsave(&sg_index_lock, iflags);
2521 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2522 	scsidp = sdp ? sdp->device : NULL;
2523 	if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2524 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2525 			   scsidp->vendor, scsidp->model, scsidp->rev);
2526 	else
2527 		seq_puts(s, "<no active device>\n");
2528 	read_unlock_irqrestore(&sg_index_lock, iflags);
2529 	return 0;
2530 }
2531 
2532 /* must be called while holding sg_index_lock */
2533 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2534 {
2535 	int k, new_interface, blen, usg;
2536 	Sg_request *srp;
2537 	Sg_fd *fp;
2538 	const sg_io_hdr_t *hp;
2539 	const char * cp;
2540 	unsigned int ms;
2541 
2542 	k = 0;
2543 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2544 		k++;
2545 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2546 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2547 			   "(res)sgat=%d low_dma=%d\n", k,
2548 			   jiffies_to_msecs(fp->timeout),
2549 			   fp->reserve.bufflen,
2550 			   (int) fp->reserve.k_use_sg,
2551 			   (int) sdp->device->host->unchecked_isa_dma);
2552 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2553 			   (int) fp->cmd_q, (int) fp->force_packid,
2554 			   (int) fp->keep_orphan);
2555 		list_for_each_entry(srp, &fp->rq_list, entry) {
2556 			hp = &srp->header;
2557 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2558 			if (srp->res_used) {
2559 				if (new_interface &&
2560 				    (SG_FLAG_MMAP_IO & hp->flags))
2561 					cp = "     mmap>> ";
2562 				else
2563 					cp = "     rb>> ";
2564 			} else {
2565 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2566 					cp = "     dio>> ";
2567 				else
2568 					cp = "     ";
2569 			}
2570 			seq_puts(s, cp);
2571 			blen = srp->data.bufflen;
2572 			usg = srp->data.k_use_sg;
2573 			seq_puts(s, srp->done ?
2574 				 ((1 == srp->done) ?  "rcv:" : "fin:")
2575 				  : "act:");
2576 			seq_printf(s, " id=%d blen=%d",
2577 				   srp->header.pack_id, blen);
2578 			if (srp->done)
2579 				seq_printf(s, " dur=%d", hp->duration);
2580 			else {
2581 				ms = jiffies_to_msecs(jiffies);
2582 				seq_printf(s, " t_o/elap=%d/%d",
2583 					(new_interface ? hp->timeout :
2584 						  jiffies_to_msecs(fp->timeout)),
2585 					(ms > hp->duration ? ms - hp->duration : 0));
2586 			}
2587 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2588 				   (int) srp->data.cmd_opcode);
2589 		}
2590 		if (list_empty(&fp->rq_list))
2591 			seq_puts(s, "     No requests active\n");
2592 		read_unlock(&fp->rq_list_lock);
2593 	}
2594 }
2595 
2596 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2597 {
2598 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2599 	Sg_device *sdp;
2600 	unsigned long iflags;
2601 
2602 	if (it && (0 == it->index))
2603 		seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2604 			   (int)it->max, sg_big_buff);
2605 
2606 	read_lock_irqsave(&sg_index_lock, iflags);
2607 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2608 	if (NULL == sdp)
2609 		goto skip;
2610 	read_lock(&sdp->sfd_lock);
2611 	if (!list_empty(&sdp->sfds)) {
2612 		seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2613 		if (atomic_read(&sdp->detaching))
2614 			seq_puts(s, "detaching pending close ");
2615 		else if (sdp->device) {
2616 			struct scsi_device *scsidp = sdp->device;
2617 
2618 			seq_printf(s, "%d:%d:%d:%llu   em=%d",
2619 				   scsidp->host->host_no,
2620 				   scsidp->channel, scsidp->id,
2621 				   scsidp->lun,
2622 				   scsidp->host->hostt->emulated);
2623 		}
2624 		seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2625 			   sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2626 		sg_proc_debug_helper(s, sdp);
2627 	}
2628 	read_unlock(&sdp->sfd_lock);
2629 skip:
2630 	read_unlock_irqrestore(&sg_index_lock, iflags);
2631 	return 0;
2632 }
2633 
2634 #endif				/* CONFIG_SCSI_PROC_FS */
2635 
2636 module_init(init_sg);
2637 module_exit(exit_sg);
2638