xref: /openbmc/linux/drivers/scsi/sg.c (revision 81de3bf3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  History:
4  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
5  *           to allow user process control of SCSI devices.
6  *  Development Sponsored by Killy Corp. NY NY
7  *
8  * Original driver (sg.c):
9  *        Copyright (C) 1992 Lawrence Foard
10  * Version 2 and 3 extensions to driver:
11  *        Copyright (C) 1998 - 2014 Douglas Gilbert
12  */
13 
14 static int sg_version_num = 30536;	/* 2 digits for each component */
15 #define SG_VERSION_STR "3.5.36"
16 
17 /*
18  *  D. P. Gilbert (dgilbert@interlog.com), notes:
19  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
20  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
21  *        (otherwise the macros compile to empty statements).
22  *
23  */
24 #include <linux/module.h>
25 
26 #include <linux/fs.h>
27 #include <linux/kernel.h>
28 #include <linux/sched.h>
29 #include <linux/string.h>
30 #include <linux/mm.h>
31 #include <linux/errno.h>
32 #include <linux/mtio.h>
33 #include <linux/ioctl.h>
34 #include <linux/slab.h>
35 #include <linux/fcntl.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/moduleparam.h>
39 #include <linux/cdev.h>
40 #include <linux/idr.h>
41 #include <linux/seq_file.h>
42 #include <linux/blkdev.h>
43 #include <linux/delay.h>
44 #include <linux/blktrace_api.h>
45 #include <linux/mutex.h>
46 #include <linux/atomic.h>
47 #include <linux/ratelimit.h>
48 #include <linux/uio.h>
49 #include <linux/cred.h> /* for sg_check_file_access() */
50 
51 #include "scsi.h"
52 #include <scsi/scsi_dbg.h>
53 #include <scsi/scsi_host.h>
54 #include <scsi/scsi_driver.h>
55 #include <scsi/scsi_ioctl.h>
56 #include <scsi/sg.h>
57 
58 #include "scsi_logging.h"
59 
60 #ifdef CONFIG_SCSI_PROC_FS
61 #include <linux/proc_fs.h>
62 static char *sg_version_date = "20140603";
63 
64 static int sg_proc_init(void);
65 #endif
66 
67 #define SG_ALLOW_DIO_DEF 0
68 
69 #define SG_MAX_DEVS 32768
70 
71 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
72  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
73  * than 16 bytes are "variable length" whose length is a multiple of 4
74  */
75 #define SG_MAX_CDB_SIZE 252
76 
77 #define SG_DEFAULT_TIMEOUT mult_frac(SG_DEFAULT_TIMEOUT_USER, HZ, USER_HZ)
78 
79 int sg_big_buff = SG_DEF_RESERVED_SIZE;
80 /* N.B. This variable is readable and writeable via
81    /proc/scsi/sg/def_reserved_size . Each time sg_open() is called a buffer
82    of this size (or less if there is not enough memory) will be reserved
83    for use by this file descriptor. [Deprecated usage: this variable is also
84    readable via /proc/sys/kernel/sg-big-buff if the sg driver is built into
85    the kernel (i.e. it is not a module).] */
86 static int def_reserved_size = -1;	/* picks up init parameter */
87 static int sg_allow_dio = SG_ALLOW_DIO_DEF;
88 
89 static int scatter_elem_sz = SG_SCATTER_SZ;
90 static int scatter_elem_sz_prev = SG_SCATTER_SZ;
91 
92 #define SG_SECTOR_SZ 512
93 
94 static int sg_add_device(struct device *, struct class_interface *);
95 static void sg_remove_device(struct device *, struct class_interface *);
96 
97 static DEFINE_IDR(sg_index_idr);
98 static DEFINE_RWLOCK(sg_index_lock);	/* Also used to lock
99 							   file descriptor list for device */
100 
101 static struct class_interface sg_interface = {
102 	.add_dev        = sg_add_device,
103 	.remove_dev     = sg_remove_device,
104 };
105 
106 typedef struct sg_scatter_hold { /* holding area for scsi scatter gather info */
107 	unsigned short k_use_sg; /* Count of kernel scatter-gather pieces */
108 	unsigned sglist_len; /* size of malloc'd scatter-gather list ++ */
109 	unsigned bufflen;	/* Size of (aggregate) data buffer */
110 	struct page **pages;
111 	int page_order;
112 	char dio_in_use;	/* 0->indirect IO (or mmap), 1->dio */
113 	unsigned char cmd_opcode; /* first byte of command */
114 } Sg_scatter_hold;
115 
116 struct sg_device;		/* forward declarations */
117 struct sg_fd;
118 
119 typedef struct sg_request {	/* SG_MAX_QUEUE requests outstanding per file */
120 	struct list_head entry;	/* list entry */
121 	struct sg_fd *parentfp;	/* NULL -> not in use */
122 	Sg_scatter_hold data;	/* hold buffer, perhaps scatter list */
123 	sg_io_hdr_t header;	/* scsi command+info, see <scsi/sg.h> */
124 	unsigned char sense_b[SCSI_SENSE_BUFFERSIZE];
125 	char res_used;		/* 1 -> using reserve buffer, 0 -> not ... */
126 	char orphan;		/* 1 -> drop on sight, 0 -> normal */
127 	char sg_io_owned;	/* 1 -> packet belongs to SG_IO */
128 	/* done protected by rq_list_lock */
129 	char done;		/* 0->before bh, 1->before read, 2->read */
130 	struct request *rq;
131 	struct bio *bio;
132 	struct execute_work ew;
133 } Sg_request;
134 
135 typedef struct sg_fd {		/* holds the state of a file descriptor */
136 	struct list_head sfd_siblings;  /* protected by device's sfd_lock */
137 	struct sg_device *parentdp;	/* owning device */
138 	wait_queue_head_t read_wait;	/* queue read until command done */
139 	rwlock_t rq_list_lock;	/* protect access to list in req_arr */
140 	struct mutex f_mutex;	/* protect against changes in this fd */
141 	int timeout;		/* defaults to SG_DEFAULT_TIMEOUT      */
142 	int timeout_user;	/* defaults to SG_DEFAULT_TIMEOUT_USER */
143 	Sg_scatter_hold reserve;	/* buffer held for this file descriptor */
144 	struct list_head rq_list; /* head of request list */
145 	struct fasync_struct *async_qp;	/* used by asynchronous notification */
146 	Sg_request req_arr[SG_MAX_QUEUE];	/* used as singly-linked list */
147 	char force_packid;	/* 1 -> pack_id input to read(), 0 -> ignored */
148 	char cmd_q;		/* 1 -> allow command queuing, 0 -> don't */
149 	unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
150 	char keep_orphan;	/* 0 -> drop orphan (def), 1 -> keep for read() */
151 	char mmap_called;	/* 0 -> mmap() never called on this fd */
152 	char res_in_use;	/* 1 -> 'reserve' array in use */
153 	struct kref f_ref;
154 	struct execute_work ew;
155 } Sg_fd;
156 
157 typedef struct sg_device { /* holds the state of each scsi generic device */
158 	struct scsi_device *device;
159 	wait_queue_head_t open_wait;    /* queue open() when O_EXCL present */
160 	struct mutex open_rel_lock;     /* held when in open() or release() */
161 	int sg_tablesize;	/* adapter's max scatter-gather table size */
162 	u32 index;		/* device index number */
163 	struct list_head sfds;
164 	rwlock_t sfd_lock;      /* protect access to sfd list */
165 	atomic_t detaching;     /* 0->device usable, 1->device detaching */
166 	bool exclude;		/* 1->open(O_EXCL) succeeded and is active */
167 	int open_cnt;		/* count of opens (perhaps < num(sfds) ) */
168 	char sgdebug;		/* 0->off, 1->sense, 9->dump dev, 10-> all devs */
169 	struct gendisk *disk;
170 	struct cdev * cdev;	/* char_dev [sysfs: /sys/cdev/major/sg<n>] */
171 	struct kref d_ref;
172 } Sg_device;
173 
174 /* tasklet or soft irq callback */
175 static void sg_rq_end_io(struct request *rq, blk_status_t status);
176 static int sg_start_req(Sg_request *srp, unsigned char *cmd);
177 static int sg_finish_rem_req(Sg_request * srp);
178 static int sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size);
179 static ssize_t sg_new_read(Sg_fd * sfp, char __user *buf, size_t count,
180 			   Sg_request * srp);
181 static ssize_t sg_new_write(Sg_fd *sfp, struct file *file,
182 			const char __user *buf, size_t count, int blocking,
183 			int read_only, int sg_io_owned, Sg_request **o_srp);
184 static int sg_common_write(Sg_fd * sfp, Sg_request * srp,
185 			   unsigned char *cmnd, int timeout, int blocking);
186 static int sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer);
187 static void sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp);
188 static void sg_build_reserve(Sg_fd * sfp, int req_size);
189 static void sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size);
190 static void sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp);
191 static Sg_fd *sg_add_sfp(Sg_device * sdp);
192 static void sg_remove_sfp(struct kref *);
193 static Sg_request *sg_get_rq_mark(Sg_fd * sfp, int pack_id);
194 static Sg_request *sg_add_request(Sg_fd * sfp);
195 static int sg_remove_request(Sg_fd * sfp, Sg_request * srp);
196 static Sg_device *sg_get_dev(int dev);
197 static void sg_device_destroy(struct kref *kref);
198 
199 #define SZ_SG_HEADER sizeof(struct sg_header)
200 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
201 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
202 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
203 
204 #define sg_printk(prefix, sdp, fmt, a...) \
205 	sdev_prefix_printk(prefix, (sdp)->device,		\
206 			   (sdp)->disk->disk_name, fmt, ##a)
207 
208 /*
209  * The SCSI interfaces that use read() and write() as an asynchronous variant of
210  * ioctl(..., SG_IO, ...) are fundamentally unsafe, since there are lots of ways
211  * to trigger read() and write() calls from various contexts with elevated
212  * privileges. This can lead to kernel memory corruption (e.g. if these
213  * interfaces are called through splice()) and privilege escalation inside
214  * userspace (e.g. if a process with access to such a device passes a file
215  * descriptor to a SUID binary as stdin/stdout/stderr).
216  *
217  * This function provides protection for the legacy API by restricting the
218  * calling context.
219  */
220 static int sg_check_file_access(struct file *filp, const char *caller)
221 {
222 	if (filp->f_cred != current_real_cred()) {
223 		pr_err_once("%s: process %d (%s) changed security contexts after opening file descriptor, this is not allowed.\n",
224 			caller, task_tgid_vnr(current), current->comm);
225 		return -EPERM;
226 	}
227 	if (uaccess_kernel()) {
228 		pr_err_once("%s: process %d (%s) called from kernel context, this is not allowed.\n",
229 			caller, task_tgid_vnr(current), current->comm);
230 		return -EACCES;
231 	}
232 	return 0;
233 }
234 
235 static int sg_allow_access(struct file *filp, unsigned char *cmd)
236 {
237 	struct sg_fd *sfp = filp->private_data;
238 
239 	if (sfp->parentdp->device->type == TYPE_SCANNER)
240 		return 0;
241 
242 	return blk_verify_command(cmd, filp->f_mode);
243 }
244 
245 static int
246 open_wait(Sg_device *sdp, int flags)
247 {
248 	int retval = 0;
249 
250 	if (flags & O_EXCL) {
251 		while (sdp->open_cnt > 0) {
252 			mutex_unlock(&sdp->open_rel_lock);
253 			retval = wait_event_interruptible(sdp->open_wait,
254 					(atomic_read(&sdp->detaching) ||
255 					 !sdp->open_cnt));
256 			mutex_lock(&sdp->open_rel_lock);
257 
258 			if (retval) /* -ERESTARTSYS */
259 				return retval;
260 			if (atomic_read(&sdp->detaching))
261 				return -ENODEV;
262 		}
263 	} else {
264 		while (sdp->exclude) {
265 			mutex_unlock(&sdp->open_rel_lock);
266 			retval = wait_event_interruptible(sdp->open_wait,
267 					(atomic_read(&sdp->detaching) ||
268 					 !sdp->exclude));
269 			mutex_lock(&sdp->open_rel_lock);
270 
271 			if (retval) /* -ERESTARTSYS */
272 				return retval;
273 			if (atomic_read(&sdp->detaching))
274 				return -ENODEV;
275 		}
276 	}
277 
278 	return retval;
279 }
280 
281 /* Returns 0 on success, else a negated errno value */
282 static int
283 sg_open(struct inode *inode, struct file *filp)
284 {
285 	int dev = iminor(inode);
286 	int flags = filp->f_flags;
287 	struct request_queue *q;
288 	Sg_device *sdp;
289 	Sg_fd *sfp;
290 	int retval;
291 
292 	nonseekable_open(inode, filp);
293 	if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
294 		return -EPERM; /* Can't lock it with read only access */
295 	sdp = sg_get_dev(dev);
296 	if (IS_ERR(sdp))
297 		return PTR_ERR(sdp);
298 
299 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
300 				      "sg_open: flags=0x%x\n", flags));
301 
302 	/* This driver's module count bumped by fops_get in <linux/fs.h> */
303 	/* Prevent the device driver from vanishing while we sleep */
304 	retval = scsi_device_get(sdp->device);
305 	if (retval)
306 		goto sg_put;
307 
308 	retval = scsi_autopm_get_device(sdp->device);
309 	if (retval)
310 		goto sdp_put;
311 
312 	/* scsi_block_when_processing_errors() may block so bypass
313 	 * check if O_NONBLOCK. Permits SCSI commands to be issued
314 	 * during error recovery. Tread carefully. */
315 	if (!((flags & O_NONBLOCK) ||
316 	      scsi_block_when_processing_errors(sdp->device))) {
317 		retval = -ENXIO;
318 		/* we are in error recovery for this device */
319 		goto error_out;
320 	}
321 
322 	mutex_lock(&sdp->open_rel_lock);
323 	if (flags & O_NONBLOCK) {
324 		if (flags & O_EXCL) {
325 			if (sdp->open_cnt > 0) {
326 				retval = -EBUSY;
327 				goto error_mutex_locked;
328 			}
329 		} else {
330 			if (sdp->exclude) {
331 				retval = -EBUSY;
332 				goto error_mutex_locked;
333 			}
334 		}
335 	} else {
336 		retval = open_wait(sdp, flags);
337 		if (retval) /* -ERESTARTSYS or -ENODEV */
338 			goto error_mutex_locked;
339 	}
340 
341 	/* N.B. at this point we are holding the open_rel_lock */
342 	if (flags & O_EXCL)
343 		sdp->exclude = true;
344 
345 	if (sdp->open_cnt < 1) {  /* no existing opens */
346 		sdp->sgdebug = 0;
347 		q = sdp->device->request_queue;
348 		sdp->sg_tablesize = queue_max_segments(q);
349 	}
350 	sfp = sg_add_sfp(sdp);
351 	if (IS_ERR(sfp)) {
352 		retval = PTR_ERR(sfp);
353 		goto out_undo;
354 	}
355 
356 	filp->private_data = sfp;
357 	sdp->open_cnt++;
358 	mutex_unlock(&sdp->open_rel_lock);
359 
360 	retval = 0;
361 sg_put:
362 	kref_put(&sdp->d_ref, sg_device_destroy);
363 	return retval;
364 
365 out_undo:
366 	if (flags & O_EXCL) {
367 		sdp->exclude = false;   /* undo if error */
368 		wake_up_interruptible(&sdp->open_wait);
369 	}
370 error_mutex_locked:
371 	mutex_unlock(&sdp->open_rel_lock);
372 error_out:
373 	scsi_autopm_put_device(sdp->device);
374 sdp_put:
375 	scsi_device_put(sdp->device);
376 	goto sg_put;
377 }
378 
379 /* Release resources associated with a successful sg_open()
380  * Returns 0 on success, else a negated errno value */
381 static int
382 sg_release(struct inode *inode, struct file *filp)
383 {
384 	Sg_device *sdp;
385 	Sg_fd *sfp;
386 
387 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
388 		return -ENXIO;
389 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
390 
391 	mutex_lock(&sdp->open_rel_lock);
392 	scsi_autopm_put_device(sdp->device);
393 	kref_put(&sfp->f_ref, sg_remove_sfp);
394 	sdp->open_cnt--;
395 
396 	/* possibly many open()s waiting on exlude clearing, start many;
397 	 * only open(O_EXCL)s wait on 0==open_cnt so only start one */
398 	if (sdp->exclude) {
399 		sdp->exclude = false;
400 		wake_up_interruptible_all(&sdp->open_wait);
401 	} else if (0 == sdp->open_cnt) {
402 		wake_up_interruptible(&sdp->open_wait);
403 	}
404 	mutex_unlock(&sdp->open_rel_lock);
405 	return 0;
406 }
407 
408 static ssize_t
409 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
410 {
411 	Sg_device *sdp;
412 	Sg_fd *sfp;
413 	Sg_request *srp;
414 	int req_pack_id = -1;
415 	sg_io_hdr_t *hp;
416 	struct sg_header *old_hdr = NULL;
417 	int retval = 0;
418 
419 	/*
420 	 * This could cause a response to be stranded. Close the associated
421 	 * file descriptor to free up any resources being held.
422 	 */
423 	retval = sg_check_file_access(filp, __func__);
424 	if (retval)
425 		return retval;
426 
427 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
428 		return -ENXIO;
429 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
430 				      "sg_read: count=%d\n", (int) count));
431 
432 	if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
433 		old_hdr = memdup_user(buf, SZ_SG_HEADER);
434 		if (IS_ERR(old_hdr))
435 			return PTR_ERR(old_hdr);
436 		if (old_hdr->reply_len < 0) {
437 			if (count >= SZ_SG_IO_HDR) {
438 				/*
439 				 * This is stupid.
440 				 *
441 				 * We're copying the whole sg_io_hdr_t from user
442 				 * space just to get the 'pack_id' field. But the
443 				 * field is at different offsets for the compat
444 				 * case, so we'll use "get_sg_io_hdr()" to copy
445 				 * the whole thing and convert it.
446 				 *
447 				 * We could do something like just calculating the
448 				 * offset based of 'in_compat_syscall()', but the
449 				 * 'compat_sg_io_hdr' definition is in the wrong
450 				 * place for that.
451 				 */
452 				sg_io_hdr_t *new_hdr;
453 				new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
454 				if (!new_hdr) {
455 					retval = -ENOMEM;
456 					goto free_old_hdr;
457 				}
458 				retval = get_sg_io_hdr(new_hdr, buf);
459 				req_pack_id = new_hdr->pack_id;
460 				kfree(new_hdr);
461 				if (retval) {
462 					retval = -EFAULT;
463 					goto free_old_hdr;
464 				}
465 			}
466 		} else
467 			req_pack_id = old_hdr->pack_id;
468 	}
469 	srp = sg_get_rq_mark(sfp, req_pack_id);
470 	if (!srp) {		/* now wait on packet to arrive */
471 		if (atomic_read(&sdp->detaching)) {
472 			retval = -ENODEV;
473 			goto free_old_hdr;
474 		}
475 		if (filp->f_flags & O_NONBLOCK) {
476 			retval = -EAGAIN;
477 			goto free_old_hdr;
478 		}
479 		retval = wait_event_interruptible(sfp->read_wait,
480 			(atomic_read(&sdp->detaching) ||
481 			(srp = sg_get_rq_mark(sfp, req_pack_id))));
482 		if (atomic_read(&sdp->detaching)) {
483 			retval = -ENODEV;
484 			goto free_old_hdr;
485 		}
486 		if (retval) {
487 			/* -ERESTARTSYS as signal hit process */
488 			goto free_old_hdr;
489 		}
490 	}
491 	if (srp->header.interface_id != '\0') {
492 		retval = sg_new_read(sfp, buf, count, srp);
493 		goto free_old_hdr;
494 	}
495 
496 	hp = &srp->header;
497 	if (old_hdr == NULL) {
498 		old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
499 		if (! old_hdr) {
500 			retval = -ENOMEM;
501 			goto free_old_hdr;
502 		}
503 	}
504 	memset(old_hdr, 0, SZ_SG_HEADER);
505 	old_hdr->reply_len = (int) hp->timeout;
506 	old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
507 	old_hdr->pack_id = hp->pack_id;
508 	old_hdr->twelve_byte =
509 	    ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
510 	old_hdr->target_status = hp->masked_status;
511 	old_hdr->host_status = hp->host_status;
512 	old_hdr->driver_status = hp->driver_status;
513 	if ((CHECK_CONDITION & hp->masked_status) ||
514 	    (DRIVER_SENSE & hp->driver_status))
515 		memcpy(old_hdr->sense_buffer, srp->sense_b,
516 		       sizeof (old_hdr->sense_buffer));
517 	switch (hp->host_status) {
518 	/* This setup of 'result' is for backward compatibility and is best
519 	   ignored by the user who should use target, host + driver status */
520 	case DID_OK:
521 	case DID_PASSTHROUGH:
522 	case DID_SOFT_ERROR:
523 		old_hdr->result = 0;
524 		break;
525 	case DID_NO_CONNECT:
526 	case DID_BUS_BUSY:
527 	case DID_TIME_OUT:
528 		old_hdr->result = EBUSY;
529 		break;
530 	case DID_BAD_TARGET:
531 	case DID_ABORT:
532 	case DID_PARITY:
533 	case DID_RESET:
534 	case DID_BAD_INTR:
535 		old_hdr->result = EIO;
536 		break;
537 	case DID_ERROR:
538 		old_hdr->result = (srp->sense_b[0] == 0 &&
539 				  hp->masked_status == GOOD) ? 0 : EIO;
540 		break;
541 	default:
542 		old_hdr->result = EIO;
543 		break;
544 	}
545 
546 	/* Now copy the result back to the user buffer.  */
547 	if (count >= SZ_SG_HEADER) {
548 		if (copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
549 			retval = -EFAULT;
550 			goto free_old_hdr;
551 		}
552 		buf += SZ_SG_HEADER;
553 		if (count > old_hdr->reply_len)
554 			count = old_hdr->reply_len;
555 		if (count > SZ_SG_HEADER) {
556 			if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
557 				retval = -EFAULT;
558 				goto free_old_hdr;
559 			}
560 		}
561 	} else
562 		count = (old_hdr->result == 0) ? 0 : -EIO;
563 	sg_finish_rem_req(srp);
564 	sg_remove_request(sfp, srp);
565 	retval = count;
566 free_old_hdr:
567 	kfree(old_hdr);
568 	return retval;
569 }
570 
571 static ssize_t
572 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
573 {
574 	sg_io_hdr_t *hp = &srp->header;
575 	int err = 0, err2;
576 	int len;
577 
578 	if (count < SZ_SG_IO_HDR) {
579 		err = -EINVAL;
580 		goto err_out;
581 	}
582 	hp->sb_len_wr = 0;
583 	if ((hp->mx_sb_len > 0) && hp->sbp) {
584 		if ((CHECK_CONDITION & hp->masked_status) ||
585 		    (DRIVER_SENSE & hp->driver_status)) {
586 			int sb_len = SCSI_SENSE_BUFFERSIZE;
587 			sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
588 			len = 8 + (int) srp->sense_b[7];	/* Additional sense length field */
589 			len = (len > sb_len) ? sb_len : len;
590 			if (copy_to_user(hp->sbp, srp->sense_b, len)) {
591 				err = -EFAULT;
592 				goto err_out;
593 			}
594 			hp->sb_len_wr = len;
595 		}
596 	}
597 	if (hp->masked_status || hp->host_status || hp->driver_status)
598 		hp->info |= SG_INFO_CHECK;
599 	err = put_sg_io_hdr(hp, buf);
600 err_out:
601 	err2 = sg_finish_rem_req(srp);
602 	sg_remove_request(sfp, srp);
603 	return err ? : err2 ? : count;
604 }
605 
606 static ssize_t
607 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
608 {
609 	int mxsize, cmd_size, k;
610 	int input_size, blocking;
611 	unsigned char opcode;
612 	Sg_device *sdp;
613 	Sg_fd *sfp;
614 	Sg_request *srp;
615 	struct sg_header old_hdr;
616 	sg_io_hdr_t *hp;
617 	unsigned char cmnd[SG_MAX_CDB_SIZE];
618 	int retval;
619 
620 	retval = sg_check_file_access(filp, __func__);
621 	if (retval)
622 		return retval;
623 
624 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
625 		return -ENXIO;
626 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
627 				      "sg_write: count=%d\n", (int) count));
628 	if (atomic_read(&sdp->detaching))
629 		return -ENODEV;
630 	if (!((filp->f_flags & O_NONBLOCK) ||
631 	      scsi_block_when_processing_errors(sdp->device)))
632 		return -ENXIO;
633 
634 	if (count < SZ_SG_HEADER)
635 		return -EIO;
636 	if (copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
637 		return -EFAULT;
638 	blocking = !(filp->f_flags & O_NONBLOCK);
639 	if (old_hdr.reply_len < 0)
640 		return sg_new_write(sfp, filp, buf, count,
641 				    blocking, 0, 0, NULL);
642 	if (count < (SZ_SG_HEADER + 6))
643 		return -EIO;	/* The minimum scsi command length is 6 bytes. */
644 
645 	buf += SZ_SG_HEADER;
646 	if (get_user(opcode, buf))
647 		return -EFAULT;
648 
649 	if (!(srp = sg_add_request(sfp))) {
650 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
651 					      "sg_write: queue full\n"));
652 		return -EDOM;
653 	}
654 	mutex_lock(&sfp->f_mutex);
655 	if (sfp->next_cmd_len > 0) {
656 		cmd_size = sfp->next_cmd_len;
657 		sfp->next_cmd_len = 0;	/* reset so only this write() effected */
658 	} else {
659 		cmd_size = COMMAND_SIZE(opcode);	/* based on SCSI command group */
660 		if ((opcode >= 0xc0) && old_hdr.twelve_byte)
661 			cmd_size = 12;
662 	}
663 	mutex_unlock(&sfp->f_mutex);
664 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
665 		"sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
666 /* Determine buffer size.  */
667 	input_size = count - cmd_size;
668 	mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
669 	mxsize -= SZ_SG_HEADER;
670 	input_size -= SZ_SG_HEADER;
671 	if (input_size < 0) {
672 		sg_remove_request(sfp, srp);
673 		return -EIO;	/* User did not pass enough bytes for this command. */
674 	}
675 	hp = &srp->header;
676 	hp->interface_id = '\0';	/* indicator of old interface tunnelled */
677 	hp->cmd_len = (unsigned char) cmd_size;
678 	hp->iovec_count = 0;
679 	hp->mx_sb_len = 0;
680 	if (input_size > 0)
681 		hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
682 		    SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
683 	else
684 		hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
685 	hp->dxfer_len = mxsize;
686 	if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
687 	    (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
688 		hp->dxferp = (char __user *)buf + cmd_size;
689 	else
690 		hp->dxferp = NULL;
691 	hp->sbp = NULL;
692 	hp->timeout = old_hdr.reply_len;	/* structure abuse ... */
693 	hp->flags = input_size;	/* structure abuse ... */
694 	hp->pack_id = old_hdr.pack_id;
695 	hp->usr_ptr = NULL;
696 	if (copy_from_user(cmnd, buf, cmd_size))
697 		return -EFAULT;
698 	/*
699 	 * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
700 	 * but is is possible that the app intended SG_DXFER_TO_DEV, because there
701 	 * is a non-zero input_size, so emit a warning.
702 	 */
703 	if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
704 		printk_ratelimited(KERN_WARNING
705 				   "sg_write: data in/out %d/%d bytes "
706 				   "for SCSI command 0x%x-- guessing "
707 				   "data in;\n   program %s not setting "
708 				   "count and/or reply_len properly\n",
709 				   old_hdr.reply_len - (int)SZ_SG_HEADER,
710 				   input_size, (unsigned int) cmnd[0],
711 				   current->comm);
712 	}
713 	k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
714 	return (k < 0) ? k : count;
715 }
716 
717 static ssize_t
718 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
719 		 size_t count, int blocking, int read_only, int sg_io_owned,
720 		 Sg_request **o_srp)
721 {
722 	int k;
723 	Sg_request *srp;
724 	sg_io_hdr_t *hp;
725 	unsigned char cmnd[SG_MAX_CDB_SIZE];
726 	int timeout;
727 	unsigned long ul_timeout;
728 
729 	if (count < SZ_SG_IO_HDR)
730 		return -EINVAL;
731 
732 	sfp->cmd_q = 1;	/* when sg_io_hdr seen, set command queuing on */
733 	if (!(srp = sg_add_request(sfp))) {
734 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
735 					      "sg_new_write: queue full\n"));
736 		return -EDOM;
737 	}
738 	srp->sg_io_owned = sg_io_owned;
739 	hp = &srp->header;
740 	if (get_sg_io_hdr(hp, buf)) {
741 		sg_remove_request(sfp, srp);
742 		return -EFAULT;
743 	}
744 	if (hp->interface_id != 'S') {
745 		sg_remove_request(sfp, srp);
746 		return -ENOSYS;
747 	}
748 	if (hp->flags & SG_FLAG_MMAP_IO) {
749 		if (hp->dxfer_len > sfp->reserve.bufflen) {
750 			sg_remove_request(sfp, srp);
751 			return -ENOMEM;	/* MMAP_IO size must fit in reserve buffer */
752 		}
753 		if (hp->flags & SG_FLAG_DIRECT_IO) {
754 			sg_remove_request(sfp, srp);
755 			return -EINVAL;	/* either MMAP_IO or DIRECT_IO (not both) */
756 		}
757 		if (sfp->res_in_use) {
758 			sg_remove_request(sfp, srp);
759 			return -EBUSY;	/* reserve buffer already being used */
760 		}
761 	}
762 	ul_timeout = msecs_to_jiffies(srp->header.timeout);
763 	timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
764 	if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
765 		sg_remove_request(sfp, srp);
766 		return -EMSGSIZE;
767 	}
768 	if (copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
769 		sg_remove_request(sfp, srp);
770 		return -EFAULT;
771 	}
772 	if (read_only && sg_allow_access(file, cmnd)) {
773 		sg_remove_request(sfp, srp);
774 		return -EPERM;
775 	}
776 	k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
777 	if (k < 0)
778 		return k;
779 	if (o_srp)
780 		*o_srp = srp;
781 	return count;
782 }
783 
784 static int
785 sg_common_write(Sg_fd * sfp, Sg_request * srp,
786 		unsigned char *cmnd, int timeout, int blocking)
787 {
788 	int k, at_head;
789 	Sg_device *sdp = sfp->parentdp;
790 	sg_io_hdr_t *hp = &srp->header;
791 
792 	srp->data.cmd_opcode = cmnd[0];	/* hold opcode of command */
793 	hp->status = 0;
794 	hp->masked_status = 0;
795 	hp->msg_status = 0;
796 	hp->info = 0;
797 	hp->host_status = 0;
798 	hp->driver_status = 0;
799 	hp->resid = 0;
800 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
801 			"sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
802 			(int) cmnd[0], (int) hp->cmd_len));
803 
804 	if (hp->dxfer_len >= SZ_256M)
805 		return -EINVAL;
806 
807 	k = sg_start_req(srp, cmnd);
808 	if (k) {
809 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
810 			"sg_common_write: start_req err=%d\n", k));
811 		sg_finish_rem_req(srp);
812 		sg_remove_request(sfp, srp);
813 		return k;	/* probably out of space --> ENOMEM */
814 	}
815 	if (atomic_read(&sdp->detaching)) {
816 		if (srp->bio) {
817 			scsi_req_free_cmd(scsi_req(srp->rq));
818 			blk_put_request(srp->rq);
819 			srp->rq = NULL;
820 		}
821 
822 		sg_finish_rem_req(srp);
823 		sg_remove_request(sfp, srp);
824 		return -ENODEV;
825 	}
826 
827 	hp->duration = jiffies_to_msecs(jiffies);
828 	if (hp->interface_id != '\0' &&	/* v3 (or later) interface */
829 	    (SG_FLAG_Q_AT_TAIL & hp->flags))
830 		at_head = 0;
831 	else
832 		at_head = 1;
833 
834 	srp->rq->timeout = timeout;
835 	kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
836 	blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
837 			      srp->rq, at_head, sg_rq_end_io);
838 	return 0;
839 }
840 
841 static int srp_done(Sg_fd *sfp, Sg_request *srp)
842 {
843 	unsigned long flags;
844 	int ret;
845 
846 	read_lock_irqsave(&sfp->rq_list_lock, flags);
847 	ret = srp->done;
848 	read_unlock_irqrestore(&sfp->rq_list_lock, flags);
849 	return ret;
850 }
851 
852 static int max_sectors_bytes(struct request_queue *q)
853 {
854 	unsigned int max_sectors = queue_max_sectors(q);
855 
856 	max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
857 
858 	return max_sectors << 9;
859 }
860 
861 static void
862 sg_fill_request_table(Sg_fd *sfp, sg_req_info_t *rinfo)
863 {
864 	Sg_request *srp;
865 	int val;
866 	unsigned int ms;
867 
868 	val = 0;
869 	list_for_each_entry(srp, &sfp->rq_list, entry) {
870 		if (val >= SG_MAX_QUEUE)
871 			break;
872 		rinfo[val].req_state = srp->done + 1;
873 		rinfo[val].problem =
874 			srp->header.masked_status &
875 			srp->header.host_status &
876 			srp->header.driver_status;
877 		if (srp->done)
878 			rinfo[val].duration =
879 				srp->header.duration;
880 		else {
881 			ms = jiffies_to_msecs(jiffies);
882 			rinfo[val].duration =
883 				(ms > srp->header.duration) ?
884 				(ms - srp->header.duration) : 0;
885 		}
886 		rinfo[val].orphan = srp->orphan;
887 		rinfo[val].sg_io_owned = srp->sg_io_owned;
888 		rinfo[val].pack_id = srp->header.pack_id;
889 		rinfo[val].usr_ptr = srp->header.usr_ptr;
890 		val++;
891 	}
892 }
893 
894 #ifdef CONFIG_COMPAT
895 struct compat_sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() */
896 	char req_state;
897 	char orphan;
898 	char sg_io_owned;
899 	char problem;
900 	int pack_id;
901 	compat_uptr_t usr_ptr;
902 	unsigned int duration;
903 	int unused;
904 };
905 
906 static int put_compat_request_table(struct compat_sg_req_info __user *o,
907 				    struct sg_req_info *rinfo)
908 {
909 	int i;
910 	for (i = 0; i < SG_MAX_QUEUE; i++) {
911 		if (copy_to_user(o + i, rinfo + i, offsetof(sg_req_info_t, usr_ptr)) ||
912 		    put_user((uintptr_t)rinfo[i].usr_ptr, &o[i].usr_ptr) ||
913 		    put_user(rinfo[i].duration, &o[i].duration) ||
914 		    put_user(rinfo[i].unused, &o[i].unused))
915 			return -EFAULT;
916 	}
917 	return 0;
918 }
919 #endif
920 
921 static long
922 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
923 {
924 	void __user *p = (void __user *)arg;
925 	int __user *ip = p;
926 	int result, val, read_only;
927 	Sg_device *sdp;
928 	Sg_fd *sfp;
929 	Sg_request *srp;
930 	unsigned long iflags;
931 
932 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
933 		return -ENXIO;
934 
935 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
936 				   "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
937 	read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
938 
939 	switch (cmd_in) {
940 	case SG_IO:
941 		if (atomic_read(&sdp->detaching))
942 			return -ENODEV;
943 		if (!scsi_block_when_processing_errors(sdp->device))
944 			return -ENXIO;
945 		result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
946 				 1, read_only, 1, &srp);
947 		if (result < 0)
948 			return result;
949 		result = wait_event_interruptible(sfp->read_wait,
950 			(srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
951 		if (atomic_read(&sdp->detaching))
952 			return -ENODEV;
953 		write_lock_irq(&sfp->rq_list_lock);
954 		if (srp->done) {
955 			srp->done = 2;
956 			write_unlock_irq(&sfp->rq_list_lock);
957 			result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
958 			return (result < 0) ? result : 0;
959 		}
960 		srp->orphan = 1;
961 		write_unlock_irq(&sfp->rq_list_lock);
962 		return result;	/* -ERESTARTSYS because signal hit process */
963 	case SG_SET_TIMEOUT:
964 		result = get_user(val, ip);
965 		if (result)
966 			return result;
967 		if (val < 0)
968 			return -EIO;
969 		if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
970 			val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
971 				    INT_MAX);
972 		sfp->timeout_user = val;
973 		sfp->timeout = mult_frac(val, HZ, USER_HZ);
974 
975 		return 0;
976 	case SG_GET_TIMEOUT:	/* N.B. User receives timeout as return value */
977 				/* strange ..., for backward compatibility */
978 		return sfp->timeout_user;
979 	case SG_SET_FORCE_LOW_DMA:
980 		/*
981 		 * N.B. This ioctl never worked properly, but failed to
982 		 * return an error value. So returning '0' to keep compability
983 		 * with legacy applications.
984 		 */
985 		return 0;
986 	case SG_GET_LOW_DMA:
987 		return put_user((int) sdp->device->host->unchecked_isa_dma, ip);
988 	case SG_GET_SCSI_ID:
989 		{
990 			sg_scsi_id_t v;
991 
992 			if (atomic_read(&sdp->detaching))
993 				return -ENODEV;
994 			memset(&v, 0, sizeof(v));
995 			v.host_no = sdp->device->host->host_no;
996 			v.channel = sdp->device->channel;
997 			v.scsi_id = sdp->device->id;
998 			v.lun = sdp->device->lun;
999 			v.scsi_type = sdp->device->type;
1000 			v.h_cmd_per_lun = sdp->device->host->cmd_per_lun;
1001 			v.d_queue_depth = sdp->device->queue_depth;
1002 			if (copy_to_user(p, &v, sizeof(sg_scsi_id_t)))
1003 				return -EFAULT;
1004 			return 0;
1005 		}
1006 	case SG_SET_FORCE_PACK_ID:
1007 		result = get_user(val, ip);
1008 		if (result)
1009 			return result;
1010 		sfp->force_packid = val ? 1 : 0;
1011 		return 0;
1012 	case SG_GET_PACK_ID:
1013 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
1014 		list_for_each_entry(srp, &sfp->rq_list, entry) {
1015 			if ((1 == srp->done) && (!srp->sg_io_owned)) {
1016 				read_unlock_irqrestore(&sfp->rq_list_lock,
1017 						       iflags);
1018 				return put_user(srp->header.pack_id, ip);
1019 			}
1020 		}
1021 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1022 		return put_user(-1, ip);
1023 	case SG_GET_NUM_WAITING:
1024 		read_lock_irqsave(&sfp->rq_list_lock, iflags);
1025 		val = 0;
1026 		list_for_each_entry(srp, &sfp->rq_list, entry) {
1027 			if ((1 == srp->done) && (!srp->sg_io_owned))
1028 				++val;
1029 		}
1030 		read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1031 		return put_user(val, ip);
1032 	case SG_GET_SG_TABLESIZE:
1033 		return put_user(sdp->sg_tablesize, ip);
1034 	case SG_SET_RESERVED_SIZE:
1035 		result = get_user(val, ip);
1036 		if (result)
1037 			return result;
1038                 if (val < 0)
1039                         return -EINVAL;
1040 		val = min_t(int, val,
1041 			    max_sectors_bytes(sdp->device->request_queue));
1042 		mutex_lock(&sfp->f_mutex);
1043 		if (val != sfp->reserve.bufflen) {
1044 			if (sfp->mmap_called ||
1045 			    sfp->res_in_use) {
1046 				mutex_unlock(&sfp->f_mutex);
1047 				return -EBUSY;
1048 			}
1049 
1050 			sg_remove_scat(sfp, &sfp->reserve);
1051 			sg_build_reserve(sfp, val);
1052 		}
1053 		mutex_unlock(&sfp->f_mutex);
1054 		return 0;
1055 	case SG_GET_RESERVED_SIZE:
1056 		val = min_t(int, sfp->reserve.bufflen,
1057 			    max_sectors_bytes(sdp->device->request_queue));
1058 		return put_user(val, ip);
1059 	case SG_SET_COMMAND_Q:
1060 		result = get_user(val, ip);
1061 		if (result)
1062 			return result;
1063 		sfp->cmd_q = val ? 1 : 0;
1064 		return 0;
1065 	case SG_GET_COMMAND_Q:
1066 		return put_user((int) sfp->cmd_q, ip);
1067 	case SG_SET_KEEP_ORPHAN:
1068 		result = get_user(val, ip);
1069 		if (result)
1070 			return result;
1071 		sfp->keep_orphan = val;
1072 		return 0;
1073 	case SG_GET_KEEP_ORPHAN:
1074 		return put_user((int) sfp->keep_orphan, ip);
1075 	case SG_NEXT_CMD_LEN:
1076 		result = get_user(val, ip);
1077 		if (result)
1078 			return result;
1079 		if (val > SG_MAX_CDB_SIZE)
1080 			return -ENOMEM;
1081 		sfp->next_cmd_len = (val > 0) ? val : 0;
1082 		return 0;
1083 	case SG_GET_VERSION_NUM:
1084 		return put_user(sg_version_num, ip);
1085 	case SG_GET_ACCESS_COUNT:
1086 		/* faked - we don't have a real access count anymore */
1087 		val = (sdp->device ? 1 : 0);
1088 		return put_user(val, ip);
1089 	case SG_GET_REQUEST_TABLE:
1090 		{
1091 			sg_req_info_t *rinfo;
1092 
1093 			rinfo = kcalloc(SG_MAX_QUEUE, SZ_SG_REQ_INFO,
1094 					GFP_KERNEL);
1095 			if (!rinfo)
1096 				return -ENOMEM;
1097 			read_lock_irqsave(&sfp->rq_list_lock, iflags);
1098 			sg_fill_request_table(sfp, rinfo);
1099 			read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1100 	#ifdef CONFIG_COMPAT
1101 			if (in_compat_syscall())
1102 				result = put_compat_request_table(p, rinfo);
1103 			else
1104 	#endif
1105 				result = copy_to_user(p, rinfo,
1106 						      SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1107 			result = result ? -EFAULT : 0;
1108 			kfree(rinfo);
1109 			return result;
1110 		}
1111 	case SG_EMULATED_HOST:
1112 		if (atomic_read(&sdp->detaching))
1113 			return -ENODEV;
1114 		return put_user(sdp->device->host->hostt->emulated, ip);
1115 	case SCSI_IOCTL_SEND_COMMAND:
1116 		if (atomic_read(&sdp->detaching))
1117 			return -ENODEV;
1118 		return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1119 	case SG_SET_DEBUG:
1120 		result = get_user(val, ip);
1121 		if (result)
1122 			return result;
1123 		sdp->sgdebug = (char) val;
1124 		return 0;
1125 	case BLKSECTGET:
1126 		return put_user(max_sectors_bytes(sdp->device->request_queue),
1127 				ip);
1128 	case BLKTRACESETUP:
1129 		return blk_trace_setup(sdp->device->request_queue,
1130 				       sdp->disk->disk_name,
1131 				       MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1132 				       NULL, p);
1133 	case BLKTRACESTART:
1134 		return blk_trace_startstop(sdp->device->request_queue, 1);
1135 	case BLKTRACESTOP:
1136 		return blk_trace_startstop(sdp->device->request_queue, 0);
1137 	case BLKTRACETEARDOWN:
1138 		return blk_trace_remove(sdp->device->request_queue);
1139 	case SCSI_IOCTL_GET_IDLUN:
1140 	case SCSI_IOCTL_GET_BUS_NUMBER:
1141 	case SCSI_IOCTL_PROBE_HOST:
1142 	case SG_GET_TRANSFORM:
1143 	case SG_SCSI_RESET:
1144 		if (atomic_read(&sdp->detaching))
1145 			return -ENODEV;
1146 		break;
1147 	default:
1148 		if (read_only)
1149 			return -EPERM;	/* don't know so take safe approach */
1150 		break;
1151 	}
1152 
1153 	result = scsi_ioctl_block_when_processing_errors(sdp->device,
1154 			cmd_in, filp->f_flags & O_NDELAY);
1155 	if (result)
1156 		return result;
1157 	return scsi_ioctl(sdp->device, cmd_in, p);
1158 }
1159 
1160 #ifdef CONFIG_COMPAT
1161 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1162 {
1163 	Sg_device *sdp;
1164 	Sg_fd *sfp;
1165 	struct scsi_device *sdev;
1166 
1167 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1168 		return -ENXIO;
1169 
1170 	sdev = sdp->device;
1171 	if (sdev->host->hostt->compat_ioctl) {
1172 		int ret;
1173 
1174 		ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1175 
1176 		return ret;
1177 	}
1178 
1179 	return -ENOIOCTLCMD;
1180 }
1181 #endif
1182 
1183 static __poll_t
1184 sg_poll(struct file *filp, poll_table * wait)
1185 {
1186 	__poll_t res = 0;
1187 	Sg_device *sdp;
1188 	Sg_fd *sfp;
1189 	Sg_request *srp;
1190 	int count = 0;
1191 	unsigned long iflags;
1192 
1193 	sfp = filp->private_data;
1194 	if (!sfp)
1195 		return EPOLLERR;
1196 	sdp = sfp->parentdp;
1197 	if (!sdp)
1198 		return EPOLLERR;
1199 	poll_wait(filp, &sfp->read_wait, wait);
1200 	read_lock_irqsave(&sfp->rq_list_lock, iflags);
1201 	list_for_each_entry(srp, &sfp->rq_list, entry) {
1202 		/* if any read waiting, flag it */
1203 		if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1204 			res = EPOLLIN | EPOLLRDNORM;
1205 		++count;
1206 	}
1207 	read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1208 
1209 	if (atomic_read(&sdp->detaching))
1210 		res |= EPOLLHUP;
1211 	else if (!sfp->cmd_q) {
1212 		if (0 == count)
1213 			res |= EPOLLOUT | EPOLLWRNORM;
1214 	} else if (count < SG_MAX_QUEUE)
1215 		res |= EPOLLOUT | EPOLLWRNORM;
1216 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1217 				      "sg_poll: res=0x%x\n", (__force u32) res));
1218 	return res;
1219 }
1220 
1221 static int
1222 sg_fasync(int fd, struct file *filp, int mode)
1223 {
1224 	Sg_device *sdp;
1225 	Sg_fd *sfp;
1226 
1227 	if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1228 		return -ENXIO;
1229 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1230 				      "sg_fasync: mode=%d\n", mode));
1231 
1232 	return fasync_helper(fd, filp, mode, &sfp->async_qp);
1233 }
1234 
1235 static vm_fault_t
1236 sg_vma_fault(struct vm_fault *vmf)
1237 {
1238 	struct vm_area_struct *vma = vmf->vma;
1239 	Sg_fd *sfp;
1240 	unsigned long offset, len, sa;
1241 	Sg_scatter_hold *rsv_schp;
1242 	int k, length;
1243 
1244 	if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1245 		return VM_FAULT_SIGBUS;
1246 	rsv_schp = &sfp->reserve;
1247 	offset = vmf->pgoff << PAGE_SHIFT;
1248 	if (offset >= rsv_schp->bufflen)
1249 		return VM_FAULT_SIGBUS;
1250 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1251 				      "sg_vma_fault: offset=%lu, scatg=%d\n",
1252 				      offset, rsv_schp->k_use_sg));
1253 	sa = vma->vm_start;
1254 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1255 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1256 		len = vma->vm_end - sa;
1257 		len = (len < length) ? len : length;
1258 		if (offset < len) {
1259 			struct page *page = nth_page(rsv_schp->pages[k],
1260 						     offset >> PAGE_SHIFT);
1261 			get_page(page);	/* increment page count */
1262 			vmf->page = page;
1263 			return 0; /* success */
1264 		}
1265 		sa += len;
1266 		offset -= len;
1267 	}
1268 
1269 	return VM_FAULT_SIGBUS;
1270 }
1271 
1272 static const struct vm_operations_struct sg_mmap_vm_ops = {
1273 	.fault = sg_vma_fault,
1274 };
1275 
1276 static int
1277 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1278 {
1279 	Sg_fd *sfp;
1280 	unsigned long req_sz, len, sa;
1281 	Sg_scatter_hold *rsv_schp;
1282 	int k, length;
1283 	int ret = 0;
1284 
1285 	if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1286 		return -ENXIO;
1287 	req_sz = vma->vm_end - vma->vm_start;
1288 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1289 				      "sg_mmap starting, vm_start=%p, len=%d\n",
1290 				      (void *) vma->vm_start, (int) req_sz));
1291 	if (vma->vm_pgoff)
1292 		return -EINVAL;	/* want no offset */
1293 	rsv_schp = &sfp->reserve;
1294 	mutex_lock(&sfp->f_mutex);
1295 	if (req_sz > rsv_schp->bufflen) {
1296 		ret = -ENOMEM;	/* cannot map more than reserved buffer */
1297 		goto out;
1298 	}
1299 
1300 	sa = vma->vm_start;
1301 	length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1302 	for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1303 		len = vma->vm_end - sa;
1304 		len = (len < length) ? len : length;
1305 		sa += len;
1306 	}
1307 
1308 	sfp->mmap_called = 1;
1309 	vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1310 	vma->vm_private_data = sfp;
1311 	vma->vm_ops = &sg_mmap_vm_ops;
1312 out:
1313 	mutex_unlock(&sfp->f_mutex);
1314 	return ret;
1315 }
1316 
1317 static void
1318 sg_rq_end_io_usercontext(struct work_struct *work)
1319 {
1320 	struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1321 	struct sg_fd *sfp = srp->parentfp;
1322 
1323 	sg_finish_rem_req(srp);
1324 	sg_remove_request(sfp, srp);
1325 	kref_put(&sfp->f_ref, sg_remove_sfp);
1326 }
1327 
1328 /*
1329  * This function is a "bottom half" handler that is called by the mid
1330  * level when a command is completed (or has failed).
1331  */
1332 static void
1333 sg_rq_end_io(struct request *rq, blk_status_t status)
1334 {
1335 	struct sg_request *srp = rq->end_io_data;
1336 	struct scsi_request *req = scsi_req(rq);
1337 	Sg_device *sdp;
1338 	Sg_fd *sfp;
1339 	unsigned long iflags;
1340 	unsigned int ms;
1341 	char *sense;
1342 	int result, resid, done = 1;
1343 
1344 	if (WARN_ON(srp->done != 0))
1345 		return;
1346 
1347 	sfp = srp->parentfp;
1348 	if (WARN_ON(sfp == NULL))
1349 		return;
1350 
1351 	sdp = sfp->parentdp;
1352 	if (unlikely(atomic_read(&sdp->detaching)))
1353 		pr_info("%s: device detaching\n", __func__);
1354 
1355 	sense = req->sense;
1356 	result = req->result;
1357 	resid = req->resid_len;
1358 
1359 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1360 				      "sg_cmd_done: pack_id=%d, res=0x%x\n",
1361 				      srp->header.pack_id, result));
1362 	srp->header.resid = resid;
1363 	ms = jiffies_to_msecs(jiffies);
1364 	srp->header.duration = (ms > srp->header.duration) ?
1365 				(ms - srp->header.duration) : 0;
1366 	if (0 != result) {
1367 		struct scsi_sense_hdr sshdr;
1368 
1369 		srp->header.status = 0xff & result;
1370 		srp->header.masked_status = status_byte(result);
1371 		srp->header.msg_status = msg_byte(result);
1372 		srp->header.host_status = host_byte(result);
1373 		srp->header.driver_status = driver_byte(result);
1374 		if ((sdp->sgdebug > 0) &&
1375 		    ((CHECK_CONDITION == srp->header.masked_status) ||
1376 		     (COMMAND_TERMINATED == srp->header.masked_status)))
1377 			__scsi_print_sense(sdp->device, __func__, sense,
1378 					   SCSI_SENSE_BUFFERSIZE);
1379 
1380 		/* Following if statement is a patch supplied by Eric Youngdale */
1381 		if (driver_byte(result) != 0
1382 		    && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1383 		    && !scsi_sense_is_deferred(&sshdr)
1384 		    && sshdr.sense_key == UNIT_ATTENTION
1385 		    && sdp->device->removable) {
1386 			/* Detected possible disc change. Set the bit - this */
1387 			/* may be used if there are filesystems using this device */
1388 			sdp->device->changed = 1;
1389 		}
1390 	}
1391 
1392 	if (req->sense_len)
1393 		memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1394 
1395 	/* Rely on write phase to clean out srp status values, so no "else" */
1396 
1397 	/*
1398 	 * Free the request as soon as it is complete so that its resources
1399 	 * can be reused without waiting for userspace to read() the
1400 	 * result.  But keep the associated bio (if any) around until
1401 	 * blk_rq_unmap_user() can be called from user context.
1402 	 */
1403 	srp->rq = NULL;
1404 	scsi_req_free_cmd(scsi_req(rq));
1405 	blk_put_request(rq);
1406 
1407 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
1408 	if (unlikely(srp->orphan)) {
1409 		if (sfp->keep_orphan)
1410 			srp->sg_io_owned = 0;
1411 		else
1412 			done = 0;
1413 	}
1414 	srp->done = done;
1415 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1416 
1417 	if (likely(done)) {
1418 		/* Now wake up any sg_read() that is waiting for this
1419 		 * packet.
1420 		 */
1421 		wake_up_interruptible(&sfp->read_wait);
1422 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1423 		kref_put(&sfp->f_ref, sg_remove_sfp);
1424 	} else {
1425 		INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1426 		schedule_work(&srp->ew.work);
1427 	}
1428 }
1429 
1430 static const struct file_operations sg_fops = {
1431 	.owner = THIS_MODULE,
1432 	.read = sg_read,
1433 	.write = sg_write,
1434 	.poll = sg_poll,
1435 	.unlocked_ioctl = sg_ioctl,
1436 #ifdef CONFIG_COMPAT
1437 	.compat_ioctl = sg_compat_ioctl,
1438 #endif
1439 	.open = sg_open,
1440 	.mmap = sg_mmap,
1441 	.release = sg_release,
1442 	.fasync = sg_fasync,
1443 	.llseek = no_llseek,
1444 };
1445 
1446 static struct class *sg_sysfs_class;
1447 
1448 static int sg_sysfs_valid = 0;
1449 
1450 static Sg_device *
1451 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1452 {
1453 	struct request_queue *q = scsidp->request_queue;
1454 	Sg_device *sdp;
1455 	unsigned long iflags;
1456 	int error;
1457 	u32 k;
1458 
1459 	sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1460 	if (!sdp) {
1461 		sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1462 			    "failure\n", __func__);
1463 		return ERR_PTR(-ENOMEM);
1464 	}
1465 
1466 	idr_preload(GFP_KERNEL);
1467 	write_lock_irqsave(&sg_index_lock, iflags);
1468 
1469 	error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1470 	if (error < 0) {
1471 		if (error == -ENOSPC) {
1472 			sdev_printk(KERN_WARNING, scsidp,
1473 				    "Unable to attach sg device type=%d, minor number exceeds %d\n",
1474 				    scsidp->type, SG_MAX_DEVS - 1);
1475 			error = -ENODEV;
1476 		} else {
1477 			sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1478 				    "allocation Sg_device failure: %d\n",
1479 				    __func__, error);
1480 		}
1481 		goto out_unlock;
1482 	}
1483 	k = error;
1484 
1485 	SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1486 					"sg_alloc: dev=%d \n", k));
1487 	sprintf(disk->disk_name, "sg%d", k);
1488 	disk->first_minor = k;
1489 	sdp->disk = disk;
1490 	sdp->device = scsidp;
1491 	mutex_init(&sdp->open_rel_lock);
1492 	INIT_LIST_HEAD(&sdp->sfds);
1493 	init_waitqueue_head(&sdp->open_wait);
1494 	atomic_set(&sdp->detaching, 0);
1495 	rwlock_init(&sdp->sfd_lock);
1496 	sdp->sg_tablesize = queue_max_segments(q);
1497 	sdp->index = k;
1498 	kref_init(&sdp->d_ref);
1499 	error = 0;
1500 
1501 out_unlock:
1502 	write_unlock_irqrestore(&sg_index_lock, iflags);
1503 	idr_preload_end();
1504 
1505 	if (error) {
1506 		kfree(sdp);
1507 		return ERR_PTR(error);
1508 	}
1509 	return sdp;
1510 }
1511 
1512 static int
1513 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1514 {
1515 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1516 	struct gendisk *disk;
1517 	Sg_device *sdp = NULL;
1518 	struct cdev * cdev = NULL;
1519 	int error;
1520 	unsigned long iflags;
1521 
1522 	disk = alloc_disk(1);
1523 	if (!disk) {
1524 		pr_warn("%s: alloc_disk failed\n", __func__);
1525 		return -ENOMEM;
1526 	}
1527 	disk->major = SCSI_GENERIC_MAJOR;
1528 
1529 	error = -ENOMEM;
1530 	cdev = cdev_alloc();
1531 	if (!cdev) {
1532 		pr_warn("%s: cdev_alloc failed\n", __func__);
1533 		goto out;
1534 	}
1535 	cdev->owner = THIS_MODULE;
1536 	cdev->ops = &sg_fops;
1537 
1538 	sdp = sg_alloc(disk, scsidp);
1539 	if (IS_ERR(sdp)) {
1540 		pr_warn("%s: sg_alloc failed\n", __func__);
1541 		error = PTR_ERR(sdp);
1542 		goto out;
1543 	}
1544 
1545 	error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1546 	if (error)
1547 		goto cdev_add_err;
1548 
1549 	sdp->cdev = cdev;
1550 	if (sg_sysfs_valid) {
1551 		struct device *sg_class_member;
1552 
1553 		sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1554 						MKDEV(SCSI_GENERIC_MAJOR,
1555 						      sdp->index),
1556 						sdp, "%s", disk->disk_name);
1557 		if (IS_ERR(sg_class_member)) {
1558 			pr_err("%s: device_create failed\n", __func__);
1559 			error = PTR_ERR(sg_class_member);
1560 			goto cdev_add_err;
1561 		}
1562 		error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1563 					  &sg_class_member->kobj, "generic");
1564 		if (error)
1565 			pr_err("%s: unable to make symlink 'generic' back "
1566 			       "to sg%d\n", __func__, sdp->index);
1567 	} else
1568 		pr_warn("%s: sg_sys Invalid\n", __func__);
1569 
1570 	sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1571 		    "type %d\n", sdp->index, scsidp->type);
1572 
1573 	dev_set_drvdata(cl_dev, sdp);
1574 
1575 	return 0;
1576 
1577 cdev_add_err:
1578 	write_lock_irqsave(&sg_index_lock, iflags);
1579 	idr_remove(&sg_index_idr, sdp->index);
1580 	write_unlock_irqrestore(&sg_index_lock, iflags);
1581 	kfree(sdp);
1582 
1583 out:
1584 	put_disk(disk);
1585 	if (cdev)
1586 		cdev_del(cdev);
1587 	return error;
1588 }
1589 
1590 static void
1591 sg_device_destroy(struct kref *kref)
1592 {
1593 	struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1594 	unsigned long flags;
1595 
1596 	/* CAUTION!  Note that the device can still be found via idr_find()
1597 	 * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1598 	 * any other cleanup.
1599 	 */
1600 
1601 	write_lock_irqsave(&sg_index_lock, flags);
1602 	idr_remove(&sg_index_idr, sdp->index);
1603 	write_unlock_irqrestore(&sg_index_lock, flags);
1604 
1605 	SCSI_LOG_TIMEOUT(3,
1606 		sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1607 
1608 	put_disk(sdp->disk);
1609 	kfree(sdp);
1610 }
1611 
1612 static void
1613 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1614 {
1615 	struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1616 	Sg_device *sdp = dev_get_drvdata(cl_dev);
1617 	unsigned long iflags;
1618 	Sg_fd *sfp;
1619 	int val;
1620 
1621 	if (!sdp)
1622 		return;
1623 	/* want sdp->detaching non-zero as soon as possible */
1624 	val = atomic_inc_return(&sdp->detaching);
1625 	if (val > 1)
1626 		return; /* only want to do following once per device */
1627 
1628 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1629 				      "%s\n", __func__));
1630 
1631 	read_lock_irqsave(&sdp->sfd_lock, iflags);
1632 	list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1633 		wake_up_interruptible_all(&sfp->read_wait);
1634 		kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1635 	}
1636 	wake_up_interruptible_all(&sdp->open_wait);
1637 	read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1638 
1639 	sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1640 	device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1641 	cdev_del(sdp->cdev);
1642 	sdp->cdev = NULL;
1643 
1644 	kref_put(&sdp->d_ref, sg_device_destroy);
1645 }
1646 
1647 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1648 module_param_named(def_reserved_size, def_reserved_size, int,
1649 		   S_IRUGO | S_IWUSR);
1650 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1651 
1652 MODULE_AUTHOR("Douglas Gilbert");
1653 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1654 MODULE_LICENSE("GPL");
1655 MODULE_VERSION(SG_VERSION_STR);
1656 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1657 
1658 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1659                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1660 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1661 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1662 
1663 static int __init
1664 init_sg(void)
1665 {
1666 	int rc;
1667 
1668 	if (scatter_elem_sz < PAGE_SIZE) {
1669 		scatter_elem_sz = PAGE_SIZE;
1670 		scatter_elem_sz_prev = scatter_elem_sz;
1671 	}
1672 	if (def_reserved_size >= 0)
1673 		sg_big_buff = def_reserved_size;
1674 	else
1675 		def_reserved_size = sg_big_buff;
1676 
1677 	rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1678 				    SG_MAX_DEVS, "sg");
1679 	if (rc)
1680 		return rc;
1681         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1682         if ( IS_ERR(sg_sysfs_class) ) {
1683 		rc = PTR_ERR(sg_sysfs_class);
1684 		goto err_out;
1685         }
1686 	sg_sysfs_valid = 1;
1687 	rc = scsi_register_interface(&sg_interface);
1688 	if (0 == rc) {
1689 #ifdef CONFIG_SCSI_PROC_FS
1690 		sg_proc_init();
1691 #endif				/* CONFIG_SCSI_PROC_FS */
1692 		return 0;
1693 	}
1694 	class_destroy(sg_sysfs_class);
1695 err_out:
1696 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1697 	return rc;
1698 }
1699 
1700 static void __exit
1701 exit_sg(void)
1702 {
1703 #ifdef CONFIG_SCSI_PROC_FS
1704 	remove_proc_subtree("scsi/sg", NULL);
1705 #endif				/* CONFIG_SCSI_PROC_FS */
1706 	scsi_unregister_interface(&sg_interface);
1707 	class_destroy(sg_sysfs_class);
1708 	sg_sysfs_valid = 0;
1709 	unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1710 				 SG_MAX_DEVS);
1711 	idr_destroy(&sg_index_idr);
1712 }
1713 
1714 static int
1715 sg_start_req(Sg_request *srp, unsigned char *cmd)
1716 {
1717 	int res;
1718 	struct request *rq;
1719 	struct scsi_request *req;
1720 	Sg_fd *sfp = srp->parentfp;
1721 	sg_io_hdr_t *hp = &srp->header;
1722 	int dxfer_len = (int) hp->dxfer_len;
1723 	int dxfer_dir = hp->dxfer_direction;
1724 	unsigned int iov_count = hp->iovec_count;
1725 	Sg_scatter_hold *req_schp = &srp->data;
1726 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
1727 	struct request_queue *q = sfp->parentdp->device->request_queue;
1728 	struct rq_map_data *md, map_data;
1729 	int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1730 	unsigned char *long_cmdp = NULL;
1731 
1732 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1733 				      "sg_start_req: dxfer_len=%d\n",
1734 				      dxfer_len));
1735 
1736 	if (hp->cmd_len > BLK_MAX_CDB) {
1737 		long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1738 		if (!long_cmdp)
1739 			return -ENOMEM;
1740 	}
1741 
1742 	/*
1743 	 * NOTE
1744 	 *
1745 	 * With scsi-mq enabled, there are a fixed number of preallocated
1746 	 * requests equal in number to shost->can_queue.  If all of the
1747 	 * preallocated requests are already in use, then blk_get_request()
1748 	 * will sleep until an active command completes, freeing up a request.
1749 	 * Although waiting in an asynchronous interface is less than ideal, we
1750 	 * do not want to use BLK_MQ_REQ_NOWAIT here because userspace might
1751 	 * not expect an EWOULDBLOCK from this condition.
1752 	 */
1753 	rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1754 			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, 0);
1755 	if (IS_ERR(rq)) {
1756 		kfree(long_cmdp);
1757 		return PTR_ERR(rq);
1758 	}
1759 	req = scsi_req(rq);
1760 
1761 	if (hp->cmd_len > BLK_MAX_CDB)
1762 		req->cmd = long_cmdp;
1763 	memcpy(req->cmd, cmd, hp->cmd_len);
1764 	req->cmd_len = hp->cmd_len;
1765 
1766 	srp->rq = rq;
1767 	rq->end_io_data = srp;
1768 	req->retries = SG_DEFAULT_RETRIES;
1769 
1770 	if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1771 		return 0;
1772 
1773 	if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1774 	    dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1775 	    !sfp->parentdp->device->host->unchecked_isa_dma &&
1776 	    blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1777 		md = NULL;
1778 	else
1779 		md = &map_data;
1780 
1781 	if (md) {
1782 		mutex_lock(&sfp->f_mutex);
1783 		if (dxfer_len <= rsv_schp->bufflen &&
1784 		    !sfp->res_in_use) {
1785 			sfp->res_in_use = 1;
1786 			sg_link_reserve(sfp, srp, dxfer_len);
1787 		} else if (hp->flags & SG_FLAG_MMAP_IO) {
1788 			res = -EBUSY; /* sfp->res_in_use == 1 */
1789 			if (dxfer_len > rsv_schp->bufflen)
1790 				res = -ENOMEM;
1791 			mutex_unlock(&sfp->f_mutex);
1792 			return res;
1793 		} else {
1794 			res = sg_build_indirect(req_schp, sfp, dxfer_len);
1795 			if (res) {
1796 				mutex_unlock(&sfp->f_mutex);
1797 				return res;
1798 			}
1799 		}
1800 		mutex_unlock(&sfp->f_mutex);
1801 
1802 		md->pages = req_schp->pages;
1803 		md->page_order = req_schp->page_order;
1804 		md->nr_entries = req_schp->k_use_sg;
1805 		md->offset = 0;
1806 		md->null_mapped = hp->dxferp ? 0 : 1;
1807 		if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1808 			md->from_user = 1;
1809 		else
1810 			md->from_user = 0;
1811 	}
1812 
1813 	if (iov_count) {
1814 		struct iovec *iov = NULL;
1815 		struct iov_iter i;
1816 
1817 #ifdef CONFIG_COMPAT
1818 		if (in_compat_syscall())
1819 			res = compat_import_iovec(rw, hp->dxferp, iov_count,
1820 						  0, &iov, &i);
1821 		else
1822 #endif
1823 			res = import_iovec(rw, hp->dxferp, iov_count,
1824 					   0, &iov, &i);
1825 		if (res < 0)
1826 			return res;
1827 
1828 		iov_iter_truncate(&i, hp->dxfer_len);
1829 		if (!iov_iter_count(&i)) {
1830 			kfree(iov);
1831 			return -EINVAL;
1832 		}
1833 
1834 		res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1835 		kfree(iov);
1836 	} else
1837 		res = blk_rq_map_user(q, rq, md, hp->dxferp,
1838 				      hp->dxfer_len, GFP_ATOMIC);
1839 
1840 	if (!res) {
1841 		srp->bio = rq->bio;
1842 
1843 		if (!md) {
1844 			req_schp->dio_in_use = 1;
1845 			hp->info |= SG_INFO_DIRECT_IO;
1846 		}
1847 	}
1848 	return res;
1849 }
1850 
1851 static int
1852 sg_finish_rem_req(Sg_request *srp)
1853 {
1854 	int ret = 0;
1855 
1856 	Sg_fd *sfp = srp->parentfp;
1857 	Sg_scatter_hold *req_schp = &srp->data;
1858 
1859 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1860 				      "sg_finish_rem_req: res_used=%d\n",
1861 				      (int) srp->res_used));
1862 	if (srp->bio)
1863 		ret = blk_rq_unmap_user(srp->bio);
1864 
1865 	if (srp->rq) {
1866 		scsi_req_free_cmd(scsi_req(srp->rq));
1867 		blk_put_request(srp->rq);
1868 	}
1869 
1870 	if (srp->res_used)
1871 		sg_unlink_reserve(sfp, srp);
1872 	else
1873 		sg_remove_scat(sfp, req_schp);
1874 
1875 	return ret;
1876 }
1877 
1878 static int
1879 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1880 {
1881 	int sg_bufflen = tablesize * sizeof(struct page *);
1882 	gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1883 
1884 	schp->pages = kzalloc(sg_bufflen, gfp_flags);
1885 	if (!schp->pages)
1886 		return -ENOMEM;
1887 	schp->sglist_len = sg_bufflen;
1888 	return tablesize;	/* number of scat_gath elements allocated */
1889 }
1890 
1891 static int
1892 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1893 {
1894 	int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1895 	int sg_tablesize = sfp->parentdp->sg_tablesize;
1896 	int blk_size = buff_size, order;
1897 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN | __GFP_ZERO;
1898 	struct sg_device *sdp = sfp->parentdp;
1899 
1900 	if (blk_size < 0)
1901 		return -EFAULT;
1902 	if (0 == blk_size)
1903 		++blk_size;	/* don't know why */
1904 	/* round request up to next highest SG_SECTOR_SZ byte boundary */
1905 	blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1906 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1907 		"sg_build_indirect: buff_size=%d, blk_size=%d\n",
1908 		buff_size, blk_size));
1909 
1910 	/* N.B. ret_sz carried into this block ... */
1911 	mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1912 	if (mx_sc_elems < 0)
1913 		return mx_sc_elems;	/* most likely -ENOMEM */
1914 
1915 	num = scatter_elem_sz;
1916 	if (unlikely(num != scatter_elem_sz_prev)) {
1917 		if (num < PAGE_SIZE) {
1918 			scatter_elem_sz = PAGE_SIZE;
1919 			scatter_elem_sz_prev = PAGE_SIZE;
1920 		} else
1921 			scatter_elem_sz_prev = num;
1922 	}
1923 
1924 	if (sdp->device->host->unchecked_isa_dma)
1925 		gfp_mask |= GFP_DMA;
1926 
1927 	order = get_order(num);
1928 retry:
1929 	ret_sz = 1 << (PAGE_SHIFT + order);
1930 
1931 	for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1932 	     k++, rem_sz -= ret_sz) {
1933 
1934 		num = (rem_sz > scatter_elem_sz_prev) ?
1935 			scatter_elem_sz_prev : rem_sz;
1936 
1937 		schp->pages[k] = alloc_pages(gfp_mask, order);
1938 		if (!schp->pages[k])
1939 			goto out;
1940 
1941 		if (num == scatter_elem_sz_prev) {
1942 			if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1943 				scatter_elem_sz = ret_sz;
1944 				scatter_elem_sz_prev = ret_sz;
1945 			}
1946 		}
1947 
1948 		SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1949 				 "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1950 				 k, num, ret_sz));
1951 	}		/* end of for loop */
1952 
1953 	schp->page_order = order;
1954 	schp->k_use_sg = k;
1955 	SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1956 			 "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1957 			 k, rem_sz));
1958 
1959 	schp->bufflen = blk_size;
1960 	if (rem_sz > 0)	/* must have failed */
1961 		return -ENOMEM;
1962 	return 0;
1963 out:
1964 	for (i = 0; i < k; i++)
1965 		__free_pages(schp->pages[i], order);
1966 
1967 	if (--order >= 0)
1968 		goto retry;
1969 
1970 	return -ENOMEM;
1971 }
1972 
1973 static void
1974 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1975 {
1976 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1977 			 "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1978 	if (schp->pages && schp->sglist_len > 0) {
1979 		if (!schp->dio_in_use) {
1980 			int k;
1981 
1982 			for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1983 				SCSI_LOG_TIMEOUT(5,
1984 					sg_printk(KERN_INFO, sfp->parentdp,
1985 					"sg_remove_scat: k=%d, pg=0x%p\n",
1986 					k, schp->pages[k]));
1987 				__free_pages(schp->pages[k], schp->page_order);
1988 			}
1989 
1990 			kfree(schp->pages);
1991 		}
1992 	}
1993 	memset(schp, 0, sizeof (*schp));
1994 }
1995 
1996 static int
1997 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1998 {
1999 	Sg_scatter_hold *schp = &srp->data;
2000 	int k, num;
2001 
2002 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2003 			 "sg_read_oxfer: num_read_xfer=%d\n",
2004 			 num_read_xfer));
2005 	if ((!outp) || (num_read_xfer <= 0))
2006 		return 0;
2007 
2008 	num = 1 << (PAGE_SHIFT + schp->page_order);
2009 	for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
2010 		if (num > num_read_xfer) {
2011 			if (copy_to_user(outp, page_address(schp->pages[k]),
2012 					   num_read_xfer))
2013 				return -EFAULT;
2014 			break;
2015 		} else {
2016 			if (copy_to_user(outp, page_address(schp->pages[k]),
2017 					   num))
2018 				return -EFAULT;
2019 			num_read_xfer -= num;
2020 			if (num_read_xfer <= 0)
2021 				break;
2022 			outp += num;
2023 		}
2024 	}
2025 
2026 	return 0;
2027 }
2028 
2029 static void
2030 sg_build_reserve(Sg_fd * sfp, int req_size)
2031 {
2032 	Sg_scatter_hold *schp = &sfp->reserve;
2033 
2034 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2035 			 "sg_build_reserve: req_size=%d\n", req_size));
2036 	do {
2037 		if (req_size < PAGE_SIZE)
2038 			req_size = PAGE_SIZE;
2039 		if (0 == sg_build_indirect(schp, sfp, req_size))
2040 			return;
2041 		else
2042 			sg_remove_scat(sfp, schp);
2043 		req_size >>= 1;	/* divide by 2 */
2044 	} while (req_size > (PAGE_SIZE / 2));
2045 }
2046 
2047 static void
2048 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
2049 {
2050 	Sg_scatter_hold *req_schp = &srp->data;
2051 	Sg_scatter_hold *rsv_schp = &sfp->reserve;
2052 	int k, num, rem;
2053 
2054 	srp->res_used = 1;
2055 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
2056 			 "sg_link_reserve: size=%d\n", size));
2057 	rem = size;
2058 
2059 	num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
2060 	for (k = 0; k < rsv_schp->k_use_sg; k++) {
2061 		if (rem <= num) {
2062 			req_schp->k_use_sg = k + 1;
2063 			req_schp->sglist_len = rsv_schp->sglist_len;
2064 			req_schp->pages = rsv_schp->pages;
2065 
2066 			req_schp->bufflen = size;
2067 			req_schp->page_order = rsv_schp->page_order;
2068 			break;
2069 		} else
2070 			rem -= num;
2071 	}
2072 
2073 	if (k >= rsv_schp->k_use_sg)
2074 		SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2075 				 "sg_link_reserve: BAD size\n"));
2076 }
2077 
2078 static void
2079 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2080 {
2081 	Sg_scatter_hold *req_schp = &srp->data;
2082 
2083 	SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2084 				      "sg_unlink_reserve: req->k_use_sg=%d\n",
2085 				      (int) req_schp->k_use_sg));
2086 	req_schp->k_use_sg = 0;
2087 	req_schp->bufflen = 0;
2088 	req_schp->pages = NULL;
2089 	req_schp->page_order = 0;
2090 	req_schp->sglist_len = 0;
2091 	srp->res_used = 0;
2092 	/* Called without mutex lock to avoid deadlock */
2093 	sfp->res_in_use = 0;
2094 }
2095 
2096 static Sg_request *
2097 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2098 {
2099 	Sg_request *resp;
2100 	unsigned long iflags;
2101 
2102 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2103 	list_for_each_entry(resp, &sfp->rq_list, entry) {
2104 		/* look for requests that are ready + not SG_IO owned */
2105 		if ((1 == resp->done) && (!resp->sg_io_owned) &&
2106 		    ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2107 			resp->done = 2;	/* guard against other readers */
2108 			write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2109 			return resp;
2110 		}
2111 	}
2112 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2113 	return NULL;
2114 }
2115 
2116 /* always adds to end of list */
2117 static Sg_request *
2118 sg_add_request(Sg_fd * sfp)
2119 {
2120 	int k;
2121 	unsigned long iflags;
2122 	Sg_request *rp = sfp->req_arr;
2123 
2124 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2125 	if (!list_empty(&sfp->rq_list)) {
2126 		if (!sfp->cmd_q)
2127 			goto out_unlock;
2128 
2129 		for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2130 			if (!rp->parentfp)
2131 				break;
2132 		}
2133 		if (k >= SG_MAX_QUEUE)
2134 			goto out_unlock;
2135 	}
2136 	memset(rp, 0, sizeof (Sg_request));
2137 	rp->parentfp = sfp;
2138 	rp->header.duration = jiffies_to_msecs(jiffies);
2139 	list_add_tail(&rp->entry, &sfp->rq_list);
2140 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2141 	return rp;
2142 out_unlock:
2143 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2144 	return NULL;
2145 }
2146 
2147 /* Return of 1 for found; 0 for not found */
2148 static int
2149 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2150 {
2151 	unsigned long iflags;
2152 	int res = 0;
2153 
2154 	if (!sfp || !srp || list_empty(&sfp->rq_list))
2155 		return res;
2156 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2157 	if (!list_empty(&srp->entry)) {
2158 		list_del(&srp->entry);
2159 		srp->parentfp = NULL;
2160 		res = 1;
2161 	}
2162 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2163 	return res;
2164 }
2165 
2166 static Sg_fd *
2167 sg_add_sfp(Sg_device * sdp)
2168 {
2169 	Sg_fd *sfp;
2170 	unsigned long iflags;
2171 	int bufflen;
2172 
2173 	sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2174 	if (!sfp)
2175 		return ERR_PTR(-ENOMEM);
2176 
2177 	init_waitqueue_head(&sfp->read_wait);
2178 	rwlock_init(&sfp->rq_list_lock);
2179 	INIT_LIST_HEAD(&sfp->rq_list);
2180 	kref_init(&sfp->f_ref);
2181 	mutex_init(&sfp->f_mutex);
2182 	sfp->timeout = SG_DEFAULT_TIMEOUT;
2183 	sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2184 	sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2185 	sfp->cmd_q = SG_DEF_COMMAND_Q;
2186 	sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2187 	sfp->parentdp = sdp;
2188 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2189 	if (atomic_read(&sdp->detaching)) {
2190 		write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2191 		kfree(sfp);
2192 		return ERR_PTR(-ENODEV);
2193 	}
2194 	list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2195 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2196 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2197 				      "sg_add_sfp: sfp=0x%p\n", sfp));
2198 	if (unlikely(sg_big_buff != def_reserved_size))
2199 		sg_big_buff = def_reserved_size;
2200 
2201 	bufflen = min_t(int, sg_big_buff,
2202 			max_sectors_bytes(sdp->device->request_queue));
2203 	sg_build_reserve(sfp, bufflen);
2204 	SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2205 				      "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2206 				      sfp->reserve.bufflen,
2207 				      sfp->reserve.k_use_sg));
2208 
2209 	kref_get(&sdp->d_ref);
2210 	__module_get(THIS_MODULE);
2211 	return sfp;
2212 }
2213 
2214 static void
2215 sg_remove_sfp_usercontext(struct work_struct *work)
2216 {
2217 	struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2218 	struct sg_device *sdp = sfp->parentdp;
2219 	Sg_request *srp;
2220 	unsigned long iflags;
2221 
2222 	/* Cleanup any responses which were never read(). */
2223 	write_lock_irqsave(&sfp->rq_list_lock, iflags);
2224 	while (!list_empty(&sfp->rq_list)) {
2225 		srp = list_first_entry(&sfp->rq_list, Sg_request, entry);
2226 		sg_finish_rem_req(srp);
2227 		list_del(&srp->entry);
2228 		srp->parentfp = NULL;
2229 	}
2230 	write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2231 
2232 	if (sfp->reserve.bufflen > 0) {
2233 		SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2234 				"sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2235 				(int) sfp->reserve.bufflen,
2236 				(int) sfp->reserve.k_use_sg));
2237 		sg_remove_scat(sfp, &sfp->reserve);
2238 	}
2239 
2240 	SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2241 			"sg_remove_sfp: sfp=0x%p\n", sfp));
2242 	kfree(sfp);
2243 
2244 	scsi_device_put(sdp->device);
2245 	kref_put(&sdp->d_ref, sg_device_destroy);
2246 	module_put(THIS_MODULE);
2247 }
2248 
2249 static void
2250 sg_remove_sfp(struct kref *kref)
2251 {
2252 	struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2253 	struct sg_device *sdp = sfp->parentdp;
2254 	unsigned long iflags;
2255 
2256 	write_lock_irqsave(&sdp->sfd_lock, iflags);
2257 	list_del(&sfp->sfd_siblings);
2258 	write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2259 
2260 	INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2261 	schedule_work(&sfp->ew.work);
2262 }
2263 
2264 #ifdef CONFIG_SCSI_PROC_FS
2265 static int
2266 sg_idr_max_id(int id, void *p, void *data)
2267 {
2268 	int *k = data;
2269 
2270 	if (*k < id)
2271 		*k = id;
2272 
2273 	return 0;
2274 }
2275 
2276 static int
2277 sg_last_dev(void)
2278 {
2279 	int k = -1;
2280 	unsigned long iflags;
2281 
2282 	read_lock_irqsave(&sg_index_lock, iflags);
2283 	idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2284 	read_unlock_irqrestore(&sg_index_lock, iflags);
2285 	return k + 1;		/* origin 1 */
2286 }
2287 #endif
2288 
2289 /* must be called with sg_index_lock held */
2290 static Sg_device *sg_lookup_dev(int dev)
2291 {
2292 	return idr_find(&sg_index_idr, dev);
2293 }
2294 
2295 static Sg_device *
2296 sg_get_dev(int dev)
2297 {
2298 	struct sg_device *sdp;
2299 	unsigned long flags;
2300 
2301 	read_lock_irqsave(&sg_index_lock, flags);
2302 	sdp = sg_lookup_dev(dev);
2303 	if (!sdp)
2304 		sdp = ERR_PTR(-ENXIO);
2305 	else if (atomic_read(&sdp->detaching)) {
2306 		/* If sdp->detaching, then the refcount may already be 0, in
2307 		 * which case it would be a bug to do kref_get().
2308 		 */
2309 		sdp = ERR_PTR(-ENODEV);
2310 	} else
2311 		kref_get(&sdp->d_ref);
2312 	read_unlock_irqrestore(&sg_index_lock, flags);
2313 
2314 	return sdp;
2315 }
2316 
2317 #ifdef CONFIG_SCSI_PROC_FS
2318 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2319 
2320 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2321 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2322 			          size_t count, loff_t *off);
2323 static const struct file_operations adio_fops = {
2324 	.owner = THIS_MODULE,
2325 	.open = sg_proc_single_open_adio,
2326 	.read = seq_read,
2327 	.llseek = seq_lseek,
2328 	.write = sg_proc_write_adio,
2329 	.release = single_release,
2330 };
2331 
2332 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2333 static ssize_t sg_proc_write_dressz(struct file *filp,
2334 		const char __user *buffer, size_t count, loff_t *off);
2335 static const struct file_operations dressz_fops = {
2336 	.owner = THIS_MODULE,
2337 	.open = sg_proc_single_open_dressz,
2338 	.read = seq_read,
2339 	.llseek = seq_lseek,
2340 	.write = sg_proc_write_dressz,
2341 	.release = single_release,
2342 };
2343 
2344 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2345 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2346 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2347 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2348 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2349 static void dev_seq_stop(struct seq_file *s, void *v);
2350 static const struct seq_operations dev_seq_ops = {
2351 	.start = dev_seq_start,
2352 	.next  = dev_seq_next,
2353 	.stop  = dev_seq_stop,
2354 	.show  = sg_proc_seq_show_dev,
2355 };
2356 
2357 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2358 static const struct seq_operations devstrs_seq_ops = {
2359 	.start = dev_seq_start,
2360 	.next  = dev_seq_next,
2361 	.stop  = dev_seq_stop,
2362 	.show  = sg_proc_seq_show_devstrs,
2363 };
2364 
2365 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2366 static const struct seq_operations debug_seq_ops = {
2367 	.start = dev_seq_start,
2368 	.next  = dev_seq_next,
2369 	.stop  = dev_seq_stop,
2370 	.show  = sg_proc_seq_show_debug,
2371 };
2372 
2373 static int
2374 sg_proc_init(void)
2375 {
2376 	struct proc_dir_entry *p;
2377 
2378 	p = proc_mkdir("scsi/sg", NULL);
2379 	if (!p)
2380 		return 1;
2381 
2382 	proc_create("allow_dio", S_IRUGO | S_IWUSR, p, &adio_fops);
2383 	proc_create_seq("debug", S_IRUGO, p, &debug_seq_ops);
2384 	proc_create("def_reserved_size", S_IRUGO | S_IWUSR, p, &dressz_fops);
2385 	proc_create_single("device_hdr", S_IRUGO, p, sg_proc_seq_show_devhdr);
2386 	proc_create_seq("devices", S_IRUGO, p, &dev_seq_ops);
2387 	proc_create_seq("device_strs", S_IRUGO, p, &devstrs_seq_ops);
2388 	proc_create_single("version", S_IRUGO, p, sg_proc_seq_show_version);
2389 	return 0;
2390 }
2391 
2392 
2393 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2394 {
2395 	seq_printf(s, "%d\n", *((int *)s->private));
2396 	return 0;
2397 }
2398 
2399 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2400 {
2401 	return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2402 }
2403 
2404 static ssize_t
2405 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2406 		   size_t count, loff_t *off)
2407 {
2408 	int err;
2409 	unsigned long num;
2410 
2411 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2412 		return -EACCES;
2413 	err = kstrtoul_from_user(buffer, count, 0, &num);
2414 	if (err)
2415 		return err;
2416 	sg_allow_dio = num ? 1 : 0;
2417 	return count;
2418 }
2419 
2420 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2421 {
2422 	return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2423 }
2424 
2425 static ssize_t
2426 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2427 		     size_t count, loff_t *off)
2428 {
2429 	int err;
2430 	unsigned long k = ULONG_MAX;
2431 
2432 	if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2433 		return -EACCES;
2434 
2435 	err = kstrtoul_from_user(buffer, count, 0, &k);
2436 	if (err)
2437 		return err;
2438 	if (k <= 1048576) {	/* limit "big buff" to 1 MB */
2439 		sg_big_buff = k;
2440 		return count;
2441 	}
2442 	return -ERANGE;
2443 }
2444 
2445 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2446 {
2447 	seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2448 		   sg_version_date);
2449 	return 0;
2450 }
2451 
2452 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2453 {
2454 	seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2455 	return 0;
2456 }
2457 
2458 struct sg_proc_deviter {
2459 	loff_t	index;
2460 	size_t	max;
2461 };
2462 
2463 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2464 {
2465 	struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2466 
2467 	s->private = it;
2468 	if (! it)
2469 		return NULL;
2470 
2471 	it->index = *pos;
2472 	it->max = sg_last_dev();
2473 	if (it->index >= it->max)
2474 		return NULL;
2475 	return it;
2476 }
2477 
2478 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2479 {
2480 	struct sg_proc_deviter * it = s->private;
2481 
2482 	*pos = ++it->index;
2483 	return (it->index < it->max) ? it : NULL;
2484 }
2485 
2486 static void dev_seq_stop(struct seq_file *s, void *v)
2487 {
2488 	kfree(s->private);
2489 }
2490 
2491 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2492 {
2493 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2494 	Sg_device *sdp;
2495 	struct scsi_device *scsidp;
2496 	unsigned long iflags;
2497 
2498 	read_lock_irqsave(&sg_index_lock, iflags);
2499 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2500 	if ((NULL == sdp) || (NULL == sdp->device) ||
2501 	    (atomic_read(&sdp->detaching)))
2502 		seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2503 	else {
2504 		scsidp = sdp->device;
2505 		seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2506 			      scsidp->host->host_no, scsidp->channel,
2507 			      scsidp->id, scsidp->lun, (int) scsidp->type,
2508 			      1,
2509 			      (int) scsidp->queue_depth,
2510 			      (int) atomic_read(&scsidp->device_busy),
2511 			      (int) scsi_device_online(scsidp));
2512 	}
2513 	read_unlock_irqrestore(&sg_index_lock, iflags);
2514 	return 0;
2515 }
2516 
2517 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2518 {
2519 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2520 	Sg_device *sdp;
2521 	struct scsi_device *scsidp;
2522 	unsigned long iflags;
2523 
2524 	read_lock_irqsave(&sg_index_lock, iflags);
2525 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2526 	scsidp = sdp ? sdp->device : NULL;
2527 	if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2528 		seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2529 			   scsidp->vendor, scsidp->model, scsidp->rev);
2530 	else
2531 		seq_puts(s, "<no active device>\n");
2532 	read_unlock_irqrestore(&sg_index_lock, iflags);
2533 	return 0;
2534 }
2535 
2536 /* must be called while holding sg_index_lock */
2537 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2538 {
2539 	int k, new_interface, blen, usg;
2540 	Sg_request *srp;
2541 	Sg_fd *fp;
2542 	const sg_io_hdr_t *hp;
2543 	const char * cp;
2544 	unsigned int ms;
2545 
2546 	k = 0;
2547 	list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2548 		k++;
2549 		read_lock(&fp->rq_list_lock); /* irqs already disabled */
2550 		seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2551 			   "(res)sgat=%d low_dma=%d\n", k,
2552 			   jiffies_to_msecs(fp->timeout),
2553 			   fp->reserve.bufflen,
2554 			   (int) fp->reserve.k_use_sg,
2555 			   (int) sdp->device->host->unchecked_isa_dma);
2556 		seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2557 			   (int) fp->cmd_q, (int) fp->force_packid,
2558 			   (int) fp->keep_orphan);
2559 		list_for_each_entry(srp, &fp->rq_list, entry) {
2560 			hp = &srp->header;
2561 			new_interface = (hp->interface_id == '\0') ? 0 : 1;
2562 			if (srp->res_used) {
2563 				if (new_interface &&
2564 				    (SG_FLAG_MMAP_IO & hp->flags))
2565 					cp = "     mmap>> ";
2566 				else
2567 					cp = "     rb>> ";
2568 			} else {
2569 				if (SG_INFO_DIRECT_IO_MASK & hp->info)
2570 					cp = "     dio>> ";
2571 				else
2572 					cp = "     ";
2573 			}
2574 			seq_puts(s, cp);
2575 			blen = srp->data.bufflen;
2576 			usg = srp->data.k_use_sg;
2577 			seq_puts(s, srp->done ?
2578 				 ((1 == srp->done) ?  "rcv:" : "fin:")
2579 				  : "act:");
2580 			seq_printf(s, " id=%d blen=%d",
2581 				   srp->header.pack_id, blen);
2582 			if (srp->done)
2583 				seq_printf(s, " dur=%d", hp->duration);
2584 			else {
2585 				ms = jiffies_to_msecs(jiffies);
2586 				seq_printf(s, " t_o/elap=%d/%d",
2587 					(new_interface ? hp->timeout :
2588 						  jiffies_to_msecs(fp->timeout)),
2589 					(ms > hp->duration ? ms - hp->duration : 0));
2590 			}
2591 			seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2592 				   (int) srp->data.cmd_opcode);
2593 		}
2594 		if (list_empty(&fp->rq_list))
2595 			seq_puts(s, "     No requests active\n");
2596 		read_unlock(&fp->rq_list_lock);
2597 	}
2598 }
2599 
2600 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2601 {
2602 	struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2603 	Sg_device *sdp;
2604 	unsigned long iflags;
2605 
2606 	if (it && (0 == it->index))
2607 		seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2608 			   (int)it->max, sg_big_buff);
2609 
2610 	read_lock_irqsave(&sg_index_lock, iflags);
2611 	sdp = it ? sg_lookup_dev(it->index) : NULL;
2612 	if (NULL == sdp)
2613 		goto skip;
2614 	read_lock(&sdp->sfd_lock);
2615 	if (!list_empty(&sdp->sfds)) {
2616 		seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2617 		if (atomic_read(&sdp->detaching))
2618 			seq_puts(s, "detaching pending close ");
2619 		else if (sdp->device) {
2620 			struct scsi_device *scsidp = sdp->device;
2621 
2622 			seq_printf(s, "%d:%d:%d:%llu   em=%d",
2623 				   scsidp->host->host_no,
2624 				   scsidp->channel, scsidp->id,
2625 				   scsidp->lun,
2626 				   scsidp->host->hostt->emulated);
2627 		}
2628 		seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2629 			   sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2630 		sg_proc_debug_helper(s, sdp);
2631 	}
2632 	read_unlock(&sdp->sfd_lock);
2633 skip:
2634 	read_unlock_irqrestore(&sg_index_lock, iflags);
2635 	return 0;
2636 }
2637 
2638 #endif				/* CONFIG_SCSI_PROC_FS */
2639 
2640 module_init(init_sg);
2641 module_exit(exit_sg);
2642