drivers/scsi/sg.c

   1 /*
   2  *  History:
   3  *  Started: Aug 9 by Lawrence Foard (entropy@world.std.com),
   4  *           to allow user process control of SCSI devices.
   5  *  Development Sponsored by Killy Corp. NY NY
   6  *
   7  * Original driver (sg.c):
   8  *        Copyright (C) 1992 Lawrence Foard
   9  * Version 2 and 3 extensions to driver:
  10  *        Copyright (C) 1998 - 2014 Douglas Gilbert
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License as published by
  14  * the Free Software Foundation; either version 2, or (at your option)
  15  * any later version.
  16  *
  17  */
  18
  19 static int sg_version_num = 30536;      /* 2 digits for each component */
  20 #define SG_VERSION_STR "3.5.36"
  21
  22 /*
  23  *  D. P. Gilbert (dgilbert@interlog.com), notes:
  24  *      - scsi logging is available via SCSI_LOG_TIMEOUT macros. First
  25  *        the kernel/module needs to be built with CONFIG_SCSI_LOGGING
  26  *        (otherwise the macros compile to empty statements).
  27  *
  28  */
  29 #include <linux/module.h>
  30
  31 #include <linux/fs.h>
  32 #include <linux/kernel.h>
  33 #include <linux/sched.h>
  34 #include <linux/string.h>
  35 #include <linux/mm.h>
  36 #include <linux/errno.h>
  37 #include <linux/mtio.h>
  38 #include <linux/ioctl.h>
  39 #include <linux/slab.h>
  40 #include <linux/fcntl.h>
  41 #include <linux/init.h>
  42 #include <linux/poll.h>
  43 #include <linux/moduleparam.h>
  44 #include <linux/cdev.h>
  45 #include <linux/idr.h>
  46 #include <linux/seq_file.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/delay.h>
  49 #include <linux/blktrace_api.h>
  50 #include <linux/mutex.h>
  51 #include <linux/atomic.h>
  52 #include <linux/ratelimit.h>
  53 #include <linux/uio.h>
  54
  55 #include "scsi.h"
  56 #include <scsi/scsi_dbg.h>
  57 #include <scsi/scsi_host.h>
  58 #include <scsi/scsi_driver.h>
  59 #include <scsi/scsi_ioctl.h>
  60 #include <scsi/sg.h>
  61
  62 #include "scsi_logging.h"
  63
  64 #ifdef CONFIG_SCSI_PROC_FS
  65 #include <linux/proc_fs.h>
  66 static char *sg_version_date = "20140603";
  67
  68 static int sg_proc_init(void);
  69 static void sg_proc_cleanup(void);
  70 #endif
  71
  72 #define SG_ALLOW_DIO_DEF 0
  73
  74 #define SG_MAX_DEVS 32768
  75
  76 /* SG_MAX_CDB_SIZE should be 260 (spc4r37 section 3.1.30) however the type
  77  * of sg_io_hdr::cmd_len can only represent 255. All SCSI commands greater
  78  * than 16 bytes are "variable length" whose length is a multiple of 4
  79  */
  80 #define SG_MAX_CDB_SIZE 252
  81
  82 #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 sg_request *nextrp;      /* NULL -> tail request (slist) */
 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         int timeout;            /* defaults to SG_DEFAULT_TIMEOUT      */
 146         int timeout_user;       /* defaults to SG_DEFAULT_TIMEOUT_USER */
 147         Sg_scatter_hold reserve;        /* buffer held for this file descriptor */
 148         unsigned save_scat_len; /* original length of trunc. scat. element */
 149         Sg_request *headrp;     /* head of request slist, NULL->empty */
 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 low_dma;           /* as in parent but possibly overridden to 1 */
 153         char force_packid;      /* 1 -> pack_id input to read(), 0 -> ignored */
 154         char cmd_q;             /* 1 -> allow command queuing, 0 -> don't */
 155         unsigned char next_cmd_len; /* 0: automatic, >0: use on next write() */
 156         char keep_orphan;       /* 0 -> drop orphan (def), 1 -> keep for read() */
 157         char mmap_called;       /* 0 -> mmap() never called on this fd */
 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, int uptodate);
 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 int sg_res_in_use(Sg_fd * sfp);
 202 static Sg_device *sg_get_dev(int dev);
 203 static void sg_device_destroy(struct kref *kref);
 204
 205 #define SZ_SG_HEADER sizeof(struct sg_header)
 206 #define SZ_SG_IO_HDR sizeof(sg_io_hdr_t)
 207 #define SZ_SG_IOVEC sizeof(sg_iovec_t)
 208 #define SZ_SG_REQ_INFO sizeof(sg_req_info_t)
 209
 210 #define sg_printk(prefix, sdp, fmt, a...) \
 211         sdev_prefix_printk(prefix, (sdp)->device,               \
 212                            (sdp)->disk->disk_name, fmt, ##a)
 213
 214 static int sg_allow_access(struct file *filp, unsigned char *cmd)
 215 {
 216         struct sg_fd *sfp = filp->private_data;
 217
 218         if (sfp->parentdp->device->type == TYPE_SCANNER)
 219                 return 0;
 220
 221         return blk_verify_command(cmd, filp->f_mode & FMODE_WRITE);
 222 }
 223
 224 static int
 225 open_wait(Sg_device *sdp, int flags)
 226 {
 227         int retval = 0;
 228
 229         if (flags & O_EXCL) {
 230                 while (sdp->open_cnt > 0) {
 231                         mutex_unlock(&sdp->open_rel_lock);
 232                         retval = wait_event_interruptible(sdp->open_wait,
 233                                         (atomic_read(&sdp->detaching) ||
 234                                          !sdp->open_cnt));
 235                         mutex_lock(&sdp->open_rel_lock);
 236
 237                         if (retval) /* -ERESTARTSYS */
 238                                 return retval;
 239                         if (atomic_read(&sdp->detaching))
 240                                 return -ENODEV;
 241                 }
 242         } else {
 243                 while (sdp->exclude) {
 244                         mutex_unlock(&sdp->open_rel_lock);
 245                         retval = wait_event_interruptible(sdp->open_wait,
 246                                         (atomic_read(&sdp->detaching) ||
 247                                          !sdp->exclude));
 248                         mutex_lock(&sdp->open_rel_lock);
 249
 250                         if (retval) /* -ERESTARTSYS */
 251                                 return retval;
 252                         if (atomic_read(&sdp->detaching))
 253                                 return -ENODEV;
 254                 }
 255         }
 256
 257         return retval;
 258 }
 259
 260 /* Returns 0 on success, else a negated errno value */
 261 static int
 262 sg_open(struct inode *inode, struct file *filp)
 263 {
 264         int dev = iminor(inode);
 265         int flags = filp->f_flags;
 266         struct request_queue *q;
 267         Sg_device *sdp;
 268         Sg_fd *sfp;
 269         int retval;
 270
 271         nonseekable_open(inode, filp);
 272         if ((flags & O_EXCL) && (O_RDONLY == (flags & O_ACCMODE)))
 273                 return -EPERM; /* Can't lock it with read only access */
 274         sdp = sg_get_dev(dev);
 275         if (IS_ERR(sdp))
 276                 return PTR_ERR(sdp);
 277
 278         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 279                                       "sg_open: flags=0x%x\n", flags));
 280
 281         /* This driver's module count bumped by fops_get in <linux/fs.h> */
 282         /* Prevent the device driver from vanishing while we sleep */
 283         retval = scsi_device_get(sdp->device);
 284         if (retval)
 285                 goto sg_put;
 286
 287         retval = scsi_autopm_get_device(sdp->device);
 288         if (retval)
 289                 goto sdp_put;
 290
 291         /* scsi_block_when_processing_errors() may block so bypass
 292          * check if O_NONBLOCK. Permits SCSI commands to be issued
 293          * during error recovery. Tread carefully. */
 294         if (!((flags & O_NONBLOCK) ||
 295               scsi_block_when_processing_errors(sdp->device))) {
 296                 retval = -ENXIO;
 297                 /* we are in error recovery for this device */
 298                 goto error_out;
 299         }
 300
 301         mutex_lock(&sdp->open_rel_lock);
 302         if (flags & O_NONBLOCK) {
 303                 if (flags & O_EXCL) {
 304                         if (sdp->open_cnt > 0) {
 305                                 retval = -EBUSY;
 306                                 goto error_mutex_locked;
 307                         }
 308                 } else {
 309                         if (sdp->exclude) {
 310                                 retval = -EBUSY;
 311                                 goto error_mutex_locked;
 312                         }
 313                 }
 314         } else {
 315                 retval = open_wait(sdp, flags);
 316                 if (retval) /* -ERESTARTSYS or -ENODEV */
 317                         goto error_mutex_locked;
 318         }
 319
 320         /* N.B. at this point we are holding the open_rel_lock */
 321         if (flags & O_EXCL)
 322                 sdp->exclude = true;
 323
 324         if (sdp->open_cnt < 1) {  /* no existing opens */
 325                 sdp->sgdebug = 0;
 326                 q = sdp->device->request_queue;
 327                 sdp->sg_tablesize = queue_max_segments(q);
 328         }
 329         sfp = sg_add_sfp(sdp);
 330         if (IS_ERR(sfp)) {
 331                 retval = PTR_ERR(sfp);
 332                 goto out_undo;
 333         }
 334
 335         filp->private_data = sfp;
 336         sdp->open_cnt++;
 337         mutex_unlock(&sdp->open_rel_lock);
 338
 339         retval = 0;
 340 sg_put:
 341         kref_put(&sdp->d_ref, sg_device_destroy);
 342         return retval;
 343
 344 out_undo:
 345         if (flags & O_EXCL) {
 346                 sdp->exclude = false;   /* undo if error */
 347                 wake_up_interruptible(&sdp->open_wait);
 348         }
 349 error_mutex_locked:
 350         mutex_unlock(&sdp->open_rel_lock);
 351 error_out:
 352         scsi_autopm_put_device(sdp->device);
 353 sdp_put:
 354         scsi_device_put(sdp->device);
 355         goto sg_put;
 356 }
 357
 358 /* Release resources associated with a successful sg_open()
 359  * Returns 0 on success, else a negated errno value */
 360 static int
 361 sg_release(struct inode *inode, struct file *filp)
 362 {
 363         Sg_device *sdp;
 364         Sg_fd *sfp;
 365
 366         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 367                 return -ENXIO;
 368         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp, "sg_release\n"));
 369
 370         mutex_lock(&sdp->open_rel_lock);
 371         scsi_autopm_put_device(sdp->device);
 372         kref_put(&sfp->f_ref, sg_remove_sfp);
 373         sdp->open_cnt--;
 374
 375         /* possibly many open()s waiting on exlude clearing, start many;
 376          * only open(O_EXCL)s wait on 0==open_cnt so only start one */
 377         if (sdp->exclude) {
 378                 sdp->exclude = false;
 379                 wake_up_interruptible_all(&sdp->open_wait);
 380         } else if (0 == sdp->open_cnt) {
 381                 wake_up_interruptible(&sdp->open_wait);
 382         }
 383         mutex_unlock(&sdp->open_rel_lock);
 384         return 0;
 385 }
 386
 387 static ssize_t
 388 sg_read(struct file *filp, char __user *buf, size_t count, loff_t * ppos)
 389 {
 390         Sg_device *sdp;
 391         Sg_fd *sfp;
 392         Sg_request *srp;
 393         int req_pack_id = -1;
 394         sg_io_hdr_t *hp;
 395         struct sg_header *old_hdr = NULL;
 396         int retval = 0;
 397
 398         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 399                 return -ENXIO;
 400         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 401                                       "sg_read: count=%d\n", (int) count));
 402
 403         if (!access_ok(VERIFY_WRITE, buf, count))
 404                 return -EFAULT;
 405         if (sfp->force_packid && (count >= SZ_SG_HEADER)) {
 406                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 407                 if (!old_hdr)
 408                         return -ENOMEM;
 409                 if (__copy_from_user(old_hdr, buf, SZ_SG_HEADER)) {
 410                         retval = -EFAULT;
 411                         goto free_old_hdr;
 412                 }
 413                 if (old_hdr->reply_len < 0) {
 414                         if (count >= SZ_SG_IO_HDR) {
 415                                 sg_io_hdr_t *new_hdr;
 416                                 new_hdr = kmalloc(SZ_SG_IO_HDR, GFP_KERNEL);
 417                                 if (!new_hdr) {
 418                                         retval = -ENOMEM;
 419                                         goto free_old_hdr;
 420                                 }
 421                                 retval =__copy_from_user
 422                                     (new_hdr, buf, SZ_SG_IO_HDR);
 423                                 req_pack_id = new_hdr->pack_id;
 424                                 kfree(new_hdr);
 425                                 if (retval) {
 426                                         retval = -EFAULT;
 427                                         goto free_old_hdr;
 428                                 }
 429                         }
 430                 } else
 431                         req_pack_id = old_hdr->pack_id;
 432         }
 433         srp = sg_get_rq_mark(sfp, req_pack_id);
 434         if (!srp) {             /* now wait on packet to arrive */
 435                 if (atomic_read(&sdp->detaching)) {
 436                         retval = -ENODEV;
 437                         goto free_old_hdr;
 438                 }
 439                 if (filp->f_flags & O_NONBLOCK) {
 440                         retval = -EAGAIN;
 441                         goto free_old_hdr;
 442                 }
 443                 retval = wait_event_interruptible(sfp->read_wait,
 444                         (atomic_read(&sdp->detaching) ||
 445                         (srp = sg_get_rq_mark(sfp, req_pack_id))));
 446                 if (atomic_read(&sdp->detaching)) {
 447                         retval = -ENODEV;
 448                         goto free_old_hdr;
 449                 }
 450                 if (retval) {
 451                         /* -ERESTARTSYS as signal hit process */
 452                         goto free_old_hdr;
 453                 }
 454         }
 455         if (srp->header.interface_id != '\0') {
 456                 retval = sg_new_read(sfp, buf, count, srp);
 457                 goto free_old_hdr;
 458         }
 459
 460         hp = &srp->header;
 461         if (old_hdr == NULL) {
 462                 old_hdr = kmalloc(SZ_SG_HEADER, GFP_KERNEL);
 463                 if (! old_hdr) {
 464                         retval = -ENOMEM;
 465                         goto free_old_hdr;
 466                 }
 467         }
 468         memset(old_hdr, 0, SZ_SG_HEADER);
 469         old_hdr->reply_len = (int) hp->timeout;
 470         old_hdr->pack_len = old_hdr->reply_len; /* old, strange behaviour */
 471         old_hdr->pack_id = hp->pack_id;
 472         old_hdr->twelve_byte =
 473             ((srp->data.cmd_opcode >= 0xc0) && (12 == hp->cmd_len)) ? 1 : 0;
 474         old_hdr->target_status = hp->masked_status;
 475         old_hdr->host_status = hp->host_status;
 476         old_hdr->driver_status = hp->driver_status;
 477         if ((CHECK_CONDITION & hp->masked_status) ||
 478             (DRIVER_SENSE & hp->driver_status))
 479                 memcpy(old_hdr->sense_buffer, srp->sense_b,
 480                        sizeof (old_hdr->sense_buffer));
 481         switch (hp->host_status) {
 482         /* This setup of 'result' is for backward compatibility and is best
 483            ignored by the user who should use target, host + driver status */
 484         case DID_OK:
 485         case DID_PASSTHROUGH:
 486         case DID_SOFT_ERROR:
 487                 old_hdr->result = 0;
 488                 break;
 489         case DID_NO_CONNECT:
 490         case DID_BUS_BUSY:
 491         case DID_TIME_OUT:
 492                 old_hdr->result = EBUSY;
 493                 break;
 494         case DID_BAD_TARGET:
 495         case DID_ABORT:
 496         case DID_PARITY:
 497         case DID_RESET:
 498         case DID_BAD_INTR:
 499                 old_hdr->result = EIO;
 500                 break;
 501         case DID_ERROR:
 502                 old_hdr->result = (srp->sense_b[0] == 0 &&
 503                                   hp->masked_status == GOOD) ? 0 : EIO;
 504                 break;
 505         default:
 506                 old_hdr->result = EIO;
 507                 break;
 508         }
 509
 510         /* Now copy the result back to the user buffer.  */
 511         if (count >= SZ_SG_HEADER) {
 512                 if (__copy_to_user(buf, old_hdr, SZ_SG_HEADER)) {
 513                         retval = -EFAULT;
 514                         goto free_old_hdr;
 515                 }
 516                 buf += SZ_SG_HEADER;
 517                 if (count > old_hdr->reply_len)
 518                         count = old_hdr->reply_len;
 519                 if (count > SZ_SG_HEADER) {
 520                         if (sg_read_oxfer(srp, buf, count - SZ_SG_HEADER)) {
 521                                 retval = -EFAULT;
 522                                 goto free_old_hdr;
 523                         }
 524                 }
 525         } else
 526                 count = (old_hdr->result == 0) ? 0 : -EIO;
 527         sg_finish_rem_req(srp);
 528         retval = count;
 529 free_old_hdr:
 530         kfree(old_hdr);
 531         return retval;
 532 }
 533
 534 static ssize_t
 535 sg_new_read(Sg_fd * sfp, char __user *buf, size_t count, Sg_request * srp)
 536 {
 537         sg_io_hdr_t *hp = &srp->header;
 538         int err = 0, err2;
 539         int len;
 540
 541         if (count < SZ_SG_IO_HDR) {
 542                 err = -EINVAL;
 543                 goto err_out;
 544         }
 545         hp->sb_len_wr = 0;
 546         if ((hp->mx_sb_len > 0) && hp->sbp) {
 547                 if ((CHECK_CONDITION & hp->masked_status) ||
 548                     (DRIVER_SENSE & hp->driver_status)) {
 549                         int sb_len = SCSI_SENSE_BUFFERSIZE;
 550                         sb_len = (hp->mx_sb_len > sb_len) ? sb_len : hp->mx_sb_len;
 551                         len = 8 + (int) srp->sense_b[7];        /* Additional sense length field */
 552                         len = (len > sb_len) ? sb_len : len;
 553                         if (copy_to_user(hp->sbp, srp->sense_b, len)) {
 554                                 err = -EFAULT;
 555                                 goto err_out;
 556                         }
 557                         hp->sb_len_wr = len;
 558                 }
 559         }
 560         if (hp->masked_status || hp->host_status || hp->driver_status)
 561                 hp->info |= SG_INFO_CHECK;
 562         if (copy_to_user(buf, hp, SZ_SG_IO_HDR)) {
 563                 err = -EFAULT;
 564                 goto err_out;
 565         }
 566 err_out:
 567         err2 = sg_finish_rem_req(srp);
 568         return err ? : err2 ? : count;
 569 }
 570
 571 static ssize_t
 572 sg_write(struct file *filp, const char __user *buf, size_t count, loff_t * ppos)
 573 {
 574         int mxsize, cmd_size, k;
 575         int input_size, blocking;
 576         unsigned char opcode;
 577         Sg_device *sdp;
 578         Sg_fd *sfp;
 579         Sg_request *srp;
 580         struct sg_header old_hdr;
 581         sg_io_hdr_t *hp;
 582         unsigned char cmnd[SG_MAX_CDB_SIZE];
 583
 584         if (unlikely(segment_eq(get_fs(), KERNEL_DS)))
 585                 return -EINVAL;
 586
 587         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 588                 return -ENXIO;
 589         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 590                                       "sg_write: count=%d\n", (int) count));
 591         if (atomic_read(&sdp->detaching))
 592                 return -ENODEV;
 593         if (!((filp->f_flags & O_NONBLOCK) ||
 594               scsi_block_when_processing_errors(sdp->device)))
 595                 return -ENXIO;
 596
 597         if (!access_ok(VERIFY_READ, buf, count))
 598                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 599         if (count < SZ_SG_HEADER)
 600                 return -EIO;
 601         if (__copy_from_user(&old_hdr, buf, SZ_SG_HEADER))
 602                 return -EFAULT;
 603         blocking = !(filp->f_flags & O_NONBLOCK);
 604         if (old_hdr.reply_len < 0)
 605                 return sg_new_write(sfp, filp, buf, count,
 606                                     blocking, 0, 0, NULL);
 607         if (count < (SZ_SG_HEADER + 6))
 608                 return -EIO;    /* The minimum scsi command length is 6 bytes. */
 609
 610         if (!(srp = sg_add_request(sfp))) {
 611                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sdp,
 612                                               "sg_write: queue full\n"));
 613                 return -EDOM;
 614         }
 615         buf += SZ_SG_HEADER;
 616         __get_user(opcode, buf);
 617         if (sfp->next_cmd_len > 0) {
 618                 cmd_size = sfp->next_cmd_len;
 619                 sfp->next_cmd_len = 0;  /* reset so only this write() effected */
 620         } else {
 621                 cmd_size = COMMAND_SIZE(opcode);        /* based on SCSI command group */
 622                 if ((opcode >= 0xc0) && old_hdr.twelve_byte)
 623                         cmd_size = 12;
 624         }
 625         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
 626                 "sg_write:   scsi opcode=0x%02x, cmd_size=%d\n", (int) opcode, cmd_size));
 627 /* Determine buffer size.  */
 628         input_size = count - cmd_size;
 629         mxsize = (input_size > old_hdr.reply_len) ? input_size : old_hdr.reply_len;
 630         mxsize -= SZ_SG_HEADER;
 631         input_size -= SZ_SG_HEADER;
 632         if (input_size < 0) {
 633                 sg_remove_request(sfp, srp);
 634                 return -EIO;    /* User did not pass enough bytes for this command. */
 635         }
 636         hp = &srp->header;
 637         hp->interface_id = '\0';        /* indicator of old interface tunnelled */
 638         hp->cmd_len = (unsigned char) cmd_size;
 639         hp->iovec_count = 0;
 640         hp->mx_sb_len = 0;
 641         if (input_size > 0)
 642                 hp->dxfer_direction = (old_hdr.reply_len > SZ_SG_HEADER) ?
 643                     SG_DXFER_TO_FROM_DEV : SG_DXFER_TO_DEV;
 644         else
 645                 hp->dxfer_direction = (mxsize > 0) ? SG_DXFER_FROM_DEV : SG_DXFER_NONE;
 646         hp->dxfer_len = mxsize;
 647         if ((hp->dxfer_direction == SG_DXFER_TO_DEV) ||
 648             (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV))
 649                 hp->dxferp = (char __user *)buf + cmd_size;
 650         else
 651                 hp->dxferp = NULL;
 652         hp->sbp = NULL;
 653         hp->timeout = old_hdr.reply_len;        /* structure abuse ... */
 654         hp->flags = input_size; /* structure abuse ... */
 655         hp->pack_id = old_hdr.pack_id;
 656         hp->usr_ptr = NULL;
 657         if (__copy_from_user(cmnd, buf, cmd_size))
 658                 return -EFAULT;
 659         /*
 660          * SG_DXFER_TO_FROM_DEV is functionally equivalent to SG_DXFER_FROM_DEV,
 661          * but is is possible that the app intended SG_DXFER_TO_DEV, because there
 662          * is a non-zero input_size, so emit a warning.
 663          */
 664         if (hp->dxfer_direction == SG_DXFER_TO_FROM_DEV) {
 665                 static char cmd[TASK_COMM_LEN];
 666                 if (strcmp(current->comm, cmd)) {
 667                         printk_ratelimited(KERN_WARNING
 668                                            "sg_write: data in/out %d/%d bytes "
 669                                            "for SCSI command 0x%x-- guessing "
 670                                            "data in;\n   program %s not setting "
 671                                            "count and/or reply_len properly\n",
 672                                            old_hdr.reply_len - (int)SZ_SG_HEADER,
 673                                            input_size, (unsigned int) cmnd[0],
 674                                            current->comm);
 675                         strcpy(cmd, current->comm);
 676                 }
 677         }
 678         k = sg_common_write(sfp, srp, cmnd, sfp->timeout, blocking);
 679         return (k < 0) ? k : count;
 680 }
 681
 682 static ssize_t
 683 sg_new_write(Sg_fd *sfp, struct file *file, const char __user *buf,
 684                  size_t count, int blocking, int read_only, int sg_io_owned,
 685                  Sg_request **o_srp)
 686 {
 687         int k;
 688         Sg_request *srp;
 689         sg_io_hdr_t *hp;
 690         unsigned char cmnd[SG_MAX_CDB_SIZE];
 691         int timeout;
 692         unsigned long ul_timeout;
 693
 694         if (count < SZ_SG_IO_HDR)
 695                 return -EINVAL;
 696         if (!access_ok(VERIFY_READ, buf, count))
 697                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 698
 699         sfp->cmd_q = 1; /* when sg_io_hdr seen, set command queuing on */
 700         if (!(srp = sg_add_request(sfp))) {
 701                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 702                                               "sg_new_write: queue full\n"));
 703                 return -EDOM;
 704         }
 705         srp->sg_io_owned = sg_io_owned;
 706         hp = &srp->header;
 707         if (__copy_from_user(hp, buf, SZ_SG_IO_HDR)) {
 708                 sg_remove_request(sfp, srp);
 709                 return -EFAULT;
 710         }
 711         if (hp->interface_id != 'S') {
 712                 sg_remove_request(sfp, srp);
 713                 return -ENOSYS;
 714         }
 715         if (hp->flags & SG_FLAG_MMAP_IO) {
 716                 if (hp->dxfer_len > sfp->reserve.bufflen) {
 717                         sg_remove_request(sfp, srp);
 718                         return -ENOMEM; /* MMAP_IO size must fit in reserve buffer */
 719                 }
 720                 if (hp->flags & SG_FLAG_DIRECT_IO) {
 721                         sg_remove_request(sfp, srp);
 722                         return -EINVAL; /* either MMAP_IO or DIRECT_IO (not both) */
 723                 }
 724                 if (sg_res_in_use(sfp)) {
 725                         sg_remove_request(sfp, srp);
 726                         return -EBUSY;  /* reserve buffer already being used */
 727                 }
 728         }
 729         ul_timeout = msecs_to_jiffies(srp->header.timeout);
 730         timeout = (ul_timeout < INT_MAX) ? ul_timeout : INT_MAX;
 731         if ((!hp->cmdp) || (hp->cmd_len < 6) || (hp->cmd_len > sizeof (cmnd))) {
 732                 sg_remove_request(sfp, srp);
 733                 return -EMSGSIZE;
 734         }
 735         if (!access_ok(VERIFY_READ, hp->cmdp, hp->cmd_len)) {
 736                 sg_remove_request(sfp, srp);
 737                 return -EFAULT; /* protects following copy_from_user()s + get_user()s */
 738         }
 739         if (__copy_from_user(cmnd, hp->cmdp, hp->cmd_len)) {
 740                 sg_remove_request(sfp, srp);
 741                 return -EFAULT;
 742         }
 743         if (read_only && sg_allow_access(file, cmnd)) {
 744                 sg_remove_request(sfp, srp);
 745                 return -EPERM;
 746         }
 747         k = sg_common_write(sfp, srp, cmnd, timeout, blocking);
 748         if (k < 0)
 749                 return k;
 750         if (o_srp)
 751                 *o_srp = srp;
 752         return count;
 753 }
 754
 755 static int
 756 sg_common_write(Sg_fd * sfp, Sg_request * srp,
 757                 unsigned char *cmnd, int timeout, int blocking)
 758 {
 759         int k, at_head;
 760         Sg_device *sdp = sfp->parentdp;
 761         sg_io_hdr_t *hp = &srp->header;
 762
 763         srp->data.cmd_opcode = cmnd[0]; /* hold opcode of command */
 764         hp->status = 0;
 765         hp->masked_status = 0;
 766         hp->msg_status = 0;
 767         hp->info = 0;
 768         hp->host_status = 0;
 769         hp->driver_status = 0;
 770         hp->resid = 0;
 771         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
 772                         "sg_common_write:  scsi opcode=0x%02x, cmd_size=%d\n",
 773                         (int) cmnd[0], (int) hp->cmd_len));
 774
 775         k = sg_start_req(srp, cmnd);
 776         if (k) {
 777                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
 778                         "sg_common_write: start_req err=%d\n", k));
 779                 sg_finish_rem_req(srp);
 780                 return k;       /* probably out of space --> ENOMEM */
 781         }
 782         if (atomic_read(&sdp->detaching)) {
 783                 if (srp->bio) {
 784                         scsi_req_free_cmd(scsi_req(srp->rq));
 785                         blk_end_request_all(srp->rq, -EIO);
 786                         srp->rq = NULL;
 787                 }
 788
 789                 sg_finish_rem_req(srp);
 790                 return -ENODEV;
 791         }
 792
 793         hp->duration = jiffies_to_msecs(jiffies);
 794         if (hp->interface_id != '\0' && /* v3 (or later) interface */
 795             (SG_FLAG_Q_AT_TAIL & hp->flags))
 796                 at_head = 0;
 797         else
 798                 at_head = 1;
 799
 800         srp->rq->timeout = timeout;
 801         kref_get(&sfp->f_ref); /* sg_rq_end_io() does kref_put(). */
 802         blk_execute_rq_nowait(sdp->device->request_queue, sdp->disk,
 803                               srp->rq, at_head, sg_rq_end_io);
 804         return 0;
 805 }
 806
 807 static int srp_done(Sg_fd *sfp, Sg_request *srp)
 808 {
 809         unsigned long flags;
 810         int ret;
 811
 812         read_lock_irqsave(&sfp->rq_list_lock, flags);
 813         ret = srp->done;
 814         read_unlock_irqrestore(&sfp->rq_list_lock, flags);
 815         return ret;
 816 }
 817
 818 static int max_sectors_bytes(struct request_queue *q)
 819 {
 820         unsigned int max_sectors = queue_max_sectors(q);
 821
 822         max_sectors = min_t(unsigned int, max_sectors, INT_MAX >> 9);
 823
 824         return max_sectors << 9;
 825 }
 826
 827 static long
 828 sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
 829 {
 830         void __user *p = (void __user *)arg;
 831         int __user *ip = p;
 832         int result, val, read_only;
 833         Sg_device *sdp;
 834         Sg_fd *sfp;
 835         Sg_request *srp;
 836         unsigned long iflags;
 837
 838         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
 839                 return -ENXIO;
 840
 841         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
 842                                    "sg_ioctl: cmd=0x%x\n", (int) cmd_in));
 843         read_only = (O_RDWR != (filp->f_flags & O_ACCMODE));
 844
 845         switch (cmd_in) {
 846         case SG_IO:
 847                 if (atomic_read(&sdp->detaching))
 848                         return -ENODEV;
 849                 if (!scsi_block_when_processing_errors(sdp->device))
 850                         return -ENXIO;
 851                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_IO_HDR))
 852                         return -EFAULT;
 853                 result = sg_new_write(sfp, filp, p, SZ_SG_IO_HDR,
 854                                  1, read_only, 1, &srp);
 855                 if (result < 0)
 856                         return result;
 857                 result = wait_event_interruptible(sfp->read_wait,
 858                         (srp_done(sfp, srp) || atomic_read(&sdp->detaching)));
 859                 if (atomic_read(&sdp->detaching))
 860                         return -ENODEV;
 861                 write_lock_irq(&sfp->rq_list_lock);
 862                 if (srp->done) {
 863                         srp->done = 2;
 864                         write_unlock_irq(&sfp->rq_list_lock);
 865                         result = sg_new_read(sfp, p, SZ_SG_IO_HDR, srp);
 866                         return (result < 0) ? result : 0;
 867                 }
 868                 srp->orphan = 1;
 869                 write_unlock_irq(&sfp->rq_list_lock);
 870                 return result;  /* -ERESTARTSYS because signal hit process */
 871         case SG_SET_TIMEOUT:
 872                 result = get_user(val, ip);
 873                 if (result)
 874                         return result;
 875                 if (val < 0)
 876                         return -EIO;
 877                 if (val >= mult_frac((s64)INT_MAX, USER_HZ, HZ))
 878                         val = min_t(s64, mult_frac((s64)INT_MAX, USER_HZ, HZ),
 879                                     INT_MAX);
 880                 sfp->timeout_user = val;
 881                 sfp->timeout = mult_frac(val, HZ, USER_HZ);
 882
 883                 return 0;
 884         case SG_GET_TIMEOUT:    /* N.B. User receives timeout as return value */
 885                                 /* strange ..., for backward compatibility */
 886                 return sfp->timeout_user;
 887         case SG_SET_FORCE_LOW_DMA:
 888                 result = get_user(val, ip);
 889                 if (result)
 890                         return result;
 891                 if (val) {
 892                         sfp->low_dma = 1;
 893                         if ((0 == sfp->low_dma) && (0 == sg_res_in_use(sfp))) {
 894                                 val = (int) sfp->reserve.bufflen;
 895                                 sg_remove_scat(sfp, &sfp->reserve);
 896                                 sg_build_reserve(sfp, val);
 897                         }
 898                 } else {
 899                         if (atomic_read(&sdp->detaching))
 900                                 return -ENODEV;
 901                         sfp->low_dma = sdp->device->host->unchecked_isa_dma;
 902                 }
 903                 return 0;
 904         case SG_GET_LOW_DMA:
 905                 return put_user((int) sfp->low_dma, ip);
 906         case SG_GET_SCSI_ID:
 907                 if (!access_ok(VERIFY_WRITE, p, sizeof (sg_scsi_id_t)))
 908                         return -EFAULT;
 909                 else {
 910                         sg_scsi_id_t __user *sg_idp = p;
 911
 912                         if (atomic_read(&sdp->detaching))
 913                                 return -ENODEV;
 914                         __put_user((int) sdp->device->host->host_no,
 915                                    &sg_idp->host_no);
 916                         __put_user((int) sdp->device->channel,
 917                                    &sg_idp->channel);
 918                         __put_user((int) sdp->device->id, &sg_idp->scsi_id);
 919                         __put_user((int) sdp->device->lun, &sg_idp->lun);
 920                         __put_user((int) sdp->device->type, &sg_idp->scsi_type);
 921                         __put_user((short) sdp->device->host->cmd_per_lun,
 922                                    &sg_idp->h_cmd_per_lun);
 923                         __put_user((short) sdp->device->queue_depth,
 924                                    &sg_idp->d_queue_depth);
 925                         __put_user(0, &sg_idp->unused[0]);
 926                         __put_user(0, &sg_idp->unused[1]);
 927                         return 0;
 928                 }
 929         case SG_SET_FORCE_PACK_ID:
 930                 result = get_user(val, ip);
 931                 if (result)
 932                         return result;
 933                 sfp->force_packid = val ? 1 : 0;
 934                 return 0;
 935         case SG_GET_PACK_ID:
 936                 if (!access_ok(VERIFY_WRITE, ip, sizeof (int)))
 937                         return -EFAULT;
 938                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
 939                 for (srp = sfp->headrp; srp; srp = srp->nextrp) {
 940                         if ((1 == srp->done) && (!srp->sg_io_owned)) {
 941                                 read_unlock_irqrestore(&sfp->rq_list_lock,
 942                                                        iflags);
 943                                 __put_user(srp->header.pack_id, ip);
 944                                 return 0;
 945                         }
 946                 }
 947                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
 948                 __put_user(-1, ip);
 949                 return 0;
 950         case SG_GET_NUM_WAITING:
 951                 read_lock_irqsave(&sfp->rq_list_lock, iflags);
 952                 for (val = 0, srp = sfp->headrp; srp; srp = srp->nextrp) {
 953                         if ((1 == srp->done) && (!srp->sg_io_owned))
 954                                 ++val;
 955                 }
 956                 read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
 957                 return put_user(val, ip);
 958         case SG_GET_SG_TABLESIZE:
 959                 return put_user(sdp->sg_tablesize, ip);
 960         case SG_SET_RESERVED_SIZE:
 961                 result = get_user(val, ip);
 962                 if (result)
 963                         return result;
 964                 if (val < 0)
 965                         return -EINVAL;
 966                 val = min_t(int, val,
 967                             max_sectors_bytes(sdp->device->request_queue));
 968                 if (val != sfp->reserve.bufflen) {
 969                         if (sg_res_in_use(sfp) || sfp->mmap_called)
 970                                 return -EBUSY;
 971                         sg_remove_scat(sfp, &sfp->reserve);
 972                         sg_build_reserve(sfp, val);
 973                 }
 974                 return 0;
 975         case SG_GET_RESERVED_SIZE:
 976                 val = min_t(int, sfp->reserve.bufflen,
 977                             max_sectors_bytes(sdp->device->request_queue));
 978                 return put_user(val, ip);
 979         case SG_SET_COMMAND_Q:
 980                 result = get_user(val, ip);
 981                 if (result)
 982                         return result;
 983                 sfp->cmd_q = val ? 1 : 0;
 984                 return 0;
 985         case SG_GET_COMMAND_Q:
 986                 return put_user((int) sfp->cmd_q, ip);
 987         case SG_SET_KEEP_ORPHAN:
 988                 result = get_user(val, ip);
 989                 if (result)
 990                         return result;
 991                 sfp->keep_orphan = val;
 992                 return 0;
 993         case SG_GET_KEEP_ORPHAN:
 994                 return put_user((int) sfp->keep_orphan, ip);
 995         case SG_NEXT_CMD_LEN:
 996                 result = get_user(val, ip);
 997                 if (result)
 998                         return result;
 999                 if (val > SG_MAX_CDB_SIZE)
1000                         return -ENOMEM;
1001                 sfp->next_cmd_len = (val > 0) ? val : 0;
1002                 return 0;
1003         case SG_GET_VERSION_NUM:
1004                 return put_user(sg_version_num, ip);
1005         case SG_GET_ACCESS_COUNT:
1006                 /* faked - we don't have a real access count anymore */
1007                 val = (sdp->device ? 1 : 0);
1008                 return put_user(val, ip);
1009         case SG_GET_REQUEST_TABLE:
1010                 if (!access_ok(VERIFY_WRITE, p, SZ_SG_REQ_INFO * SG_MAX_QUEUE))
1011                         return -EFAULT;
1012                 else {
1013                         sg_req_info_t *rinfo;
1014                         unsigned int ms;
1015
1016                         rinfo = kmalloc(SZ_SG_REQ_INFO * SG_MAX_QUEUE,
1017                                                                 GFP_KERNEL);
1018                         if (!rinfo)
1019                                 return -ENOMEM;
1020                         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1021                         for (srp = sfp->headrp, val = 0; val < SG_MAX_QUEUE;
1022                              ++val, srp = srp ? srp->nextrp : srp) {
1023                                 memset(&rinfo[val], 0, SZ_SG_REQ_INFO);
1024                                 if (srp) {
1025                                         rinfo[val].req_state = srp->done + 1;
1026                                         rinfo[val].problem =
1027                                             srp->header.masked_status &
1028                                             srp->header.host_status &
1029                                             srp->header.driver_status;
1030                                         if (srp->done)
1031                                                 rinfo[val].duration =
1032                                                         srp->header.duration;
1033                                         else {
1034                                                 ms = jiffies_to_msecs(jiffies);
1035                                                 rinfo[val].duration =
1036                                                     (ms > srp->header.duration) ?
1037                                                     (ms - srp->header.duration) : 0;
1038                                         }
1039                                         rinfo[val].orphan = srp->orphan;
1040                                         rinfo[val].sg_io_owned =
1041                                                         srp->sg_io_owned;
1042                                         rinfo[val].pack_id =
1043                                                         srp->header.pack_id;
1044                                         rinfo[val].usr_ptr =
1045                                                         srp->header.usr_ptr;
1046                                 }
1047                         }
1048                         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1049                         result = __copy_to_user(p, rinfo,
1050                                                 SZ_SG_REQ_INFO * SG_MAX_QUEUE);
1051                         result = result ? -EFAULT : 0;
1052                         kfree(rinfo);
1053                         return result;
1054                 }
1055         case SG_EMULATED_HOST:
1056                 if (atomic_read(&sdp->detaching))
1057                         return -ENODEV;
1058                 return put_user(sdp->device->host->hostt->emulated, ip);
1059         case SCSI_IOCTL_SEND_COMMAND:
1060                 if (atomic_read(&sdp->detaching))
1061                         return -ENODEV;
1062                 if (read_only) {
1063                         unsigned char opcode = WRITE_6;
1064                         Scsi_Ioctl_Command __user *siocp = p;
1065
1066                         if (copy_from_user(&opcode, siocp->data, 1))
1067                                 return -EFAULT;
1068                         if (sg_allow_access(filp, &opcode))
1069                                 return -EPERM;
1070                 }
1071                 return sg_scsi_ioctl(sdp->device->request_queue, NULL, filp->f_mode, p);
1072         case SG_SET_DEBUG:
1073                 result = get_user(val, ip);
1074                 if (result)
1075                         return result;
1076                 sdp->sgdebug = (char) val;
1077                 return 0;
1078         case BLKSECTGET:
1079                 return put_user(max_sectors_bytes(sdp->device->request_queue),
1080                                 ip);
1081         case BLKTRACESETUP:
1082                 return blk_trace_setup(sdp->device->request_queue,
1083                                        sdp->disk->disk_name,
1084                                        MKDEV(SCSI_GENERIC_MAJOR, sdp->index),
1085                                        NULL,
1086                                        (char *)arg);
1087         case BLKTRACESTART:
1088                 return blk_trace_startstop(sdp->device->request_queue, 1);
1089         case BLKTRACESTOP:
1090                 return blk_trace_startstop(sdp->device->request_queue, 0);
1091         case BLKTRACETEARDOWN:
1092                 return blk_trace_remove(sdp->device->request_queue);
1093         case SCSI_IOCTL_GET_IDLUN:
1094         case SCSI_IOCTL_GET_BUS_NUMBER:
1095         case SCSI_IOCTL_PROBE_HOST:
1096         case SG_GET_TRANSFORM:
1097         case SG_SCSI_RESET:
1098                 if (atomic_read(&sdp->detaching))
1099                         return -ENODEV;
1100                 break;
1101         default:
1102                 if (read_only)
1103                         return -EPERM;  /* don't know so take safe approach */
1104                 break;
1105         }
1106
1107         result = scsi_ioctl_block_when_processing_errors(sdp->device,
1108                         cmd_in, filp->f_flags & O_NDELAY);
1109         if (result)
1110                 return result;
1111         return scsi_ioctl(sdp->device, cmd_in, p);
1112 }
1113
1114 #ifdef CONFIG_COMPAT
1115 static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg)
1116 {
1117         Sg_device *sdp;
1118         Sg_fd *sfp;
1119         struct scsi_device *sdev;
1120
1121         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1122                 return -ENXIO;
1123
1124         sdev = sdp->device;
1125         if (sdev->host->hostt->compat_ioctl) {
1126                 int ret;
1127
1128                 ret = sdev->host->hostt->compat_ioctl(sdev, cmd_in, (void __user *)arg);
1129
1130                 return ret;
1131         }
1132
1133         return -ENOIOCTLCMD;
1134 }
1135 #endif
1136
1137 static unsigned int
1138 sg_poll(struct file *filp, poll_table * wait)
1139 {
1140         unsigned int res = 0;
1141         Sg_device *sdp;
1142         Sg_fd *sfp;
1143         Sg_request *srp;
1144         int count = 0;
1145         unsigned long iflags;
1146
1147         sfp = filp->private_data;
1148         if (!sfp)
1149                 return POLLERR;
1150         sdp = sfp->parentdp;
1151         if (!sdp)
1152                 return POLLERR;
1153         poll_wait(filp, &sfp->read_wait, wait);
1154         read_lock_irqsave(&sfp->rq_list_lock, iflags);
1155         for (srp = sfp->headrp; srp; srp = srp->nextrp) {
1156                 /* if any read waiting, flag it */
1157                 if ((0 == res) && (1 == srp->done) && (!srp->sg_io_owned))
1158                         res = POLLIN | POLLRDNORM;
1159                 ++count;
1160         }
1161         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1162
1163         if (atomic_read(&sdp->detaching))
1164                 res |= POLLHUP;
1165         else if (!sfp->cmd_q) {
1166                 if (0 == count)
1167                         res |= POLLOUT | POLLWRNORM;
1168         } else if (count < SG_MAX_QUEUE)
1169                 res |= POLLOUT | POLLWRNORM;
1170         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1171                                       "sg_poll: res=0x%x\n", (int) res));
1172         return res;
1173 }
1174
1175 static int
1176 sg_fasync(int fd, struct file *filp, int mode)
1177 {
1178         Sg_device *sdp;
1179         Sg_fd *sfp;
1180
1181         if ((!(sfp = (Sg_fd *) filp->private_data)) || (!(sdp = sfp->parentdp)))
1182                 return -ENXIO;
1183         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1184                                       "sg_fasync: mode=%d\n", mode));
1185
1186         return fasync_helper(fd, filp, mode, &sfp->async_qp);
1187 }
1188
1189 static int
1190 sg_vma_fault(struct vm_fault *vmf)
1191 {
1192         struct vm_area_struct *vma = vmf->vma;
1193         Sg_fd *sfp;
1194         unsigned long offset, len, sa;
1195         Sg_scatter_hold *rsv_schp;
1196         int k, length;
1197
1198         if ((NULL == vma) || (!(sfp = (Sg_fd *) vma->vm_private_data)))
1199                 return VM_FAULT_SIGBUS;
1200         rsv_schp = &sfp->reserve;
1201         offset = vmf->pgoff << PAGE_SHIFT;
1202         if (offset >= rsv_schp->bufflen)
1203                 return VM_FAULT_SIGBUS;
1204         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1205                                       "sg_vma_fault: offset=%lu, scatg=%d\n",
1206                                       offset, rsv_schp->k_use_sg));
1207         sa = vma->vm_start;
1208         length = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1209         for (k = 0; k < rsv_schp->k_use_sg && sa < vma->vm_end; k++) {
1210                 len = vma->vm_end - sa;
1211                 len = (len < length) ? len : length;
1212                 if (offset < len) {
1213                         struct page *page = nth_page(rsv_schp->pages[k],
1214                                                      offset >> PAGE_SHIFT);
1215                         get_page(page); /* increment page count */
1216                         vmf->page = page;
1217                         return 0; /* success */
1218                 }
1219                 sa += len;
1220                 offset -= len;
1221         }
1222
1223         return VM_FAULT_SIGBUS;
1224 }
1225
1226 static const struct vm_operations_struct sg_mmap_vm_ops = {
1227         .fault = sg_vma_fault,
1228 };
1229
1230 static int
1231 sg_mmap(struct file *filp, struct vm_area_struct *vma)
1232 {
1233         Sg_fd *sfp;
1234         unsigned long req_sz, len, sa;
1235         Sg_scatter_hold *rsv_schp;
1236         int k, length;
1237
1238         if ((!filp) || (!vma) || (!(sfp = (Sg_fd *) filp->private_data)))
1239                 return -ENXIO;
1240         req_sz = vma->vm_end - vma->vm_start;
1241         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sfp->parentdp,
1242                                       "sg_mmap starting, vm_start=%p, len=%d\n",
1243                                       (void *) vma->vm_start, (int) req_sz));
1244         if (vma->vm_pgoff)
1245                 return -EINVAL; /* want no offset */
1246         rsv_schp = &sfp->reserve;
1247         if (req_sz > rsv_schp->bufflen)
1248                 return -ENOMEM; /* cannot map more than reserved buffer */
1249
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                 sa += len;
1256         }
1257
1258         sfp->mmap_called = 1;
1259         vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP;
1260         vma->vm_private_data = sfp;
1261         vma->vm_ops = &sg_mmap_vm_ops;
1262         return 0;
1263 }
1264
1265 static void
1266 sg_rq_end_io_usercontext(struct work_struct *work)
1267 {
1268         struct sg_request *srp = container_of(work, struct sg_request, ew.work);
1269         struct sg_fd *sfp = srp->parentfp;
1270
1271         sg_finish_rem_req(srp);
1272         kref_put(&sfp->f_ref, sg_remove_sfp);
1273 }
1274
1275 /*
1276  * This function is a "bottom half" handler that is called by the mid
1277  * level when a command is completed (or has failed).
1278  */
1279 static void
1280 sg_rq_end_io(struct request *rq, int uptodate)
1281 {
1282         struct sg_request *srp = rq->end_io_data;
1283         struct scsi_request *req = scsi_req(rq);
1284         Sg_device *sdp;
1285         Sg_fd *sfp;
1286         unsigned long iflags;
1287         unsigned int ms;
1288         char *sense;
1289         int result, resid, done = 1;
1290
1291         if (WARN_ON(srp->done != 0))
1292                 return;
1293
1294         sfp = srp->parentfp;
1295         if (WARN_ON(sfp == NULL))
1296                 return;
1297
1298         sdp = sfp->parentdp;
1299         if (unlikely(atomic_read(&sdp->detaching)))
1300                 pr_info("%s: device detaching\n", __func__);
1301
1302         sense = req->sense;
1303         result = rq->errors;
1304         resid = req->resid_len;
1305
1306         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sdp,
1307                                       "sg_cmd_done: pack_id=%d, res=0x%x\n",
1308                                       srp->header.pack_id, result));
1309         srp->header.resid = resid;
1310         ms = jiffies_to_msecs(jiffies);
1311         srp->header.duration = (ms > srp->header.duration) ?
1312                                 (ms - srp->header.duration) : 0;
1313         if (0 != result) {
1314                 struct scsi_sense_hdr sshdr;
1315
1316                 srp->header.status = 0xff & result;
1317                 srp->header.masked_status = status_byte(result);
1318                 srp->header.msg_status = msg_byte(result);
1319                 srp->header.host_status = host_byte(result);
1320                 srp->header.driver_status = driver_byte(result);
1321                 if ((sdp->sgdebug > 0) &&
1322                     ((CHECK_CONDITION == srp->header.masked_status) ||
1323                      (COMMAND_TERMINATED == srp->header.masked_status)))
1324                         __scsi_print_sense(sdp->device, __func__, sense,
1325                                            SCSI_SENSE_BUFFERSIZE);
1326
1327                 /* Following if statement is a patch supplied by Eric Youngdale */
1328                 if (driver_byte(result) != 0
1329                     && scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, &sshdr)
1330                     && !scsi_sense_is_deferred(&sshdr)
1331                     && sshdr.sense_key == UNIT_ATTENTION
1332                     && sdp->device->removable) {
1333                         /* Detected possible disc change. Set the bit - this */
1334                         /* may be used if there are filesystems using this device */
1335                         sdp->device->changed = 1;
1336                 }
1337         }
1338
1339         if (req->sense_len)
1340                 memcpy(srp->sense_b, req->sense, SCSI_SENSE_BUFFERSIZE);
1341
1342         /* Rely on write phase to clean out srp status values, so no "else" */
1343
1344         /*
1345          * Free the request as soon as it is complete so that its resources
1346          * can be reused without waiting for userspace to read() the
1347          * result.  But keep the associated bio (if any) around until
1348          * blk_rq_unmap_user() can be called from user context.
1349          */
1350         srp->rq = NULL;
1351         scsi_req_free_cmd(scsi_req(rq));
1352         __blk_put_request(rq->q, rq);
1353
1354         write_lock_irqsave(&sfp->rq_list_lock, iflags);
1355         if (unlikely(srp->orphan)) {
1356                 if (sfp->keep_orphan)
1357                         srp->sg_io_owned = 0;
1358                 else
1359                         done = 0;
1360         }
1361         srp->done = done;
1362         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
1363
1364         if (likely(done)) {
1365                 /* Now wake up any sg_read() that is waiting for this
1366                  * packet.
1367                  */
1368                 wake_up_interruptible(&sfp->read_wait);
1369                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_IN);
1370                 kref_put(&sfp->f_ref, sg_remove_sfp);
1371         } else {
1372                 INIT_WORK(&srp->ew.work, sg_rq_end_io_usercontext);
1373                 schedule_work(&srp->ew.work);
1374         }
1375 }
1376
1377 static const struct file_operations sg_fops = {
1378         .owner = THIS_MODULE,
1379         .read = sg_read,
1380         .write = sg_write,
1381         .poll = sg_poll,
1382         .unlocked_ioctl = sg_ioctl,
1383 #ifdef CONFIG_COMPAT
1384         .compat_ioctl = sg_compat_ioctl,
1385 #endif
1386         .open = sg_open,
1387         .mmap = sg_mmap,
1388         .release = sg_release,
1389         .fasync = sg_fasync,
1390         .llseek = no_llseek,
1391 };
1392
1393 static struct class *sg_sysfs_class;
1394
1395 static int sg_sysfs_valid = 0;
1396
1397 static Sg_device *
1398 sg_alloc(struct gendisk *disk, struct scsi_device *scsidp)
1399 {
1400         struct request_queue *q = scsidp->request_queue;
1401         Sg_device *sdp;
1402         unsigned long iflags;
1403         int error;
1404         u32 k;
1405
1406         sdp = kzalloc(sizeof(Sg_device), GFP_KERNEL);
1407         if (!sdp) {
1408                 sdev_printk(KERN_WARNING, scsidp, "%s: kmalloc Sg_device "
1409                             "failure\n", __func__);
1410                 return ERR_PTR(-ENOMEM);
1411         }
1412
1413         idr_preload(GFP_KERNEL);
1414         write_lock_irqsave(&sg_index_lock, iflags);
1415
1416         error = idr_alloc(&sg_index_idr, sdp, 0, SG_MAX_DEVS, GFP_NOWAIT);
1417         if (error < 0) {
1418                 if (error == -ENOSPC) {
1419                         sdev_printk(KERN_WARNING, scsidp,
1420                                     "Unable to attach sg device type=%d, minor number exceeds %d\n",
1421                                     scsidp->type, SG_MAX_DEVS - 1);
1422                         error = -ENODEV;
1423                 } else {
1424                         sdev_printk(KERN_WARNING, scsidp, "%s: idr "
1425                                     "allocation Sg_device failure: %d\n",
1426                                     __func__, error);
1427                 }
1428                 goto out_unlock;
1429         }
1430         k = error;
1431
1432         SCSI_LOG_TIMEOUT(3, sdev_printk(KERN_INFO, scsidp,
1433                                         "sg_alloc: dev=%d \n", k));
1434         sprintf(disk->disk_name, "sg%d", k);
1435         disk->first_minor = k;
1436         sdp->disk = disk;
1437         sdp->device = scsidp;
1438         mutex_init(&sdp->open_rel_lock);
1439         INIT_LIST_HEAD(&sdp->sfds);
1440         init_waitqueue_head(&sdp->open_wait);
1441         atomic_set(&sdp->detaching, 0);
1442         rwlock_init(&sdp->sfd_lock);
1443         sdp->sg_tablesize = queue_max_segments(q);
1444         sdp->index = k;
1445         kref_init(&sdp->d_ref);
1446         error = 0;
1447
1448 out_unlock:
1449         write_unlock_irqrestore(&sg_index_lock, iflags);
1450         idr_preload_end();
1451
1452         if (error) {
1453                 kfree(sdp);
1454                 return ERR_PTR(error);
1455         }
1456         return sdp;
1457 }
1458
1459 static int
1460 sg_add_device(struct device *cl_dev, struct class_interface *cl_intf)
1461 {
1462         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1463         struct gendisk *disk;
1464         Sg_device *sdp = NULL;
1465         struct cdev * cdev = NULL;
1466         int error;
1467         unsigned long iflags;
1468
1469         disk = alloc_disk(1);
1470         if (!disk) {
1471                 pr_warn("%s: alloc_disk failed\n", __func__);
1472                 return -ENOMEM;
1473         }
1474         disk->major = SCSI_GENERIC_MAJOR;
1475
1476         error = -ENOMEM;
1477         cdev = cdev_alloc();
1478         if (!cdev) {
1479                 pr_warn("%s: cdev_alloc failed\n", __func__);
1480                 goto out;
1481         }
1482         cdev->owner = THIS_MODULE;
1483         cdev->ops = &sg_fops;
1484
1485         sdp = sg_alloc(disk, scsidp);
1486         if (IS_ERR(sdp)) {
1487                 pr_warn("%s: sg_alloc failed\n", __func__);
1488                 error = PTR_ERR(sdp);
1489                 goto out;
1490         }
1491
1492         error = cdev_add(cdev, MKDEV(SCSI_GENERIC_MAJOR, sdp->index), 1);
1493         if (error)
1494                 goto cdev_add_err;
1495
1496         sdp->cdev = cdev;
1497         if (sg_sysfs_valid) {
1498                 struct device *sg_class_member;
1499
1500                 sg_class_member = device_create(sg_sysfs_class, cl_dev->parent,
1501                                                 MKDEV(SCSI_GENERIC_MAJOR,
1502                                                       sdp->index),
1503                                                 sdp, "%s", disk->disk_name);
1504                 if (IS_ERR(sg_class_member)) {
1505                         pr_err("%s: device_create failed\n", __func__);
1506                         error = PTR_ERR(sg_class_member);
1507                         goto cdev_add_err;
1508                 }
1509                 error = sysfs_create_link(&scsidp->sdev_gendev.kobj,
1510                                           &sg_class_member->kobj, "generic");
1511                 if (error)
1512                         pr_err("%s: unable to make symlink 'generic' back "
1513                                "to sg%d\n", __func__, sdp->index);
1514         } else
1515                 pr_warn("%s: sg_sys Invalid\n", __func__);
1516
1517         sdev_printk(KERN_NOTICE, scsidp, "Attached scsi generic sg%d "
1518                     "type %d\n", sdp->index, scsidp->type);
1519
1520         dev_set_drvdata(cl_dev, sdp);
1521
1522         return 0;
1523
1524 cdev_add_err:
1525         write_lock_irqsave(&sg_index_lock, iflags);
1526         idr_remove(&sg_index_idr, sdp->index);
1527         write_unlock_irqrestore(&sg_index_lock, iflags);
1528         kfree(sdp);
1529
1530 out:
1531         put_disk(disk);
1532         if (cdev)
1533                 cdev_del(cdev);
1534         return error;
1535 }
1536
1537 static void
1538 sg_device_destroy(struct kref *kref)
1539 {
1540         struct sg_device *sdp = container_of(kref, struct sg_device, d_ref);
1541         unsigned long flags;
1542
1543         /* CAUTION!  Note that the device can still be found via idr_find()
1544          * even though the refcount is 0.  Therefore, do idr_remove() BEFORE
1545          * any other cleanup.
1546          */
1547
1548         write_lock_irqsave(&sg_index_lock, flags);
1549         idr_remove(&sg_index_idr, sdp->index);
1550         write_unlock_irqrestore(&sg_index_lock, flags);
1551
1552         SCSI_LOG_TIMEOUT(3,
1553                 sg_printk(KERN_INFO, sdp, "sg_device_destroy\n"));
1554
1555         put_disk(sdp->disk);
1556         kfree(sdp);
1557 }
1558
1559 static void
1560 sg_remove_device(struct device *cl_dev, struct class_interface *cl_intf)
1561 {
1562         struct scsi_device *scsidp = to_scsi_device(cl_dev->parent);
1563         Sg_device *sdp = dev_get_drvdata(cl_dev);
1564         unsigned long iflags;
1565         Sg_fd *sfp;
1566         int val;
1567
1568         if (!sdp)
1569                 return;
1570         /* want sdp->detaching non-zero as soon as possible */
1571         val = atomic_inc_return(&sdp->detaching);
1572         if (val > 1)
1573                 return; /* only want to do following once per device */
1574
1575         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
1576                                       "%s\n", __func__));
1577
1578         read_lock_irqsave(&sdp->sfd_lock, iflags);
1579         list_for_each_entry(sfp, &sdp->sfds, sfd_siblings) {
1580                 wake_up_interruptible_all(&sfp->read_wait);
1581                 kill_fasync(&sfp->async_qp, SIGPOLL, POLL_HUP);
1582         }
1583         wake_up_interruptible_all(&sdp->open_wait);
1584         read_unlock_irqrestore(&sdp->sfd_lock, iflags);
1585
1586         sysfs_remove_link(&scsidp->sdev_gendev.kobj, "generic");
1587         device_destroy(sg_sysfs_class, MKDEV(SCSI_GENERIC_MAJOR, sdp->index));
1588         cdev_del(sdp->cdev);
1589         sdp->cdev = NULL;
1590
1591         kref_put(&sdp->d_ref, sg_device_destroy);
1592 }
1593
1594 module_param_named(scatter_elem_sz, scatter_elem_sz, int, S_IRUGO | S_IWUSR);
1595 module_param_named(def_reserved_size, def_reserved_size, int,
1596                    S_IRUGO | S_IWUSR);
1597 module_param_named(allow_dio, sg_allow_dio, int, S_IRUGO | S_IWUSR);
1598
1599 MODULE_AUTHOR("Douglas Gilbert");
1600 MODULE_DESCRIPTION("SCSI generic (sg) driver");
1601 MODULE_LICENSE("GPL");
1602 MODULE_VERSION(SG_VERSION_STR);
1603 MODULE_ALIAS_CHARDEV_MAJOR(SCSI_GENERIC_MAJOR);
1604
1605 MODULE_PARM_DESC(scatter_elem_sz, "scatter gather element "
1606                 "size (default: max(SG_SCATTER_SZ, PAGE_SIZE))");
1607 MODULE_PARM_DESC(def_reserved_size, "size of buffer reserved for each fd");
1608 MODULE_PARM_DESC(allow_dio, "allow direct I/O (default: 0 (disallow))");
1609
1610 static int __init
1611 init_sg(void)
1612 {
1613         int rc;
1614
1615         if (scatter_elem_sz < PAGE_SIZE) {
1616                 scatter_elem_sz = PAGE_SIZE;
1617                 scatter_elem_sz_prev = scatter_elem_sz;
1618         }
1619         if (def_reserved_size >= 0)
1620                 sg_big_buff = def_reserved_size;
1621         else
1622                 def_reserved_size = sg_big_buff;
1623
1624         rc = register_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1625                                     SG_MAX_DEVS, "sg");
1626         if (rc)
1627                 return rc;
1628         sg_sysfs_class = class_create(THIS_MODULE, "scsi_generic");
1629         if ( IS_ERR(sg_sysfs_class) ) {
1630                 rc = PTR_ERR(sg_sysfs_class);
1631                 goto err_out;
1632         }
1633         sg_sysfs_valid = 1;
1634         rc = scsi_register_interface(&sg_interface);
1635         if (0 == rc) {
1636 #ifdef CONFIG_SCSI_PROC_FS
1637                 sg_proc_init();
1638 #endif                          /* CONFIG_SCSI_PROC_FS */
1639                 return 0;
1640         }
1641         class_destroy(sg_sysfs_class);
1642 err_out:
1643         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0), SG_MAX_DEVS);
1644         return rc;
1645 }
1646
1647 static void __exit
1648 exit_sg(void)
1649 {
1650 #ifdef CONFIG_SCSI_PROC_FS
1651         sg_proc_cleanup();
1652 #endif                          /* CONFIG_SCSI_PROC_FS */
1653         scsi_unregister_interface(&sg_interface);
1654         class_destroy(sg_sysfs_class);
1655         sg_sysfs_valid = 0;
1656         unregister_chrdev_region(MKDEV(SCSI_GENERIC_MAJOR, 0),
1657                                  SG_MAX_DEVS);
1658         idr_destroy(&sg_index_idr);
1659 }
1660
1661 static int
1662 sg_start_req(Sg_request *srp, unsigned char *cmd)
1663 {
1664         int res;
1665         struct request *rq;
1666         struct scsi_request *req;
1667         Sg_fd *sfp = srp->parentfp;
1668         sg_io_hdr_t *hp = &srp->header;
1669         int dxfer_len = (int) hp->dxfer_len;
1670         int dxfer_dir = hp->dxfer_direction;
1671         unsigned int iov_count = hp->iovec_count;
1672         Sg_scatter_hold *req_schp = &srp->data;
1673         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1674         struct request_queue *q = sfp->parentdp->device->request_queue;
1675         struct rq_map_data *md, map_data;
1676         int rw = hp->dxfer_direction == SG_DXFER_TO_DEV ? WRITE : READ;
1677         unsigned char *long_cmdp = NULL;
1678
1679         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1680                                       "sg_start_req: dxfer_len=%d\n",
1681                                       dxfer_len));
1682
1683         if (hp->cmd_len > BLK_MAX_CDB) {
1684                 long_cmdp = kzalloc(hp->cmd_len, GFP_KERNEL);
1685                 if (!long_cmdp)
1686                         return -ENOMEM;
1687         }
1688
1689         /*
1690          * NOTE
1691          *
1692          * With scsi-mq enabled, there are a fixed number of preallocated
1693          * requests equal in number to shost->can_queue.  If all of the
1694          * preallocated requests are already in use, then using GFP_ATOMIC with
1695          * blk_get_request() will return -EWOULDBLOCK, whereas using GFP_KERNEL
1696          * will cause blk_get_request() to sleep until an active command
1697          * completes, freeing up a request.  Neither option is ideal, but
1698          * GFP_KERNEL is the better choice to prevent userspace from getting an
1699          * unexpected EWOULDBLOCK.
1700          *
1701          * With scsi-mq disabled, blk_get_request() with GFP_KERNEL usually
1702          * does not sleep except under memory pressure.
1703          */
1704         rq = blk_get_request(q, hp->dxfer_direction == SG_DXFER_TO_DEV ?
1705                         REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, GFP_KERNEL);
1706         if (IS_ERR(rq)) {
1707                 kfree(long_cmdp);
1708                 return PTR_ERR(rq);
1709         }
1710         req = scsi_req(rq);
1711
1712         scsi_req_init(rq);
1713
1714         if (hp->cmd_len > BLK_MAX_CDB)
1715                 req->cmd = long_cmdp;
1716         memcpy(req->cmd, cmd, hp->cmd_len);
1717         req->cmd_len = hp->cmd_len;
1718
1719         srp->rq = rq;
1720         rq->end_io_data = srp;
1721         rq->retries = SG_DEFAULT_RETRIES;
1722
1723         if ((dxfer_len <= 0) || (dxfer_dir == SG_DXFER_NONE))
1724                 return 0;
1725
1726         if (sg_allow_dio && hp->flags & SG_FLAG_DIRECT_IO &&
1727             dxfer_dir != SG_DXFER_UNKNOWN && !iov_count &&
1728             !sfp->parentdp->device->host->unchecked_isa_dma &&
1729             blk_rq_aligned(q, (unsigned long)hp->dxferp, dxfer_len))
1730                 md = NULL;
1731         else
1732                 md = &map_data;
1733
1734         if (md) {
1735                 if (!sg_res_in_use(sfp) && dxfer_len <= rsv_schp->bufflen)
1736                         sg_link_reserve(sfp, srp, dxfer_len);
1737                 else {
1738                         res = sg_build_indirect(req_schp, sfp, dxfer_len);
1739                         if (res)
1740                                 return res;
1741                 }
1742
1743                 md->pages = req_schp->pages;
1744                 md->page_order = req_schp->page_order;
1745                 md->nr_entries = req_schp->k_use_sg;
1746                 md->offset = 0;
1747                 md->null_mapped = hp->dxferp ? 0 : 1;
1748                 if (dxfer_dir == SG_DXFER_TO_FROM_DEV)
1749                         md->from_user = 1;
1750                 else
1751                         md->from_user = 0;
1752         }
1753
1754         if (iov_count) {
1755                 struct iovec *iov = NULL;
1756                 struct iov_iter i;
1757
1758                 res = import_iovec(rw, hp->dxferp, iov_count, 0, &iov, &i);
1759                 if (res < 0)
1760                         return res;
1761
1762                 iov_iter_truncate(&i, hp->dxfer_len);
1763                 if (!iov_iter_count(&i)) {
1764                         kfree(iov);
1765                         return -EINVAL;
1766                 }
1767
1768                 res = blk_rq_map_user_iov(q, rq, md, &i, GFP_ATOMIC);
1769                 kfree(iov);
1770         } else
1771                 res = blk_rq_map_user(q, rq, md, hp->dxferp,
1772                                       hp->dxfer_len, GFP_ATOMIC);
1773
1774         if (!res) {
1775                 srp->bio = rq->bio;
1776
1777                 if (!md) {
1778                         req_schp->dio_in_use = 1;
1779                         hp->info |= SG_INFO_DIRECT_IO;
1780                 }
1781         }
1782         return res;
1783 }
1784
1785 static int
1786 sg_finish_rem_req(Sg_request *srp)
1787 {
1788         int ret = 0;
1789
1790         Sg_fd *sfp = srp->parentfp;
1791         Sg_scatter_hold *req_schp = &srp->data;
1792
1793         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1794                                       "sg_finish_rem_req: res_used=%d\n",
1795                                       (int) srp->res_used));
1796         if (srp->bio)
1797                 ret = blk_rq_unmap_user(srp->bio);
1798
1799         if (srp->rq) {
1800                 scsi_req_free_cmd(scsi_req(srp->rq));
1801                 blk_put_request(srp->rq);
1802         }
1803
1804         if (srp->res_used)
1805                 sg_unlink_reserve(sfp, srp);
1806         else
1807                 sg_remove_scat(sfp, req_schp);
1808
1809         sg_remove_request(sfp, srp);
1810
1811         return ret;
1812 }
1813
1814 static int
1815 sg_build_sgat(Sg_scatter_hold * schp, const Sg_fd * sfp, int tablesize)
1816 {
1817         int sg_bufflen = tablesize * sizeof(struct page *);
1818         gfp_t gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
1819
1820         schp->pages = kzalloc(sg_bufflen, gfp_flags);
1821         if (!schp->pages)
1822                 return -ENOMEM;
1823         schp->sglist_len = sg_bufflen;
1824         return tablesize;       /* number of scat_gath elements allocated */
1825 }
1826
1827 static int
1828 sg_build_indirect(Sg_scatter_hold * schp, Sg_fd * sfp, int buff_size)
1829 {
1830         int ret_sz = 0, i, k, rem_sz, num, mx_sc_elems;
1831         int sg_tablesize = sfp->parentdp->sg_tablesize;
1832         int blk_size = buff_size, order;
1833         gfp_t gfp_mask = GFP_ATOMIC | __GFP_COMP | __GFP_NOWARN;
1834
1835         if (blk_size < 0)
1836                 return -EFAULT;
1837         if (0 == blk_size)
1838                 ++blk_size;     /* don't know why */
1839         /* round request up to next highest SG_SECTOR_SZ byte boundary */
1840         blk_size = ALIGN(blk_size, SG_SECTOR_SZ);
1841         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1842                 "sg_build_indirect: buff_size=%d, blk_size=%d\n",
1843                 buff_size, blk_size));
1844
1845         /* N.B. ret_sz carried into this block ... */
1846         mx_sc_elems = sg_build_sgat(schp, sfp, sg_tablesize);
1847         if (mx_sc_elems < 0)
1848                 return mx_sc_elems;     /* most likely -ENOMEM */
1849
1850         num = scatter_elem_sz;
1851         if (unlikely(num != scatter_elem_sz_prev)) {
1852                 if (num < PAGE_SIZE) {
1853                         scatter_elem_sz = PAGE_SIZE;
1854                         scatter_elem_sz_prev = PAGE_SIZE;
1855                 } else
1856                         scatter_elem_sz_prev = num;
1857         }
1858
1859         if (sfp->low_dma)
1860                 gfp_mask |= GFP_DMA;
1861
1862         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
1863                 gfp_mask |= __GFP_ZERO;
1864
1865         order = get_order(num);
1866 retry:
1867         ret_sz = 1 << (PAGE_SHIFT + order);
1868
1869         for (k = 0, rem_sz = blk_size; rem_sz > 0 && k < mx_sc_elems;
1870              k++, rem_sz -= ret_sz) {
1871
1872                 num = (rem_sz > scatter_elem_sz_prev) ?
1873                         scatter_elem_sz_prev : rem_sz;
1874
1875                 schp->pages[k] = alloc_pages(gfp_mask, order);
1876                 if (!schp->pages[k])
1877                         goto out;
1878
1879                 if (num == scatter_elem_sz_prev) {
1880                         if (unlikely(ret_sz > scatter_elem_sz_prev)) {
1881                                 scatter_elem_sz = ret_sz;
1882                                 scatter_elem_sz_prev = ret_sz;
1883                         }
1884                 }
1885
1886                 SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1887                                  "sg_build_indirect: k=%d, num=%d, ret_sz=%d\n",
1888                                  k, num, ret_sz));
1889         }               /* end of for loop */
1890
1891         schp->page_order = order;
1892         schp->k_use_sg = k;
1893         SCSI_LOG_TIMEOUT(5, sg_printk(KERN_INFO, sfp->parentdp,
1894                          "sg_build_indirect: k_use_sg=%d, rem_sz=%d\n",
1895                          k, rem_sz));
1896
1897         schp->bufflen = blk_size;
1898         if (rem_sz > 0) /* must have failed */
1899                 return -ENOMEM;
1900         return 0;
1901 out:
1902         for (i = 0; i < k; i++)
1903                 __free_pages(schp->pages[i], order);
1904
1905         if (--order >= 0)
1906                 goto retry;
1907
1908         return -ENOMEM;
1909 }
1910
1911 static void
1912 sg_remove_scat(Sg_fd * sfp, Sg_scatter_hold * schp)
1913 {
1914         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1915                          "sg_remove_scat: k_use_sg=%d\n", schp->k_use_sg));
1916         if (schp->pages && schp->sglist_len > 0) {
1917                 if (!schp->dio_in_use) {
1918                         int k;
1919
1920                         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1921                                 SCSI_LOG_TIMEOUT(5,
1922                                         sg_printk(KERN_INFO, sfp->parentdp,
1923                                         "sg_remove_scat: k=%d, pg=0x%p\n",
1924                                         k, schp->pages[k]));
1925                                 __free_pages(schp->pages[k], schp->page_order);
1926                         }
1927
1928                         kfree(schp->pages);
1929                 }
1930         }
1931         memset(schp, 0, sizeof (*schp));
1932 }
1933
1934 static int
1935 sg_read_oxfer(Sg_request * srp, char __user *outp, int num_read_xfer)
1936 {
1937         Sg_scatter_hold *schp = &srp->data;
1938         int k, num;
1939
1940         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
1941                          "sg_read_oxfer: num_read_xfer=%d\n",
1942                          num_read_xfer));
1943         if ((!outp) || (num_read_xfer <= 0))
1944                 return 0;
1945
1946         num = 1 << (PAGE_SHIFT + schp->page_order);
1947         for (k = 0; k < schp->k_use_sg && schp->pages[k]; k++) {
1948                 if (num > num_read_xfer) {
1949                         if (__copy_to_user(outp, page_address(schp->pages[k]),
1950                                            num_read_xfer))
1951                                 return -EFAULT;
1952                         break;
1953                 } else {
1954                         if (__copy_to_user(outp, page_address(schp->pages[k]),
1955                                            num))
1956                                 return -EFAULT;
1957                         num_read_xfer -= num;
1958                         if (num_read_xfer <= 0)
1959                                 break;
1960                         outp += num;
1961                 }
1962         }
1963
1964         return 0;
1965 }
1966
1967 static void
1968 sg_build_reserve(Sg_fd * sfp, int req_size)
1969 {
1970         Sg_scatter_hold *schp = &sfp->reserve;
1971
1972         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1973                          "sg_build_reserve: req_size=%d\n", req_size));
1974         do {
1975                 if (req_size < PAGE_SIZE)
1976                         req_size = PAGE_SIZE;
1977                 if (0 == sg_build_indirect(schp, sfp, req_size))
1978                         return;
1979                 else
1980                         sg_remove_scat(sfp, schp);
1981                 req_size >>= 1; /* divide by 2 */
1982         } while (req_size > (PAGE_SIZE / 2));
1983 }
1984
1985 static void
1986 sg_link_reserve(Sg_fd * sfp, Sg_request * srp, int size)
1987 {
1988         Sg_scatter_hold *req_schp = &srp->data;
1989         Sg_scatter_hold *rsv_schp = &sfp->reserve;
1990         int k, num, rem;
1991
1992         srp->res_used = 1;
1993         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, sfp->parentdp,
1994                          "sg_link_reserve: size=%d\n", size));
1995         rem = size;
1996
1997         num = 1 << (PAGE_SHIFT + rsv_schp->page_order);
1998         for (k = 0; k < rsv_schp->k_use_sg; k++) {
1999                 if (rem <= num) {
2000                         req_schp->k_use_sg = k + 1;
2001                         req_schp->sglist_len = rsv_schp->sglist_len;
2002                         req_schp->pages = rsv_schp->pages;
2003
2004                         req_schp->bufflen = size;
2005                         req_schp->page_order = rsv_schp->page_order;
2006                         break;
2007                 } else
2008                         rem -= num;
2009         }
2010
2011         if (k >= rsv_schp->k_use_sg)
2012                 SCSI_LOG_TIMEOUT(1, sg_printk(KERN_INFO, sfp->parentdp,
2013                                  "sg_link_reserve: BAD size\n"));
2014 }
2015
2016 static void
2017 sg_unlink_reserve(Sg_fd * sfp, Sg_request * srp)
2018 {
2019         Sg_scatter_hold *req_schp = &srp->data;
2020
2021         SCSI_LOG_TIMEOUT(4, sg_printk(KERN_INFO, srp->parentfp->parentdp,
2022                                       "sg_unlink_reserve: req->k_use_sg=%d\n",
2023                                       (int) req_schp->k_use_sg));
2024         req_schp->k_use_sg = 0;
2025         req_schp->bufflen = 0;
2026         req_schp->pages = NULL;
2027         req_schp->page_order = 0;
2028         req_schp->sglist_len = 0;
2029         sfp->save_scat_len = 0;
2030         srp->res_used = 0;
2031 }
2032
2033 static Sg_request *
2034 sg_get_rq_mark(Sg_fd * sfp, int pack_id)
2035 {
2036         Sg_request *resp;
2037         unsigned long iflags;
2038
2039         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2040         for (resp = sfp->headrp; resp; resp = resp->nextrp) {
2041                 /* look for requests that are ready + not SG_IO owned */
2042                 if ((1 == resp->done) && (!resp->sg_io_owned) &&
2043                     ((-1 == pack_id) || (resp->header.pack_id == pack_id))) {
2044                         resp->done = 2; /* guard against other readers */
2045                         break;
2046                 }
2047         }
2048         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2049         return resp;
2050 }
2051
2052 /* always adds to end of list */
2053 static Sg_request *
2054 sg_add_request(Sg_fd * sfp)
2055 {
2056         int k;
2057         unsigned long iflags;
2058         Sg_request *resp;
2059         Sg_request *rp = sfp->req_arr;
2060
2061         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2062         resp = sfp->headrp;
2063         if (!resp) {
2064                 memset(rp, 0, sizeof (Sg_request));
2065                 rp->parentfp = sfp;
2066                 resp = rp;
2067                 sfp->headrp = resp;
2068         } else {
2069                 if (0 == sfp->cmd_q)
2070                         resp = NULL;    /* command queuing disallowed */
2071                 else {
2072                         for (k = 0; k < SG_MAX_QUEUE; ++k, ++rp) {
2073                                 if (!rp->parentfp)
2074                                         break;
2075                         }
2076                         if (k < SG_MAX_QUEUE) {
2077                                 memset(rp, 0, sizeof (Sg_request));
2078                                 rp->parentfp = sfp;
2079                                 while (resp->nextrp)
2080                                         resp = resp->nextrp;
2081                                 resp->nextrp = rp;
2082                                 resp = rp;
2083                         } else
2084                                 resp = NULL;
2085                 }
2086         }
2087         if (resp) {
2088                 resp->nextrp = NULL;
2089                 resp->header.duration = jiffies_to_msecs(jiffies);
2090         }
2091         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2092         return resp;
2093 }
2094
2095 /* Return of 1 for found; 0 for not found */
2096 static int
2097 sg_remove_request(Sg_fd * sfp, Sg_request * srp)
2098 {
2099         Sg_request *prev_rp;
2100         Sg_request *rp;
2101         unsigned long iflags;
2102         int res = 0;
2103
2104         if ((!sfp) || (!srp) || (!sfp->headrp))
2105                 return res;
2106         write_lock_irqsave(&sfp->rq_list_lock, iflags);
2107         prev_rp = sfp->headrp;
2108         if (srp == prev_rp) {
2109                 sfp->headrp = prev_rp->nextrp;
2110                 prev_rp->parentfp = NULL;
2111                 res = 1;
2112         } else {
2113                 while ((rp = prev_rp->nextrp)) {
2114                         if (srp == rp) {
2115                                 prev_rp->nextrp = rp->nextrp;
2116                                 rp->parentfp = NULL;
2117                                 res = 1;
2118                                 break;
2119                         }
2120                         prev_rp = rp;
2121                 }
2122         }
2123         write_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2124         return res;
2125 }
2126
2127 static Sg_fd *
2128 sg_add_sfp(Sg_device * sdp)
2129 {
2130         Sg_fd *sfp;
2131         unsigned long iflags;
2132         int bufflen;
2133
2134         sfp = kzalloc(sizeof(*sfp), GFP_ATOMIC | __GFP_NOWARN);
2135         if (!sfp)
2136                 return ERR_PTR(-ENOMEM);
2137
2138         init_waitqueue_head(&sfp->read_wait);
2139         rwlock_init(&sfp->rq_list_lock);
2140
2141         kref_init(&sfp->f_ref);
2142         sfp->timeout = SG_DEFAULT_TIMEOUT;
2143         sfp->timeout_user = SG_DEFAULT_TIMEOUT_USER;
2144         sfp->force_packid = SG_DEF_FORCE_PACK_ID;
2145         sfp->low_dma = (SG_DEF_FORCE_LOW_DMA == 0) ?
2146             sdp->device->host->unchecked_isa_dma : 1;
2147         sfp->cmd_q = SG_DEF_COMMAND_Q;
2148         sfp->keep_orphan = SG_DEF_KEEP_ORPHAN;
2149         sfp->parentdp = sdp;
2150         write_lock_irqsave(&sdp->sfd_lock, iflags);
2151         if (atomic_read(&sdp->detaching)) {
2152                 write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2153                 return ERR_PTR(-ENODEV);
2154         }
2155         list_add_tail(&sfp->sfd_siblings, &sdp->sfds);
2156         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2157         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2158                                       "sg_add_sfp: sfp=0x%p\n", sfp));
2159         if (unlikely(sg_big_buff != def_reserved_size))
2160                 sg_big_buff = def_reserved_size;
2161
2162         bufflen = min_t(int, sg_big_buff,
2163                         max_sectors_bytes(sdp->device->request_queue));
2164         sg_build_reserve(sfp, bufflen);
2165         SCSI_LOG_TIMEOUT(3, sg_printk(KERN_INFO, sdp,
2166                                       "sg_add_sfp: bufflen=%d, k_use_sg=%d\n",
2167                                       sfp->reserve.bufflen,
2168                                       sfp->reserve.k_use_sg));
2169
2170         kref_get(&sdp->d_ref);
2171         __module_get(THIS_MODULE);
2172         return sfp;
2173 }
2174
2175 static void
2176 sg_remove_sfp_usercontext(struct work_struct *work)
2177 {
2178         struct sg_fd *sfp = container_of(work, struct sg_fd, ew.work);
2179         struct sg_device *sdp = sfp->parentdp;
2180
2181         /* Cleanup any responses which were never read(). */
2182         while (sfp->headrp)
2183                 sg_finish_rem_req(sfp->headrp);
2184
2185         if (sfp->reserve.bufflen > 0) {
2186                 SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2187                                 "sg_remove_sfp:    bufflen=%d, k_use_sg=%d\n",
2188                                 (int) sfp->reserve.bufflen,
2189                                 (int) sfp->reserve.k_use_sg));
2190                 sg_remove_scat(sfp, &sfp->reserve);
2191         }
2192
2193         SCSI_LOG_TIMEOUT(6, sg_printk(KERN_INFO, sdp,
2194                         "sg_remove_sfp: sfp=0x%p\n", sfp));
2195         kfree(sfp);
2196
2197         scsi_device_put(sdp->device);
2198         kref_put(&sdp->d_ref, sg_device_destroy);
2199         module_put(THIS_MODULE);
2200 }
2201
2202 static void
2203 sg_remove_sfp(struct kref *kref)
2204 {
2205         struct sg_fd *sfp = container_of(kref, struct sg_fd, f_ref);
2206         struct sg_device *sdp = sfp->parentdp;
2207         unsigned long iflags;
2208
2209         write_lock_irqsave(&sdp->sfd_lock, iflags);
2210         list_del(&sfp->sfd_siblings);
2211         write_unlock_irqrestore(&sdp->sfd_lock, iflags);
2212
2213         INIT_WORK(&sfp->ew.work, sg_remove_sfp_usercontext);
2214         schedule_work(&sfp->ew.work);
2215 }
2216
2217 static int
2218 sg_res_in_use(Sg_fd * sfp)
2219 {
2220         const Sg_request *srp;
2221         unsigned long iflags;
2222
2223         read_lock_irqsave(&sfp->rq_list_lock, iflags);
2224         for (srp = sfp->headrp; srp; srp = srp->nextrp)
2225                 if (srp->res_used)
2226                         break;
2227         read_unlock_irqrestore(&sfp->rq_list_lock, iflags);
2228         return srp ? 1 : 0;
2229 }
2230
2231 #ifdef CONFIG_SCSI_PROC_FS
2232 static int
2233 sg_idr_max_id(int id, void *p, void *data)
2234 {
2235         int *k = data;
2236
2237         if (*k < id)
2238                 *k = id;
2239
2240         return 0;
2241 }
2242
2243 static int
2244 sg_last_dev(void)
2245 {
2246         int k = -1;
2247         unsigned long iflags;
2248
2249         read_lock_irqsave(&sg_index_lock, iflags);
2250         idr_for_each(&sg_index_idr, sg_idr_max_id, &k);
2251         read_unlock_irqrestore(&sg_index_lock, iflags);
2252         return k + 1;           /* origin 1 */
2253 }
2254 #endif
2255
2256 /* must be called with sg_index_lock held */
2257 static Sg_device *sg_lookup_dev(int dev)
2258 {
2259         return idr_find(&sg_index_idr, dev);
2260 }
2261
2262 static Sg_device *
2263 sg_get_dev(int dev)
2264 {
2265         struct sg_device *sdp;
2266         unsigned long flags;
2267
2268         read_lock_irqsave(&sg_index_lock, flags);
2269         sdp = sg_lookup_dev(dev);
2270         if (!sdp)
2271                 sdp = ERR_PTR(-ENXIO);
2272         else if (atomic_read(&sdp->detaching)) {
2273                 /* If sdp->detaching, then the refcount may already be 0, in
2274                  * which case it would be a bug to do kref_get().
2275                  */
2276                 sdp = ERR_PTR(-ENODEV);
2277         } else
2278                 kref_get(&sdp->d_ref);
2279         read_unlock_irqrestore(&sg_index_lock, flags);
2280
2281         return sdp;
2282 }
2283
2284 #ifdef CONFIG_SCSI_PROC_FS
2285
2286 static struct proc_dir_entry *sg_proc_sgp = NULL;
2287
2288 static char sg_proc_sg_dirname[] = "scsi/sg";
2289
2290 static int sg_proc_seq_show_int(struct seq_file *s, void *v);
2291
2292 static int sg_proc_single_open_adio(struct inode *inode, struct file *file);
2293 static ssize_t sg_proc_write_adio(struct file *filp, const char __user *buffer,
2294                                   size_t count, loff_t *off);
2295 static const struct file_operations adio_fops = {
2296         .owner = THIS_MODULE,
2297         .open = sg_proc_single_open_adio,
2298         .read = seq_read,
2299         .llseek = seq_lseek,
2300         .write = sg_proc_write_adio,
2301         .release = single_release,
2302 };
2303
2304 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file);
2305 static ssize_t sg_proc_write_dressz(struct file *filp,
2306                 const char __user *buffer, size_t count, loff_t *off);
2307 static const struct file_operations dressz_fops = {
2308         .owner = THIS_MODULE,
2309         .open = sg_proc_single_open_dressz,
2310         .read = seq_read,
2311         .llseek = seq_lseek,
2312         .write = sg_proc_write_dressz,
2313         .release = single_release,
2314 };
2315
2316 static int sg_proc_seq_show_version(struct seq_file *s, void *v);
2317 static int sg_proc_single_open_version(struct inode *inode, struct file *file);
2318 static const struct file_operations version_fops = {
2319         .owner = THIS_MODULE,
2320         .open = sg_proc_single_open_version,
2321         .read = seq_read,
2322         .llseek = seq_lseek,
2323         .release = single_release,
2324 };
2325
2326 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v);
2327 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file);
2328 static const struct file_operations devhdr_fops = {
2329         .owner = THIS_MODULE,
2330         .open = sg_proc_single_open_devhdr,
2331         .read = seq_read,
2332         .llseek = seq_lseek,
2333         .release = single_release,
2334 };
2335
2336 static int sg_proc_seq_show_dev(struct seq_file *s, void *v);
2337 static int sg_proc_open_dev(struct inode *inode, struct file *file);
2338 static void * dev_seq_start(struct seq_file *s, loff_t *pos);
2339 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos);
2340 static void dev_seq_stop(struct seq_file *s, void *v);
2341 static const struct file_operations dev_fops = {
2342         .owner = THIS_MODULE,
2343         .open = sg_proc_open_dev,
2344         .read = seq_read,
2345         .llseek = seq_lseek,
2346         .release = seq_release,
2347 };
2348 static const struct seq_operations dev_seq_ops = {
2349         .start = dev_seq_start,
2350         .next  = dev_seq_next,
2351         .stop  = dev_seq_stop,
2352         .show  = sg_proc_seq_show_dev,
2353 };
2354
2355 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v);
2356 static int sg_proc_open_devstrs(struct inode *inode, struct file *file);
2357 static const struct file_operations devstrs_fops = {
2358         .owner = THIS_MODULE,
2359         .open = sg_proc_open_devstrs,
2360         .read = seq_read,
2361         .llseek = seq_lseek,
2362         .release = seq_release,
2363 };
2364 static const struct seq_operations devstrs_seq_ops = {
2365         .start = dev_seq_start,
2366         .next  = dev_seq_next,
2367         .stop  = dev_seq_stop,
2368         .show  = sg_proc_seq_show_devstrs,
2369 };
2370
2371 static int sg_proc_seq_show_debug(struct seq_file *s, void *v);
2372 static int sg_proc_open_debug(struct inode *inode, struct file *file);
2373 static const struct file_operations debug_fops = {
2374         .owner = THIS_MODULE,
2375         .open = sg_proc_open_debug,
2376         .read = seq_read,
2377         .llseek = seq_lseek,
2378         .release = seq_release,
2379 };
2380 static const struct seq_operations debug_seq_ops = {
2381         .start = dev_seq_start,
2382         .next  = dev_seq_next,
2383         .stop  = dev_seq_stop,
2384         .show  = sg_proc_seq_show_debug,
2385 };
2386
2387
2388 struct sg_proc_leaf {
2389         const char * name;
2390         const struct file_operations * fops;
2391 };
2392
2393 static const struct sg_proc_leaf sg_proc_leaf_arr[] = {
2394         {"allow_dio", &adio_fops},
2395         {"debug", &debug_fops},
2396         {"def_reserved_size", &dressz_fops},
2397         {"device_hdr", &devhdr_fops},
2398         {"devices", &dev_fops},
2399         {"device_strs", &devstrs_fops},
2400         {"version", &version_fops}
2401 };
2402
2403 static int
2404 sg_proc_init(void)
2405 {
2406         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2407         int k;
2408
2409         sg_proc_sgp = proc_mkdir(sg_proc_sg_dirname, NULL);
2410         if (!sg_proc_sgp)
2411                 return 1;
2412         for (k = 0; k < num_leaves; ++k) {
2413                 const struct sg_proc_leaf *leaf = &sg_proc_leaf_arr[k];
2414                 umode_t mask = leaf->fops->write ? S_IRUGO | S_IWUSR : S_IRUGO;
2415                 proc_create(leaf->name, mask, sg_proc_sgp, leaf->fops);
2416         }
2417         return 0;
2418 }
2419
2420 static void
2421 sg_proc_cleanup(void)
2422 {
2423         int k;
2424         int num_leaves = ARRAY_SIZE(sg_proc_leaf_arr);
2425
2426         if (!sg_proc_sgp)
2427                 return;
2428         for (k = 0; k < num_leaves; ++k)
2429                 remove_proc_entry(sg_proc_leaf_arr[k].name, sg_proc_sgp);
2430         remove_proc_entry(sg_proc_sg_dirname, NULL);
2431 }
2432
2433
2434 static int sg_proc_seq_show_int(struct seq_file *s, void *v)
2435 {
2436         seq_printf(s, "%d\n", *((int *)s->private));
2437         return 0;
2438 }
2439
2440 static int sg_proc_single_open_adio(struct inode *inode, struct file *file)
2441 {
2442         return single_open(file, sg_proc_seq_show_int, &sg_allow_dio);
2443 }
2444
2445 static ssize_t
2446 sg_proc_write_adio(struct file *filp, const char __user *buffer,
2447                    size_t count, loff_t *off)
2448 {
2449         int err;
2450         unsigned long num;
2451
2452         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2453                 return -EACCES;
2454         err = kstrtoul_from_user(buffer, count, 0, &num);
2455         if (err)
2456                 return err;
2457         sg_allow_dio = num ? 1 : 0;
2458         return count;
2459 }
2460
2461 static int sg_proc_single_open_dressz(struct inode *inode, struct file *file)
2462 {
2463         return single_open(file, sg_proc_seq_show_int, &sg_big_buff);
2464 }
2465
2466 static ssize_t
2467 sg_proc_write_dressz(struct file *filp, const char __user *buffer,
2468                      size_t count, loff_t *off)
2469 {
2470         int err;
2471         unsigned long k = ULONG_MAX;
2472
2473         if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2474                 return -EACCES;
2475
2476         err = kstrtoul_from_user(buffer, count, 0, &k);
2477         if (err)
2478                 return err;
2479         if (k <= 1048576) {     /* limit "big buff" to 1 MB */
2480                 sg_big_buff = k;
2481                 return count;
2482         }
2483         return -ERANGE;
2484 }
2485
2486 static int sg_proc_seq_show_version(struct seq_file *s, void *v)
2487 {
2488         seq_printf(s, "%d\t%s [%s]\n", sg_version_num, SG_VERSION_STR,
2489                    sg_version_date);
2490         return 0;
2491 }
2492
2493 static int sg_proc_single_open_version(struct inode *inode, struct file *file)
2494 {
2495         return single_open(file, sg_proc_seq_show_version, NULL);
2496 }
2497
2498 static int sg_proc_seq_show_devhdr(struct seq_file *s, void *v)
2499 {
2500         seq_puts(s, "host\tchan\tid\tlun\ttype\topens\tqdepth\tbusy\tonline\n");
2501         return 0;
2502 }
2503
2504 static int sg_proc_single_open_devhdr(struct inode *inode, struct file *file)
2505 {
2506         return single_open(file, sg_proc_seq_show_devhdr, NULL);
2507 }
2508
2509 struct sg_proc_deviter {
2510         loff_t  index;
2511         size_t  max;
2512 };
2513
2514 static void * dev_seq_start(struct seq_file *s, loff_t *pos)
2515 {
2516         struct sg_proc_deviter * it = kmalloc(sizeof(*it), GFP_KERNEL);
2517
2518         s->private = it;
2519         if (! it)
2520                 return NULL;
2521
2522         it->index = *pos;
2523         it->max = sg_last_dev();
2524         if (it->index >= it->max)
2525                 return NULL;
2526         return it;
2527 }
2528
2529 static void * dev_seq_next(struct seq_file *s, void *v, loff_t *pos)
2530 {
2531         struct sg_proc_deviter * it = s->private;
2532
2533         *pos = ++it->index;
2534         return (it->index < it->max) ? it : NULL;
2535 }
2536
2537 static void dev_seq_stop(struct seq_file *s, void *v)
2538 {
2539         kfree(s->private);
2540 }
2541
2542 static int sg_proc_open_dev(struct inode *inode, struct file *file)
2543 {
2544         return seq_open(file, &dev_seq_ops);
2545 }
2546
2547 static int sg_proc_seq_show_dev(struct seq_file *s, void *v)
2548 {
2549         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2550         Sg_device *sdp;
2551         struct scsi_device *scsidp;
2552         unsigned long iflags;
2553
2554         read_lock_irqsave(&sg_index_lock, iflags);
2555         sdp = it ? sg_lookup_dev(it->index) : NULL;
2556         if ((NULL == sdp) || (NULL == sdp->device) ||
2557             (atomic_read(&sdp->detaching)))
2558                 seq_puts(s, "-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\t-1\n");
2559         else {
2560                 scsidp = sdp->device;
2561                 seq_printf(s, "%d\t%d\t%d\t%llu\t%d\t%d\t%d\t%d\t%d\n",
2562                               scsidp->host->host_no, scsidp->channel,
2563                               scsidp->id, scsidp->lun, (int) scsidp->type,
2564                               1,
2565                               (int) scsidp->queue_depth,
2566                               (int) atomic_read(&scsidp->device_busy),
2567                               (int) scsi_device_online(scsidp));
2568         }
2569         read_unlock_irqrestore(&sg_index_lock, iflags);
2570         return 0;
2571 }
2572
2573 static int sg_proc_open_devstrs(struct inode *inode, struct file *file)
2574 {
2575         return seq_open(file, &devstrs_seq_ops);
2576 }
2577
2578 static int sg_proc_seq_show_devstrs(struct seq_file *s, void *v)
2579 {
2580         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2581         Sg_device *sdp;
2582         struct scsi_device *scsidp;
2583         unsigned long iflags;
2584
2585         read_lock_irqsave(&sg_index_lock, iflags);
2586         sdp = it ? sg_lookup_dev(it->index) : NULL;
2587         scsidp = sdp ? sdp->device : NULL;
2588         if (sdp && scsidp && (!atomic_read(&sdp->detaching)))
2589                 seq_printf(s, "%8.8s\t%16.16s\t%4.4s\n",
2590                            scsidp->vendor, scsidp->model, scsidp->rev);
2591         else
2592                 seq_puts(s, "<no active device>\n");
2593         read_unlock_irqrestore(&sg_index_lock, iflags);
2594         return 0;
2595 }
2596
2597 /* must be called while holding sg_index_lock */
2598 static void sg_proc_debug_helper(struct seq_file *s, Sg_device * sdp)
2599 {
2600         int k, m, new_interface, blen, usg;
2601         Sg_request *srp;
2602         Sg_fd *fp;
2603         const sg_io_hdr_t *hp;
2604         const char * cp;
2605         unsigned int ms;
2606
2607         k = 0;
2608         list_for_each_entry(fp, &sdp->sfds, sfd_siblings) {
2609                 k++;
2610                 read_lock(&fp->rq_list_lock); /* irqs already disabled */
2611                 seq_printf(s, "   FD(%d): timeout=%dms bufflen=%d "
2612                            "(res)sgat=%d low_dma=%d\n", k,
2613                            jiffies_to_msecs(fp->timeout),
2614                            fp->reserve.bufflen,
2615                            (int) fp->reserve.k_use_sg,
2616                            (int) fp->low_dma);
2617                 seq_printf(s, "   cmd_q=%d f_packid=%d k_orphan=%d closed=0\n",
2618                            (int) fp->cmd_q, (int) fp->force_packid,
2619                            (int) fp->keep_orphan);
2620                 for (m = 0, srp = fp->headrp;
2621                                 srp != NULL;
2622                                 ++m, srp = srp->nextrp) {
2623                         hp = &srp->header;
2624                         new_interface = (hp->interface_id == '\0') ? 0 : 1;
2625                         if (srp->res_used) {
2626                                 if (new_interface &&
2627                                     (SG_FLAG_MMAP_IO & hp->flags))
2628                                         cp = "     mmap>> ";
2629                                 else
2630                                         cp = "     rb>> ";
2631                         } else {
2632                                 if (SG_INFO_DIRECT_IO_MASK & hp->info)
2633                                         cp = "     dio>> ";
2634                                 else
2635                                         cp = "     ";
2636                         }
2637                         seq_puts(s, cp);
2638                         blen = srp->data.bufflen;
2639                         usg = srp->data.k_use_sg;
2640                         seq_puts(s, srp->done ?
2641                                  ((1 == srp->done) ?  "rcv:" : "fin:")
2642                                   : "act:");
2643                         seq_printf(s, " id=%d blen=%d",
2644                                    srp->header.pack_id, blen);
2645                         if (srp->done)
2646                                 seq_printf(s, " dur=%d", hp->duration);
2647                         else {
2648                                 ms = jiffies_to_msecs(jiffies);
2649                                 seq_printf(s, " t_o/elap=%d/%d",
2650                                         (new_interface ? hp->timeout :
2651                                                   jiffies_to_msecs(fp->timeout)),
2652                                         (ms > hp->duration ? ms - hp->duration : 0));
2653                         }
2654                         seq_printf(s, "ms sgat=%d op=0x%02x\n", usg,
2655                                    (int) srp->data.cmd_opcode);
2656                 }
2657                 if (0 == m)
2658                         seq_puts(s, "     No requests active\n");
2659                 read_unlock(&fp->rq_list_lock);
2660         }
2661 }
2662
2663 static int sg_proc_open_debug(struct inode *inode, struct file *file)
2664 {
2665         return seq_open(file, &debug_seq_ops);
2666 }
2667
2668 static int sg_proc_seq_show_debug(struct seq_file *s, void *v)
2669 {
2670         struct sg_proc_deviter * it = (struct sg_proc_deviter *) v;
2671         Sg_device *sdp;
2672         unsigned long iflags;
2673
2674         if (it && (0 == it->index))
2675                 seq_printf(s, "max_active_device=%d  def_reserved_size=%d\n",
2676                            (int)it->max, sg_big_buff);
2677
2678         read_lock_irqsave(&sg_index_lock, iflags);
2679         sdp = it ? sg_lookup_dev(it->index) : NULL;
2680         if (NULL == sdp)
2681                 goto skip;
2682         read_lock(&sdp->sfd_lock);
2683         if (!list_empty(&sdp->sfds)) {
2684                 seq_printf(s, " >>> device=%s ", sdp->disk->disk_name);
2685                 if (atomic_read(&sdp->detaching))
2686                         seq_puts(s, "detaching pending close ");
2687                 else if (sdp->device) {
2688                         struct scsi_device *scsidp = sdp->device;
2689
2690                         seq_printf(s, "%d:%d:%d:%llu   em=%d",
2691                                    scsidp->host->host_no,
2692                                    scsidp->channel, scsidp->id,
2693                                    scsidp->lun,
2694                                    scsidp->host->hostt->emulated);
2695                 }
2696                 seq_printf(s, " sg_tablesize=%d excl=%d open_cnt=%d\n",
2697                            sdp->sg_tablesize, sdp->exclude, sdp->open_cnt);
2698                 sg_proc_debug_helper(s, sdp);
2699         }
2700         read_unlock(&sdp->sfd_lock);
2701 skip:
2702         read_unlock_irqrestore(&sg_index_lock, iflags);
2703         return 0;
2704 }
2705
2706 #endif                          /* CONFIG_SCSI_PROC_FS */
2707
2708 module_init(init_sg);
2709 module_exit(exit_sg);